Hayes Introductory Linguistics 2021

Introductory Linguistics
A textbook by
Bruce P. Hayes
Department of Linguistics
University of California, Los Angeles
Copyright 2021 by Bruce P. Hayes
This copyrighted textbook may be freely read by anyone. It may be used by any linguistics teacher
for teaching purposes, under the condition that you notify the author by email
(bhayes@humnet.ucla.edu) that you are using it. Comments and corrections, including from
students, are welcome.
Hayes Introductory Linguistics p. 2
Contents
Preface p. 3
Chapter 1: What is Linguistics? p. 4
Chapter 2: Morphology p. 19
Chapter 3: Normative views of language p. 67
Chapter 4: Syntax I — Phrase Structure p. 77
Chapter 5: Syntax II — Transformations p. 159
Chapter 6: Syntax III — Subcategorization and Wh- Movement p. 175
Chapter 7: Language Acquisition p. 256
Chapter 8: Review of Morphology and Syntax p. 266
Chapter 9: Semantics p. 311
Chapter 10: Phonetics p. 378
Chapter 11: Phonology I — Phonological Rules and Phonemic Analysis p. 420
Chapter 12: Phonology II — Alternation and Neutralization p. 481
Chapter 13: Historical Linguistics p. 507
Chapter 14: Applications and Outlook p. 563
Chapter 15: More review problems p. 571
Preface
This text has been written by me over the years for the course “Linguistics 20: Introduction to
Linguistic Analysis”, which I teach in my home department at UCLA. The course is meant to be a
short introduction to “core” linguistics, by which I mean the analysis of language data using
theory. (My department covers broader issues, such as language in society, in a separate course,
“Introduction to Language”.) To the extent that my text is successful, students will get a clear idea
of the goals and character of linguistic analysis and will be well prepared to take on the various
subfields of linguistics in later specialized courses.
My text follows mainstream thinking in linguistics in assuming that learning the field is best
done through exercises in which students deal with language data, trying to discover the pattern
and express it clearly with rules. The course I teach includes weekly homeworks of this kind; these
homeworks include the most ambitious problems. In this text there also 90 Study Exercises; some
are interspersed at appropriate moments in the presentation; others are placed in Chapters 8 and 15,
meant for pre-midterm and pre-final review. I have arranged the page breaks to make it convenient
for students to try to solve the exercises themselves before consulting the printed answer.
The main purpose of the exercises is to help students make the essential transition from
passive knowledge (material makes perfect sense when the professor or text explains it) to active
knowledge (student can apply the theory in new contexts and make independent assessments). In
truth, I also hope that the exercises will be not just a way of achieving control over the material but
at least occasionally a source of intellectual pleasure. Most linguists I know enjoy the puzzles
presented by language data and I hope that for the reader it will be the same.
Over the years I have worked at filling gaps in the material and as a result the book now
exceeds the time available for teaching it in the 10-week academic term at UCLA; this leads me
sometimes to skip chapters 3, 7, and/or 13. I believe it would not be hard to get through the whole
text in a course taught on a semester system.
My thanks go to the many students who have read through earlier versions of this text, often
usefully pointing out errors (you can become one of them — my email is
bhayes@humnet.ucla.edu). I also thank my many teaching assistants for their wisdom and first-
hand experience, along with my colleagues Sandra Disner, Craig Melchert, and Jessica Rett for
expert advice.
Chapter 1: What is Linguistics?
1. What this book will be like
Linguistics is the science of language; it studies the structure of human languages and aims to
develop a general theory of how languages work. Linguists over the years have worked out
theories and analytical techniques that can help us to understand the structure of a language at a
deeper level. The purpose of this text is to help you learn some of these theories and techniques —
enough to get an idea of what linguistics is like, and a basis for going on to more advanced
coursework in the field if that is where you are headed.
There are basically three things I hope you will get out of this book.
First, there is the subject matter itself, which is useful to know for people in many different
fields, such as education, psychology, and computation. The course is also an introduction to
linguistics for those who are going to major in it.
Second, the course involves some mental exercise, involving analysis of data from English
and other languages. I doubt that anyone who doesn’t go on in linguistics will remember much of
the course material five years after they have graduated, but the analytical skills in which you will
get practice will be (I hope) both more permanent and more useful.
Third, the course is intended to give a participatory role in science than is possible in most
science courses. The reason we can do this in linguistics is that it is a fairly primitive science,
without an enormous body of well-established results. Because of this, we are less interested in
teaching you a body of established knowledge; rather, our focus is on teaching you to decide what
is right on your own, by looking at the data, or (for languages you speak) coming up with new data
on your own. All sciences are in this state of uncertainty at their frontiers; linguistics can give you
a more authentically scientific experience in a beginning course.
2. Implicit and explicit knowledge of language; working with consultants
Linguists are constantly asked the question “How many languages do you speak?” This
question is a little irritating, because it is not all that relevant to what linguists are trying to do. The
goals of linguistics are to describe and understand the structure of human languages; to discover
the principles by which human languages work, both at the level of specific languages and at the
level of language in general. Even if one could speak all 8000 or so of the world’s languages, one
would not have solved any of the problems of linguistics.
The reason is this: speaking a language and knowing its structure are two different things. In
speaking a language, one uses thousands of grammatical rules without being aware of them; they
are unconscious knowledge. Linguists attempt to make explicit this unconscious knowledge by
looking closely at the data of language. That is, they attempt to make the implicit knowledge of
native speakers into explicit knowledge.
This goal implies one of the central methods of doing linguistic research, the consultant
session. Quite often, a linguist will study the structure of language she does not speak; this is done
by finding a native-speaker consultant to provide the data. The linguist normally asks the
consultant a great number of questions. Some of them are simple and establish basic knowledge:
(1) “What is the word for “duck” in your language?”
Others look for the various different grammatical forms of the same word:
(2) “How would you say “two ducks”?
This is probably meant to discover how plurals are formed. Others queries involve whole sentences
and often their meanings as well.
The crucial idea in a consultant session is that the linguist is thinking about structure—is
making and checking hypotheses. The native speaker is most often trying simply to provide an
accurate report of how she speaks the language, and of her intuitions about meaning and other
matters.
Obviously, the lines can be blurred a bit: sometimes the consultant (especially if she knows
some linguistics), may want to suggest some hypotheses herself. And linguists sometimes “work
on themselves,” so that the dialogue across the consultant table becomes an internal dialogue in the
mind of the linguist.1
The following example illustrates the method: for one particular area of English grammar, we
get some native speaker intuitions, and work out a series of hypotheses for what the rules of
English are. We’ll assume without comment that we are working with a native speaker of English,
and indeed, I believe that the data below are characteristic of intuitions of English native speakers.
The point of the analysis will be to illustrate a consistent truth about linguistics: the native
speaker consultant doesn’t know the answer. You cannot effectively ask the consultant to provide
the linguistic analysis. However, the native speaker does have the tacit, intuitive knowledge that
makes it possible to find the answer, or at least to get closer to it.
3. The reference of each other
I will illustrate with an example, where the language is English and the item under study is the
rather special expression each other. In the sentence (3), a native speaker of English is likely to tell
you that each other refers to we, and that it means something like, “I like you and you like me.”
(3) We like each other.

(4) Alice and Sue like each other.
1
In practice I and probably other linguists find this hard to do; it’s just too much going on in your head at once.
More important, it poses methodological problems; the data are likely to be contaminated by wishful thinking.
In linguistics this is often called the reciprocal reading; i.e. it says we are in a state of reciprocal
liking. Sentence (4) has a similar reciprocal reading. You could think of each other as sort of
pronoun that has this special reciprocal reading (the reference of pronouns is covered in more
detail in Chapter 9).
Sentence (5) is a bizarre sentence, in that each other cannot logically refer to I.
(5) *I like each other.
The native speaker responds to it by saying, “That’s weird” or “That’s bizarre” or “You can’t say
that in English.” We will say for present purposes that (5) is ungrammatical; that is, ill-formed.
Following standard practice, I will place an asterisk before sentences that are ungrammatical. We
can also record a rule, perhaps trivial in this context but at least it is explicit.
(6) Restriction on reciprocals
Each other can only refer to plural entities (collections of more than one thing).
In (5), the ungrammaticality can be traced to the absence of any plausible interpretation for the
sentence; since each other describes reciprocal actions, like this:
(7) X Y
X Y
Each other cannot be used unless the agent of the action is plural. But not all cases can be
explained in this way. In (8), you can think of a meaning that the sentence could in principle have,
but this meaning is not allowed by the rules of English grammar (think through what this meaning
would be, then check yourself by reading this footnote2):
(8) *Alice and Sue think I like each other.
In other words, being grammatical and having a sensible meaning are two different things.
Sentence (9) shows the same thing: you can think up two logically possible meanings, but
only one meaning is allowed by the rules of English.
(9) We believe they like each other.
Again, reason it through for yourself then read the footnote.3
2
Alice thinks I like Sue and Sue thinks I like Alice.
3
Possible meaning: if they refers, for instance to Bill and John, then We believe that Bill likes John
and John likes Bill. Impossible meaning: I believe that John and Bill like you and you believe that John and
Bill like me.
We’ve now reached our basic point: there must be some rule of English that accounts for what
each other can refer to, but it is a tacit rule. No one can look inside their mind to find out what the
rule is; one can only look at the data and try to figure the rule out. Linguists have worked on this
particular rule for some time, and have gradually made progress in stating the rule accurately.
To make sense of this rule, we need some theoretical apparatus to work with — the elements
of language structure that the rules refer to. A very important such element is defined (rather
casually) in (10).
(10) Defn.: clause
A clause is either a whole sentence or a sentence within a sentence
You can identify clauses because they generally have a subject and a verb, and they express some
sort of proposition or complete thought. One way to depict clauses is to draw brackets around
them, labeled with subscript “S” for “sentence”.
(11) [ We like each other. ]S

(12) [ John and Bill like each other. ]S
(13) *[ I like each other. ]S
(14) *[ John and Bill think [I like each other. ]S]S
(15) [ We believe [ they like each other. ]S]S
Notice that clauses can have clauses inside them. In (14), there is a clause that expresses the
content of John and Bill’s thoughts (I like each other), and the whole thing is an (ungrammatical)
clause that describes a state (John and Bill are having a particular thought.) Typically, but not
always, clauses inside clauses will occur when the verb of the main clause is a verb of thinking or
saying.
We also need to define noun phrases.
(16) Defn.: noun phrase
A noun phrase is a complete syntactic unit that refers to a thing or a set of things.
So, in (11), we is a noun phrase, and each other is a noun phrase. In (14), John and Bill is a
noun phrase4 and again so is each other.
With these definitions, we can write a tentative rule for what each other refers to:
4
Also, John is a noun phrase, and Bill is a noun phrase; they are noun phrases that are inside another
noun phrase, just as there are sentences inside sentences.
(17) Each Other Reference Rule
Each other must take its reference from a noun phrase that inside the smallest clause
containing it.
Like all proposed linguistic rules, this should be applied with care, checking to see if it works.
We work more carefully by employing appropriate graphics, like underlining the noun phrases5
and putting brackets around the clauses.
Consider first (15), We believe they like each other. We want each other to refer only to they,
and not to we. We can underline and bracket in the appropriate way, and try drawing arrows
indicating candidates for the reference of each other, like this:
(18) [We believe [they like each other.]S]S

ok
impossible
Since the smallest clause containing each other is the smaller one, and it contains the noun
phrase they, the theory predicts that each other should be able to refer to they and not to we. This
seems to be correct; i.e. so far the theory is working.
Considering next John and Bill think I like each other. I suggest at this point you jot down the
structure yourself (brackets, underlines, arrows), and check what you wrote against what you see in
(19) on the next page.
5
The underlining method will be used only in this chapter; later will be develop a more fluent notation.
(19) *[John and Bill think [I like each other.]S]S

impossible
impossible
Why is this sentence ungrammatical? The reasoning here is more subtle. Each other cannot
take its reference from John and Bill because of the Each Other Reference Rule, it can only takes
its reference from noun phrases inside the smallest clause that contains it. But then why not let
each other refer to I? Here, we run into the Restriction on Reciprocals given earlier in (6): each
other can only refer to plural noun phrases. So no matter how you attempt to interpret each other,
the sentence gets blocked, and in the end it is simply ungrammatical.
Cases (11)-(13) are easy: there is only one noun phrase for each other to refer to, and the rule
permits this.
3.1 Ambiguous sentences with each other
Notice that in a sentence with just one clause, but two noun phrases in addition to each other,
there will be two possibilities for what each other might refer to:
(20) [We wrapped the ropes around each other.]S
ok
ok
This is just what the Each Other Reference Rule predicts. Because of this, the sentence has two
possible meanings. Try making up similar cases (a few are given in the footnote6).
3.2 Chains of reference
Here is a slightly delicate case; it is special because it contains the following: a full Noun
Phrase, John and Bill, an ordinary pronoun they, and the reciprocal expression each other.:
(21) [John and Bill say [they like each other.]S]S

ok
??
The sentence is perfectly fine, but what should we say about the arrow labeled “??”? John and Bill
is not inside the smallest clause that contains each other. So, according to the Each Other
Reference Rule, each other cannot “take its reference from” John and Bill. What is special about
this sentence is that (under the most typical reading), we get a sort of chain: the pronoun they takes
its reference from John and Bill, and the reflexive pronoun each other takes its reference from
6
We assigned the representatives to each other. We instructed Fred and Sue on behalf of each other.
Alice and Sue introduced the students to each other’s mother. We prepared the cannibals for each other.
they. Because of this chain of referring, each other ends up referring to the same thing as John and
Bill. But it does so indirectly, via the intermediate pronoun. The confusion around the “??” arises
from the existence of this chain. (My own preference would be to use the label “ok”, noting the
special case of the reference chain.
We will cover the behavior of ordinary pronouns like they in Chapter 9.
4. The role of implicit noun phrases
Here are some further relevant data, which are perhaps syntactically the most interesting:
(22) [We consulted two detectives in order [ to find out about each other]S]S
(23) [They seem to us [to like each other]S]S
These sentences are mysterious: it looks like there is no noun phrase at all that occurs inside
the smallest clause containing each other (other than each other itself). You might think that the
sentences would be ungrammatical, since the Each Other Reference Rule says that each other can
only takes it reference from within the same clause. No other Noun Phrase from which it could
take its reference is apparent.
In cases like this, linguists are often willing to go out on a limb, setting up an abstract
structure, which has no pronunciation but can be justified more subtly from the behavior of the
language.
Consider more carefully the meaning sentences (22) and (23): someone is doing the finding
out in (22), namely, “we”, and someone is doing the liking in (23), namely “they”. Thus, the
peculiar clauses to find out about each other and to like each other appear to have implicit noun
phrases. They have a meaning, but they’re not pronounced.
In fact, a detail of English grammar is matched to this pattern: in both (22) and (23) the verb
is it so-called “infinitive” form (no suffix like -s or -ed), and the grammatical element to is present.
This is one of the ways that English indicates that a sentence has an implicit noun phrase as its
subject.
Exploring this line of analysis, let us fill in the explicit subjects, inserting overt noun phrases
that designate what the implicit noun phrases mean:
(24) [We consulted two detectives in order [(we) to find out about each other]S]S
(25) [They seem to us [(they) to like each other]S]S
With the implicit subjects filled in, we can explain what is going on. The Each Other
Reference Rule needs slight revising:
(26) Each Other Reference Rule (revised)
Each other must take its reference from noun phrases (including implicit noun phrases) that
are inside the smallest clause containing it.
So now, to apply the Each Other Reference rule properly, we need to evaluate the reference of
implicit noun phrases. Here is an analytic diagram for sentence (20):
(27) [We consulted two detectives in order [(we) to find out about each other]S]S
ok
impossible
We will do more on this kind of rule later. The major gap in the analysis as given so far is that
we haven’t said anything at all about what causes the implicit noun phrase to take on a particular
meaning—for instance, why does the implicit noun phrase in (25) have to mean we, and not two
detectives? This problem, too, has at least a partial solution in linguistics. But we can’t go into it
now, for reasons of time and space.7 A full-scale textbook on syntax would, of course, take on this
question.
The analysis of each other also makes a general point about analytic practice in linguistics:
linguists are, in general, willing to propose structural analyses that include inaudible entities like
implicit noun phrases. Such analyses are always more controversial, since they rest on inference
rather than directly observable facts. But they can be supported, notably by referring to the
meaning of sentences and to the overall coherence of the language system that they make possible.
5. Each other: the bottom line
The example of each other has two purposes here. First, it is meant to give you a flavor of
linguistic analysis: we assume that utterances have structure, like clauses and noun phrases and
implicit noun phrases. We also assume that the language has rules, like (17), the Each Other
Reference Rule. Using these Much of the work of linguistics involves analysis, finding out the
structures and rules that can do justice to the facts of a language.
The other main point so far concerns the question of unconscious knowledge. Any native
speaker of English will have the intuitions about grammaticality and what refers to what; that is,
the crucial information that the linguist will to justify the analysis. So knowing English means that
you “know” the Each Other Reference Rule, in an intuitive, unconscious sense. But it does not
7
If you’d like a one-paragraph answer, here it is: (1) When the verb of the main clause is persuade (and other
verbs like tell or ask), the implicit noun phrase must refer to the object of the verb: [ Sue persuaded Alice [ (Alice) to
go ]S]S. (2) When the verb of the main clause is promise (at least in some dialects of English), the implicit noun phrase
must refer to the subject of the verb: [ Sue promised Alice [ (Sue) to go ]S]S. (3) In the in order to construction, the
implicit noun phrase must refer to the subject of the main clause, as in (22).
mean that you know it explicitly. There is no English speaker on earth who can just “look inside
her head” and say what the rule is. The only way to make progress on language structure is the
more indirect way laid out here: we make hypotheses about structure and about rules, then refine
and improve them as we encounter more language data. The procedure is actually much the same
as in other sciences: we gather data (here, from the native speaker) and formulate hypotheses. The
hypotheses will enable us to make predictions about what we should find in additional data, which
we can then gather from a native speaker. Sometimes the new data makes our hypotheses look
good, increasing our confidence in them; and sometimes the new data is problematic for our
hypotheses, forcing us to modify them or even abandon them and start over. With patience, we can
achieve gradual progress. Though we cannot directly access the speaker’s unconscious knowledge
of her language, our repeated inquiries can achieve an ever better approximation of it.
In the long run, we want linguistics (like any science) to have strong predictive capacity. With
a really good theory, we can make an accurate prediction about the intuitions of a native speaker
about a sentence before we ever elicit it. This is a hard goal and unlikely to be achieved very soon.
6. The field of linguistics
Linguistics as covered in this introductory text focuses on the (largely implicit) knowledge
people have when they speak a language. This focus extends to four subareas that can each, to
some extent, be covered separately (although they all interrelate):
Syntax study of rules for forming sentences

Semantics study of rules for meaning
Morphology study of rules for forming words
Phonology study of rules of pronunciation
In all four subfields, the rules being studied are of the kind known implicitly by native speakers, as
discussed above, not the kind learned in school.
Linguists attempt to arrive at explicit knowledge of all the world’s languages, and the method
of querying a native speaker consultant is applicable to all four areas listed above. Ideally, this
method should be applied to all the languages of the world, in order to obtain a complete picture.
Unfortunately, it is unlikely that this task will never be completed. First, there are over 8000
different languages, many of which are endangered and likely will soon be no longer available for
study. Second, the amount of explicit knowledge contained even in a single language would fill a
whole library. Linguists find it both frustrating and astonishing that a small child can acquire
implicitly in just a few years the same knowledge that takes decades of hard work for linguists to
figure out explicitly.
While working out the systems of individual languages, linguists are simultaneously interested
in developing general theories of language, which provide a framework for language-specific
analysis. In principle, a complete general theory of language will lay out the properties that all
languages have in common, and the ways in which individual languages may vary.
Linguistics has other major subfields that relate the four core fields just mentioned to various
aspects of external reality:
Phonetics study of how sounds are produced and perceived

Historical linguistics study of how languages change and evolve
Sociolinguistics study of how the structure of languages relates to the structure
of societies
Acquisition study of how linguistics rule systems are acquired in early childhood
Computational mathematical properties and practical implementation of rule
systems
This textbook will allude to these fields from time to time, but for the most part you will have to
take other courses or do outside reading to learn about them.
7. What’s this about “native speakers”?
I mentioned native speakers at the start of this chapter and would like to fill in a bit of
information for why native speakers are considered important in linguistics.
A native speaker, to give a very strict definition, is someone who has heard their language
continuously since birth, learned it in the natural way from exposure (as we say, “at his mother’s
knee”; “at her father’s knee”), and continues to speak it regularly in everyday life, so they stay in
practice, as it were. You can be a native speaker of one language, or sometimes of two or more, or
of none. The latter usually happens when someone switches languages in mid-childhood.
I wish to emphasize: there are no value judgments being made when linguists talk about native
speakers! We don’t think native speakers are better (or worse) than other people. However, like
scientists everywhere, we want our data to be replicable: another scientist should be able to
conduct the same research and find out if the original results were correct. The native speaker idea
is meant to assist replicability. Linguist #1 can say, “the results of this study come from four native
speakers of Language X, all of whom speak the dialect characteristic of middle-class inhabitants of
city Y.” Then, if Linguist #2 wants to carry out further study, or check #1’s results, she can go to
city Y, find middle-class native speakers, and do this additional research or checking.
The reason using native speakers as consultants makes this easier is that native speakers tend
to be more uniform in their linguistic systems, have more confidence in the forms of their speech,
and typically speak their language in its richest and most intricate form. Non-native speakers are
more vulnerable to arbitrary, external factors. These include interference from their native
language or the accidents of what sort of data they encountered learning their nonnative language.
Study Exercises for “each other”

The first exercise is a very basic one: finding clauses and implicit subjects. This will matter a lot
later on. The others are more directly related to the each other phenomenon.
Study Exercise #1
This exercise simply asks you to find the clauses and the Noun Phrases. Put [ … ]S brackets around
the clauses and underline the Noun Phrases. Be sure to get all of them. If the subject is implicit, put
“( )” where the subject would be and say what it stands for.
Examples:
i. Alice believes that Fred sang. Answer: [ Alice believes that [ Fred sang ]S ]S
ii. Alice hopes to climb Everest. Answer [ Alice hopes [ (Alice) to climb Everest ]S ]S
Exercises:
a. I believe that turtles can swim.
b. The fact that Fred left bothers Alice.
c. Bill said that Jane sang and Fred danced.
d. I persuaded Fred to buy a telescope.
e. I promised Fred to buy a telescope.
f. To appear on television is her fondest dream.
g. Joe said that he wants to leave.
h. That Jane can sing tenor makes no difference.
i. Bill left because he was tired.
j. the idea that truth is obtainable
Answer to Study Exercise #1
a. [ I believe that [ turtles can swim. ]S ]S

b. [ The fact that [ Fred left ]S bothers Alice. ]S
Comment: the fact that Fred left is a Noun Phrase with a Noun Phrase inside it; hence double
underlines. We’ll come back to this; but let’s not worry about it for a while.
c. [ Bill said that [ [ Jane sang ]S and [ Fred danced. ]S ]S ]S
Comment: Jane sang and Fred danced are two simple sentences; Jane sang and Fred danced
is a more complex sentence that expresses what Bill said; and the whole thing is a sentence.
d. [ I persuaded Fred [ (Fred) to buy a telescope. ]S ]S

e. [ I promised Fred [ (me) to buy a telescope. ]S ]S
f. [[ (her) to appear on television ]S is her fondest dream. ]S
g. [ Joe said that [ he wants [ (Joe) to leave. ]S ]S ]S
h. [ That [ Jane can sing tenor ]S makes no difference. ]S
i. [ Bill left because [ he was tired. ]S ]S
j. the idea that [ truth is obtainable ]S
Comment: the whole thing is not a sentence; it’s a Noun Phrase; hence the double underlining.
When people speak, they use a mixture of sentences, noun phrases, interjections, and various
other linguistic forms.
———————————————————————————————————————
Study Exercise #2
This sentence is ambiguous:
My sister and I gave our parents books about each other.
Explain each possible meaning and illustrate it with a diagram (brackets and arrows) like the
ones given above.
My sister and I gave our parents books about each other.
One meaning: “My sister gave our parents books about me and I gave our parents books about
my sister.”
Other meaning: “My sister and I gave our mother a book about our father and gave our father
a book about our mother.”
Diagram:
[ My sister and I gave our parents books about each other. ]S

ok
ok
—————————————————————————————————————
Study Exercise #3
This sentence is ambiguous:
Bill and Fred persuaded Alice and Sue ( ) to buy telescopes in order ( ) to find out more
about each other.
For example, in one reading, you could continue: “In fact, as it turned out, Bill succeeded in
finding out more about Fred, but Fred did not succeed in finding out more about Bill.” In the other
reading, you could continue, “In fact, as it turned out, Alice succeeded in finding out more about
Sue, but Sue did not succeed in finding out more about Alice.”
For each meaning, fill in the implicit subjects shown with ( ). Then draw diagrams for the
reference of each other. (So you’ll end up with two diagrams.)
Answer to Study Exercise #3:
The impossible readings are the ones where each other refers to something outside the
smallest clause containing it.8
[Bill and Fred persuaded Alice and Sue [(Alice and Sue) to buy telescopes ]S in order [(Bill and Fred) to find out more about each other]S]S.
ok
impossible
impossible
[Bill and Fred persuaded Alice and Sue [(Alice and Sue) to buy telescopes in order [(Alice and Sue) to find out more about each other]S]S]S.
ok
impossible impossible
————————————————————————————————————
Study Exercise #4
For this sentence:
My parents tell my sister and me every day to write books about each other.
there’s only one meaning: “My parents tell my sister every day to write a book about me and tell
me every day to write a book about my sister.” It can’t mean “My mother tells my sister and me
every day to write a book about my father and my father tells my sister and me every day to write a
book about my mother.” Explain why, giving diagrams for both the possible and the impossible
meaning.
8
An additional detail is that the bracketing of the clauses is different in the two sentences. To find out more
about each other is a purposive clause, and in the first sentence it describes the purpose of persuading, in the second
sentence the purpose of buying. This difference does not affect the reference of each other, which is confined to its
own clause in any event.
The crucial part is to identify the clauses and the implicit subject, which must mean “my sister
and me” and not “my parents”. Once you’ve got this, then it follows straightforwardly from the
Each Other Reference Rule that each other can refer only to my sister and me.
[ My parents tell my sister and me every day [ (my sister and me) to write books about each other ]S]S
ok
impossible
Chaptersu2: Morphology
1. Orientation
In linguistics, “morphology” means “the study of word structure.” We’re interested in the
structure of individual words, as well as the grammatical rules with which words are formed.
First, we need terminology to be able to discuss the parts of words. The stem of a word is its
core, the part that bears its central meaning. In the English word undeniable, the stem is deny; and
in insincerity the stem is sincere.
Material that is added to the stem, thus modifying its meaning in some way, consists
principally of prefixes and suffixes. The suffix -able is suffixed to deny to form deniable; 9 and the
prefix un- is added to the result to obtain undeniable. Often, multiple prefixes and suffixes can be
added to the same stem, producing ever longer and more elaborate words: undeniability,
hyperundeniability.
Sometimes it is useful to have a term that covers both prefixes and suffixes. The standard
word for this is affix. More generally still, suppose we want a term that generalizes over stems,
prefixes, and suffixes — over all the building blocks from which words are assembled. The term
used for this is morpheme. It is often defined as follows:
(28) Defn.: morpheme
A morpheme is the smallest linguistic unit that bears a meaning.
In undeniable, un-, deny, and -able are the three morphemes. Deniable is not a morpheme because
it can be split into meaningful deny and -able. De and ny are not morphemes because they are
meaningless. More precisely, the sequence de often is a morpheme when it appears in other words,
for instance declassify, decompose, and delouse; but is it not a morpheme when it appears in deny.
To show how a word is divided into morphemes, one can separate the morphemes with
hyphens: un-deni-abil-ity. When discussed by themselves, prefixes and suffixes are indicated with
hyphens: prefixes as in un-, suffixes as in -ity. You can think of the hyphen as a bit of imaginary
“glue” with which a morpheme attaches to the stem.
2. Two kinds of morphology
Most linguists acknowledge at least a rough distinction between two kinds of morphology:
word formation vs. inflectional morphology. Word formation is what it says: the use of
prefixes, affixes, etc. to create new words; thus from the starting point good we can create the
word goodness by attaching the suffix -ness.
9
We’ll ignore the change of y to i, which follows a rule of English spelling.
Inflectional morphology is grammatical morphology, normally closely related to the syntax.

Here are some examples to start, from English:
• tense on verbs (present tense jumps, past tense jumped)

• number on nouns (singular cow, plural cows)
• a small amount of person and number agreement in verbs (She sings. vs. They sing.)
English is actually not a very good language for studying inflectional morphology, because it
doesn’t have very much of it (Mandarin is a similar case). But other languages, such as Swahili,
Russian, or Turkish, have a great deal, and students of these languages can spend years learning
the complete inflectional system.
3. Morphological analysis
One of the tasks of linguistic analysis that must be carried out to make sense of any newly-
encountered language is to figure out the structure of the morphology. This involves gathering
data, determining what morphemes are present, and writing the rules that form the words from
their constituent morphemes.
The fundamental method for this is as follows: one must compile a collection of
morphologically similar words and their meanings, then scrutinize it to determine which sound
sequences remain the same whenever the meaning remains the same. It is this criterion that will
isolate the meaningful chunks, i.e. morphemes.
We will do this now for a fairly simple case, namely a fragment of the nominal morphology
(=morphology for nouns) in Turkish. Here are the data:
(29) Turkish morphology data
1. el ‘hand’ ev ‘house’ zil ‘bell’

2. eli ‘hand (object)’ evi ‘house (object)’ zili ‘bell (object)’
3. ele ‘to (a) hand’ eve ‘to (a) house’ zile ‘to (a) bell’
4. elde ‘in (a) hand’ evde ‘in (a) house’ zilde ‘in (a) bell’
5. elim ‘my hand’ evim ‘my house’ zilim ‘my bell’
6. elimi ‘my hand (object)’ evimi ‘my house (object)’ zilimi ‘my bell (object)’
7. elime ‘to my hand’ evime ‘to my house’ zilime ‘to my bell’
8. elimde ‘in my hand’ evimde ‘in my house’ zilimde ‘in my bell’
9. elin ‘your hand’ evin ‘your house’ zilin ‘your bell’
10. elini ‘your hand (object)’ evini ‘your house (object)’ zilini ‘your bell (object)’
11. eline ‘to your hand’ evine ‘to your house’ ziline ‘to your bell’
12. elinde ‘in your hand’ evinde ‘in your house’ zilinde ‘in your bell’
13. elimiz ‘our hand’ evimiz ‘our house’ zilimiz ‘our bell’
14. elimizi ‘our hand (object)’ evimizi ‘our house (object)’ zilimizi ‘our bell (object)’
15. elimize ‘to our hand’ evimize ‘to our house’ zilimize ‘to our bell’
16. elimizde ‘in our hand’ evimizde ‘in our house’ zilimizde ‘in our bell’
17. eliniz ‘your (plur.) hand’ eviniz ‘your (plur.) house’ ziliniz ‘your (plur.) bell’
18. elinizi ‘your (pl.) hand (obj.)’ evinizi ‘your (pl.) house (obj.)’ zilinizi ‘your (pl.) bell (obj.)’
19. elinize ‘to your (pl.) hand’ evinize ‘to your (pl.) house’ zilinize ‘to your (pl.) bell’
20. elinizde ‘in your (pl.) hand’ evinizde ‘in your (pl.) house’ zilinizde ‘in your (pl.) bell’
21. eller ‘hands’ evler ‘houses’ ziller ‘bells’
22. elleri ‘hands (object)’ evleri ‘houses (object)’ zilleri ‘bells (object)’
23. ellere ‘to hands’ evlere ‘to houses’ zillere ‘to bells’
24. ellerde ‘in hands’ evlerde ‘in houses’ zillerde ‘in bells’
25. ellerim ‘my hands’ evlerim ‘my houses’ zillerim ‘my bells’
26. ellerimi ‘my hands (obj.)’ evlerimi ‘my houses (obj.)’ zillerimi ‘my bells (obj.)’
27. ellerime ‘to my hands’ evlerime ‘to my houses’ zillerime ‘to my bells’
28. ellerimde ‘in my hands’ evlerimde ‘in my houses’ zillerimde ‘in my bells’
29. ellerin ‘your hands’ evlerin ‘your houses’ zillerin ‘your bells’
30. ellerini ‘your hands (obj.)’ evlerini ‘your houses (obj.)’ zillerini ‘your bells (obj.)’
31. ellerine ‘to your hands’ evlerine ‘to your houses’ zillerine ‘to your bells’
32. ellerinde ‘in your hands’ evlerinde ‘in your houses’ zillerinde ‘in your bells’
33. ellerimiz ‘our hands’ evlerimiz ‘our houses’ zillerimiz ‘our bells’
34. ellerimizi ‘our hands (obj.)’ evlerimizi ‘our houses (obj.)’ zillerimizi ‘our bells (obj.)’
35. ellerimize ‘to our hands’ evlerimize ‘to our houses’ zillerimize ‘to our bells’
36. ellerimizde ‘in our hands’ evlerimizde ‘in our houses’ zillerimizde ‘in our bells’
37. elleriniz ‘your (pl.) hands’ evleriniz ‘your (pl.) houses’ zilleriniz ‘your (pl.) bells’
38. ellerinizi ‘your (pl.) hands (obj.)’ evlerinizi ‘your (pl.) houses (obj.)’ zillerinizi ‘your (pl.) bells (obj.)’
39. ellerinize ‘to your (pl.) hands’ evlerinize ‘to your (pl.) houses’ zillerinize ‘to your (pl.) bells’
40. ellerinizde ‘in your (pl.) hands’ evlerinizde ‘in your (pl.) houses’ zillerinizde ‘in your (pl.) bells’
We have here three columns, indicating inflected forms of the three nouns meaning “hand”,
“house”, and “bell”. Abbreviations and grammatical conventions are as follows:
• “plur.” or “pl.” abbreviate “plural”.

• “your (pl.)” is second person plural. Here, as a possessive, it means “belonging to you,
there being more than one of you.”
• (object) or (obj.) means that that form would be used as the object of a verb. Thus, if one
were to say in Turkish something like “I saw (the) hand”, one would use #3, eli,10 namely
eli gjørdym.
Each column of (29) constitutes a partial paradigm for the noun shown. A paradigm consists
of a single stem (above, the stems meaning ‘hand’, ‘house’ and ‘bell’), along with a group of its
grammatically different forms — often called its “inflected” forms. Usually, a paradigm is
displayed in a carefully thought-out order that helps reveal the structure at hand. A common sort of
problem in linguistics is to take a paradigm (either a problem set, or fieldwork data in real life),
and figure out the morphological system from it.
4. Breaking up the words into morphemes
To analyze a paradigm, we first search for invariant form paired with invariant meaning. In
the first column of (27), every single form begins with the sounds el and has a meaning involving
hands. It seems inconceivable that “hand” could be anything other than el, or that el could be
anything other than “hand”—note in particular the first line, where el means “hand” by itself.
4.1 Finding the stems
The columns for “house” and “bell” are completely identical to the column for “hand”, except
that where column has el, columns 2 and 3 have ev and zil as stems. It is plain that ev means
“house” and zil means “bell”.
Moving on, we can compare:
1. el ‘hand’
2. eli ‘hand (object)’
3. ele ‘to (a) hand’
Subtracting out el from the second and third forms, it appears that -i and -e must be suffixes. We
can confirm this by casting an eye over the remainder of the data: -e “goes together” with the
English word “to” given in the translations; and likewise -i with “(object)”.
4.2 Case
The -e and -i suffixes apparently denote the grammatical role that the noun plays in a Turkish
sentence, a phenomenon called case. Let’s briefly digress with the basics of case.
10
The reference source on Turkish I’m using gives another sentence with -i and a verbal object:
Beş adam heykel-i kırdi

five man statue-accusative broke
‘Five men broke the statue’
• Case is fundamentally an inflectional category of nouns (though often adjectives and

articles agree with their noun in case).
• Case tells us, intuitively, who is doing what to whom — it identifies the basic semantic
roles of the participants in a clause. In many languages, “Man bites dog” is Man-
nominative bites dog-accusative”, and “Dog bites man” is “Dog-nominative bites man-
accusative.” Nominative and accusative are probably the two most common cases.
• Case is not the only way to show who is doing what to whom. In languages with no case, or
ill-developed case systems (English), the work done by case is taken over by strict word
order and by prepositions.11
• Some typical cases in languages (each language is different in its cases and their usage):
➢ Nominative, usually for subjects of sentences or the citation form of a word

➢ Accusative, usually for objects of verbs
➢ Dative, conveying the notion of “to” in English: I gave the book to the student.
➢ Locative, conveying the notion of “at”, “in”, “on”, etc.
There are many other cases; Finnish is analyzed as having fifteen. This isn’t really that remarkable,
since many of these are simply that way of expressing notions that are expressed in other
languages by prepositions.12
In Turkish:
➢ -e is the suffix for the dative case

➢ -i is the suffix for the accusative case
➢ -de is the suffix for the locative case.
4.3 Plural
That covers the cases. Then, if we further inspect the data in rows 21-40 of (27), it is plain that
every plural noun has the suffix -ler.
4.4 Possessive suffixes
Lastly, there is a set of possessive suffixes, which express essentially the same information as
what in English is expressed by possessive pronouns like my and your. 13 There are four possessive
suffixes present in the data (Turkish has more, but these are not included here.)
11
Still other ways exist—in Tagalog, much of this information is given using prefixes or suffixes on
the verb.
12
Or their counterpart, postpositions, which follow their object noun phrase.
13
Their usage is not quite the same, because if there is a noun possessor, you use the suffix as well.
Thus, in English, we say (for example) Ayşe’s bell; but in Turkish Ayşe-nn zil-si, which is literally Ayşe’s
bell-her; similarly biz-im zil-imiz, literally “us’s bell-our”.
➢ -im ‘my’
➢ -in ‘your’
➢ -imiz ‘our’
➢ -iniz ‘your-plural’
We can classify the possessive suffixes on the dimensions of person and number. Number, in this
context, is simply the distinction between singular vs. plural. Person takes (as a first
approximation) three values:
• “First person” refers to pronouns and grammatical endings that involve the speaker, either
alone or with others. Thus in English I is a first-person singular pronoun, we is first person
plural.
• “Second person” refers to pronouns and grammatical endings that involve the hearer,
either alone or with others. In Spanish tú is a second-person singular pronoun, used to
address one person, and vosotros is a second-person plural pronoun, used to address more
than one person.14
• “Third person” refers to pronouns and grammatical endings that involve neither the
speaker nor the hearer. Thus he/she/it are third-person singular pronouns, they third person
plural.
4.5 Hyphens and glosses
Once we’ve found all the parts, we can restate the original data, putting in hyphens to separate
out the morphemes. I’ll do this just for the “hand” forms. I’ve also added a morpheme-by-
morpheme translation, also separated out by hyphens; this is called a gloss. Glosses are a sort of
micro-translation; they are meant to clarify structure, rather than give an idiomatic reading.
Word Gloss Idiomatic translation

1. el ‘hand ‘hand’
2. el-i ‘hand-acc. ‘hand (object)’
3. el-e ‘hand-dative ‘to (a) hand’
4. el-de ‘hand-locative’ ‘in (a) hand’
5. el-im ‘hand-1 sg. ‘my hand’
6. el-im-i ‘hand-1 sg.-acc.’ ‘my hand (object)’
7. el-im-e ‘hand-1 sg.-dat.’ ‘to my hand’
8. el-im-de ‘hand-1 sg.-loc.’ ‘in my hand’
9. el-in ‘hand-2 sg.’ ‘your hand’
10. el-in-i ‘hand-2 sg.-acc.’ ‘your hand (object)’
11. el-in-e ‘hand-2 sg.-dat.’ ‘to your hand’
12. el-in-de ‘hand-2 sg.-loc.’ ‘in your hand’
13. el-imiz ‘hand-1 plur.’ ‘our hand’
14
Standard English doesn’t make the distinction between singular and plural in the second person;
though many regional dialects have a special plural pronoun, “yall”, used whenever the addressee is plural.
14. el-imiz-i ‘hand-1 plur.-acc.’ ‘our hand (object)’

15. el-imiz-e ‘hand-1 plur.-dat.’ ‘to our hand
16. el-imiz-de ‘hand-1 plur.-loc.’ ‘in our hand’
17. el-iniz ‘hand-2 plur.’ ‘your (plur.) hand’
18. el-iniz-i ‘hand-2 plur.-acc.’ ‘your (pl.) hand (obj.)’
19. el-iniz-e ‘hand-2 plur.-dat.’ ‘to your (pl.) hand’
20. el-iniz-de ‘hand-2 plur. -loc.’ ‘in your (pl.) hand’
21. el-ler ‘hand-plural’ ‘hands’
22. el-ler-i ‘hand-plural-acc.’ ‘hands (object)’
23. el-ler-e ‘hand-plural-dat.’ ‘to hands’
24. el-ler-de ‘hand-plural-loc.’ ‘in hands’
25. el-ler-im-i ‘hand-plural-1 sg.-acc.’ ‘my hands (obj.)’
26. el-ler-im-e ‘hand-plural-1 sg.-dat.’ ‘to my hands’
27. el-ler-im ‘hand-plural-1 sg.’ ‘my hands’
28. el-ler-im-de ‘hand-plural-1 sg.-loc.’ ‘in my hands’
29. el-ler-in ‘hand-plural-2 sg.’ ‘your hands’
30. el-ler-in-i ‘hand-plural-2 sg.-acc.’ ‘your hands (obj.)’
31. el-ler-in-e ‘hand-plural-2 sg.-dat.’ ‘to your hands’
32. el-ler-in-de ‘hand-plural-2 sg.-loc.’ ‘in your hands’
33. el-ler-imiz ‘hand-plural-1 plur.’ ‘our hands’
34. el-ler-imiz-i ‘hand-plural-1 plur.-acc.’ ‘our hands (obj.)’
35. el-ler-imiz-e ‘hand-plural-1 plur.-dat.’ ‘to our hands’
36. el-ler-imiz-de ‘hand-plural-1 plur.-loc.’ ‘in our hands’
37. el-ler-iniz ‘hand-plural-2 plur.’ ‘your (pl.) hands’
38. el-ler-iniz-i ‘hand-plural-2 plur.-acc.’ ‘your (pl.) hands (obj.)’
39. el-ler-iniz-e ‘hand-plural-2 plur.-dat.’ ‘to your (pl.) hands’
40. el-ler-iniz-de ‘hand-plural-2 plur.-loc.’ ‘in your (pl.) hands’
5. Classifying the suffixes and discovering order
It is useful at this point to sort all the suffixes discovered according to their function:
Case endings
-i accusative
-e dative
-de locative
Possessive suffixes
-im ‘my’
-in ‘your’
-imiz ‘our’
-iniz ‘your-plural’
Plural
-ler
In particular, if you scan the data (now greatly clarified with hyphens and glosses), you can
find two important generalizations:
• No word contains more than one possessive suffix, or more than one case.
• Suffix order is invariant, and goes like this: Plural precedes Possessive Suffix precedes
Case.
With a word processor, it’s not hard to prove these relationships by lining up the relevant
morphemes into columns with tabs. Here the data once more, displayed in this way.
(30) Turkish data laid out in columns
Stem Plural Poss. Case

1. el ‘hand ‘hand’
2. el i ‘hand-acc. ‘hand (object)’
3. el e ‘hand-dative ‘to (a) hand’
4. el de ‘hand-locative’ ‘in (a) hand’
5. el im ‘hand-1 sg. ‘my hand’
6. el im i ‘hand-1 sg.-acc.’ ‘my hand (object)’
7. el im e ‘hand-1 sg.-dat.’ ‘to my hand’
8. el im de ‘hand-1 sg.-loc.’ ‘in my hand’
9. el in ‘hand-2 sg.’ ‘your hand’
10. el in i ‘hand-2 sg.-acc.’ ‘your hand (object)’
11. el in e ‘hand-2 sg.-dat.’ ‘to your hand’
12. el in de ‘hand-2 sg.-loc.’ ‘in your hand’
13. el imiz ‘hand-1 plur.’ ‘our hand’
14. el imiz i ‘hand-1 plur.-acc.’ ‘our hand (object)’
15. el imiz e ‘hand-1 plur.-dat.’ ‘to our hand
16. el imiz de ‘hand-1 plur.-loc.’ ‘in our hand’
17. el iniz ‘hand-2 plur.’ ‘your (plur.) hand’
18. el iniz i ‘hand-2 plur.-acc.’ ‘your (pl.) hand (obj.)’
19. el iniz e ‘hand-2 plur.-dat.’ ‘to your (pl.) hand’
20. el iniz de ‘hand-2 plur. -loc.’ ‘in your (pl.) hand’
21. el ler ‘hand-plural’ ‘hands’
22. el ler i ‘hand-plural-acc.’ ‘hands (object)’
23. el ler e ‘hand-plural-dat.’ ‘to hands’
24. el ler de ‘hand-plural-loc.’ ‘in hands’
25. el ler im ‘hand-plural-1 sg.’ ‘my hands’
26. el ler im i ‘hand-plural-1 sg.-acc.’ ‘my hands (obj.)’
27. el ler im e ‘hand-plural-1 sg.-dat.’ ‘to my hands’
28. el ler im de ‘hand-plural-1 sg.-loc.’ ‘in my hands’
29. el ler in ‘hand-plural-2 sg.’ ‘your hands’

30. el ler in i ‘hand-plural-2 sg.-acc.’ ‘your hands (obj.)’
31. el ler in e ‘hand-plural-2 sg.-dat.’ ‘to your hands’
32. el ler in de ‘hand-plural-2 sg.-loc.’ ‘in your hands’
33. el ler imiz ‘hand-plural-1 plur.’ ‘our hands’
34. el ler imiz i ‘hand-plural-1 plur.-acc.’ ‘our hands (obj.)’
35. el ler imiz e ‘hand-plural-1 plur.-dat.’ ‘to our hands’
36. el ler imiz de ‘hand-plural-1 plur.-loc.’ ‘in our hands’
37. el ler iniz ‘hand-plural-2 plur.’ ‘your (pl.) hands’
38. el ler iniz i ‘hand-plural-2 plur.-acc.’ ‘your (pl.) hands (obj.)’
39. el ler iniz e ‘hand-plural-2 plur.-dat.’ ‘to your (pl.) hands’
40. el ler iniz de ‘hand-plural-2 plur.-loc.’ ‘in your (pl.) hands’
Study Exercise #5
Reexamine these suffixes and propose a different — finer-grained — analysis.
-im ‘my’
-in ‘your’
-imiz ‘our’
-iniz ‘your-plural’
-imiz is really -im + -iz, and -iniz is really -in + -iz.
We can make this work if we give the suffixes slightly more abstract meanings: -im doesn’t
mean “my”, but more generally, “first person”. -in doesn’t mean “your”, but more generally,
“second person”. Then, -iz means “plural possessor”. Singular possessor is indicated by including
no suffix.
———————————————————————————————————————
6. Position classes in inflectional morphology
When we looked at the Turkish data, the primary finding was that the morphemes could be
arranged in a linear order, which could be expressed as five slots.
(31) Five slots in Turkish morphology
Stem Plural Possessor Possessor Case

Person Number
el ‘hand’ -ler -im 1st -iz plural  nominative

ev ‘house’ -in 2nd -i accusative
zil ‘bell’ -e dative
-de locative
For nominative case, I’ve used the symbol , meaning null. This means that nominative is a case
like any other, but happens not to have a suffix.
In a long word like ellerimizde ‘in our hands’, all five slots mentioned in (31) get filled:
(32) All five slots occupied

Stem Plural Possessor Possessor Case
Person Number
el -ler -im -iz de

hand plural 1st plur. poss. locative
In analysis, long words like ellerimizde are very useful, since they demonstrate the need for all five
slots at once.
The slots in a system like this are sometimes called position classes. Each position is an
abstract location in the word, which can be filled by a particular morpheme or set of morphemes.
In the analysis given earlier, we derived position classes using blocks of rules, one block per class.
An important check on a position class analysis is that there should be no contradictions of

ordering in the data, if the analysis is correct. We can look through the data and see that (for
example) -in, -iz, and -de never precede -ler; that -iz and -de never precede in; that -de never
precedes -iz; and similarly with the other morphemes.
Position classes can be defined simply by looking at the morphemes and checking their mutual
ordering. But most often, the classes can be related to morphological function. For example, it is
hardly an accident that the two suffixes in the third Turkish slot are both possessor person suffixes.
The normal pattern is: position reflects function.
This said, it should be noted that there are exceptions; the occasional language will take the
same function and put some of the morphemes into different positions; or fill a position with
morphemes of variegated function. For instance, the Swahili morpheme cho, which means roughly
“which”, gets put in a different position for positive and negative verbs:
(33) ki-tabu a-ki-taka-cho Hamisi

book SUBJ-OBJ-want-which Hamisi ‘the book which Hamisi wants’
ki-tabu a-si-cho-ki-taka Hamisi

book SUBJ-NEG-which-OBJ-want Hamisi ‘the book which Hamisi
doesn’t want’
7. Formalizing with a grammar
Linguists seek to make their analyses as explicit as possible, by expressing the pattern of the
language with rules. The rules taken together form a grammar. We’ll start with a very simple
grammar for Turkish nominal inflection.
The “architecture” of this grammar is conceptually about as simple as it could be: we’ll start
with a representation of (roughly) meaning, and set up rules that input this meaning and output
sound.
In particular, let’s assume that the stem (el, ev, zil, or whatever) comes with morphological
features specifying its grammatical content. The bundle of features is called the morphosyntactic
representation.15 The job of our grammar will be to manifest this content with actual material. For
example, we can start out with something like this for #40 from (28) above (on p. 26):
el:[Number:plural, PossessorPerson:2, PossessorNumber:plural, Case:Locative]
The el part is the stem meaning ‘hand’. The part in [ ] is the morphosyntactic representation. It
contains four morphological features:
(34) Number
PossessorPerson
PossessorNumber
Case
15
Why? We’ll see later on: the morphosyntactic representation transfers information over from the
syntax to the morphology.
Each feature has a value, which is shown by placing it after a colon. So you can read the formula
el:[Number:plural, PossessorNumber:plural, PossessorPerson:2, Case:Locative] as: “the stem el,
with a morphosyntactic representation indicating plural Number, plural PossessorNumber, second
PossessorPerson, and Locative Case”. We’ll return later on to the question of where these features
come from.
The grammar itself consists of four rules. The order in which the rules are stated is significant
and is part of the grammar. Only the first rule is stated in full.
(35) A Grammar for Inflectional Morphology in Turkish Nouns
a. Number Rule
Suffix /-ler/ if the morphosyntactic representation bears the feature [Number:plural]
b. Possessor Person Rule
Add a possessor suffix, as follows:
-im if [PossessorPerson:1person]
-in if [PossessorPerson:2person]
c. Possessor Number Rule
Add a possessor suffix, as follows:
-iz if [PossessorNumber:plural]
d. Case Rule
Add a case suffix, as follows:
-i if [Case:Accusative]
-e if [Case:Dative]
-de if [Case:Locative]
The reason that the rules must apply in the order given is that by doing this, we construct the
word from “inside out”, adding a bit more to the material we’ve already accumulated. This “inside
out” character will be shown immediately below.
You can show how the rules apply to a particular form by giving a derivation. In linguistics, a
derivation shows each rule applying in succession, and justifies the rules by showing that they
correctly derive the observed forms. You’ve probably seen something like a derivation before, as
similar devices are used for proofs in math or illustrate an inference in logic — each line is
justified by the rule that is applied.
For the Turkish form ellerinizde ‘in your (plur.) hands’ (#40 in the data of (27) above), the
derivation would look like this (I use underlines, etc. simply to increase legibility).
(36) Morphological derivation for ellerinizde
el:[Number:plural, PossPers:2, PossNum:plural, Case:Locative] ‘hand’ with its

morphosyntactic representation
eller Number Rule
ellerin Possessor Person Rule
elleriniz Possessor Number Rule
ellerinizde Case Rule
At each stage, the relevant rule “sees” the right feature, and adds the appropriate suffix.
The derivation just given is a bit extravagant, in that I repeated the morphosyntactic
representation at each phase, to show what was being referred to. A shorter answer, which I think
would be fine to submit in a homework, would be like this:
(37) A terser morphological derivation for ellerinizde
el:[Number:plural, PossPers:2, PossNum:plural, Case:Locative] ‘hand’ with its

eller Number Rule
ellerin Possessor Person Rule
elleriniz Possessor Number Rule
ellerinizde Case Rule
Study Exercise #6
Derive #34, ellerimizi ‘our hands-accusative,’ starting with an appropriate morphosyntactic

representation and using the rules above. Answer on next page.
el:[Number:plural, PossPers:1, PossNum:plural, Case:Accusative] ‘hand’ with its

eller Number Rule
ellerim Possessor Person Rule
ellerimiz Possessor Number Rule
ellerimizi Case Rule
_______________________________________________________________________
8. A note on grammars and applying them
The rules of (33) above are the very first grammar we have discussed; there will be quite a
few more grammars as we proceed.
Grammars, written with formalized rules, are a central element of linguistics; they are the
method that linguists use to make explicit hypotheses that can be tested and improved. The
formalization of grammars has a purpose: we want the grammars to be unambiguous, so we can
always agree on how the rules apply.
The role of formalized rules in linguistics is not that different from the role of explicit theory
in any other science. The theory is supposed to be applicable, capable of making predictions about
new data. When the predictions are right, we feel good about it; if the predictions are wrong, we go
back to the drawing board, either modifying or replacing the theory.
Students new to formal grammars may need to master a skill that I will facetiously call
“turning your brain into a computer”: you do nothing but look at the rule and the forms it applies
to, and derive the result, as if you were a machine. For a moment, don’t think about what the
grammar ought to be deriving; just follow the rules and see what it does derive.16
The payoff for such behavior is that the grammars become tools in our hands; tools for
understanding in explicit terms how a language is working. We treat our tools with respect when
we take them for what they are, namely, utterly mechanical principles.
I hasten to add: the need to turn your brain into a computer is temporary. All you want to do,
at this moment, is check the outcome and see if it matches the true data of the language. But the
other tasks of a linguist are actually quite creative: they include (a) thinking of better rules when
the old rules fail; (b) looking at language data to detect the patterns that the rules should be
capturing; (c) thinking of new ways (data to elicit in fieldwork, designing experiments) to get the
data that will reveal interesting truths about a language. Linguistics is indeed a field that welcomes
creativity; the “turning your brain into a computer” bit I’ve described here occurs only at the stage
of checking how the rules apply.
16
This is actually not such a bad skill to cultivate; there aren’t all that many people in the world who
are capable of following complex instructions to the letter; you can be one of them!
9. The bigger picture
Grammars like the one we are working on can derive quite a bit of data. It’s worth pondering,
for instance, how many forms a Turkish noun can have. There are several choices to be made:
• Number: singular or plural, thus two possibilities

• Possessor Person: any of three: 1, 2, 3 (“his or her”)
• Possessor Number: any of two (singular, plural)
• Case: nominative (no ending), accusative, dative, locative, plus ablative (“from”), genitive
“’s”, instrumental (“with”), thus seven possibilities17
Multiplying these out, every Turkish noun can appear in (at least) 2 x 3 x 2 x 7 = 84 forms, of
which we covered only 40. It seems likely that Turkish speakers often must produce a new form
for a noun, when they haven’t heard a particular combination before.
The Turkish nominal system is a fairly simple one; Turkish verbs, for instance, are quite a bit
more complex; one estimate for the total size of the Turkish verbal paradigm is 40,000. The most
elaborate system I know of is the verbal system of Shona (Bantu, Zimbabwe), where (according to
the linguist David Odden), the typical verb has about 10 trillion possible forms. Odden has
developed a system that generates these forms using a rather complicated set of rules; most of the
complications arise in getting the tones right.
It seems also likely that Turkish children or Shona children must also be using some kind of a
grammar; a grammar they internalize somehow in their heads. This is so because they could not
possibly memorize every form of every word. We cannot know — yet — to what extent their
grammars resemble our grammars, but the idea that through analysis and research we can get close
to what they learn is a central idea of contemporary linguistics.
10. The source of morphosyntactic representations
The discussion in the last chapter showed how we can write a set of rules that create
morphologically well-formed words through the successive addition of prefixes and suffixes by
rule. But what do these rules apply to? There are various answers given by various linguists; here,
we will examine just one fairly representative one.
The idea is that the syntax of a language builds up a feature structure for every stem that
appears in a sentence. Here are the data we address:
I jump. We jump.
You jump.
He/she/it/Fred jumps. They jump.
17
You can learn about all these cases, and much more, in a good grammar book. For Turkish, I have consulted
Geoffrey Lewis (2000) Turkish Grammar (Oxford: Oxford University Press), which I find exceptionally clear.
The suffix -s is seen, and only seen, when the subject is one of the pronouns he/she/it, or a
singular noun phrase. This indicates the fact that in English the verb “agrees with” its subject. A
simple way to do this syntactically is to set up a rule that copies the morphosyntactic features of
the subject onto the verb. Specifically, in a sentence like Fred jumps this rule must cause the
features [Person:3, Number:Singular] to appear on the stem jump. Looking ahead to syntax, we can
draw a syntactic structure18 and the process of agreement:
(38) S S
NP VP NP VP
| | | |
N V → N V
| | |
Fred Fred jump
 Number:sg jump Number:sg Tense:Pres 
Person:3  [Tense:Pres] Person:3  Number:sg
Person:3 
feature copying
We can assume that Fred is inherently [Number:sg, Person:3], since it is a proper name. The
[Tense:Pres] must be assumed at the start as well, since it is part of the meaning of the sentence.
The operation above is part of syntax. Once the rules of the morphology get to apply, the
presence of these feature will cause a suffixation rule to apply, which attaches the suffix that we
spell -s. Here is a sample rule:
(39) 3rd Sing. Present Rule
Suffix -s when the morphosyntactic representation contains

[Tense:Pres,Person:Sing.,Number:3].
In sum, we have quite a bit of descriptive work to do in a complete grammar: the syntactic
component arranges words in correct order and builds up the morphosyntactic representations,
while the morphological component refers to the morphosyntactic representation in order to add
the appropriate affixes.19
This is looking ahead, so don’t be alarmed if the diagrams aren’t clear. To clue you in a bit: S =
18
Sentence, NP = Noun Phrase, VP = Verb Phrase, N = Noun, V = Verb, vertical line means “is part of”.
19
The definition of the term “affix” was given on p. 19.
11. More than one feature per morpheme: an example from German
Consider the person-number endings of German, in the present and past. In the data below I
give these forms accompanied by a subject pronoun in parentheses, intended to make the meaning
clearer.20
(40) A German verb paradigm
Present Past
1 sg. (ich) warte ‘I wait’ (ich) wartete ‘I waited’
2 sg. (du) wartest ‘you-sg. wait’ (du) wartetest ‘You waited’
3 sg. (sie/er) wartet ‘she/he waits’ (sie/er) wartete ‘she/he waited’
1 pl. (wir) warten ‘we wait’ (wir) warteten ‘we waited’

2 pl. (ihr) wartet ‘you-plur. wait’ (ihr) wartetet ‘you-plur. waited’
3 pl. (sie) warten ‘they wait’ (sie) warteten ‘they waited’
Looking at the paradigm of (37), we find a tricky issue: is the stem warte, with endings
like -(zero), -st, -t, -n, -t, -n; or is it wart, with endings like -e, -est, -et, -en, -et, -en? To decide this
, we can appeal to a slightly bigger paradigm that includes the imperative: wart by itself means,
‘wait!’, as a command. This suggests that wart really is the stem.
Accepting this view, we can now give the forms broken up into position classes (shown with
vertical alignment):
(41) The same paradigm arranged as position classes

Present Past
1 sg. (ich) wart e (ich) wart et e
2 sg. (du) wart est (du) wart et est
3 sg. (sie/er) wart et (sie/er) wart et e
1 pl. (wir) wart en (wir) wart et en

2 pl. (ihr) wart et (ihr) wart et et
3 pl. (sie) wart en (sie) wart et en
An important first thing to notice here is that unlike in Turkish, we are not going to be able to
put forth an analysis in which the inflectional rules mention just one feature each—that is, with
endings for person, endings for number, and endings for tense. Rather, German “bundles” the
features, in the sense that one single suffix manifests more than one feature at a time. Thus, for
instance, the suffix -est is simultaneously the realization of second person and singular number. As
A caution here: the discussion here holds true only for stems ending in [t]; I’m doing this to keep
20
the morpheme divisions simple. The problem is that to get German verb morphology really right you have to
do some phonology, which won’t happen for quite a few chapters yet…
a result, in the analysis below, I have mostly written rules that mention more than one feature at a
time. For the six person/number combinations, one needs (at least six rules). Here is a grammar:
(42) Rules for verbal inflection in German
I. Tense Marking
Suffix -et when the morphophosyntactic representations contains [Tense:Past]
II. Person/Number Marking
Suffix:
-e if [Person:1, Number:Singular]
-est if [Person:2, Number:Singular]
-e if [Tense:Past, Person:3, Number:Singular]
-et if [Tense:Present, Person:3, Number:Singular]
-en if [Person:1, Number:Plural]

-et if [Person:2, Number:Plural]
-en if [Person:3, Number:Plural]
In fact, things are even more complicated than this. In precisely one place in the system—the
3rd person singular—the person-number ending is different in the past than in the present. The
analysis takes account of this with the rules in boldface, which mention three features at once.
Systems of inflectional morphology are well known for including asymmetries of this kind; and
the normal analytic approach in this kind of analysis is simply to include enough features in the
rule to get the right result.
English has a very similar instance: the -s of jumps, seen earlier, simultaneously manifests
[Number:Singular, Person:3, Tense:Present]. In fact, such “tangling” is found in languages all over
the world.
Subparadigms often involve partial overlap: thus, the German present and past verb
paradigms overlap in all but the third singular. From the viewpoint of rules, this is because it is
only in the third singular that the rules are sensitive to tense. The overlapping cases will derive
from the more general rules for the rest of the paradigm, which don’t mention tense.
12. Irregular forms and spell-out
Morphology is notorious for the amount of irregularity it involves. All the examples seen
above were selected from instances where the rules work more or less perfectly. But very often,
the rules have exceptions.
To give a very simple example, English verbs most often realize the feature [Tense:past] by
suffixing a morpheme that in spelling appears as -ed, as in kiss/kissed, jump/jumped, vote/voted.21
We might write a rule like this:
(43) Past Tense Rule
Suffix -ed when the morphosyntactic representation contains [Tense:Past].
Yet English has about 180 verbs that don’t work this way: here are some sample pairs of
present tense verbs with their irregular pasts:
sing sang (not *singed)

cling clung (not *clinged)
write wrote (not *writed)
keep kept (not *keeped)
have had (not *haved)
make made (not *maked)
go went (not *goed)
You can see that sometimes the vowel changes, or consonants drop out (had, made), or a
completely unrelated form is used, as in went. (Such cases of total replacement are described in
linguistics with the term suppletion; went is said to be the “suppletive past” of go.)
A common way to deal with irregulars is to adopt word-specific rules of spell-out. These
don’t try to be general or insightful, but merely state the facts. Here are spell-out rules for went and
kept:
(44) Went spell-out

go [Tense:past] is spelled out as went.
Kept spell-out
keep [Tense:past] is spelled out as kept.
A spell-out rule is assumed to preempt any regular rule that mentions the same features; we need to
do this to make sure that we don’t apply the Past Tense Rule to went, yielding *wented.
We will have further uses for spell-out rules as we proceed.
12.1 The origin of irregular inflected forms
Irregular inflected forms are widespread in the world’s languages. They have three sources.
21
By invoking the spelling, I can ignore some differences in pronounciation that arise from phonology: -ed is
[t] in kissed, [d] in rubbed, and [əd] in voted. We will pay not attention to such differences until we cover phonology in
Chapter 12.
(1) Often, irregulars are relics of changes in pronunciation that took place in the language
centuries (or even millenia) before. For instance, around the year 1000, the Middle English version
of keep was cēpan, where ē spelled a long vowel. At this time, long vowels came to be shortened
before two consonants, as in cepte, formerly cēpte. Our own keep and kept are phonetically
evolved versions of cēpan and cepte. But the shortening principle is no longer active in the
language. So, for example, the past tense of beep has to be beeped, not *bept.
(2) A second source of irregular forms is very different. Written languages often include
technical or scholarly vocabulary, and it is quite common for this vocabulary to be borrowed from
a prestigious foreign language. Korean and Japanese borrow from Chinese, Persian from Arabic,
English from Latin and Greek. So stimulus, irregular plural stimuli, is used in scholarly articles and
in lecturing by psychologists. This word comes from Latin, and stimuli is simply the Latin plural
carried over into English. Persian likewise uses a whole group of complex plurals borrowed from
Arabic; so the word ketaːb ‘book’22 can pluralize either with its Arabic-derived plural kotob or
with the regular Persian plural suffix -haː, as in ketab-haː.
(3) Suppletion occurs when an inflected form of Word A somehow comes to be reinterpreted
as an inflected form of Word B. Thus, went was once the past tense of wend, but somehow was
reinterpreted to be the past tense of go. English be, am, and was were, thousands of years ago,
three separate verbs, but now constitute different spell-outs of the same verb.23
13. What are the characteristic inflectional categories?
Every language has a set of inflectional categories, though the sheer amount of inflection can
vary quite a bit. Mandarin Chinese has very little; Turkish and Finnish are quite richly inflected;
English is closer to the Mandarin end of the scale.
Each inflectional category is expressed (in the theory we are using) as a feature within the
morphosyntactic representations.
Here is a quick survey of some inflectional categories.
13.1 Nominal Inflection
Nouns and pronouns are often inflected for number (singular, plural, and occasionally dual,
meaning exactly two; or even trial, exactly three). Pronouns are in addition inflected for person
(first = includes speaker; second = includes hearer; third = neither).
22
[ː] means the preceding vowel is long.
23
All information in this paragraph obtained from what is probably the most authoritative source, the Oxford
English Dictionary, www.oed.com.
13.1.1 Gender
In a number of languages nouns are inflected for gender; for instance, in German nouns can
be masculine, feminine or neuter (as we can tell by the definite articles they take). In some cases,
gender is semantically quite sensible:
(45) der Mann ‘the-masculine man’

die Frau ‘the-feminine woman’
Extraordinarily, this system carries over—often quite arbitrarily — to the whole vocabulary of
nouns, irrespective of meaning. Thus each of the three common items of silverware is a different
gender in German:
(46) der Löffel ‘the-masculine spoon

die Gabel ‘the-feminine fork’
das Messer ‘the-neuter knife’
Thus gender is for the most part a purely formal device, not an expression of meaning.
Gender involves many other semantic correlations that have nothing to do with biological sex.
From a web page intended to help learners of German24 I quote the following rules:
(47) 60. Fabrics are predominantly masculine (der Gingham, der Kaschmir).
61. Heavenly bodies are predominantly masculine (der Mond [moon], der Stern [star]).
62. Forms of precipitation are predominantly masculine (der Regen [rain], der Schnee
[snow]).
63. Bodies of water (restricted to inland streams, currents, and stagnant bodies) are
predominantly masculine (der See [sea], der Teich [pond]).
64. Words denoting sound or loud noise or phonetic speech sounds are masculine (der
Donner [thunder], der Dental [dental sound], der Diphthong).
65. Dance steps and popular music forms are masculine (der Jazz, der Tango).
Such generalizations are pervasive in gender languages. However, since there are usually
exceptions of various sorts, it seems that people who know gender languages have probably
memorized the gender of every word.
Gender is not just a property of familiar European languages; for example, it is also found in
Semitic languages, and a kind of system rather like gender (but with at least a dozen types) is
found in Bantu languages.
13.1.2 Case
Nouns, and the syntactic phrases they occur in, are marked for case, which marks their role in
the sentence. See p. 22 above for discussion of case.
24
https://sites.google.com/site/meyersde/Home/determinants-of-gender-in-german; sadly, no longer
available.
13.2 Verbal Inflection
Very common is tense, which gives the time of action relative to the present: past (I jumped),
present (I jump), future (I will jump), and other (for example, “remote past”) tenses.
Aspect sets the boundaries of the time of the action of the verb, for instance, completed vs.
non-completed action.
Verbs often agree with their subjects (and sometimes their objects as well) in features for
nouns (as shown above in section 10 of this chapter). These features include person (I am, you are,
she is), number (I am, we are), gender.
Verbs, particularly second person forms (see below) can also be inflected for the degree of
familiarity of the addressee; thus English used to make a distinction between (say) thou believest,
addressed to intimates, children, and animals; and you believe, for less familiar addressees. Most
European languages, Javanese, Persian, Japanese, and Korean have such systems today.
In various languages verbs are inflected for degree of belief. German, for instance, has an
indicative (for full endorsement), a weak subjunctive (for weak endorsement), and a strong
subjective (full skepticism):
(48) Indicative and two subjunctives in German
Er sagte, dass er krank ist.

He said that he sick is-indicative
“He said he is sick” (acknowledging a belief held by all)
Er sagte, dass er krank sei.

He said that he sick is-weak subjunctive
“He said he is sick” (and it’s not necessarily true)
Er sagte, dass er krank wäre.

He said that he sick is-strong subjunctive
“He said he is sick” (and the speaker doubts it)
Related to this is the category of verbal inflection in many languages which marks information
known only by hearsay rather than by direct witness; this is common in American Indian
languages. Turkish has both a “simple past” and an “inferential past”, illustrated in the pair below
(The letter ş is used in Turkish to spell the sound written in English as sh, phonetic symbol [ʃ]).
(49) Verbal inflection for direct evidence vs. hearsay in Turkish
bir turist vapuru gel-di

a tourist ship come-past
‘A tourist ship has come’ (witnessed by the speaker)
bir turist vapuru gel-miş

a tourist ship come-inferential past
‘I gather that a tourist ship has come’ (learned from other people)
13.3 Adjectival Inflection
Adjectives typically don’t have their own inflectional categories, but acquire inflection by
agreeing with the nouns they modify; thus German:
(50) ein guter Löffel ‘a-masculine good-masculine spoon’

eine gute Gabel ‘a-feminine good-feminine fork’
ein gutes Messer ‘a-neuter good-neuter knife’
14. The principle of obligatory expression
An important aspect of inflectional morphology is that it often involves obligatory choices.

When in English one says “I bought the book”, it specifically means “one book”, not “any old
number of books”. Likewise, “books” necessarily implies the plural. To avoid the obligatory
choice, one must resort to awkward circumlocutions like “book or books”. There are other
languages (for example, Mandarin) that work quite differently. Thus, the following sentence:
(51) wɔ3 mai3 ʂu1

I buy book
我买书
is quite noncommittal about how many books are bought.25 Thus an important aspect of the
grammar of languages is the set of choices they force speakers to make when speaking; this is
determined by their systems of inflectional morphology.
Fundamentally, there is a bifurcation between the two ways that thought is embodied in
language. The following diagram tries to make this clearer.
25
I have queried many Mandarin speakers about whether the sentence is also noncommittal about time (verb
tense), and get differing answers. There is some subtlety here I am missing …
(52) Thought
concepts, classifications, ideas…
Grammar Content
[Number: singular, plural] one vs. more than one: “two”, “three”
[Tense: present, past] overt statements about time: “now”, “then”
[Honorific: formal, informal] overt labels of respect: “Mr.”
[Mood: indicative, subjunctive] overt statements of degree of belief: “I doubt
etc. …”
Languages differ: each one takes a subset of the fundamental ideas, and grammatically
codifies them. By this I mean that in some particular language, a particular concept gets expressed
as grammatical features, and that these features are included in the morphosyntactic
representations and thus integrated into the grammar. Whenever this happens, the expression of the
concept in question becomes obligatory — since you have to obey the grammar of your language
when you speak. In some other language, the same concept could remain uncodified
grammatically, and the speaker is free to express it or not as she chooses, through choice of words
and other means.
On the whole, the forms of thought that can get integrated into grammar are, as we might
expect, the ones that are omnipresent in our lives: time, number, belief vs. doubt, and the
fundamental aspects of conversations (speaker/hearer/other and their social relations.)
15. The typology of inflection
As we saw earlier in section 2.11 above, it’s possible for there to be inflectional rules that
apply (attaching their affix) only if two features are present in the morphosyntactic representation.
This is true, even more so, of the following Latin data:
(53) A Latin noun paradigm
Case Singular Plural Meaning of case

Nominative somn-us somn-i for subjects
Genitive somn-i somn-orum for possessors
Dative somn-o somn-is “to”
Accusative somn-um somn-os for objects
Ablative somn-o somn-is “from”
Locative somn-i somn-is “at, on, in”
Vocative somn-e somn-i for calling to someone
Taking just the suffix -us, we can see that it packs a considerable bundle of information: it
tells us that somnus is nominative, that it is singular, and (with a few exceptions we will ignore)
that it is masculine. We could write the rule like this:
(54) A rule for the Latin inflectional suffix -us
Add -us if [Case:Nominative, Number:singular, Gender:masculine]
Indeed, most of the Latin nominal inflectional suffixes require three features for their proper
attachment.
There is some very traditional classificatory terms for the morphological type of a language,
which, in the context of the theory we are using, can be expressed in terms of features.
(55) Isolating / agglutinative / inflecting
• A language is isolating to the extent that it has little or no inflectional morphology.

Examples: English, Chinese.
• A language is agglutinating if it has a rich inflectional morphology, and each morpheme

expresses a single morphosyntactic feature. Thus, words have a very transparent structure.
Examples: Turkish, Swahili.
• A language is inflecting (bad term, since it’s ambiguous) if it has a rich morphology, and
morphemes typically express multiple features. Example: Latin.
Of course, these terms are just matters of degree; Turkish is famous for being really quite
agglutinating, and Latin is famous for being highly inflecting (in the relevant sense). There are
many less-clear cases.
All else being equal, inflecting languages will tend to have shorter words than agglutinating
languages, since each affix is “doing more work”. However, there is usually a cost to this
terseness: typically, in an inflecting language the same ending often serves multiple purposes, so
words tend to be inflectionally ambiguous.
MORPHOLOGY, PART II: WORD FORMATION

16. Rules of word formation
The discussion in this chapter so far is of inflection; the morphology that is related to
grammar. The other side of morphology is the system of rules used to to expand the stock of words
by forming new words from old. Often linguists refer to this process as derivational morphology
or derivation; I use the term word formation since it is more precise.
For example, given that identify is an existing word of English, a rule of English word
formation can create a new word, identifiable. From this another rule can provide identifiability,
and from this yet another rule can create unidentifiability.
16.1 Rules of word formation
Consider some words formed with the English suffix -able:
(56) -able: washable, lovable, thinkable, growable, doable
We wish to write the word formation rule that attaches -able to an existing word to form a
new one. There are three kinds of information that must be included in the rule.
First, there is a change of form; the existing word is augmented by the suffix. This could be
expressed with the formalism below.
(57) X → X + -able
Capital X is meant to be a variable, designating “any string” (of letters, or later on, sounds). So the
rule says, “augment any string with the string -able”.
Second, there is a change of meaning: Xable means “able to be Xed”. We will not formalize
this, since the task of representing meaning is far too big to take on in this context. Finally, there is
often a change in part of speech. -able attaches to Verbs (e.g. wash, love, think, etc.) and forms
Adjectives. We can do this by adding in appropriate brackets to the primitive version of the rule
seen in (51). Let’s put this all together:
(58) -able Rule
[ X ]Verb → [ [ X ]Verb -able ]Adjective Meaning: “able to be V’ed”
You can read this, approximately, as follows: “-able may be attached to verbs to form adjectives
with the meaning, “able to be Verb’ed”.
“”
Rules of word formation can be shown applying in a derivation. As before, we label each line of
the derivation according to the rule that applies. Thus, for instance, here is a derivation for
washable:
(59) [ wash ]Verb Stem

[ [ wash ] Verb able ]Adjective -able Rule
If you want to figure out how to express a word formation rule of English, the first step is just
to find the right data: a set of representative words that have the relevant prefixes and suffixes (or
more generally, that have the same word formation process).
Here are some further word formation rules of English. To express the derivation of words in -
ity, (for example, divinity, obscurity, obesity, insanity, sensitivity), we could write the rule
(60) -ity Rule
[ X ]Adjective → [ [ X ]Adjective ity ]Noun Meaning: “quality of being Adjective”
This creates structures like [[ obes ]Adjective ity ]Noun, which we can abbreviate as [[ obes ]A ity ]N. 26
To handle words formed with the prefix un-, (unfair, unkind, unjust, unspoken, unattested,
unidentifiable) we could write the following rule:
(61) un- Rule
[ X ]Adjective → [ un [ X ]Adjective ]Adjective Meaning: “not Adjective”
The rule creates structures like [ un [ kind ]A]A.
To solve problems involving writing of rules of word formation in English, it’s clear that the
first task is to think of a set of words that have the relevant prefix or suffix, then generalize over
what you find. You can get help with prefixes just by consulting a dictionary, where words with
the same prefix alphabetize together. To find words with the same suffix, there are dictionaries that
alphabetize from the end of the word rather than the beginning. In either cases, it is sometime
possible to use software. 27
17. Productivity
The -able Rule, (58), is often described as productive, defined as follows:
(62) Defn.: productivity
A rule of word formation is productive to the extent that it can be used by speakers to create
novel words.
26
English spelling generally drops the letter e before suffixes that begin with a vowel; let us ignore this fact for
purposes of the rule.
The two I know are entitled Walker’s Rhyming Dictionary and The English Word Speculum. There is also a
27
great deal of software that can do this; for example

http://www.linguistics.ucla.edu/people/hayes/EnglishPhonologySearch
The evidence that -able is a productive suffix is that it is fairly easy to suffix it to a verb and
thereby create a completely novel adjective. You may never have heard the -able adjectives in
these sentences, but I believe they are intelligible and sound not all that awkward as English
words.
(63) Some examples suggesting that -able is productive

This concept is diagramable in many ways.
These extremely delicate vegetables are not especially sautéable.
The criminal had disappeared and was not apprehendable.
This is the sort of evidence that can be used to justify calling a rule productive. One can take a step
further and check if native speakers will even accept attaching the suffix to a completely made-up
stem, and I believe that -able passes this test. For instance, if we agree to pretend for purposes of
elicitation that wug is a verb of English, then it seems not too hard to imagine that wuggable could
be formed from it. (I leave it to you to make up a meaning…).
In contrast, the -ity Rule (60) is not especially productive. If you try to use it to make new
words, they will usually sound very odd (I marked them below with asterisks, as we did for the bad
sentences of the previous chapter).
(64) Some examples suggesting that -ity is not productive

The *abstractity of this theory was a barrier to our understanding it.
The *maroonity of his sweater paradoxically made it stand out, since the rest of us were
wearing bright colors.
The *steepity of the roof made the re-roofing job perilous.
It still makes sense to include an -ity Rule as part of our description of English, since English
speakers clearly recognize divinity, obscurity, obesity, insanity, sensitivity, and so on as being
derived from divine, obscure, obese, insane, and sensitive, respectively. But a proper description of
English would acknowledge that the rule is not productive.
The outputs of an unproductive word formation rule will exist in a language only if the words
it derives are accepted, individually, as words by the speech community. An English-learning child
almost certainly learns a word like divinity as a individual, memorized form, but also is
(eventually) able to perceive that it is derived from divine.28
Grammar books for particular languages often include a chapter on word formation, and
sometimes attempt to differentiate the word formation processes of a language according to their
productivity.
28
One of the unsolved questions of theoretical linguistics is how language learners come to know which rules
of word formation are productive and which aren’t; current theories of this process suggest that word frequency is
involved in some way.
18. Stacked derivation
At least in English, the idea of the position class, covered above for inflection, is not relevant
for derivation. Rather, the rules of derivation can apply freely, provided their requirements are met.
For example, we can derive the long word unmindfulness by applying the following rules in
succession:
(65) Derivation of unmindfulness
[mind]Noun stem
[[mind]Noun ful]A -ful Rule: [ X ]N → [[ X ]N ful ]Adj
[un[[mind]Noun ful]Adj] un- Rule ((61))
[[un[[mind]Noun ful]Adj]ness]Noun -ness Rule: [ X ]Adj → [[ X ]Adj ness ]Noun
With a bit of strain, it’s even possible to have the same inflectional rule apply twice in the
same form. Here is an outline derivation for the (novel) word industrializational.
industry
industrial
industrialize
industrialization
industrializational
Although the last word is a bit of stretch, you can see that the result has “double application” of the
rule that attaches -al.
Study Exercise #7: First find batches of words that include the suffixes -al, -ize, and -
ation. Then give the rules and derivation for industrializational. Answer on next page.
Words including the affixes (these are used to justify the rules):
-al: cynical, coastal, epochal, triumphal

-ize: humanize, criminalize, socialize, legalize
-ation: visitation, accreditation, limitation, condensation
[industry]Noun stem
[[industri]Noun al]A29 -al Rule:

[ X ]N → [[ X ]N al ]Adj
“[[ X ]N al ]Adj” means “pertaining to X”
[[[industri]Noun al]A ize ]V -ize Rule:

[ X ]Adj → [[ X ]Adj ize ]Verb
“[[ X ]Adj ize ]Verb” means “to render X”
[[[[industri]Noun al]A iz ]V ation ]N -ation Rule:

[ X ]Verb → [[ X ]Verb ation ]Noun
“[[ X ]Verb ation ]Noun” means “the process of Xing”
[[[[[industri]Noun al]A iz ]V ation ]N al]A -al Rule (again)
Here are some words that justify these rules:
-al: cynical, coastal, epochal, triumphal

-ize: humanize, criminalize, socialize, legalize
-ation: visitation, accreditation, limitation, condensation
—————————————————————————————————————
The repetition of the same suffix in the word is fairly good evidence that English word
formation does not involve position classes. The multiple appearances result from the inherent
property of word formation, that the rules apply where they can. In contrast, in the position-class
systems seen in inflection, the rules apply in a strict order.
18.1 Ambiguity from stacked derivation
A theme we will repeatedly return to in this book is ambiguity and the ways it emerges from
the rules of a language. We begin with an example from word formation.
To start, let’s set up a bit of the English morphological system. We have already dealt with the
following rules:
29
We may ignore the spelling change, assuming our focus is on spoken English.
(66) Negative un- Rule (repeated from (61), but renamed)
[ X ]Adjective → [ un [ X ]Adjective ]Adjective Meaning: “not Adjective”
(67) -able Rule (repeated from (58))
[ X ]Verb → [ [ X ]Verb -able ]Adjective Meaning: “able to be V’ed”
We will also need a new rule. If you consider pairs like the following:
seat unseat
attach unattach
do undo
twist untwist
it should be clear that there is some kind of rule attaching un-. However, this rule cannot be the
same rule as (66), since it attaches un- to verbs. Moreover, its meaning is not really negative
(unseat doesn’t mean “to not seat”) but rather something more specific, which we might call
reversive; each of the un- verbs in the list more or less reverses the action of the simple verb.
Write the rule yourself, then check your work on the next page.
(68) Reversive un- Rule
[ X ]Verb → [ un [ X ]Verb ]Verb Meaning: “reverse the action of X”
We now have the apparatus we need to characterize an ambiguity, namely untieable (as in,
“My shoes are untieable.”). Just to be clear about it, we could provide contexts that make both
meanings clear (the usual term for this is “disambiguating context”).
(69) Two meanings of “untieable”
a. My shoelaces were untieable, so I went ahead and untied them.

b. I tried and tried to tie my shoelaces, but it turned out that they were simply untieable.
We could say, following the bits of meaning we put into our word formation rules, that the
meaning in (70)b is something like “not able to be tied”, and the meaning in (70)a is “able to be
untied” or (more explicitly) “able to be undone with respect to tieing”. This is meant to lead up to
the actual morphological derivations that generate the two meanings.
(70) Two meanings of “untieable”: derivations
a. [ tie ]Verb stem

[un [ tie ]Verb]Verb ‘undo the action of tieing’ Reversive un- Rule
[[ un [ tie ]Verb ]Verb able ]Adj ‘able to be undone with respect to tieing’ -Able Rule
b. [ tie ]Verb stem

[[ tie ]Verb able ]Adj ‘able to be tied’ -Able Rule
[ un [[ tie ]Verb able ]Adj]Adj ‘not able to be tied’ Negative un- Rule
19. What meanings are expressed by word formation rules?
The short answer here is “almost anything,” as we’ll see shortly. But there are some core
meanings.
19.1 Changing syntactic category
Perhaps the most common purpose of word formation rules is to change syntactic category;
we may want to say pretty much the same idea, but using the stem as a noun instead of a verb:
He had trouble concentrating. (verb)

He had trouble with his concentration. (noun)
In English, there are word formation processes that can change between any pair of the three
major syntactic categories of verb, noun, and adjective:
(71) Part-of-speech-changing word formation in English
Verb to noun: -ation, as above (as in visitation)

usual meaning: “the process of Verbing”
Noun to verb: -ify, -ize (as in classify, demonize)
usual meaning: “to do something involving Noun”
Adjective to noun: -ness, -ity (as in goodness, perversity)

usual meaning: “the quality of being Adjective”
Noun to adjective: -esque (as in picturesque)
usual meaning: “resembling Noun”
Verb to adjective: -ent (as in effevescent)

usual meaning: “tending to Verb”
Adjective to verb: -ify (as in clarify, humidify)
usual meaning: “render Adjective”
The following terminology is sometimes used to describe such cases:
Visitation is a deverbal noun. (noun derived from verb)

Classify is a denominal verb (verb derived from noun)
Perversity is a deadjectival noun.
Picturesque is a denominal adjective.
Humidify is a deadjectival verb.
Effervescent is a deverbal adjective.
19.2 Changing the number of participants in the action of a verb
Verbs often have rules of word formation that change the number of participants involved in
the action they describe. Consider the Persian verbs below.
(72) Word formation processes the change the number of participants in a verb
res-idan ‘reach-infinitive’
res-aːn-idan ‘send-infinitive’
xaːb-idan ‘to sleep’

xaːb-aːn-idan ‘to put to sleep’30
Here, we can take a verb that has just one participant (the one who is reaching, or sleeping),
and make from it a verb that has an additional participant (the one who causes to reach, or causes
to sleep). This is called a causative verb. English has no such word formation process, and uses
syntactic constructions to express causation (“He made them sleep”).
30
In the International Phonetic Alphabet, the symbol [] means that the preceding vowel is long.
For Persian, the rule could be expressed as follows.
(73) Persian Causative Verb Formation

[ X ]Verb → [ [ X ]Verb aːn ]Verb Meaning: “to cause to Verb”
19.3 The grand miscellaneous
Although the two purposes of word formation rules just given are probably the most common
across languages, individual languages can include word formation rules that are strikingly
specific. Among my favorites is one in Ilokano (Philippines), with a process that derives from a
verb a new verb meaning “to pretend to be verbing”.
dait ‘to sew’

agindadait ‘to pretend to be sewing’
sait ‘to cry’
aginsasait ‘to pretend to cry’
d͡anitor ‘to work as a janitor’
agind͡ad͡anitor ‘to pretend to be a janitor’
As you can see, the rule involves copying the first two sounds of the stem (a process to be
discussed below), followed by addition of the prefix [agin-].31
English has some very specific word formation processes:
bowl-arama, carpet-arama ‘grand emporium for X or X-ing’

Stalin-ism, Mao-ism ‘doctrine propounded by X’
pay-ola, shin-ola, plug-ola ‘bribery involving X’
20. The ordering of word formation and inflection
It is at least a strong cross-linguistic tendency — perhaps a universal pattern in language —

that rules of word formation apply before inflectional rules. Thus, for instance, in English it is
possible to have words like nullifies, which are derived as follows (I’ll use IPA transcription to
skirt issues involving spelling):
[nl]Adj. root: ‘null’

[[nl]Adj.fa]Verb Word formation rule: Adjective + /-fa/ → Verb
[[nl]Adj.fa+z]Verb Inflectional rule:
X → Xz if [Verb, +3rd person, +singular, +present]
31
[] is IPA for the glottal stop, a sound that English speakers make in the middle of “uh-oh” [o]
The opposite rule ordering would have derived *[nlzfa], so that the inflectional suffix would
appear “inside” the derivational suffix. Cases of this sort are rare at best.
This has implications for when you analyze a new language: typically it is possible to work
out the inflection — appearing on the “outside” of the word, and then work with the leftover
material and find the word formation rules.
21. Compounding
A widespread view of compound words is that they are a form of word formation. They differ
in that rather than attaching an affix to a stem, they concatenate (chain together) two stems.
Here is a same rule of compounding for English nouns:
(74) English Noun Compounding Rule
[ X1 ]Noun + [ X2 ]Noun → [ [ X1 ]Noun [ X2 ]Noun ]Noun

Meaning: “an X2 that has something to do with X1.”
Thus: boat house (structure: [[ boat ]Noun [ house ]Noun ]Noun) is a house that has something to
do with boats (for example, you keep boats inside it). A houseboat is a boat that functions as a
house.
The word tigerbird is probably not familiar to you, but you can guess part of its meaning
simply by knowing how to speak English: you know it is a kind of bird (and not a kind of tiger),
and that it has something to do with tigers (perhaps it is striped like a tiger, or it likes to roost on
top of sleeping tigers, or that it fights like a tiger, and so on).
Compounds like houseboat, boathouse, and tigerbird, derived by the rule given above, are
said to be headed: the “head” of houseboat is boat, because a houseboat is a boat. Likewise house
is the head of boathouse, because a boathouse is a house, and bird is the head of tigerbird.
In English, most compounds have at most one head, but other languages allow “double-
headed” compounds, for instance when “mother-father” is used to mean “parents.” One rare
English example is Austria-Hungary, which designated a country of Europe (1867-1918) that
included both Austria and Hungary. Double-headed compounds can be derived with a rule that is
exactly like the compound rule given above, except that the meaning has to be stated differently;
something like “an entity that includes both X1 and X2.”
It is possible to form a compound from two words one of which is itself a compound. For
example, we can combine the compound law degree with the word requirement to get the complex
compound law degree requirement. This compound can in turn be combined with changes to get
law degree requirement changes; and so on. The following example suggests that the process is
essentially unlimited:
(75) Unlimited compounding in English
a. Data
eggplant ‘plant shaped like an egg’
eggplant plant ‘factory for manufacturing eggplants’
eggplant plant plant ‘factory for manufacturing factories for
manufacturing eggplants’
etc.
b. Derivation
[ egg ]N + [ plant ]N input stems: egg, plant
[ [ egg ]N [ plant ]N ]N English Noun Compounding Rule
‘plant that has something to do with
eggs’
[ [ egg ]N [ plant ]N ]N + [ plant ]N adding a second input stem: plant

[ [ [ egg ]N [ plant ]N ]N [ plant ]N ]N English Noun Compounding Rule
eggplants’
[ [ [ egg ]N [ plant ]N ]N [ plant ]N ]N + [ plant ]N adding a third input stem: plant

[ [ [ [ egg ]N [ plant ]N ]N [ plant ]N ]N [ plant ]N ]N English Noun Compounding Rule
eggplant plants’
The glosses given in (75) are what can be deduced from the rule itself; the glosses in (75) reflect
what normally happens in English compounding: the bare meaning from the rule is elaborated into
something sensible in context.
A consequence of this sort of repeated compounding is further support for our claim about
rules of word formation: they are applied freely (potentially, over and over), rather than in the
strict “assembly line” fashion seen in rules of inflectional morphology.
21.1 The spelling of compounds
The spelling system of English is inconsistent with regard to compounds; some are spelled
without a space between the component words and some are spelled with a space. It is important to
realize that an expression spelled with a space can still be a compound.
One can argue for this in two ways. First, consider German: it is customary in German to
spell all compounds without a space between the component words. That is, the English practice is
more or less an accident, given that other languages go the other way.
More important, there are linguistic arguments that compounds spelled with spaces are just
like compounds spelled without them. Note first that, in the case of a genuine NP of the form
Adjective + Noun, it is possible to insert an extra adjective between the adjective and the noun. For
example, we can take the NP large cake and add an additional adjective to get large round cake.
But if we start with a compound, it is impossible to get an additional adjective in the middle. For
example, starting from pancake, we cannot get *pan round cake. The basic point is the while the
noun of a NP can be modified by an additional immediately preceding adjective, a noun that is the
second word of a compound cannot.
This fact provides us with a test to determine whether an expression really is a compound,
even if it is spelled with a space. For instance, we can show that carrot cake is a compound by
trying to place an adjective in the middle: *carrot large cake. Other examples also show that
expressions spelled with a space can be compounds:
coal scuttle *coal dirty scuttle

lap dog *lap slobbery dog
can opener *can handy opener
22. Summing up so far
The picture of morphology thus far drawn is like this.
First, languages have means of expanding their inventory of words (more precisely: of stems).
The rules of word formation add affixes to stems to derive new stems, which have new meanings.
These meanings can be common, characteristic ones (like “the quality of being Adjective”, “to
cause to Verb”), or exotic ones (like “emporium for selling Noun”). Compounding likewise
expands the stock of stems, creating either single-headed compounds (like boathouse) or, in some
languages, two-headed ones (like Austria-Hungary).32 There is in principle no limit to “when” a
derivational rule can apply; it simply looks for the right kind of base form and applies optionally.
The stems that result, whether they are basic or derived, are used in sentences. In a sentence
context, the rules of the syntax (as yet undiscussed) provide each stem with a morphosyntactic
representation, that is to say, a bundle of inflectional features. These features are specific to a
particular language, although a number of features like [Case:Accusative] or [Number:Plural]
occur repeatedly in languages. The features are referred to by the rules of inflectional
morphology, which add affixes in order to express their content overtly. It is generally possible to
arrange the affixes of an inflectional system into “slots”, where each word has at most one affix
per slot. In terms of rules, the slots are expressed by having one rule per slot; each rule attaches the
affix that corresponds to the features given in the morphosyntactic representation of the stem.
As a consequence of this scheme, inflectional morphology, being attached by rules that apply
“later”, occurs on the “outside” of a word; that is to say, further from the stem than word
formation.
32
There are also compounds with implied heads, like airhead. These typically have an unstated head,
usually meaning “person” or “thing”. Thus airhead means, essentially, “air-headed person” = “person with
head filled with air” = “person who doesn’t think”. Similarly: pick-pocket “person who picks pockets”;
stopgap “thing that stops gaps”.
23. Phonological realization in morphology
When I say “phonological realization”, I mean the arrangement of the phonological material
(speech sounds) that realizes the morphological categories, whether they be derivation or
inflectional. I would guess that a large majority of all morphology (in the narrow sense that
excludes compounding) is prefixation, suffixation, or compounding. All three are concatenative,
in the sense that they string together sequences of speech sounds.
But concatenation is not the only way you could carry out an inflectional or derivational
process: segments can be interpolated, or copied, or altered in their phonetic content. Below, I will
give some cases, and present ways that explicit rules can be written for them.
Note that all of these non-concatenative forms of morphology can be used for both inflection
and derivation — on the whole, the functions of morphology (grammatical or derivational) can be
studied independently of the changes in phonological material that carry out these functions.
23.1 Infixation
The following data from Bontoc (Philippines) illustrate infixation, which can be defined as
insertion of segments into some location inside the base:
(76) Bontoc infixation
fikas ‘strong’ fumikas ‘he is becoming strong’

kilad ‘red’ kumilad ‘he is becoming red’
bato ‘strong’ bumato ‘he is becoming stone’
fusul ‘enemy’ fumusul ‘he is becoming an enemy’
It’s reasonably clear that this is a derivational process, forming deadjectival (first two
examples) or denominal (last two examples) verbs. The brackets we’ll need are something like
this:
(77) Rules for Bontoc infixation (first draft)
[ X ]Adj → [ X ]Adj]Verb
Meaning: “become Adjective”
[ X ]Noun → [ X ]Noun]Verb
Meaning: “become Noun”
But how to express the infixation? The important part here is to be precise about just where the
infixed material should be inserted. We will use here a method that makes uses of variables and
subscripts.
The variables we have seen already with simple rules in prefixation and suffixation, as in
[ X ]Adj → [[ X ]Adj ness ]Noun. Here, instead of simply expressing the speech sounds of the base
with a simple variable X (meaning: any sequence), we will give this part of the rule more
structure, sufficient structure to specify where the infix goes. The needed symbols, widely used in
linguistics, are C for consonant, V for vowel. Doing just the adjective case (the noun example
works the same), we have:
(78) Formalizing the deadjectival Bontoc infixation rule
[ C V X ]Adj → [[ C um V X ]Adj ]Verb

1 2 3 1 2 3
Meaning: “become Adjective”
You can read the rule above as follows:
(79) “An adjective base consisting of a consonant, followed by a vowel, followed by anything,
is converted to an verb by inserting the sequence -um- after the consonant.”
Some details: the numbers under the terms of the rule are included to make sure we are clear
on what matches up with what (important if, for example, a rule contains more than one C).
Applying the rule to the form, we have the following matchup:
[ f i kas ]Adj [[ f um i kas ]Adj ]Verb
[ C V X ]Adj → [[ C um V X ]Adj ]Verb

1 2 3 1 2 3
Infixation is not common in English. You are probably familiar with the colloquial expression
fan[ˈfkən]tastic,33 in which a taboo word is placed in the middle of the stem, as a kind of infix.
Cockney dialect recognizes similar constructions like abso-bloody-lutely. From time to time
linguists have proposed analyses that predict, for any given word, where the expletive can be
infixed; this turns out to be a surprisingly difficult area for analysis.34
Infixes are normally written with both preceding and following hyphens, since they have two
“joining points”: -um-.
A caution concerning infixes: not all morphemes in the middle of a word are infixes. Many of
them are prefixes/suffixes that happen to have had additional material added to their left/right: in
ex-vice-president, vice- is a prefix, not an infix. You can identify the infixes by their ability to
occur in the middle of a morpheme.
33
IPA symbols: [] is the vowel of cut, [ə] is the second vowel of taken.
34
A pretty good analysis appears in John McCarthy (1981) “Prosodic structure and expletive
infixation,” Language 58, 574–590, available at
http://scholarworks.umass.edu/cgi/viewcontent.cgi?article=1062&context=linguist_faculty_pubs.
Infixes are normally considered to be affixes (like prefixes and suffixes); the English cases
above, a curious sort of “compounding infixation”, are an exception.
23.2 Reduplication
Reduplication is a morphological process in which all or part of a word is copied. We have

already seen an example of reduplication in Ilokano, for which I repeat the data below:
dait ‘to sew’

agindadait ‘to pretend to be sewing’
sait ‘to cry’
aginsasait ‘to pretend to cry’
d͡anitor ‘to work as a janitor’
agind͡ad͡anitor ‘to pretend to be a janitor’
(d͡ is IPA for the “j” sound, heard twice in English judge.) We want to use our subscript notation
to produce a copy of the first consonant-vowel sequence; and also we need to attach the additional
prefix agin-. The following rule can do this:
(80) The “pretend to Verb” rule of Ilokano formalized
[ C V X ]V → [ agin C V [ C V X ]V]V
1 2 3 1 2 1 2 3
Meaning: ‘to pretend to Verb’
The point of the repeated subscripts is to designate what is getting copied. Here is a derivation for
agindadait; brackets omitted for clarity.
‘to sew’ ‘to pretend to sew’
d a it agin d a d a it
C V X → C V C V X
1 2 3 1 2 1 2 3
Here is a second example. In Samoan (S. Pacific), the plural form of a verb is formed by
reduplicating the second-to-last syllable of a verb:
(81) Samoan plural reduplication
nofo ‘he sits’ nonofo ‘they sit’

pese ‘he sings’ pepese ‘they sing’
savali ‘he walks’ savavali ‘they walk’
atamai ‘he is wise’ atamamai ‘they are wise’
We can use our numerical subscript notation to express the Samoan process unambiguously:
(82) Samoan Plural Subject Reduplication as a rule
X C V C V → X C V C V C V
1 2 3 4 5 1 2 3 2 3 4 5
when the morphosyntactic representation contains [Number:Plural]
The rule tell us to count off the final CVCV of a word, and copy its first CV sequence (what is
numbered “23” in the rule). Here is a derivation for savavali:
savali[Number: Plural]
sa v a l i sa v a v a l i
X C V C V → X C V C V C V
1 2 3 4 5 1 2 3 2 3 4 5
In Yidi,35 an aboriginal language of Australia, the intensive form of verbs is created by

adding to the beginning of a word a copy of the first two syllables of the word:
(83) Yidiɲ reduplication
madjindan ‘walk up’

madjinmadjindan ‘keep walking up’
djadjaman ‘jump’
djadjadjadjaman ‘jump a lot’36
We can’t formalize the Yidi rule (yet) because we haven’t developed a theory of syllables.
The meanings expressed by reduplication are often “symbolic,” designating plurals,

intensiveness, repeated action — things present in quantity.
35
ɲ is a symbol of the IPA. Imagine make a sound like the first sound in English jump, except that it’s
a nasal (similar to m or n).
36
dj is the IPA for a voiced lamino-palatal stop, similar to the English “j” sound.
Study Exercise #8
Write the rule for forming causatives in Ateso (Nilotic family, Uganda). 37
duk ‘to build’ tuduk ‘to cause to build’

lel ‘to be glad’ telel ‘to gladden’
ɲam ‘eat’ taɲam ‘feed’
wadik ‘write’ tawadik ‘cause to write’
cak ‘throw’ tacak ‘cause to throw’
IPA phonetic symbols: vowels more or less as in Spanish, [ɲ] is rather English “ny”, [c] is rather
37
like English “ch” only made further back in the mouth.

In words, you begin the word with /t/, then copy the first vowel, the conclude with the rest.
Using the notation taught here, this is:
[ C V X ]Verb → [ tV [ C V X ]Verb ]Verb

1 2 3 2 1 2 3
Meaning: “cause to verb”
So, in wadik/tawadik, 1 is w, 2 is a, 3 is dik.
—————————————————————————————————————
23.3 Segment Mutation
Morphology sometimes is carried out by finding a particular segment (perhaps in a particular

context) and changing it into something else. For instance, a fair number of English verbs form
their past tense by changing the stem vowel [] into [] (the value of the IPA symbols should be
clear from context).
(84) Present Past

cling clung
sling slung
fling flung
string strung
wring wrung
While these are irregular verbs (and thus are probably memorized), the process is nevertheless
a little bit productive: forms have arisen in dialects like sing - sung, ring - rung, bring - brung; the
latter is explored for a period even by many children whose parents say brought. In experiments,
people asked to provide a past tense for the made-up verb spling often volunteer splung.
We can state this rule as follows, noting that a crucial element in (most of) these verbs is the
presence of a following [], the “ng” sound:
(85) A segment-mutation rule for English past tenses
[ X   ] → [ X   ]
1 2 3 1 2 3
when the morphosyntactic representation contains [Tense:Past]

23.4 Morphological conversion
Consider the following data, which illustrate a process that forms deverbal nouns.
(86) Examples of verb-to-noun conversion
Fred likes to jump. His last jump was 20 feet.

Chomsky likes to talk. His last talk was attended by 500 people.
We need to think. We had a good long think.
These cases have simplest possible string operation of all; that is, nothing changes. Such rules,
which are often called rules of morphological conversion, can be expressed as follows:
(87) Verb-Noun Conversion Rule
[ X ]Verb → [ [ X ]Verb ]Noun

meaning: “an instance of Verbing”
There is also a rule that goes in the opposite direction, for data like these:
(88) Examples of noun-to-verb conversion
Noun Verb
a mop I mopped the floor.
a fax I faxed the message.
a hammer I hammered the nail in.
The rule needed is something like this:
(89) Noun-Verb Conversion Rule
[ X ]Noun → [ [ X ]Noun ]Verb

meaning: “do something crucially involving Noun”
These are simply word formation rules that carry out no affixation (or any other change).
There is no reason to exclude them from the theory, and indeed they seem to be pretty common
among languages.
One might ask why we want rules going in both directions. The best answer, perhaps, is that
the morphological base form in each case is somehow semantically primary: a jump is what
happens when you engage in jumping (not: “jumping is what happens when you execute a jump”);
mopping is the activity you do with a mop (not: “a mop is the device you mop with”).
Occasionally in older linguistic works one will find the claim “Language X lacks a distinction
between nouns and verbs.” This is currently viewed as rather implausible; instead, one could say
that in Language X, morphological conversion between nouns and verbs is highly productive, so
most nouns stems can be used as verb stems and vice versa. In any language, there are good
syntactic reasons to want to have a distinction between nouns and verbs.
The conversion rules of English will turn out to matter quite bit starting in the next chapter, as
we start to parse (assign syntactic structure to sentences). You have to treat a word like hammer in
a context-dependent way, so that for example it is a verb in Please hammer the nail in and a noun
in This is an excellent hammer.
Study Exercise #9
Does English have adjective-to-noun conversion? Try to find examples. Specify the meaning
that this process imparts. Write a formalized rule for it. Answer on next page.
You can get adjective-to-noun conversion in English most easily in a particular context,
namely with a preceding definite article the.
Adjective Noun derived by conversion

lame (the) lame
able (the) able
meek (the) meek
lonely (the) lonely
tall (the) tall
It’s clear that there is a rather particular meaning here: “the tall” means “people who are tall”.
Formalizing:
(90) Adjective-Noun Conversion Rule
[ X ]Adjective → [[ X ]Adjective]Noun
Meaning: “those people who are Adjective”
This analysis is incomplete in that it doesn’t indicate the special context of occurring after the.
For further reading

There are many textbooks in introductory morphology, among which I have appreciated Geert
Booij’s (2012) The grammar of words. An introduction to linguistic morphology. 3rd edition.
Oxford: Oxford University Press.
The estimate for the size of the Turkish verbal paradigm on p. 34 comes from a well-known
introduction to computational linguistics, Daniel Jurafsky and James Martin’s book Speech and
Language Processing (Prentice-Hall 2000), where it is attributed to the computational linguist
Kemal Oflazer.
Chapter 3: Normative views of language
1. Introduction
Let us start with some examples. I suspect there are plenty of people in the world who believe
each of the following assertions:
(91) Examples of normative views
a. “French has a more beautiful sound than German.”

b. “Brung as the past participle of bring is bad English.”
c. “It is better to say ‘it is I’ than ‘it is me’.”
d. “[Such and such an ethnic group ] speaks a substandard dialect of the language”
e. “Australian accents sound cool.”
f. “Southwestern Mandarin sounds rural and unsophisticated.”
g. “Southern accents sound friendly.”
h. “Southern accents sound ignorant and uneducated.”
Such beliefs are called normative, because they imply a judgment of “good” or “bad” in some
way. The basis of the judgment might be esthetic (among (91), perhaps (a)), a sense of correctness
((b), (c)), or indeed an opinion about a whole class of people, defined by ethnicity (d) or geography
(e-h). As the examples show, normative beliefs can be about some particular word (a) or grammatical
construction (c), or about whole languages or dialects (all others).
For linguists, normative beliefs impinge on their work in two ways.
First, there is methodology: given that it is basically inevitable that any human, including a
linguist, will hold normative beliefs, how might we make sure they don’t harm the quality of
linguistic research?
Second, normative beliefs about are a (relatively minor) topic within the field, that give rise to
research questions, for instance:
• How do we find out about normative beliefs and assess them?

• Where do normative beliefs come from? Why do they arise?
• Are normative beliefs ever “justified” in a factual sense?
2. The professional practice of linguists concerning normative beliefs
Since as linguists we want to do good science, the primary issue here is the worry that our
own normative beliefs might impede our objectivity. My own favorite metaphor for this is the
clean white lab coat — the emblem that a laboratory scientist wants to keep the samples clean and
uncontaminated. As linguists, we keep our lab coats clean (in part) by ignoring what we feel about
language, and concentrating on the data.
Scientific objectivity is of course a goal that cannot always be attained. Everyone, including
experienced linguists, has normative beliefs, and we can’t make them go away. To speak
personally on this point: I find that whenever I encounter a phrase like “very unique,” or the
pronunciation [ˈnukjulər] (“nucular”) for nuclear, I experience real, unavoidable normative
feelings — a brief, concealed, inner squirm. Both cases just cited are instances in which the
normative belief is one that favors the older meaning or pronunciation (see more on this below).
But as a scholar I know there is nothing inherently wrong with them — the world would not come
to an end if everyone started saying [ˈnukjulər]! And when I am doing linguistics, I can try to
factor out my feelings from my thoughts and analysis.
The fact that even linguists are vulnerable to normative feelings has consequences for how
linguistics is conducted. First, the ability to maintain a poker face is valuable: a careful fieldworker
will be able to conceal from their consultant any normative feelings that they may have about what
the consultant is saying. Any expression of such feelings is likely to distort the material given by
the consultant later on, as (s)he seeks to avoid the embarrassment of being laughed at or otherwise
negatively evaluated. (Suppressing normative opinions about the consultant’s speech is, of course,
also a good way to retain a good working relationship with the consultant.)
Linguists also display their commitment to objectivity by employing professional vocabulary

that shows that (at least when at work) that they do not buy into the normative beliefs that may be
are held by the speakers of the language being investigated. For instance, a linguist would be likely
to use the term “nonstandard” rather than “substandard”; “prestigious forms” for “correct speech”.
3. Investigating normative beliefs
We turn next to the study of normative beliefs as a sub-area of linguistics.
To learn about normative beliefs, a good starting point is simply to attend to what people say
about language. For instance, the “Cockney” dialect of English is that historically spoken by
working-class people in poorer neighborhoods of London. It is fairly familiar to Americans
because we hear it in mouths of fictional characters in film and drama. Here is a reported opinion
of Cockney from about a century ago:
‘inspectors and teachers of English in London elementary schools who met in conference in 1906
declared that “The Cockney [London lower-class] mode of speech, with its unpleasant twang, is a
modern corruption without legitimate credentials, and is unworthy of being the speech of any
person in the capital city of the Empire.’38
The description may surprise Americans, since the Cockney they hear is sentimentalized; usually
placed in the mouths of fictional characters who are uneducated but have a heart of gold.
All over the world, there are dialects that are considered (by many people) to be prestigious
and dialects that are considered (by many people) to be non-prestigious — often the way that this
38
Source: Does Accent Matter? (1989) by John Honey.
is put is that the speakers do not speak “correct” English (or Spanish, or Korean, etc.). Cockney, at
at least historically, was a clear example of a low-prestige dialect of a language.
What sort of dialect is likely to have low prestige? There seem to be at least three criteria:
• Social class, as in Cockney. A similar case is caste, the formal system of social hierarchy
found in India.39
• Minority ethnicity, as in the German-influenced varieties of English spoken in North
Dakota and neighboring states, or African-American Vernacular English (Black English).
• Geography: the varieties of Korean spoken outside Seoul, and some varieties of Japanese
spoken outside Tokyo, tend to be stigmatized.
To some degree, you can get an idea of the prestige of varieties of language just by asking
people, but social psychologists have tried to be more systematic about it. A favored research
method is the so-called matched-guise experiment:40 you find a perfect bilingual or bidialectal,
and have them say (more or less) the same thing in both of the language varieties in question. You
also mix in many other voices, so that, if all goes well, the experimental subjects who listen to the
recording aren’t aware that one person is speaking twice. The subject are asked to rate the speakers
on various scales, for instance:
• intelligence
• suitability for employment
• trustability
• likelihood to be a friend
The measurement of interest concerns how these ratings differ for the recordings of the same
speaker saying (essentially) the same thing in two languages or dialects.
By now, dozens of matched-guise experiments have been carried out around the world.
Generally, they show what you might expect: that people who are speaking a prestigious dialect
are judged as more intelligent and suited to positions of responsibility. For the more intimate
criteria of trustability and friendliness, the less prestigious variety sometimes wins, but quite often
the more prestigious variety does. Often enough, prestigious varieties are preferred even by the
native speakers of the non-prestigious variety — at least when they are giving reponses in an
experimental setting, which is itself an academic, prestige-oriented environment.
This is what matched-guise experiments teach us. However, they are limited in their scope,
due to the artificiality, just noted, of the experimental setting. A more nuanced view would be that
there are different kinds of prestige. Nonstandard varieties are valued, at least by their speakers, as
39
A classic article is John J. Gumperz (1958) Dialect differences and social stratification in a north Indian
village. American Anthropologist 60: 668-682.
40
A good review is in Ralph Fasold (1984) The Sociolinguistics of Society, Basil Blackwell, Oxford,
Chapter 6.
badges of community membership, and members of a community with a non-standard dialect who
speak the standard dialect to their peers are “sending a message” of some sort.41
Educators, particularly of young children, often have an extremely delicate task: they judge
that teaching a standard variety to non-standard speakers will help their students with their lives
and careers, but the tacit message “your dialect is inferior” that may come with this training is not
a very nice — or, as we will see, valid — message to give to kids. Enlightened educators try to
steer a course between the need to teach a standard, and the need not to alienate their students
when teaching the standard.
4. The origin of normative beliefs
Why do people have normative beliefs about language? This question is in need of further
study, but it seems reasonable to point out two possible sources.
4.1 Societal division
Many normative beliefs seem to stem from the divisions found in a society. I don’t think it is
controversial to say that every society is in a state of conflict, ranging from mild to extreme. The
divisions can be ethnic, economic, or geographic. In general, the varieties of language that are
affiliated with power will be the more prestigious ones. This includes varieties spoken by wealthier
and better-educated people; the varieties spoken in the capital city of a country; and varieties
spoken by a politically powerful ethnicity.
An interesting comparison of this sort can be made when the very same language has different
status in different locations. French has an exalted status in France, where it is the dominant
language, but until recent decades it had low status in Quebec, which was economically dominated
by English speakers. German once had very high status in Latvia, where it was the language of an
economically dominant foreign-based minority. German was less prestigious in 19th century
America, where it was widely spoken but gradually abandoned by its speakers in favor of English.
I believe that the key point is this: feelings about language varieties are often projected from
feelings about the people who speak them. The social structure and inner conflicts of a society can
be diagnosed to some degree by querying its members about their normative linguistic beliefs.
Nevertheless, often feelings about language are felt to be specifically feelings about language; the
causal connection to feelings about people is not grasped.
4.2 Linguistic conservatism
A rather different, and less political, source of normative beliefs results from the ever-present
process of language change. Typically, speakers will feel that the older forms of a language are
inherently “correct” and that the innovating forms are wrong. For example, “it is I” is the older
form; “it is me” is an innovation. Putting the stress on the first syllable of compensate and
41
The web cartoonist Randall Munroe points out that there is no way to avoid sending some sort of message by
the way you talk; you can’t “opt out”: https://xkcd.com/1735/
confiscate was sometimes considered vulgar in the 18th century, since at that time many people
still used the old pronunciation with stress on the second syllable.42
The example I gave above of my own normative beliefs (“very unique”, [ˈnukjulər]) are of
this kind: the older meaning of “unique” is “exactly one” (so it makes no sense to modify it with
“very”); nowadays, “unique” is coming to mean “unusual” (so it’s perfectly sensible to say “very
unique”). The older pronunciation of nuclear is [ˈnukliər].
5. The labels “language” and “dialect” as used by linguists and in ordinary life
Once we have a clear picture of normative beliefs and their basis, we can define the terms
language and dialect. It is helpful to provide two definitions for each word: one as they are
commonly used in linguistics and one for the way they are commonly used in ordinary language.
In linguistics:
• We start with the concept of idiolect. An idiolect is the version of a language spoken by
one single person. For example, my own idiolect of English represents the large set of rules
of this language in the version that is currently stored in my mind/brain, ready for the
creation of novel utterances. My idiolect would also include my lexicon; the store of
English words and idioms I have memorized in the course of my lifetime.
• A dialect is a relatively uniform set of idiolects; people who speak the same dialect can
communicate fluently with minimal possibility of misunderstanding. Naturally, the identity
of idiolects is never perfect, so the concept of dialect is not a precise one.
• A language is a collection of mutually-intelligible dialects. Here again, the concept is

imprecise since there are dialects so different that mutual intelligibility is possible but
marginal.
Among dialects, one speaks of “standard dialects”, which in general are the dialects associated
with power, education, and prestige. Thus, for instance, in Mexico the variety of Spanish spoken in
Mexico City is generally considered to be standard. However, from the viewpoint of the linguist
the “standard” dialect is just another dialect.
In ordinary speech of non-linguists, the difference between a “language”and a “dialect” is

quite different, and it is totally bound up with normative belief. Going out on a limb, we might say
that in ordinary usage, a “dialect” is:
• a language variety that is (according to prevailing normative beliefs) non-important.

• a language variety that is non-standard (see below on this term).
42
The shift can be observed in progress an influential 1791 dictionary by John Walker, who specifically offered
his book to (native speakers of) English as a way of improving their speech. Walker often carefully deliberates which
of two stress variants is better, citing older authorities and using words such as “propriety”.
An example of “non-important” occurs when the World Almanac lists the languages of some
faraway country as “French, English, African dialects.”43 An example of “non-standard” is in a
sentence like I asked the farmer for directions and was amused when he replied in dialect.
An old saying shared among linguists44 is, A language is a dialect with an army and a navy.
This is meant to be silly but is often surprisingly accurate when applied to the real world.
Cantonese, which by the linguist’s mutual-intelligibility criterion is a language, has no army or
navy of its own and indeed is commonly called a “dialect” of Chinese. Norwegian, Swedish, and
Danish, which by the linguist’s criterion would constitute divergent dialects of Scandinavian, all
have their own armies and navies and all get counted by the public (including their own speakers)
as languages.
6. Standard and vernacular languages
A special kind of language — which was absent from the world before the rise of writing,
education, and nation-states — is the standard language. Standard languages as used today are
found in education, journalism, literature, and government. Standard languages always have a
system of writing. A standard language is usually the required medium of instruction in schools,
and there is normally an army of (mostly hidden) copy editors who labor to make sure that the
written output of authors and journalists conforms to the standard language prior to publication.
A standard language normally co-exists in the same territory with one or more vernacular
varieties of language. Vernacular varieties belong to the community; they are spoken among
family and with friends. Usually, people do not write in a vernacular; they use their standard
language instead for this purpose. Throughout the world, it is extremely common to be bilingual or
bidialectal, capable of speaking both a standard language (often first encountered when a child
enters school) and a home vernacular.45 The home vernacular can either be a dialect of the standard
language or a completely different language. Vernaculars are often endangered; standard
languages hardly ever.
The choice of a standard language reflects a society’s history — typically, it is the descendent
of some vernacular language that happened to be spoken by the winners in a military, political, or
cultural struggle of long ago. The ancient Romans proved very skilled at warfare and conquest, and
it is uniquely this fact that explains how the little Italic dialect they happened to speak (Latin)
emerged as a mighty pan-European standard. There is nothing about the contending languages
themselves that influences the possibility that they will emerge as standards.
43
Here is another one: all the languages of China that are genetically related to Mandarin are called
“dialects” in English. So, in ordinary English speech Cantonese is a “dialect” but for a linguist it is a
language, being mutually unintelligible with other varieties of Chinese.
44
There’s a not-so-bad article on the Wikipedia about the origin of this saying;
https://en.wikipedia.org/wiki/A_language_is_a_dialect_with_an_army_and_navy.
45
Due to migration, many people speak a language that would count for them as a home vernacular, but is a
standard language elsewhere. Sometimes such speakers seek out language training so that they can bring to their home
language the skills (literacy, vocabulary size, detailed normative principles) that are hallmarks of a standard variety.
To return to the main thread of normative beliefs: a standard language is usually the most
prestigious form of language spoken in its territory; obviously, exclusive use in education,
government, and publication provides a great deal of prestige.
7. Are normative beliefs ever justified?
Let us now take on the most loaded question of all: is it really true that one language or
dialect could legitimately be called inferior to another?
It seems unlikely to me that any language could be significantly simpler than any other. The
reason I believe this is that field workers who go to work on a language never believe that they’re
done. A responsible and accurate reference grammar of a language46 will go on for hundreds of
pages, and still be giving just a rough outline of many areas. The languages for which the only
grammars are thin ones are the languages that haven’t been studied much. What we know about
English would probably fill a large bookshelf. There’s little reason to doubt that the same would
hold of any other language that was submitted to the same degree of study.
Often, grammatical issues in a particular language are subtle or complex, and thus difficult for
the linguist to establish confidently. This holds true just as much for vernacular languages spoken
by peoples with simple material culture as for standard languages spoken in large industrialized
countries.
A related point is that all languages seem to be about equally expressive: roughly speaking,
whatever can be thought, can be said in any language; though the degree of effort needed might
vary in certain cases.
This claim is probably true for dialects as well. A famous article by the linguist William
Labov, “The logic of nonstandard English,”47 made a case for the grammatical integrity of African
American Vernacular English as a system (a well known fact about the dialect is that it has
distinctions of verbal tense not available in the standard dialect), and also for the distinction
between being a speaker of a standard dialect and being an articulate speaker (there are both
articulate and inarticulate speakers of both standard and nonstandard dialects).
Naturally, languages differ greatly in vocabulary. A language will normally include a

vocabulary suitable for the culture within which it is spoken; and indeed, experience suggests that
it is not at all easy just to take a random vernacular language “off the shelf” and adapt it instantly
as a standard language serving the needs of an industrialized society. But this seems to be a rather
superficial difference, as with time languages can acquire new vocabulary (through borrowing and
the use of word formation rules) to accommodate any culturally-novel concept. After all, every
standard language was once a vernacular.
46
A reference grammar is a grammar book written for linguists: its goal is to give the structure of a
language clearly. Other grammars are organized to help teach the language, or are addressed to the lay public.
For more on reference grammars see p. 435 below.
47
In his book Language and the Inner City (University of Pennsylvania Press, Philadelphia 1974).
Languages differ a great deal in their morphological complexity. But it would be a mistake to
equate morphology with overall complexity. In English, for instance, the inflectional morphology
is simple, but the choice of articles (the vs. a) is monstrously complex and difficult; it just happens
to be a problem in syntax and semantics, not morphology.
Earlier in this text (p. 42) I very tentatively suggested that there may be some virtue in
inflectionally-impoverished languages like Mandarin Chinese, which don’t force their speakers to
make commitments they don’t want or need to make. Yet as a native speaker of a mildly
inflectional language, I feel it is implausible that the inflectional choices of English are somehow
hampering my ability to communicate, and I’m sure that native speakers of richly inflected
languages like Turkish or Finnish would feel the same.
7.1 Putative cases of “illogicality” in language
It is sometimes said that stigmatized languages or dialects are “illogical.” For example, in
many dialects of English (including African American Vernacular English), the sentence
corresponding to standard English “You don’t know anything” is “You don’t know nothing”.
Some people believe that this makes the non-standard dialects “illogical”, in that they are “really
saying” something they don’t mean, namely “it is not the case that you know nothing.”
The absurdity of this is revealed by looking at other, non-stigmatized languages, which do the
same thing without being looked down upon. For example, in French “You don’t know anything”
would be translated as “Tu ne sais rien”, literally “You not know nothing.”
In fact, in non-standard English dialects, “You don’t know nothing” is completely clear and
unambiguous. The way one would say “It is not the case that you know nothing” would be “You
don’t know nothing”, with a heavy accent placed on nothing. There is no possibility for confusion.
So, for instance, the following sentence is a possible one in African American Vernacular English:
(92) I didn’t say nothing — I just said it very softly.48
The “illogicality” accusation is based on a fundamental analytic error, that of analyzing other
languages or dialects from the viewpoint of one’s own language or dialect. Every language and
dialect has a grammar, which to be understood has to be studied in its own terms.
I conclude that at present there seems to be very little justification for any claims that one
language or dialect is superior to another. Naturally, since I have my “white lab coat” on (see
above), I would not want to exclude the possibility that such justification could be discovered in
the future, but this is at present a hypothetical possibility.
48
Example taken from this article: Stefan Martin and Walt Wolfram (1998) “The sentence in AAVE,”
in Guy Bailey, John Baugh, Salikoko S. Mufwene, John R. Rickford (eds.) African-American English:
Structure, History and Use, pp. 11-36. New York: Routledge.
8. Normative beliefs: who is in charge?
The normative beliefs that arise from societal divisions, particularly those involving standard
languages, raise tough political questions. Notably, the speakers of nonstandard languages and
dialects face the dilemma that because of negative normative feeling on the part of the standard
speakers, their mother tongue — the vehicle of their innermost thoughts and of conversation with
their nearest and dearest — has negative consequences (such as loss of prestige, or worse) when
they speak it among members of the majority community. Naturally, this is not an easy outcome to
accept.
Rebellions have been known to occur: the speakers of a non-standard variety join in insisting
on the right to have their language treated as a standard (examples: French in Quebec, Catalan and
Basque in Spain, Irish in Ireland). Often the linguistic uprising goes hand in hand with a political
one. There are also minor, individual rebellions, consisting of speakers of non-standard dialects
choosing (consciously or unconsciously) not to alter their speech when talking with standard
dialect speakers.
9. Summary: normative beliefs
Summing up: normative beliefs about languages and dialects are found everywhere. They can
be measured in matched-guise experiments, and typically are a reflection of the structure of a
society; notably the difference between standard and non-standard varieties. With regard to
particular grammatical constructions, words, and pronunciations within a single dialect, normative
beliefs usually involve adherence to slightly archaic variants, that is to say, resistance to change.
Linguists, aspiring to be scientists, seek to be aware of their own normative beliefs, in order to
be able to guard against bias. A number of scholars are actively interested in the nature and causes
of normative beliefs and examine them as a research topic.
—————————————————————————————————————
Study Exercise #10
Go through the text of this chapter and make a list identifying all of the normative beliefs
mentioned. You can say “X is bad” if you like, but where possible be more specific. Answer
below.
Study Exercise #11
This is open ended, so can’t have a single correct answer. On p. 70 above, the text says
“Nonstandard varieties are valued, at least by their speakers, as badges of community
membership.” This is not the whole story, however. Sometimes nonstandard varieties are valued
by speakers outside the community. Try to think of examples. Use a search engine to try to find
some justification for your claim. (Note: Google Books and Google Scholar are more likely to get
you answers from people who have actually done some research on the question.)
I believe the following list is fairly complete.
1. Brung as past tense of bring is bad.

2. All items listed in (91)
3. “very unique” is bad
4. [ˈnukjulər] (“nucular”) for nuclear is bad
5. The Cockney dialect of English is unworthy.
6. Dialects of English with German influences are bad.
7. Non-Seoul dialects of Korean are bad
8. Non-Tokyo dialects of Japanese are bad.
9. French is good (for France)
10. French is bad (for Canada)
11. German is good (for Latvia)
12. German is bad (for America)
13. Compensate and confiscate ought to be pronounced with stress on the second syllable.
14. “It is me” is bad.
15. Double negatives are bad.
16. Double negatives are illogical (speaker isn’t saying what he really means).
Sample answer to Study Exercise #11
Here are a couple possible examples.
1. There are European-American youths who attempt to speak African American Vernacular
English. Naturally enough, this comes with an affinity for African American vernacular culture,
particularly as it appears in works of popular culture like music and film. Here is one reference:
Chapter 6 of Mary Buchholtz’s book White Kids: Language, Race, and Styles of Youth Identity
(2010) Cambridge: Cambridge University Press.
2. It’s apparently quite common for British popular musicians to sing, at least in part, in
American accents (from my youth I remember the Beatles singing [ˈdæns] for dance, rather than
British [ˈdɑns]). Of course, American English is the standard variety in America, but it isn’t in
Britain. Malchow (p. 108) puts this “singing in American” phenomenon in a broader context.
(Howard Malchow (2011) Special Relations: The Americanization of Britain? Stanford: Stanford
University Press.
Chapter 4: Syntax I — Phrase Structure
1. Knowledge of syntax
A theme of Chapter 1 was implicit knowledge: people show they possess such knowledge in
that it is reflected in the patterning of their language, but they cannot directly intuit the form of that
knowledge. Here, we will focus on the kinds of implicit knowledge encountered in studying
syntax, which is the study of sentence structure. What do speakers know when they know the
syntax of a language?
(1) They have intuitions about grammaticality. A sentence is grammatical if it is syntactically

well-formed; if it counts as “part of the language.” Grammaticality is distinct from merely making
sense. Consider, for example, the following series of sentences:
She wonders if Alice is going to like Bill.

Who does she wonder if Alice is going to like? (answer: Bill)
*Who does she wonder if is going to like Bill? (answer: Alice)
As far as meaning goes, the third sentence is as sensible as the second. It is only
ungrammatical. Similarly, sentences like *John and Bill think I like each other (p. 6) have a
perfectly sensible interpretation, but are ungrammatical. Sentences like Colorless green ideas sleep
furiously, however, are quite grammatical but are nonsense.
(2) Our implicit knowledge of syntax cannot possibly take the form of a list of sentences. No
such list could be stored in a finite mind, as there are an infinite number of grammatical sentences
in English (or any other human language). It is easy to show this. A list of sentences like the
following:
(93) A potential infinity of sentences
Alice likes Fred

John said that Alice likes Fred
Bill believes that John said that Alice likes Fred
can be extended onward to infinity. The basis of this particular potential infinity, as we will see, is
that clauses can occur inside clauses.
Since syntactic knowledge cannot take the form of a list, we are led to the hypothesis that we
implicitly know a set of syntactic rules; the rules enable us to create novel sentences (a potentially
infinite supply of them) on the spot.49 Just what sort of rules could do this will become clear later
on.
49
This book already argued for the necessity of rules on the basis of huge paradigms in some inflectional
systems, like the 10 trillion verb forms of Shona (p. 33); now we move up to infinity.
(3) Speakers have the ability to recognize and manipulate systematic relations among
sentences. For example, the following set of four sentences:
Bill shaved Fred (active statement)

Did Bill shave Fred? (active question)
Fred was shaved by Bill (passive statement)

Was Bill shaved by Fred? (passive question)
forms a clear pattern that can be duplicated by a speaker of English for an indefinite number of
other sentences.
(4) Sentences are not simply strings of words; they also involve grouping of words into larger
units. The easiest way to show this is with sentences that have two meanings, traceable to two
different groupings of the words:
(94) Four ambiguous sentences, with ambiguity traceable to different word groupings
There were (old)(men and women)

There were (old men)(and women)
They (danced) and (sang the first number)

They (danced and sang)(the first number)
Sue saw (the man)(with the telescope) (she used the telescope to help see him)
Sue saw (the man with the telescope) (the particular man she saw had a telescope)
The ambassador (gave)(the Chinese vases) (...to a friend who likes Chinese vases)
The ambassador (gave)(the Chinese)(vases)
(...when he arrived in China, at the welcoming ceremony)
This is essentially the same point we made for word structure in Chapter 2; see for instance the
discussion in (70)-(69) of the ambiguous word untieable. Structurally ambiguous sentences are far
more common, though, than structurally-ambiguous words.
2. Constituent structure
The first step in developing a syntactic theory is to devise a formal notation for the structure of
sentences. We wish to express the fact that the words of a sentence form groups of various kinds;
that the groups are themselves grouped into larger units, so that a sentence forms a single complex
structure. Linguists normally use a tree notation to do this.
Trees are actually applicable to morphology as well as syntax, so I’ll illustrate the idea with a
morphological example done earlier. On p. 48 above we gave a derivation for the word
unmindfulness, as follows.
(95) Derivation of unmindfulness (repeated)
[mind]Noun stem
[[mind]Noun ful]Adj -ful Rule:

[ X ]N → [[ X ]N ful ]Adj
[un[[mind]Noun ful]Adj] Adj] Negative un- Rule

[ X ]Adjective → [ un [ X ]Adj]Adj
[un[[mind]Noun ful]Adj] Adj]ness]Noun -ness Rule:

[ X ]Adj → [[ X ]Adj ness ]Noun
The output, [un[[mind]Noun ful]Adj] Adj]ness]Noun, can be shown more clearly with the alternative, but
equivalent, tree notation, which is as shown below:
(96) Unmindfulness as a tree
Noun
Adj. ness
un Adj.
Noun ful
mind
As you can see, the tree metaphor is a bit odd, since linguistic trees are drawn upside down,
relative to biological ones.
Definition: any unit in a tree is called a node. The nodes in the tree above are as follows:
Noun, Adj., -ness, un-, Adj. (again), Noun, -ful, and mind .
When you combine a node with all the material you can reach by going “downhill” from that
node, the result is called a constituent. The constituents of the tree just given are:
(97) All the constituents of unmindfulness
(a) Noun
Adj. ness
un Adj.
Noun ful
mind
(b) Adj.
un Adj.
Noun ful
mind
(c) Adj.
Noun ful
mind
(d) Noun
mind
In addition, the elements un, mind, ful, and ness, each of which is at the “bottom” of the tree,
are called terminal nodes. The terminal nodes are constituents, too, though in informal practice
they are often left out of a list of constituents.
If you compare the tree with the bracketed version of unmindfulness given above, you’ll see
that every constituent that isn’t a terminal node corresponds to a bracketed unit.
(98) All the constituents of unmindfulness: shown with brackets
(97)a = [[un[[mind]Noun ful]Adj]ness]Noun

(97)b = [un[[mind]Noun ful]Adj]
(97)c = [[mind]Noun ful]Adj]
(97)d = [mind]Noun
So the two notations are equivalent. For syntax, we’ll mostly use trees, because syntactic structures
tend to be quite a bit more complex than morphological structures, and the tree notation is much
more readily apprehended by the eye. Bracketed notation is convenient for simple cases because it
is compact.
You can name a constituent by pronouncing its terminal nodes in order. So, for example, you
can say things like: “in the word unmindfulness, unmindful is a constituent, and mindfulness is not
a constituent.”
3. Trees in syntax
Drawing the syntactic trees for sentences depends in part on our knowledge of the meaning of
the sentence, and in part on our knowledge of the grammar (the syntactic part of the grammar) of
the language. The idea is to think through the meaning, and locate the syntactic units.
Consider the sentence Sue saw the man with the telescope. This sentence actually has two
meanings: either Sue used a telescope for her observations, or the man was carrying one. Often,
different meanings correspond to different trees, so let us for present purposes assume the meaning
in which the man was carrying the telescope. I will build the tree from the bottom up.
I believe it is pretty intuitive that the telescope is a linguistic unit. We show this with a tree
diagram.
NP
Art N
| |
the telescope
What does this diagram mean? The basic idea is that the is classified as an Article, and telescope as
a Noun, abbreviated N; and the entire constituent is a Noun Phrase, abbreviated NP.50 This NP
can stand alone, for instance as the answer to the question “What did the man have with him?”
Let’s move on to the next larger unit. If we want the answer to “which man”, we could say
(rather tersely):
50
I’ll assume you learned in school how to identify articles, nouns, verbs, helping verbs, adjectives, and
prepositions. If you’d like to review this material, please consult this help page:
http://www.linguistics.ucla.edu/people/hayes/20/resources/CheckingPartsOfSpeech.pdf.
PP
NP
P Art N
| | |
with the telescope
This is a prepositional phrase (PP), with the preposition with (P). The constituent the telescope is
contained within the constituent with the telescope. One can also say it like this: the telescope is
embedded in with the telescope.
We can continue, building up the structure as follows:
NP
PP
NP
Art N P Art N
| | | | |
the man with the telescope
This is a bigger Noun Phrase, involving a man, further identified with the article the and the
Prepositional Phrase with the telescope. It could answer the question, “Which man are we speaking
of?
VP
NP
PP
NP
V Art N P Art N
| | | | | |
saw the man with the telescope
This is a Verb Phrase (VP), whose verb is saw. What we created before can now be seen to be the
object of this verb. The Verb Phrase could answer the question, “What did Sue do?”.
Ultimately we arrive at a structure for the complete Sentence, abbreviated S:

VP
NP
PP
NP NP
|
N V Art N P Art N
| | | | | | |
Sue saw the man with the telescope
Here, we have a subject, in the form of the NP Sue, and a predicate, in the form of the VP saw the
man with the telescope.51
Trees in syntax are also referred to as phrase structure diagrams.
4. Phrase structure and ambiguity
As mentioned above, one of the first and most obvious descriptive benefits of constituent
structure is that it provides a clear account of the ambiguity of many sentences and phrases. For
example, with the tree just given, the meaning we had in mind was that “with the telescope”
identifies the particular man that Sue saw (for example, he was walking down the street holding
the telescope in its carrying case). For the (probably more obvious) meaning that Sue used the
telescope to see the man, we would have:
51
For why we are treating Sue as a full NP, not just an N, see below.
VP
PP
NP NP NP
|
N V Art N P Art N
| | | | | | |
Sue saw the man with the telescope
What is at issue is where “with the telescope” is attached in the tree: is it part of VP or of NP?
We can clarify this concept a bit further with some terminology.
5. Heads and modifiers
Many (but not all) syntactic constituents possess a head. In a Noun Phrase (NP), the head is a
Noun, and similarly the head of a Verb Phrase is a Verb, of a Prepositional Phrase is a Preposition,
and (as we’ll see) of an Adjective Phrase is an Adjective. Intuitively, the head is the “core” of a
constituent, what expresses the essence of its meaning. With just a few exceptions, heads are in
one-to-one correspondence with phrases; every XP has a head X and vice versa.52
You can think of heads either formally (as a property of tree structures), or semantically.
Semantically, the thing denoted by NP is a Noun, where Noun is the head of NP; thus, the tall boy
is a boy. The action denoted by VP “is an” instance of Verb-ing, where Verb is the head of VP.
Thus, in the VP “slowly eat pies”, the action described is an act of eating.
Everything within a phrase that is not the head can be termed a modifier, so long as we are
willing to use the word “modifier” in a rather loose sense. This terminology may differ from what
you learn in later linguistics courses, but it will be useful for our purposes.
Getting the concept of head and modifier right is, in my teaching experience, one of the
trickier parts of learning syntax, so here are some examples.
(a) tall women
The head of this NP is the N women (tall women are instances of women). The word tall is a
modifier, specifying what kind of women.
52
The exceptions, in this book, are that S has no head; and that for Comp, Conj, Art, and Aux there is no phrase
of which they are the head. If you study more syntax, you may encounter theories that rearrange the system of
categories so that the one-to-one correspondence works exceptionlessly.
(b) the book
The head of this NP is the N book; when we say the book we are speaking of a book. The
meaning of the is somewhat elusive, but essentially its purpose is to tell the listener that the
speaker expects that she will be able to know (through overt presence, prior discourse, or
reasoning) which book is being discussed — it says, “You know, somehow, which book I am
talking about”. The “opposite” of the is a, which signals that a book of which the listener is not
necessarily aware is under discussion.
(c) the man with the telescope
The head of this NP is the noun man, and both the article the and the PP with a telescope are
modifiers.
(d) read the book
The head of this VP is the V read; the VP describes an instance of reading, and the book is in
some sense a modifier; it indicates what sort of reading-event took place by specifying one of the
participants.
(d) on Sepulveda
The head of this PP (prepositional phrase) is the P(reposition) on. The meaning or function of
the PP is to express location, and the word on serves to express this core meaning (Sepulveda has
no inherent locative meaning; one can say “Sepulveda is a busy street”, “They are repaving
Sepulveda”, and so on.)
(e) very tall
Looking ahead a bit, this is an Adjective Phrase, with an Adjective head tall, preceded by an
Adverb modifier very.
6. Parsing sentences
The starting point for syntactic analysis of a language is to parse (provide a parse for; find the
tree structure of) a variety of sentences. In the theory taught here, the basic principles of parsing
are quite simple.
(99) Principles of Parsing
a. For the phrases NP, VP, PP, AP, locate the head, and include all its modifiers in the same
phrase.
b. Sentences (including sentences inside sentences; see Chapter 1) are assumed to consist of a
subject (which is an NP), and a predicate (which is a VP.)
Just as in traditional school grammar, the subject indicates what the sentence is about, and the
predicate says something about the subject.
The hard part seems to be to make sure you find all the modifiers of each head, and include
them in the phrase of which it is the head; so exercise care here.
Returning to the two structures of Sue saw the man with the telescope, the crucial distinction is
what the PP with a telescope is a modifier of: in one reading, it modifies man (that is, it specifies
which man), and thus belongs as part of NP; in the other reading, it modifies see (that is, it
specifies what kind of act of seeing took place), and thus belongs as part of VP.
Study Exercise #12
Diagram (that is, parse) both readings of the sentence The ambassador gave the Chinese
vases. The answer is given on the next page.
This ambiguity is slightly more complex than the previous one, since it hinges not just on tree
structure but also on the fact that Chinese can serve as either a noun (as in the Chinese, meaning
“the Chinese people”) or an adjective. 53 With both readings, we can get two parses, as follows:
(That is, gave the Chinese vases to someone, unspecified. Chinese is an adjectival modifier of
vases.)
(That is, the ambassador gave vases to the Chinese, as a diplomatic gift. The noun Chinese is
the head of a NP; gave as head of VP takes two modifying NP, one the recipient of the giving, the
other the thing given.)
The two “versions” of the word Chinese are the result of morphological conversion. Chinese is
53
fundamentally an Adjective, but from it is derived the Noun Chinese, by the conversion rule given in Chap. 4.
[ X ]Adj → [[ X ]Adj]Noun meaning: “person who is X”
A curious property of the rule is that the output can only be used in the plural; hence it should also attach the
inflectional feature [Number:plural] to the morphosyntactic representation of its output.
——————————————————————————————————————
The example illustrates the point that differing parses of the same string are only one source of
ambiguity in language. To mention some others in passing:
• Homophonous words: We walked to the bank.

• Multiple quantifier words: Three men were examined by each doctor.
• Phonological merger: We were patting/We were padding, which are identical in
many North American English dialects
Study Exercise #13
Here is one more ambiguity with its two parses (check that you know the answer before you
look). The sentence is: The hungry bear fishes, and the answer is on the next page.
The far more likely reading uses the noun meaning of bear, making it the head of an NP;
fishes is the head of VP.54
NP VP
Art AP N V
| | | |
The A bear fishes
|
hungry (, scooping the salmon with its paw.)
In the less likely reading, hungry is treated as a noun,55 and bear as a verb:
VP
NP NP
|
Art N V N
| | | |
The hungry bear fishes (holding them in both hands because they are slippery)
——————————————————————————————————————
7. Possessive constructions
People are taught in school that adjectives are words that modify nouns. I think this is
basically true; provided that you don’t say they are the only words that modify nouns; there are
quite a few other possibilities.
One very common noun modifier is the possessive construction, as in the tall student’s books.
The tall student’s modifies books, but in its internal structure it looks just like an NP. (except for
the extra material ’s). It couldn’t possibly be an Adjective; an Adjective is a word, but The tall
student’s is a whole phrase.
54
You may be wondering why we bother with a VP symbol when there is no modifier present; see
below on phrase structure rules for some justification.
55
Hungry as a noun is, just as with Chinese, derived in the morphology by the conversion
rule [ X ]Adjective → [[ X ]Adjective]Noun.
We will assume here that the tall student’s is in fact an NP, and it sits inside the larger NP the
tall student’s books, modifying the head books (i.e., it says in effect, “whose books?”). Thus the
structure is:
NP
NP
|
Art AP N N
| | | |
the A student’s books
|
tall
There’s a debt to pay here: where does the ’s morpheme come from, and where should it sit in
the tree? We’ll cover this in detail later on; the brief answer is that the ’s is inflectional
morphology, genitive case. What we will need is a way to relate the inflectional morphology to the
syntax.
8. Conjoined structures
Conjunctions like and and or are fairly straightforward: we’ll assume that they link together
two identical units, forming a large unit of the same kind. Thus the boy and the girl is
NP
NP NP
Art N Conj Art N

| | | | |
the boy and the girl
We say that the two NPs the boy and the girl are conjoined by and into a larger NP, the entire
structure. Similarly, Sue chopped wood and made syrup has a VP made of two conjoined VP’s:
VP
VP VP
NP NP NP
| | |
N V N Conj V N
| | | | | |
Sue chopped wood and made syrup
Several other categories, including Adjective Phrases, PP, and S, can participate in this
construction: examples of these (same order) are very tall but quite thin; over the river and
through the woods; I like coffee and you like tea.
Study Exercise #14
Parse the king and the queen’s throne in both meanings.

NP
NP
NP NP
Art N Conj Art N N

| | | | | |
the king and the queen’s throne [= the throne of the king and the queen]
NP
NP
NP NP
Art N Conj Art N N

| | | | | |
the king and the queen’s throne [= a person (the king) and a throne]
————————————————————————————————————
9. Terminology for trees
Here is some terminology that will be useful in referring to trees. I will use the tree in (100) to
illustrate the various terms:
(100) A tree used to illustrate terminology for constituent structure
VP
NP
PP
NP NP
Art N V Art N P AP
| | | | | | |
the man bought a book about A N
| |
Chinese vases
(101) Defn.: dominate
Node X dominates node Y if you can get from X to Y by going “downhill” in the tree and
never uphill. For example, S dominates everything in the tree; the NP on the right dominates
an A, an N, and the words Chinese and vases. The NP on the right does not dominate the VP,
nor does it dominate the word man.
(102) Defn.: constituent
A constituent is a node, plus all the nodes that it dominates.
Constituent was already defined earlier, but (102) is a more careful definition. As noted above, one
usually refers to constituents by the words they contain. Thus one can say that the following:
the man a book about Chinese vases bought bought a book about Chinese vases
are constituents (in sentence (100)). Note that sequences like
bought a book
the man bought a book
a book
are not constituents in sentence (100), though they could be in other sentences.56
(103) Defn.: directly dominate
X directly dominates Y if Y is “one node down the tree” from X. Thus the NP a book about
Chinese vases directly dominates the PP about Chinese vases.
(104) Defns.: daughter, mother
If X directly dominates Y, then X is Y’s mother and Y is X’s daughter.
(105) Defn.: sister
Two daughters of the same node are sisters.
We’ve used the term head casually, but can now give the tree-based formal version:
(106) Defn.: head
The head of an XP is the X that it directly dominates.
56
As mentioned above, this is one of the principal difficulties in parsing; that is, not to get distracted
by mere “potential” constituents like these, and instead choose complete constituents.
Thus, the head of a VP is the V that it directly dominates; the head of a NP is the N that it
directly dominates, and so on. Therefore, the head of the NP a book about Chinese vases is book;
the head of the VP bought a book about Chinese vases is bought; the head of the AP very tall is the
Adjective tall.
10. Becoming a skilled parser
Of the practical skills needed to do linguistic analysis well, parsing sentences is probably #1.
Many things we will do with grammar and meaning depend on having the right parse. So it’s
worth practicing your parsing, especially if it doesn’t come to you naturally. This section offers
some principles that will help you become a fluent and reliable parser.
The first principle is:
(107) Start with the obvious constituents you can get by proceeding “from the bottom up.”
By “bottom up”, I mean, first of all, to label each word for it part of speech. So if you trying to
parse these two sentences:
(108) Two constrasting sentences to illustrate parsing
A. Alice owns the book on the table B. Alice placed the book on the table
it makes sense to begin with the very low-level structure that assigns each word to its part of
speech:
N V Art N P Art N N V Art N P Art N

| | | | | | | | | | | | | |
That’s the very bottom. But moving upward, you could then start grouping the words into
bigger phrases, like this:
P P
NP NP NP NP
| |
N V Art N P Art N N V Art N P Art N
| | | | | | | | | | | | | |
Observe what I’ve done here. Nouns have to belong to Noun Phrases, and there is nothing else in
sight that could plausibly be part of the same Noun Phrase as Alice, so we’ve got an NP node more
or less for free. The table is plainly a simple NP, with the common Article + Noun structure, and
moreover it is the object of the preposition on, so we have a Prepositional Phrase, too.
The second principle of accurate parsing requires that you think consistently about heads, and
about grouping modifiers into the same phrase as their heads. For example, in diagramming
sentence (108)A above, the crucial question is what on the table belongs to. If you think about the
meaning of the sentence, it is clear that on the table modifies book; that is, it specifies which book
is under discussion. The rest of the reasoning goes like this: ‘book’ is a noun; it must be the head
of a NP; anything that modifies it (namely ‘the’ and ‘on the table’ must be its sister; therefore the
full NP is the book on the table.
NP
PP
NP NP
|
N V Art N P Art N
| | | | | | |
Alice owns the book on the table
From there on, the diagramming is straightforward; you just need a VP (verb and object) plus
the whole sentence:
(109) Final structure for “Alice owns the book on the table”
VP
NP
PP
NP NP
|
N V Art N P Art N
| | | | | | |
Note that in this sentence, the book is not an NP; it is only part of an NP because the head is
missing one of its modifiers. More on this below.
Suppose this time that you are diagramming (108)B above, Alice placed the book on the table.
In this case, the PP on the table modifies the verb placed (it indicates the target of placing).
Accordingly it must be the sister of the verb within the VP. The book is left as an NP on its own.
(110) Final structure for “Alice placed the book on the table”
Let us now codify more carefully the principle we’ve been following:
(111) The modifiers of a head must be sisters to the head.
In Alice owns the book on the table, the PP on the table tells you what book it is; it modifies book,
it must be a sister of book, and in (109) this is so. In Alice placed the book on the table, the PP on
the table tells you the destination of the act of placing; it modifies place, it must be a sister of
place, and in (110) this is so.
We can illustrate the principle (111) with a more complex sentence that has has two PP’s.
They get placed in different positions according to what they modify:
(112) A sentence with PP’s at two different levels
The PP to the president says what kind of letters are being discussed; it modifies letters and
must be a sister of letters within NP. The PP in the wastebasket specified the destination of the act
of putting; it modifies the verb put and must be a sister of put within VP.
Here is another point that sometimes troubles beginners learning to parse:
(113) Just because some word sequence is a constituent in some other sentence, it is
not necessarily a constituent in the sentence you are trying to parse.
Look at example (109) (Alice owns the book on the table), and think about the sequence owns the
book. It is unquestionably true that in all sorts of English sentences, owns the book is a constituent
(example: in Alice owns the book, owns the book is a VP). But it is not a constituent in (109), for it
is incomplete. We could say the very thing about the sequence the book. In (112), the president in
the wastebasket is not a constituent, though it would be in some other (probably insulting)
sentence.
Here is one other handy hint for parsing. For long sentences, once you’ve done the low-level
constituents according to principle (107), it’s often helpful to parse English sentences going
backwards, starting with the end of a sentence.57
Why so? It has to do with a property of English called “right-branchingness”. When a constituent
57
has two daughters, rather often the daughter on the left is a single word, whereas the daughter on the right has
11. Constituency tests
The discussion so far because it consists simply of directions to you, the student, on how to
parse sentences. This is arbitrary unless it can be shown that the structures we’re creating have
some scientific validity.
One sort of evidence involves what linguists call “it-cleft sentences:”
(114) Some it-cleft sentences
It’s Bill that they don’t like.

It’s on Mulberry street that they live.
It was the flowers that Mary sent to Bill.
Such sentences are clearly related to simpler sentences, such as
They don’t like Bill.

They live on Mulberry Street.
Mary sent the flowers to Bill.
We can express the relation between simple sentences and it-cleft sentences by writing a
syntactic rule (we’ll cover this more formally later on):
(115) It-Clefting Rule
To form an it-cleft sentence, take a simple sentence and perform the following operations on
it:
1. Add it and an appropriate form of the verb be to the beginning.

2. Find a NP or PP constituent inside the sentence and reorder it so that it directly follows be.
3. Insert the word that just after the reordered NP or PP.
You can see for yourself that the cleft sentences cited above are derived from the corresponding
simple sentences.
The crucial part of the It-Clefting Rule is where it says “find an NP or PP constituent”. It
predicts that if we apply It-Clefting to a sequence of words that is not a constituent, the result
should be ungrammatical. If you look at the tree drawn earlier in (110) for Alice placed the book
on the table, you will see that the book on the table is not a constituent; it is two.
some internal structure. When you have a right-branching system, right-to-left implies bottom-up. Japanese,
which is mostly left-branching, is probably easier to parse left-to-right.
The analysis thus correctly predicts that if we attempt to carry out It-Clefting with this
sequence of words, the result will be ungrammatical:
*It was the book on the table that Alice placed.
On the other hand, in (109) Alice owns the book on the table, the sequence the book on the
table is a constituent:
VP
NP
PP
NP NP
|
N V Art N P Art N
| | | | | | |
Therefore, It-Clefting produces a grammatical result:
It’s the book on the table that Alice owns.

It-Clefting is often called a constituency test; it is used to verify the correctness of a proposed
syntactic parse.
Neither the principle (111) (modifiers form constituents with their heads), nor the rule of It-
Clefting can be assumed in advance to be correct. We can only test them out against facts. The
more correct predictions they make, the greater is our confidence that they are true. If we want to
be really confident about these principles, we must test them out against a much larger set of facts,
and we will carry out part of this task later on.
Study Exercise #15
(a) In They sent the king to Barataria is the king to Barataria a constituent? Support your answer
with evidence from It-Clefting.
(b) Replace to with of in the same sentence and answer the same question.
(c) What are the grammatical it-clefted versions of Alice put the book on the table? (There are
about four).
Answers on next page.

Answers to Study Exercise #15
(a) In They sent the king to Barataria’, the sequence the king to Barataria is not a constituent.
If it were, the rule of It-Clefting could apply to it, producing the sentence
*It was the king to Barataria that they sent
We can explain the ungrammaticality of this sentence by supposing that ‘the king to Barataria’ is
not a constituent.
(b) We have seen that It-Clefting can move only constituents. Since when we apply It-Clefting
to They sent the king of Barataria we get a grammatical sentence:
It was the king of Barataria that they sent.
we infer that the king of Barataria must be a constituent.
(c)
It was Alice that put the book on the table. (clefting the NP Alice)
It was the book that Alice put on the table. (clefting the NP the book)
It was on the table that Alice put the book. (clefting the PP on the table)
It was the table that Alice put the book on.58 (clefting the NP the table)
—————————————————————————————————————
12. Grammars for syntactic structure I: Phrase structure rules
The discussion so far has been about structures; we now turn to the grammars that are
responsible for these structures.
Phrase structure is language specific. To be sure, it does appear to be true that all languages
have S and NP. However, the VP, AP, and PP that we have in English appears to be missing in
certain languages,59 and moreover the order of the constituents of a phrase varies from language to
language. Perhaps other languages include phrase types that English lacks.
For the “stranded preposition” on in this sentence, see Chapter 3 above. Normative feeling in
58
English (Chapter 3) is that you should not leave prepositions at the end of a sentence.
59
It seems pretty clear that all languages have NP and S. It’s less clear that there is a Verb Phrase in
languages where the subject comes between the verb and the object (for example, Verb-Subject-Object order,
as in Malagasy). AP clearly cannot exist in a language without adjectives, and PP cannot exist in a language
without prepositions (or postpositions). Korean has been claimed to be an adjectiveless language; like most
adjectiveless languages it uses verbs for the same purpose; see
http://webpages.acs.ttu.edu/minjkim/KimHUMIT02.pdf. Klamath (N. California) has been claimed to be a
language without prepositions or postpositions; see
https://www.academia.edu/3876363/Adposition_as_a_non-universal_category.
Persian (also called Farsi) is distantly related to English and has a similar inventory of phrasal
categories in its syntax. But the order of elements within constituents is often different, as is
illustrated by the following English sentence and its literal Persian translation:
S S
VP VP
PP NP PP
NP NP AP NP
| | | |
Art AP N V P N Art N A P N V
| | | | | | | | | | | |
That A student went to America n dnedu-je xub be mrik ræft60
| that student good to America went
good
Because of this, every language must include rules that specify its grammatical word orders.
In the theory covered in this text, the rules that are responsible for word order are called
phrase structure rules. Some examples of phrase structure rules are as follows:
English: NP → Art AP N Persian: NP → Art N AP

VP → V PP VP → PP V
You can read the rules as follows: “an NP may consist of the sequence Art, AP, N.”
There is another way of interpreting phrase structure rules. If we have a complete set of them
for a given language, we can think of the set of rules as an abstract machine that generates
syntactic structures. For example, assume for the moment the following (incomplete) set of phrase
structure rules for English:
S → NP Aux VP (Aux = “helping verb”, like can, will, be, etc.)

NP → Art AP N
VP → V NP
AP → A61
In this respect, the phrase structure rules are like the rules of inflectional morphology given
earlier: given a starting point, they generate a sentence. For inflectional morphology, the starting
point is a stem with its morphosyntactic representation. For syntax, the starting point is a single
symbol, such as NP or (most often) S, which designates the category that we wish to generate.
60
IPA symbols: ɑ = somewhat like ah; ʃ = sh, dʒ = j, x as in ch of German Bach, æ = the vowel of
cat.
61
This rule looks trivial right now—we’ll beef it up a bit later by allowing Adverbs and Preposition
Phrases.
Here is the procedure.
(a) Provide the rules with the symbol S (or NP, or whatever) to start out with.
(b) Whenever a symbol appears in a tree that is found on the left side of the rule, give that
symbol daughters according to what the rule says;
(c) Repeat step (b) until you no more rules can be applied.
If we start with S, the derivation would look this like:
1. Starting point:
2: apply S → NP Aux VP
NP Aux VP
3: apply NP → Art AP N
NP Aux VP
|
Art AP N
4: Apply VP→ V NP
NP Aux VP
|
Art AP N V NP
5: Apply NP→ Art AP N (again)
NP Aux VP
|
Art AP N V NP
|
Art AP N
6: Apply AP→ A
NP Aux VP
|
Art AP N V NP
| |
A Art AP N
7: Apply AP→ A (again)
NP Aux VP
|
Art AP N V NP
| |
A Art AP N
|
A
All that remains is to insert actual words into the tree (a process called lexical insertion), and
you get sentences:
NP Aux VP
|
Art AP N V NP
| |
A Art AP N
| | | |
The short lumberjack will chop the A tree
|
tall
or, with different choices for lexical insertion:
The sleepy student might ignore the noisy alarm.

The green idea will paint the blue intellect.
Note that these sentences will not necessarily be sensible; a grammar only tells us that they are
grammatical.
A note on applicability of phrase structure rules: we saw that in inflectional morphology, the
rules apply in a rigid order, like machines on an assembly line. Rules of word formation are more
“opportunistic”; they can reapply freely when they get the chance (see Chapter 2, §18). Phrase
structure rules are likewise opportunistic; they keep applying until there are no more nodes to be
expanded.
The phrase structure rules just proposed are obviously primitive, since they generate only one
single structure. We can improve the rules by observing that some of the daughters introduced by a
rule are optional. In particular, the NP rule has to introduce a N, but it doesn’t have to introduce an
Art or an A. The notation in this approach for expressing optional elements is parentheses:
S → NP (Aux) VP
NP → (Art) (AP) N
VP → V (NP)
AP → A
These more flexible rules can provide the syntactic structures of sentences like these:
The lumberjack will chop the tree

Frogs will eat flies
Fish can see
Those students read books.
Sue won
and so on. (Diagram these if it is not obvious what the structure is.)
We can also make our AP rule less trivial, so that Adverbs are allowed.
AP → (Adv) A
For instance: very tall.
13. Curly brackets for “or”
One other complication in the notation for phrase structure rules. We find that a NP can begin
either with an Article or with a possessive NP, but not both.
Article:
the book, a book, this book, those books
NP:
Fred’s book, the king of England’s book, my book
not both:
*the Fred’s book, *the king of England’s this book, *those my books
Here is a simple way to account for this: we use curly brackets in the rules to mean “one or
the other, but not both” (logicians call this “exclusive or”). The basic NP phrase structure rule for
English comes out something like this:
Art 
NP → NP (AP) N (PP)
 
A AE
This means that you can start out an NP with an Article, or an NP, then continue with the rest
(optional Adjective, obligatory Noun, optional PP). Examples of each type:
the long book about linguistics (beginning with Article)

the king’s long book about linguistics (beginning with NP)
14. Phrase structure rules for English sentences given so far
As we continue through syntax in this text, we will gradually build up an ever-improving

grammar of phrase structure rules. Just to catch us up so far, I believe the following set of rules can
generate most of the examples given in this book so far, as well as the sentences in the Study
Exercises.
(116) Phrase structure rules for English: Version I
S → NP (Aux) VP
Art 
NP → NP (AP) N (PP)
A AE
NP → Pronoun
VP → V (NP)(NP)(PP)
PP → P NP
AP → (Adv) A
S → S Conj S
NP → NP Conj NP
VP → VP Conj VP
PP → PP Conj PP
AP → AP Conj AP
V → V Conj V
15. Parsing: using the phrase structure rules as a guide
Once you’ve got a grammar like this to work with, then in principle it becomes easier to parse
sentences—any particular set of rules represents a claim about the inventory of phrase types a
language allows, and thus constrains what kind of structures you can set up. Thus:
(117) When diagramming sentences, make sure every structure you set up is licensed by
the rules.
More specifically, every mother node must appear on the left side of the arrow in a phrase
structure that allows for the set of daughters that the mother node dominated. For example, you
can’t set up an NP whose daughters are N AP, unless there is a phrase structure rule that specifies
this sequence (either directly, or by leaving out parenthesized material). Thus, in parsing, you can
be guided to an answer by both the meaning of the sentence and by the rules of the grammar.
Example: here is an incorrect structure (according to our grammar) for the king of England:
*NP
PP
NP NP
Art N P N
| | | |
the king of England
You can tell it’s not right because the grammar in (116) contains no rule that permits NP to
dominate NP followed by PP.62 This is an easy principle to apply; you just need a printed copy of
the grammar on your desk, then you can check every single node in your tree to see if it is legal.
There actually is one way you can legitimately diagram a structure that the grammar doesn’t
allow—namely, change the grammar. In other words, you have to say something like “This
sentence shows that our grammar was wrong, and has to be fixed like this [offer substitute rules
here].” In this book I have included only sentences that can be parsed with the grammar given so
far at that point in the text. But of course real life is different: a grammar that could parse all of
English would be quite large and a big challenge to create. Expanding the grammar so it can cover
more of the legal sentences of a language is something linguists do all the time when they work
with language data.
16. Further details of our current grammar
A few of our phrase structure rules in (116) need clarification.
16.1 Pronouns
The phrase structure rule of (116) that introduces Pronouns is very simple:
Note that there is a rule NP → NP  (AP) N (PP). But it won’t help, because it requires there to
62 Art
 
be an N daughter.
NP → Pronoun
Pronoun, appearing in trees, is often abbreviated as Pro. Thus:
VP
NP NP
| |
Pro V Pro
| | |
She saw him
The reason to have a separate rule for pronouns is that, unlike nouns, they do not admit
modifiers, except in special circumstances we’ll defer for now.63 This is one reason to give them
their own phrase structure rule, rather than just calling them a kind of Noun. The other reason is
that, later on, we will need rules of semantic interpretation that indicate what the pronouns refer to,
and these rules need to identify the pronouns.
Incidentally, pronouns in English are unusual in that they are inflected for case. English has a
three-way case system, with Nominative, Objective, and Genitive. Objective covers what in many
other languages (including English, centuries ago) was Accusative or Dative. Different authors
will give different names to these cases.
(118) English pronouns in outline form
Nominative Objective Genitive

Singular 1 I me my
2 you you your
3 he/she/it him/her/it his/her/its
Plural 1 we us our
2 you you your
3 they them their
Part of what a grammar must do is ensure that the correct case form of each pronoun is used in the
right context; we will turn to the sort of rules that are needed later on.
16.2 Aux
“Aux”, meaning “auxiliary verb”, is the “helping verb” taught in school. In our phrase
structure rules, it is the optional second daughter of S (S → NP (Aux) VP). Here is a list of auxes:
63
Examples: Poor me, a “frozen” memorized expression; He who dares to go…, with a relative
clause.
“Modal” verbs: can, could, shall, should, may, might, will, would
Example: I can go.
Forms of have: have, has, had
Example: I have gone.
Forms of be: be, am, are, is, was, were
Example: I am going.
You can see that the choice of Aux also determines the inflectional morphology of the following
verb—this involves rules we haven’t yet covered.
Be aware that have and be can serve as either Auxes or main Verbs. Thus:
He is having a fit
involves be as an Aux and have as a main Verb.
He has been President.
has have as an Aux and be as a main Verb.
Study Exercise #16
Parse. Give one parse for each meaning. The number of parses is given in parentheses.64 I
suggest you keep a printed copy of the Version I grammar ((116) above) next to you as you parse.
a. We have captured the turtle with the net. (2)

i. ... it was quite interesting because most turtles don’t carry a net.
ii. ... we felt sad to see the noble reptile caught in the webbing.
b. The great import sinks. (2)

i. ... because, like many imports, it weighs a lot.
ii. ... because powerful people feel that their sinks ought to come from overseas.
c. We offered the mighty bruins. (2)

i. ... because the mighty bruins were a good choice to offer them.
ii. ... because mighty people like to keep bruins in their personal zoo
d. UCLA and USC’s departments (2)

i. the departments that belong to UCLA and to USC
ii. a university and a batch of departments
e. We robbed the King of England of his money. (1)
64
Through maybe you can think of more …
f. We despise the King of England and France (2)

i. We despise the person who is the king all at once of two countries.65
ii. We despise both a person and a country.
g. We despise the King of England and of France (1)

i. We despise the person who is the king all at once of two countries.
h. Bill and Alice and Sue (2)

i. a group consisting of Bill and Alice, and also a single person, Sue
ii. a single person, Bill, and also a group consisting of Alice and Sue
i. the cover of Bill’s book about the very dry climate of Australia (1)
j. They might cook and serve the vegetables. (2)

i. what they do to the vegetables is cook and serve them
ii. They do two things: cook, and serve the vegetables
k. a long and very unpleasant episode (1)
m. Bill’s mother’s brother’s shaggy dog (1)
n. We sent the ambassador to Venezuela to Colombia. (1)
o. We sent the ambassador to Venezuela (2)

i. We needed somebody to send, so we picked the ambassador to Venezuela
ii.Venezuela is where we sent the ambassador
p. We have cooked her turnips. (2)

i. It’s her turnips that we cooked.
ii. We cooked turnips for her.
Answers begin on the next page.
65
This happened once, more or less, and the Hundred Years’ War was the result.
a. We have captured the turtle with the net. (2)

i. ... it was quite interesting because most turtles don’t carry a net.
ii. ... we felt sad to see the noble reptile caught in the webbing.
b. The great import sinks. (2)

i. ... because this wonderful import weighs so much
ii. ... because powerful people feel that their sinks ought to come from overseas.
c. We offered the mighty bruins. (2)

i. ... because the mighty bruins were a good choice to offer them.
ii. ... because mighty people like to keep bruins in their personal zoo
d. UCLA and USC’s departments (2)

i. the departments that belong to UCLA and to USC
ii. a university and a batch of departments
e. We robbed the King of England of his money. (1)

f. We despise the King of England and France (2)

ii. We despise both a person and a country.

g. We despise the King of England and of France (1)

h. Bill and Alice and Sue (2)

i. a group consisting of Bill and Alice, and also a single person, Sue
ii. a single person, Bill, and also a group consisting of Alice and Sue
i. the cover of Bill’s book about the very dry climate of Australia (1)
j. They might cook and serve the vegetables. (2)

i. what they do to the vegetables is cook and serve them
ii. they do two things: cook, and serve the vegetables

k. a long and very unpleasant episode (1)
m. Bill’s mother’s brother’s shaggy dog (1)

n. We sent the ambassador to Venezuela to Columbia. (1)
o. We sent the ambassador to Venezuela (2)

i. We needed somebody to send, so we picked the ambassador to Venezuela
ii. Venezuela is where we sent the ambassador
p. We have cooked her turnips. (2)

i. It’s her turnips that we have cooked.
ii. We cooked turnips for her.
———————————————————————————————————————
16.3 Complementizers and subordinate clauses
Much of the most intricate syntax arises when one “puts a sentence inside a sentence”; that is,
when one uses a subordinate clause. This showed up in Chapter 1 when we looked at the
patterning of each other. Thus, *[ John and Bill think [ I like each other ]S ]S is impossible, because
each other is allowed to refer only to Noun Phrases that are within the smallest clause containing it
— in this case, the clause [ I like each other ]S. Subordinate clauses often occur when the verb of
the main clause is a verb of saying or belief — the subordinate clause serves to express the content
of whatever is said or believed. With the notions of syntax we’ve developed so far, we can now be
more explicit about subordinate clauses than we were in Chapter 1.
To analyze subordinate clauses, we need to provide a slot in phrase structure for the
grammatical words that often introduce them—that in sentences like:
I think that [ John and Bill like each other ]S
There is also for, as in:
I would prefer for [ John and Bill to like each other]S.
Such words are called subordinating conjunctions in traditional terminology. Linguists use the
slightly shorter term complementizer,66 abbreviated Comp. Other complementizers include if,
(al)though, when, whether, and some others we’ll mention later.
For why the complementizers is so called: the subordinate clause is sometimes classified as a “sentential
66
complement”, meaning it functions as the object of the verb. A complementizer renders a simple S suitable for
appearing as a complement; the bare sentence is made into a possible object.
With this apparatus, we can set up rules like these (I’m omitting optional material; see below
for the full rules):
VP → V CP
CP → Comp S
CP is the category that provides the syntactic “slot” for the complementizer. Here is an example
sentence that can be generated by these rules:
S Derived by: CP → Comp S
VP
CP
VP
NP NP NP
| | |
N V Comp N V N
| | | | | |
16.4 Phrase structure rules for subordinate clauses
Subordinate clauses in English most often occur the last constituent of the VP, indicating what
was said or thought. Here are some examples:
(119) Four sentences with subordinate clauses
We said [ that we were going ]CP

We told Alice [ that we were going ]CP
We gave Bill notice [ that we were going ]CP
We sent word to Jane [ that we were going ]CP
From these sentences, you can see that the Verb Phrase can, in addition to its subordinate
clause, include one or two NP objects and a PP, all of them preceding the CP. Thus the phrase
structure rule needed is something like this:
(120) A phrase structure rule for VP that incorporates CP
VP → V (NP)(NP)(PP)(CP)
Study Exercise #17
Parse the four sentences given in (119) above. Answers on next page.
VP
CP
NP NP VP
| | |
Pro V Comp Pro Aux V
| | | | | |
We said that we were going
VP
CP
NP NP NP VP
| | | |
Pro V N Comp Pro Aux V
| | | | | | |
We told Alice that we were going
VP
CP
NP NP NP NP VP
| | | | |
Pro V N N Comp Pro Aux V
| | | | | | | |
We gave Bill notice that we were going
VP
CP
PP S
NP NP NP NP VP
| | | | |
Pro V N P N Comp Pro Aux V
| | | | | | | | |
We sent word to Jane that we were going
———————————————————————————————
The rule for CP needs to let Comp be optional, since we have sentences like (121):
(121) We said we were going.
The conditions under which the Comp can be left out are somewhat complicated and will not be
covered here.
Note that the tree for given below must have a “vacuous” CP node, at least under the phrase
structure rules we are using, since with those rules only CP, not S, can be a daughter of VP.
VP
CP
|
S
VP
NP NP NP
| | |
Pro V Pro V N
| | | | |
We said we like Fred
17. Recursive application of phrase structure rules
I mentioned above (p. 77) that the speaker’s knowledge of syntax is large but finite (that is, it
fits somehow encoded in a single brain), yet permits the creation of an infinite number of
sentences. The following partial list was meant as a demonstration:
Alice likes Fred

Bill believes that John said that Alice likes Fred
…etc.
We can now examine the cause of this infinite property. It results, for the most part, from a
particular property of phrase structure rules, namely that they permit application in loops. In (122),
I demonstrate one of these loops, taken from the phrase structure rules already given. That is, by
looping through application of phrase structure rules given above in (116) and (120), we can
generate structures as large as we please.
(122) A loop consisting of three phrase structure rules
S → NP (Aux) VP
VP → V (NP) (NP) (PP) (CP)
CP → (Comp) S
If we employ loop (122) in deriving a sentence and lexically insert appropriate words, we can
generate a sentence as long as we like:
This is because there is an infinite number of places where we could stop the loop. Thus there are
an infinite number of possible sentences that the grammar can generate.
In this sense, the phrase structure rules of (116) and (120) may be considered recursive; that
is, their application “recurs” when they apply in loops to create structures of unlimited length.
As far as is known, every human language allows an infinite number of sentences. In every
case, the principal reason is the same: the phrase structure rules of all languages contain recursive
loops, which allow infinitely long syntactic trees to be generated. The recursive loop of phrase
structure rules is the device that allows a finite number of rules to generate an infinite number of
structures.
Study Exercise #18
The phrase structure grammar in (116) (p. 106) has several other loops in it. The loop in (122)
is a three-rule loop; find a two-rule loop and a one-rule loop and for each one give an example of
the long structures they can generate. Draw trees derived using the loop.
a. Two rule loop
Art 
NP → NP  (AP) N (PP)
 
A AE
and
PP → P NP
form a two-rule loop (see boldface items)
example: books about topics in linguistics
b. Any of the phrase structure rules with a conjunction forms a one-rule loop. For the rule
V → V Conj V
you can get a ambiguous structures like danced and sang and acted.
Art 
Also, the rule NP → NP  (AP) N (PP) forms a one-rule loop. This produces strings of
 
A AE
genitives, like Sue’s mother’s brother’s dog.
18. Relating syntax to inflectional morphology
We are now in a position to tie together our two course units so far (morphology and syntax).
The crucial notion is the morphosyntactic representation, covered in Chapter 2. You can think of
the morphosyntactic representation as the means by which the syntax communicates essential
information to the inflectional morphology.
The features in a morphosyntactic representation can have three sources.
18.1 Inherent features
First, some features of a morphosyntactic representation are inherent. They are properties of
particular words or stems.
As noted in Chapter 3, it is normal to use the word lexicon to refer to the speaker’s mental
dictionary; their store of memorized stems, words, and other entities.67 Since a feature like
[Gender] on nouns is memorized, it must be listed in the lexicon. Here are three examples of
inherent inflectional features.
67
We also memorize a great number of word sequences, often called idioms.
I. Gender in German. The German word Messer (knife) is inherently, and arbitrarily, neuter.
Its lexical entry must look something like this:
Messer [Gender:Neuter]
That is, attached to Messer is a partial morphosyntactic representation that indicates that Messer is
a neuter noun.
II. The English pronoun his is inherently [Case:Genitive,Gender:Masculine].
his [Case:Genitive,Gender:Masculine]
III. All nouns derived by the English word formation rule [ X ]Adj → [[ X ]Adj]Noun (example:
The French care a lot about food) are inherently [Number:Plural].68 This is also true for a small
number of words for “pairlike” things, such as trousers, scissors, and so on, which must be
lexically listed as [Number:plural].
18.2 Speaker-selected features
Other features of the morphosyntactic representation are meaningful; they represent choices
made by the speaker, as part of the meaning of what they are trying to say. When we say book in
English we are implicitly conveying the partial morphosyntactic representation [Number:singular],
and when we say books we are similarly conveying [Number:plural]. (This raises the question of
how linguistic entities bear meaning, a question addressed in Chapter 9.)
18.3 Features derived by syntactic rules
The remaining source for the features in morphosyntactic is syntactic rules. These attach the
features that depend on what else occurs in the tree. Two important kinds of rule in syntax are rules
of case marking and rules of agreement
19. Case marking
19.1 Genitive case in English
Genitive case in English is the case that we spell with the suffix -’s. Semantically, it denotes
the relationship of possession. To derive it, we need both a syntactic case marking rule and a
morphological suffixation rule.
Here is a tree to serve as an example. The phrase structure rules given so far generate this:
68
Thus, a fully explicit version of the conversion rule would actually attach a partial morphosyntactic
representation: [ X ]Adj → [[ X ]Adj ]Noun,[Number:Plural].
(123) A sample English configuration that requires genitive case
Art
Choices employed: for NP1: NP → (NP  )(AP) N (PP) (CP)
 
E
Art
for NP2: NP → (NP  )(AP) N (PP) (CP)
 
E
The syntactic rule of case marking that is needed is as follows:
(124) Genitive Case Marking (English)
In NP1
NP2
where NP2 is leftmost in NP1, assign the feature [Case:Genitive] to the rightmost word of NP2.
Genitive Case Marking can be applied to the above as shown. I use dotted lines to show what
part of the rule matches up to what part of the form
(125) Applying the rule of Genitive Case Marking
In NP1
NP2
Genitive Case Marking can be applied to the above as shown. I use dotted lines to show what
part of the rule matches up to what part of the form
NP
NP
|
Art AP N A N
| | | | |
the A student heavy books
| [Case:Gen]
tall
where [Case:Genitive] is the morphosyntactic representation of student.
That is the most complicated part. Once the syntactic rules have placed the feature [Case:Genitive]
on the word student, then we move on to the inflectional morphology. Here, it is straightforward to
get the suffix in place, with an ordinary rule of inflectional suffixation, as follows:
Inflectional morphology: Genitive Inflection (English)
Suffix -s if [Case:Genitive].
Thus the full NP the tall student’s is the combined result of syntactic and morphological rules.
19.2 Where to inflect? Edges vs. heads
Case is fundamentally a property of NP’s, reflecting their grammatical role in a sentence.

However, the realization of case is usually with some affix, such as English -’s, and that affix must
appear on some particular word of the NP (and sometimes, more than one word). In the theory
assumed here, the choice is made at the level of the case marking rule; and this choice is assumed
to be “put the feature on the rightmost word”.
This position is actually a bit tricky to defend, for other approaches might in principle also
work. For instance, if in (124) we had said not “to the rightmost word of NP2”, but instead “to the
head of NP2”, the rule would have work just as well for the example given, correctly attaching the
Genitive case feature (and ultimately, -’s) to student, which is the head of NP2 as well as its
rightmost word.
The correct choice emerges when we look at more complicated cases like [ the king of
England’s ]NP hat, where England is the rightmost word of its NP but not the head (the actual head
is king). The matchup is shown below:
Genitive Case Marking
In NP1
NP2
NP
NP N
|
Art N PP hat
| |
the king P NP
| |
of N
|
England
[Case:Gen]
In fact, the distinction just made represents a real distinction among languages. German,
which is fairly closely related to English, nevertheless uses the distinct method of placing case on
the head of the NP that is required to bear case. Let us consider an example from German. On
German Amazon I found an entry for a book with this title:
(126) A German book title illustrating case marking on heads
Schliemanns Erben, Von den Herrschern der Hethiter

Schliemann’s legacy from the ruler-dative-plural the Hittite-genitive plural
zu den König-en der Khmer

to the king-dative plural the Khmer-genitive plural
‘Schliemann’s legacy: from the rulers of the Hittites to the kings of the Khmers’69
69
Hermann Schliemann was the archaeologist who excavated the ruins of Troy.
We’re interested in zu den König-en der Khmer, meaning ‘to the kings of the Khmers’.70
Prior to case marking, the structure looks like this (for this particular construction, the relevant
phrase structure rules of German are the same as in English):
PP
P NP
| |
zu Art N NP
| |
den König Art N
[Num:Pl] | |
der Khmer
[Number:Plural] is already attached to König ‘king’; this reflects a semantic choice made by the
person who made up this title.
A crucial fact about German is that the various prepositions take (more formally: govern)
different cases. The preposition zu, pronounced [tsu] and meaning ‘to’, is one of the prepositions
that governs the dative case. A partial dative-case marking rule for German can be written as
follows:
(127) German Dative Case Marking
In the configuration shown:
PP
P NP
where P is one of { zu, aus, ausser, bei, mit, nach, seit, von },71 assign [Case:Dative] to the
morphosyntactic representation of the head of NP.
This rule targets the head of NP for dative case realization, hence applies to our example as
follows. You should check every arrow in the diagram to make sure it makes sense.
70
The Khmers are the Cambodians.
71
‘to’, ‘from’, ‘except’, ‘at X’s home’, ‘with’, ‘after’, ‘since’, ‘of’
PP
P NP
where P is one of { zu, aus, ausser, bei, mit, nach, seit, von }, assign [Case:Dative] to the
morphosyntactic representation of the head of NP.
This rule targets the head of NP for dative case realization, hence applies to our example as
follows:
PP
P NP
| |
zu Art N NP
| |
den König Art N
 Num:Pl 
Case:Dat E
der Khmer
The dative plural is then realized in the morphology with suffixation:
Inflection morphology: Dative Plural Realization (German)

Suffix -en if morphosyntactic representation contains [Num:Plur., Case:Dative].
This will derive the boldfaced material in zu den Königen der Khmer.
There are further details about German we’ll pass over here quickly. Case is generally also
realized, through additional agreement rules (see below), on the Article beginning a Noun Phrase.
Thus, den is in fact the dative plural form of the definite article.
The crucial distinction illustrated here is the edge-based case marking of the English genitive
vs. the head-based marking of German datives. If each language used the opposite language’s
strategy, we’d get very different results: *the king’s of England hat (marking of genitive on the
head), and *zu den König der Khmeren (marking of dative on the rightmost word).
Study Exercise #19
Look at example (126) again and find a second case in which Dative case is realized on a head
(and not on the rightmost word) of NP.
Answer to Study Exercise 19
The relevant passage is this:
Von den Herrschern der Hethiter

from the ruler-dative-plural the Hittite-genitive plural
The preposition von is on the list in rule (127) of prepositions that require their NP objects to
be in the Dative case. The relevant NP is Herrschern der Hethiter, ‘ruler of the Hittites’. The head
of this NP is Herrschern, ‘ruler-dative plural’, and the Dative suffix can be identified on it. It’s a
little bit obscured for phonological reasons; as it happens, the dative is spelled and pronounced -en
after g, as in the earlier König-en, but simply -n after r, as in Herrscher-n.
———————————————————————————————————
There are other differences between edge-based and head-based case marking. Marking on
heads tends to get complicated, with different affixes for different nouns and so on; marking on
edges tends to be a simple, single morpheme like English -‘s. Marking on heads probably is more
often accompanied by agreement on modifying adjectives and articles.
20. Agreement
Features also get assigned in syntax when one phrase agrees with another. For instance, in
English we have a very simple agreement paradigm in verbs.
I jump we jump
you jump you jump
he/she/it jumps they jump
There is only one ending, -s, which marks three features at once; occurring when the subject is
[Person:3, Number:Singular, Tense:Present]. Note, however, that for the special verb be there is a
slightly richer system, with a special form for the first person singular:
I am we are
you are you are
he/she/it is they are
The point at hand is that agreement with the subject is inherently syntactic; the verb needs to
“know,” what the subject is in order to bear the right inflectional features.
Again, our strategy is to write a syntactic rule that assigns the features of the morphosyntactic
representation, then a rule of inflectional morphology to add the appropriate affix.
The syntactic rule can be written provisionally as follows:

Verbal Agreement (English)
In S
NP VP
assign the [Person] and [Number] features of the head of NP to the head of VP.
An application is shown as follows.
Verbal Agreement (English)
In S
NP VP
assign the [Person] and [Number] features of the head of NP to the head of VP.
S S
NP VP NP VP
| | | |
N V → N V
| | | |
Fred jump Fred jump
Tense:Pres 
Number:sg Number:sg Number:sg
Person:3  E [Tense:pres] Person:3   E
 E
Person:3 
The rule of inflectional morphology that generates the -s suffix is given below:
3rd Singular Present Rule
Suffix -s if [Person:3, Number:Singular, Tense:Present]
This will convert the stem jump in the tree above to the correct output form jumps.
Compare: I jump, they jump, etc.; these do have their subject features copied, but no rule of
the morphology refers to their features, so nothing is added morphologically.
20.1.1 Agreement in general
In languages with rich inflection, agreement rules like the above copy a great deal of
information around the tree: verbs agree with their subjects (and sometimes their objects, too),
adjectives and articles agree with the nouns they modify, and in at least one language (Lardil,
Australia) nouns agree with the verb of their clause in tense.
Here are some characteristic data from Swahili, a language with a rich inflectional system.
Swahili has a rough equivalent of the genders found in European and Semitic languages, but it is
much richer: instead of two or three genders, there is a system of about a dozen noun classes,
which occur in pairs that consist of one singular and one plural prefix (traditionally numbered
separately, rather than as a group). In the sentences below, you can see the “1/2” class and the 3/4
class; and in each the class is marked not just on the noun (toto, ‘child’; or fuko ‘bag’), but also
through agreement rules on the Article le ‘that’ and on the past tense verb li-anguka. The forms are
shown with a plausible choice of features and morphosyntactic representations. The features
employed are [NounClass] and [Number]. NounClass can take the values 1/2, 3/4, etc., following
the traditional singular/plural numbering of the classes, but grouping the singular-plural pairs
together in a single feature value.
(128) Noun class agreement in Swahili
a. [NounClass:1/2, Number: singular]

m-toto yu-le a-li-anguka
Class 1-child Cl. 1-that Cl. 1 subj.-past-fall
‘that child fell’
b. [NounClass:1/2, Number: plural]

wa-toto wa-le wa-li-anguka
Class 2-child Cl. 2-that Cl. 2 subj.-past-fall
‘those children fell’
c. [NounClass:3/4, Number: singular]

m-fuko u-le u-li-anguka
Class 3-bag Cl. 3-that Cl. 3 subj.-past-fall
‘that bag fell’
d. [NounClass:3/4, Number: plural]

mi-fuko i-le i-li-anguka
Class 4-bag Cl. 4-that Cl. 3 subj.-past-fall
‘those bags fell’
As the data show, these features get copied by rule from the noun that inherently bears them onto
the neighboring Art and Verb, and later are realized in the inflectional component by rules that
attach the various prefixes. It is usually the same prefix that realize these features on nouns,
articles, and verbs, but as you can see in some cases they differ; so more than one prefixation rule
would be needed.
Summing up, agreement and case marking are the main phenomenon in which syntax
determines the features of the morphosyntactic representation. Other features represent choices
made by the speaker, or are inherent to particular words. These three sources combine to produce
the completed representation, which is then fed into the rules of inflectional morphology to provide
the realization of these features in sound.
21. Syntactic analysis: creating an adequate set of phrase structure rules for a language
Languages differ quite a bit in their word order, a fact which can be described in grammars by
writing different phrase structure rules.
One kind of analytic skill to be developed here is to formulate the phrase structure rules
needed to analyze any particular language. Assuming you have a representative batch of sentences
to work with, this involves three steps:
• Parse the sentences.

• Look at the trees, and see which daughters any given type of node can have.
• Express what you find with a reasonably economical set of rules.
21.1 Parsing without a grammar?
Initially, you may have a sense of circularity when I say “parse the sentences”. Since parsing
is, essentially, the assignment of syntactic structure according to the rules of a grammar, how can
one parse without already having the grammar in place? But in fact, there is probably enough
information in the general theory of syntax that we are assuming for you to achieve a reasonably
good parse. In particular, we are assuming the principles in (129):
(129) Some principles that assist parsing in languages where the phrase structure rules are
unknown
• Languages always have clauses (which linguists conventionally label as S).
• Clauses generally have a subject and a predicate, analyzed as the NP and the VP.72
• All languages have Nouns and Verbs. Many languages also have Adjectives, either
Prepositions or Postpositions (like prepositions, but follow their objects), and perhaps other
parts of speech.
• Nouns, Verbs, and other words (for instance, Adjectives and Pre/Postpositions) are
normally heads of phrases, so that N is head of NP, V is head of VP, and so on.
• These syntactic principles are accompanied by loosely formulated semantic principles: the
VP of S predicates something of the head, and in XP (e.g., NP, VP, etc.) the sisters of the
head X are semantically modifiers of the head.
Generally, assuming this principles will suffice to obtain at least a rough characterization of
the structure of sentences, which can be refined later. You should also remember that the principles
above are a bare minimum; languages will have all sorts of different orders of the daughter nodes
72
An interesting challenge for this view is the existence of languages with the subject in the middle, with
normal sentence order Verb-Subject-Object (as in Tagalog), or Object-Subject-Verb (as in Jamamadi) — where is the
VP? We won’t try to handle these cases in this text.
of a given phrase, and differ in what daughters are allowed. Here, you have to rely on the patterns
in the data.
21.2 Phrase structure rules for Hittite
The data below involve sentences in Hittite, taken from an exercise created by Prof. Jay
Jasanoff of Harvard University. The transcription and syntactic analysis were guided by input from
my UCLA colleague Prof. Craig Melchert; both are experts on this language.
Hittite was spoken in early ancient times in what is now Turkey. It is known from a hoard of
about 25,000 cuneiform tablets discovered early in the last century and deciphered in the decades
that followed. Some of the texts date back to about 1700 BCE and thus count as the oldest
attestation of any Indo-European language.73 We accept here on Jasanoff’s authority that the
sentences below, which he made up, would be grammatical to real Hittite speakers if we could
somehow bring them back.
Phonetic symbols are necessarily based on educated guesses. [x] is as in Spanish jamon or
German Bach (voiceless velar fricative).
1. nu xassus salli parn-i anda estsi

comp king big-dative house-dative in is
‘The king is in the big house.’
2. nu antuxsas akuwakuwan istamastsi

comp man-nominative frog-accusative hears
‘The man hears the frog.’
3. nu antuxsas sallin akuwakuwan parn-a pexutetsi

comp man-nominative big-accusative frog-accusative house-allative brings
‘The man brings the big frog home.’
4. nu akuwakuwas westar-i assun memijan tetsi

comp frog-nominative shepherd-dative good-accusative word-acc. says
‘The frog says a good word to the shepherd.’
5. nu westaras sallin akuwakuwan pir-i anda xassussar-i katta

istamastsi
comp shepherd-nom. big-acc. frog.-acc. house-dative in queen-dative with
hears
‘The shepherd hears the big frog in the house with the queen.’
6. nu akuwakuwas antuxsan natta istamastsi

comp frog-nominative man-accusative not hears
73
Indo-European is the very large language family that includes (for example) English, Russian,
Hindi, Latin, Irish, etc. See Chapter 13.
‘The frog doesn’t hear the man.’
7. nu xassussaras xassui piran salli akuwakuwi katta tijatsi

comp queen-nom. king-dat. before big-dative frog-dative with comes
‘The queen comes before the king with the big frog.’
8. nu westaras assui xassui akuwakuwan pexutetsi

comp shepherd good-dative king-dative frog-accusative brings
‘The shepherd brings the frog to the good king.’
One can do both syntactic and morphological analysis on these texts. At the level of
morphology, it is possible to collect some partial noun paradigms, as follows.
xassu-s king nominative

xassu-i king-dative
antuxsa-s man-nominative
antuxsa-n man-accusative
akuwakuwa-s frog-nominative
akuwakuwa-n frog-accusative
akuwakuw-i frog-dative
westara-s shepherd-nominative
westar-i shepherd-dative
memija-n word-accusative
parn-i house-dative
parn-a home-allative
xassussara-s queen-nominative
xassussar-i queen-dative
It looks at least roughly that the nominative suffix is -s, the accusative suffix is -n, and the
dative suffix is -i. This predicts *akuwakuwa-i and *xassussara-i for the datives of “frog” and
‘queen’; in fact, there’s a bit of phonology going on: the vowel a is dropped before this suffix. We
express the rules of inflectional morphology as follows.
Hittite Nominal Inflection (sketch)

Suffix -s when [Case:Nominative]
Suffix -n when [Case:Accusative]
Suffix -i when [Case:Dative]
There also appears to be verbal inflection, for which we can conjecture this rule:
Suffix -tsi when [Person:3, Number:singular, Tense:present]
But in fact we know almost nothing about -tsi from these few data.
Turning now to the phrase structure rules, the idea is to inspect the sentences, parse them
according to the principles of the theory, and generalize over what we see to produce the rules.
An intriguing aspect of the sentences is that they all begin with nu. This is most likely a
complementizer: Hittites usually spoke in CP’s, not S’s, though in certain contexts it was possible
to say just a plain S. Thus we will start our derivations with CP and assume this phrase structure
rule:
CP → Comp S
NP could be derived by the following rule:
NP → (AP) N
AP → A
We don’t have any instances of A modified within AP, but we include the AP simply to maintain a
consistent theory, in which X and XP are always in one-to-one correspondence.
Another simple rule is for PP, which is this language is evidently not a phrase for prepositions
but for postpositions, which are just like prepositions but come after their NP rather than before.
The phrase structure rule needed is:
PP → NP P
In sentences, the subject evidently comes before the predicate, justifying the rule
S → NP VP
The trickiest phrase structure rule to write here is for VP. Here there is a question of
methodology: what is the best way to figure out simple phrase structure rules when we are given
the data? I suggest putting together what I will call a phrase structure table. The table puts each
example in a separate row, and aligns the contents of the phrase (here, as VP) in columns.
(130) A phrase structure table for the Hittite VP
NP NP PP PP Adv V
1 salli piri anda estsi
big-dat. house-dat. in is
2 akuwakuwan istamastsi
frog-acc. hears
3 sallin akuwakuwan parn-a pehutetsi
big-acc. frog-acc. house-dative brings
4 westari assun memijan tetsi
shepherd-dat. good-acc. word-acc. says
5 sallin akuwakuwan piri anda hassussari katta istamastsi
big-acc. frog.-acc. house-dat. in queen-dat. with hears
6 antuhssan natta istamastsi
man-acc. not hears
7 hassui piran salli akuwakuwi katta tijatsi
king-dat. before big-dat. frog-dat. with comes
8 assui hassui akuwakuwan pehutetsi
good-dat. king-dat. frog-acc. brings
If we collect all of the various items that evidently fit within a VP, and (going out on a limb)
put them in a single rule, we get:
VP → (NP)(NP)(PP)(PP)(Adv) V
This completes the set of phrase structure rules, stated all in one place thus:
Phrase structure rules for Hittite

CP → Comp S
S → NP VP
VP → (NP)(NP)(PP)(PP)(Adv) V
NP → (AP) N
AP → A
PP → NP P
“P” must be read “postposition”, rather than “preposition.”
21.3 Example diagrammed sentence
The rules suffice to generate all the sentences; here is one particularly long example.
CP
VP
NP PP PP
NP AP NP NP
| | | |
Comp N A N N P N P V
| | | | | | | | |
nu westara-s salli-n akuwakuwa-n pir-i anda xassussar-i katta istamas-tsi
the shepherd-nom. big-acc. frog.-acc. house-dative in queen-dative with hears

‘The shepherd hears the big frog in the house with the queen.’
Study Exercise #20
Parse the Hittite sentence
nu akuwakuwas westari assun memijan tetsi

the frog-nominative shepherd-dative good-accusative word-acc. says
Answer on next page.

CP
VP
NP NP NP
| |
Comp N N Adj N V
| | | | | |
nu akuwakuwas westari assun memijan tetsi
———————————————————————————————————
21.4 Hittite as a head-final language
It can be seen that, at least in these data, Hittite is a head-final language: N is last in NP, P is
last in PP, V is last in VP (and we don’t know about AdjP).
Some other well-known head-final languages are Japanese, Korean, Bangla (=Bengali), and
Turkish. The Bantu languages, such as Swahili and Zulu, tend to be strongly head-initial. English
tends towards being head-initial, but is conflicted, in the sense that it puts adjectives before the
head noun in NP. Hence some English noun phrases have the head noun in the middle:
[ the long [ book ]N about linguistics ]NP
21.5 Case marking in Hittite
Hittite has a richer case system than English, with overt suffixes marking the Nominative,
Dative, and Accusative. We can write syntactic rules that place the appropriate value for the
feature [Case], based on the configuration of the tree.
For instance, Dative case is assigned in Hittite by postpositions. It can be attached by a similar
rule:
Dative Case Marking (Hittite)
In the configuration
PP
NP P
add [Case:Dative] to the morphosyntactic representation of the head of NP.
Getting Accusative and Dative objects right is trickier, and we also have very few data, so the
following is really something of a guess:
Case Marking for Objects (Hittite)
In verb phrases containing one or more NP, then

• if there are two NP, assign [Case:Dative] to the head of the first and [Case:Accusative] to
the head of the second.
• if there is just one NP, assign [Case:Accusative] to its head.
Here is an example with two NPs inside the VP:
VP
NP NP NP
| AP
Art N N | N V
| | | A | |
nu akuwakuwa-s westar-i | memija-n te-tsi
[Case:Nom] [Case:Dat] assu-n [Case:Acc]
“assign [Case:Dative] to the head of the first and [Case:Accusative] to the head of the second”
A further rule, not stated here, would cause adjectives (such as assun above) to agree with
their head nouns in case.
——————————————————————————————————————
Study Exercise #21: Turkish Phrase Structure Rules
Examine the Turkish sentences below. Provide a syntactic tree for each (you might find it
more convenient to use the English glosses in the tree, rather than that actual Turkish words). Then
examine all your trees and come up with a terse, economical set of phrase structure rules that
derive all of them.
Spelling: I’ve replaced some Turkish letters with English equivalents or IPA symbols:
• sh is IPA [ʃ], Turkish letter ş.

• ch is IPA [tʃ], Turkish letter ç.
• [ɨ] is IPA, pronounced somewhat like the sound of the second vowel in roses; Turkish letter
ı.
• Turkish letter ü (which in IPA is [y]) is made by uttering the vowel [i] (“ee”) and
simultaneously rounding the lips.
• Turkish letter ö (which in IPA is [ø]) is made by uttering the vowel [e] (“bay”) and
simultaneously rounding the lips.
• [dʒ] is IPA for the consonants that occur at the beginning and end of English judge. In
Turkish it is spelled with a c.
1. Shirin uyudu
Shirin slept
‘Shirin slept’
2. Vezir uyudu
vizier slept
‘The vizier slept’
3. Uzun vezir uyudu

tall vizier slept
‘the tall vizier slept’
4. Chok uzun vezir uyudu

very tall vizier slept
‘the very tall vizier slept’
5. Sarɨshɨn uzun vezir uyudu

blond tall vizier slept
‘The tall blond vizier slept’
6. Yorgun sarɨshɨn uzun vezir uyudu

tired blond tall vizier slept
‘The tired tall blond vizier slept’
7. Defter düshtü
notebook fell
‘the notebook fell’
8. Shekspir-in defter-i düshtü

Shakespeare-GEN notebook-Poss fell
‘Shakespeare’s notebook fell’
9. Shekspir-in uzun defter-i

Shakespeare-GEN long notebook-Poss
‘Shakespeare’s long notebook’
10. uzun vezir-in shiir-i

tall vezir-GEN poem-Poss
‘the tall vizier’s poem’ (i.e. the poem by him)
11. uzun vezir-in uzun shiir-i

tall vizier-GEN long poem-Poss
‘the tall vizier’s long poem’
12. uzun vezir-in uzun sɨkɨdʒɨ shiir-i74

tall vizier-GEN long boring poem-Poss
‘the tall vizier’s long boring poem’
13. *uzun vezir-in Shekspir-in shiir-i

tall vizier-GEN Shakespeare-GEN poem-Poss
‘the tall vizier’s Shakespeare’s poem’
14. Shirin Kerem-i öptü

Shirin Kerem-ACC kissed
‘Shirin kissed Kerem’
15. Uzun vezir Shekspir-in shiir-i-ni okudu

tall vizier Shakespeare-GEN poem-Poss-ACC read
‘The tall vizier read Shakespeare’s poem’
16. Shirin vezir-in uyuduuna inanɨyor

Shirin vizier-GEN is-sleeping believes
‘Shirin believes the vizier is sleeping’
17. Kerem Shirin-e vezir-in uyuduuna söyledi

Kerem Shirin-DAT vizier-GEN is-sleeping told
‘Kerem told Shirin that the vizier is sleeping’
18. Kerem Shirin-in vezir-in uyuduuna inandɨɨnɨ zannediyor

Kerem Shirin-GEN vizier-GEN is-sleeping believes imagines
‘Kerem imagines that Shirin believes that the vizier is sleeping’
74
dʒ is IPA for the English “j’ sound
19. Orhan Kerem-in Shirin-in vezir-in uyuduuna inandɨɨnɨ zannettiini

itirafetti
Orhan Kerem-GEN Shirin-GEN vizier-GEN is-sleeping believes imagines
confessed
‘Orhan confessed that Kerem imagines that Shirin believes that the vizier is sleeping’
1. Shirin uyudu
Shirin slept ‘Shirin slept’
2. Vezir uyudu
vizier slept ‘The vizier slept’
3. Uzun vezir uyudu

tall vizier slept ‘the tall vizier slept’
4. Chok uzun vezir uyudu

very tall vizier slept ‘the very tall vizier slept’
4. Sarɨshɨn uzun vezir uyudu

blond tall vizier slept ‘The tall blond vizier slept’
5. Yorgun sarɨshɨn uzun vezir uyudu

tired blond tall vizier slept
‘The tired tall blond vizier slept’
6. Defter düshtü
notebook fell ‘the notebook fell’
7. Shekspir-in defter-i düshtü

Shakespeare-GEN notebook-Poss fell
‘Shakespeare’s notebook fell’
8. Shekspir-in uzun defter-i

Shakespeare-GEN long notebook-Poss
‘Shakespeare’s long notebook’
9. uzun vezir-in shiir-i

tall vezir-GEN poem-Poss
‘the tall vizier’s poem’ (i.e. the poem by him)
10. uzun vezir-in uzun shiir-i

tall vizier-GEN long poem-Poss
‘the tall vizier’s long poem’
11. uzun vezir-in uzun sɨkɨdʒɨ shiir-i

tall vizier-GEN long boring poem-Poss
‘the tall vizier’s long boring poem’
12. *uzun vezir-in Shekspir-in shiir-i

tall vizier-GEN Shakespeare-GEN poem-Poss
‘the tall vizier’s Shakespeare’s poem’
13. Shirin Kerem-i öptü

Shirin Kerem-ACC kissed
‘Shirin kissed Kerem’
14. Uzun vezir Shekspir-in shiir-i-ni okudu

tall vizier Shakespeare-GEN poem-POSS-ACC read
‘The tall vizier read Shakespeare’s poem’
15. Shirin vezir-in uyuduuna inanɨyor

Shirin vizier-GEN is-sleeping believes
‘Shirin believes the vizier is sleeping’
16. Kerem Shirin-e vezir-in uyuduuna söyledi

Kerem Shirin-DAT vizier-GEN is-sleeping told
‘Kerem told Shirin that the vizier is sleeping’
17. Kerem Shirin-in vezir-in uyuduuna inandɨɨnɨ zannediyor

Kerem Shirin-GEN vizier-GEN is-sleeping believes imagines
‘Kerem imagines that Shirin believes that the vizier is sleeping’
18. Orhan Kerem-in Shirin-in vezir-in uyuduuna inandɨɨnɨ zannettiini itirafetti

Orhan Kerem-GEN Shirin-GEN vizier-GEN is-sleeping believes imagines confessed
‘Orhan confessed that Kerem imagines that Shirin believes that the vizier is sleeping’
PS Rules for Turkish
S → NP VP
NP → (NP) (AP)* N
VP → (P) (S) V
AP → (Adv) A
If you’re curious you might try to figure out how Turkish assigns Genitive, Possessive,
Accusative, and Dative case. The data aren’t really sufficient to solve the problem but they are
suggestive.
————————————————————————————————————
22. A bit more on phrase structure rules: Kleene star
Let us beef up the system of phrase structure rules once more. Some phrase structure rules
allow for any number of daughters of a certain type. For example, the rule for NP allows for an
unlimited number of Adjective Phrases preceding the noun, as in ‘a very long, dull, unusually
boring movie’. A formalism for this often employed is to enclose in brackets the element that can
be repeated indefinitely, and place an asterisk after the right bracket (the asterisk is known as
“Kleene star”, after the computer scientist who invented the notation). For example, the phrase
structure rule for NP can be written as follows:
NP → (Art) (AP)* N (PP)*.
An NP that uses both (AP)* and (PP)* would be the very big blue book about linguistics on the
counter. AP’s: very big and blue; PP’s: about linguistics, on the counter.
Quite a few of the items on our previous phrase structure grammar would be more accurately
depicted with Kleene star; the following is a list:
(131) Phrase structure rules for English: Version II, improved with Kleene star
S → NP (Aux) VP
Art 
NP → NP   (AP)* N (PP)* Example: his noble, wonderful inspiring gift of
 
A A E
$1,000,000 to X on Tuesday
NP → Pronoun
AP → (Adv) A
VP → V (NP) (NP) (PP)* (CP) Ex.: sold books to students for $50 on Wednesdays
PP → P NP
CP → (Comp) S
NP → NP (Conj NP)* Ex. Alice and Sally and Bill left.

VP → VP (Conj VP)* Ex. We sang the song and danced the dance and
played the tune.
PP → PP (Conj PP)* Ex. We tossed it over the fence and through the
window and under the house.
S → S (Conj S)* Ex. He said that he was sick and he would go and he
would get better soon.
CP → CP (Conj CP)* Ex. He said that he was sick and that he would go
and that he would get better soon
V → V (Conj V)* Ex. They washed and diced and sliced the
vegetables.
Plainly, Kleene star, just like recursion, can generate infinite numbers of structures, but in
what is arguably a less interesting way: it simply lets us have an unbounded number of sister nodes
at particular places in the tree; in contrast, recursion creates “deeper” trees, with much more (and
less trivial) structure.75
Lastly, there is the issue of how to detect the need for Kleene star in actual language data. The
presence of a variable number of phrases of the same type occurring in sequence does not always
mean that Kleene star is justified. In particular, the English verb phrase allows for zero, one, or two
NP objects — but no more, as the following sentences suggest:
0: Alice sings.
1: Alice sold mittens.
2: Alice sold Fred mittens.
3: *Alice sold Fred mittens money.
*(any longer set of NP)
I would guess that if a language allows at least three of some category, then there probably is no
upper limit and that Kleene star is justifiable.
23. Summary
In this chapter, we have covered four topics:
• Recursion, the basis by which grammars can generate infinite numbers of structures.
• How syntax determines aspects of morphosyntactic representations, through rules of case
marking and agreement.
• How to go about writing phrase structure rules for novel language data.
• How to augment phrase structure rules with Kleene star, to obtain unlimited numbers of
daughter nodes where this is appropriate.
75
In further courses about syntax you may see an approach that eliminates Kleene star by introducing more
categories and using recursion. For instance, a grammar like NP → Det N′, N′ → (AP) N′, N′ → N can place an
unbounded number of AP’s before N within NP.
Chapter 5: Syntax II — Transformations
1. Syntax beyond phrase structure: the need for transformations
As seen already, our overall goal is to beef up the grammar so that it becomes an ever better
approximation to the grammar internalized by speakers of English. We have done this by
amplifying the system of phrase structure rules, and also by adding rules of agreement and case
marking to govern the distribution of inflectional features. This section introduces the next major
type of syntactic rule, the transformation, and argues for why it is needed.
English contains a construction called the Tag Question. Tag questions appear after the
comma in the following examples:
(132) Examples of Tag Questions
Frogs can eat flies, can’t they?

The president has resigned, hasn’t she?
Bill was watching the stew, wasn’t he?
As the data show, a tag question contains three parts in order:
• A copy of the Aux of the main sentence (can…can, has … has, was … was).
• A contracted form of the word not
• A pronoun expressing the person and number of the subject of the main sentence.
1.1 Digression: spell-out rules for word sequences
We introduced spell-out rules in Chapter 2 as a means of covering irregular inflected forms

like kept or went. Spell-out rules are also a sensible option for contracted forms of English like
can’t. It seems intuitively very reasonable that can’t is the normal realization of can not, and
similarly that hasn’t is the realization of has not, and (more interestingly) won’t is the realization
of will not. For such contractions (as traditional grammar calls them), we employ spell-out rules, of
which the following are a partial list:
(133) Some spell-out rules of English that create negative contractions
will not → won’t

can not → can’t
am not → aren’t76
do not → don’t
shall not → shan’t77
76
As in I’m tall, aren’t I?, used only in vernacular speech.
77
Archaic, at least for Americans.
etc.
1.2 Phrase structure rules inadequate for tags
It is in the nature of phrase structure rules that they can’t copy: they specify the daughter
nodes of a particular kind of mother node, as well as the order in which the daughters appear, but
that is all. If we naively attempted to generate tag questions simply by extending our set of phrase
structure rules, we would derive many ungrammatical instances with a mismatched Aux, because
these rules lack the copying capacity. Here is the failed approach in detail:
(134) A Failed Grammar for English Tags
(a) Change the phrase structure for S to:
S → NP (Aux) VP (Tag)
(b) Add the phrase structure rule:
Tag → Aux not Pro
This hypothesis derives Alice will kiss Bill, won’t she? as follows:
(The tree shows the pre-spelled-out version of the sentence; the spell-out rule would convert will
not to won’t.)
This hypothesis fails because it doesn’t enforce the copying requirement. We can apply the
very same rules and derive preposterous sentences:
By Spell-Out: *Alice will kiss Bill, hasn’t she?
and similarly:
*Alice will kiss Bill, can’t he?

*Alice will kiss Bill, hasn’t it?
*Alice will kiss Bill, won’t they?
1.3 Terminology of failed grammars
Linguists are always dealing with failed grammars like the one just given, taking them back to
the drawing board and trying either to improve them or replace them with a better approach. Failed
grammars are not a pointless activity; they lead us to explore the data more thoroughly and force
us to refine or replace the failed analysis.
Some terminology for failed grammars that is widely used:
Undergenerate: A grammar undergenerates when there are grammatical sentences

that it cannot generate.
Overgenerate: A grammar overgenerates when it generates ungrammatical
sentences
The grammar we just looked at overgenerates, as the starred examples above indicate. A grammar
that can’t generate tag questions at all (which is what we had before) undergenerates. We are
looking for a grammar that does neither.
1.4 Diagnosing the failure
As already noted, the failed grammar given in (107) above fails because nothing in the rule
apparatus developed so far can copy. Grammar (107) can be thought of as providing a poor
substitute for copying: it copies the structure, but not the actual words involved, which is what we
really need. Plainly, we need more kinds of rules.
More generally, phrase structure grammars don’t allow for cases where the constituents
present in one part of a tree depend on the constituents present in another part, which may be some
distance away. In fact, tag questions are a rather out-of-the-way instance of this phenomenon; the
really important cases are yet to come. The tag questions will suffice, however, to give the basic
idea.
1.5 Transformations
Faced with phenomena like tag questions, linguists generally assume that phrase structure
rules do not alone suffice as a grammar formalism for languages. An additional kind of rule takes
as its input a sentence generated by the phrase structure rules and alters it in some way.
• A syntactic rule that alters a tree structure is called a transformation.

• A grammar that includes transformations as well as phrase structure rules is a
transformational grammar.
The rules of case marking and agreement given in the preceding chapter could be considered
to be transformations, although their effects are not as dramatic as the copying and movement
transformations we will cover in what follows; case marking and agreement only change the
morphosyntactic representation, not the tree as a whole.
The general strategy seen in transformations is to let the phrase structure rules define the
“basic inventory” of sentences in the language, and let the transformations apply to generate the
wider variety of sentences that go beyond the capacity of phrase structure rules. For example, the
sentence Alice will kiss Bill is in some sense a basic sentence (being generable by phrase structure
rules alone), and Alice will kiss Bill, won’t she is a syntactic elaboration of the simple sentence.
What can transformations do? This is a rather open question, whose answer forms a large part
of the theory of syntax. At the moment, it’s best to simply formulate the transformations we need
and later on see what general theoretical principles are applicable.78
1.6 A transformation for tag questions
Here is a copying transformation that can derive tag questions. As you can see, it uses notation
seen earlier in morphology, where we used numeral subscripts to make clear what changes into
what for rules of infixation and reduplication. However, the syntactic transformation also contains
reference to the tree structure that is manipulated.
78
To be honest, we’re really going to stop at the first step here; for a deeper theory of transformations
you’ll have to take more advanced syntax courses.
(135) Tag Question Transformation

S S
|
NP Aux VP → NP Aux VP Tag where [MS] = morphosyntactic
[MS] [MS] representation
1 2 3 1 2 3 Aux not Pro of head of NP
[MS]
2 1
Here is an explication of this rule. It assumes you have an S, consisting of an NP, an Aux, and a
VP. The NP is assumed to have a morphosyntactic representation, that is, a feature bundle located
on the head of the NP. These three items (NP, Aux, VP) are subscripted 1, 2, and 3.
On the right side of the arrow in the rule, the change is shown. A new daughter of S is added
at the right edge, with the category Tag. Its internal content consists of an Aux, the word not, and a
Pronoun, in that order. The Aux is a copy of the Aux in the original sentence (this is shown by its
bearing the number 2), and the Pronoun is assigned a copy of the morphosyntactic representation
of the subject (this is indicated by the numerical subscript 1). Assuming that the features [Gender],
[Person], [Number], and [Case] are part of the morphosyntactic representation, this will place the
appropriate kind of pronoun into the tag; masculine subjects will get masculine pronouns, plural
subjects will get plural pronouns, and so on.
For explicitness, here are the nominative pronouns of English with their morphosyntactic
representations (for the non-nominative pronouns, see p. 108 above).
(136) Morphosyntactic representations for English nominative pronouns
I [Case:Nominative, Person:1, Number:Singular] (Gender is free)79

you [Case:Nominative, Person:2] (Gender and number are free)
he [Case:Nominative, Person:3, Number:Singular, Gender:Masculine]
she [Case:Nominative, Person:3, Number:Singular, Gender:Feminine]
it [Case:Nominative, Person:3, Number:Singular, Gender:Neuter]
we [Case:Nominative, Person:1, Number:Plural] (Gender is free)
they [Case:Nominative, Person:3, Number:Plural] (Gender is free)
Indeed, in what we are about to do, it is sensible to think of the pronouns simply as the way that
the English language happens to spell out the category Pronoun when it bears one of these
morphosyntactic representations. Thus, for instance, a rule that take the features of she and spells
out the pronoun she is given below.
79
By “free”, I mean that you can use this pronoun no matter what the specification in the
morphosyntactic representation.
(137) A spell-out rule that creates the word she
Pronoun
Person:3
Case:Nominative

Number:Singular  → she E
Gender:Feminine
With this apparatus in place, we can provide a full derivation for the sentence Alice can sing,
can’t she?
First step: application of the phrase structure rules to derive Alice can sing
S
|
NP Aux VP
| | |
N can V
| |
Alice sing
 Pers:3 
Num:sg 
 
E
Gen:fem
Note that Alice, by its very meaning, is inherently 3rd person, singular, and feminine.
Second step: since Alice is the subject, a rule of case marking makes it Nominative:
S
|
NP Aux VP
| | |
N can V
| |
Alice sing
Num:sg 
Pers:3
Gen:fem  E
Case:nom
Third step: application of Tag Question Transformation (stated in (135)); matchup shown with
dotted lines):
(138) Tag Question Formation applied

S S
|
NP Aux VP → NP Aux VP Tag where [MS] = morphosyntactic
[MS] [MS] representation
1 2 3 1 2 3 Aux not Pro of head of NP
[MS]
2 1
S S
|
NP Aux VP → NP Aux VP Tag
| | | | | | |
N can V N can V Aux not Pro
Num:sg
Pers:3

Gen:fem 
E
Case:nom
Alice sing Alice sing can
Num:sg
Pers:3
 Num:sg
Pers:3

   
Gen:fem  Gen:fem 
E E
Case:nom Case:nom
The dotted lines may look initially like spaghetti. Yet, you may find it worthwhile to inspect every
strand. This is the way to make sure you are correctly applying a transformation to a tree.
Fourth step: apply rule the Spell-Out Rule for she in (137) above. It spells out a pronoun that is
[Pers:3, Num:sg, Gen:fem, Case:nom] as she (tree omitted):
NP Aux VP Tag
| | | |
N can V Aux not Pro
| | | |
Alice sing can she
Num:sg
Pers:3

 
Gen:fem 
E
Case:nom
Last step: using one of the spell-out rules in (133), we spell out the sequence can not as can’t:
Alice can sing, can’t she?

Study Exercise #22
Derive The frogs will sing, won’t they?
The steps should include Phrase Structure rules, Tag Question formation, spelling out of the
pronoun they, spell-out of won’t, and attachment of the plural suffix (inflectional morphology) to
frog.
I. Phrase structure rules and lexical insertion:
S
|
NP Aux VP
| |
Art N will V
| | |
the frog sing
 Pers:3 
Num:plur 
  E
Gen:neuter
II. Nominative Case Marking:
S
|
NP Aux VP
| |
Art N will V
| | |
the frog sing
Num:plur 
Pers:3
Gen:neuter E
Case:Nom 
III. Tag Question Transformation (see (135))
S
|
NP Aux VP Tag
| | |
Art N will V Aux not
| | | |
the frog sing will Pro
Pers:3
Num:plur 
 Num:plur 
Pers:3
  E
Gen:neuter E
Gen:neuter Case:Nom 
1 2 3 2 1
IV. Spell out the pronoun that is [Pers:3, Num:plur, Gen:neuter, Case:nom] as they:
The frog[Pers:3,Num:plur,Gen:Neuter] will sing, will not they?
V. Spell out will not as won’t:
The frog[Pers:3,Num:plur,Gen:Neuter] will sing, won’t they?
VI. Inflectional morphology: Suffix -s when [Number:Plural]
The frogs will sing, won’t they?
___________________________________________________________________
2. Another transformation: Yes/No Question Formation
Consider a second transformation. Every sentence in English that is a statement has a

corresponding question. Thus for
Bill is leaving
we have
Is Bill leaving?
and for
The frog might hop.
we have
Might the frog hop?
Such questions are called Yes/No questions, to distinguish them from questions that begin
with ‘who’, ‘what’, ‘where’, etc., which are called Wh- questions. It is plausible to regard a
yes/no question as a syntactic variant of the corresponding statement; thus the phrase structure
rules will derive the statement, which is converted to a yes/no question. The crucial transformation
is as follows:
(139) Yes/No Question Formation

S S
| |
NP Aux VP → Aux NP VP
1 2 3 2 1 3
Below is a derivation of the yes/no question Will the king of England sit down?. Dotted lines show
the matchup between rule and form.
(140) Yes/No Question Formation applied to a sentence

S S
| |
1 2 3 2 1 3
S S
| |
NP Aux VP Aux NP VP
| | |
Art N PP will V Part → will Art N PP V Part
| | | | |
the king P NP sit down the king P NP sit down?
| | | |
of N of N
| |
England England
You may be worried at this point that we have no way of forming Yes/No questions from a
sentence that lacks an Aux. This issue addressed in the next section.
For now, it’s worth considering Yes/No Question Formation as a transformation. In this case
(unlike for tag questions), it would be quite possible to derive the sentences just by using phrase
structure rules, something along the lines of:
S → Aux NP VP
However, there seem to be at least two reasons that at least suggest that the transformational
approach is better. First, speakers seem to recognize that (for example) Is Bill leaving? is the
yes/no question that “goes with”, or is appropriately paired with, Bill is leaving. We can
characterize this sense of relatedness if we derive the question from the statement. Moreover,
Auxes in English agree with their subjects (see section 20 above for English agreement):
Bill is leaving. (3rd person singular)

Bill and Fred are leaving. (3rd person plural)
This agreement is carried over into the questions:
Is Bill leaving?
Are Bill and Fred leaving?
A clean analysis of this is possible, in which we only state the agreement rule once, if the
questions are derived from the statements. In brief, the derivation would like this:
NP VP
| |
N Aux V
| | |
Bill BE leaving output of phrase structure rules
Pers:3 
Num:Sg E features of Bill are inherent in this noun
NP VP
| |
N Aux V
| | |
Bill BE leaving Agreement: copy the subject features onto
 Pers:3  Pers:3 
Num:Sg E
Num:Sg E the Aux
NP VP
| |
Aux N V
| | |
BE Bill leaving Yes/No Question formation
 Pers:3   Pers:3 
Num:Sg E
Num:Sg E
Is Bill leaving? Spellout of BE as is.

 Pers:3 
Num:Sg E
This, then, is at least some justification for saying that Yes/No questions are formed by a
transformation.
3. Inserted do in English
It’s clear that tag questions and yes/no questions can be formed, even if there is no Aux in the
base sentence. The method used in English is to insert the verb do, which could be described as the
“default Aux” of the language.
For tag questions:
John likes turnips, doesn’t he?

We left early, didn’t we?
For yes/no questions:
Does John like turnips?

Did we leave early?
This phenomenon is an unusual aspect of English, and seems to be completely general. To

give three additional examples:
• Negation: English negates a sentence by placing not directly after its Aux.
I have not done my practicing.

I will not take out the trash.
He cannot play this concerto.
Where the basic sentence has no aux (as in “He likes turnips”), do is inserted to provide
one:
I do not like turnips.
• Polarity focus: one can emphasize the truth of what one is saying (for example, to
contradict someone who doubts it) by putting a strong accent on the Aux.80
I have done my practising.

You will take out the trash.
He can play “Chopsticks”.
Where there would otherwise be no Aux, do is provided:
I do like turnips.
• VP elision. The second of two identical Verb Phrases can be elided, provided an Aux is left
behind:
You should take up hang gliding. Sue has. [that is, has taken up hang gliding]
This Aux will be do if no other Aux is present:
80
Meaning of “polarity focus”: focus is emphasis on one particular item in a sentence as the new material
being contributed by the speaker. “Polarity” refers here to the “poles” yes and no, affirmative vs. negative.
I wonder if there are any people who grow turnips around here. / Well, Bill does.
It would appear then, that some kind of process provides the aux do as the “backup Aux”
whenever a syntactic transformation is applicable that requires an Aux to apply. A number of
ways to formalize this idea have appeared, but I will not attempt this here, simply noting the
general point that do is the “backup Aux” of English. We can at least state “what happens” as
follows:
For all syntactic rules of English that refer to Aux, the Aux do is inserted prior to their
application when the input sentence contains no Aux.
Having said this, I will mostly avoid sentences that require this unformalized operation in what
follows, for simplicity.
4. Summing up so far
Transformations have been posited to perform a variety of functions, as follows:
• Assignment of inflectional features to morphosyntactic representations (agreement and

case marking, as in the previous chapter)
• Copying of material (as in tags)
• Movement of parts of the tree (as in Yes/No Question Formation)
• Insertion (not formalized here) of semantically empty words like the Aux do.
What remains to be covered are the most dramatic of transformations, the so-called long-
distance movements. These will be covered in the next chapter.
5. Architecture of the theory: deep structure and surface structure
With these rules in place, we can consider the architecture of the theory as developed so far.
By “architecture” I mean the various kinds of rules and the order in which they are arranged; or the
“direction of information flow” that the theory assumes. Such information can be expressed with
diagrams containing boxes and arrows. The following diagram of this sort incorporates the terms
deep structure and surface structure.
(141) An architecture for grammatical theory with deep and surface structure
S (initial symbol fed to the grammar)
Phrase structure rules
(bare tree)
Lexical insertion lexicon
Deep structure
Transformations
Surface structure
(and on to other components: inflectional morphology, phonology)
[phonetic form]
The terms “deep” and “surface” involve no notion of profundity or superficiality; they are defined
as follows:
• Deep structure is the output of the phrase structure rules, with words plugged in by lexical
insertion.
• Surface structure is the output of the syntax as a whole.
In a sentence in which no transformations are applicable, the deep and surface structures would be
the same.
A caution to bear in mind is that a diagram of this sort is simply depicting the logical structure
of the model; we are not (necessarily) making any claim that this represents the time course of
sentence production in the human brain, but only a claim about the structure of the language; that
what we observe can be described in terms of a fixed number of perturbations of a simple structure
that is generable by a phrase structure grammar.
For further reading

The classical work on transformations was done by the linguist Noam Chomsky more than
half a century ago in his book Aspects of the theory of syntax (Cambridge, MA: MIT Press, 1965),
which had a huge influence on the field. The short treatment given in this text is very conservative
in mostly following Aspects. A textbook that goes into much greater depth, and would prepare you
for more modern approaches in syntax, is Dominique Sportiche, Hilda Koopman, and Edward
Stabler (2013) An introduction to syntactic analysis and theory (Oxford: Wiley-Blackwell).
Chapter 6: Syntax III — Subcategorization and Wh- Movement
1. One more update of the phrase structure rules
For purposes of this chapter we will need a slightly more powerful set of phrase structure
rules.
1.1 CP as daughter of NP
In English it is possible to have an NP that contains a CP as its daughter. One place where this
occurs is when the head noun of the NP is one that express a belief or a statement — such nouns
include belief, claim, assertion, and so on. Some examples of CP-within-NP are given in (142).
(142) Some examples of NPs that have CP as a daughter
a. Fred’s belief [CP that he is a genius ]

b. George’s insistence [CP that he be included ]
c. the claim [CP that this territory is an island ]
d. Sally’s assertion [CP that we should eat pasta ]
It should be clear that the fundamental principle of phrase structure we have been working with,
“the modifier of the head is the sister to the head” ((99) above) is obeyed by these examples; for
example that he was a genius specifies what particular belief Fred holds.
We can also establish that the normal position of CP within the NP that includes it is final
position:
George’s insistence [PP to John ] [CP that he be included ]

*George’s insistence [CP that he be included ][PP to John ]
I suggest the following version of the main phrase structure rule for NP:
(143) A modified version of the NP phrase structure rule (allows CP)
Art 
NP → NP   (AP)* N (PP)*(CP)
A A E
To justify this rule, we can consider a long NP that includes every possibility for the main NP
phrase structure rule:
NP
CP
PP VP
NP AP NP NP NP
| | | |
N A N P N Comp N V Art N
| | | | | | | | | |
Bill’s forthright assertion to Fred that television has no future.
Many of the nouns that occur with CP sisters share a morphological property: they are derived (by
a word formation rule) from verbs. This can hardly be a coincidence, and we will discuss this
further below. There are, however, a few nouns that take a CP that are not derived from verbs:
hypothesis, hunch.
1.2 AP as daughter of VP
VP can sometime include an AP (Adjective Phrase). The most common instance of this is
when the verb is some form of be: Alice is quite tall. Sentences with other verbs are given below.
(144) Some sentences in which the VP includes an Adjective Phrase
a. Fred is sick.
b. Bill looks tired.
c. Alice seems very friendly.
d. Jack appeared angry to Sam.
As far has handling such cases in the grammar, it seems sensible not to amplify our existing
phrase structure rule for VP (VP → V (NP) (NP) (PP)* (CP), given in (131)), but rather to
introduce a new rule that only allows V and AP:
(145) A new phrase structure rule for VP, permitting an AP daughter
V → V AP (PP)*
The alternative of beefing up our existing rule VP → V (NP) (NP) (PP)* (CP) with an AP position,
as in VP → V (NP) (NP) (AP) (PP)* (CP), would overgenerate (as with *Bill told Fred Sam very
angry.)
1.3 Final set of phrase structure rules
Putting all of these together, we have the phrase structure rules shown:
(146) Final version of phrase structure rules for English, improved with two new rules
S → NP (Aux) VP
Art 
NP → NP   (AP)* N (PP)*(CP)
 
A A E
NP → Pronoun
AP → (Adv) A
VP → V (NP) (NP) (PP)* (CP)
VP → V AP (PP)*
PP → P NP
CP → (Comp) S
NP → NP (Conj NP)*
VP → VP (Conj VP)*
PP → PP (Conj PP)*
S → S (Conj S)*
CP → CP (Conj CP)*
V → V (Conj V)*
To be sure, this set is “final” only with respect to this textbook; further study of English grammar
would necessitate further additions.
2. Lexical insertion and subcategorization
The phrase structure rules for English as we have developed them so far generate, among
many others, the following trees:
(147) Two trees generated by the phrase structure rules
a. S b. S
NP VP NP VP
| | |
N V N V NP
Art N
These trees are filled by the process of lexical insertion, which is constrained to match
categories: only nouns may be inserted under N, verbs under V, and so on. Yet this appears to be
not the only condition that needs to be imposed on lexical insertion; there are other factors that
constrain it. We can see this when we try out various examples and observe ungrammatical results,
in particular, overgeneration.
Thus, for instance, a verb like sigh may appear in tree (147)a above, but not tree (147)b:
Fred sighed.
*Fred sighed his fate.
A verb like destroy behave in the reverse fashion: it can appear in (147)b but not in (147)a:
Bill destroyed his car.

*Bill destroyed.
To avoid overgenerating in the way just shown, the theory needs a means of specifying the
requirements of particular words the tree structures they may appear in. This problem becomes
especially acute for the two new structures introduced in the preceding section. Only a few nouns
in English (like belief, insistence, claim) can occur with a CP sister. Only a very small number of
verbs (like be, seem, appear) may occur with an AP sister. So our theory currently has a big gap in
it; we need some way of specifying what trees particular words are allowed to occur in. The
method we’ll cover in this book is called subcategorization frames.
To start, let us try to be more specific about the process of lexical insertion, introduced earlier
on p. 104. This concept relies on the idea that speakers possess a mental dictionary, i.e. their
mental lexicon; and that lexical insertion consists of extracting a word from the lexicon and
inserting it into a syntactic tree. To handle subcategorization, we must assume that the lexical
entries of a word contains about what kinds of tree the words can be inserted into. Formalized, this
information is called a subcategorization frame.
Under this approach, the lexical entry for destroy would be like this:
destroy:
Pronunciation: /dsˈtrɔɪ/ (more on this later)

Meaning: “violently cause no longer to exist” (we lack a better way to represent meaning)
Syntactic category: Verb
subcategorization frame: [ ___ NP ].
The subcategorization frame indicate the sisters that must be present in order for the word to
be legally inserted into the tree. Destroy, being a verb, will be inserted as the head of a VP. The
subcategorization frame says that for insertion of destroy to be legal, the VP must contain an NP,
occurring immediately to the right of V within VP. The diagram in (148) is meant to explicate this
notation:
(148) Matching a tree and subcategorization frame
VP
V NP
___ Art N
| |
the city tree
[ ___ NP ] subcategorization frame for destroy
Since destroy subcategorizes for an object NP, a sentence like *John destroyed fails to match
the subcategorization frame of its verb and is therefore ungrammatical. Moreover, a grammar that
includes appropriate subcategorization restrictions will never generate this bad sentence, since it
cannot insert destroy into the tree.
It goes the other way as well: where a subcategorization frame does not include some
particular type of phrase, then lexical insertion is impossible, and ungrammatically is predicted,
when that phrase type is present. Thus, for instance, the intransitive verb sigh would have the
following subcategorization frame:
Here is a lexical entry for sigh:
sigh: pronunciation: /saɪ/

meaning: “exhale loudly to express sorrow”
syntactic category: verb
subcategorization: [ ___ ]
The frame [ ___ ] indicates that sigh may not have sisters in the VP. It accounts for the
ungrammaticality of *John sighed the misfortune.
Verbs like sing, which can occur in either structure (Jane sang, Jane sang the song) are
assigned subcategorization frames that employ parentheses to show the optionality. Here is a
lexical entry for sing:
sing: pronunciation: /sŋ/

meaning: “use one’s voice to produce music”
syntactic category: verb
subcategorization: [ ___ (NP) ]
Grammaticality results if some version of the frame (leaving out, or keeping in, parenthesized
material) matches the sentence. Thus Jane sang and Jane sang the song are both good.
The question of whether a verb subcategorizes for an object comes up sufficiently often that
special terminology exists for it. Verbs like destroy that must take an object are called transitive
verbs; verbs like sigh that cannot take an object are called intransitive. Verbs like sing, which fit
into both categories, could called “optionally transitive”.
2.1 Other instances of subcategorization
Verbs of saying and belief often subcategorize for a CP. For example, say has the
subcategorization [ ___ (PP) CP ] and tell has the subcategorization [ ___ NP (CP)]. This can be
justified by the following sentences:
(149) Data to justify the subcategorization frames of say and tell
a. *Alice said.
*Alice said to Bill.
Alice said that she would be going.
Alice said to Bill that she would be going.
b. *Fred told.
*Fred told that he would be going.
Fred told us.
Fred told us that he would be going.81
Nouns also have subcategorization frames. For example, here are the data that could be used
to justify the subcategorization frame of gift as [ ___ (PP) (PP) ]
(150) Data to justify the subcategorization frame of gift
a. a gift of $10 to the Red Cross

b. a gift of $10
c. a gift to the Red Cross
d. a gift
The subcategorization frame of picture is [ ___ (PP) ], as in picture of Alice (again the PP is
optional since picture is fine by itself.) The subcategorization of dog is [ ___ ] (there are no noun
phrases like, say, *dog of teeth). A curious fact about noun subcategorization is that the
subcategorized phrases seldom if ever obligatory; all the phrases mentioned will typically appear
with parentheses.
81
Tell also has a second subcategorization [ ___ (NP)(NP) ]: Fred told us his sorrows, Alice told them
her name.
I introduced the two changes in the phrase structure rules of the previous section
Art 
(NP → NP   (AP)* N (PP)*(CP) and VP → V AP (PP)*) precisely because these rules seem to
 
A A E
be especially “sensitive”, as it were, to subcategorization. There are only a few nouns that take CP
and only a few verbs that take AP; see (142) and (144) for examples.
With regard to N that subcategorize for CP, there is an intriguing relationship to the rules of
word formation (see Chapter 2). Specifically, when a noun is morphologically derived from a verb
that subcategorizes for CP, the resulting noun tends also to subcategorize for CP:
(151) Parallel subcategorization of some verbs and their derived nouns
a. They believe that Sue left

their belief that Sue left
b. They assert that Sue left.

their assertion that Sue left
c. They claim that Sue left.

their claim that Sue left.
2.2 Relative clauses are not subcategorized
Before we leave the topic of N subcategorizing for CP, it’s important to note that there is
another kind of CP that can occur as part of an NP. These structures are called relative clauses,
and they fall outside the scope of this text. You can detect relative clauses because they have a
silent location (rather like the “implicit noun phrases” of Chapter 1, section 4) that refers to the
head noun of the NP; often called a gap.
(152) Examples of relative clauses (gaps shown as ___)
a. the turtles [CP that we caught ___ in the pond ]

b. the beliefs [CP that we hold ___ ]
c. the king [CP who the peasants deposed ___ last year ]
Thus (123) means something like “the turtles such that we caught (those turtles) in the pond”; I’ve
filled the gap informally by spelling it out as “those turtles.”
A key point about relative clauses is that they are not subcategorized. Pretty much any noun,
occurring as the head of NP, can have a relative clause. To give an example, the noun book does
not subcategorize for a CP, even if we can think of a meaning that might be expressed by such as
construction:
*the book that linguistics should be taught in high school.
Yet book can occur as the sister of relative clause (notice the gap):
the book [CP that we read ___ in school ]
Another reason to distinguish subcategorized CP from relative-clause CP is that the same

noun can have both.
(153) An NP that has both a subcategorized CP and a relative clause
the assertion [CP that we should eat pasta ] [CP that you made ___ ]
head of NP
Subcategorized CP CP forming a relative clause
Though we have no space for relative clauses here, they are a major topic in syntax that are likely
to encounter if you go on to take a syntax course. The usual analysis for them is rather similar to
that of “wh-word” questions, which also have gaps and which we will cover below.
2.3 Items not included in the subcategorization frame
Some constituents evidently get to appear “for free” in the syntactic tree; they don’t have to be
subcategorized. This is true for PP’s with general adverbial meaning of place, time, or manner can
occur with virtually any verb:
(154) Some unsubcategorized PP
John sighed on Tuesday.

John sighed in the garden.
John sighed with great feeling.
The general practice for subcategorization is this: if any element is always able to occur as a
sister, then we don’t bother to mention it in the subcategorization frame. Basically, we are
interested only in the restrictions that hold of individual words. This aspect of the grammar will not
be formalized in this text.82
What is true of verbs is also true of nouns: PP’s of place, time, and manner are ignored in
determining noun subcategorizations, so cases like (126) would not justify a frame like [ ___ PP ]
for their nouns.
(155) Some unsubcategorized PP within NP
the dog in the garden

the party on Tuesday
a person in a good mood
82
In a more thorough grammar, we might adopt a bit more structure: some kind of node higher than
VP but lower than S, which would contain the unsubcategorized PP.
Likewise, articles and possessors are not considered in the subcategorization frame, since they are
possible for any noun (the dog, Alice’s dog).
2.4 Subcategorization and meaning
It’s a somewhat vexed question to what extent subcategorization should be treated (as it is
above) as a straightforward matter of syntax. An alternative view is that heads occur in particular
syntactic locations simply because of what they mean. For example, the verb say is entitled to
occur in the syntactic frame [ ___ PP CP ] because an act of saying generally has someone who is
being spoken to (in I said to Fred that I was leaving, this is Fred), and a thing which is said (I was
leaving). Similarly, put occurs [ ___ NP PP ] because it involves a thing that is put, and location
into which the thing is put. Sigh occurs [ ___ ] because nothing is affected when you sigh.
Although there is probably a grain of truth to this “semantics, not subcategorization” view,
there are also reasons to treat it with skepticism.
First, there are cases of verbs that have very similar meanings, but different patterns of
occurrence. Consider for instance tell and say.
I told Bill that I was leaving.

*I told to Bill that I was leaving.
*I said Bill that I was leaving.
I said to Bill that I was leaving.
It’s not clear how semantics alone could tell us which verb requires an NP object and which a PP.
Rather, it is a somewhat arbitrary fact that tell has the subcategorization frame [ ___ NP (CP) ] and
say has the frame [ ___ (PP) CP ]. Likewise, the pattern below:
I like jumping. I like to jump.

I prefer jumping. I prefer to jump.
I enjoy jumping. *I enjoy to jump.
where only one of the three similar verbs can’t take an infinitive subordinate clause (see Chapter
1), suggests that meaning won’t suffice to tell us everything about subcategorization.
The verbs give and donate are semantically similar, but have different syntactic behavior:
She gave the library $1,000,000. She gave $1,000,000 to the library.
*She donated the library $1,000,000. She donated $1,000,000 to the library.
Only give has the subcategorization [ ___ NP NP ].
There is one more phenomenon that suggests that subcategorization cannot be reduced to
meaning. Consider verbs like these:
He ate.
She sang.
We raked.
These have what are sometimes called “implicit arguments”—it’s understood that he ate
something; and that likewise she sang something (song unspecified), and we raked (leaves or grass
unspecified). In other words, the syntax does not always have to provide overt expression for all
the participants in an act. Yet in other cases, an implicit argument evidently is not allowed:
*We took.
*We own.
We deal with this by letting NP be optional in the subcategorization frames for eat, sing, and rake
([ ___ (NP) ] but not for take or own ([___ NP]).
2.5 Solving subcategorization problems
The best method seems to be the following:
• Think of words and sentences that include the word you’re considering. 83
• Look at the phrase structure rule that introduces the word (for example, if you’re dealing
with a noun, look at the phrase structure rule NP → (Art)(AP)* N (PP)* (CP)). This will
tell you the sisters that at least might be present.
• Remember that a subcategorized expression usually has a kind of intimate relation to the
meaning of the word that subcategorizes it. The noun claim subcategorizes for an CP
because the CP is used to designate the conceptual content of the claim.
• Try collecting as many individual frames for the word that you can, then use parentheses to
collapse them into one or more simpler expressions.
• Don’t be distracted by PP’s of place, manner and time that can occur with anything; they
don’t belong in the subcategorization frame.
Study Exercise #23
Give subcategorizations for the following words, justifying them with example sentences.
a. Verbs: elapse, award, tell, shout, die

b. Nouns: turtle, bowl, announcement, reason
83
I admit that this is harder for non-native speakers, a problem hard to avoid in linguistics teaching. If
you don’t have native intuitions in English, I suggest doing one of two things when you solve
subcategorization problems on your assignments: either find a native speaker consultant and get their
intuitions, or else add verbal discussion to your answer, with wording like “assuming that xxx is grammatical
in English; I’m not sure.” It would be fair to grade your answer based on the facts as you give them.
a. Verbs:
elapse: [ __ ]
Time elapsed
*Time elapsed me
*Time elapsed to the losing team.
*Time elasped that it was a great misfortune.
award: [ ___ NP (PP)]

*They awarded.
They awarded a prize
They awarded a prize to the winner.
*They award to the winner.
[ ___ (NP) NP ]
They awarded the winner a prize.
tell: [ ___ NP (CP) ]

They told Bill that they were leaving.
They told Bill.
*They told that they were leaving.
[ ___ NP (PP) ]
They told the truth to Bill.
They told the truth.
They told Bill.
*They told to Bill.
[ ___ NP (NP)]
They told Bill the truth.
They told Bill.
They told the truth.
shout: [ ___ (PP)(CP)]

They shouted.
They shouted to Sally.
They shouted that they were leaving.
They shouted to Sally that they were leaving.
[ ___ (NP)(PP)]
They shouted the words.
They shouted the words to Sally.
die: [ ___ ]
Jefferson died.
*Jefferson died Washington.
*Jefferson died to Washington.
(Note: Jefferson died in 1826, Jefferson died in Virginia don’t count, since PP’s of place
and time can occur with any verb.)
b. Nouns
turtle: [ ___ ]
turtle
*turtle of shell
*turtle that they were leaving
bowl: [ ___ (PP) ]

bowl
bowl of soup
announcement: [ ___ (PP) (CP) ]

the announcement
the announcement to Bill
the announcement that they were leaving
the announcement to Bill that they were leaving
reason: [ ___ (CP) ]

the reason
the reason that we are going
*the reason to Fred
*the reason to Fred that we are going
————————————————————————————————————
3. Wh-Movement
3.1 Backdrop
This section returns to the topic of transformations. Thus far, we’ve seen two reasons to move
beyond simple phrase structure grammars to transformational grammars:
• Phrase structure rules cannot copy material—only a copying transformation can generate
the legal array of tag questions.
• Phrase structure rules cannot relate sentences to one another (for example, simple
statements to yes-no questions).
We now move on to what many linguists would probably agree is the most important basis for
transformations, sometimes called “long distance filler-gap dependencies”. The first example of
such a case will be Wh- Movement.
3.2 Basics and terminology
A wh- word is one of a fixed inventory of words used for asking questions. They are so called
because most of the wh-words in English begin with these letters.
The wh- words of English can be various parts of speech:
which Article
whose Article
who Pronoun
whom Pronoun
what either an Article or Pronoun
how Adverb
when Adverb
why Adverb
where Adverb
A wh- question is a question that involves a wh-word. For example, the following are wh-
questions:
Who did you see?

What book did you read?
Which chocolates did you like?
In which hotel are you staying?
How do you feel?
You can see that the wh- word usually comes at or near the beginning of the sentence. It
constitutes, or is part of, a phrase that (intuitively), the sentence is about; i.e. the item that is being
questioned.
A wh- phrase is an NP, PP, or AdvP (Adverb Phrase) that contains a wh- word and is placed
at the beginning of a clause. In the wh- questions just mentioned, the wh- phrases are
PP
NP NP NP NP AdvP
| |
Pro Art N Art N P Art N Adv
| | | | | | | | |
who what book which chocolates in which hotel how
This permits a more precise definition of wh- question; it is a question that begins with a wh-
phrase.
3.3 Wh- questions, subcategorization, and gaps
Wh- questions are interesting in that they appear to violate otherwise-valid principles of
subcategorization. Here is an example. The verb put has the subcategorization frame
[ ___ NP PP ]. Because of this a sentence like the following:
Fred will put the chicken in the oven
is grammatical; the subcategorization of put is satisfied. But
*Fred will put in the oven
is ungrammatical because of the missing NP, and
*Fred will put the chicken
is ungrammatical because of the missing PP.84 This is an unusual case in which there are two
subcategorized elements and both are obligatory.
In light of the subcategorization fact, it is a bit surprising that the following sentences, both
Wh- questions, should be grammatical:
(156) Two Wh-questions that appear to violate subcategorization
Into what oven will Fred put the chicken?

What chicken will Fred put into the oven?
These sentences contain gaps: instead of the NP or PP that the subcategorization of put calls for,
one finds nothing. The gaps are shown below, denoted with an underscore:
84
As elsewhere we are ignoring extended uses of verbs, which often change the subcategorization.
John put the chicken is fine in a fantasy world in which Olympic medals are awarded in the chicken-put.
(157) Two Wh-questions that appear to violate subcategorization — marked for their gaps
What chicken will Fred put ___ into the oven?

Into what oven will Fred put the chicken ___ ?
Intuitively, it seems that these gaps are “filled” by the wh- phrase. We understand what chicken to
be the object of put in the first sentence, and in the second sentence we understand into what oven
to be the PP indicating where Fred put the chicken.
Let us define “gap”, for precision:
• A gap is a location in syntactic structure where the subcategorization requirements would

lead one to expect a phrase, but none occurs.
Such gaps are widely observed in English and in many other (not all) languages.
There is an intimate connection between wh- phrases and gaps: to a rough approximation,
gaps are allowed only when a wh- phrase is present; recall
*Fred will put in the oven

*Fred will put the chicken
This goes the other way around: if there is no gap, then we can’t have the Wh-phrases either:
*What chicken will Fred put the dinner into the oven?
*Into what oven will Fred put the chicken into the pan?
Moreover, most people who ponder the question will judge that gaps are somehow “filled” by the
wh- phrase. In (158):
(158) What chicken will Fred put ___ into the oven?
we understand what chicken to be the object of put, and in (159):
(159) Into which oven will Fred put the chicken ___ ?
we understand into which oven to be the PP indicating where Fred put the chicken.
Summing up, wh-questions in English have what are often called filler-gap dependencies,
which we can detect more rigorously by working out subcategorizations.
The two questions that need to be answered, then, are
• Why should wh- questions, and only wh- questions, permit gaps?
• How do we account for the filler-gap dependency; that is, the fact that the wh- phrase at the
beginning of the sentence intuitively fills the gap?
As you might be imagining already, the answer will involve a transformation.
3.4 Further background: echo questions
Before we proceed to the analysis, let us ponder a further phenomenon of English syntax, the
so-called echo question. These are questions that contain a Wh- phrase, but have no gap; the Wh-
phrase occurs in the ordinary position for its type, and satisfies the subcategorization requirements
of the relevant head. Echo questions are not all that common, because they can only be used to
offer an astonished reply to a parallel statement:
(160) Examples of echo questions
The Romans destroyed the television set.

The Romans destroyed what?
I saw Marilyn Monroe in Westwood last Saturday.

You saw who?
Fred will put the chicken in the Socratic Oven.

Fred will put the chicken in what oven?
Echo questions make an important point: it is possible to generate a wh-phrase in the ‘normal’
position for an NP or PP; wh- phrases do not always have to appear at the beginning of sentence.
A bit of terminology: the wh- phrases of echo questions are sometimes said to be in situ,
which is Latin for “in its original position”.85
3.5 A transformation for Wh- questions
The grammatical problem at hand is that Wh- questions have subcategorization gaps that
match up with the initial wh- phrases. This is a dependency that cannot be expressed with the
phrase structure rules we have been using. These rules can only say what daughters a node may
have, and thus they have no ability to regulate matchups between elements in the tree that are far
apart. A transformation is needed.
The intuitive idea behind our transformation analysis will be to let normal questions be
derived from deep structures that look like echo questions (for “deep structure”, see p. 173 above),
so that the Wh- phrase occurs in situ. To these deep structures, we will apply a transformation that
moves the wh- phrase out of its deep structure position (where it satisfies the subcategorization of
the verb) to the beginning of the sentence. In (161) I give preliminary version; this will be refined
later on.
(161) Wh- Movement (preliminary version)
Move a wh- phrase to the beginning of the sentence as daughter of S, leaving a trace.
85
Situ is an inflected form ([Case:ablative, Number:singular]) of situs ‘place’.
The term “trace” will be defined shortly.
Very shortly, we will use this transformation to derive some wh- questions, but a detail must
first be attended to. It’s clear that in typical Wh- questions (such as (127)), the Aux comes before
the subject: What chicken [ will ]Aux [ Fred]NP put in the oven? This is hardly something new,
because we have already seen this in Yes/No questions, discussed in the previous chapter. It seem
that the flipping of the order of subject and Aux is more general than we had imagined; it occurs in
all normal (non-echo) questions, not just Yes/No questions. So a first act of tidying up will be to
rename our earlier transformation of Yes/No Question Formation (from (111)), and assume that it
applies in all normal questions. Here is the transformation with its new name.
(162) Subject/Aux Inversion
S S
| |
1 2 3 2 1 3
Applies in all non-echo questions.
We will assume, moreover, that Subject/Aux Inversion is ordered to apply before Wh-
Movement; the concept of ordering the rules is one we’ve already seen for inflectional morphology
in Chapter 2.
Now that we have the rules and their ordering, we can do a full derivation of a Wh- question
under the proposed analysis. We begin with the phrase structure rules (on the left), then do lexical
insertion (on the right):
(163) Application of phrase structure rules and lexical insertion

S S
| |
NP Aux VP NP Aux VP
| | |
N V NP PP N will V NP PP
| |
Art N P NP Fred put Art N P NP
| | |
Art N what chicken in Art N
| |
the oven
This creates the stage of deep structure, with what chicken in situ. The crucial point at this stage is
that we have not violated the subcategorization of put, which in deep structure does have the
required NP and PP sisters. In fact, with the theory we are working on, ultimately this will be seen
to be true even in surface structure (more on this below).
Following our assumed rule ordering, the first transformation to apply is Subject/Aux
Inversion, from (162):
(164) Application of Subject/Aux inversion

S S
| |
1 2 3 2 1 3
Deriving:
S S
| |
1 2 3 2 1 3
Now we apply Wh-Movement. I show this below first by drawing arrows to show what moves
where, then showing the surface structure that results. A caution: the destination of what chicken
is provisional; we will change the analysis a bit below.
(165) a. Output of Subject/Aux Inversion, with arrow showing application of (130) Wh-
Movement
S
|
Aux NP VP
| |
will Fred V NP PP
|
put Art N P NP
| | |
what chicken in Art N
| |
the oven
b. Surface structure
S
NP Aux NP VP
| |
Art N will Fred V NP PP
| | | |
what chicken put t P NP
|
in Art N
| |
the oven
As stated in the Wh- Movement rule of (130), the movement of what chicken is assumed to
leave a trace. A trace is more or less our formalization of a gap: it is an empty copy of what got
moved; it has the same category, but it contains no phonetic material. To show that a trace is
empty, we use the letter t, as the daughter of the trace’s category. For now, the trace is just an
arbitrary choice, but we later on it will play an important role in the semantics of wh- questions
and similar constructions. For now, we can observe that the trace NP means that the
subcategorization requirements of put are satisfied (albeit by an empty, abstract entity) at surface
structure as well as deep structure.
To summarize: the goal of the analysis has been to provide a solution to the problem of
subcategorization gaps, and why these gaps characteristically are matched with a Wh-phrase at the
start of the sentence. In this approach, gaps only arise from movement,86 so the fronted wh- phrase
will always match the gap. This ability to capture a long-distance dependency (“X over here only
if Y over there”) is a common justification for a transformational analysis.
86
A caution: there are many other sources of gaps, such as the subject gaps mentioned in Chapter 1,
or the dropped subject pronouns of Spanish, Persian, and many other languages. But these tend to have a
special distribution, so the general point still holds.
Study Exercise #24
Provide a step-by-step syntactic derivation, mimicking that just given in the text, for
Into what oven will Fred put the chicken?
Note: as elsewhere in linguistics, when you’re asked to provide a derivation, a useful strategy is to
first do the derivation “backwards” in your head, to figure out the starting point (here, deep
structure). Then do the derivation “forwards” on paper.
Study Exercise #25
Explain how our grammar predicts that
*What city have the Romans destroyed Carthage?
is ungrammatical.
Study Exercise #26
Explain how our grammar predicts that
*Who will the princess sigh?
is ungrammatical.
Study Exercise #27
Derive the question
Who will leave?
using the rules given above. What is odd about this derivation?
Phrase structure rules:
Lexical insertion:
Application of Subject/Aux inversion:

S S
| |
1 2 3 2 1 3
Deriving:
S S
| |
1 2 3 2 1 3
Output of Subject/Aux Inversion, with arrow showing application of (130) Wh-Movement
Surface structure results from Wh-Movement, including trace:

This sentence is a wh- question. Accordingly to our analysis, the wh- phrase in such a question
must have originated in deep structure in some position inside the sentence. But there cannot be
any such position. The subject position is already filled by the Romans, and the verb destroy
subcategorizes for only one sister NP position, which is already occupied by the NP Carthage.
Since our grammar cannot generate an appropriate deep structure, it is unable to generate the
surface structure. It therefore predicts that the sentence should be ungrammatical.
This sentence has essentially the same problem as in the previous Study Exercise: there is no
place that the NP ‘who’ could have come from: the subject position is already taken up by the
princess, and sigh doesn’t subcategorize for any sister NPs. Thus there is no possible deep
structure, so our grammar cannot generate the surface structure. It therefore predicts
ungrammaticality.
a. Deep structure, showing Subject/Aux Inversion:
S
|
NP Aux VP
| | |
Pro will V
| |
who leave
b. Result of Subject/Aux Inversion, showing Wh-movement
S
|
Aux NP VP
| | |
will Pro V
| |
who leave
c. Surface structure
S
NP Aux NP VP
| | | |
Pro will t V
| |
who leave
What’s odd? The input reads just like the output! This is sometimes called a string-vacuous
derivation; the surface structure word order hasn’t changed (trace being silent), but the structure is
different.
String-vacuous derivations test the ability of the student to “think like a computer” (see p. 33
above). As humans, we may sometimes feel that derivations that in the end do essentially nothing
waste our time. But the real payoff here is that we’ve applied the rules of the grammar, showing
we got the right answer — what we win from the seemingly pointless activity is reassurance. And
the rules, of course, aren’t pointless because often the derivations they create are not string-
vacuous at all.
_________________________________________________________________________
4. The “landing site” of Wh- Movement
Wh- Movement doesn’t always move words to the beginning of the sentence. In so-called
embedded Wh- questions, movement is to the beginning of a subordinate clause, as in the
following examples.
(166) Some instances of embedded Wh- questions
I wonder what city the Romans destroyed.

We asked for whom the bell tolls.
They are found when the main clause has a verb like wonder and ask, which takes a question as its
sister node. I’ll assume that these verbs have a special categorization, not formalized here, under
which they take a CP that is not a declarative (the usual case), but a wh- question.
A further observation about embedded questions is that they don’t occur with the
complementizer that:
*I wonder what city that the Romans destroyed.

*We asked for whom that the bell tolls.
*I wonder that what city the Romans destroyed.

*We asked that for whom the bell tolls.
These facts suggest a refinement of our analysis of Wh- Movement. An influential idea in
syntactic theory is that the order of words in sentences can be explicated in terms of slots, which
the words compete to fill. We’ve already said that the Complementizer that occupies the position
Comp, a daughter of CP. The idea to be developed here is that in an embedded Wh- question, the
moved Wh- phrase actually occupies the Comp slot. When Comp is thus occupied, there is no
room for that (there’s no problem in leaving it out; it is semantically empty in any event).
Under this approach, we can arrange lexical insertion simply to leave Comp empty for
embedded clauses introduced by verbs like wonder and imagine. Then, Wh- Movement acts to fill
the empty slot by moving the wh- phrase into it, as follows:
(167) A derivation, using empty Comp, of I wonder what city the Romans destroyed?
a. Deep structure
S
VP
CP
VP
NP NP NP
|
Pro V Comp Art N V Art N
| | | | | | |
I wonder the Romans destroyed what city
b. Wh- Movement and surface structure
VP
CP
VP
NP NP
|
Pro V Comp Art N V NP
| | | | | | |
I wonder NP the Romans destroyed t
Art N
| |
what city
In this approach, the empty Comp node provides a kind of “landing site” for the moved Wh-
phrase.
4.1 Cleaning up the analysis I: fixing Subject/Aux inversion
Before going on, I should confess to a minor cheat: in the sentence just derived, I simplified
matters by leaving out an Aux. In fact, there’s a nice puzzle at hand here: if there is an Aux in a
subordinate clause, it does not flip with the subject as it would in a main clause. Here is the crucial
comparison:
(168) Aux inverts with subject only in main clauses
a. Main clause
What city [have]Aux [ the Romans ]NP destroyed?
b. Subordinate clause
I wonder what city [ the Romans ]NP [have]Aux destroyed?
Let us modify Subject/Aux Inversion to acknowledge this fact.

(169) Subject/Aux inversion (revised)

S S
| |
1 2 3 2 1 3
Restrictions:
Applies in non-echo questions.
Applies in main clauses only
Study Exercise #28
Give an example, based on the sentence in the text, of what ungrammatical sentences would be
generated if we allowed Subject/Aux Inversion to apply in subordinate clauses.
*I wonder what city [have]Aux [ the Romans ]NP destroyed.
Note that the intended reading is different from the one you would spell with a comma, colon,
or three periods:
(Y’know), I wonder, what city have the Romans destroyed?

(Y’know), I wonder: what city have the Romans destroyed?
(Y’know), I wonder … what city have the Romans destroyed?
It seems that in these cases, the sequence is being treated as two sentences: I wonder and What
city have the Romans destroyed? In these case the use of Subject/Aux Inversion is completely
expected.
————————————————————————————————
4.2 Cleaning up the analysis II: Non-embedded Wh-questions
So at this point we have a working analysis for embedded Wh- questions. But to be consistent,
we also need to cover the wh- questions that are not embedded, that is, the ones we started out
with. There is a fairly reasonable tack that can be taken here, namely that these sentences also have
Comp, which provides the landing site for the sentence-initial wh-phrase. Specifically, the
assumptions we need to make are as follows:
• Wh- questions are not instances of S, but of CP.

• They require (by means not stated here) that the initial Comp be empty in Deep Structure.87
Under this analysis, the derivation of What chicken will Fred put in the oven? comes out
slightly differently:
87
There are alternatives to this, for instance letting the moved Wh- phrase displace a that, and adding
a transformation that deletes that from the topmost complementizer of the sentence.
(170) Revised syntactic derivation for What chicken will Fred put in the oven?, using empty Comp
a. Deep structure
CP
Comp VP
PP
NP NP NP
|
N Aux V Art N P Art N
| | | | | | | |
Fred will put what chicken in the oven
b. Subject-aux inversion
CP
Comp VP
PP
NP NP NP
|
Aux N V Art N P Art N
| | | | | | | |
will Fred put what chicken in the oven
c. Wh- movement (and Subject-Aux inversion), yielding surface structure
CP
Comp VP
| PP
NP
NP NP
Art N |
| | Aux N V NP P Art N
what chicken | | | | | | |
will Fred put t in the oven
With this in mind, we can express the Wh- Movement transformation more explicitly.
(171) Wh- Movement (revised version)
Move a wh-phrase into an unfilled Comp, leaving a trace.
5. Typology of Wh-movement
Many languages other than English form Wh- questions by moving the wh- phrase to the
beginning of the sentence. Here are three examples:
French: Tu as vu Paris (normal statement)

you have seen Paris
Quelle ville as- tu vu t ? (wh- question)

what city have you seen (trace)?
Chamorro (Austronesian, Guam (South Pacific)):
Hafahan si-Maria i-sanhilo gi tenda (normal statement)

bought Maria the-blouse at the-store
‘Maria bought the blouse at the store’
Hafa hafahan si-Maria t gi tenda (wh- question)

what bought Maria (trace) at store
‘What did Maria buy at the store?’
Vata (Ivory Coast, West Africa) Tones: 1 = highest, 4 = lowest
41 3 2 1
Kofi le saka (normal statement)
Kofi ate rice
3 41 3 2
yi Kofi le t la (wh- question)
what Kofi eat (trace) question-particle
‘What did Kofi eat?’
Many other languages work in the same way; for example Modern Hebrew, Russian, and Spanish.
However, a large number of languages do not have Wh-Movement. These languages form
Wh- questions simply by leaving the Wh- phrase in situ. An example of a non-Wh- Movement
language is Persian:
Ali an ketab-ra xand

Ali that book read
‘Ali read that book’
Ali tʃe ketab-i xand?88

Ali what book-indef. read
‘What book did Ali read?’
*tʃe ketab-i Ali xand?

what book Ali read
Japanese is similar:
John-wa naze kubi-ni natta no?

John-Topic why was fired question particle
‘Why was John fired?’
Bill-wa [ John-ga naze kubi-ni natta tte ] itta no?

Bill-TOP John-nom why was fired Comp said question particle
‘Why did Bill say that John was fired?’89
88
If you’re thinking about case marking here, the answer to your question is that the Accusative suffix
-ra only attaches to definite Noun Phrases, the kind that would be translated with the in English. In the wh-
question, the expression for ‘what book’ is indefinite and takes the indefinite suffix.
89
An odd custom of linguists writing in English about Japanese syntax is to use English first names.
It’s striking that the languages seem to pattern together; for instance, unbounded movement to
the right is apparently exceedingly rare.90 Moreover, there are logical possibilities for Wh-
movement that seem to be unattested:
• *Move a wh- phrase to the exact middle of a sentence.

• *Move a wh- phrase so that it follows the second word of a sentence.
• *Move all the words that precede the wh- phrase so that they follow the wh- phrase, and
move all the words that follow the wh- phrase so that they precede the wh- phrase.
No such rules have been found in any language. We will discuss such cross-linguistic patterns in
greater detail later on.
6. Why Wh- Movement?
It’s something of a puzzle why languages have Wh- Movement at all—why not adopt the
sensible Persian/Japanese/Chinese strategy, and just leave your Wh- words in situ? Surely it would
be clearer for the listener to interpret the wh- word in its proper syntactic location.91
A clue, I think, can be found in pairs of sentences that have the same gap, but where the Wh-
phrase appears in a different location:
[ What song ] can Sue imagine that Bill sang t ?

Sue can imagine [ what song ] Bill sang t ?
Such pairs are often said to illustrate a difference of scope: the location of the wh-phrase indicates
the domain in which the wh- phrase is acting as a logical operator. Thus, in the first sentence
above, the wh- phrase what song is used to ask something about the content of Sue’s imaginings—
its scope is the entire sentence. The second sentence reports a thought of Sue’s. Within this
thought, what song is being used to ask something about Bill’s singing (that is, Sue is mentally
answering the question, “What song did Bill sing?”). Therefore, the scope of what song in the
second sentence is just the subordinate clause. It can be seen, then, that the linear position of the
wh- phrase is suited to expressing a distinction of scope. (We will cover more about scope later on
when we turn to semantics.)
What emerges, if this speculation is correct, is that there’s no perfect design available.
Languages without wh- movement make it clear where the inherent location of the wh- phrase is,
but are less clear in indicating scope; languages with wh- movement mark scope clearly, but
impose a burden on listeners, who need to carry out gap detection.
90
Proposed instances are in Navajo, Circassian, and American Sign Language. Such claims often
trigger scholarly replies suggesting alternative interpretations of the data.
91
Indeed, experimental work by psycholinguists has documented the increased cognitive load and
memory burden that listeners experience when they have heard a wh- phrase and are “looking for” the
corresponding gap later in the sentence.
7. The unbounded nature of Wh- Movement
An important aspect of Wh- Movement is that it can move a wh- phrase over very long
stretches of syntactic structure. Consider the following deep structures and corresponding surface
structures:
You have seen who.

[ Who ] have you seen t ?
Joan thinks that you have seen who.

[ Who ] does Joan think that you have seen t ?
Bill would imagine that Joan thinks that you have seen who.
[ Who ] would Bill imagine that Joan thinks that you have seen t ?
Sally believes that Bill would imagine that Joan thinks that you have seen who.
Who does Sally believe that Bill would imagine that Joan thinks that you have seen?
Study Exercise #29
Provide a syntactic derivation (that is, deep structure, arrows showing what moves where, surface
structure) for the sentence ‘What city will Fred say that Judy thinks that you live in?’.
Deep structure. The wh- phrase is in situ, so that the preposition has an object. The arrow
shows the movement attributed to Subject/Aux Inversion:
CP
Comp S
VP
CP
VP
CP
VP
PP
NP NP NP NP
| | |
N Aux V Comp N V Comp Pro V P Art N
| | | | | | | | | | | |
Fred will say that Judy thinks that you live in what city
Output of Subject/Aux Inversion, with arrow showing action of Wh- Movement:
CP
Comp S
VP
CP
VP
CP
VP
PP
NP NP NP NP
| | |
Aux N V Comp N V Comp Pro V P Art N
| | | | | | | | | | | |
will Fred say that Judy thinks that you live in what city
Surface structure, with trace of what city:
CP
Comp S
NP VP
Art N CP
what city S
VP
CP
VP
PP
NP NP NP
| | |
Aux N V Comp N V Comp Pro V P NP
| | | | | | | | | | |
will Fred say that Judy thinks that you live in t
———————————————————————————————————————————————————————————————————
Study Exercise #30
Provide a syntactic derivation (that is, deep structure, arrows showing what moves where,
surface structure) for the following sentences:
a. Which book will Sue ask that we study?

b. Sue will ask which book we should study.
Assume the transformations of Subject/Aux Inversion and Wh- Movement.

a. Which book will Sue ask that we study?
Deep structure, with which book in situ.

Note presence of empty Comp, the “landing site” for wh-phrases.
Arrow shows application of Subj/Aux Inversion:
CP
Comp S
VP
CP
VP
NP NP NP
| |
N Aux V Comp Pro V Art N
| | | | | | | |
Sue will ask that we study which book
Output of Subject/Aux Inversion. Arrow shows application of Wh-Movement, moving the

wh- phrase into Comp.
CP E
Comp S
VP
CP
VP
NP NP NP
| |
Aux N V Comp Pro V Art N
| | | | | | | |
will Sue ask that we study which book
Surface structure, with wh- phrase in Comp and a trace left behind:
CP
Comp S
|
NP VP
Art N CP
| |
which book S
VP
NP NP NP
| | |
Aux N V Comp Pro V t
| | | | | |
will Sue ask that we study
b. Sue will ask which book we should study
Deep structure, with which book in situ.

Now the empty Comp is an embedded Comp.
The full sentence, being a statement, is an S, not a CP.
Subject/Aux Inversion ((201)) does not apply in embedded questions.
Arrow shows application of Wh-Movement:
VP
CP
VP
NP NP NP
| |
N Aux V Comp Pro Aux V Art N
| | | | | | | |
Sue will ask we should study which book
Surface structure:
S
VP
CP
VP
NP NP NP
| | |
N Aux V Comp Pro Aux V t
| | | | | | |
Sue will ask NP we should study
Art N
| |
which book
________________________________________________________________________
8. Another transformation: Topicalization
English has a number of transformations similar to Wh- Movement. Perhaps the simplest is
the so-called Topicalization rule, used to account for sentences like these:
Linguistics, I can teach.

Those guys we would never give our credit cards to.
In that oven you should never put a chicken.
The name of the rule is from that fact that the fronted NP serves as the “topic” of its sentence; what
it is about.92 These sentences have a distinctly rhetorical character, and often sound best if you
imagine that the topic is being contrasted with some other topic:
Postmodernism, I’m clueless about, but linguistics, I can teach.
The “landing site” for fronted topics is not Comp, since you can get both that and the fronted
topic in sequence:
I’d say that linguistics, I can teach.
This is of course very different from Wh- Movement, where you never get both at once, leading us
to set up an analysis in which the wh- phrase moves into Comp (see (171), on p. 204).
Thus I will state the rule of Topicaliziation as simply moving a phrase to the left edge of S, as
follows:
(172) Topicalization
Move an NP or PP to the left edge of an S, making it the daughter of S, and leaving a trace.
The justification for Topicalization is much the same as that for Wh- Movement: the presence of a
topicalized element is correlated with a subcategorization gap later in the sentence.
Topicalization, like Wh- movement, appears to be unbounded, though the examples that show
this tend to be a bit less natural:
John, I don’t think a lot of people would like.

Fred, I’d imagine that you’d think that a lot of people wouldn’t like.
As unbounded transformations, Wh- Movement and Topicalization (as well as others to come)
have some crucial similar behaviors, which we’ll examine later on in discussing “islands”.
92
I noted earlier that in normal NP + VP structure, the NP is what the sentence is about. Topicalization more or
less overrides this, letting some other constituent be designated as what the sentence is about.
Study Exercise #31
Derive I’d say that linguistics, I can teach.

I’d say that linguistics, I can teach.
Deep structure, with linguistics in situ.

Application of Topicalization is shown.
Note that movement is simply to the beginning of S, not into Comp as with Wh-Movement.
VP
CP
VP
NP NP NP
| | |
Pro Aux V Comp Pro Aux V N
| | | | | | | |
I would say that I can teach linguistics.
Surface structure, with trace.
VP
CP
VP
NP NP NP NP
| |
Pro Aux V Comp N Pro Aux V
| | | | | | | |
I would say that linguistics I can teach t
________________________________________________________________________
9. It-Clefting
We will cover one more long-distance movement rule, the “It-Clefting” process that was
briefly discussed above in Chapter 4, section 11. The sort of data that justify the rule are given
below.
a. Sean loaded the tricycles into the truck.

It was Sean that loaded the tricycles into the truck.
or: It was the tricycles that Sean loaded into the truck.
or: It was into the truck that Sean loaded the tricycles.
or: It was the truck that Sean loaded the tricycles into.
b. Fred thinks Alice climbed up Mt. Everest.

It is Fred that thinks Alice climbed up Mt. Everest.
or: It is Alice that Fred thinks climbed up Mt. Everest.
or: It is up Mt. Everest that Fred thinks Alice climbed.
or: It is Mt. Everest that Fred thinks Alice climbed up.
The idea is that the It-Clefting transformation “cleaves” the sentence, by moving one of its
constituents into a high clause containing it plus BE. Intuitively it works like this:
“cleaving”
here
It was the tricycles that Sean loaded ______ into the truck
The second through fourth sentences in each group are all clearly related to the first sentence, and
can be derived from it with a transformation.
It-clefted sentences are clearly not neutral in their rhetorical force; they place strong emphasis
of some kind (often called focus) on the clefted NP or PP.
I give a version of the transformation below.
(173) It-Clefting
• Input: an S containing NP or PP.

• Construct “on top” of S a new S, with subject it, verb be, and an CP sister to V within its
VP.
• Copy onto be the same value of the feature [Tense] as the original S.
• Move the NP or PP to a position following be as daughter of the new S.
The mention of the copying of the feature [Tense] onto be (which you might think of as a form of
agreement) is for completeness;93 we will generally skip this step in the derivations to follow.
It Clefting is another instance of an unbounded dependency, and for the same reason as in
Wh- Movement requires a movement analysis.
Here is an example of how It Clefting applies. Structure added by the rules is shown in italics.
Deep structure and movement:
VP
PP
NP NP NP
|
N V Art N P Art N
| | | | | | |
Sean loaded the tricycles into the truck.
Surface structure, with trace:
VP
CP
PP VP
NP NP NP NP PP
| | |
Pro V P Art N Comp N V Art N t
| | | | | | | | | |
It was into the truck that Sean loaded the tricycles
Like Wh- Movement and Topicalization, It Clefting is an unbounded rule:
93
Detail: it seems that you actually don’t have to copy tense in every case; you can also use a kind of default
[Tense:present] on be, as in sentences like It is Alice that left.
It was [ the tricycles ]NP that Tom thinks Sue knows that Bill loaded ___ onto the truck.
Study Exercise #32
Provide deep structure, arrows showing movement, and surface structure for this case of
unbounded movement:
It was the king that we told the knights that they must fight for.
Deep structure, with arrow showing movement:
Surface structure, with inserted it as subject and be as verb; moved NP is replaced by trace:
10. Island constraints
Consider the following deep structure:
You have seen Alice and who?
This is clearly a possible deep structure, as it can be an echo question (p. 190) if nothing
applies to it. However, if we make it part of a CP, in anticipation of making it into a Wh- question
(see section 4.2 of this chapter for why), and then apply Wh- movement to this deep structure, the
result is unexpectedly ungrammatical:
Deep structure, Subject/Aux Inversion, and Wh- Movement:
CP
Comp S
|
NP Aux VP
| |
Pro have V NP
| | |
you seen NP Conj NP
| | |
N and Pro
| |
Alice who
Surface structure:
CP
Comp S
|
NP Aux NP VP
| | |
Pro have Pro V NP
| | |
*who you seen NP Conj NP
| | |
N and t
|
Alice
Note that we really are dealing with ungrammaticality rather than nonsense; the question is
perfectly reasonable and could mean roughly Who did you see Alice with?
The ungrammaticality of this sentence is a serious overgeneration problem: we have applied

the rules of the grammar in a perfectly legitimate way, but have derived a bad result. Here are
some further data of the same sort:
a. Conjoined NP
You have seen who and Alice (okay as echo question)
*Who have you seen t and Alice?
b. Conjoined NP
Bill will take pictures of Fred and Alice (not the same deep structure, but close
enough)
*Who will Bill take pictures of t and Alice?

*Who will Bill take pictures of Fred and t?
c. Conjoined PP
Jay jumped onto the trampoline and into the pool.
*What did Jay jump onto the trampoline and into t?
*What did Jay jump onto t and into the pool?
d. Conjoined VP
Phil loves coffee and abhors tea.
*What beverage does Phil love coffee and abhor t?
*What beverage does Phil love t and abhor tea?
e. Conjoined S
Phil might thinks that Sue loves coffee and Alice abhors tea.
*What beverage might Phil think that Sues loves coffee and Alice abhors t?
*What beverage might Phil think that Sues loves t and Alice abhors?
The generalization here is that Wh- Movement produces an ungrammatical result if it tries to
move a wh- phrase outside a structure in which two constituents are joined by a conjunction.
Structures of this sort are called coordinate structures. In the four groups of sentences above, the
structures are as follows:
(174) Some coordinate structures

a,b: NP c: PP d: VP
| | |
NP Conj NP PP Conj PP VP Conj VP
e: S
|
S Conj S
A general notation for coordinate structures is as follows:94
X
|
X Conj X
The next step is to fix the grammar so that it will no longer generate sentences in which
extraction has taken place from a coordinate structure. The most obvious move would be to add a
94
We could generalize this to cover the multiple conjuncts generated with Kleene star (p. 155; as in
NP and NP and NP...), but won’t take the time.
complication to the Wh- movement rule that would simply block the rule from doing this.
However, we will see later on that all the other long-distance transformations are blocked in the
same way. If we added exactly the same complication to all the other rules, we would be missing a
generalization.
11. Constraints in grammar
The more general solution would be to add to grammar a constraint on possible derivations.
A constraint could be thought of as a “filter” on the operation of the grammar: if the derivation of a
sentence violates the constraint, then the constraint marks the sentence as ungrammatical, and it is
eliminated from the set of sentences that the grammar generates.95
The general organization of such a grammar can be imagined as follows:
95
If you study more linguistics you will likely find a major role for constraints in the theories taught to you.
Indeed, some theoretical approaches eliminate rules entirely: in such theories a well-formed linguistic structure is
simply one that obeys all the constraints.
(175) An architecture for grammatical theory with deep and surface structure plus constraints
S (initial symbol fed to the grammar)
(bare tree)
Lexical insertion lexicon
Deep structure
Constraints (acting as filter)
Transformations
Surface structure
(and on to other components: inflectional morphology, phonology)
[phonetic form]
This conception includes three of (what I take to be) the four basic formal mechanisms of
linguistic theory: (a) generative rules (here, phrase structure rules); (b) transformations
(converting one structure to another); (c) filters (throwing out the result of a derivation).96
For the data under discussion, the constraint we need is the following:
96
The fourth rule type is interpretation, which we will cover when we get to semantics.
(176) Coordinate Structure Constraint
Mark as ungrammatical any sentence in which a constituent has been extracted from inside a
coordinate structure.
* X * X
| |
X Conj X X Conj X
The triangle notation seen here means, “any constituent of the typeX”.
The Coordinate Structure Constraint is called an island constraint. A coordinate structure

acts as an “island,” in that it is inaccessible to the efforts of transformations to extract things from
it. The Coordinate Structure constraint will correctly rule out the ungrammatical sentences given
earlier.
Here is one way to demonstrate how a constraint works: you draw the deep structure of a
sentence, outline the constituent that moves, outline the island that contains it, and draw an arrow
showing that the movement does indeed move a constituent outside of the island. (One also adds
an asterisk, to indicate that this movement results in ungrammaticality.)
Deep structure (empty Comp is the landing site for Wh- Movement)
Arrow shows application of Subject/Aux Inversion.
CP
VP
NP
Comp PP
NP
NP NP NP
| | |
N Aux V N P N Conj Pro
| | | | | | | |
Bill will take pictures of Fred and who
Output of Subject/Aux Inversion:
CP
VP
NP
Comp PP
NP
NP NP NP
| | |
Aux N V N P N Conj Pro
| | | | | | | |
will Bill take pictures of Fred and who
2nd transformation (attempted): Wh-Movement:
CP
VP
NP Island: coordinate
structure
Comp PP
NP
NP NP NP
| | |
Aux N V N P N Conj Pro
| | | | | | | |
will Bill take pictures of Fred and who
*
Result: *Who will Bill take pictures of Fred and?
Study Exercise #33
Explain why the sentence
*What city have the Romans destroyed and attacked Athens?
is ungrammatical. Illustrate with a derivation.

What city have the Romans destroyed and attacked Athens?
This sentence is a violation of the Coordinate Structure Constraint. The coordinate structure
consists of the two conjoined VP’s destroyed what city and attacked Athens. The wh- phrase what
city is extracted out from inside the coordinate structure, resulting in an ungrammatical sentence.
Deep structure, showing Subject/Aux Inversion:
CP
Comp S
|
NP Aux VP
| |
Art N have VP Conj VP
| | |
the Romans V NP and V NP
| | |
destroyed Art N attacked N
| | |
what city Athens
Output of Subject/Aux Inversion:
CP
Comp S
|
Aux NP VP
| |
have Art N VP Conj VP
| | |
| | |
| | |
what city Athens
Wh- Movement, moving outside coordinate structure:
CP
Comp S
|
Aux NP VP
| |
have Art N VP Conj VP
| | |
| | |
| | |
what city Athens
*
Illegal surface structure:
CP
Comp S
|
NP Aux NP VP
| |
Art N have Art N VP Conj VP
| | | | |
what city the Romans V NP and V NP
| | | |
destroyed t attacked N
|
Athens
—————————————————————————————————————————————————————
Study Exercise #34
Explain why the sentence ‘What city have the Romans attacked and destroyed?’ is
grammatical. Illustrate with a derivation. Hint: take a look at the phrase structure rules (146) on p.
177.
What city have the Romans attacked and destroyed?
Here there is a coordinate structure, but the wh- phrase is not inside it. The coordinate
structure is the two verbs attacked and destroyed. Since the wh- phrase what city is not extracted
from inside the coordinate structure island, the sentence does not violate the Coordinate Structure
Constraint and thus is grammatical. The following deep structure + movement arrows shows that
the “extractee” is not inside the island:
CP
Comp S
|
NP Aux VP
|
Art N have V NP
| | |
the Romans V Conj V Art N
| | | | |
attacked and destroyed what city
This makes a useful point: it is not the presence of islands as such that creates
ungrammatically under extraction; it is specially the extraction of material from the island.
—————————————————————————————————————
Study Exercise #35
Explain why the sentence
Which city and which province will the Romans destroy?’
is grammatical. Illustrate with a derivation.

Which city and which province will the Romans destroy?
In this case, the wh- phrase (namely which city and which province) is again not inside the
coordinate structure; rather, it is the coordinate structure. Thus applying Wh- movement does not
extract a wh- phrase from inside a coordinate structure, and the Coordinate Structure Constraint is
not violated. Here is the deep structure with movement shown.
CP
Comp S
NP Aux VP
|
Art N will V NP
| | | |
the Romans destroy NP Conj NP
|
Art N and Art N
| | | |
which city which province
———————————————————————————————————————
12. The universality of the Coordinate Structure Constraint
One of the goals sought by linguistics in writing formalized grammars is to locate universals
of language. A linguistic universal is a property shared by all human languages. The explanation
of linguistic universals is one of the key tasks of linguistic theory.
Linguistic universals are proposed and tested against data from the languages of the world;
there are thought to be about 8000 of them.97 No universal has been checked against all 8000,
however, at least some proposed universals look fairly promising.
Some universals that have been proposed are fairly superficial, for example:
• All languages have nouns and verbs.

• All languages have wh- questions.
• All languages have consonants and vowels.
• All languages use the vowel [a]98 or something phonetically close to it.
97
The number is declining steadily. Probably the best list of languages is the Ethnologue, at
http://www.ethnologue.com/.
98
IPA [a] is more or less the [a] vowel of Spanish, or in some dialects of English the vowel of hot.
Others universals are more subtle, and emerge only when we have submitted a large number of
languages to formal analysis—that is, have constructed partial grammars for them.
As you might expect, it is common for linguists to propose universals, then be forced to
abandon or modify their proposal in the face of falsifying evidence. This is only natural, and
indeed one might argue that part of the job of the linguist is to be a bit “out on a limb”, creating
hypotheses about language that are interesting enough to be worth checking.
The Coordinate Structure Constraint was first noticed and proposed as a universal by the
linguist John R. Ross, who pioneered the study of syntactic islands in the 1960’s. The phenomenon
of islands attracted a great deal of attention and has been extensively studied and analyzed since
then. Today, there seems to be a consensus, based on study of a fair number of languages, is that
the Coordinate Structure Constraint is universal. (The doubtful cases are instances in which we’re
not sure that the structure in question is really a coordinate structure.) To be more precise: in all
languages that can be tested (because they have wh- movement; in situ languages can’t be tested),
extraction from coordinate structures is impossible. Here are some sample data from other
languages:
German Simple sentence: Du hast Fritz gesehen

you have Fritz seen
‘You have seen Fritz.’
Wh- question (with gap): Wen hast du ___ gesehen?

who have you (gap) seen
Extraction from *Wen hast du Fritz und ___ gesehen?

coordinate structure: who have you Fritz and (gap) seen
*‘who have you seen Fritz and?’
CP
Comp S
|
NP Aux VP
| |
Pro hast NP V
| have | |
du NP Conj NP gesehen
you | | | seen
N und Pro
| and |
Fritz wen
Fritz who
French: *Quelle journal as- tu lu ___ et ce livre?

what newspaper have you read (gap) and this book
*Quelle livre as- tu lu ce journal et ___?
what book have you read this newspaper and (gap)
Russian: *Kovo Ivan videl Petra ili ___?

who Ivan saw Peter or (gap)
*‘Who did Ivan see Peter or?’
*Kovo Ivan videl ___ ili Petra?

who Ivan saw (gap) or Peter
Arabic (examples involve Topicalization)
*Zayd-an ʔintaqada xalīl-un ____ wa-ʕamr-an

Zaid criticized Khalil-nom and-Amr-acc.
*‘Zaid, Khalil criticized and Amr’
* ʕamr-an ʔintaqada xalīl-un ____ Zayd-an wa

Amr-acc criticized Khalil-nom Zayd-acc. and
*‘Amr, Khalil criticized Zayd and’
Choctaw (Oklahoma)
*Katah-oosh John-at taloowa-tok anoti ____ hilhah-tok?
who-focus John-nom. sing-past and dance-past
*‘Who did John sing and dance’ (= ‘Who was the person such that that person
sang and John danced?’)
American Sign Language (example involves Topicalization)
t
*FLOWER 2GIVE1 MONEY, jGIVE1 ____
Flowers, he-gave-me money but she-gave-me
*‘Flowers, he gave me money but she gave me ___.’99
Formal universals like the Coordinate Structure Constraint have inspired a fair amount of
theorizing about language and language learning, which we’ll take on in the next chapter.
12.1 Looking ahead
The combination of long-distance transformations and islands leads to a fair amount of
99
The underline represents a non-manual element produced in synchrony with the sign FLOWERS. Subscripts
indicate, roughly, starting and ending points of verbal signs that serve the function of pronouns.
analytical work: for each transformation, you want to show that it respects the island, and for
each island (we’ll cover more) you want to show that the transformations all respect it.
13. Two more islands
To complete our general account of long-distance transformations and island constraints, here
are two more islands (there are quite a few more, varying from language to language, but we will
stick with just three total). The general point that emerges is that all the long-distance
transformations obey all of the island constraints (since there will be three of each, we will need to
check a total of nine cases).
13.1 The Wh- Island Constraint
Recall embedded wh- questions, like I know what Bill saw. We already have the means to
derive this (see Chapter 6, section 4.2) and the example is reviewed below.
Deep structure and Wh- Movement:
VP
CP
VP
NP NP NP
| | |
Pro V Comp N V Pro
| | | | |
I know Bill saw what
Surface structure:
VP
CP
VP
NP NP
| |
Pro V Comp N V NP
| | | | | |
I know NP Bill saw t
|
Pro
|
what
But now consider the following scenario: what if, at the level of deep structure, there were
two wh- phrases in the same clause? This is not so absurd, since we actually have sentences like
the following:
I know [ who saw what ]CP
Here, the wh- word what remains in situ, as the object of say. We won’t be able to cover here just
what circumstances permit a wh- phrase to remain in situ in English, but for now this sentence
suffices to show that it is possible to have a clause with two wh- phrases.
Now, consider this scenario: we take the above sentence as a deep structure, move what into
the “lower” Comp, and who into the “higher” Comp, as follows:
Deep structure and first application of Wh- Movement:
CP
VP
CP
Comp S
VP
NP NP NP
| | |
Pro Aux V Comp N V Pro
| | | | | |
You would know who saw what
Result of first application of Wh- Movement, with arrow showing Subject-Aux inversion:
CP
VP
CP
Comp S
VP
NP NP
| |
Pro Aux V Comp N V NP
| | | | | | |
You would know NP who saw t
|
Pro
|
what
Result of Subject-Aux Inversion, showing second application of Wh- Movement:
CP
VP
CP
Comp S
VP
NP NP
| |
Aux Pro V Comp N V NP
| | | | | | |
would you know NP who saw t
|
Pro
|
what
Surface structure:
CP
VP
CP
Comp Aux S
| |
NP would VP
|
Pro NP
| |
who Pro V Comp NP V NP
| | | | |
you know NP t saw t
|
Pro
|
what
The result is *Who would you know what saw?, which most speakers find crashingly bad. It is
worth emphasizing that this is not due to its lacking a meaning; it’s clear that it should mean the
following:
“What is the person such that you know what that person saw?”
The meaning is hard to access, given the extreme ungrammaticality of the sentence.
Linguists have proposed to explain the ill-formedness of sentences like Who do you know
what saw? by positing yet another island, along the following lines:
(177) Wh- Island Constraint
Mark as ungrammatical any sentence in which a constituent has been extracted from inside an
CP whose Comp contains a wh- phrase.
CP
*
Comp S
|
wh-
This island constraint is slightly different from the Coordinate Structure Constraint, because
the island is actually created by a transformation. The “lower down” Wh- Movement forms an
island that blocks any further Wh- movements higher up in the tree.
To illustrate: returning to the derivation given above, but this time drawing in the island, we
can see that it is correctly excluded by the Wh- Island Constraint. The sequence what who saw is
covered by the description of the island, and thus the sentence is ruled out.
CP
VP
CP
Comp S
VP
NP NP
| |
Aux Pro V Comp N V NP
| | | | | | |
would you know NP who saw t
|
Pro
|
what
*
The Wh- Island constraint covers a fair amount of data; here are some other sentences that it
excluded. I’ve put brackets in to illustrate the CP that begins with a wh- phrase and thus forms a
Wh- Island.
Bill admitted who committed this crime.

*[ What crime ] did Bill admit [ who committed ___ ]CP ?
Fred doesn’t care how long you take on this exam.

*[ Which exam ] doesn’t Fred care [ how long you take on ___ ]CP?
Alice doesn’t care which exam you take a long time on.
*[ How long ] doesn’t Alice care [ which exam you take ___ on ]CP?
Observe further that there is nothing wrong with having two wh- phrases in the same sentence.
It’s only when one wh-phrase is moved out of the CP that the other one begins that you get a bad
result. Here is an example. In the sentence
Which student would you tell who you saw?
the two instances of wh- movement are non-overlapping. The movement that goes to the higher
Comp is not out of the island, so the sentence comes out fine. Here is the full derivation.
Deep structure with lower wh- movement:
CP
VP
CP
VP
NP NP NP NP
| | |
Comp Pro Aux V Art N Comp Pro V Pro
| | | | | | | |
you would tell which student you saw who
Output of lower wh- movement, showing Subject-Aux Inversion:
CP
VP
CP
Comp VP
|
NP NP NP NP NP
| | | |
Comp Pro Aux V Art N Pro Pro V t
| | | | | | | |
you would tell which student who you saw
Output of upper wh- movement.
Note that this is not movement outside of the wh- island, shown in blue.
CP
VP
CP
Comp VP
|
NP NP NP NP NP
| | | |
Comp Aux Pro V Art N Pro Pro V t
| | | | | | | |
would you tell which student who you saw
Surface structure:
CP
VP
CP
Comp Comp VP
| |
NP NP NP NP NP
| | | |
Art N Aux Pro V NP Pro Pro V t
| | | | | | | | |
which student would you tell t who you saw
You can see this all at once if we put the material on just one line, showing only the two
instances of Wh- Movement and the island:
[ Who ] would you tell ___ [ [ who ] you saw ___ ]CP?
Note finally that Topicalization and It-Clefting obey the Wh- Island Constraint:
Topicalization:
*[ Kate ], I realized [ to whom they would send ___ ]CP.
It-Clefting:
*It was [ Oliver ] that I wondered [ which book ___ would read ]CP.
13.2 The Complex NP Constraint
Another kind of island is the so-called “complex noun phrase”. Recall (p. 175) the main
phrase structure rule in English for NP, the one to which we added a possible CP daughter:
NP 
NP → (Art }) (AP)* N (PP)* (CP)
 
E
A complex NP is an NP having CP as a daughter (there may also be other modifiers). You get
a complex NP if you include the boldface items below in applying the rule.
NP 
NP → (Art }) (AP)* N (PP)* (CP)
 
E
Some examples of complex NPs include
Sue’s belief that Sam is leaving Omaha

Alice’s hunch that the burglar used this window
Bill’s inane hypothesis that Frieda saw Jack100
The island constraint for complex NPs, called the Complex NP Constraint, is stated as
follows:
(178) Complex NP Constraint
complex NP.
* NP
... N ... CP
... X ...
To demonstrate that complex NPs are islands, one does the following. (a) Set up a deep
structure that contains a complex NP; (b) make sure that in this deep structure, there is a wh-phrase
contained within the complex NP; (c) apply Subject Aux Inversion and Wh- Movement to the deep
structure and see if the result is grammatical. I have done this in the following example. The
Complex NP is circled, and the arrows show what moves where.
100
The relative clauses mentioned above (p. 178) are also islands; for instance: *What apples will you
see the man who picked ___?
Deep structure, Subject-Aux Inversion, and Wh- Movement:

CP
Comp S
|
NP Aux VP
| |
Pro have V NP
| | |
you discounted Art N CP
| |
many rumors Comp S
| |
that NP Aux VP
| |
N is V NP
| |
Sam leaving Art N
| |
* what city
Surface structure:
CP
Comp S
NP Aux NP VP
| |
Art N have Pro V NP
| | | | |
*what city you discounted Art N CP
| |
many rumors Comp S
| |
that NP Aux VP
| | |
N is V NP
| | |
Sam leaving t?
The fact that the surface structure is ungrammatical supports the existence of the Complex NP
Constraint. Similar ungrammatical sentences would be
*Which window would you disagree with Alice’s hunch that the burglar used?
*Who might you hear Bill’s inane hypothesis that Frieda saw?
———————————————————————————————————————
Study Exercise #36
Give a derivation, with boxes, arrows for movement, and a circled island, for the two sentences
just given.
*Which window would you disagree with Alice’s hunch that the burglar used?
Deep structure, with island shown with circle and arrows for Subject-Aux Inversion and Wh-
Movement:
CP
VP
PP
NP
CP
VP
NP NP NP NP
| |
Comp Pro Aux V P N N Comp Art N V Art N
| | | | | | | | | | | |
You would disagree with Alice’s hunch that the burglar used which window
Ungrammatical surface structure:
CP
VP
PP
NP
CP
Comp S
|
NP NP NP NP VP
| |
Art N Aux Pro V P N N CompArt N V NP
| | | | | | | | | | | | |
*Which window would you disagree with Alice’s hunch that the burglar used t
Who might you hear Bill’s inane hypothesis that Frieda saw?
Deep structure, with Complex NP island, arrows for Subject-Aux Inversion
CP
VP
NP
CP
VP
NP NP AP NP NP
| | | | |
Comp Pro Aux V N A N Comp N V Pro
| | | | | | | | | |
You might hear Bill’s inane hypothesis that Frieda saw who
Ungrammatical surface structure:
CP
VP
NP
CP
Comp S
|
NP NP NP AP NP VP
| | | | |
Pro Aux Pro V N A N Comp N V NP
| | | | | | | | | | |
Who might you hear Bill’s inane hypothesis that Frieda saw t
____________________________________________________________________________
Study Exercise #37
Why is the sentence
Whose theory that Sam is crazy could you believe?
grammatical? (You have to imagine a scenario in which all sorts of people are presenting theories
that Sam is crazy.)
Deep structure, with Subject-Aux Inversion and Wh-Movement shown with arrows:
CP
Comp S
VP
NP
CP
VP
NP NP NP AP
| | | |
Pro Aux V Pro N Comp N V A
| | | | | | | | |
you could believe whose theory that Sam is crazy
The point is that the whole island is extracted. Island constraints are violated when you extract
from within an island.
———————————————————————————————————————
13.3 The Complex NP Constraint and other syntactic transformations
As the following labeled sentences show, the Complex NP Constraint is obeyed by

Topicalization and by It-Clefting:
Topicalization
*[ Kate ], I discounted [ many rumors that they would elect ___ ]NP.
It-Clefting
*It was [ Kate ] that I discounted [ many rumors that they would elect ___ ]NP.
13.4 The complex NP constraint is not universal
When it was noticed and first formalized by Ross in the 1960’s, it was thought that the
Complex NP Constraint is a linguistic universal, just like the Coordinate Structure Constraint is.
Shortly thereafter, however, Scandinavian linguists began studying the island constraints of their
native languages, and noticed that neither Norwegian nor certain dialects of Swedish and Danish
respect the constraint. The linguist Jens Allwood offers the following data from Swedish,101 which
he checked with a number of speakers; the complex NP is shown in brackets.
Simple sentence:
Herodes levde i [ hopp-et om att Salome skulle förföra den mannen. ]NP
Herod lived in hope-the of that Salome should seduce that man
‘Herod lived in the hope that Salome should seduce that man.’
Wh- Movement out of Complex NP:

[ Vem ] levde Herodes i [ hopp-et om att ___ skulle förföra den mannen ]NP?
Who lived Herod in hope-the of that should seduce that man
[ Vem ] levde Herodes i [ hopp-et om att Salome skulle förföra ___ ]NP?
Who lived Herod in hope-the of that Salome should seduce
Topicalization out of Complex NP:

[ Salome ] levde Herodes i [ hopp-et om att ___ skulle förföra den mannen. ]NP
Salome lived Herod in hope-the of that should seduce that man
[ Den mannen ] levde Herodes i [ hopp-et om att Salome skulle förföra ___ ]NP.
That man lived Herod in hope-the of that Salome should seduce
This is unusual; most languages that have these rules do respect complex NPs. Thus, here are
some French data, very much like English:
*[ Qui ] as- tu proposé [ l’idée que Marie aime ___ ]NP?
‘Who have you suggested the-idea that Marie loves (gap)?
14. Why these islands?
To review, we have discussed three separate island constraints:
101
The reason for the Biblical subject matter is not clear to me. You can make up example sentences
about whatever you like, of course.
Coordinate Structure Constraint
coordinate structure.
* X * X
| |
X Conj X X Conj X
Wh- Island Constraint
Mark as ungrammatical any sentence in which a constituent has been extracted from inside an
S̄ whose Comp contains a wh- phrase.
E
CP
*
Comp S
|
wh-
Complex NP Constraint
complex NP.
* NP
... N ... CP
... X ...
Of these, the Coordinate Structure Constraint seems to be a good candidate for being a
linguistic universal; the other two are probably not universal but seem to be found in many
languages.
The question that arises when one lines up the islands in a row like this is: “Why these
islands?” That is, why should island-hood be found for just this particular configuration of
syntactic structures? The three islands seem to have little in common with each other.
The view of most linguists who consider this question is that the islands as formulated above
are a first-pass approximation. That is, it’s a good idea to formulate the islands in this way, for the
sake of explicitness of analysis, but in the long term it seem desirable to seek more abstract
principles to explain the data.
One approach that seems fruitful is to invert the problem: one specifies what places it is legal
to extract from rather than what places it is illegal. You may encounter approaches of this type if
you study syntax in future course work.
A final point is that the islands may be in some sense useful to the speakers. Psycholinguistic
experimentation (including with brain-scanning devices) suggests there is a cognitive burden for
the listener whenever the sentence heard involves a filler-gap constructions such as those created
in the transformations described here. When a language has island constraints, they in effect tell
the language user, “don’t bother to look for gaps here” — perhaps this reduces the burden on
speech perception, and thus reflects a principle of good “language design”.
For further reading

Islands were discovered, and launched as a major syntactic research topic, in a doctoral
dissertation written at MIT in 1967 by John Robert Ross, entitled Constraints on variables in
syntax. MIT makes this massive work available on their university website:
http://hdl.handle.net/1721.1/15166, and it is quite readable.
The research literature that has built up around islands in the intervening years is enormous
and would probably best be undertaken by the student after more coursework in syntax. The
linguist Dennis Ott has offered a sobering assessment of just how hard the island problem has
proven to be in a brief book review: https://muse.jhu.edu/article/539657.
Chapter 7: Language acquisition
1. The relationship of linguistic theory to language acquisition
Linguistics exists in a kind of dual mode: at the level of language data, linguists are endlessly
engaged in analysis, trying to develop better grammars as well as better general theories in which
such grammars can be laid out. But behind all this activity are ponderings at a different level,
concerning how the strikingly elaborate grammars of human languages arise.
Let us assume (for purposes of argument) that the grammar we’ve been developing does in
some way characterize the native speaker’s knowledge. Our starting point here is that the native
speaker must learn the grammar, too. Children do this in infancy and childhood, over the course of
just a few years, usually without overt instruction, but instead simply by inhabiting a community
where the language is spoken, listening intently, and trying to speak. Moreover, what the child
learns is not the toy grammar we have been working with, but something much larger.
Language acquisition is an important area of linguistics and linguistic theory. There are three
research methodologies. First, observational study is the longest-standing method: one arranges
to be in a situation where one can hear little kids talking and records what they say, ideally with
audio as well as transcription.102 Much of the data from such study has been gathered into a large
corpus, known as CHILDES, from which investigators can gather new and important
generalizations.103 Second, infants and children are the subjects in experiments, which gather their
reactions to carefully-planned language material presented to them. In my own department at
UCLA, an active infant and child laboratory carries out experiments with children brought into the
lab by their parents. Third, research proceeds by simulation: linguists attempt to develop computer
software that can learn the grammatical and phonological patterns of language on exposure to
language data representative of what children hear; the grammars learned by the simulator can then
be compared with evidence about what real people know about their language.
What results are being obtained by this research program? I think the most important are two.
First, production lags perception: children, and even infants, have considerable linguistic
knowledge that can be detected only in their reactions to experimental stimuli, not in their own
productions. The extreme case of this is phonological knowledge in infants, who evidently know
the speech sounds and the principles of legal sound sequencing from the age of about six months.
Going even further, even newborns can in some cases identify their mother’s language from its
characteristic patterns of syllable timing and pitch; presumably they can do this because such
auditory properties are available to them in utero.
The other well-established result of the study of language acquisition is that children are
virtuosic: they are prodigiously capable acquirers of language, and most noticeably, they
102
A classical procedure, decades old, is for a linguist to keep a detailed diary on the linguistic productions of
his or her own child. Alternatively, repeated visits are made to the same children in their home or daycare center.
103
http://childes.psy.cmu.edu/
outperform the efforts of linguists. Kids can exposed to a language for a few years become fluent
native speakers, with extensive production abilities and nuanced, subtle judgments of well-
formedness. Linguists, toiling away at analysis for many years, still struggle to obtain grammars
that properly match what the native speaker knows. Moreover, to the extent that the linguists’
theories are incorporated into machine-implemented systems that actually learn language, these
systems cannot learn with anything like the speed or accuracy that children do. All of this causes
linguists to believe that the ability to acquire language is an extraordinary aspect of human beings,
well worth study.
The broad scientific debates surrounding grammar and language learning can be outlined as
follows:
Outline:
• the quality of the evidence available to the child for learning

➢ the “no negative evidence” issue
• the possibility of innate knowledge serving as an aid to learning
➢ the alternative of powerful inductive mechanisms
• innate knowledge/abilities as a biological phenomenon
➢ critical periods
➢ genetic defects specific to language
• innate knowledge and linguistic universals
• linguistic abilities and human evolution
2. The quality of the evidence available to the child
To begin, it seems clear that however children learn language, conscious instruction (say, by
parents) must play very little role. Not all parents instruct their children in language, and the
parents who do are likely focusing on bits of normative grammar (Chap. 3), peripheral to the
language as a whole.
In fact, there’s even less reason to consider overt instruction as a factor, because it appears that
small children don’t even pay much attention to it. Textbooks on language acquisition often
include entertaining little dialogs between parents and toddlers showing the futility of overt
instruction, of which the following is brief sampling:
C: Nobody don’t like me C: Want other one spoon, daddy.

A: No, say ‘nobody likes me’ D: You mean, you want the other spoon
C: Nobody don’t like me C: Yes, I want the other one spoon, please, daddy.
A: No, say ‘nobody likes me’ D: Can you say ‘the other spoon’?
C: Nobody don’t like me C: Other...one...spoon
A: No, say ‘nobody likes me’ D: Say ‘other’
C: Nobody don’t like me C: Other.
A: No, say ‘nobody likes me’ D: Spoon.
C: Nobody don’t like me C: Spoon.
A: No, say ‘nobody likes me’ D: Other spoon.

C: Nobody don’t like me C: Other...spoon. Now give me other one spoon?
A: No, say ‘nobody likes me’
C: Nobody don’t like me
A: Now listen carefully: ‘nobody likes me’
C: Oh, nobody doesn’t likes me.104
These examples make a further point: at any given point in the child’s acquisition period, she
has a relatively stable, internalized, wrong grammar, which she tends to stick to until it evolves in
the natural way to the next, more accurate stage.
Leaving aside the case of overt instruction, we might also ask if children are somehow given a
special linguistic diet by their parent, which makes acquisition possible. Such a diet might perhaps
consist of a simplified version of the language, which some linguists (more or less jokingly) have
referred to as called Motherese. Reasons to be skeptical of the effectiveness of Motherese are the
apparent existence of children who learn their native language without it; and the fact that
Motherese is often ungrammatical, a pattern that could hardly help acquisition in the long run.105
Scholars also differ on whether the input to the child is in general grammatical: Noam
Chomsky106 has repeatedly insisted that it is not, as in quotes like the following:
“Thus, it is clear that the language each person acquires is a rich and complex construction
hopelessly underdetermined by the fragmentary evidence available.” (Reflections on
Language, 1975, p. 10)
“Knowledge arises on the basis of very scattered and inadequate data and ... there are
uniformities in what is learned that are in no way uniquely determined by the data itself..”
(Cartesian Linguistics, 1966, p. 65)
Various experts in child language development have disagreed with Chomsky’s claim. It
seems worth remembering that any one error in the learning environment (for example, if a
speaker gets tangled up and inadvertently produces an island violation) could be very dangerous to
the task of getting the grammar right, so even a modestly error-ful ambient environment might still
suffice to make Chomsky’s point.
104
From: McNeill, D. (1966). Developmental psycholinguistics. In Smith, F., and Miller, G. A.
(eds.),The Genesis of Language, MIT Press, Cambridge, Mass.
105
I make no claims here on whether or not it is desirable to speak Motherese to one’s children.
106
Chomsky invented most of the content of the syntax unit of this text in some extremely influential work
from the 1950s to the 1970s. Devices attributable to him include phrase structure rules, subcategorization,
transformations, morphosyntactic representations, and an early version of island constraints.
3. Cases where “negative evidence” would be required for learning; theoretical approaches
One particularly intriguing aspect of language learning is this: how do we learn that sentences
are ungrammatical? As noted above, actual correction of error is rare and ineffective, and for the
more interesting cases like learning not to violate islands, it seems extremely unlikely that there
would be sufficient overt correction for a child to learn the pattern.
3.1 Innate knowledge
For some linguists, the no-negative evidence problem provides indirect support for the
hypothesis of innate knowledge. If some grammatical knowledge is simply not accessible to
direct learning from the data, the only reasonable explanation for how we come by this knowledge
as children is that we bring the knowledge to the task with us. In other words, our genome,
physically embodied in our DNA, contains grammatical information, information crucial to
acquisition.
It may seem counterintuitive to suppose that knowledge could be innate; some people get used
to thinking that the genes control only the form of the body, and not of the mind. But examples of
innate knowledge are easy to find in the animal kingdom. For example, some species of birds have
a song that does not vary at all across individuals, and which even birds raised apart from their
species will sing. The ability to sing these songs surely would count as innate knowledge. In many
species of mammals, newborns, shortly after birth, get up and walk, an act that requires a huge
amount of computation (we know this because getting a robot to walk is one of the biggest
challenges facing human engineers today).
Chomsky is well known for his strong views on the innateness question for language. Here is
a sampling, from his Language and Mind (1968):
To repeat: Suppose that we assign to the mind, as an innate property, the general
theory of language that we have called “universal grammar.” This theory encompasses
the principles that I discussed in the preceding lecture and many others of the same sort,
and it specifies a certain subsystem of rules that provides a skeletal structure for any
language and a variety of conditions, formal and substantive, that any further
elaboration of the grammar must meet. The theory of universal grammar, then, provides
a schema to which any particular grammar must conform. Suppose, furthermore, that
we can make this schema sufficiently restrictive so that very few possible grammars
conforming to the schema will be consistent with the meager and degenerate data
actually available to the language learner. His task, then, is to search among the
possible grammars and select one that is not definitely rejected by the data available to
him. What faces the language learner, under these assumptions, is not the impossible
task of inventing a highly abstract and intricately structured theory on the basis of
degenerate data, but rather the much more manageable task of determining whether
these data belong to one or another of a fairly restricted set of potential languages.
3.2 Inductivism
Chomsky’s view is near one pole of an intellectual continuum at whose other extreme are
scholars with a strongly inductivist point of view. In this alternative, what makes the child capable
of the feat of language acquisition is her possession of formidable techniques of inductive learning
— that is, grasping the pattern through intensive processing of the learning data available.
Bruce Derwing of the University of Alberta wrote:107
An alternative model of language acquisition ought to be developed on the assumption that

the human organism is initially completely un-informed both as to the essential nature of
language and as to the best way to learn a language. My initial assumption is that no other
‘special mechanisms’ or ‘secret abilities’ are required for learning language than for
learning anything else. In short, I shall assume that language acquisition can be fully
accounted for in terms of such known (i.e. empirically established) capacities as the ability
of the human organism to discriminate among and generalize from the sense-data to which
he is exposed and most importantly (since this capacity seems to be species-specific), to
extract regularity—or induce a latent structure—from his experience.
Observe that Derwing is not opposed to innate abilities, but emphasizes that they are abilities to
learn (not pre-formed knowledge); and he wants these abilities to not be specifically linguistic.
4. The learning of ill-formedness: two viewpoints
4.1 Learning well-formedness using innate knowledge
If we imagine that some aspects of linguistic knowledge are innate, a good place to focus
research efforts might be linguistic universals. People wouldn’t need to learn (somehow) that
Coordinate Structure Constraint violations are ungrammatical if the Coordinate Structure
Constraint (or, one might hope, something from which the Coordinate Structure Constraint follows
as a consequence) is innate. This would also be the reason why the Coordinate Structure Constraint
is universal.
We might also imagine a more subtle scenario, concerning how innate knowledge could
permit English speakers to come to know that violations of the Complex NP Constraint are
ungrammatical, whereas speakers of Swedish and Norwegian know that such violations are are
acceptable in their own languages. The idea is that the Complex NP Constraint (or again,
something more abstract from which it follows) is innate, but in a form that would permit “data
override”: if you actually hear data that tell you the constraint is violable in your language, you
override your innate knowledge. This would be the case for Swedish and Norwegian children, who
presumably hear during their childhoods a number of sentences that violate this constraint.
Children in English-speaking environments never hear the data that would justify an override, so
they never push aside the innate pattern.
107
Transformational Grammar as a Theory of Language Acquisition (1973). Cambridge: Cambridge
University Press, pp. 200-201.
4.2 Learning ill-formedness inductively without negative evidence
Although the learning of ill-formedness without negative evidence has been sometimes
presented as an insuperable difficulty to inductive approaches, in fact inductivism is not entirely
helpless in such cases. Over the last few decades, learning has been approached — notably by
computer scientists — as a problem in computation, and the research findings in this area have in
some cases engaged precisely with the negative evidence problem.
To give an idea of what is involved, consider one very simple strategy that has been used:
collect a lot of data, then compute some form of this statistic: observed/expected — that is, the
number of observed instances of a structure, divided by some kind of informed estimate of how
many instances one should observe, under neutral assumptions.
For Complex NP Constraint violations, the observed/expected strategy might be something

like this (I’m quoting the hypothetical unconscious inner monologue of the language-learning
child):
• “Thus far in my experience I have heard 4,947 complex NPs, that is to say, in 10% of all
sentences.
• “I have also noticed 6,823 gaps, created by rules such as Wh- Movement and
Topicalization,” in 30% of all sentences.
• “Thus, by multiplying, I estimate that 3% of the ambient sentences should have occurred
with a gap inside a complex NP. This would be about 600 sentences.”
• “But in fact, not a single gap has yet occurred inside a complex NP.”
• “I therefore infer there is something wrong about extracting constituents from inside
complex NPs”.
This is just a simple intuitive comparison, but more rigorous implementations of the same basic
idea are available by invoking probability theory, as is now standard in statistics and computer
science.108
Is this scenario a fantasy? It has in fact been applied to simpler data, in phonology, with fairly
good results. Moreover, there is evidence that people can keep track of such statistics in syntax:
psycholinguistic studies of how people understand sentences indicate that people’s guesses about
where in the tree a new word should go are guided by the statistics of subcategorization: their first
guesses are those that match the most frequent subcategorization frame of the last syntactic head
they heard. Thus (to use an example from earlier), the guess for the structure of Fred ran up a
big... will depend on the relative frequency in real life with which run is followed by a particle
(thus ran up a big bill) vs. a prepositional phrase (thus ran up a big hill).109
108
A key idea is that of choosing among models according to how much they maximize likelihood, the
probability that any given model assigns to the observed data A standard reference that covers this, currently on line, is
Speech and Language Processing (2020) by Dan Jurafsky and James Martin; https://web.stanford.edu/~jurafsky/slp3/.
109
For a general review of this and related literature, see
http://lcnl.wisc.edu/people/marks/pubs/SeidenbergMacDonald.1999.CogSci.pdf
The apparent ability of people to count the statistics of subcategorization frames is particularly
relevant because these frames have been put forth as a negative evidence problem.110 If a child
learns a wrong subcategorization, nothing in the ambient data can directly tell her it is wrong. Yet
in fact, children rather frequently make subcategorization errors—and ultimately, of course,
recover from them. Here are a couple of examples:
“I filled the salt into the bear.”111

“She came it over there” (Meaning: “She made it come over there”, a causative)
The first sentence indicates a wrong subcategorization for fill (probably acquired by wrongly
generalizing from put and other verbs), which almost certainly was corrected prior to adulthood—
mostly likely by gradually noticing that no occurrences of fill uttered by qualified individuals used
this frame; the observed number of [ ___ NP PP ] cases (probably zero) was smaller than the
child’s expected value, and ultimately led her to abandon this frame. The same reasoning would
hold for the second example.
In general, I would judge that in recent year inductivism has made something of a comeback
in linguistics, primarily due to experimental findings suggesting that people are very good at
inductive learning.
Inductivism nevertheless faces a huge and still largely unanswered challenge. A statistic like
observed/expected requires you to have, in effect, a set of “bins” into which you sort your
linguistic experience, so as to be able to compute these values. A complex NP is a nontrivially
complicated structure to describe—might there be a large variety of equally complex structures
that also have to have their statistics monitored. Even for subcategorization, there is a danger of
irrelevant bins: one hardly wants to waste counting how many sentences with an even number of
words a verb has occurred in, and similarly for other utterly pointless contexts. Inductivism must
either rely on innate knowledge to know what bins experience is sorted into, or find some way, not
yet established, to get them “for free”.
4.3 Universals
Lastly, a purely inductivist approach to language offers no account of true linguistic

universals, if such exist—in inductivism, a language violating the Coordinate Structure Constraint
would be learnable.112 Innatist approaches offer a quick and cheap explanation of universals;
they’re true because languages are spoken by people, who being biologically arranged the way
they are, have no choice but to obey them.
For instance, by C. L. Baker (1979) in “Syntactic Theory and the Projection Problem,” Linguistic
110
Inquiry 10.4.
111
Bowerman, M. (1982). Evaluating competing linguistic models with language acquisition data:
Implications of developmental errors with causative verbs. Quaderni di Semantica, 3, 5-66.
112
There is of course a temptation to do an experiment: have a team of skilled and charismatic
research assistants spend a great deal of time in a day care center uttering Coordinate Structure Constraint
violations, and see if the children who attend it develop a “universal-violating” grammar. It’s not clear what
the Human Subjects Protection Committee would think of this one ...
5. Innateness and biology
Turning, then, to the issue of the biology of human language, it’s important to note that the
view that people are biologically equipped for language in a special way unique to our species is
not entirely tied to the idea of innate linguistic knowledge. In particular, we could be highly
adapted to learning and use of language, but not possess any innate knowledge of the content of
language per se. (This seems to be view taken, for example, in the quote from Bruce Derwing
given above). Under this assumption, the view that humans are specialized for language is rather
less controversial, though again it is hardly agreed upon universally.
A common way to make a case in this area is to compare language with other abilities (of
various species) that plainly are part of their biological endowment.
5.1 Irrelevance of deliberate training
Achievement of fluency in language does not seem to depend on training (compare, for
instance, playing the piano, or studying math). Most children become fluent speakers on their own,
on schedule, by their own more or less automatic efforts. This is similar to the process of learning
to walk, likewise documented to occur spontaneously, follows a standard time course, and (by
experiment) has been shown not to be particularly aided by instruction.
5.2 Critical period
Language appears to involve a critical period, that is, a span of time after which complete
acquisition of the skill becomes difficult or impossible. The critical period is widely documented
for language; we see it (anecdotally) in families of immigrants, where the youngest members of a
family often become the best speakers of the new language, despite equal exposure for all.
Vision in cats is apparently a similar ability: kittens who have one eye temporarily closed
when young fail to “wire up” their neural circuitry for that eye, and do not make up the deficit
later. Experimentation (cortical probes) indicates that the circuitry does not grow in. “Accidental
experiments” on humans (surgery on congenital cataracts, misguidedly delayed to lessen risk)
show that same is probably true for us.113
Some forms of birdsong reflect a critical period; young birds reared away from their species
fail to acquire the song upon being returned to their original habitat.114
There’s clearly a range of opinion, some of it impassioned, on whether such experiments on

kittens and birds should be done. A much greater consensus exists that they should not be
performed on young humans. However, cases of lunatic or criminal behavior by parents have
occasionally created such experiments. “Genie,” studied in the 1970’s by my colleague Susan
Curtiss,115 was forced by a psychopathic father to spend the first thirteen years of her life strapped
113
For details on this work see http://neuro.med.harvard.edu/site/dh/b50.htm (David Hubel).
114
See http://www.neuro.duke.edu/faculty/mooney/.
115
Genie: A Psycholinguistic Study of a Modern-Day “Wild Child” (1977) Academic Press.
into a high chair, with little human contact and no linguistic input. When she was rescued and
released into a more normal environment, her linguistic development was slow and difficult.
Although she acquired a fair amount of vocabulary, she was unable to acquire the syntactic rules of
English. When Genie was last studied, around age 20, she still was essentially not a speaker of
English. Prof. Curtiss contends that she was unable to acquire language fully because she began
the task after the end of the critical period.
The psychologist Steven Pinker has conjectured that critical periods occur when the members
of the species learn the skill when young; the neural apparatus for learning is programmed to
atrophy at the end of the critical period, to avoid the metabolic cost of retaining it.
5.3 Genetic defects of language
It is by now fairly well established that the syndrome called Specific Language Impairment,
which is marked by inferior ability to use language (but normal intelligence), has at least some
genetic component; indeed, investigators have located families in which multiple members suffer
from the syndrome, and at least one specific gene has been located that is implicated in Specific
Language Impairment. The relevance of all this is called into question, however, by some scholars,
who note the possibility that the impairment may involve some fairly “low-level” defect of
phonetic process that could be the cause of the higher-level language difficulties.
6. Language and evolution
Certainly, if it is true that humans are biologically specialized for language it would be
reasonable to attribute this to natural selection, the source of all adaptive specializations in species.
Steven Pinker and Paul Bloom (1990)116 offer reasons why an innate ability for language would
have conferred a selectional advantage on our distant ancestors and thus shaped their evolution.
Pinker and Bloom endorse in passing a theory due to the phonetician Philip Lieberman that
our vocal tracts (mouth, throat, larynx) were evolutionarily shaped to permit speech. Lieberman’s
idea is that in evolving a long, arched vocal tract, we slightly increased our risk of choking to death
while swallowing (the food and air paths cross in our elongated pharynxes). The evolutionary
payoff, Lieberman claims, was highly intelligible speech. His theory remains controversial among
paleontologists.117
7. Summary
The scope, complexity, and sheer difficulty of human language grammars has led to the
hypothesis that children learn it with the aid of innate mechanisms. This innateness hypothesis
collides with the rival point of view that language can be learned with virtuosic inductive
116
From the journal Behavioral and Brain Sciences; on line at
http://www.bbsonline.org/documents/a/00/00/04/99/.
117
P. Lieberman and E. S. Crelin (1971) “On the speech of Neanderthal man,” Linguistic Inquiry
2:203-22.
mechanisms (which themselves may or may not be innate, or specifically linguistic). Innate
mechanisms of any sort assume that language is a biological specialization of humans, a claim
supported by the existence of a critical period, specific language impairment, and other evidence.
Lastly, such mechanisms have led scholars to try to speculate in as informed a way as they can
about the evolution of language.
Chapter 8: Review of Morphology and Syntax

1. Summary: grammar fragment
This is as far as we’re going to get concerning the syntactic analysis of English. It may be
useful at this point to summarize the rules and constraints as we developed them.
Our grammar has:
• an overall architecture (see p. 173), providing for the following:

• a set of phrase structure rules (final version given above in (146))
• a lexicon, with words bearing subcategorizations, and principles of lexical insertion
• a set of word formation rules (Chapter 2) that can be used to expand the inventory of
words in the lexicon
• A set of transformations:
➢ Tag Question Formation
➢ Subject/Aux Inversion
➢ Wh-Movement
➢ Topicalization
➢ It-Clefting
• Further transformations that form morphosyntactic representations:
➢ Genitive Case Marking
➢ Verbal Agreement
• A set of three island constraints
➢ Coordinate Structure Constraint
➢ Complex NP Constraint
➢ Wh- Island Constraint
• A postsyntactic module of inflectional morphology, including rules of
➢ Spell-Out (for contractions)
➢ Genitive Inflection (adding -’s)
This grammar suffices to cover a fragment of English. As mentioned earlier, a full grammar of
English would be vast—and not all the data have even been gathered yet.
The rest of this chapter consists of study exercises; these hopefully will be helpful if your
teacher puts a midterm exam in the middle of the course. Answers will be found at the end of the
chapter. These exercises cover everything up to, but not including, the unbounded transformations
and island constraints.
2. Morphology: Areas and affiliated skills
• Word formation:
➢ Writing word formation rules, which specify the base, what change in meaning and
(perhaps) part of speech is involved.
➢ Constructing iterated derivations, generally “inside out”, including multiple
derivations for ambiguous words like untieable.
• Inflectional morphology
➢ Finding morphemes and arranging them in position classes
➢ Writing inflectional rules, specifying the relevant features of the morphosyntactic
representation
➢ Ordering the rules correctly to obtain the right affix order
• The phonological form of inflection and word formation
➢ What change in the string of sounds is used to realize the word formation or
inflectional process?
——————————————————————————————————
Study Exercise #38: Persian verbal inflection
Positive present indicative
mixaram ‘I buy’ mixarim ‘we buy’

mixari ‘you-sing. buy’ mixarid ‘you-plur. buy’
mixarad ‘he/she buys’ mixarand ‘they buy’
Negative present indicative
nemixaram ‘I do not buy’ nemixarim ‘we do not buy’

nemixari ‘you do not buy’ nemixarid ‘you-plur. do not buy’
nemixarad ‘he/she does not buy’ nemixarand ‘they do not buy’
Positive past indicative
xaridam ‘I bought’ xaridim ‘we bought’

xaridi ‘you-sing. bought’ xaridid ‘you-plur. bought’
xarid ‘he/she bought’ xaridand ‘they bought’
Negative past indicative
naxaridam ‘I did not buy’ naxaridim ‘we did not buy’

naxaridi ‘you-sing. did not buy’ naxaridid ‘you-plur. did not buy’
naxarid ‘he/she did not buy’ naxaridand ‘they did not buy’
Positive subjunctive
bexaram ‘that I buy’ bexarim ‘that we buy’

bexari ‘that you-sing. buy’ bexarid ‘that you-plur. buy’
bexarad ‘that he/she buys’ bexarand ‘that they buy’
Negative subjunctive
naxaram ‘that I not buy’ naxarim ‘that we not buy’

naxari ‘that you-sing. not buy’ naxarid ‘that you-plur. not buy’
naxarad ‘that he/she not buy’ naxarand ‘that they not buy’
a. Find and gloss the morphemes

b. Arrange the morphemes into five position classes.
c. Make a set of inflectional rules to derive all these forms. You may assume these features:
[Polarity: positive, negative]
[Mood:indicative, subjunctive]
[Tense: past, present]
[Person: 1, 2, 3]
[Number: Singular, Plural]
d. Show that your rules work by deriving all six third singular forms.
e. Which orderings of the rules are needed?
Answer to Study Exercise #38, Persian Verbal Inflection
a. Morphemes:
ne-, na- negative

mi- present
be- subjunctive
xar ‘buy’
-id past
-am first person singular
-i second person singular
-ad, - third person singular
-im first person plural
-id second person plural
-and third person plural
b. There are five slots.
Negative Misc. Stem Tense Agreement

ne- mi- xar -id -am
na- be- -i
-ad/-
-im
-id
-and
c. Rules, in order
Mood Rule
Prefix be- when [Mood:Subjunctive, Polarity:Positive]
mi- when [Mood:Indicative, Tense:Present]
Negative Rule
Prefix ne- when [Polarity:Negative, Tense:Present]
na- all other [Polarity:Negative]
Tense Rule
Suffix -id when [Tense:Past]
Agreement Rule
Suffix -am when [Person: 1, Number:Singular]
-i when [Person: 2, Number:Singular]
- when [Person: 3, Number:Singular, Tense:Past]
-ad all other [Person: 3, Number:Singular]
-im when [Person: 1, Number:Plural]
-id when [Person: 2, Number:Plural]
-and when [Person: 3, Number:Plural]
d. Deriving “(s)he buys”:

xar [ Tense:Present, Polarity:Positive, Mood:Indicative, Person:3, Number:Singular]
mixar Misc.: , 
— Negative rule
— Tense Rule
mixarad Agreement Rule (, , )
Deriving “(s)he does not buy”:

xar [ Tense:Present, Polarity:Negative, Mood:Indicative, Person:3, Number:Singular]
mixar Misc. Rule (, )
nemixar Negative rule (, )
— Tense Rule
nemixarad Agreement Rule (, , )
Deriving “(s)he bought”:

xar [ Tense:Past, Polarity:Positive, Mood:Indicative, Person:3, Number:Singular]
— Misc. Rule
— Negative Rule
xarid Tense Rule ()
— Agreement Rule
Deriving “(s)he did not not buy”:

xar [ Tense:Past, Polarity:Negative, Mood:Indicative, Person:3, Number:Singular]
— Misc. Rule
naxar Negative Rule (, )
naxarid Tense Rule ()
— Agreement Rule
Deriving “that (s)he buy”:

xar [ Polarity:Positive, Mood:Subjunctive, Person:3, Number:Singular]
bexar Misc. Rule (, )
— Negative Rule
— Tense Rule
bexarad Agreement Rule
Deriving “that (s)he not buy”:

xar [ Polarity:Negative, Mood:Subjunctive, Person:3, Number:Singular]
— Misc. Rule
naxar Negative Rule ()
— Tense Rule
bexarad Agreement Rule (, )
e. Ordering:
The prefix rules must apply in the order shown, else would get *mi-ne- rather than the correct
form ne-mi-.
The suffix rules must apply in the order shown, else we would get (for first singular forms)
*-am-id, rather than the correct -id-am.
——————————————————————————————————
Study Exercise #39: English word formation with -ness
full fullness
squeamish squeamishness
lurid luridness
profound profoundness
Write a word formation rule in the format of Chapter 2.

Answer to Study Exercise #38, English Word Formation
[ X ]Adj → [ [ X ]Adj ness ]Noun
Meaning: “the quality of being Adjective”
——————————————————————————————
Study Exercise #40: Compounding in English
Use this rule, taken from Chapter 2:
English Compound Formation (word formation rule)
[ X1 ]Noun + [ X2 ]Noun → [ [ X ]Noun [ X ] Noun ]Noun

Meaning: “an X2 that has something to do with X1.”
to derive
a. tigerbird
b. law degree requirements (watch for inflection)
c. eggplant plant
Answer to Study Exercise #39, Compounding
a. tigerbird:
Given the existence of [ tiger ]Noun and [ bird ]Noun, we obtain [ [ tiger ]Noun [ bird ] Noun ]Noun,
which means “a bird having something to do with tigers”
b. law degree requirements:

Step 1: Given the existence of [ law ]Noun and [ degree ]Noun, we obtain
[ [ law ]Noun [ degree ] Noun ]Noun, which means “a degree having something to do with the law”, in
this case “degree awarded for the study of law”
Step 2: Given the existence of [ [ law ]Noun [ degree ] Noun ]Noun and [ requirement ]Noun, we
obtain [ [ [ law ]Noun [ degree ]Noun ]Noun [ requirement ]Noun ]Noun, which means “a requirement have
to do with a law degree”, in this case “requirements needed to obtain a law degree”
Step 3: a rule of inflection morphology gives us the plural law degree requirements.
c. eggplant plant
Step 1: Given the existence of [ egg ]Noun and [ plant ]Noun, we obtain
[ [ egg ]Noun [ plant ] Noun ]Noun, which means “a plant having something to do with eggs.” In this
case, the “having something to do with” is, “shaped like”, so we get the familiar vegetable.
Step 2: Given the existence of [ [ egg ]Noun [ plant ] Noun ]Noun and [ plant ]Noun, we obtain [
[ [ egg ]Noun [ plant ] Noun ]Noun [ plant ]Noun ]Noun which means “a plant having something to do with
eggplants.” In this case, the “having something to do with” is, premably, “suited for the
manufacture of”; i.e. a hypothetical future factory capable of manufacturing eggplants.
——————————————————————————————
Study Exercise #41: Hungarian word formation
Phonetic symbols: [ɔ] = “aw”, with lip rounding; [] is rather like “ny”; [] marks a long
vowel, [ˈ] goes before the stressed syllable; [ø] is like German “ö” or French “eu”.
ˈall ‘stand’ ˈallva ‘stand, scaffolding’

ˈigɔzol ‘justify’ ˈigɔzolva ‘certificate’
kiɔd ‘publish’ ˈkiɔdva ‘publication’
ˈmutɔt ‘show’ ˈmutɔtva ‘specimen, spectacle’
with a different vowel; ignore the vowel issue for now.
køt ‘tie’ køtve ‘bond, security’

ker ‘ask for’ kérveː ‘questionnaire’
Write a rule of word formation. The hardest part is specifying the meaning.
Answer to Study Exercise #40, Hungarian Word Formation
[ X ]Verb → [ [ X ]Verb va ]Noun
Meaning: “the result or instrument of Verbing”

-ve is due to phonology, a rule of Vowel Harmony
—————————————————————————————————————
Study Exercise #42: English word formation
a. Write a rule of word formation that can derive the italicized items.
a merry chase
a fifty-foot drop
The canoeists found that between Racquette Lake and Forked Lake was not a difficult carry.
He reached the water fountain and took a good long drink.
b. Write a rule of word formation that can derive the items in the right column.
kitchen kitchenette
pipe pipette118
rose rosette
statue statuette
118
This example works just fine in the spoken domain (pipe = [paɪp], pipette = [paɪpɛt]), but in the written
domain we have to assume a spelling rule. It’s pretty general in English that final letter e is dropped before a vowel-
initial suffix, as in ride ~ riding, dispense ~ dispensation, and so on. Prior to the application of this spelling rule,
pipette is pipe+ette.
Answer to Study Exercise #41, English Word Formation
a. [ X ]Verb → [ [ X ]Verb ]Noun
Meaning: “an instance of Verbing”
b. [ X ]Noun → [ [ X ette ]Noun ]Noun
Meaning: “a little Noun or thing having to do with Noun”
—————————————————————————————————————
Study Exercise #43: Luiseño Inflection(S. California)
Write morphological rules to cover inflection. You will have to make up your own
morphosyntactic representations.
Hint: think about whether a noun is something you could ever lose, and make up a feature to
describe this.
uˈmakinaki ‘your car’ uˈmatʃa ‘your back’

nuˈmakinaki ‘my car’ nuˈmatʃa ‘my back’
uˈpanki ‘your bread’ uˈkami ‘your son’
puˈpanki ‘his bread’ puˈkami ‘his son’
nuxarˈdinki ‘my garden’ nuˈna ‘my father’
puxarˈdinki ‘his garden’ uˈna ‘your father’
nuˈnotki ‘my boss’
uˈaki ‘your wife’119
119
To these may be added the somewhat startling nuˈpeʃli ‘my dish’, puˈpeʃli ‘her dish’; grammar has
an arbitrary side…
Answer to Study Exercise #42, Luiseño
The data illustrate the concept of inalienability, an inflectional category in many languages. A
thing is inalienably possessed if you could never truly be rid of it: your relatives, the parts of the
body.
Two rules are needed, which could apply in either order:
Person-Number Marking
X → nu X where morphsyntactic representation contains [Number:Sing, Pers:1]

X → u X where morphsyntactic representation contains [Number:Sing, Pers:2]
X → pu X where morphsyntactic representation contains [Number:Sing, Pers:3]
Inalienability Marking
X → X ki where morphosyntactic representation contains [Alienable:True]
_______________________________________________________________________
3. Study Exercise #44: Applying Word Formation Rules in Order
Find two meanings for overfillable and provide derivations for both.
Answer to Study Exercise #43, Applying Rules of Word Formation in Order
fill root
fillable [ X ]Verb → [ [ X ]Verb able ]Adj Meaning: ‘able to be Verbed’
overfillable [ X ]Adj → [ over [ X ]Adj ]Adj Meaning: ‘excessively A’
‘too fillable’; said perhaps of a conveniently wide-brimmed whiskey flask belonging to a

heavy drinker. “Alas for Smith, the flask proved overfillable, and he filled it again and again.”
compare: dreamable, drinkable; overeager, overproud
fill root
overfill [ X ]Verb → [ over [ X ]Verb ]Verb Meaning: ‘Verb too much’
overfillable [ X ]Verb → [ [ X ]Verb able ]Adj Meaning: ‘able to be Verbed’
‘liable to be overfilled’; said perhaps of a car engine that admits a dangerous excess of motor
oil because the dipstick gives an inaccurate reading: “my car engine is just way too overfillable”
compare: overeat, overdress; (same examples for -able)
————————————————————————————————————
4. Study Exercise #45: Chamorro Infixation
Formalize this rule of word formation using the symbols V, C, and numeral subscripts. State
in words what your rule does. [ʔ] is a “glottal stop”, a kind of consonant. It can be identified as the
little silence created by closing the vocal cords, heard in the middle of “uh-oh”.
lii ‘to see’ liniiʔ ‘(the) thing seen’

hatsa ‘to lift’ hinatsa ‘(the) thing lifted’
saan ‘to tell’ sinaan ‘(the) thing told’
hasu ‘to think’ hinasu ‘thought’
faisin ‘to ask’ finaisin ‘question’
Answer to Study Exercise #44, Chamorro Infixation
Assumed to be word formation.
[ C X ]Verb → [[ C in X ]Verb ]Noun Meaning: “thing that is Verbed”

1 2 1 2
In other words, “count off the first consonant, and place -in- right after it.”
————————————————————————————————————
Study Exercise #46: Yucatec Reduplication
Symbols:
t͡ʃ as in church
t͡s like Betsy but is just one sound, not two
k’ is k with extra oral pressure (“ejective”), and similarly for other sounds.
j is IPA for y
ʔ is glottal stop, heard in the middle of uh-oh.
sak ‘white’ sasak ‘very white’

t͡ʃit͡ʃ ‘hard’ t͡ʃit͡ʃit͡ʃ ‘very hard’
t͡ʃak ‘red’ t͡ʃat͡ʃak ‘very red’
k’as ‘bad’ k’ak’as ‘very bad’
nat͡s’ ‘near’ nanat͡s’ ‘very near’
haj ‘thin’ hahaj ‘very thin’
nat͡ʃ ‘far’ nanat͡ʃ ‘very far’
sis ‘cool’ sisis ‘very cool’
suut͡s’ ‘sour’ suusuut͡s’ ‘very sour’
jaab ‘many’ jaajaab ‘very many’
Answer to Study Exercise #45, Yucatec Reduplication
Assumed to be word formation.
[ X C ]Adj → [[ X X C ]Adj ]Adj Meaning: “very Adjective”

1 2 11 2
in other words: “copy all but the last consonant, and put the copy before the original.”
—————————————————————————————————————
Study Exercise #47: Obligatory Expression
In
My cat jumped.
specify two cases of obligatory expression (inflectional system of English forces you to
communicate particular information)
Answer to Study Exercise #46, Obligatory Expression
Cat is singular—the sentence means specifically one cat. This is because English nouns must
appear with either [Number:Singular] or [Number:Plural] in their morphosyntactic representations.
Jumped is past tense; tense must be marked in the morphosyntactic representation of the main
verb of a sentence.
———————————————————————————————————————
Study Exercise #48: Organization of the grammar: morphology
In the view of some linguists, the following is not only an impossible word of English, but
violates a fundamental principle of grammar. Explain.
*personsology ‘the study of collections of more than one person’
Answer on next page.
Study Exercise #49: Normative grammar
Briefly describe a matched-guise experiment.

Answer to Study Exercise #47, Organization of the Grammar: Morphology
In *personsology, a suffix for word formation, -ology, has been added “outside” of (hence,
“after”) an inflectional suffix. If word formation precedes lexical insertion and inflection follows
it, this should not be possible
Answer to Study Exercise #48, Normative Grammar
The core of a matched guise experiment is to have a bilingual or bidialectal person say
essentially the same thing in both of the language varieties she speaks, and then have experimental
subjects rate both voices for various traits—honesty, intelligence, friendliness, etc.—without
knowing that the “two” speakers are actually one. The idea is to get a controlled evaluation of what
people think about the varieties as such.
5. Syntax: Some skills
• Parsing sentences, particularly knowing where to attach modifiers.

• Using phrase structures to check if your structure is “compliant”.
• Establishing the phrase structure rules needed for new languages.
➢ Method: parse the sentences first, based on meaning, then make a short, clean set
of rules using ( ) and ( )*.
• Applying various transformations, given in the text.
• Determining the deep structure, given the surface structure, so you can give a whole
derivation.
• Show how the various island constraints rule out sentences.
• Demonstrate that a movement transformation applied unboundedly.
• Demonstrate that a transformation obeys a particular island constraint.
————————————————————————————————————
Study Exercise #50: Syntax: parsing
English phrase structure rules:
S → NP (Aux) VP
Art 
NP → NP   (AP)* N (PP)*(CP)
 
A A E
NP → Pronoun
VP → V (NP) (NP) (PP)* (CP)
VP → V AP
PP → P NP
CP → (Comp) S
NP → NP (Conj NP)*
VP → VP (Conj VP)*
PP → PP (Conj PP)*
S → S (Conj S)*
CP → CP (Conj CP)*
V → V (Conj V)*
Parse:
a. His brother and his wife’s book’s excessive length meant that it would cost a lot.
b. They awarded the key to the city (explicate both meanings)
Answer to Study Exercise #49, Syntax: Parsing
a. Note that since the inflectional suffix -’s is added by rules of morphology, it is not placed in
the deep structure tree. See below for how it is added.
VP
NP
NP
CP
NP S
NP NP VP
NP NP NP NP
| | |
Pro N Conj Pro N N Adj N V CompPro Aux V Art N
| | | | | | | | | | | | | | |
His brother and his wife book excessive length meant that it would cost a lot.
For the -’s, see a later exercise in this section.
b.
S
VP
NP
PP
NP NP
|
Pro V Art N P Art N
| | | | | | |
They awarded the key to the city
This is the meaning, “They awarded the key to the city (to someone, as an honor).” To the
city specifies what kind of key.
VP
PP
NP NP NP
|
Pro V Art N P Art N
| | | | | | |
They awarded the key to the city
This is the meaning, “The city was award the key (perhaps an important historical artifact for
the municipal museum.” To the city specifies what kind of act of awarding.
________________________________________________________________________
Study Exercise #51: Constituency Testing with It Clefting
A. As noted earlier, It-Clefting can be used to show what is an NP or PP, since it “targets”
these phrases; that is, it is a potential constituency test. Use this test to justify the constituency of
the two meanings of the sentence (b) in Study Exercise #49.
B. Use the It-Clefting constituency test to determine if the underlined sequences of words are
constituents.
a. I put the key under the mat.

b. I own the key under the mat.
Answer to Study Exercise #50, Syntax: Constituency Testing
Part A. In the first reading, the key to the city is held to be an NP; that is a constituent. It-
Clefting can only apply to constituents. When we cleft the key to the city:
It was [ the key to the city ]NP that they awarded ___.
we only get the reading where to the city specifies which key.
In the second reading, the key and to the city are separate constituents, and they can each be It-
Clefted on their own:
It was [ the key ]NP that they awarded ___ to the city.
It was [to the city ]PP that they awarded the key ___.
However, in each case, Clefting removes the ambiguity. It can only affect constituents; so it
reveals the constituent structure of the basic sentence for each of the two meanings.
Part B.
a. I put the key under the mat.
*It was the key under the mat I put.
So, the key under the mat is not a constituent. (It’s actually an NP followed by a separate PP.)
b. I own the key under the mat.
It is the key under the mat that I own.
So, the key under the mat is a constituent. (under the mat is part of this NP, modifying key)
——————————————————————————————————————
Study Exercise #52: Case Marking
Apply the case marking rule below to the structure you gave for sentence (a) in Study Exercise
#49 above.
English Genitive Case Marking
In the configuration shown:
NP1
NP2 ...
assign the feature [Case:Genitive] to the morphosyntactic representation of the rightmost word
in NP2.
You may assume that when the sentence is turned over to the component of inflectional
morphology, the following morphological rule applies:
Genitive Realization
Suffix -’s when the morphosyntactic representation contains [Case:Genitive].
Answer to Study Exercise #51, Syntax: Case Marking
I’ll show just the relevant NP. Items referred to in the rule are shown in boldface. We are
looking for:
NP1
NP2 ...
and are putting the feature [Case:Genitive] on the rightmost word of NP2. Here is one
application:
NP1
NP2
NP
NP NP
NP NP
| |
Pro N Conj Pro N N Adj N
| | | | | | | |
His brother and his wife book excessive length
[Case:Gen]
Here is the other application:
NP
NP1
NP2
NP NP
NP NP
| |
Pro N Conj Pro N N Adj N
| | | | | | | |
His brother and his wife book excessive length
[Case:Gen][Case:Gen]
These are realized later in the morphology as wife’s and book’s.

Study Exercise #53: Subcategorization
I.
Provide and justify subcategorization frames for:
a. opinion
b. transform (as a verb)
c. expire
II.
a. Explain why the grammar in this book would not generate these sentences:
*We took.
*We own.
b. Suppose for the moment that we had a grammar that did generate these sentences. Would
this be a case of overgeneration or ungeneration?
Answer to Study Exercise #52, Subcategorization
I.
a. opinion
[ ___ (CP) ] and [ ___ (PP) ]

Bill’s opinion
Bill’s opinion of Fred
Bill’s opinion that Fred will win.
b. transform (as a verb)

[ ___ NP (PP) ] and [ ___ PP ]
*Fred transformed.
Fred transformed the field.
Fred transformed the apple into a pear.
Fred transformed into a dwarf.
c. expire
[ ___ ]
Time expired.
*Time expired the men.
*Time expired to (or: of, above) the men.
II.
a. *We took is bad because take subcategorizes for an obligatory NP object. Its frame is:
[ ___ NP ]
The grammar won’t generate *We took because take cannot be inserted into the relevant tree,
which is:
NP VP
| |
Pro V
| |
we ___
The explanation for *We own is identical.
If the grammar did generate *We took, *We own, it would be overgeneration: outputting
examples that are ungrammatical.
Study Exercise #54: Syntax: Writing your own phrase structure rules
The following data are from a problem set book by Jeannette Witucki. It’s a pretty good book
(sadly, never formally published), but you should remember that Witucki isn’t necessarily teaching
exactly the same syntactic theory as me, and not all the loose ends will necessarily get tied up here.
The language here is Sango, a creole120 language spoken in the Central African Republic. The
word-by-word glosses are by me, guessing as best as I could from the sentence glosses, which are
Witucki’s.
1. mbi tɛ mbeni yama

I eat some meat
‘I eat some meat’
2. mbi kɛ tɛ yama la so
I prog. eat meat day this
‘I am eating meat today’
3. mbi mu na mɔ mersi mingi

I give to you thanks many
‘I give you many thanks’
4. lo mu na lo ngu
he give to him water
‘He gives him water’
5. lo kɛ mu na mɔ nginza
he prog. give to you money
‘He is giving you money’
6. i mu mbeni atɛmɛ ka
we give some stones there
‘We take some stones there’
7. mbi mu na mɛrɛngɛ ti lo nginza

I give to child of him money
‘I gave money to a child of his.’
8. mɔ zia ngu na wa
you put water to fire
120
A creole language arises when a simple, spur-of-the-moment contact language arising among
speakers of distinct languages (here, French and Ngbandi) is learned by children and elaborated (via
Universal Grammar, some think) into a full-fledged, fully-expressive language with native speakers.
‘You put water on the fire’
9. mɔ zia mbeni yama na sese

you put some meat to ground
‘You put some meat on the ground’
10. ala mu na lo ere so

they give to him name this
‘They gave this name to him.
11. ala zia yama so na sese

they put meat this to ground
‘They put this meat on the ground’
12. lo zia ala na kanga

he put them to jail
‘He put them in jail’
13. mbi zia mbeni atɛmɛ da

I put some stone here
‘I put some stones here’
14. mbi zia ala da la so

I put them here day this
‘I put them here today’
15. i faa na yama

we kill to meat
‘We kill animals’
16. i na mɔ faa na yama

we to you kill to meat
‘you and I kill animals’
17. i na ala faa na yama

we to them kill to meat
‘They and I kill animals’
18. i na lo faa kɔndɔ

we to him kill chicken
‘He and I kill chickens’
19. i faa kɔndɔ mingi

we kill chicken many
‘We kill many chickens’
20. ala faa woga mingi ti tɛ

they kill antelope many of eating?
‘They kill many antelope to eat’
21. mbi faa kɔndɔ ka na keke

I kill chicken there to tree
‘I killed a chicken there in the tree’
22. mɔ wara mbi ka la so

you find I there day this
You found me there today’
23. mbi wara kɔli ti mbi

I find man of I
‘I found my husband’
24. mbi wara ita ti mbi ti wale

I find sibling of I of woman
‘I found my sister’
25. mbi wara ita ti mbi ti kɔli

I find sibling of I of man
‘I found my brother’
Answer to Study Exercise #53, Sango Phrase Structure Rules
Hoping for a slightly cleaner answer I made a couple of perhaps dubious assumptions:
• la so ‘today’ is a fixed expression, i.e. a compound, as in English. I am classifying this as

an Adverb.
• I am treating kɛ as an Aux.
• I am treating the prenominal expression mbeni ‘some’ as an Adjective, letting postnominal
so and mingi be in the position for Articles.
The analytic strategy is:
• Provide conjecture trees for each sentence.

• Write down “sketch” phrase structure rules, that is, exactly what is needed to generate the
proposed tree.
• At the end, collate, generalize and simplify the rules using ( ) and ( )*.
1. mbi tɛ mbeni yama

I eat some meat
Pro V A N
‘I eat some meat’
Rules needed:
S → NP VP
NP → Pro
VP → V NP
NP → AP N
AP → A
Out on a limb: “some” as Adjective, since it looks like in general, the Articles follow the
noun.
2. mbi kɛ tɛ yama la so
I prog. eat meat day this
Pro Aux V N Adv.............
‘I am eating meat today’
S → NP Aux VP
NP → Pro (lots of these, I won’t repeat this one)
VP → V NP AdvP
AdvP → Adv
NP → N
3. mbi mu na mɔ mersi mingi

I give to you thanks many
Pro V P Pro N Art
‘I give you many thanks’
V → V PP NP
PP → P NP
NP → N Art
4. lo mu na lo ngu
he give to him water
Pro V P Pro N
‘He gives him water’
VP → V PP NP
PP → P NP (many of these, won’t repeat)
NP → N (many of these, won’t repeat)
5. lo kɛ mu na mɔ nginza
he prog. give to you money
Pro Aux V P Pro N
‘He is giving you money’
S → NP Aux VP (many of these, won’t repeat)

VP → V PP NP
6. i mu mbeni atɛmɛ ka
we give some stones there
Pro V Adj N Adv
‘We take some stones there’
V → V NP AdvP
AdvP → Adv (won’t repeat)
7. mbi mu na mɛrɛngɛ ti lo nginza

I give to child of him money
Pro V P N P Pro N
‘I gave money to a child of his.’
VP → V PP NP
NP → N PP
8. mɔ zia ngu na wa
you put water to fire
Pro V N P N
‘You put water on the fire’
VP → V NP PP
9. mɔ zia mbeni yama na sese

you put some meat to ground
Pro V Adj N P N
‘You put some meat on the ground’
VP → V NP PP
10. ala mu na lo ere so

they give to him name this
Pro V P Pro N Art
‘They gave this name to him.
V → V PP NP
NP → N Art
11. ala zia yama so na sese

they put meat this to ground
Pro V N Art P N
‘They put this meat on the ground’
VP → V NP PP
NP → N Art (won’t repeat this one)
12. lo zia ala na kanga

he put them to jail
Pro V Pro P N
‘He put them in jail’
VP → V NP PP
13. mbi zia mbeni atɛmɛ da

I put some stone here
Pro V Adj N Adv
‘I put some stones here’
VP → V NP AdvP
14. mbi zia ala da la so

I put them here today........
Pro V Pro Adv Adv
‘I put them here today’
VP → V NP AdvP AdvP
15. i faa na yama

we kill to meat
Pro V P N
‘We kill animals’
VP → V PP
16. i na mɔ faa na yama

we to you kill to meat
Pro P Pro V P N
‘you and I kill animals’
NP → Pro PP
VP → V PP
Note unusual construction, with a PP modifying a Pronoun within NP; not possible in
English.
17. i na ala faa na yama

we to them kill to meat
Pro P Pro V P N
‘They and I kill animals’
Nothing new in PS rules.
18. i na lo faa kɔndɔ

we to him kill chicken
Pro P Pro V N
‘He and I kill chickens’
Ditto.
19. i faa kɔndɔ mingi

Pro V N Art
‘We kill many chickens’
No new rules.
20. ala faa woga mingi ti tɛ

they kill antelope many of eating?
Pro V N Art P N
‘They kill many antelope to eat’
VP → V NP PP
21. mbi faa kɔndɔ ka na keke

I kill chicken there to tree
Pro V N Adv P N
‘I killed a chicken there in the tree’
I suggest that the PP ne keke is modifying the Adverb ka. Thus we need to put our Adverbs
inside Adverb Phrases (adjusting the previous rules that used bare Adverbs), and set up an Adverb
Phrase rule.
VP → V NP AdvP
AdvP → Adv PP
22. mɔ wara mbi ka la so

you find I there today.........
Pro V Pro Adv Adv
‘You found me there today’
A fairly clear case of two Adverbs.

23. mbi wara kɔli ti mbi

I find man of I
Pro V N P Pro
‘I found my husband’
NP → N PP
24. mbi wara ita ti mbi ti wale

I find sibling of I of woman
Pro V N P Pro P N
‘I found my sister’
NP → Pro PP PP
I assume that each PP is independently a modifier of ita ‘sibling’:
25. mbi wara ita ti mbi ti kɔli

I find sibling of I of man
Pro V N P Pro P N
‘I found my brother’
Just like #24.
This completes the gathering of the “sketch” phrase structure rules. We first collate them,
removing duplicates, like this:
S → NP Aux VP
S → NP VP
NP → AP N
NP → N
NP → N Art
NP → N PP
NP → Pro
NP → Pro PP
NP → Pro PP PP
VP → V NP
VP → V NP AdvP
VP → V NP PP
VP → V PP
VP → V PP NP
AdvP → Adv
AdvP → Adv PP
AP → A
then we can use the abbreviatory notations, and a little guess work, to produce a more general
grammar:
S → NP (Aux) VP
N 
NP → (AP)* Pro (PP)* (Art)
 
E
PP → P NP
VP → V (PP)(NP)(PP)(AdvP)*
AdvP → Adv (PP)
The most interesting of these is the VP rule. There are evidently VP’s with both NP PP and PP NP
order. My guess would be that this is determined by subcategorization; that is
Subcategorizes for [ ___ NP PP ]: zia ‘put’

Subcategorizes for [ ___ PP NP ]: mu ‘give’
Study Exercise #55: Recursiveness in Sango
Find a recursive loop in the phrase structure rules you just developed for Sango in Study
Exercise #53. If there is none, so state.
Answer to Study Exercise #54, Recursiveness
The one loop appears to be a two-rule case:
N 
NP → (Adj)* Pro (PP)* (Art)
 
E
PP → P NP
It’s virtually certain that Sango has subordinate clauses, which would produce at least one
further loop, as in English.
————————————————————————————————————
Study Exercise #56: Case marking in Pseudo-English
Write rules to mark case in this pseudo-English. You should write both syntactic rules of case
marking, to put the right morphosyntactic features in the right places, and rules of inflectional
morphology, to actually add the suffixes.
You will find it helpful first to parse the sentences. Other than the case marking, the language
is just like real English.
a. Johnwa gave Maryni a booko.

b. The king of Englandwa sold the queenni a book of poemso.
c. The kingwa sent the princeni of Wales a lettero.
d. The key to the doorwa is of great importance.
e. Fredwa thinks that the seller of fishwa read the papero.
f. Wewa told the daughterni of Mary a story about miceo.
g. Iwa sang a song about birdso to Alice.
h. Alicewa made the claim that shewa was leavingo.
Answer to Study Exercise #55, Case marking in Pseudo-English
Syntactic rules of case marking

Mark the rightmost word of the subject NP (daughter of S) as [Case:Nominative].
Mark the rightmost word of the last object NP (last NP daughter of VP) as
[Case:Accusative].
Mark the head of the first NP, when there are two of them in VP, as [Case:Dative].
Rules of inflectional morphology

Suffix -wa when [Case:Nominative]
Suffix -o when [Case:Accusative]
Suffix -ni when [Case:Dative]
Note: the nominative and the accusative here (but not the dative) are very roughly as in
Japanese.
Chapter 9: Semantics
1. Goals of semantics
Semantics is the branch of linguistics that studies meaning, particularly meaning as it is

conveyed by language. We can start out by asking what meaning is.
Meaning is a characteristic of symbolic systems. Language is by far the most elaborate and
powerful symbolic system that has ever been found. Our sentences are complex symbols,
physically realized in speech or writing, which bear meanings and express our thoughts.
Clearly, there is more to thought than the language that expresses it. Thought can exist in the
absence of language, since many animals can behave in a sophisticated and rational fashion
without having anything like human language.121 It also seems clear that we sometimes experience
thought in ways that are very direct and not linguistic. There is no need for thought to occur in a
linear sequence, as our words must; and moreover our visual thoughts are not particularly
expressible in language.
The development of a theory of thought is at present an active but speculative activity,

involving psychologists, philosophers, cognitive scientists, and experts in artificial intelligence.
One vindication for a proposed theory of thought would be if it could be embodied in a system that
could think and reason like a person. This remains a distant goal.
The focus in semantics is not as grand; we just want to know how language expresses thought.
The problem faced by semanticists is to study the ways in which language embodies thought,
without a well-developed theory of thought to go by. This problem has not stymied research,
however, because there are plenty of ways to conduct careful research that don’t require a full
theory of thought to make progress. For instance, one strategy that has been followed (it originates
in the field of philosophy) is to develop formal systems that determine the truth conditions of
sentences (properties of the world that must hold for sentences to be true), often in a small,
artificially-constructed world. This kind of approach requires a fair amount of development and
will not be taught here; instead, in the interest of a unified text I want to cover aspects of semantics
that interact most closely with syntax.
We will cover three aspects of linguistic meaning: predicate-argument structure,

anaphora, and operators and scope.
2. Predicate-argument structure
A predicate-argument structure is a simple semantic representation that limits itself to

depicting “who is doing what to whom.” For example, in a sentence like
121
A book on this topic I have enjoyed, written from a sober but exploratory viewpoint, is Animal
Minds, by Donald Griffin (University of Chicago Press, 1992). This continues to be an active area of research
as scientists document the ability of various animals to plan, to use tools, and to infer the mental states of
other beings.
John cooked the egg.
an act of cooking is described. We could characterize this act with the following predicate-
argument structure.
(179) An example predicate-argument structure
COOK ( (Cooker John ), (Cook-ee the egg ) )
In this structure, COOK is a “predicate”, which has “arguments”, in this case filling the slot of
Cooker and Cook-ee. The labels for the argument slots are arbitrary, and in fact I will sometimes
be choosing slightly silly ones, simply because they are short and clear.122 Argument slots are
sometimes designated with the term thematic roles. Predicates are capitalized simply to make
them easier to spot in the formula.
Predicate-argument structure contains both more and less information than a syntactic tree. It
contains more information because it identifies the argument slots for each participant in the
action. Syntactic structure instead places the participants (designated by NPs) in syntactic roles,
such as subject (NP daughter of S) or object (NP daughter of VP) — and, as we will see, there is
only a loose, flexible connection between thematic roles and syntactic roles.
Predicate-argument structure contains less information than syntactic structure for various
reasons. Most notably, predicate-argument structure is not meant to convey linear order; in (179)
above, COOK “has” the two arguments given, but there is nothing in the thought being expressed
that requires this order. The order that appears on the page is selected purely for convenience.
Linear order is a property of language, not of thought. Different languages have idiosyncratic
orders, including all six logically possible orders for simple two-NP sentences like John cooked the
egg. Here are all six, with examples of each.
SVO English, Spanish, Swahili, Mandarin

SOV Japanese, Korean, Turkish, Persian
VSO Tagalog, Biblical Hebrew, Irish
VOS Fijian
OSV Xavante (Brazil)
OVS Hixkaryana (Brazil)123
122
More ambitious theories try to generalize over slots, with widely-applicable terms. For instance,
Agent is used for any slot occupied by an entity that controls the action, Theme is used for objects in motion,
and so on. The details needed for this kind of generalization are not agreed upon by all linguists.
123
The last three orders, with object before subject, are rare.
2.1 Loose correlations between syntax and predicate-argument structure
Linguists have long noticed that aspects of meaning, as we might express them in a predicate-
argument structure, show a loose correlation with syntactic structure. Here are some common
generalizations.
Many predicate-argument structures involve some sort of actor; an entity that is in control and
performs the action. Most often, the actor is expressed syntactically as the subject (NP daughter of
S). This is true, for instance, in the following sentences.
(180) The norm: subject as actor
Alice sang.
Susan built the transmission.
Fred and Bill opened the package.
Many predicate-argument structures involve some sort of patient or “undergoer” of the

action. These are most often expressed as direct objects of the verb (in English, NP daughter of
VP. So, this would hold true of the transmission or the package in the sentences just given.
Verbs of giving or sending often have a recipient or beneficiary. These are often expressed
as an object (daughter of VP124), or as the object of a preposition, as below:
(181) The norm: recipient/beneficiary as object or PP object
Alice sent Fred a dozen red roses.

Alice sent a dozen red roses to Fred.
Bill baked Sheila some cookies.
Bill baked some cookies for Sheila.
These are only loose correlations. The verb undergo is striking in that its subject is usually the
patient of the action.
(182) Exception: subject as patient
John underwent surgery.
The verb experienced is unusual because its subject is the mental experiencer of the event;
normally experiencers are expressed in prepositional phrases.
(183) Norm: experiencer as PP object
The world seemed sad to Sam.
124
First daughter, when there are two.
(184) Exception: experiencer as subject
Sam experienced sorrow.
There are many other exceptional cases.
2.2 Variation in how syntax manifests predicate-argument structure
The grammar of particular languages often provide multiple possibilities for how a predicate-
argument structure is realized syntactically. These patterns hold good for many or even all of the
verbs of the language.
2.2.1 Passives
A well-known example of this kind is the passive construction, found in many languages:
(185) An active sentence and its corresponding passive
a. The doctor examined John.

b. John was examined by the doctor.
The first of these sentences is said to be in the “active voice” and the second in the “passive
voice.” For both sentences, the predicate-argument structure is something like this:
(186) Predicate-argument structure for both (185)a and (185)b
EXAMINE ( (Examiner Doctor ), (Examinee John ) )
The active voice for examine is probably more frequent; it makes the subject the agent of
examining, and the object into the thing examined. I suspect that this is the most common form of
expression for this verb. In the passive voice, the thing examined is expressed as the subject, and
the agent of examining is expressed (if it is expressed at all) as the object of the preposition by
within the VP.125 In the passive, the agent can also be simply suppressed; that is, omitted:
(187) A passive sentence with suppressed agent
John was examined.
We might plausibly give such a sentence a predicate-argument structure with a null argument,
something like this:
(188) Predicate-argument structure for (187)
EXAMINE ( (Examiner ), (Examinee John ) )
125
There is one other syntactic difference: passive sentences contain be as an Auxiliary, and the verb is
inflected in its past participle form.
The null would be interpreted as meaning that someone did the examining but the sentence does
not specify what.
Why might languages offer more than one way to connect up the thematic roles with the
grammatical positions? One view is that these variations are related to discourse structure: when
we converse or tell a story, we are not producing sentences in isolation; rather, each sentence
builds on a body of information that already exists and adds a new bit of information. Quite often,
at least in English, the subject NP embodies the pre-existing information, and the VP is what adds
something new. Thus, The doctor examined John is most naturally used where one is already
talking about the doctor, and John was examined by the doctor is most naturally used when one is
already talking about John. Thus, the passive construction permits the speaker to organize
information in a dialogue or narrative in a coherent way that builds on older information, by
making the old information the subject.
As already noted, the passive is also used to avoid the expression of certain arguments, as in
(187)-(188), John was examined, in which the Examiner argument is omitted. In English, this
pattern of omossion is confined to transitive verbs, and is implemented by leaving out the
prepositional phrase with by. German goes beyond English in allowing arguments to be
unexpressed even when the verb is intransitive:
Es wurde getanzt.
It was danced
‘There was dancing, people danced.’
DANCE ( (Dancer ) )
This construction is often called an impersonal passive.
2.2.2 Dative constructions
Another instance in which the same predicate argument structure has more than one syntactic
expression occurs with verbs of giving. Here is an example:
VP S
PP VP
NP NP NP NP NP NP
| | | |
N V Art N P N N V N Art N
| | | | | | | | | | |
Mary gave the book to Sue Mary gave Sue the book
The first tree illustrates the NP PP construction, in which the item given is the NP object and
the recipient is in the PP. The second tree illustrates the NP NP construction, in which the recipient
is the first NP and the item given is the second NP. Both have the same predicate-argument
structure:
GIVE ( (Giver Mary ), (Gift book ), (Recipient Sue ) )
As with passive, the variation may be related to the form of a discourse: the first sentence
would be more natural when one is already talking about the book, the second would more natural
when one is already talking about Sue. As in passive constructions, the new information comes
later in the sentence.
2.3 Propositions as filling argument slots
The following sentence has a predicate-argument structure in which one of the participants is a
Proposition — in this case, the content of Mary’s thoughts. To treat such cases, we need to allow
predicate-argument structures to have nested form, similar to the embedding structures of syntax.
Mary thinks that Bill jumped.

THINK ( (Thinker Mary ), (Proposition JUMP ( (Jumper Bill ) ) ) )126
In this structure, think has two arguments, namely Mary and the proposition that forms the content
of her thoughts.
A case of a type we will examine further later on is the causative construction. Here, the agent
carries out some (unspecified) action that causes an event to take place. Syntactically, English uses
the verb have to express the causation.
(189) A causative sentence and its predicate-argument structure.
Mary had John cook the egg.

CAUSE ( (Causer Mary ), (Proposition COOK ( (Cooker John ), (Cook-ee the egg ) ) ) )
In this sentence, Mary, the agent, caused the state of events described in Proposition to come into
being.
2.4 Cases of mismatch between syntax and predicate-argument structure I: weather it and
pleonastic it
Consider the following sentence, shown with a proposed predicate-argument structure.
It rained.
RAIN
In this and some later predicate-argument structures, I’ve used color to make sure that brackets
126
match up correctly. For correct bracket structure: every argument is surrounded by parentheses, and every list
of arguments is surrounded by parenthesis (even if there is just one argument).
What is special about such a case is that there are no arguments—raining is a thing that just
happens (nobody rains!).127 The it we get in syntactic structure is meaningless, and is evidently
present simply to satisfy the grammatical requirement (S → NP VP) that sentences of English must
have subjects. Such semantically empty elements are a mismatch between syntax and predicate-
argument structure. They illustrate that syntax involves demands of “pure form” that have nothing
to do with the expression of meaning.
The it that occurs as the subject of rain, snow, etc. is sometimes called “weather it.”
Here is another such case:
(190) A sentence with pleonastic it
It seems that Mary gave the book to Sue.

SEEM( (Proposition GIVE( (Giver Mary ), (Gift book ), (Recipient Sue ) ) ) )
Here again we have a semantically empty it, present to give the main clause a subject. This it is
sometimes called pleonastic it.128
A related construction gives the main clause a subject by taking the logical subject of the
embedded clause and expressing it “in the wrong position”:
Mary seems to have given the book to Sue.

SEEM ( (Proposition GIVE( (Giver Mary ), (Gift book ), (Recipient Sue ) ) ) )
In this grammatical construction, often called “Subject Raising”, the NP Mary occurs in a syntactic
location that is intuitively “higher” than its location in predicate-argument structure. Note that the
predicate-argument structure is assumed here to be exactly the same as in (190), though just as
with passive and datives, there are nuances of use related to discourse.
Study Exercise #57
Give predicate-argument structures for the following sentences. Be brave about labeling the
argument slots; this is somewhat arbitrary.
a. John appears to have been given a book by Sue.

b. It seems to have rained.
c. It seems that it rained.
d. It is felt that Bill rants.
127
Observe that this is different from John was examined and Es wurde getanzt, discussed above.
Someone really did examine John, and someone really did dance (we’re just not saying who). But no one
rains.
128
“Pleonastic” comes from the Greek for “superfluous”; the it is felt to be somehow unnecessary
(though it’s necessary for the sentence to be grammatical.)
a. John appears to have been given a book by Sue.
APPEAR ( (Proposition GIVE( (Giver Sue ), (Gift book ), (Recipient John) ) ) )
b. It seems to have rained.
SEEM ( (Proposition RAIN ) )
c. It seems that it rained.
SEEM ( (Proposition RAIN ) )
d. It is felt that Bill rants.
FEEL ( (Experiencer  ), (Proposition RANT( (Ranter Bill ) ) ) )
————————————————————————————————————
2.5 Cases of mismatch between syntax and predicate-argument structure II: causative verbs
We noted above ((189)) that English normally expresses causative sentences using the verb
have, with a separate, subordinate clause to depict the caused event. This is not the only way that
causation can be expressed; strikingly, other languages employ derivational morphology. The
relevant kind of morphological construction is known as a causative verb. Here, a prefix or suffix
appears on the verb (attached, in the theory assumed here, by a word formation rule), and the
whole construction is interpreted in predicate-argument structure with essentially the same kind of
structure we saw for (189), with a propositional argument for the caused event.
Turkish is one language that has causative verbs. The data pattern is illustrated below with
some sentence paired with their suggested predicate-argument structures.
(191) Morphological causatives in Turkish
a. Simple sentence with its predicate-argument structure
müdür mektub-u imzala-dı

director letter-ACC sign-PAST
‘The director signed the letter’
SIGN ( (Signer director ), (Signee letter) )

b. Causative sentence based on (a), with its predicate-argument structure
dişçi mektub-u müdür-e imzala-t-tı

dentist letter-ACC director-DAT sign-CAUS-PAST
‘The dentist made the director sign the letter’
CAUSE ( (Causer dentist ), (Event SIGN ( (Signer director ) (Signee letter) ) )
Syntactically, there is really just one clause, since there is only one verb present, and it assigns
case to the NPs in the usual way for a Turkish clause (Accusative for the first object, Dative for the
second). You can see that in the formation of causatives, the number of NPs allowed in the clause
goes up by one; the additional syntactic slot is needed to express the agent of causation; usually as
the subject.
Study Exercise #58
Provide plausible predicate-argument structures for the following two sentences of Turkish:
müdür Hasan-a mektub-u göster-di

director Hasan-DAT letter-ACC show-PAST
‘The director showed the letter to Hasan’
dişçi Hasan-a mektub-u müdür tarafından göster-t-ti

dentist Hasan-DAT letter-ACC director by show-CAUS-PAST
‘The dentist made the director show the letter to Hasan’
How is this construction similar to the English passive?

First sentence:
SHOW( (Show-er director ), (Shown letter), (Witness Hasan ) )
(or some other reasonable labels for the argument slots)
Second sentence:
CAUSE ( (Causer dentist ), (Event SHOW( (Show-er director ), (Shown letter), (Witness Hasan ) ) ) )
The parallel to English that I had in mind was the use of the postposition tarafından, which means
(roughly) ‘by’. In English passives, we provide no simple slot for the agent of the action (there is
no object position, and the subject position is taken up by the recipient of the action), so an added
by-phrase is used to express the subject. In Turkish causatives like the one in this exercise, the slots
provided by Nominative, Accusative, and Dative case are all “used up”, as it were, so the language
opts for the equivalent of the English by-phrase to express the fourth argument.
_____________________________________________________________________________
2.6 Predicate-argument structure in linguistic theory
There are several possibilities for integrating predicate-argument structure into linguistic
theory. One is to create a set of rules that inputs syntactic trees and derives the predicate-argument
structure from them. Another approach that has been taken is to let the predicate-argument
structure be the starting point of the derivation—embodying the message the speaker wishes to
communicate—and let the grammar find an appropriate tree structure or structures for
communicating this message. Still other approaches exist.
Study Exercise #59: The predicate-argument structure of As-phrases
The particle as has interesting syntactic and semantic behavior, in which the phrase structure
again mismatches the semantics. Some sample sentences:
(192) Four English sentences with as-phrases
a. We consider him as being eccentric.

b. They regard him as praiseworthy.
c. We judge it as unfortunate that he visited Mary.
d. We regard him as appearing to be sick.
We could accommodate as phrases in our English grammar with the following phrase
structure rules:
VP → V (NP) AsP (CP)

NP
AsP → as AP
VP
E
Furthermore, we must add rules of inflectional morphology that would ensure that in the VP that is
part of an as-phrase, the verb is marked to be a present participle (V-ing). Only a few verbs such as
regard and consider subcategorize for as-phrases.
What is interesting semantically about as-phrases is that express propositions without

including any extra CP — somewhat reminiscent of the Turkish causatives discussed earlier. For
example, in sentence (192)a above, we are not doing anything to him; rather, we are holding a
belief about him, which means that we are the mental experiencers of a proposition involving him.
This idea could be expressed with the predicate-argument structure below:
CONSIDER ( (Experiencer we ) (Proposition ECCENTRIC ( (Characterized he ) ) ) )
The proposition is, essentially, “he is eccentric”, without any verb or CP encoding this proposition.
Assign predicate-argument structures to (192)b-c above. Note that the it in (192)c is

pleonastic.
b. REGARD ( (Experiencer they ) (Proposition PRAISEWORTHY ( (Characterized he ) ) ) )

c. JUDGE ( (Experiencer we ) (Proposition UNFORTUNATE ( (Proposition VISIT ( (Visitor he ), (Visitee Mary ) ) ) ) ) )
d. REGARD ( (Experiencer we ) (Proposition APPEAR ( (Proposition SICK ( (IllPerson he ) ) ) ) ) )
————————————————————————————————
ANAPHORA
3. Defining anaphora
All languages have pronouns. For example, (193) gives the pronouns of English in their
various forms (this is an extended version of (118)).
(193) English pronouns
Nominative
I we
you you
he/she/it they
Objective
me us
you you
him/her/it them
Genitive
my our
your your
his/her/its their
Predicative Genitive129
mine ours
yours yours
his/hers/— theirs
Used after be, as in It is mine. There is no 3rd pers. singular inanimate form; for example, you can’t
129
say *That fuel pump is its, referring to a particular car. This is known as a “paradigm gap” and is widely
found in more heavily inflected languages.
Reflexive
myself ourselves
yourself yourselves
himself/herself/itself themselves130
Reciprocal (see Chapter 1)

each other
Pronouns are like nouns, but they get their reference from context, either the linguistic context, or
the situational context of speech. As already noted, the English pronouns are distinguished by
morphosyntactic features of Number, Case, and Person, and in the third person, for gender. Their
meanings are determined entirely by these features.
There are also pro-forms for other parts of speech. The phrases do it and do so are pro-forms
for Verb Phrases:
(194) Do it and do so as VP pro-forms
a. Sue wanted to [ teach Linguistics 865 ]VP but was too busy with other courses to [ do so ]VP.
b. Bill had to [ teach Linguistics 497 ]VP because no one else would [ do it ]VP.
Thus is a somewhat archaic proform for Adverb Phrases:
He did it thus.
The term anaphora refers, in linguistics, to the process whereby a pro-form gets its reference
from the meaning of another phrase; thus in:
Bill thinks he’s a genius.
we say that he makes anaphoric reference to Bill; likewise, above do so makes anaphoric reference
to teach Linguistics 865.
4. The Pronominalization Hypothesis and why it fails
A tempting analytical option for pronouns, assuming that we need transformations anyway, is
to suppose that pronouns are the result of applying a “Pronominalization” transformation.
Consider a sentence like (195):
130
It would also be sensible to include here the wh- pronouns: Nominative who, objective who
(normative English whom), Genitive whose, Predicative Genitive whose, missing the Reflexive.
(195) Alice told Sue that she was a genius.
Here, the pronoun she can refer either to Alice or to Sue. The sentence is therefore ambiguous. The
Pronominalization theory would say that when she means Alice, then the deep structure would be
as in (196) (tree omitted).
(196) Alice told Sue that Alice was a genius.
Analogously, when she means Sue, then the deep structure is as in (197).
(197) Alice told Sue that Sue was a genius.
According to the Pronominalization Hypothesis, there is a Pronominalization transformation that

converts the second instance of two identical noun phrases into the appropriate pronoun:
(198) Pronominalization (hypothetical rule)
When two NP occur in sequence, replace the second one with a pronoun whose
morphosyntactic representation bears matched values for the features [Number], [Animacy],
and [Gender].
It is easy to see that Pronominalization will convert both of the deep structures (196) and (197)
into the same surface structure, namely:
Alice told Sue that she was a genius
Assuming that the meaning of pronouns is determined by consulting their deep structure form, the
Pronominalization Hypothesis therefore succeeds in accounting for the ambiguity of sentence
(195), and indeed for sentences in general that are ambiguous because of pronoun reference.
Although the Pronominalization Hypothesis initially may seem reasonable (and indeed
enjoyed a brief vogue among linguists in the early 1960’s), it suffers from several problems.
First, there are sentences in which the deep structure that the Pronominalization Hypothesis
provides doesn’t mean what we want it to. If all pronouns are derived from full noun phrases, then
the deep structure of
Everyone thinks he is a genius
would be
Everyone thinks everyone is a genius.
But this deep structure clearly means something quite different from the surface structure. The
problem here evidently lies in the quantifier word everyone; we will return to quantifiers later on
in section 11 of this chapter.
A second problem with the Pronominalization Hypothesis is that there are pronouns that it
can’t derive, because the essential sequence of two identical NPs, as referred to in the
Pronominalization rule (198), is not present. Specifically, there are instances in which one uses a
pronoun in the total absence of any other NP.
The following example was invented by the linguist Howard Lasnik. Imagine a cocktail party
at which a man arrives, a stranger to all, who starts drinking heavily and getting into heated,
unpleasant discussions with all he encounters. After an hour of unpleasantness, he storms out of
the room, slamming the door behind him. At this point one could, without knowing the man’s
name, say:
(199) Well, he’s left.
Indeed, in this particular example it would be fine to say this sentence without even knowing the
name of the man that he refers to. The point is that if some pronouns are interpreted as referring to
a salient person in the context (that is, the pragmatic, real-life context), then we should consider the
possibility that even the she in Alice thinks she’s a genius is similarly interpreted — Alice is a
plausible person for she to refer to, since, after all, we’re talking about her.
A final problem with the Pronominalization Hypothesis is that, curiously enough, it appears to
lead us to infinite deep structures.131 Here is an example:
The girl who deserves it will get the prize she wants.
This sentence contains two pronouns, it and she. According to the Pronominalization Hypothesis,
we can get the deep structure by replacing these pronouns with copies of the full NPs to which
they refer. Doing this yields:
The girl who deserves [the prize she wants] will get the prize [the girl who deserves it] wants.
But this sentence also contains pronouns! Thus, to arrive at the true deep structure we will
have to substitute for these as well:
The girl who deserves [the prize [the girl who deserves it] wants] will get the prize [the girl
who deserves [the prize she wants]] wants.
And we are still not done, so:
The girl who deserves [the prize [the girl who deserves [the prize she wants]] wants] will get
the prize [the girl who deserves [the prize [the girl who deserves it] wants]]] wants.
No matter how long we keep going, we are still going to have uninterpreted pronouns in our
representation, so it’s clear that this process is never going to yield an interpreted representation.
131
The problem was noticed in the 1960’s by the linguists Emmon Bach and Stanley Peters, and is
sometimes called the Bach-Peters paradox.
The upshot is that deriving pronouns from full-NP deep structures does not seem promising as an
account of their semantics.
5. Interpretive rules for pronouns
Given what we’ve just seen, one might think that the right way to handle the meaning of
pronouns would be just to let them be pronouns; that is, nouns whose meaning is determined by
referring to a salient (highly noticeable) entity in the context (either linguistic context, or real-life
context), which matches the requirements of number (she vs. they), gender (she vs. he), and
animacy (she vs. it). In this approach, interpreting pronouns is relegated largely to the domain of
thought, not language—pretty much every sentence would be interpreted the way we interpret the
sentence Well, he’s left given above.
This is an appealingly simple theory, but it likewise cannot work. Research on the possibilities
of how pronouns refer has shown that there is indeed a heavy linguistic contribution to their
interpretation.
Consider the following very simple sentence:
John likes him.
Fluent speakers of English will assert pretty firmly that him cannot refer to John, even though there
is no logical reason why it could not. Similar sentences are:
He likes John.
He likes John’s brother.
He thinks John is a genius.
The reason why he cannot refer to John in these sentences turns out, as we’ll see shortly, to be
linguistic; that is, grammatical. Curiously, there seem to be linguistic rules that tell you what
certain pronouns cannot refer to. In what follows, we will work out the basics of these rules, and
find that they depend on syntax.
Our rules will not change the syntactic structure or words of sentences in any way; they
simply specify possible (or impossible) meanings. Thus, they are called interpretive rules.
We have already covered, informally, an interpretive rule for English, the Each Other
Reference rule, given as rule (26) in Chapter 1. Here, we will cover further rules, with a more
ambitious formalization of them.
6. Formal preliminaries
6.1 Subscripts and coreference
In what follows, we will use a standard notation for designating what pronouns refer to,
namely, subscripting. When I write this:
(200) Billi thinks hei is a genius.
I will mean: the reading of this sentence in which he is understood as referring to Bill. This is
denoted by the use of identical letters as subscripts.
When I write this:
(201) Billi thinks hej is a genius.
the nonidentical subscripts should be taken to mean that he, in this reading, refers to someone other
than Bill.
It will important later on to suppose that the indices are attached to the NP node, not further
down (like the Pronoun or Noun node). Thus the tree for (200) is as follows:
NPi VP
|
N V CP
| | |
Bill thinks S
NPi VP
|
Pro V NP
| |
he is Art N
| |
a genius
Terminology: in (200), Bill and he are said to be coreferent, meaning that they refer to the
same thing. In (201), Bill and he are not coreferent. Also, in the first sentence, Bill is taken to be
the antecedent for he, which means that it supplies the information about what he refers to.
We will also make use of a three-way distinction, which refers to the pronoun system as given
in (193).
• Reflexive pronouns are members of the set {myself, yourself, ourselves, ...}
• Regular pronouns are members of the set {I, me, you, he, them, ... }
• Full noun phrases are Noun Phrases that are neither reflexive pronouns or regular
pronouns; such as Sue, the president, my brother, etc.
Lastly, a technical definition:
(202) Definition: c-command
In a syntactic tree, constituent A c-commands constituent B if the mother of A dominates B.
In other words, A c-commands B if there is a path that
➢ starts at A
➢ moves up one node from A to A’s mother node
➢ travels exclusively downward through the tree and arrives at B.
Thus in the following example:
VP
CP
VP
NP NP NP
| |
N V Comp Pro V Art N
| | | | | |
Bill thinks that he is a genius
the NP Bill c-commands the pronoun he because you can go upward by one from the NP Bill,
arrive at S, then move downward through VP, CP, S, and thence to the NP he; see dotted arrows.
In the same example, the NP he does not c-command the NP Bill because once you’ve gone
uphill once from he, you can’t get to Bill by going just downhill:
VP
CP
VP
NP NP NP
| |
N V Comp Pro V Art N
| | | | | |
Bill thinks that he is a genius
He does c-command genius, however.
In general, we will speak of c-command only for NPs. In drawing these little arrows, you want
to start with the NP node, not the lexical item, in order to get the right result.
The term c-command apparently means “constituent-command”. It emerges from a period of

syntactic research that tried out a number of similar definitions, of which c-command appears, at
least for now, to be the simplest and most effective. We’ll see the relevance of c-command to
pronouns shortly.
6.2 Clausemates
Following up on the discussion in Chapter 1, we will also make use of the term clausemates,
defined as follows.
(203) Definition: clausemates
Constituents X and Y are clausemates if every S node that dominates X also dominates Y,
and vice versa.
Clausemates are often said to be in the same clause, which means the same thing.
In the following sentence:

VP
CP
VP
NP NP NP NP
| | | |
N V N Comp N V N
| | | | | | |
Bill told Sue that Fred likes Alice
the clausemate pairs are: Bill-Sue, Fred-Alice. Non-clausemates: Bill-Alice, Bill-Fred, Sue-Alice,
Sue-Fred.
A quick informal way to show clausemates is to bracket the sentences into domains of
clausematehood, like this:
VP
CP
VP
NP NP NP NP
| | | |
N V N Comp N V N
| | | | | | |
Bill told Sue that Fred likes Alice
Within brackets, any two NP are clausemates.

Study Exercise #60: Clausemates
Mary assumes that Fred will tell Sam that Alice saw Tom.
Parse the sentence, and draw the informal brackets to show the clausemate domains. Then
consider every pair of NP (there are ten pairs) and specify whether they are clausemates. Answer
on next page.
Pairwise:
Mary-Fred: no
Mary-Sam: no
Mary-Alice: no
Mary-Tom: no
Fred-Sam: yes
Fred-Alice: no
Fred-Tom: no
Sam-Alice: no
Sam-Tom: no
Alice-Tom: yes
_____________________________________________________________________
7. An interpretive analysis for reflexive pronouns
Using the approach just described, we can write the following rule of interpretation for
reflexives:
(204) Reflexive Interpretation
A reflexive pronoun must be coreferent with an NP that

(a) is its clausemate; and
(b) c-commands it.
Here are examples, labeled for how the rule works. As you read these examples, I suggest you
draw the tree, consult the definitions of c-command and clausemate, and check the rule is working
correctly.
*Himself sings.
Here, there’s no NP for himself to be coreferent with, so it receives no interpretation. The

standard assumption, which we will follow, is that a sentence with an uninterpretable pronoun is
ungrammatical.
Next consider:
Maryi congratulated herselfi.
This one is fine; the NP Mary c-commands the NP herself and, since there is just one clause,
the two are clausemates. The correct indexation (note: on the NPs, not lower down) is shown in
the tree above.
Next consider:
*[Maryi’s brother] congratulated herselfi.

Here, Mary is a clausemate of herself, but doesn’t c-command it—the mother of Mary is the
higher NP Mary’s brother; so Mary is not “high enough” in the tree to c-command herself.
Next consider:
*[Mary's brother]i congratulated herselfi.
This has the same tree, but different indices. Here, the NP Mary’s brother does c-command
the NP herself, and is a clausemate. The problem here is not with Reflexive Interpretation, but
rather with the morphosyntactic representation: brothers are always male, and herself is
[Gender:feminine]), so the sentence is still ungrammatical. Let us record this feature-matching
principle for future reference:
(205) Feature-matching in pronouns
In order to be conindexed with an NP, a pronoun must bear a morphosyntactic

representation that matches the NP in the features [Gender], [Number], and [Person].
Now consider:
(206) [Mary’s brother]i congratulated himselfi.

This one matches all requirements (gender match, c-command, clausemate condition), and is fine.
The next case to consider is:
(207) Maryi said that Tom congratulated herselfi.
Bad: Mary is not the clausemate of herself (herself is in the lower S, Mary is not).
We can try a different indexation for the same sentence:

(208) Mary said that Tomi congratulated herselfi.
Bad: Tom is a c-command clausemate but because it is a name for males there is a featural
mismatch with the pronoun, following (205).
Study Exercise #61

*The fact that Maryi lost the race surprised herselfi.
This one is bad; give the explanation. Answer on next page.

This one is bad for two reasons. First, Mary doesn’t c-command herself; specifically, the
mother of Mary is the S Mary lost the race, which doesn’t dominate herself. Second, Mary is not
the clausemate of herself.
————————————————————————————————————
7.1 Each other
The phrase each other, discussed in Chapter 1, is a reciprocal pronoun, not a reflexive. For
reasons of meaning, it requires a plural antecedent, but as far as the conditions on its reference it
works essentially like a reflexive and is normally analyzed using the same sort of rule. Thus:
[ John and Mary ]i like [each other]i. Ok.

*[ Each other ]i like [ John and Mary ]i. Bad: c-command condition violated
*[ John and Mary ]i think I like [each other] i. Bad: clausemate condition violated
8. An interpretive analysis for regular pronouns
The regular pronouns (like she, him, us, our, etc.) are used quite differently from reflexives.
For one thing, they can be used without any linguistic Noun Phrase to refer to at all—as in the
“Well, he’s left” example given earlier in (199) above.
The key to these pronouns, in the view of many linguists, is that you specify not what they can
refer to, but rather what they cannot refer to. Here is a version of the rule commonly proposed:
(209) Regular Pronoun Interpretation132

A regular pronoun cannot be coreferent with a c-commanding clausemate.
Some examples follow.
8.1 Pronouns alone
Hei left.
This is fine: there is no NP in the sentence that he is required to be non-coreferent with, and
the sentence is freely usable whenever there’s an obvious enough male entity available for he to
refer to. This could be someone mentioned in a previous sentence, or someone noticed in the
physical surroundings, as discussed earlier for sentence (199). Compare the ungrammatical
*Himself left, where the reflexive requires an overt NP to refer to.
8.2 No c-command, coreference ok
[Maryi's brother] congratulated heri.
This is fine, because Mary doesn’t c-command her.
8.3 No coreference — always ok
[Maryi’s brother] congratulated herj.
In the linguistics literature this rule is often called “Principle B.” The rules for reflexives and
132
reciprocals are subsumed together under “Principle A”. In this introductory text I have opted for descriptive
rule names instead.
The subscript j means that the her refers to a female person other than Mary. This is fine, too,
since Regular Pronoun Interpretation doesn’t actually require that pronouns be coreferent with any
other NP in the sentence. Thus, this sentence could appear in a context like this:
Alice sang incredibly well, enough to convince her sternest critics. In fact, even Mary’s
brother congratulated her.
I think it’s pretty clear that in this sentence it would be possible for her to refer to Alice.
8.4 C-commanding clausemate: coreference impossible
VP
NPi NPi
| |
N V Pro
| | |
*Alice congratulated her.
This one is no good: Alice is the clausemate of her, and also c-commands her, so it can’t be
coreferent. However, with distinct reference, the following reading is ok:
Alicei congratulated herj.

8.5 C-command but not clausemate: coreference ok
VP
CP
VP
NPi NP NPi
| | |
N V Comp N V Pro
| | | | | |
Mary said that Tom congratulated her.
This one is fine: Mary does c-command her, but it is not the clausemate of her, so
Regular Pronoun Interpretation doesn’t rule out this reading.
8.6 C-command but not clausemate: non-coreference ok
Maryi said that Tom congratulated herj.
This is likewise fine, herj refers to some female person mentioned earlier or physically present.
8.7 No c-command, not clausemate: coreference ok
The fact that Maryi lost the race surprised heri.
Ok, Mary is neither a clausemate of her, nor does her c-command Mary, so the coreference is
allowed.
8.8 Backwards coreference
An intriguing prediction of the analysis is that you could, in principle, get sentences in which
the pronoun actually comes before the full NP with which it is coreferent. These do in fact arise,
though because of additional factors they won’t be found in all places you would expect them.
Here is an example:
The fact that shei lost the race surprised Maryi.
This sounds best only under particular conditions of emphasis and intonation. In particular,
you can’t utter Mary with a full phrasal stress, as if the name were being introduced to the
conversation for the first time — if Mary were new information, you wouldn’t have been referring
to her with a pronoun! The sentence sounds ok if you say:
The fact that shei lost the race SURPRISED Maryi.
Of course, since Regular Pronoun Interpretation only forbids coreference, the following
reading is also acceptable:
The fact that shej lost the race surprised Maryi.
9. An interpretive analysis for full noun phrases
One wouldn’t think that there need to be any rules for the meaning of full noun phrases, but
these are in fact needed. Consider a sentence like:
*Hei thinks Billi is a genius.
The coreference shown is impossible, even though nothing we’ve said so far rules it out. The
rule commonly used is this one:
(210) Full Noun Phrase Interpretation133

A full noun phrase cannot be coreferent with a c-commanding Noun Phrase.
This rules out *Hei thinks that Billi is a genius because he c-commands Bill and Bill is a full
NP.
VP
CP
VP
NPi NPi NP
| |
Pro V Comp N V Art N
| | | | | | |
He thinks that Bill is a genius
Indeed, the same rule predicts that in
Bill thinks that Bill is a genius.

Bill saw Bill.
we must interpret the two Bill’s as being different people; that is, these sentences must be
interpreted:
Billi thinks Billj is a genius.

Billi saw Billj.
If neither copy of Bill c-commands the other, then coreference becomes more or less ok:
[ well, at least ] Billi’s mother likes Billi.
133
In the linguistics literature this rule is often called “Principle C.” Principles A and B were
mentioned in footnote 132.
Study Exercise #62: in the following, why can the two Bill’s be the same person? Show the
relevant structure.
The idea that Bill might have the lowest score bothers Bill.
This is ok because neither instance of Bill c-commands the other. The mother of the first Bill
is S, which doesn’t dominate the second Bill; and the mother of the second Bill is VP, which
doesn’t dominate the first Bill.
_______________________________________________________________________
10. Summary of anaphora
We’ve now done a particular corner of English semantics, setting out rules of semantic
interpretation for anaphoric expressions. Dividing all NPs into the categories of Reflexive
Pronouns (with their close relative Reciprocal Pronouns), Regular Pronouns, and Full NPs, we
developed three rules, one of which requires coreference in certain contexts, the other two of
which forbid it:
Reflexive Interpretation (204)
A reflexive pronoun must be coreferent with an NP that (a) is its clausemate; and (b) c-
commands it.
Regular Pronoun Interpretation (209)
A regular pronoun cannot be coreferent with a c-commanding clausemate.
Full Noun Phrase Interpretation (210)
A full noun phrase cannot be coreferent with a c-commanding Noun Phrase.

OPERATORS AND SCOPE

11. Operators and scope in formal logic
The idea of operators and scope was incorporated into linguistics from the field of symbolic
logic, a branch of philosophy.134 Logicians express (certain aspects of) meaning with formulas like
the following.
For all x
P is true of x
x(P(x))
The meaning of the formula is, “for all x, P is true of x”. If we were applying this formula to a real-
life situation, we might imagine a universe that consists of the students in Linguistics 20, and P
represents “has the flu”. The formula could then be interpreted as “For every student in Linguistics
20, it is the case that that student has the flu.” Or, more fluently: “Every student in Linguistics 20
has the flu.” In the formula, x is an operator, x is a variable, and P is a predicate (just like we
saw with predicate-argument structure).
To see the concept of scope, let us compare two formulae that are more complex. In (211) I
give the first one. There is a page break to facilitate comparison.
134
At UCLA you can study the basics of this field in Philosophy 31; indeed almost every university
has an introductory logic course.
(211) A logical formula containing an operator and a variable, in one scope relation
For all x
P is true of x
implies that
Q is true
x(P(x)) → Q
Pursuing our real-life interpretation, we might suppose that Q means “the professor postpones the
exam”. The symbol → means “if … then”. The interpretation would then be “If every one of the
students in Linguistics 20 has the flu, then the professor will postpone the exam.”
In (212) there is a similar formula, but with a different location for the red-colored right
parenthesis.
(212) A similar logical formula involving a different scope relation

For all x
P is true of x
implies that
Q is true
x(P(x) → Q)
With the parenthesis relocated, “for all” now covers the entire rest of the formula, rather than just
P(x). Thus, in the real-life interpretation of the formula, this would be “For every student, if that
student has the flu, then the professor will postpone the exam.” — this would imply that the
professor will postpone the exam even if there is just one case of the flu in the class.
One can speak here of an operator having scope. In the first formula, the scope of the operator
x is just P(x) (informally, “x has the flu”) whereas in the second formula the scope of the operator
x is P(x) → Q (informally, “if x has the flu, the professor will postpone the exam”).
The operator x is of a particular kind, called a quantifier. It means “all” (symbol: inverted
A). The other quantifier most often used in elementary logic is x, which means “at least one x”
(inverted E, “exists”).
In logic, these concepts are employed in the study of the principles of valid reasoning. For
example, the formula ~x(P(x)) → y(~P(y)) (which means “If it is not the case that P is true of
all x, then there must exist some y of which P is not true”) represents a case of valid reasoning. It is
true irrespective of what we mean in the real world by the entities x and y or the predicate P. Over
the centuries, logicians have provided mathematical proofs for a vast number of such formulae,
thus providing a mathematically valid, more trustable, basis for deductive reasoning.135
12. Operators and scope in language: some examples
In linguistics, the focus is less on proofs of validity, and more on using logic to provide a
precise and interpretable characterization of meaning. In fact, linguistic meaning is much richer
than what can be expressed with the logic taught in beginning logic courses, and finding a rich
enough formal system to characterize human language continues to be a challenge for logicians
and linguists alike.
We can start by seeing that the logical notions of quantifier, scope, and variable are expressed
fairly directly in English (or indeed in any other language). Here is an example:
Every boy sang.
Here, we have the following:
Every boy a kind of (restricted) universal quantifier (x, x a boy)

x sang a predicate
Putting these together, we get something like (213):
(213) (x, x a boy) (x sang)
If you want to read (213) aloud, you can say “For all x such that x is a boy, x sang.”
The restricted form of the quantifier, (x, x a boy), is very characteristic of human language:
it is quite rare that we would want to quantify over absolutely everything (boys, turtles, personal
qualities, months, neutron stars …), and typically quantifiers hold over some modest sub-domain,
such as the class of boys.
Like logicians, linguists generally place operators at the left of the domain over which they
have scope; this is a matter of convenience and is an arbitrary convention. So, for instance, a
sentence like:
Jane taught every student.
would be expressed as:
(x, x a student) (Jane taught x)
135
Deductive reasoning goes from premises to conclusions. For inductive reasoning, which goes from
observations to inferences, there are other theories of logic.
principle, we could integrate such expressions with the predicate-argument structure

In
developed earlier in this chapter. Under this approach, the meaning would appear like this:
(x, x a student) (TEACH ( (Teacher Jane ) (Teachee x) )
For brevity, in what follows I will skip this step and simply place the quantifiers and variables
into ordinary syntactic structure.
The material below covers three topics related to quantifiers in language: their interaction with
pronouns, their interaction with each other, and their interaction with clause structure.
12.1 Pronouns as variables
So far, we have treated pronouns as NPs that refer to things. When a pronoun is coindexed
with another NP (Billi thinks hei is tall) it is meant to refer to the same real-world thing as that NP.
When a pronoun has its own distinct index (Well, hei left), it is meant to refer to some real-world
thing assumed to be identifiable by the context but not mentioned linguistically.
However, not all pronouns refer to things. The other use of pronouns is as the linguistic
manifestation of logical variables. This can happen when there is a logical operator, such as a
quantifier, elsewhere in the sentence. Consider the following sentence.
Every boy thinks that he is smarter than average.136
There is a boring reading:
[ Every boy ]i thinks that [ he ]j is smarter than average.
where he is someone else, like, say, Fred. We focus here on the interesting reading:
[ Every boy ]i thinks that [ he ]i is smarter than average.
This sentence would hold true in a world in which Fred thinks Fred is smarter than average, Bill
thinks Bill is smarter than average, Sam thinks Sam is smarter than average, and so on. In this
reading, the pronoun he does not refer to anyone. Instead, it acts as a logical variable, and indeed
we have two instances of the same variable under the scope of a single quantifier.
Every boy is the quantifier (x, x a boy)

he is a bound variable (x)
x thinks x is smarter than average is a complex predicate, with two variables
Putting these together, we get the structure in (214):
136
You may recall from section 4 of this chapter that sentences like this are part of the reason not to derive
pronouns from full NP’s by a transformation.
(214) (x, x a boy) (x thinks x is smarter than average)
In sum, the pronoun he is not referential but rather is the linguistic means for expressing the
second instance of the variable. (The first variable simply occurs in the syntactic location of the NP
containing the quantifier; see rules below for how this can be derived).
Let us return briefly to the “boring” reading mentioned above: the pronoun he does not have
to act as a bound variable, but can also be an ordinary pronoun, which can refer to some male
person who happens to be under discussion, such as Fred. Thus, the boring reading could be
represented as in (215):
(215) (x, x a boy) (x thinks hei is smarter than average)
where hei is a pronoun referring to someone in the environment, in the ordinary way.
Some terminology: we say that in the first reading, he acts as a variable that is bound by the
quantifier. The pronouns of a language play at least two roles: they either simply refer to other
entities, or they act as bound variables.
What is the mechanism whereby pronouns get interpreted as bound variables? As a rough
approximation, we can make use of the discussion of pronoun reference from earlier in this
chapter. There, we studied rules that assign indices to pronouns and their antecedents, to express
ordinary coreference and non-coreference. The extension of this idea in the present context is this:
if a pronoun gets coindexed with a quantified NP, then the relationship is then semantically
interpreted not as coreference, but as an operator-variable relationship. Thus, for instance, the rules
of anaphoric interpretation permits the following coindexation for the NPs in the sentence we are
working with (he is not the clausemate of every boy, so Regular Pronoun Interpretation ((209)) is
satisfied; and he does not c-command every boy, so Full Noun Phrase Interpretation ((210)) is
satisfied).137
137
The tree below is not compliant with our phrase structure rules. The change needed is pretty straightforward:
AP → (Adv) A (PP). All comparative adjectives (“X-er”) can take a PP with than.
Because every boy is a quantified NP, this must be further translated to
(x, x a boy) (x thinks x is smarter than average)
More specifically, when a quantified NP is logically interpreted as an operator-variable

combination, any pronouns coindexed with it must be assigned the same variable. The following
match-up illustrates this:
[ Every boy ]i thinks that [ he ]i is smarter than average.
(x, x a boy) (x thinks x is smarter than average)
The pattern holds not just for regular pronouns but for coindexed reflexives.
[ Every cat ]i washed [ itself ]i.
(x, x a cat) (x washed x)
We will give explicit rules for doing this later on.

13. Logical form
It is time now to integrate the discussion into a general approach to semantics. Note that the
following is just one (well represented) viewpoint among many.
The core idea is that the rules of the semantics create from syntactic representation138 a
separate representation of the sentence’s meaning (or, in cases of ambiguity, more than one
representation). Such a semantic representation is often called the logical form of a sentence.
Logical form is meant to be specifically linguistic in character; it only represents the contribution
of language to meaning and is certainly not the “language of thought”, if such a thing exists — as
noted already, our thoughts involve all sorts of non-linguistic inferences and associations,in
addition to language.
Here are some of the steps that would be needed to construct a logical form from a syntactic
structure. As some (probably early) stage we would establish the possible references of pronouns
and reflexives through the assignment of indices, using the rules of Reflexive Interpretation,
Regular Pronoun Interpretation, and Full Noun Phrase Interpretation, given earlier in this chapter.
Another step would be to convert quantified NPs into operator-variable pairs, to indicate scope, as
described in the previous section; at this stage coindexed pronouns must be converted to variables
under the scope of the same quantifier. Yet another step would be to establish precisely “who is
doing what to whom” by replacing the syntactic tree with an appropriate predicate-argument
structure.
Here are a couple of examples of how all this might work. In
Every boy thinks he is smart
the rule of Regular Pronoun Interpretation would (as one of its options) coindex every boy and he,
thus:
[ Every boy ]i thinks [ he ]i is smart
Next, the quantified NP every boy would be converted to an operator-variable combination. Since
he is coindexed with every boy, it is a assigned the same variable x (this will be done more
explicitly below in section 19):
(x, x a boy) ( x thinks x is smart )
Then the whole expression could be converted to a predicate-argument structure, yielding a logical form:
(x, x a boy)((THINK((Thinker x), (Proposition SMART((Assessee x)))))
For the sentence Mary seems to like every boy, the same processes would yield:
138
Most likely, from surface structure. The traces left by movement rules generally make it possible to
cover the effects of deep structure on meaning; they serve as a “memory” for the location of phrases at the
deep structure level.
(x, x a boy) ( Mary seem to like x )
and from this we would derive:
SEEM ( (Proposition (x, x a boy)( LIKE ( (Liker Mary ) (Lik-ee x ) ) ) ) )
This is at present only an outline scheme. In the pages below, I’ll discuss briefly the rules for
converting quantified NPs into operator-variable pairs, which will flesh out the scheme a bit.
However, we will henceforth skip the step of creating predicate-argument structure from syntax.
14. Sentences with two operators
It is possible for a sentence to include two logical operators. Here, the two operators often
interact with each other, yielding different meanings. For example, speaking of an archery
tournament, we might utter sentence (216):
(216) A sentence ambiguous because of two interacting quantifiers
Two arrows hit every target.
This sentence is ambiguous, in the following way. Suppose that the archers are so impoverished
that between them they could bring a total of only five arrows to the tournament. Thus, each arrow
has to be used repeatedly. Suppose further that the archers used a total of five targets. Here is one
reading: two of the arrows (perhaps the straightest ones) were used so successfully that during the
course of the tournament they penetrated every one of the five targets.
(217) “Two arrows hit every target”: Scenario I
In the other reading, we would find that inspecting the targets at the end of the tournament,
each has at least two holes in it.139
139
I’m saying “at least” because this seems to be the default interpretation of numerals like two; we could get a
different interpretation by saying exactly two.
(218) “Two arrows hit every target”: Scenario II
A nice challenge to one’s ability to write clearly is the task of expressing, in a single sentence,
just one of the two meanings of a a double-quantifier sentence. Often, to do this well, it pays to use
“philosopher’s language” — highly formal, stilted wordings that achieve precision. For example,
the two readings of Two arrows hit every target can be summarized as follows:
(219) ‘There were two arrows such that they hit every target.’
True of Scenario I; diagram (217)
(220) ‘For every target, it is the case that two arrows hit it.’
True of Scenario II; diagram (218)
The paraphrases in (219) and (220) employ two common phrases that form part of this
“philosopher’s language” are such that and it is the case that — both of them have the effect of
canceling out an unwanted scope reading, so the paraphrase becomes unambiguous.
To handle the ambiguity we have just seen within the formal system we are developing, we
can employ two operators. The word every is a real-language version of the universal quantifier
x, ‘for all x’. Two is not an operator that is normally taught in introductory logic, but I think it is
intuitively clear that it is an operator of some kind. Thus, by putting the operators in the right
structural locations, we can characterize the ambiguity.
(221) Representing the two scopes for “Two arrows hit every target”
a. Scenario I
( For two x, x an arrow ) ( ( for every y, y a target ) ( x hit y ) )

‘There were two arrows such that they hit every target.’
b. Scenario II
( For every y, y a target ) ( ( for two x, x an arrow ) ( x hit y ) )

‘For every target, it is the case that two arrows hit it.’
This is an example of a scope ambiguity. In (221)a, the scope of the operator ( For two x, x an
arrow ) is ( ( for every y, y a target ) ( x hit y ) ). In (221)b, the scope of the operator ( For every y, y
a target ) is ( ( for two x, x an arrow ) ( x hit y ) ).
Another way of saying that same thing is that in (221)a, ( For two x, x an arrow ) takes scope
over ( for every y, y a target ), because ( for every y, y a target ) is inside the scope of ( For two x, x
an arrow ). In (221)b, ( for every y, y a target ) takes scope over ( For two x, x an arrow ).
15. Operator scope in multiclause sentences
Operators can have scope not just over other operators, but over particular clauses in a
sentence that has more than one clause. These cases are of special interest for us because they can
be used to show the close relationship of operator scope with syntactic structure.
Here is an example. The sentence at hand is:
(222) Sue shouted [ for us to give water to each runner ]
We need briefly to cover the syntax here. In one commonly-adopted analysis, for us to give water
to each runner is a CP, for is a Comp, us is the NP subject of for us to give water to each runner
and to is a particular sort of Aux used only in verbal infinitives. Shout is a verb that subcategorizes
for this particular kind of CP (often called an “infinitival clause”, since to give is the infinitive
form of give) Here is the proposed parse:
Now, let us consider the meanings at hand. The easy reading here, which I will call Narrow
Scope, is the one where Sue shouts just once, at the beginning of a marathon, the utterance (223):
(223) Hey! Give water to each runner!
In this reading, the scope of each is the embedded clause that reports what Sue shouted. Here is a
possible logical structure for this reading:
(224) Narrow scope reading of “Sue shouted for us to give water to each runner”
Sue shouted ( ( for each x, x a runner ) ( for us to give water to x ) )
For the other reading, imagine it’s a bit late in the day in a marathon, and the stragglers are
coming by the water station at Mile 23, spaced about two minutes apart. Whenever this happens,
Sue shouts (225):
(225) (Happens many times:) Hey! Give water to that runner!
Call this the Wide Scope reading. It could be represented like this:
(226) Wide scope reading of “Sue shouted for us to give water to each runner”
( For each x, x a runner ) (Sue shouted ( for us to give water to x ) )
In other words, for each passing runner, there was a “shouting event”, in which Sue directed the
workers to give that runner some water.
Here is an effort to provide single-sentence paraphrases of the two readings.
Narrow-scope reading ((224)):
The message expressed by Sue’s shouting was such that for each runner, we should give water
to that person.
Wide-scope reading ((225)):
For each runner, Sue emitted a shout to the effect that we should give water to that runner.
The general point of this example is that we can have a sentence that has just one variable, but
is ambiguous. This is because the sentence has two clauses, and thus two locations for the operator
to go.
Study Exercise #63

Provide paraphrases for both the wide and narrow scope readings of these sentences, which
are ambiguous in the same way as (222). Also, provide logical notation similar to the ones just
given.
a. I signed an order for each soldier to be given a medal.

b. I announced that progress has been made on every front.
a. Wide scope: “For each soldier, I signed an order that that soldier be given a medal.”
( For each x, x a soldier ) (I signed an order that ( x be given a medal ) )
Narrow scope: “I signed an order, whose content was that each soldier should be given a
medal.”
I signed an order such that ( ( for each x, x a soldier ) ( x be given a medal ) )
b. Wide scope: “For every front, I announced that progress was being made on that front.”
( For each x, x a front) (I announced that ( progress had been made on x ) )
Narrow scope: “I made an announcement, whose content was the progress was being made
on every front.”
I announced that ( ( For each x, x a front) ( progress had been made on x ) )
————————————————————————————————————
16. Creating operator-variable pairs from quantifiers in logical form
We have now covered three phenomena in which quantifiers interact with other elements:
quantifiers binding pronouns (section 12.1), quantifiers taking scope over one another (section 14),
and quantifiers taking scope over different clauses (section 15). With this in hand, we can now
consider what rules could be used to derive the logical form of quantified sentences. We know, up
front, that the rules need to have some flexibility, because often a single syntactic structure can
yield two different interpretations for quantification.
We first need a rule that translates quantified NPs into operators. This is not the final form of
the rule, but it will suffice for now
(227) Quantifier Translation (preliminary version)
Replace
[ every N ]NP with [ for every x, x an N]NP
[ some N ]NP with [ for some x, x an N]NP
and similarly for other quantified expressions. If the variable x is already in use, use y
instead; etc.
The other rule we need is more dramatic: it lets us pick the clause over which the operator will
have scope, moves it there, and creates a variable in the location that the moved NP left behind.
(228) Quantifier Raising
Left-adjoin a quantified NP to S, leaving behind a variable in its original location.

This rule has an undefined concept in it, adjunction¸ which is defined as follows:
(229) Left-Adjunction
Given a constituent A, containing a B, and (optionally) C, the mother of A:

Form a new constituent, which is:
➢ has the same node label as A
➢ has as its daughter nodes a copy of B, followed by A
➢ if A was the daughter of C, the new constituent becomes the daughter of C
Here are two simple cases of left adjunction.
left adjoin B (A is not the

to A daughter of any
→ node)
left adjoin B (A is daughter

to A of C)
→
The purpose of left adjunction is simply to provide a slot in which the logical operator can reside,
at a higher level in the tree than the clause over which it takes scope.
17. Deriving distinct meanings with Quantifier Raising
Let us return to (222) Sue shouted for us to give water to each runner, whose surface structure
is repeated below.
First applying Quantifier Translation to each runner, we get the following. A triangle is used
to avoid worrying about the inner details of the quantifiers.
Next, we note that the clue to the multiple meanings is that the sentence has two clauses,
hence two S nodes that the Quantifier Raising can adjoin each runner to. If we pick the lower S,
adjunction will look like this:
Adjoin here
Expression
to be moved
Inserting the new S node, and rearranging the tree in the way required, we get the following:
New copy
(adjunction)
Old copy
moved quantifier variable

inserted by
Quantifier
Raising
Note the variable: it is the logical place marker formerly occupied by each runner, and it is bound
(shown by the shared index x) by the raised operator each runner. This yields a logical form for
one of the meanings, that is, a single act of shouting, telling us to attend to all of the runners. This
is the Narrow-scope reading given above.
If we pick the upper clause, we will end up deriving the Wide-scope reading. The stage of
adjunction will look like this:
Adjoin to this node
Quantifier
to be adjoined
Study Exercise #64

Show the final output of the derivation.
This is the Wide-scope reading.

————————————————————————————————————
18. Logical form in sentences with two quantifiers
Let us now return to the topic of section 14, namely sentences that include two quantifiers.
Equipped with the rules just give, we can now provide the derivations that apply to the syntactic
representation to create the alternative readings at the level of Logical Form.
Suppose we start with a simplified version of our “arrow-target” sentence:
Many arrows hit every target.
This sentence is ambiguous, and could mean either “Many were the arrows that hit every target”;
or “For every target, many arrows hit it.” The syntactic surface structure (as well as deep structure)
would be as shown below; many and every are both Articles syntactically.
We first translate the NP with quantifiers into appropriate operators, with the rule of Quantifier
Translation (227). Following the exact statement of this rule, we insert different variables (here, x
and y) for the different noun phrases.
Although the order in which we perform the operations turns out not to matter here, we can
arbitrarily choose first to left-adjoin many x, x an arrow to the sentence, as follows:
The result has a new S node, copying the original one, and the moved quantifier is the sister of the
original S:
In the next step, we need to apply the same rule of Quantifier Raising again, this time to
every y, y a target, which likewise is a quantified NP. Assuming (again arbitrarily) that it left-
adjoins to the highest available S node, the application would look like this:
Here is the result:
Note that a second variable, y, now appears in the clause. This is the reading we wanted: “For
every target, many were the arrows that hit it”. In this reading, every has scope over many, and this
can be seen directly in the structure of the logical form.
Study Exercise #65

Derive the other reading.
Syntactic structure:
Output of Quantifier Translation. I also show an arrow that indicates the application of
Quantifier Raising to the quantified expression every y, y a target.
The structure that results is given below.
Next, we apply Quantifier Raising to many x, x arrows. This is shown with the arrow below:
The final structure that results is shown below:
As the diagram shows, in this reading Many has scope over every. Hence, the meaning is
something like “Many were the arrows that hit every target.”
————————————————————————————————————
19. Quantifier translation with coindexed pronouns
In this section we will circle back to the discussion in section 13 of pronouns that are
coindexed with quantified NP, and bring them fully into the picture. For this purpose we will
examine the sentence Every cat thinks it deserves prompt feeding. This is ambiguous, but by far
the most likely reading is that cat Fluffy thinks Fluffy deserves prompt feeding, cat Fritzie thinks
Fritzie deserves prompt feeding, cat Oscar thinks Oscar deserves prompt feeding, and so on. We
cover this reading first.
Ordinary application of the rule Regular Pronoun Interpretation (209) would permit it to be
coindexed with the subject of the sentence; this is so because they are not clausemates. Here is the
structure thus derived, with every cat and it coindexed with i.
(230) A sentence with a quantified NP coindexed with a pronoun
Now, when we convert every cat to a quantified expression, there is an extra step that must
take place, for we must deal with the coindexed pronoun it. The key idea is to add a fairly simple
additional principle to the Quantifier Translation rule, shown in italics below.
(231) Quantifier Translation (revised)
Replace
instead; etc. Let pronouns coindexed with NP be replaced by the same variable.
The italicized new provision in the rule essentially transfer the concept of coreference into the
similar-but-distinct concept of co-binding — two variables will end up bound by the same
quantifier. Quantifier Translation applied to sentence (230) will yield (232).
(232) Output of Quantifier Translation as applied to a sentence with a coindexed pronoun
The new variable x can be seen in the former position of the subordinate clause it. For simplicity I
have removed the former indices indicating coreference, since they have served their function. The
new representation now depicts co-binding, not coreference.
The final output will derive by applying Quantifier Raising to the quantifier, like this:
(233) Output of Quantifier Raising as applied to a sentence with a coindexed pronoun

This could be read, “For every cat, it is true of that cat that it thinks that it deserves prompt
feeding.” The single quantifier now binds two identical variables.
To tie up a loose end, recall that the sentence is ambiguous: if at the stage shown in (230) we
had exercised our choice (legal when applying Regular Pronoun Interpretation) to assign a distinct
index to the pronoun it, then the special provision of Quantifier Translation (231) would not be
applicable. The it does not become a variable, and in the final output the quantifier binds just one
variable, the one in the position from which it was raised:
(234) Output of Quantifier Raising as applied to a sentence with a distinctly indexed pronoun
The pronoun it bears the “extraneous” index j, meaning it designates some other entity, already
mentioned in the conversation, that every cat thinks deserves prompt feeding.140
20. Wh-phrases are operators
Chapter 5, section 6 of this text discussed the fact that Wh- questions can differ in the scope of
the Wh- phrase, giving the following example:
[ What song ] can Sue imagine that Bill sang ___?

Sue can imagine [ what song ] Bill sang ___
140
This scenario might seem odd but I could imagine it being the canary.
We can now express this idea more precisely by giving these sentences logical forms similar
to the quantifier sentences already discussed. The key idea is that wh- phrases are logical
operators, which are requests for the listener to fill in the missing information that the variable
stands for. Thus we might have the following two logical forms:
a. Wide scope of what song
b. Narrow scope of what song
You can see that the syntactic transformation of Wh- Movement is a kind of observable, syntactic
analogue of Quantifier Raising, and has the function of placing the wh- phrase where it bears its
logical scope. The landing site for Wh-Movement is different (Comp vs. adjoined to S), but this is
a relatively superficial difference.
In languages where Wh- phrases syntactically remain in situ, things will work differently.
Here, Quantifier Raising must apply to wh- phrases, so that their scope will be correctly expressed
in logical form. Here is an example from Mandarin Chinese, an in-situ language:
(235) A sentence of Mandarin with ambiguous scope
a. IPA transcription
[S tʂáŋsán tsái [S lìsɯ̂ ɕìhwán ʂěi ]

Zhangsan guess Lisi like who
b. Pinyin Romanization with integers for tone number
1 1 1 3 4 3 1 2
Zhangsan cai Li si xi-huan shui
This sentence is ambiguous. It can mean “Who does Zhangsan guess that Lisi likes?” This
meaning involves raising the Wh- phrase to adjoin to the highest S in logical form, a wide-scope
reading. The sentence can also mean “Zhangsan guessed who Lisi likes”, a narrow-scope reading.
This meaning involves raising the Wh- phrase only to the lower S in logical form.141 Here are
derivations demonstrating the two meanings.142
a. Surface syntactic structure
141
Thanks to UCLA graduate students Kristine Yu and Grace Kuo for constructing this example for
me.
142
Mandarin apparently has no complementizers for embedded clauses, so I am omitting the CP node.
b. Quantifier translation of wh-phrase
c. Quantifier Raising to lower S (narrow-scope reading):
d. OR Quantifier Raising to higher S (broad-scope reading):

21. Summary of operators, variables, and scope
• Constructions with operators and variables are among the most intricate of semantic
phenomena. A basic analysis of them is possible using the rules of Quantifier Translation
and Quantifier Raising. These rules apply during the creation of logical form, a
hypothesized grammatical level that explicitly characterizes linguistic aspects of meaning.
• Scope differences can be of various kinds: a single operator can be raised to different
levels (as in Sue shouted for us to give water to each runner), or there can be two operators
that vary in their scope relative to each other (as in At least two arrows hit every target.).
• Pronouns coindexed with quantified NPs often turn into additional variables in logical form
(as in Every cat thinks it deserves prompt feeding)
• The constructions created in logical form by Quantifier Raising are abstract and not directly
observable. Yet they are mimicked by observable constructions in language: Wh- phrases
are a sort of quantifier, which in languages like English really do move to the appropriate
scope location in surface structure. Contrariwise, languages without Wh-movement must
be assumed to have a “hidden” version of Wh-movement, needed to handle ambiguous
sentences and similar in form to Quantifier Raising.
For further reading
Here are two standard textbooks in modern linguistic semantics:
• Chierchia, Gennaro, and Sally McConnell-Ginet. Meaning and grammar: An introduction

to semantics. MIT press, 2000.
• Kratzer, Angelika, and Irene Heim. Semantics in generative grammar. Oxford: Blackwell,
1998.
A lively and readable discussion of Quantifier Raising is given in:
• Hornstein, Norbert (1984) Logic as grammar. Cambridge, MA: MIT Press.

Chapter 10: Phonetics
This point in the text marks a discontinuity in the subject matter. Language is a system
relating sound and meaning, and the previous chapter ended with the relatively subtle aspects of
meaning involved in quantifiers. We turn now to the other extreme: physical sounds, created by
the organs of speech and apprehended by the ear. Later, we will see that phonetics is the primary
working material of phonology, a field that in turn relates to morphology and to syntax, and hence
ultimately to semantics. So language is like an arch, primarily abstract but anchored at either end
to observable aspects of the world. Semantics relates language structure to situations in the real
world; phonetics relates language structure to the physical events in the vocal tract and the
atmosphere on which we depend to communicate.
1. Phonetic description
The first task of a linguist trying to study a new language is to be able to hear its sounds
correctly and take down utterances in accurate and reliable fashion. It’s a familiar experience for
everyone to have heard a foreign language as a babbling stream of sound—it seems to go by very
fast, and is hard to imitate and remember. Often, a language will include crucial but subtle
distinctions between sounds that escape the linguist entirely in the period of initial efforts.
Therefore, it’s a fundamental skill of linguists to be able to listen to other languages with a trained
ear and to take down what is said accurately in a phonetic transcription. Transcription is taught to
beginning linguists all over the world.143
Here is an example of phonetic transcription:
[ðɪs ɪz ə fənɛɾɪk tɹænskrɪpʃən əv ən ɪŋglɪʃ sɛntn̩s pɹənaʊnst ɪn maɪ on daɪəlɛkt]

(This is a phonetic transcription of an English sentence pronounced in my own dialect.)
In fieldwork, the task of transcribing a new language usually begins with slow and modest
steps: listening to one single, perhaps short, word at a time, and only gradually building up to the
point of being able to provide an accurate transcription for any utterance. It also helps to focus on
very short utterances when you are learning to transcribe.
2. The International Phonetic Alphabet
The standard form of phonetic transcription is the International Phonetic Alphabet, a large
symbol set promulgated by a scholarly society called the International Phonetic Association. Both
the alphabet and the association may be abbreviated “IPA”. The IPA is the form of transcription
that will be covered in this text. The Association offers much information, either free or
inexpensive, on its Alphabet and how to use it:
143
At UCLA we offer both Linguistics 102 (applications-centered) and Linguistics 103 (fieldwork-
centered) for this purpose.
• Their printed guide, Handbook of the International Phonetic Association : A Guide to the
Use of the International Phonetic Alphabet (Cambridge University Press, and probably,
your university library)
• The IPA website: https://www.internationalphoneticassociation.org/
• The IPA phonetic chart, which, despite the continuing discovery of new sounds, still fits on
just one page. Below, I’ve split it up for greater legibility.
(236) The IPA chart

It would not be reasonable to teach the entire IPA chart in an introductory linguistics course,
but I’ve included it to show what is needed to cover (most of) the world’s languages.144 Given the
very brief time available, the only language that we will cover will be American English. This is
144
The IPA is revised and improved from time to time, but still needs work. For instance, it still lacks
symbols for the sounds of Korean commonly Romanized as pp, tt, and kk.
actually a rather complex language phonetically, and once you have it down, it makes transcribing
the others easier.
3. Vowel and consonant charts for English
The following are charts, based on the IPA chart, giving just the vowels and consonants of
English. Below each symbol is a keyword of English meant to identify and illustrate the sound.
The row and column labels give the phonetic terminology, to be covered below, that says what
kind of sound each symbol designates.
(237) IPA symbols for the sounds of English
a. Consonants
Bilabial Labio- Dental Alveolar Palato- Palatal Velar Glottal

dental alveolar
Stops voiceless p t k
pin tin kin
voiced b d g
bin din gill
Affricates voiceless tʃ
chin
voiced dʒ
gin
Fricatives voiceless f θ s ʃ h
fin thin sin shin hymn
voiced v ð z ʒ
vim this zip vision
Nasals m n ŋ
mitt nip sing
Approxi- lateral l
mants Lynn
central w ɹ j
win rim ying
b. Vowels and diphthongs
Front Back Diphthongs

Unrounded Unrounded Rounded
Upper high i u aɪ, aʊ, ɔɪ
beat boot bite, bout, Coit
Lower high ɪ ʊ
bit foot
Upper mid e o Rhotacized upper
bait boat mid central
Lower mid ɛ ə unrounded
bet but, abbot ɚ
Bert
Low æ ɑ ɔ
bat father taut
(only
some
dialects)
If you want to hear the keywords pronounced (by me) visit

http://www.linguistics.ucla.edu/people/hayes/103/charts/english/chartsforEnglishbroadtranscription.htm
Here is a quick identification of the symbols. The IPA was created in its first version by a
panel of European linguists in the late 19th century (first publication 1888). Naturally enough, as
much as possible they used as the basis of IPA the typical phonetic values that Roman letters have
in European languages, and you can see this in the charts above. However, some of the symbols
are invented.
(238) Key for English consonants in IPA
• [ʃ] is the “sh” sound, heard in she.

• [ʒ] is the “zh” sound, heard in the middle of vision.
• [tʃ] is the “ch” sound, heard at the beginning and end of church. If you listen carefully,
you’ll hear that the choice of “t” + “ʃ” is a sensible one here.145
• [dʒ] is the “j” sound, heard at the beginning and end of judge. It, too, is
“compositional”, made up of [d] plus [ʒ]
• [ŋ] is the “ng” sound, heard at the end of sing.
• [ɹ] is the “r” sound of English. (We can’t use [r], because in IPA [r] is a tongue-tip
trill.)
• [j] is the “y” sound (the IPA symbol is based on how the “y” sound is spelled in
German, Dutch, Polish, and many other languages)
A few of the symbols that aren’t letters have informal names used by linguists to refer to them
orally. For example [ʃ] is “esh” (or even “snake”); [ʒ] is “ezh”; [ŋ] is “angma” or “eng”. In my
own teaching I refer to [ɹ] (as it appears in English) as “English r”.
For vowels, it is uncommon for the IPA symbol to match the English spelling. Letter i in
English often spells what in IPA is [aɪ] ([baɪt] bite), whereas the IPA sound [i] represents a sound
fairly close to what English often spells as ee (as in [bi] bee). Similarly, IPA [u] is what English
often spells as oo ([mun] moon), whereas letter u is often [ju] in IPA ([mjut] mute). The IPA
founders went essentially by “majority rule”: the Roman vowel letters have mostly pan-European
(now: pan-global) meanings, but for historical reasons English diverges greatly from this
consensus.146 If you know a non-English language that uses the Roman alphabet, you can probably
use it to help you learn IPA. For example, the IPA symbols [i], [e], [a], [o], and [u] have meanings
not far from the same letters as used in Spanish spelling.
145
Though in close detail, you can notice that the [tʃ] in gray chip is not really the same as the [t] + [ʃ]
in great ship. If you want to show that a single sound is meant, you can link up the [t] and the [ʃ] with a
ligature: [t͡ʃ].
146
How did this come to be? The original phonetic values of the Roman letters come from Latin, and most
European languages preserve, roughly, these old phonetic values. English underwent massive phonetic changes in the
decades starting around 1500, which greatly altered the phonetic values of its letters.
In addition, the Roman alphabet has very few vowel letters, so a greater number of novel
symbols needed to be used for the IPA. The ones used for English are given below.
(239) Non-Roman symbols for English vowels in IPA
• [ɪ] is easy for native speakers but can be hard for second language speakers; it is close to [i]
but not as long, and with the tongue pressing less firmly against the roof of the mouth. Try
listening to pairs such as peat, pit ([ˈpit, ˈpɪt]).
• [ʊ] is to [u] as [ɪ] is to [i]. Try listening to pairs such as kook, cook [ˈkuk, ˈkʊk].
• [ə] is used here for the stressed vowel of cut [ˈkət] as well as the very short, indistinct
vowel occurring in stressless syllables (as in abbot [ˈæbət]). Some texts use a separate
symbol, [ʌ], for the stressed vowel, but we will ignore the difference here.
• [ɚ] is much like [ɹ], only it acts as a vowel rather than consonant. Compare furry [ˈfɚi]
with free [ˈfɹi].
In light of the potentially-confusing differences between the letter name of the symbol and its
sound, it is useful to adopt conventions for referring to the symbol. In my own teaching, I identify
vowel symbols orally as follows.
[i] “letter i”
[u] “letter u”
[e] “letter e”
[o] “letter o”
[ɪ] “small cap i”
[ʊ] “small cap u” or “horseshoe”
[ə] “schwa”
[ɚ] “rhotacized147 schwa”
[ɛ] “epsilon”
[ɑ] “script a”
[æ] “a - e digraph”
[ɔ] “backwards c” or “open o”
The remaining vowels of English are diphthongs, which means a vowel that changes during
its time course. IPA transcribes diphthongs by providing two symbols; one for the start, the other
for the end of the vowel. Try pronouncing these diphthongs very slowly, and hearing the starting
or ending points—is your [aɪ] like Spanish [a] plus English [ɪ]?148
147
“Rhotacized” = “rendered r-like”; after Greek rho “r”.
148
If not: the probable cause is that the diphthongs vary greatly across different speaking rates and
styles. [aɪ] is a “medium” pronunciation; “fast” would be [ae], and “slow and careful” would be [ai].
Probably, when you listen carefully to yourself, your speech is slow and careful. Another possibility (rather
unlikely if you are a UCLA student), is that your own dialect doesn’t have an [aɪ], using (for example) the
sound [aː] instead.
Depending on your dialect, you may also notice that your [e] and [o] are actually closer to [eɪ]
and [oʊ], at least in some contexts. This text will use the [e] and [o] symbols, though you may see
diphthong transcriptions in other texts.149
4. Stress
Most of the information in a phonetic transcription will consist of symbols standing for
individual speech sounds. However, in a language like English, it is also important to transcribe
stress, which roughly speaking, is the amount of articulatory effort or loudness found on a
syllable. Stress must be included because you can have different words that are phonetically
distinguished only by their stress pattern, as in the examples given below. These illustrate the IPA
diacritic for stress: [ˈ], placed just before the stressed syllable:
differ [ˈdɪfɚ]
defer [dɪˈfɚ]
permit [pɚˈmɪt] verb: ‘to allow’

permit [ˈpɚmɪt] noun: ‘a kind of document giving permission’150
5. Transcription technique
There are various methods you can use to become a skilled phonetic transcriber.
5.1 The keyword method
A very useful method is the use of keywords, which I will illustrate with an example. Suppose
you have trouble hearing the distinction between [i] and [ɪ], but you are trying to transcribe the
word mitt. The correct transcription happens to be [mɪt]. You already know, having examined chart
(237) above, that the English word beat has [i] and the word bit has [ɪ]. These can serve as
keywords for the [i]/[ɪ] distinction. The dialogue below illustrates the method.
Linguist: “Please say the next word on our list.”

Native speaker: “[mɪt]”
Linguist (can’t hear it, but is coping): “Please say the word spelled b - i - t.”
Native speaker: “[bɪt]”
Linguist: Please say b - e - a - t.
Native speaker: “[bit]”
Linguist: “Now say the word we are working on.”
Native speaker: “[mɪt]”
Linguist: “Now say all three words in a row.”
Native speaker: “[bɪt] … [bit] … [mɪt]”
149
There is variation even among texts that use IPA. The problem is that there can be more than one reasonable
way to deploy the symbols of the IPA to approximate what is said.
150
The example works for the majority of Americans, though there are many who say [pɚˈmt] for
both.
The idea should be plain: it’s easier for your ear to compare a new word to known words than
it is to transcribe “out of the blue”. This holds not just for the more delicate distinctions of English
but for all difficult distinctions, in any language.
When I do English phonetic dictations in class, I sometimes encourage students to raise their
hand and ask me to pronounce keywords. If you want to use keywords in doing a homework, you
can find them (with sound files) at this web address:
http://www.linguistics.ucla.edu/people/hayes/103/charts/english/chartsforEnglishbroadtranscription.htm
5.2 Other general hints
When you’re transcribing a language that you know, and which has a spelling system, it’s
important not to be influenced by the spelling of a word. In my experience teaching English
transcription, this is a major source of errors.
It’s also sometimes a good idea to listen to a word more than once. You can do this
indefinitely with a recording; with a live speaker you have to size up how patient they are.
Lip reading can be very useful, particularly for the difference between [θ] and [f] (thin vs.
fin), and for vowels that have lip rounding. Lower vowels have more jaw opening.
5.3 Hints specific to English
The inflectional suffixes -ed and -s are spelled in a constant way, but are pronounced
differently in different environments. For example, latched = [lætʃt], not *[lætʃd]; blades = [bledz],
not *[bleds].
The letter s is often ambiguous between [s] and [z]: compare goose [gus] with lose [luz].
The sequence ng in spelling can spell either one sound or two (for most dialects). For
example, finger is [ˈfgɚ] (two sounds), but singer is [ˈsɚ] (one sound). In the less-widely-
spoken dialect, spelled ng is [ŋg] in the middle of a word (finger [ˈfgɚ], singer [ˈsgɚ]) and [ŋ]
otherwise. For what we should think about this dialect, see Chapter 3.
It is difficult to hear schwa; often people transcribe a full vowel that corresponds to the
spelling. For example: tenacious = [təˈneʃəs], not *[tɛˈneʃəs]; connection = [kəˈnɛkʃən], not
*[koʊˈnɛkʃən]; childless = [ˈtʃaɪldləs], not *[ˈtʃaɪldlɛs]. Schwas can be spotted because they tend to
be very short and rather “indistinct” in their quality.
As noted above, the letter u often represents a sequence of [j]+[u]: use = [ˈjuz]; fugue =
[ˈfjug]; spectacular = [spɛkˈtækjulɚ] or [spɛkˈtækjəlɚ].
The letter x can represent [ks] (Texas = [ˈtɛksəs]) or [gz] (exact = [əgˈzækt]).
The sequence th can represent either [θ] (ether = [ˈiθɚ]) or [ð] (brother = [ˈbrəðɚ]).
A note on my own teaching practice: where I have provided more than one way of
transcribing the same sound, either way is acceptable. I do not require that you memorize the
symbols; phonetic charts are provided for exams.
Study Exercise #66
Visit the following web page. It has a list of English words. When you click on a word, it will
launch a sound file in .wav format, which (if your Web browser is set up properly), should play on
your computer. (I recommend you use headphones in a quiet place.)
• http://www.linguistics.ucla.edu/people/hayes/20/sounds/English/.
Answers below.

1. [ˈpɪn]
2. [ˈtɪn]
3. [ˈkɪn]
4. [ˈbɪn]
5. [ˈdɪn]
6. [ˈgɪl]
7. [ˈtʃɪn]
8. [ˈdʒɪn]
9. [ˈfɪn]
10. [ˈθɪn]
11. [ˈsɪn]
12. [ˈʃɪn]
13. [ˈhɪm]
14. [ˈvɪm]
15. [ˈðɪs]
16. [ˈzɪp]
17. [ˈvɪʒən]
18. [ˈdeɾə]
19. [ˈmɪt]
20. [ˈnɪp]
21. [ˈsɪŋ] (many speakers say [ˈsiŋ] instead)
22. [ˈlɪn]
23. [ˈwɪn]
24. [ˈɹɪm]
25. [ˈjɪŋ] (many speakers say [ˈjiŋ] instead)
26. [ˈbut]
27. [ˈbʊk]
28. [ˈbot]
29. [ˈbit]
30. [ˈkɑt]
31. [ˈbit]
32. [ˈbut]
33. [ˈkɑt]
34. [ˈbæt]
35. [ˈbɪt]
36. [ˈfʊt]
37. [ˈbet]
38. [ˈæbət]
39. [ˈbot]
40. [ˈbɛt]
41. [ˈbət]
42. [ˈbɔt]
43. [ˈbæt]
44. [ˈfɑðɚ]
45. [ˈbaɪt]
46. [ˈbaʊt]
47. [ˈkɔɪt]
48. [ˈbɚt]
49. [ˈdɪfɚ]
50. [dɪˈfɚ]
51. [pɚˈmɪt]
52. [ˈpɚmɪt] (some speakers have final stress for this word)
53. [ˈpɑp]
54. [ˈtɑt]
55. [ˈkɪk]
56. [ˈbɑb]
57. [ˈdæd]
58. [ˈgæg]
59. [ˈfaɪf]
60. [ˈθɪn]
61. [ˈsɪs]
62. [ˈʃu]
63. [ˈhi]
64. [ˈvæt]
65. [ˈðau]
66. [ˈzu]
67. [ˈeʒə]
68. [ˈtʃɚtʃ]
69. [ˈdʒədʒ]
70. [ˈmɑm]
71. [ˈnən]
72. [ˈjəŋ]
73. [ˈjuθ]
74. [ˈwɪtʃ]
75. [ˈtɔɪ]
76. [ˈdʒunəpɚ]
77. [ˈwɪʃ]
78. [ˈpɚʒən]
79. [ˈθætʃɚ]
80. [ˈjɑt]
81. [ˈkwɛstʃən]
82. [ˈtɛnθ]
83. [ˈəðɚ]
84. [ˈʃælo]
85. [ˈbɛltʃ]
86. [ˈmjuzɪk]
87. [ˈlætʃt]
88. [ˈbledz]
89. [ˈfɪŋgɚ]
90. [ˈsɪŋɚ] (some speakers have [ˈsɪŋgɚ])
91. [ˈjuz]
92. [ˈfjug]
93. [spɛkˈtækjulɚ]
94. [spɛkˈtækjəlɚ]
95. [ˈtɛksəs]
96. [ɪgˈzækt]
97. [ˈiθɚ]
98. [ˈbɹəðɚ]
99. [ˈfit]
100. [ˈfɪt]
101. [ˈfɪət]
102. [ˈluk]
103. [ˈlʊk]
104. [ˈlʊək]
105. [ˈðaɪ]
106. [ˈθaɪ]
107. [ˈɹəɪɾɚ]
108. [ˈɹaɪɾɚ]
109. [ˈkɔt]
110. [ˈkɑt]
111. [ˈʔəʔo]
112. (misnumbered, no word here)
113. (misnumbered, no word here)
114. [ˈɹɔɹ]
115. [ˈdeɾə]
116. [ˈθɪn]
117. [ˈðɛn] (some speakers have [ðɪn])
118. [ˈʃu]
119. [ˈvɪʃən]
120. [ˈtʃɚtʃ]
121. [ˈdʒədʒ]
122. [ˈjɪɹ] (some speakers have [ˈjiɹ])
123. [ˈlɔ] (some speakers have [ˈlɑ] for this word)
124. [ˈlɑ]
125. [ˈkɔt] (some speakers have [ˈkɑt] for this word]
126. [ˈkɑt]
127. [ˈpɔli] (some speakers have [ˈpɑli] for this word)
128. [ˈpɑli]
129. [ˈbɔɪ]
130. [təˈmeɾo]
131. [əˈmɛɹəkə] (some speakers have [əˈmɛɹɪkə] for this word)
132. [kəˈnɛɾəkət]
133. [ˈbɚd]
134. [ˈɹaɪd]
135. [ˈbɔɪ]
136. [ˈhaʊ]
137. [ˈtɪkəl]
138. [ˈbətn̩]
6. Other places to practice
If you want to get practice in learning the symbols, you might try reading passages of
transcription; I have posted a couple of them at
• http://www.linguistics.ucla.edu/people/hayes/103/PracticeReadingTranscription.pdf
• http://www.linguistics.ucla.edu/people/hayes/103/PracticeReadingTranscriptionII.pdf
Some further practice can be obtained from an exercise I’ve posted for another course:
• http://www.linguistics.ucla.edu/people/hayes/103/EnglishTranscriptionPractice/
7. Some toughies from English
The hardest factor in phonetic transcription is that we tend to hear best the phonetic
distinctions of languages we speak. In fact, it’s typically the distinctions heard in infancy and
toddlerhood that are the most noticeable — experiments in infant labs have shown that the neural
circuitry for vowel detection, for example, is already being “tuned” to the ambient language by the
age of six months.
Thus, if there are English distinctions that you didn’t acquire early on, you may find them
tough. I only apologize a little bit for this: linguistics training necessarily involves practice in
hearing such distinctions, even if it’s hard. In my own teaching I usually include some “exotic”
cases from American dialects (which I hope will be equally hard for everybody!).
Here are cases of distinctions that may be difficult. They are posted at the same Web page
mentioned above.
feet [fit]
fit [ft] Clues: [] shorter than [i]. Spoken slowly, [] becomes [ə].
Luke [luk]
look [lk] Clues: [] shorter than [u]. Spoken slowly, [] becomes [ə].
thy [a]
die [da] Clue: sit up close and lip-read. [] when pronounced carefully
usually has some tongue protrusion.
writer [ə]
rider [aə] Clue: [a] has more jaw lowering.
caught [kɔt]
cot [kt] Clue: [] has a fish-like lip-rounding gesture.
THE PRODUCTION OF SOUNDS IN THE VOCAL TRACT

The human vocal tract can produce thousands of audibly distinct sounds. Of these, only a
subset are actually used in human languages. Of this subset, some sounds are much more common
than others. For example, almost every language has a [t]-like sound, while very few languages
have a labial trill. Any one language uses only a fairly small inventory of distinct speech sounds,
usually just a few dozen.
8. Vocal tract anatomy
To understand how sounds are made, one needs to have an idea of the location and shape of
the articulatory organs. Here is a diagram; a so-called “mid-sagittal”151 section:
(240) Diagram: Midsaggital section of the vocal tract
hard palate
nasal cavity
alveolar ridge
upper lip oral cavity
velum (soft palate)

teeth
velar port
uvula
tongue body
lower lip (dorsum)
tongue tip
tongue blade
pharynx
jaw
larynx
trachea
The above is a schematic diagram; the hypothetical speaker is saying something like []
(nasalized “uh”).
The information for images has traditionally been obtained by dissection of cadavers, or later,
from X-rays. More recently, magnetic resonance imaging makes possible the safe examination of
living subjects, with images like the following:
151
This word apparently comes from “sagittal suture”, the arrow-shaped join between bones forming the skull,
running through the middle of the head from front to back. “Section” means “diagram from cutting.”
(241) MRI images of the vocal tract (midsaggital section)
American English [l]
American English []
A evidently nasalized [], showing more of the head and

neck152
8.1 The parts of the vocal tract
The three major regions of the vocal tract are the nasal cavity, the oral cavity (less
pretentiously, the mouth), and the pharynx, which is located behind the tongue but above the
larynx.
The most crucial organ of speech is the tongue. Bear in mind that just looking in a mirror
gives you a poor idea of the shape of the human tongue, because you can only see the tongue’s
forward extension. In reality, the tongue is more of a lump; when at rest it is fairly round in shape
152
First two images: from www.linguistics.ubc.ca/isrl/Gick_Whalen_Kang(SPS5); research from
Haskins Laboratories, New Haven, CT. Last image: http://web.mit.edu/albright/www/; the Web page image
of Prof. Adam Albright, Massachusetts Institute of Technology.
except for visible flange up front. The round main section is quite mobile and flexible, and can
move in all directions. Terminology for parts of the tongue are as follows: the tip (or apex), the
blade (= the forward flange), and the body (the main rounded part).
We will now cover the roof of the mouth, going from front to back. The lips and teeth need
no comment other than that they are both important for speech. The next important landmark,
going backward, is the alveolar ridge. Most people can feel this ridge by placing the tongue a little
further back in the mouth than the upper inside edge of the front teeth. The alveolar ridge forms a
useful descriptive “boundary line” on the upper surface of the mouth.
The expanse behind the alveolar ridge is called the palate. The palate is divided into a hard,
bony section in front called the hard palate and a soft fleshy section in back called the soft palate
or velum (Latin for “sail”). The velum is mobile. If you know how to produce nasalized vowels (as
in French or Portuguese), you can see it moving by looking in a mirror, placing your tongue as low
as possible, and alternating between saying nasalized and normal vowels. The main function of the
velum in speech is to control nasality. Most often, the velum is raised up to block of the nasal
passage. When it is lowered, air may pass out the nose and we get a nasal sound.
The little hanging object at the tip of the velum, made famous by screaming cartoon
characters, is called the uvula. It is used in consonant production in many languages (for example,
French, Persian, and Arabic), but not in English.
The pharynx is the space behind the tongue, invisible to us unless we use a mirror. This space
can be made smaller by retracting the tongue body down into it.
8.2 The larynx
At the bottom of the pharynx is the larynx, or voice box. This is a highly complex structure of
cartilage, muscle, and ligaments. The crucial elements of the larynx are the vocal cords.153 These
are not really cords, but flaps that come in from both sides. The vocal cords can close off the flow
of air to varying degrees. The gap between the vocal cords is called the glottis.
(242) The larynx
Larynx with vocal cords in position to

vibrate:
glottis (narrow slit

suitable for vibration)
vocal cords
153
Not: chords.
Open larynx (vocal cords spread):
glottis (wide open)
vocal cords
There are basically four things that the vocal cords can do. (1) If they are spread far apart, we
get normal breathing. (2) If they are brought tightly together, the airflow is blocked. If the
blockage is then quickly released, we get what is called a glottal stop, IPA symbol []. This is the
sound that begins each syllable of the expression “uh-oh” [ˈo]. (3) If the vocal cords are
brought close but not touching, we get an [h]. (4) If the vocal cords are just barely touching, they
vibrate, producing what is called voicing. Voicing accompanies most vowels and many consonants
(except when we whisper), and is the most important source of sound in speech.
Numerous speech organs are actively controlled by the speaker in the production of speech. In
normal speech, the following organs are active: the lips, the tongue blade, the tongue body, the
velum, the jaw, the larynx (up and down), and the vocal cords. Video imaging of speech shows
that these speech organs move extremely rapidly and with great precision.154 Speaking is one of
the most complex physical feats people can perform, yet we do it without even thinking about it.
A striking aspect of the speech stream, in all its physical complexity, is that the linguistic
system underlying it can be usefully viewed simply as a sequence of sounds, all taken from the
inventory of a few dozen sounds (phonemes) specific to a a language. In speaking, this sequence is
extensively distorted, attenuated, and smoothed out — yet it is the underlying sound sequence that
is the most important for linguistic study. Hence, phonetic description begins one sound at a time,
treating each sound as a static idealization. This is what we will now do for English.
9. Describing consonant articulation
The description of a consonant normally includes three things:
Place of articulation. All consonants involve a constriction somewhere in the vocal tract. To
specify a consonant one must state where this constriction is made; this is the place of articulation.
Manner of articulation. This indicates the kind of constriction that is made—roughly, how
narrow it is, and the acoustic result.
Voicing— whether the vocal cords are vibrating during the production of a consonant. A good
way to detect voicing is to put your hand firmly on top of your head when you say a word. If you
154
Videos are accessible on line; I suggest a search on “mri video vocal tract”.
do this while you say “za”, you will feel buzzing all the way through. If you do this for “sa”, you
will feel buzzing only after the [s] is over. So [z] is a voiced sound and [s] is voiceless.
9.1 Manner of articulation
We will cover six manners of articulation.
(a) In a stop, the airflow is momentarily blocked off completely (i.e. “stopped”), then
released. The stops of English are
[p] [t] [k]

(Keywords:) pop tot kick
[b] [d] [g]

Bob Dad gag
Note that I have arranged the six stops in rows and columns, going by place of articulation and
voicing.
(b) In a fricative, one forms a narrow constriction at the place of articulation. The air passing
through the constriction makes a hissing noise. English has nine fricatives:
[f] [] [s] [ʃ] [h]

fife thin Sis shoe he
[v] [ð] [z] [ʒ]

vat thou zoo Asia
(c) An affricate is a rapid sequence of a stop and a fricative made at roughly the same place of
articulation with a single gesture. Affricates can usually be considered a subclass of the stops.
English has two affricates, [tʃ] (as in church) and [dʒ] (as in judge).
(d) In a nasal consonant, the velum is lowered, allowing air to escape out the nose. The great
majority of nasals have a complete blockage within the mouth at the same time. The places of
articulation for nasals are usually the same as those for stops. The nasal consonants of English are
[m] (Mom), [n] (none), and [ŋ], which is the last sound of young.
(e) In an approximant, the vocal tract is relatively open, so that air flows freely and there is
no frication noise. Approximants are normally divided into lateral and central. In a lateral
approximant, the air flows around the sides of the tongue; [l] is a lateral. In a central approximant,
air flows through a central channel. English has three central approximants:
• [j] occurs in words like youth

• [w] occurs in words like witch155

• [ɹ] occurs in words like roar
Approximants are often divided up in a different way: liquids are the “r” and “l” sounds; in
English [l] and []. Glides (also called “semivowels”) are central approximants like [j] and [w] that
are closely similar to vowels (see below).
(f) In a tap, the tongue tip brushes very briefly against the roof of the mouth—too short a
closure to count as a stop. The tap of English is found in words like data (North American
dialects), and is symbolized []. The tap is generally voiced.
9.2 Place of Articulation
By combining information about place of articulation with information about voicing and
manner, we can arrive at complete descriptions of English consonants. I will cover the places of
articulation going from the front to the back of the mouth. In reading the following refer to the
midsaggital section diagram in (240) above (p. 394).
(a) Bilabial sounds are made by touching the upper and lower lips together. English has a
voiceless bilabial stop [p], a voiced bilabial stop [b], and a (voiced) bilabial nasal [m].
Note the standard form for describing a consonant: the formula is VOICING-PLACE-MANNER. In
the case of nasals and approximants, which are almost always voiced, it is usual to specify only
place and manner.
(b) Labio-dental sounds are made by touching the lower lip to the upper teeth. English has a
voiceless labio-dental fricative, [f], and a voiced one, [v]. Labio-dental stops and nasals are rare,
though English speakers make them if they try to say [p], [b], or [m] while smiling.
(c) Dental sounds are made by touching the tongue to the upper teeth. This can be done in a
number of ways. If the tongue is stuck out beyond the teeth, the sound is called an interdental,
though we will not worry about such fine distinctions. English has a voiceless dental fricative [θ]
(as in thin), and a voiced one [ð] (as in then).
(d) Alveolar sounds are made by touching the tip or blade of the tongue to a location just
forward of the alveolar ridge. English has several alveolar consonants. There is a voiceless alveolar
stop [t], a voiced alveolar stop [d], voiceless and voiced alveolar fricatives [s] and [z], an alveolar
nasal [n], and an alveolar lateral liquid [l].
(e) Palato-alveolar sounds are made by touching the blade of the tongue to a location just
behind the alveolar ridge. English has a voiceless palato-alveolar fricative [ʃ] (as in shoe), a voiced
155
A small number of American English speakers have an additional central approximant, [ʍ], which
is a voiceless version of [w]. It occurs in words spelled with wh, like which.
palato-alveolar fricative [] (as in vision), a voiceless palato-alveolar affricate [tʃ], (as in church),
and voiced palato-alveolar affricate [dʒ] (as in judge).
(f) Palatal sounds are made by moving the body of the tongue forward toward the hard palate.
English has just one palatal sound, the palatal glide [j], as in year.
(f) Velar sounds are made by touching the body of the tongue to the velum. English has three
velar sounds: a voiceless velar stop [k] (as in cat or king), a voiced velar stop [g] (as in goat), and a
velar nasal [ŋ] (as in sing).
(g) Glottal sounds are made by moving the vocal cords close to one another. English has a
voiceless glottal fricative [h].
9.3 Interpreting the consonant chart
The consonant chart for English, given above in (237) (p. 383) can now be better understood,
as it arranges the consonants of English place, manner, and voicing. The arrangement of the chart
is traditional: the columns depict place, going from front to back in the vocal tract, and the rows
depict manner, going from top to bottom by increasing sonority (loudness). Take a look at the
chart again, examining its rows and columns, to see how these group together sounds of similar
phonetic properties.
Study Exercise #67: Pronouncing the “gaps”
Look at this copy of the English consonant chart. It has eight digits in it, located in cells that
are blank for English but not for languages in general. Try to pronounce each of the gaps. The
answer key identifies the gaps in IPA and gives languages that have them.
Bilabial Labio- Dental Alveolar Palato- Palatal Velar Glottal

dental alveolar
Stops voiceless p 1 t k
pin tin kin
voiced b 2 d g
bin din gill
Affricates voiceless 3 tʃ
chin
voiced 4 dʒ
gin
Fricatives voiceless 5 f θ s ʃ 7 h
fin thin sin shin hymn
voiced 6 v ð z ʒ 8
vim this zip vision
Nasals m n ŋ
mitt nip sing
Approxi- lateral l
mants Lynn
central w ɹ j
win rim ying
Sound files of the author pronouncing these sounds can be listened to at

www.linguistics.ucla.edu/people/hayes/103/Charts/CChartWithSounds/CChart.pdf.
1. This is a voiceless dental stop, made by putting the tongue tip in the location for [θ] and
“squeezing” the constriction enough to yield a stop articulation. The IPA symbol uses a diacritic, a
little subscript platform for the t: [t̪ ]. [t̪ ] is the normal pronunciation in Spanish and French for the
sounds that are spelled with letter t in these languages.
2. This is a voiced dental stop; as [t̪ ] is to [t], so is [d̪] to [d].
3 and 4: The alveolar voiceless and voiced affricates can be pronounced by putting together a [t]
and an [s] (for voiceless) and a [d] and a [z] (for voiced) and saying them rapidly as a single sound:
[ts], [dz] (or, if precision is needed, [t͡s] and [d͡z]). [ts] is the sound spelled with letter z in German.
[dz] occurs for some English speakers as a variant (allophone; see below) of the basic [z] sound
after [n], in words like lens (IPA [lɛnd͡z]).
5 and 6: If you make a fricative instead of a stop at the bilabial location, you will get [ɸ]
(voiceless) and [β] (voiced). Both occur in Ewe (Ghana); listen at to real versions at
www.phonetics.ucla.edu/course/chapter7/ewe/ewe.html.
7 and 8: If you make a fricative instead of a stop at the velar location, you will get [x] (voiceless)
and [ɣ] (voiced). [x] occurs in German, where is spelled ch, as in Bach [bax]; and in Spanish,
where it is spelled j, as in baja [baxa] ‘fall’. [ɣ] occurs in Spanish as the g sound when between
vowels, as in lago [ˈlaɣo] ‘lake’.
___________________________________________________________________________
10. The tap
In the chapters that follow we will need to deal with a peculiar sound of North American
English, called the tap. I list it separately here because it is not a basic sound of the language —
once we’ve gotten some phonology in place, it will emerge that tap is always derived from other
sounds by phonological rules.
Tap is transcribed in IPA with the symbol [ɾ], and is found in words like butter [ˈbəɾɚ], city
[ˈsɪɾi], and battle [ˈbæɾəl].156 In these words, [ɾ] is essentially a “type of t”, as reflected in the
spelling; and indeed in the next chapter, once we can write phonological rules, we will actually
derive [ɾ] from [t]. Tap is also found in words like bidder [ˈbɪɾɚ], paddy [ˈpæɾi], and puddle
[ˈpəɾəl], where it is phonologically derived from [d].
156
As they are pronounced by most North Americans. At least until recently, the tap has been one of the really
distinctive traits that sets off North American dialects from other forms of English, which tend to use [t] (or some other
sound) in the same set of words.
Like [d], tap is a voiced sound, but is different from a [d] in that it is very short: the tongue tip
touches the alveolar ridge only for a brief instant, thus gently “tapping” the roof of the mouth.
In fact, English tap is actually fairly similar to the “r sound” employed in Spanish, Italian, and
many other languages. While tap is not a contrastive speech sound in North American English, it is
worth including here because some of the most noticeable phonology of this dialect involves taps.
11. Describing vowels
Vowels differ from consonants in that they do not have real “places of articulation”, that is to
say, points of major constriction in the vocal tract. Rather, the vocal tract as a whole acts as a
resonating chamber. By modifying the shape of this chamber using movements of the tongue, jaw,
and lips, one imparts different sound qualities to the basic noise made by the vocal cords.
An analogy can be made with brass instruments. The vocal cords by themselves make a rather
ugly buzz, just like the mouthpiece of a trumpet does when played by itself. The buzz is given its
more pleasant characteristic quality by being passed through a resonating chamber (for example, a
trumpet or a vocal tract). The quality of the sound is determined by the shape of the chamber; thus
vowels of English are similar to notes played by the same trumpet with different mutes placed
inside.
There are three basic modifications that one can make to the shape of the vocal tract. Vowels
are described by specifying the amount of each modification used.
11.1 Rounding
One way to modify the shape of the vocal tract is to round the lips, narrowing the passage at
the exit. This happens, for example, in the vowels of boot [u], book [ʊ], and boat [o]. These are
called rounded vowels. Other vowels, such as the [i] of beet or the [] of cot, are called unrounded.
11.2 Height
Another modification one can make to the shape of the vocal tract is to make passage through
the mouth wider or narrower. Widening is accomplished by opening the jaw and/or lowering the
body of the tongue towards the bottom of the mouth. Narrowing is accomplished by raising the jaw
and raising the body of the tongue.
The terminology for describing these changes is based on the height of the tongue body
(without regard to whether this is due to jaw movement or tongue movement). Vowels are
classified as high, mid, or low, depending on tongue body position. In effect, high vowels have a
narrow passage for the air to pass through, and low vowels have a wide passage.
Examples of high vowels in English are [i], the vowel of beat, and [u], the vowel of boot.
Example of low vowels are [ɑ], the vowel of cot, and [æ], the vowel of bat. You can feel the oral
passage widening and narrowing if you pronounce a sequence of vowels that alternates between
high and low, such as [i æ i æ i æ i æ].
11.3 Backness
The third primary way of changing the vocal tract shape is to place the body of the tongue
towards the front part of the mouth or towards the back. Vowels so made are called front and back
vowels.157 For example, [i] (beat) is a high front vowel, and [u] (boot) is a high back vowel (which
is also rounded). You can feel the tongue moving forwards and backwards if you pronounce the
sequence [i u i u i u i u].
11.4 Vowel chart
We now have three “dimensions” for classifying vowels, each based on a particular
modification of the vocal tract shape: rounding, height, and backness. The three dimensions allow
us to describe vowels clearly, and also to organize them in a chart. The chart below is simply a
repetition of (237)b, repeated here for convenience.
Front Back Diphthongs

Upper tense i u aɪ, aʊ, ɔɪ
beat boot bite, bout, Coit
Lower ɪ ʊ
nontense bit foot
Upper tense e ə o Rhotacized upper
bait abbot boat mid back
Lower ɛ  ɔ unrounded:
nontense bet but bought ɚ
Bert
Low æ ɑ ɔ
bat father taut
(dialectal)
Note that this chart is an abstraction, since in physical reality the vowels do not line up vertically in
tongue body position. In particular, the high front vowels are considerably more forward than the
high back vowels, owing to the space available for tongue movement. Because of this, the chart
should be interpreted as saying “relatively more front” or “relatively more high” rather than
specifying actual physical tongue positions.
Vowels are usually identified with formula HEIGHT-BACKNESS-ROUNDNESS. For example, [u]
is an “upper high back rounded vowel.”
11.5 Dialect variation
English dialects differ most noticeably in their vowel systems. Here are differences you may
find in your speech:
157
A more refined classification recognizes central vowels; neither front nor back. Here we will
employ just two degrees of backness.
(1) I included the lower mid back rounded vowel [ɔ] on the chart, but probably about half of
Americans don’t have this vowel in their speech—there is an ongoing change in American English
that is gradually wiping out this vowel. Speakers of the newer, [ɔ]-less dialect use [] in the words
that speakers of the older dialect say with [ɔ]; thus:
Old dialect New dialect158

law [ˈlɔ] la [l] law, la [ˈl]
caught [ˈkɔt] cot [ˈkt] caught, cot [ˈkt]
Pauley [ˈpɔli] Polly [ˈpli] Pauley, Polly [ˈpli]
Speakers who don’t have an [ɔ] as a separate sound do usually have it as part of diphthong, as in
[ɔ] boy.
(2) Many Americans have a high central rounded vowel, IPA [ʉ], instead of [u].
11.6 Schwa ([ə]): a reduced vowel
English has a so-called “reduced vowel”, which appears in the underlined position in the
following words:
tomato [təˈmeo]
America [əˈmeəkə]
Connecticut [kəˈnɛəkət]
This vowel varies in its quality and is quite short, so it is hard to transcribe. We will simplify
things by always transcribing the reduced vowel as [ə] (the vowel called “schwa”). In transcribing,
if you hear a very short, indistinct “blurry” vowel, I suggest you transcribe it as schwa.
11.7 The rhotacized vowel [ɚ]
[ɚ], the vowel of bird, is rather like the schwa, except that the tongue blade is curved upward
in the manner of an [] (see images above). This upward curvature is called rhotacization; thus [ɚ]
is classified as a rhotacized upper mid central unrounded vowel. Its name “rhotacized schwa” fits
its visual form.
11.8 Diphthongs
As noted above, a diphthong is a vowel (that is, a single sound) during which the articulators
are in motion. A common way to represent diphthongs in IPA is to give a sequence of vowel
symbols, one representing the starting point and the other the ending point. English has numerous
158
These labels should be interpreted with caution: some speakers of the “old” dialect are three years
old, some speakers of the “new” dialect are 100. Language change happens fairly slowly.
diphthongs. The three most obvious ones are [aɪ], which appears in ride; [ɔɪ], which appears in
boy; and [aʊ], which appears in how. The diphthong [aʊ] is pronounced [æʊ] by many speakers.
11.9 Syllabic consonants
English also has what are called “syllabic consonants”. These are sounds that are articulated
like consonants, but form the nucleus of a syllable as if they were vowels. Syllabic consonants are
transcribed by putting a [ ̩ ] underneath the symbol for the appropriate consonant. The following
transcriptions illustrate this:
tickle [ˈtkl̩ ]
button [ˈbtn̩]
It is possible to think of the rhotacized schwa, [ɚ], in a different way: as [ɹ̩ ]. These quite different
symbols depict essentially the same sound from different points of view. [ɹ̩ ] is the consonant [ɹ],
rendered syllabic, whereas [ɚ] is a schwa vowel [ə] with added tongue retroflexion.
FEATURES
12. Features and the link to phonology
We will shortly shift from phonetics to phonology, which studies the legal arrangements of
speech sounds (phonological grammaticality) as well as the changes of sounds as they appear in
context. Phonology involves writing rules and constraints, and to do this, it will be essential to
have a system of features, rather like we used for inflectional morphology in Chapter 2.
The features of phonology are based on phonetics. For example, in informal terms, the
features of [d] are that it is a stop, that it is alveolar, that it is voiceless, and further, that it is not
round, not nasal and so on. When we have specified enough features for a particular sound, then
we have complete and explicit description, properly distinguishing it from any other sound. This
will make it possible to write explicit rules and constraints.
In phonology, features are generally given a more compact notation than what we used for
morphological features: a plus sign, placed before the feature name, means that a segment has the
relevant property; minus means that it lacks it. Thus [i] is said to be [+high, +tense, –round,
–nasal, –back]. If you wanted to, you could read this as [High:Plus, Tense:Plus, Round:Minus,
Nasal:Minus, Back:Minus], just like we did for inflection, but in practice no one actually expresses
phonological features in this way.
As with inflectional features, phonological features normally have brackets placed around the
feature names. But more often than not, the features are arranged in a column rather than a list:
+high
+tense 
–round
–nasal 
–back 
Such an expression is often called a feature matrix (plural: matrices). The matrices are more
easily strung together in sequence, facilitating the expression of rules and of morphemes and
words.
13. A feature set
In this text we will use the following features for phonology; each should be assumed to take
the values plus or minus:
(243) A phonological feature set159
[syllabic] [high] [stop] [bilabial]

[voiced] [low] [affricate] [labiodental]
[nasal] [back] [fricative] [dental]
[round] [liquid] [alveolar]
[tense] [glide] [palato-alveolar]
[stressed] [palatal]
[aspirated] [velar]
[lateral] [glottal]
Here is some discussion of the individual features:
[syllabic], roughly, distinguishes vowels from consonants; the vowels are [+syllabic], the
consonants [−syllabic]. In the unusual case of syllabic consonants (like [l̩ ] and [n̩]; see §11.9
above), the value is taken to be [+syllabic].
[tense] is a feature of vowels, corresponding to what is sometimes taught in school as “long”.

The vowels [ɪ], [ɛ], [æ] [], [] are [−tense]; the others are [+tense]. The lower high, lower mid,
and low vowels are [−tense], the others [+tense]. The distinction may seem arbitrary, but is useful
for phonology — consider, for instance, that it is precisely the [−tense] vowels that may not occur
before another vowel, or at the end of a word.160
To distinguish the three basic vowel height categories (high, mid, and low), we only need two
features, not three: high vowels are [+high, –low]; low vowels are [−high, +low]; and mid vowels
159
Please note that this set of features is something of a makeshift, intended to be teachable in a short period of
time. If you take a course in phonology you will probably learn a more effective, and also more standard feature
system.
160
The actual phonetic definition of tense is subtle and non-unitary. Roughly, a tense vowel relative to a lax
one will be slightly longer, will be articulated more peripherally in the mouth, and will often be slightly diphthongized,
becoming higher throughout its duration.
are [−high, –low]. (A vowel that was [+high, +low] would be a articulatory impossibility; you
can’t put the tongue in both high and low positions at once.)
Other than the above, the features are simply restatements of the traditional phonetic
terminology already covered above.
In my own teaching I ask that students understand the meaning of the features but not
memorize them; exams include feature charts where needed. On the other hand, the more you have
the features in your head, the easier it becomes to navigate phonology.
14. Feature charts
Here are the features employed in this book, with the sounds of English defined according to
the features.
Notes:
• [aspirated]: for stops, this means “accompanied by a little puff of breath on release.” In
English, initial [ptk] are aspirated: pin [pɪn], tin [tɪn], kin [kɪn].
• [stressed]: This is treated as a feature of vowels; vowels can be either stressed or stressless.
The value is not given in the chart, but (for example) when you see stressless [i] (as in
[ˈhæpi] happy) you should assume [−stress] and when you see stressed [i] (as in [ˈdivə]
diva) you should assume [+stress]. This use of [stressed] as a vowel feature is in conflict
with IPA transcription practice, which is just something we have to cope with; it is needed
for phonology.
• When a blank appears in the chart, it means that the feature is not essential to the definition
of the sound. For example, there are no values under [p] for [high], [low], [back], [round],
or [tense]. The actual position of the tongue and lips for [p] will vary depending on the
context. This practice in feature theory is sometimes called “underspecification”.
• The major diphthongs [a a ɔ] would be treated as two-vowel sequences, so they don’t
appear in the chart.
(244) Feature charts for English sounds
a. Consonants
Manner of Features mostly for Place of articulation features
Laryngeal [aspirated]
features
articulation features vowels
[syllabic]
[stop]
[affricate]
[fricative]
[liquid]
[glide]
[voiced]
[nasal]
[high]
[low]
[back]
[round]
[tense]
[stressed]
[bilabial]
[labiodental]
[dental]
[alveolar]
[palato-alveolar]
[palatal]
[velar]
[glottal]
[lateral]

p − + − − − − − − − + − − − − − − − −
t − + − − − − − − − − − − + − − − − −
k − + − − − − − − − − − − − − − + − −
b − + − − − − + − − + − − − − − − − −
d − + − − − − + − − − − − + − − − − −
g − + − − − − + − − − − − − − − + − −
tʃ − − + − − − − − − − − − − + − − − −
dʒ − − + − − − + − − − − − − + − − − −
f − − − + − − − − − − + − − − − − − −
θ − − − + − − − − − − − + − − − − − −
s − − − + − − − − − − − − + − − − − −
ʃ − − − + − − − − − − − − − + − − − −
h − − − + − − − − − − − − − − − − + −
v − − − + − − + − − − + − − − − − − −
ð − − − + − − + − − − − + − − − − − −
z − − − + − − + − − − − − + − − − − −
ʒ − − − + − − + − − − − − − + − − − −
m − − − − − − + + − + − − − − − − − −
n − − − − − − + + − − − − + − − − − −
ŋ − − − − − − + + − − − − − − − + − −
l − − − − + − + − − − − − + − − − − +
ɹ − − − − + − + − − − − − − + − − − −
j − − − − − + + − − − − − − − + − − −
w − − − − − + + − + − − − − − − − − −
b. Vowels
Manner of Features mostly for Place of articulation features
Laryngeal [aspirated]
features
[syllabic] articulation features vowels
[stop]
[affricate]
[fricative]
[liquid]
[glide]
[voiced]
[nasal]
[high]
[low]
[back]
[round]
[tense]
[stressed]
[bilabial]
[labiodental]
[dental]
[alveolar]
[palato−alveolar]
[palatal]
[velar]
[glottal]
[lateral]

i + − − − − − + − + − − − + − − − − − − − − −
ɪ + − − − − − + − + − − − − − − − − − − − − −
e + − − − − − + − − − − − + − − − − − − − − −
ɛ + − − − − − + − − − − − − − − − − − − − − −
æ + − − − − − + − − + − − − − − − − − − − −
u + − − − − − + − + − + + + − − − − − − − − −
ʊ + − − − − − + − + − + + − − − − − − − − − −
o + − − − − − + − − − + + + − − − − − − − − −
ɔ + − − − − − + − − + + + − − − − − − − − −
ə + − − − − − + − − − + − − − − − − − − − − −
ɚ + − − − − − + − − − + − + − − − − + − − − −
ɑ + − − − − − + − − + + − − − − − − − − − −
14.1 Description of sounds using their features
As already noted, the features allow us to describe a segment phonetically using the notation
of a feature matrix. For example, a totally- feature matrix for the vowel [i] would be as in (245):
(245) A fully-specified feature matrix for [i]

+syllabic
 −stop 
 −affricate
−fricative

 −liquid 
 −glide
+voiced 
 −aspirated
−nasal

 +high 
 −low
−back 
 
E
−round
 
+tense
−bilabial
 −labiodental
−dental

 −alveolar 
 −palato-alveolar
−palatal 
 −velar 
 −glottal
−lateral

I obtained this, in essence, by taking the row of chart (243) for [i] and turning it on its side.
The matrix in (245) is distressingly long, and it is seldom that one will ever see a matrix this
big in discussions of phonology. Nevertheless, it might appear, for instance, in the files created by
a computer program meant to apply phonological rules.
In actual practice, we will work mainly with partially-specified feature matrices. Suppose in
particular that we want to consider [i] specifically as a sound of English. Here, it would make
sense to give only the features necessary to distinguish it from all the other English sounds. In
these terms, [i] is specified as in (246):
(246) An English-specific, partly-specified feature matrix for [i]
–back 
+high 
  E
+tense
You can establish the content of partly-specified feature matrix by examining the full matrix
(obtainable from the rows of (244)) and taking away features one by one where they are not
needed to distinguish the sound from any other sound in the same language. In the above example,
[−low] is not needed, since no high vowel can be low. [−round] is not needed, since English no
vowel can be both front and rounded vowels. [−nasal] is not needed, since nasality is not a
distinctive property of English vowels.
Study Exercise #68
Using the features above, describe the sounds [tʃ] and [n] in the same way that [i] was
described, that is, enough to distinguish them from other sounds of English.
I believe the shortest possible descriptions are as follows.
–voiced 
[tʃ]: +affricate E
[−voiced] is needed to distinguish [t] from [d]. [+affricate] is obviously needed since there
are many other voiced sounds in English (but only two affricates).
+nasal 
[n]: +alveolar E
[+nasal] is needed since [t, d, s, z] are also [+alveolar]. [+alveolar] is needed because [m, ]
are also [+nasal].
————————————————————————————————————
Feature notation also allows us to refer to whole classes of sounds at a time. This is similar to
the use of features in inflectional morphology, where they permit use to refer to classes of inflected
forms. For example, the expression [+voiced, +alveolar] would pick out the segments [d, z, n, ɹ, ɾ,
l] if we were dealing with English. Similarly, the expression [+syllabic, +high] picks out the
vowels [i, , u, ] from the set of all English sounds. The expression [+syllabic], with just feature,
picks out the set of all vowels. This is one of the essential purposes of a phonetic feature system,
since in phonology it is quite typically for whole groups of sounds to behave alike, based on their
shared phonetic properties. This is really no different from we made of features in inflectional
rules. For example, [Tense:past] might designate a whole set of possible morphosyntactic
representations, which might differ from each other, for example, in person and number, but all of
which are equally deserving to receive (for example) a past tense suffix when inflection is derived.
Any representation that is designated by the expression [Tense:past], whatever its features for
person and number, would be eligible for attachment of this suffix.
14.2 The role of features in language: preview of phonology
Often, when we put morphemes together into words (Chapter 2) or put words together into
sentences (Chapters 4-6), the resulting sequences get pronounced in a way that is not the “sum of
their parts”. Rather, there are phonological changes that adjust the basic sequences in some way.
Here is a simple example. The distinguished UCLA linguist Russell Schuh was known to his
colleagues as “Russ Schuh”. Prof. Schuh, a skilled transcriber, noted that he had become used to
hearing his name pronounced as [ɹəʃʃu], as if he were “Rush Schuh”. The crucial data are:161
161
I will leave off the stress marks for simplicity.
Russ [ɹəs]
Schuh [ʃu]
Russ Schuh [ɹəʃʃu]
Similarly English speakers say miss [mɪs], Sheila [ʃilə], but miss Sheila [mɪʃʃilə]; nice [naɪs],
shadow [ʃædo], but [naɪʃʃædo], and so on. The substitution is to replace [s] by [ʃ] when another [ʃ]
follows, resulting in phonetic double [ʃʃ].162
[z] is similar to [s] (its voiced partner) and it undergoes a parallel process, becoming [ʒ], the
voiced partner of [ʃ].
use [juz]
Schuh’s book [ʃuz bʊk]
use Schuh’s book [juʒ ʃuz bʊk]
We can think of [ʃ] as the “trigger” of this process; it causes the [s] to become [ʃ] and the [z]
to become [ʒ]. With this in mind, we might ask if [ʒ], the voiced partner of [ʃ], can likewise act as
a trigger for the change. This is hard to check, since words in English cannot begin with [ʒ]. But
pushing things a bit, we can try some Russian loanwords:
Russ [ɹəs]
Zhirinovsky [ʒɪrənɔfski]
Russ Zhirinovsky [ɹəʃ ʒɪrənɔfski]
use [juz]
Zhivago’s book [ʒɪvɑgoz bʊk]
use Zhivago’s book [juʒ ʒɪvɑgoz bʊk]
So it looks like [ʒ] is indeed a possible trigger. Summing up, we want a rule like this:
Alveolar Fricative Palatalization163
s  ʃ  ʃ 
  → ʒ when ʒ immediately follows.
z    
E E E
We have two sets of sounds here, {s, z} and {ʃ, ʒ}. These sets are hardly arbitrary; they have a
basis in the phonetic properties of these sounds. Thus, the phonological features become relevant.
The set {s, z} consists of all and only the alveolar fricatives. The notation here:
+fricative
+alveolar  E
If you wonder what a single [ʃ] sounds like, try the sentence Rush oodles of food to the meeting
162
room. It begins [ɹəʃ u...], with a single [ʃ], which is simply the final [ʃ] of rush.
163
“Palatalization” is a common name for any rule that shifts sounds into the (roughly) palatal region.
means “all and only the sounds (of the language under study) that are [+fricative] and [+alveolar]”.
Moreover, it is sensible to let rules alter the value of individual features. We can do this for
Alveolar Fricative Palatalization by stating it formally as follows:
(247) Alveolar Fricative Palatalization (restated with features)
+fricative −alveolar  +fricative 

+alveolar  → +palato-alveolar when +palato-alveolar immediately follows.
E E E
As we work out the basics of our phonological theory, we will see that this rule does what it is
supposed to do. The key assumption, carried over from our earlier work with features, is that any
feature not mentioned in a rule is assumed to stay the same. So, for instance, if we start out with
[s], which is [−voice], and apply Alveolar Fricative Palatalization, we end up with [ʃ], which is
likewise [−voice]. If we start out with [z], which is [+voice], and apply the rule, we end up with
[ʒ], which is likewise [+voice]. This is what permits us to describe symmetrical changes such as
that of Alveolar Fricative Palatalization.
The more general idea at hand here here is that phonological rules do not apply to arbitrary
lists of sounds, but to groups of sounds defined setting the values of some group of phonetic
features. Indeed, a term often used in this context is the following:
A natural class is the complete set of sounds in some language that share the
values for one or more features.
Thus, Alveolar Fricative Palatalization applies to the English natural class of alveolar fricatives,
+fricative
denoted {s, z} or +alveolar  .
E
There is second role that the features play in phonological rules: the change made by a rule is
usually not some massive change of sound, but merely a change in some small number of the
features. Alveolar Fricative Palatalization only changes the values of [alveolar] and [palato-
alveolar].
For now, what is important as an analytic skill is to be able to use the features to identify
natural classes of sounds, and to execute parallel changes when they occur in a rule.
The precise use of features can be tricky; and I suggest the use of appropriate software. The
program “Pheatures Spreadsheet” enables precise calculations and can be obtained (in a version
using the features presented here) from the author’s website at
linguistics.ucla.edu/people/hayes/20/. All feature calculations in this text have been checked using
this software.
Study Exercise #69
Suppose the following sets appear on the left side of the arrow or in the context of a
phonological rule of English. What features would you use to characterize the set? (Or, to put it
differently, what features define the natural class?)
a. [u,i]
b. [i, , e, ɛ, æ]
c. [v, , z, ]
d. [w, u, , o, ɔ]
e. [m, n, ]
f. [l]
g. [p, t, k, tʃ, f, , s, ʃ, h]
h. [, ]
i. [b, d, g, dʒ, v, ð, z, ʒ, m, n, ŋ, l, r, ɾ, j, w]
Try to use the minimum number of features needed. Use the feature charts given in (244) on p. 409
above.
+high 
a. [u,i] +tense 
  E
+syllabic
−back 
b. [i, , e, ɛ, æ] +syllabic E
+voiced 
c. [v, , z, ] +fricative E
d. [w, u, , o, ɔ] [+round]
e. [m, n, ] [+nasal]
f. [l] [+lateral]
g. [p, t, k, tʃ, f, , s, ʃ, h] [−voice]
h. [, ] [+dental]
i. [b, d, g, dʒ, v, ð, z, ʒ, m, n, ŋ, l, r, ɾ, j, w] [+voiced, −syllabic]
———————————————————————————————————
Study Exercise #70
Same instructions as previous exercise.
a. [d, n, z, l, ɾ]
b. [w]
c. [h]
d. [æ,ɑ,ɔ]
e. [e, ɛ, o, ə, ɚ]
f. [ɛ,ə,ɚ]
g. [ɛ,ə]
h. [æ, ɪ, ʊ, e, ɛ, o, ə, ɔ, ɚ, ɑ, i, u, b, d, g, dʒ, v, ð, z, ʒ, m, n, ŋ, l, ɹ, j, w]
i. [f, θ, s, ʃ, h]
a. [d, n, z, l, ɾ] [+voice, +alveolar]

b. [w]
In the features used here, there are many answers: [+round, +velar], [+back, +velar],
[+glide,+velar], [+glide,−palatal], [+glide,+round], [+glide,+back], [−syllabic,+round], [−syllabic,
+back]
c. [h] [+glottal] or [+aspirated]

d. [æ,ɑ,ɔ] [+low]
e. [e, ɛ, o, ə, ɚ] [−high,−low]
f. [ɛ,ə,ɚ] [−tense,−high]
g. [ɛ,ə] [−tense,−high,−palatoalveolar]
h. [æ, ɪ, ʊ, e, ɛ, o, ə, ɔ, ɚ, ɑ, i, u, b, d, g, dʒ, v, ð, z, ʒ, m, n, ŋ, l, ɹ, j, w] [+voice]
i. [f, θ, s, ʃ, h] [−voice,+fricative]
———————————————————————————————————
Study Exercise #71
In this exercise, you need to specify not just a class of sounds, but the whole rule. As before,
assume that the rules are in English, and use the feature charts given in (244) on p. 409 above. You
should use as few features as you can, but make sure you specify the change of the rule in full
(give all changing features).
I suggest using the software at https://github.com/BrucePHayes/Pheatures-for-Linguistics-20.
a. t, d become tʃ, dʒ before j, w.

b. i, ɪ become u, ʊ after w.
c. tʃ, dʒ become ʃ, ʒ after a vowel
+stop 
+affricate
−stop 
a. +alveolar → +palato-alveolar before [+glide]
−alveolar 
E E
+high  +back  +glide  164

b. +syllabic → +round after +round
E E E
+fricative
c. [+affricate] → −affricate after [+syllabic]
E
—————————————————————————————
For further reading
A highly recommended textbook is A Course in Phonetics, written by an eminent phonetician,
the late Peter Ladefoged. Ladefoged prepared a website, still posted at his home university, that
relates to his text and can be used for studying phonetics:
www.phonetics.ucla.edu/course/contents.html.
The web site of the International Phonetic Association has much useful material on descriptive
phonetics, particularly its descriptions of particular languages:
https://www.internationalphoneticassociation.org/. The Handbook of the IPA is their reference
work on the International Phonetic Alphabet and on phonetic transcription and includes on line
sound files: https://www.internationalphoneticassociation.org/content/handbook-ipa.
−syllabic
Instead of +round you could use variety of other combinations, such as +round  .
164
+glide
E E
Chapter 11. Phonology I: Phonological Rules and Phonemic Analysis

1. Phonology
Phonetics studies speech sounds as physical events; whereas phonology studies the (mostly
unconscious) rules that govern the use of sounds in language. That is, phonology studies the
“grammar of sound.” For instance, the rule of Alveolar Fricative Palatalization, treated at the end
of the previous chapter ((247) on p. 415), would count as part of this grammar.
2. What is the content of phonology?
Phonologists study, more or less, four aspects of sound patterning.
First, they study how sounds change in context. Alveolar Fricative Palatalization describes
how the basic sounds [s] and [z] vary when they occur just before a [ʃ] or [ʒ]. This change of
sounds by context is often called alternation and is discussed further below.
Second, phonologists study the principle of legal sequencing of speech sounds — essentially,
phonological grammaticality. To give on example, it is “phonologically legal” for English words to
begin [bl] (and plenty of them do: blend, blood, black, bliss, and so on). But it is phonologically
impossible for English words to begin with *[bn]: a word like bnick [bnɪk] is judged by English
speakers to be aberrant; and English speakers often have great trouble even in saying it (they tend
to “repair” the bad sequence by putting in a schwa: [bənɪk]). The study of legal sound sequencing
is often called phonotactics and, in analysis, is usually done by means of setting up a system of
constraints, which use the phonological features just like the rules do.
Third, phonologists study how the realization in sounds is related to other components of
the grammar. Here is a simple example. In the variety of American English I speak, the word
bonus (similarly onus, phonograph, persona), the sequence [o n ə] has a nasalized [õ] and a very
short [n] (in IPA it would be transcribed [ɾ]̃ ). But in slowness or lowness the [o] is not nasalized,
and the [n] is an ordinary regular-length [n]. Here is the full comparison:
bonus [ˈbõɾə̃ s] slowness [ˈslonəs]

onus [ˈõɾə̃ s] lowness [ˈlonəs]
phonograph [ˈfõɾə̃ gɹæf]
persona [pɚˈsõɾə̃ ]
These differences evidently have to do with the fact that slowness and lowness are derived by a
word formation rule, namely the -ness Rule seen in (65) in Chapter 2 ([ X ]Adj →
[[ X ]Adj ness ]Noun). Quite often, the syntactic or morphological source of an utterance will have
some kind of effect on its phonology.
Lastly, phonologists are interested in the contextual variation observed for most speech
sounds. They attempt to discover this variation and analyse it in the usual manner of linguistics
with a set of formalized rules and representations. We will cover this topic starting in section 7
below.
3. Formalization of phonological rules
The formalization of phonology quickly gets us into really massive representations, and the
expression of rules and derivations must proceed with care. In this section, I will give a pretty
careful version of formalization; once we have it in hand, we can proceed more casually (and with
less effort).
The rule we will formalize and deploy is Alveolar Fricative Palatalization, already given in the
previous chapter.
Alveolar Fricative Palatalization

+alveolar  → +palato-alveolar when +palato-alveolar immediately follows.
E E E
Recall that the rule involves two natural classes, expressed with features: the input set [s, z] and
the output set [ʃ ʒ]. All four possibilities drawn from these sets actually occur, so the rule derives
the following:
a. Russ Schuh [ɹəʃʃu] from [s] + [u]

b. use Schuh’s book [juʒ ʃuz bʊk] from [juz] + [ʃu]
c. Russ Zhirinovsky [ɹəʃ ʒɪrənɔfski] from [s] + [ʒɪrənɔfski]
d. or use Zhirinovsky’s book [juʒ ʒɪrənɔfskiz bʊk] from [juz] + [ʒɪrənɔfski]
At this point we can aim for a higher level of precision, and the first task is to develop a
notation for phonological contexts. We will use the slash / to mean “in the environment, in the
context”. We will also use underscores, ___, to mean “imagine that the sound targeted by the rule
occurs here.” So now:
/ ___ X means “before X”
Similarly:
/ X ___ means “after X”

/ X ___ Y means “between X and Y, in that order”
In this notation, Alveolar Fricative Palatalization is stated like this:
(248) Alveolar Fricative Palatalization (formalized)

+alveolar  →
E
+palato-alveolar / ___
E
+palato-alveolar E
In general, many phonological rules can be formalized with the scheme

A → B / C __ D
where A, B, C, and D are partially specified feature matrices. C, or D, are often not present (for
instance, Alveolar Fricative Palatalization has no C).
3.1 How features are treated in phonological rule application
Let us continue to be precise about rules and their application.
3.1.1 Left side of the arrow

+fricative
A matrix of features, such as +alveolar  , whenever it appears on the left side of the arrow
E
has very specific meaning: it designates any sound that possesses all of the feature values
+fricative
indicated. So, for English, +alveolar  designates the set { s, z }. This is because [s] and [z] are all
E
and only the sounds that are [+fricative] and [+alveolar] in English.
+fricative
Note that this is an English-particular fact. If we were using the feature matrix +alveolar  for E
the analysis of Korean, we would have to bear in mind that Korean actually has three alveolar
fricatives, { s, z, s }, where [s] is the IPA symbol for an aspirated s, with a strong puff of breath
that distinguishes it from regular [s].165 So, feature matrices can be taken to have both a general
meaning — the set of properties they designate — and a meaning specific to the language under
analysis — the set of sounds that the language has that are specified by these properties.
3.1.2 Right side of the arrow
On the right side of the arrow in a rule, a feature matrix actually means something quite
different. It does not designate a set of sounds, but rather is treated as an instruction to change
every feature value of the input segment in the way specified in the matrix. I think it fair to say that
the features on the right side of an arrow aren’t really a matrix at all, even though they are notated
that way. The use of the same notation is just a matter of convenience.
To restate a point made earlier: the theory assumes that any feature not specified in a rule as
needing to be changed is left the same; no changes unless the rule says so. We will see an example
of this shortly.
3.1.3 Rule environment
Lastly, in the environment of a rule (after the slash), we use feature matrices in their normal
way, to designate a set of sounds. Just like matrices on the left side of the arrow, matrices in the
contex stand for any sound that possesses the indicated feature values (and again, we don’t care
about any other values that might be present.)
165
If you know Korean, here is the relevant background: [s] is spelled as ㅆ. [sʰ] is spelled as ㅅ. [z] in Korean
is the usual pronunciation of ㅅ when surrounded by voiced sounds (it is an allophone; see below).
3.2 Formalized example of phonological rule application
Putting these concepts together, we can do (just once) a fully formal and careful application of
a phonological rule. Recall that the name Russ Schuh is basically [ɹəs] + [ʃu]. Looking at the
feature chart under (244), we can translate each symbol into a feature matrix, producing five
matrices in a row.
(249) Russ Schuh ([ɹəs] + [ʃu]) in features
−syllabic +syllabic −syllabic −syllabic

 
+syllabic
−stop −stop −stop  −stop  −stop

−affricate
−fricative  −affricate
−fricative  +fricative  +fricative  
−affricate −affricate −affricate
−fricative 
+liquid
−glide
 −liquid
−glide  −−liquid   −liquid
−glide  −liquid
−glide

+voiced   −−voiced  −−voiced  
glide
+voiced +voiced
−nasal −nasal −nasal
0high    0high   0high  
nasal nasal
−high +high
0back 
−low
 −0back  −0back  
0low 0low 0low −low
+back +back
0tense 
−round
 0tense  0tense  
E
−round E
round E
round E
+round E
−tense +tense
−labiodental  
−bilabial −bilabial
−labiodental  −labiodental  −labiodental  
−bilabial −bilabial −bilabial
−labiodental 
−−dental
alveolar  −dental
−alveolar  −+alveolar
dental
 −−dental
alveolar  −dental
−alveolar 
−+palato −alveolar
 −palato−alveolar  −−palato −alveolar
 +palato
−palatal
−alveolar 
 
−palato−alveolar
−palatal
−−velar 
palatal −palatal
 −−velar  −−velar  
palatal
−velar −velar
−lateral
glottal
 −glottal
−lateral
 −lateral
glottal
 −lateral
glottal
 −glottal
−lateral

ɹ ə s ʃ u
The zeros seen in some instances of [high], [low], and [back] indicate that these segments have no
value for these features; this is the “underspecification” discussed in the previous chapter on p.
408.
I repeat below the formalized statement of Alveolar Fricative Palatalization.
Alveolar Fricative Palatalization (formalized)

+alveolar  → +palato-alveolar / ___ +palato-alveolar
E E E
A rule can be imagined as a little machine that scans across the input, seeking sequences that are
eligible to be altered. Alveolar Fricative Palatalization affects only those sounds that are
+fricative
+alveolar  , and in applying the rule, one seeks sound that fit this description. If you look at the
E
feature matrix for [ɹəs] + [ʃu] in (249), you will see that it does indeed possess these two feature
values. Moreover, Alveolar Fricative Palatalization also requires that, for it to apply, the following
+fricative 
sound must possess the features +palato-alveolar . Inspecting (249), you can see that this is
E
likewise so. Lastly, the rule states that if these two conditions are met, then features of the target
segment (left side of arrow) must be changed to [−alveolar] and [+palato-alveolar]. This is shown
happening in (250).
(250) Changing Russ Schuh [ɹəs] + [ʃu] to [ɹəʃ] + [ʃu] by Alveolar Fricative Palatalization
−syllabic −syllabic −syllabic
 
+syllabic +syllabic
 −stop
 −stop −stop

−stop

−stop
+liquid   +fricative  +fricative  

−affricate −affricate −affricate −affricate −affricate
−fricative −fricative −fricative
+voiced    glide   glide  +voiced 

−liquid −
liquid liquid − −liquid
−glide −glide − − −glide
0high    0high   0high  +high 

+voiced −
voiced voiced −
−nasal −nasal nasal− nasal − nasal −
0back    0low   0low  +back 

−high
0low −low low −
0tense    0tense   0tense  +tense 

+back 0back 0back
−round E
−round
E
round− round E
− +round E E
 bilabial
−  −tense
−bilabial   bilabial
−   bilabial −  labiodental
bilabial − 
 alveolar    −alveolar   dental   dental 
−labiodental −labiodental −
labiodental labiodental − −
−dental −dental −
dental − −
+palato alveolar 
−
−
−alveolar
  +palato−alveolar  +palato alveolar  palato alveolar 
−palato−alveolar
alveolar − alveolar
−
−
− −
 palatal
−
−velar  −palatal
−velar   palatal
−
−   palatal
velar
−
−  velar
palatal −
− 
 lateral    lateral   lateral   lateral 
velar
−glottal −glottal −
glottal glottal − glottal −
− −lateral − − −
ɹ ə ʃ ʃ u
The full representation (250) is in fact just [ɹə ʃu] expressed in feature notation (as I have aligned
it under the feature matrices), so the rule does indeed derive the correct output.
As already noted, any features not changed by the rule are left unchanged. Thus, [s] starts out
as [+fricative], and, after it is changed to [], it is still [+fricative], and similarly with all the other
features not mentioned in the rule.
The derivation just given is actually very unusual in being so complete; I have put every
feature in and thus been completely explicit about how the rule applies. In actual practice, linguists
are reluctant to expend this much effort, and they leave the irrelevant features out for convenience.
However, in general even when this is done, it is tacitly assumed that the “real” representations are
detailed, more or less in the way given here; and that shorter versions are being used just as
convenient abbreviations. In contrast, when a phonology is programmed into a computer, as has

been repeatedly done, there is no problem with including all the features and being completely
explicit.
Recall from above that the rule of Alveolar Fricative Palatalization is more general than just s
→ ʃ / ___ ʃ. It also applies to [z], converting it in parallel fashion to the palato-alveolar counterpart
of [z], namely [], as in use Schuh’s book [juʒ ʃuz bʊk]. The features that change in making [s]
into [] are the same features that change in making [z] into []; and likewise when the rule
(unusually) applies before [ʒ] (as in Russ Zhirinovsky).
The Alveolar Fricative Palatalization rule says nothing about the feature [voice], even though
[voice] is the feature that distinguishes [s] from [z], and [] from []. Under the theory, it is a good
thing that voice is not mentioned, because it is precisely by not mentioning it that we can capture
the parallel changes of s → , z → .
3.3 More practice in formulating rules with features
In this textbook, the philosophy for use of features is: make the rule as terse as it can be while
still deriving the correct output. This was in fact done with the formalization of Alveolar Fricative
Palatalization above. Every feature included in the rule is necessary, in the sense that the rule
would derive wrong results if you left it out. No redundant features are included. This is part of the
scholarly ethos of linguistics: linguists aim for maximally concise, yet accurate, analyses.166
Study Exercise #72
Here is Alveolar Fricative Palatalization again, but with all six of its features numbered.
Alveolar Fricative Palatalization
 −alveolar
1 3 5
 +fricative    +fricative 
 2+alveolar →
E
 4+palato-alveolar / ___  6+palato-alveolar

E E
For each feature, state in words what goes wrong if it is omitted. If possible, give an example
of a bad result derived.
166
There is a more serious reason to be concise: a concise rule is more general and makes broader predictions.
For instance, our rule of Alveolar Fricative Palatalization predicts that if English were ever to borrow a new sound [sʰ]
(say, from Korean), then [sʰ] would likewise turn into [ʃʰ] before [ʃ] or [ʒ]. It’s good for the rule to make testable
predictions of this kind.
Answer to Study Exercise 72
1. (leave out [+fricative])
The rule would apply not just to [s] and [z] but to all alveolars, including the stops [t] and [d]
as well as [n] and [l]. These would all be turned into palato-alveolar sounds that do not exist in
English. (Indeed, palato-alveolar stops, nasals, and laterals are rare and can only be transcribed in
IPA by using special diacritics.)
2. (leave out [+alveolar])
The rule would wrongly convert the non-alveolar fricatives [θ] and [f] to [ʃ], and also [ð] and
[v] into [ʒ]. Example: with Schuh [wɪθ ʃu] → *[wɪʃ ʃu].
3. (leave out the change to [−alveolar])
If you change a sound to [+palato-alveolar] without taking away the [+alveolar], you get a
sound that is both [+palato-alveolar] and [+alveolar]: the tongue would touch the roof of the
mouth over a very extensive region. This is probably possible, but it is certainly not right for
English and there are no IPA symbols for such sounds.
4. (leave out the change to [+palato-alveolar])
If you do this without providing any new place of articulation, you get a combination of
features that designates no actual sound (because it has no place of articulation). So the output
would not even be interpretable.
5. (leave out [+fricative] in the context of the rule)
If this is done, then any [+palato-alveolar] sound should trigger the rule. This would include
the affricates [tʃ] and [dʒ]. So we should get palatalization in phrases like Russ Church or Russ
Johnson. Perhaps there is a dialect of English in which this actually occurs, but it does not occur in
the target dialect (*[ɹəʃ tʃɚtʃ], *[ɹəʃ dʒɑnsən]).167
6. (leave out [+palato-alveolar] in the context of the rule)
If this is done, then any [+fricative] sound should trigger the rule. The set of triggers would
include the fricatives [f], [v], [θ], [ð], [s], [z], [h]. This makes a great number of wrong predictions,
such as Russ Smith (*[ɹəʃ smɪθ]).
———————————————————————————————————————
167
In the feature system adopted here, [ɹ] is also [+palato-alveolar], so we should also get palatalization in Russ
Richards. Again, not so in the target dialect: *[ɹəʃ ɹɪtʃɚdz].
Let us ponder a minute what it means to use features minimally in expressing a rule. In actual
practice, you can often get away with very few features on the left side of the arrow (also in the
rule context, when there is one). These parts of the rule only need to single out a group of sounds
from the set of sounds that the language already has. But, on the right side of the arrow, it is often
necessary to specify quite a few features, since we want the rule to produce the intended sounds—
precisely—in its outputs. A motto that might help here is “terse on the left side, verbose on the
right side” of the arrow.
Example 1: suppose that in some hypothetical dialect of English we wanted i u ɪ ʊ → ĩ ũ ɪ ̃ ʊ̃,

where the tilde indicates nasalization on vowel; the feature being [+nasal]. The left side of the rule
turns out to be [−syllabic,+high] ([−syllabic] is needed because the glide [j, w] are also high). The
right side of the rule need only mention [+nasal], since that is all that changes; hence, [+high] →
[+nasal].
Example 2: suppose in English we want k g → p b. The left side is [+velar,+stop]. The right
+bilabial
side must specify −velar  . Why [−velar]? Because plenty of languages (especially in West
E
Africa) have sounds that are both [+velar] and [+bilabial]—they have two articulations at once.
IPA renders these sounds as [k͡p g͡b ŋ͡m].168 If we didn’t change [+velar] to [−velar], we’d end up
with one of the “labial-velar” sounds as the incorrect output.
Example 3: suppose in English we want this:
ptk→mnŋ
+stop 
The left side of the arrow in the formalized rule has to have −voice , since we don’t want [b] [d]
E
+nasal 
[g] to undergo the rule. On the right side of the arrow we need to have −stop  , since nasals are
E
+voice
nasal, and they aren’t stops, and they are voiced (look at feature chart (185) on p. 409, to see that
these are indeed exactly the three features that need to change).169
Example 4: suppose in English we want ɛ ɪ → e i. Let’s do the change first: this clearly has
to be X → [+tense]. Now, what is the simplest characterization for X? Taken in isolation, the
answer would be [−back,−low,−tense]. Yet this is not actually simplest answer; the simplest
168
And if you are pondering a theory like “automatically change the other features so that you arrive at
the closest sound in the phoneme inventory compatible with the change”, then ponder the rule of /ɹ/
Rounding, (190) on p. 432. There is no /ɹʷ/ phoneme in English, and the closest phoneme compatible with
adding [+round] to /ɹ/ is /w/. This works great for describing the phonology of many small children (rabbit =
[ˈwæbɪt]), but not ordinary adult English.
169
Note in particular that it counts for nothing under the theory that English has no voiceless nasals; we might
suppose that voiceless nasals somehow get “fixed” to voiced nasals, but this is asking for more than the theory actually
says. We really need to provide the [+voiced] feature.
answer is actually [−back,−low]. For ɪ ɛ, this works straightforwardly. And for i e, the rule takes
an input that is already [+tense] and harmlessly “turns it into” an output that is [+tense].
Application that harmlessly makes no change is sometimes called vacuous application. Vacuous
application sometime implies that rules can be simplified even more than even (perhaps) common
sense would dictate, but it seems sensible to follow a consistent principle of maximal simplicity.
Study Exercise #73
Formulate these rules using feature chart (185) on p. 409. Assume that the inventory of sounds is
as in English. As before, I believe you will find it far more reliable to use the software “Pheatures”,
mentioned earlier; location is
https://linguistics.ucla.edu/people/hayes/20/PheaturesForLinguistics20.htm.
a. t, d become tʃ, dʒ before j, w.170

b. l becomes ɹ when another l comes before it, separated by a vowel.171
c. tʃ, dʒ become ʃ, ʒ between vowels.172
d. v, ð, z, ʒ become f, θ, s, ʃ at the end of a word.173
e. ɪ, ɛ, æ are deleted at the end of a word.
f. ɪ, ʊ become voiceless vowels (IPA [ɪ̥ ], [ʊ̥]) when they occur between {p t k tʃ f θ s ʃ h} and
{p t k tʃ f θ s ʃ h}.174
g. u, ʊ become i, ɪ everywhere175
170
This one is more or less real: get you [ˈgɛtʃju], said you would go [ˈsɛdʒjuwʊdˈgo], twin [tʃwɪn],
dwell [dʒwʒɛl]. A later rule normally deletes [j] after palato-alveolars, yielding [ˈgɛtʃu], [ˈsɛdʒuwʊdˈgo].
171
Based on Latin phonology. This is why we say nav-al, but sol-ar (words borrowed into English
from Latin).
172
More or less real, as a rule of the vernacular dialect of Italian spoken in Florence.
173
This is close to being true of English, though it really happens only at the ends of phrases, and the
devoicing is sometimes partial.
174
Not unlike a phonological rule of Japanese.
175
A change that was once a phonological rule in Yiddish.
+stop 
+affricate
−stop 
a. +alveolar → +palato-alveolar / ___ [+glide]
−alveolar 
E E
−lateral 
b. [+lateral] → +palato-alveolar / [+lateral][+syllabic] ___
E
−alveolar 
+fricative
c. [+affricate] → −affricate / [+syllabic] ___ [+syllabic]
E
d. [+fricative] → [−voice] / ___ ]word Note the use of vacuous application (to [f, θ, s, ʃ]) to
simplify the rule.
−back 
e. −tense →  / ___ ]word
E
+high 
f. −tense → [−voice] / [−voice] ___ [−voice]
E
−back 
g. [+high] → −round E Note the use of vacuous application (to [u,ʊ]) to
simplify the rule.
3.4 Derivations in phonology
Derivations in phonology follow the two-column format used elsewhere in linguistics. Rules
apply in succession, changing the representation, and the linguist justifies each step of the
derivation by providing the name of a rule in the same row, in the right column.
We have already seen what a full, explicit derivation would look like — it would be
enormous! In actual practice, phonologists use the feature matrices only for their rules, and for the
words that the rules are applying to, they use IPA symbols.
Phonological derivations must begin with a starting point, which for phonology is called the
underlying representation.176 The choice of a correct phonological underlying representations is
something we will need to develop in detail below. For now, let us simply assume that the
underlying representation of Russ is s, and the underlying representation of Schuh is u. It is a
standard convention in phonology to surround underlying representations with slant brackets; thus
the underlying representation for Russ Schuh is /s u/.
So, a phonological derivation in standard format would be as follows:
176
One could draw an analogy: the phonological underlying representation is to the phonological rules what
the syntactic deep structure is to the set of transformations; each is the input to a series of rules that transforms it.
Russ Schuh
/s u/ Underlying representation
 u Alveolar Fricative Palatalization
[ u] Surface representation (output)
As with the morphological derivations we have done earlier, there is enough space to give the
representation in two-column format, with representations on the left and names of rules applied
on the right. It is assumed that the reader can check the values of symbols like [ʃ] and [s] against a
feature chart, and verify that the rule does indeed make the changes shown with the phonetic
symbols of the derivation.
Here are a couple more details. For “output”, one can use the more explicit terms “surface
representation”, or simply “phonetic representation”. Also, it is often clearer when showing how a
rule applies to indicate only the segment that changes, aligning it vertically and leaving blank
space elsewhere. With these two changes, the derivation about would come out as follows:
Russ Schuh
/s u/ underlying representation
 Alveolar Fricative Palatalization
[ u] phonetic representation
In a larger-scale example, multiple phonological rules would be applied — in fact, as we will see,
in a particular order. Each rule gets its own line in the derivation.
4. Some phonological formalisms
Phonological rules can be written with some of the notations that we used in Chapter 2 for
some of the more interesting types of morphology (that is, not prefixation or suffixation).
First, the symbols C (for “consonant”) and V (for “vowel”), used earlier for morphology, are
applicable to phonology, with the refinement that we now defined them as abbreviations: C is a
shorter way to write [−syllabic] and V is a shorter way to write [+syllabic].
Second, the use of labeled brackets carries over to phonology, though usually the bracket
employed simply designates the edge of any word, rather than some particular category like Noun
or Verb. So, for instance, we can write a rule that converts [i, u] to [e, o] at the end of a word to
something like this:177
177
Final Lowering rules of this kind are found in Chamorro (Austronesian, Guam) and Lardil (Pama-Nyungan,
Australia).
Final Lowering
+syllabic
+high  → [−high] / ___ ]
  word
+tense 
In some languages, vowels become long when they are separated by the sequence CV from the end
of a word; as in [tapi] → [taːpi]. This can be written with CV notation as follows:
Penultimate Lengthening178
V → [+long] / ___ CV ]word
Lastly, the use of the variable X, meaning “a string of sounds of any length, including zero,” is
sometimes useful for phonology. For instance, the rule of Alveolar Fricative Palatalization applies
only when the triggering sound [ʃ] or [ʃ] immediately follows the target sound [s] or [z]. But there
are a number of languages in which palatalization will apply even if the triggering [ʃ] or [ʒ] is
several segments away in the word. Such a “long-distance” palatalization rule could be written:
Alveolar Fricative Palatalization (long-distance version)

+alveolar  → +palato-alveolar / ___ X +palato-alveolar
E E E
“Replace an alveolar fricative with the corresponding palato-alveolar fricative when a

palato-alveolar fricative occurs late in the same word.”
Such a rule would convert English sushi [suʃi] to [ʃuʃi], or sunshine [sənʃaɪn] to [ʃənʃaɪn].179
5. Optional rules in phonology
In the discussion of the English rule of Alveolar Fricative Palatalization given so far, the data
have been somewhat simplified: the examples given earlier can also be pronounced without
178
“Penultimate lengthening” = “lengthening of the vowel of the second-to-last syllable”. Languages with
Penultimate Lengthening include Italian (Romance, Italy) and Mohawk (Iroquoian, New York and Ontario).
179
In real languages, rules of this type are called Sibilant Harmony; they are found, for instance, in
Kinyarwanda (Bantu, Rwanda) or Chumash (Chumashan, Southern California). Young children, in their personal
phonologies, often show Sibilant Harmony; the examples [ʃuʃi] and [ʃənʃaɪn] given above were uttered by my son
when he was little.
applying the rule. So, if we were to present the data in greater detail, we would give options of
pronunciation for all of the forms in question.
Russ [ɹəs]
Schuh [ʃu]
Russ Schuh [ɹəsʃu], [ɹəʃʃu]
use [juz]
Schuh’s book [ʃuz bʊk]
use Schuh’s book [juz ʃuz bʊk], [juʒ ʃuz bʊk]
Russ [ɹəs]
Zhirinovsky [ʒɪrənɔfski]
Russ Zhirinovsky [ɹəs ʒɪrənɔfski], [ɹəʃ ʒɪrənɔfski]
use [juz]
Zhivago’s book [ʒɪvɑgoz bʊk]
use Zhivago’s book [juz ʒɪvɑgoz bʊk], [juʒ ʒɪvɑgoz bʊk]
The variation seems to be an element of speaking style: the rule is applied in fluent, ordinary
speech, but could be suppressed in formal, careful speech.
When a phonological rule is optional, I will indicate this simply by placing the word
“(optional)” in parentheses after the rule:
Alveolar Fricative Palatalization (restated as optional)

+alveolar  →E
+palato-alveolar / ___ +palato-alveolar (optional)

E E
5.1 Derivations for optional rules
There is no standard way to write derivations for optional rules, but in this textbook I will use
an ad hoc notation, which I will call a “branching derivation.” The format gives arrows indicating
the two possibilities for when an optional rule does or does not apply; for example:
(252) A branching derivation for Russ Schuh
Russ Schuh
/ɹəs ʃu/ underlying representation
— ɹəʃ ʃu Alveolar Fricative Palatalization

[ɹəs ʃu] [ɹəʃ ʃu] surface representation
Where there are multiple applicable rules, the branches will multiply, producing a tree of
greater size. Some commercial speech recognition devices use rules to generate alternate forms of
the words to be recognized; their derivations can culminate in hundreds of branches.
5.2 Some further optional rules in English
5.2.1 Tapping
Another optional rule (of North American English180) is Tapping, which derives the tap [] as
a variant of [t].181 The data look like this:
Forms that can be tapped
butter [ˈbɚ]
attic [ˈæk]
heritability [hɛɪəˈbləi]
motto [ˈmo]
Forms that cannot be tapped
attain [əˈten] (*[əˈen])

Tommy [ˈtmi] (*[ˈmi])
cat [ˈkæt]182 (*[ˈkæ])
actor [ˈæktɚ] (*[ˈækɚ])
Atkins [ˈætknz] (*[ˈæknz])
Inspection of these and similar data indicate a very particular environment for Tapping,
namely: between two vowels (or other syllabic sounds; diphthongs and syllabic consonants), of
which the second must stressless:
(253) Tapping (preliminary version)
+syllabic
t →  / [+syllabic] ___ –stress  (optional) E
Now, different speakers will vary, but my impression is that most speakers of North American
dialects can, in very slow and careful speech, “turn off” tapping and produce [t]’s in the relevant
words:
180
Tapping is unusual outside North America. It occurs natively in some Irish speech and is apparently
currently in the process of spreading into overseas dialects such as Australian. Non-tapping dialects often
have Glottaling instead: butter [ˈbəʔə], [ˈbʊʔə].
181
And, as we’ll see later on, of /d/.
182
Or, optionally, [ˈkæʔt], not relevant here.
butter [ˈbɚ], [ˈbtɚ]

attic [ˈæk], [ˈætk]
heritability [hɛəəˈbləi], [hɛətəˈbləti]
motto [ˈmo], [ˈmto]
Hence the rule is (marginally) optional.
5.2.2 /æ/ Diphthongization
A third optional rule, which is found in the dialect of many but not all American English
speakers, is a rule of /æ/ Diphthongization, which applies before nasals. Here are data:
No diphthongization:
cat [ˈkæt]
pack [ˈpæk]
lap [ˈlæp]
lab [ˈlæb]
pal [ˈpæl]
Diphthongization:
can [ˈkæn], [ˈkɛən]

man [ˈmæn], [ˈmɛən]
Spanish [ˈspænʃ], [ˈspɛənʃ]
dance [ˈdæns], [ˈdɛəns]
spam [ˈspæm], [ˈspɛəm]
tram [ˈtɹæm], [ˈtɛəm]
Assuming that the vowel in question is basically [æ], we can write the rule as follows:
(254) /æ/ Diphthongization
æ → ɛə / ___ [+nasal]
5.3 Optional rules and speaking style
Different optional rules tend to apply in differing speech styles. In most people’s speech,
Tapping is very close to obligatory, and “turning it off” (as in pity [ˈpti]) is appropriate only in the
most formal of speaking styles. I find that /æ/ Diphthongization can be “turned off” in somewhat
more casual contexts than Tapping can be; and Preglottalization can be turned off even in fairly
relaxed contexts. On the other end of this continuum, there are rules that (for me at least), only get
to apply in the most casual speech, for instance the rule (not formalized) that monophthongizes
/a/ to [æ] (as in, for instance: Get out of here! [gɛɾˈæɾəhiɹ].)
Research by sociolinguists indicates that when we speak, we unconsciously regulate the

application of a great number of optional phonological rules. Presumably, we do this for purposes
of making the appropriate impression on the people with whom we are speaking: as already noted
in Chapter 3, there are contexts that call for colloquial speech and contexts that call for formal
speech. Most people command a range of styles,183 and the ability to turn optional rules off or on is
part of this command.
5.4 Obligatory rules?
Obligatory phonological rules do in fact exist and are numerous. But to justify them we will
need to do a bit of further theoretical development, the task of the next section.
6. Phonology as related to morphology
Our main “toy” example so far, Russ Schuh (/ɹəs ʃu/ → [ɹəʃ ʃu]) illustrates an elementary
point: it appears, at least at first glance, that phonology is post-syntactic. In other words,
phonological rules apply after the syntactic rules have already joined words together into
sentences. Were this not so, then the sentential context of a word could never affect its
pronunciation, but this is plainly not so. Phonology is thus thought of as an “interpretative
component”; it provides a detailed specification of how utterances are pronounced, thus filling in
complete information that is present only in broad outlines in the syntax.
What about morphology? It can easily be seen from a rule we already have that phonology
must apply after morphology as well. I repeat below the North American English Tapping rule as
developed above:
Tapping (preliminary version)

+syllabic
t →  / [+syllabic] ___ –stress  (optional)
E
To see how this interacts with morphology, we can observe that quite a few suffixes in English
begin with a stressless vowel. Thus, when the morphological attaches them, they can in principle
“provide” the essential environment that causes Tapping to be applicable. And indeed, Tapping
really does take place under these circumstances; see data below.
write [ˈat]
writer [ˈaɚ]
183
By this I mean even monolingual, monodialectal people. Obviously, the ability to speak more than
one dialect or language increases the range of impressions that a speaker can create.
pat [ˈpæt]
patting [ˈpæɾɪŋ]
orbit [ˈoɹbɪt]
orbital [ˈoɹbɪɾəl]
Very often, to understand the patterning of morphologically related words, we need to

consider both the morphology and the phonology, and to be totally explicit about this, we can
provide a bicomponential derivation for the present example. The morphological rules are much
as we set them up earlier in Chapter 2, except that now we are in a position to write them with IPA
symbols, rather than the orthography we had earlier used as a stopgap:
(255) Deriving six words through morphology-phonology interaction: the morphological stage
-er Rule
[ X ]V → [[ X ]V ɚ ] N “person who Verbs”
-ing Rule184
[ X ]V → [[ X ]V ɪŋ ] N “process or instance of Verbing”
-al Rule (p. 49)

[ X ]N → [[ X ]N əl ]Adj “[[ X ]N al ]Adj” means “pertaining to X”
Within the word formation component, we can apply these rules apply as follows:
‘write’ ‘pat’ ‘orbit’

[ ˈɹaɪt ]V [ ˈpæt] V [ˈoɹbɪt]N Stems (listed in lexicon)
[[ ˈɹaɪt ]V ɚ ]N [ ˈpæt] V [ˈoɹbɪt]N -er Rule

‘one who writes’ — —
[[ ˈɹaɪt ]V ɚ ]N [[ ˈpæt ]V ɪŋ ]N [ˈoɹbɪt]N -ing Rule

— ‘process of patting’ —
[[ ˈɹaɪt ]V ɚ ]N [[ ˈpæt ]V ɪŋ ]N [[ˈoɹbɪt]N əl ]N -al Rule

— — ‘pertaining to orbits’
184
This is the -ing Rule for word formation, creating process nouns from verbs. The same suffix is also
attached by rule of inflectional morphology, not stated here, triggered by the feature [Aspect:PresentProgressive]; as in
She is dancing.
With this done, we now have six words that (taking them to be normal one-word utterances), we
can feed into the phonology. From the perspective of phonology, these are now input forms; i.e.
underlying representations, and as such we will surround them with slant brackets. For now, we
will also assume that the morphological brackets disappear at the start of the phonology; at the
very least, this will unclutter our representations. Hence the derivation for our six words within
the phonological component will look like this:
(256) Deriving six words through morphology-phonology interation: the phonological stage
‘write’ ‘writer’ ‘pat’ ‘patting ‘orbit’ ‘orbital’

/ˈɹaɪt/ /ˈɹaɪtɚ/ /ˈpæt/ /ˈpætɪŋ/ /ˈoɹbɪt/ /ˈoɹbɪtəl/ Underlying forms
— ˈɹaɪɾɚ — ˈpæɾɪŋ — ˈoɹbɪɾəl Tapping
[ˈɹaɪt] [ˈɹaɪɾɚ] [ˈpæt] [ˈpæɾɪŋ] [ˈoɹbɪt] [ˈoɹbɪɾəl] Surface forms
The pattern seen here is actually very common. In a great number of cases, a morphological
rule does not always display its effect in surface (pronounced) forms. Rather, what we get is a
somewhat modified surface form, resulting from the phonological rules. A word like patting is not
the “sum of its parts” in North American English ([ˈpæɾɪŋ]  [ˈpæt] + [ɪŋ]); rather it deviates from
this because of the effects of phonology.
7. Phonemic analysis
So far, we have assumed a very informal approach to the question of derivations in

phonology. The underlying representations are just the form of the words or morphemes of the
language; syntax or phonology strings them together, and phonology happens. In this sense, the
morphological and syntactic rules give us a very straightforward sense of what an underlying form
is.
This is not the whole story, however. Phonologists have further suggested that underlying
forms deviate from surface phonetic forms in a more subtle and abstract way; a concept that will
lead us toward analysis with phonemes and allophones. We will be doing phonology, with rules;
but in a way that in principle has nothing to do with the morphology or the syntax. In my teaching
experience, phonemes and allophones are much trickier to grasp, and are the most difficult part of
elementary phonology. I will try to make the presentation clear here.
A good way to start the discussion is to pose the following question:
How many speech sounds does English have?
At first blush, this seems a trivial question: in studying phonetics, we have already enumerated the
sounds of English (see charts on p. 383). We count them up here:
a. Consonants
/p, t, k, b, d, g, tʃ, dʒ, f, θ, s, ʃ, h, v, ð, z, ʒ, m, n, ŋ, l, ɹ, w, j/

/i, ɪ, e, ɛ, æ, u, ʊ, o, ɑ, ə, ɚ, aɪ, aʊ, ɔɪ/
Inspection indicates that the total is 38. The count is dialect-specific; as always in linguistics we
can only analyze one dialect at a time.
However, this count is deceptive, because the symbols are being used only to indicate broad
categories of sound. Not every /ɹ/ is pronounced exactly as [ɹ]; some are pronounced as rounded
[ɹʷ] (see below). Not every /ɑ/ is pronounced exactly as [ɑ]; some are pronounced as nasalized [ɑ̃]
(see below). In other words, phonology in detail will give us a very different picture from what we
get from just looking at the phoneme inventory, and to explain all the patterns that are found in
detail, we will normally need a large set of phonological rules.
That is the basic idea illustrated; let us now cover it in more abstract terms.
7.1 The Phonemic Principle
A widely adopted principle of phonology is the following:
(257) Phonemic Principle
Every language has a limited set of phonemes (= basic speech sounds); and every word in
the language consists solely of phonemes of that language.
As basic speech sounds, phonemes are assumed to be the elements of underlying representation.
By convention, when we listed phonemes (as opposed to phonetic segments in general), we
designate them as such by surrounding them with slant brackets (like /t/) rather than square
brackets (like [t]). Hence, it would be fairly normal to list a phoneme inventory with slant brackets,
as we have already seen for English
The phoneme set found in the speech of many Californians (and elsewhere in North America)
is as follows:
(258) The phonemes of one dialect of English
a. Consonants
/p, t, k, b, d, g, tʃ, dʒ, f, θ, s, ʃ, h, v, ð, z, ʒ, m, n, ŋ, l, ɹ, w, j/
/i, ɪ, e, ɛ, æ, u, ʊ, o, ɔ, ɑ, ə, ɚ, aɪ, aʊ, ɔɪ/

A system of phonemes can be construed as a complete set of building blocks for words in a
language. All the words of English (in the relevant dialect) are made up of the sounds in (258) and
no others. Thus an English speaker instantly recognizes that [ˈblk] (“blick”) could be English and
that [ˈq’ø] could not, even if she has never heard either word before. In phonological analysis,
we set up a phoneme inventory that is large enough to encompass the target language — but, as we
will see, no larger than necessary. Thus, for example, if an analyst somehow had tried to
phonemicize English but forgotten to include /ʊ/, then encountering a word like book [ˈbʊk] would
force the analyst to expand the inventory by one, since otherwise it will not suffice to cover the
language.
Languages vary in the number of phonemes they have. The record low is believed to be held
by Rotokas (South Pacific), with 11, and the record high is claimed to be held by !Xoo (Namibia),
with 160.185 The average across languages is about 30.
To see the point of the phonemic principle, you have to imagine a language that did not obey
it. In such a language: every word would have its own unique phonetic content, and would not be
decomposable into a sequence of units. Such a system might be rather like the vocal
communication systems of certain animal species, consisting of a fixed inventory of calls. No one
has yet encountered a language that works this way, and at this point no one expects to.
There is a clear advantage for a language in having a phonemic design. As noted earlier,
speech articulation is highly complex, with many articulators moving very rapidly. It would be
difficult to learn to pronounce all the thousands of words of a language if each one were a unique
phonetic sequence. Presumably it is easier to proceed phonemically; that is, to learn only a limited
number of sounds and form all the words by stringing these sounds together.
The sign languages of the deaf could, in principle, be suggested as a counterexample to the
Phonemic Principle. However, research on sign language suggests that even this form of language
can be analyzed into gestural “phonemes”, even though these phonemes are quite different from
the phonemes of spoken language.186
That the Phonemic Principle holds true of English is vividly confirmed by the behavior of
English speakers (particularly monolingual ones) in a multilingual environment. The English
speaker in such a context is likely to hear new words from speakers of other languages that contain
phonemes that are absent from English. His imitations of these words will systematically warp
them so as to consist solely of English phonemes. Thus French “r”, which is phonetically [ʁ] (an
approximant consonant made at the uvula), will be replaced by English [ɹ]. German [x] (the
voiceless velar fricative [x], spelled ch in Bach) will be replaced by English [k]. The four
phonemic tones of Mandarin, which correspond to nothing whatever that is phonemic in English,
185
This should be read with caution, as some scholars consider it the result of misanalysis: many of the
putative !Xoo consonant phonemes might actually be consonant clusters.
186
And more generally, such research has found morphemes, words, phrases, sentences, grammar,
intonation, etc. in sign languages; sign languages are a serious area of linguistics with a large research
program.
will be ignored completely.187 It is very hard for speakers to learn new words made up of sounds
not in their own phonemic inventories, which confirms the the phonemic inventory as a
phonological concept.
7.2 Allophones, phonemes, and rules
A second key principle, which will be the basis of much of the analyses to follow, is the
Allophonic Principle:
(259) Allophonic Principle
Phonemes vary; and the variation is rule-governed.
A allophone is a variant of phoneme.
Consider an example. We consider two variants of the English phoneme //—a distinction that
typically is not transcribed, but seems to be widely found.188 One variant is simply the plain palato-
alveolar central approximant [], already discussed. However, many instances of this phoneme are
pronounced as [],with simultaneous lip rounding. The superscript [] is the IPA symbol for
simultaneous rounding, which is also called labialization.
Inspecting my own speech (and checking with other speakers), I transcribed the following
data:
187
This is not to say that English speakers cannot control the pitch of their voices; they do this with exquisite
precision in rendering the intonation patterns of their language. But the idea of a pitch contour as an inherent property
of a word (i.e., as phonemic) is a tough one for English speakers to assimilate, and thus a serious part of the first-year
curriculum of instruction in Chinese (or any other tone language).
188
Reference: Daniel Jones (1918) An outline of English phonetics. For a study with physical
measurement, see Delattre, Pierre C., and Donald C. Freeman. 1968. “A Dialect Study of American r’s by X-
ray Motion Picture.” Linguistics 44: 29–68.
real [ˈil] par [ˈpr]

write [ˈat] core [ˈko]
rope [ˈop] ear [ˈi]
reveal [əˈvil] part [ˈprt]
arrange [əˈend] bear [ˈbɛ]
apparition [æpəˈʃən] lure [ˈlu]
carry [ˈkei] forming [ˈfom]
era [ˈeə] tornado [toˈneo]
Erie [ˈii]
What is needed with such data is an inspection that reveals the pattern, which is the most
central aspect of phonological analysis. If you don’t see the pattern yet, look some more before
you turn the page.
From the data, it is evident that
[] occurs only before a vowel

[] occurs elsewhere; that is, before a consonant or pause
This is in one sense a mechanical difference: it’s an authentic detail of English pronunciation, and
if you don’t respect it your English will sound unnatural. But the difference between [] and [ɹ]
has no communicative value, the way the difference between (for example) [t] and [p] has. [t] vs.
[p] is “communicative” because tin and pin are not the same word; because mat and map are not
the same word, because still and spill are not the same word, and so on — you could not say the
same thing for [] and [ɹ].
The idea, then, is that at some abstract level, [] and [] belong to the same category — they
are predictable variants of the same underlying sound. A diagram suggesting this idea is given
below.
// phonemic (underlying) level
[] [] phonetic (= allophonic, surface) level
This fundamental sound designated as // is a phoneme of English.
To express this idea with rules, we can adapt the basic model of underlying forms and
derivations that we have already set up (for the purpose of relating phonology to syntax and
morphology), and apply this model to the description of phonemes. We will place the phonemes at
the level of underlying representations, and derive the allophones as part of the surface
representation, using the same kinds of rules we already were using. As before, we assume that the
underlying representations are what is stored in the lexicon, and the surface (phonetic)
representations are what is submitted to the speech articulators for pronunciation.
As noted above, phonemes are normally placed in slant brackets to distinguish them from
ordinary phonetic transcription, in particular to designate them as the elements of underlying
representation. This carries over to phonemic analysis; thus, we speak of “the phoneme /ɹ/”. The
allophones, the contextual variants that are derived from the underlying phonemes, are placed in
ordinary square brackets, as in [], [].
Here is what we have so far for theory, stated in one place.

(260) Some key definitions for phonemic analysis
a. Phoneme
Assume a particular language L.
A phoneme is one member of a limited set of sounds in L which may occur in the
underlying representations of L, and which constitute the units of which words in L are
phonologically formed.
b. Phoneme inventory
The phoneme inventory of L is its complete set of phonemes.
c. Allophone
Assume a particular language L, and a phoneme of L to be called P.
An allophone of P is a variant of a phoneme that occurs in the surface representations of L

and is derived from P by the system of phonological rules.
These definitions make no reference at all to spelling. It is just as possible to analyze the
phonemes of an unwritten language as it is to analyze the phonemes of a written language. There
will, of course, be some connection between the phonemes and the spelled letters, but
understanding the rules that govern this connection is a different research topic.189 The written
language can actually be quite deceptive with regard to the phonemes, since often spelling is quite
inconsistent (consider e.g. beat vs. beet, both with the phoneme /i/).
8. Treating phonemes with rules and derivations
Let us return to the distribution of the sounds [] and [] in English:
[] occurs only before a vowel

[] occurs elsewhere; that is, before a consonant or pause
Proceeding to the formal analysis, we might suppose that the underlying form is //, and that [] is
derived from // by rule. Alternatively, we suppose that the underlying form is //, and that [] is
derived from // by rule. Often, both analyses can be made to work, but the choice that is
generally made is to pick the underlying form on the basis of which choice will produce the
simplest rules. In the present case, choosing // would require us to set up two phonological rules;
one to turn // to [] before a consonant (as in part [ˈpt]), and another to turn // to [] at the
end of a word (as in par [p]. The simpler choice is to pick // as the underlying form, a choice
189
In fact, it is a topic only rarely treated by phonologists, even though it is often interesting and intricate.
There is something of a research gap here.
which permits us to us just a single rule turning // into [] before a vowel. In sum, under this
approach the underlying representation (or “phonemic representation”) for a phoneme is the single
sound from which all the contextual allophones can be derived, using the simplest rules.
Since phonemes are the building blocks of morphemes in this theory, they lead to a definition
for these larger units:
(261) Defn.: Phonemic representation of morphemes
The phonemic representation of a morpheme is the string of phonemes from which it is

formed.
For example, under the analysis proposed, the phonemic representation of real (phonetically
[ˈil]) is /ˈil/. The phonemic representation of par (phonetically [ˈp]) is /ˈp/.
From the phonemic representation we derive a phonetic representation, the linguistic

characterization of the actual pronunciation of a word. In the case we are considering, there is just
one phonological rule, which can be stated as follows:
(262) // Rounding
 → [+round] / ___ [+syllabic]
In words, “If the sound  occurs in the environment before a [+syllabic] sound, change its features
so that it is [+round].”190
Given a phonemic form and one or more rules, we can apply the rules in a derivation of the
kind given earlier for examples Russ Schuh (syntax interaction) or orbital (morphology
interaction). The derivation derives the allophones from the phonemes. Here are derivations for
real, par, and part:
real par part

/il/ /pɑɹ/ /pɑɹt/ underlying representations
 — — // Rounding
[il] [pɑɹ] [pɑɹt] surface representations
In this type of phonemic analysis, the underlying representation could be thought of as an abstract,
idealized version of the pronunciation, embodying only the essential aspects, and the surface
representation is what one obtains after filling in all the detail through the application of rules.
190
The rule is expressed casually, using a phonetic symbol instead of a feature matrix. It could be formalized as
 +liquid 
+palatoalveolar → [+round] / ___ [+syllabic]
Note that including “vacuous” derivations for words like par and part is not a waste of time,
but helpful in the cause of rigor — we need to show that the rule properly avoids applying where it
should not. It is assumed for phonology that all words are submitted to all rules, like objects
passing down an assembly line. In this respect, phonology is like inflectional morphology,
discussed in Chapter 2. It is not like word formation, where rules apply freely, and optionally,
whenever they can.
8.1 Some terminology for allophone types
Very often, when there are two allophones, one of them is the one that we would sensibly
derived by a particular rule. For the case of the /ɹ/ phoneme, this is [ɹʷ], which occurs in the special
environment of before a vowel. Such an allophone is sometimes called a contextual allophone.
The other allophone, which is just what we get when the rule is not applicable, is usually called the
elsewhere allophone. In the /ɹ/ phoneme of English, [ɹ] is the elsewhere allophone. The
terminology comes from ordinary prose descriptions of allophone distribution; we say things like
“[ɹʷ] is a contextual allophone, occuring before vowels; whereas [ɹ] occurs elsewhere.” Indeed, the
use of the word “elsewhere” often makes such descriptions far easier and more intelligible.
Both contextual allophones and elsewhere allophones are allophones. It is tempting to use
different usage, say, calling [ɹ] “the phoneme” and [ɹʷ] “the allophone”. But this is nonstandard
and I will follow standard usage here.
The concepts of contextual vs. elsewhere allophone do have connections to analysis. In

analyzing, we seek maximally simple accounts. This will normally be available if we select the
elsehw
8.2 “Elsewhere” and selection of underlying forms
When one chooses an underlying form for a phoneme, the elsewhere allophone (provided one
can be identified) is usually the sensible choice to make, for it will lead to the simplest rule set.
Thus, in our /ɹ/ example, we can get a simply analysis, with just one rule needed, to derive [ɹʷ].
Had we made the opposite choice, we would have needed two rules; one like this:
ɹʷ → [−round] at the end of word
and the other like this:
ɹʷ → [−round] / ___ C
8.3 Rules and derivations for optional rules
When an allophone rule applies optionally, usually the segment derived in the environment of
the rule will be treated as a contextual allophone. To illustrate, we return to the data for /æ/
Diphthongization given earlier:
No diphthongization: Diphthongization:
cat [ˈkæt] can [ˈkæn], [ˈkɛən]

pack [ˈpæk] man [ˈmæn], [ˈmɛən]
lap [ˈlæp] Spanish [ˈspænʃ], [ˈspɛənʃ]
lab [ˈlæb] dance [ˈdæns], [ˈdɛəns]
pal [ˈpæl] spam [ˈspæm], [ˈspɛəm]
tram [ˈtɹæm], [ˈtɛəm]
The elsewhere allophone must be [æ]: it is quite easy to derive the contextual allophone [ɛə]
by the /æ/ Diphthongization rule already given in §5.2.2, but it would be quite a mess to do it the
other way (an obligatory rule ɛə → æ before all non-nasal consonants, and an optional ɛə → æ
rule before nasal ones).191 So we set up /æ/ as the underlying form of the phoneme, keeping /æ/
Diphthongization as before (repeated below).
/æ/ Diphthongization
/æ/ → [ɛə] / ___ [+nasal] (optional)
With these assumptions, we can now do the illustrative derivations, including both an
environment where the rule optionally applies and one where it does not.
can cat
/kæn/ /kæt/ underlying representations
kɛən — — /æ/ Diphthongization (optional)

[kɛən] [kæn] [kæt] surface representations
8.4 Moe on the justification for obligatory rules
/ɹ/ Rounding (262) is stated as an obligatory rule (by our conventions, if it were optional we
would have to say so). The basis for obligatory rules is purely distributional; you can’t actually
“see” the rule in effect as you can with optional rules. Yet the justification for the rule is just as
strong. The rule, being obligatory, enforces a kind of “phonological grammaticality” on words.
Specially, if you say a word with the wrong allophone for one of the phonemes, typically it will
sound strange — phonetically not right. I find this to be true with the words above. The following
forms reflect the outcome if one “neglects to apply” /ɹ/ Rounding:
real *[ˈil]
write *[ˈat]
191
In fact, the argument is stronger: there are marginal words like baa [bæ] ‘sound that a sheep makes’
demonstrating that [æ] is the allophone that occurs when word-final.
rope *[ˈop]
My own judgment is that these don’t really sound right as English (intuitively: “there’s not
enough like an r”, “a lazy r”).
The opposite type of “wrong allophone” is given below; as it were we “applying /ɹ/ Rounding
where it should not be”:
par *[ˈp]
core *[ˈkɔ]
ear *[ˈi]
To me at least, these forms sound peculiar (intuitively: “adding a w where it doesn’t belong”).
In this sense, a phonemic analysis is a partial theory of what is “sayable” in a language. For a
word to sound right, it must (a) be composed of solely phonemes from the language; (b) properly
submit to all applicable obligatory phonological rules. In English, *[ˈil] is ungrammatical because
the speaker has neglected to apply // Rounding where it should be applied; *[ˈp] is
ungrammatical because [ɹʷ] is neither permissible as a basic phoneme (it is not in the inventory)
nor can it be derived by any legitimate rule.192
8.5 The large-scale descriptive goal of a phonemic analysis
With patience and experience, a trained phonetician can detect hundreds of phonetically
distinguishable sounds in a language. To deal with them all in our discussions would be chaotic,
tedious, and error prone. Once a full phonemicization is in place, the phonetic system is reduced to
order: instead of hundreds of sounds to deal with, there are no more than a few dozen. Phonemic
analysis is thus considered a basic method for bringing order to phonological systems.
9. Practical uses of phonemic analysis
Here are practical uses of phonemes.
First, linguists sometime write reference grammars, intended to be a thorough account of the
structure of a language, covering phonetics, phonology, morphology, syntax, and semantics. Often
the first few pages of a reference grammar give the examples in full IPA transcription, setting forth
a phonemic analysis with its phoneme inventory. Once this is done, all future examples can be
given in phonemic transcription. It is assumed that the reader can apply the allophone rules to any
such transcription to get the desired pronunciation. This eliminates unnecessary detail from the
transcriptions and makes them easier to read.193 You can easily imagine how un-useful (indeed,
192
“Sayability” in phonology also involves legal phoneme orders (phonotactics), mentioned above.
193
Indeed, most reference grammars go one step further and produce a practical orthography; a
spelling system that follows the phonemic principle but uses only alphabetic letters.
annoying) it would be if a reference grammar of English always specified the difference between
[ɹ] and [ɹw].194
Dictionaries usually use a phonemic transcription in their specifications of pronunciation.

This avoids redundancy and clutter. There is often a further advantage: speakers of different
dialects of the dictionary’s language often have the same phonemes but different systems of
allophones. Such speakers can make use of the same phonemic transcription but render it
(probably, without even thinking about it) in their own accent. The reason this (usually) works is
that language change most often takes place at the allophonic level: the detailed allophones get
shifted, while the underlying phonemic categories persist unaltered. So dialects inherit much of
their phonemic system from their ancestral forms, while altering the detailed rendition.
Phonemic analysis is also important in alphabet design. A sensible alphabet will have a
separate symbol for each phoneme of the language, and no other symbols. This makes it possible
for the spelling to specify, in principle, all aspects of the pronunciation of a word, without
including any additional redundant information. The rules will suffice to fill in all the allophonic
details. Alphabet design is a continuing activity worldwide as ever more languages are provided
with writing systems.
10. Phonemic analysis and features
Everything we said earlier about features and natural classes (see Chapter 10, section 14.2)
carries over to the rules used in phonemic analysis: many rules apply to more than one segment, or
apply in environments that include more than one segment, or involve a parallel phonetic change;
and all of these can suitably be expressed using features. As noted earlier, the scheme is: (a) on the
left side of the arrow, we set up a group of features to single out the class of sounds that undergo
the rule; (b) on the right side of the arrow, we specify all and only the features that change their
value. The result is a kind of parallel shift of whole classes of sounds.
Here is a characteristic examples of an allophonic rules of English that applies to natural

classes. As a rough approximation, we can say that the basic phonemes /p t k/ get an aspiration (a
little puff of breath, [+aspirated]) when they occur in word-initial position.
(263) Initial Aspiration
+stop 
−voice → [+aspirated] / [word ___
E
“Voiceless stops become aspirated word-initially.”
Here are representative underlying forms and derivations:
194
This sounds like I am making a joke, but old 19th century reference grammars can be found in libraries that
do exactly this.
pit spit top stop kick Scot

/pɪt/ /spɪt/ /tɑp/ /stɑp/ /kɪk/ /skɑt/ underlying representation
p — t — k — Initial Aspiration
[pɪt] [spɪt] [tɑp] [stɑp] [kɪk] [skɑt] surface representation
Note the parallel shift, /p t k/ → [p tʰ kʰ]. As noted earlier, the assumption made in the theory
is that only the features specified in the rule are changed in the form. Thus /p/ starts out [+bilabial]
and [−voice], and ends up with these features because nothing has changed them (and similarly for
all of the features of /p/, see features chart (185) on p. 409).
Here is another example:
(264) Vowel Nasalization
[+syllabic] → [+nasal] / ___ [+nasal]
“A vowel becomes nasalized when it precedes a nasal sound.”
Derivations:
bun bud doom do sing sit

/bən/ /bəd/ /dum/ /du/ /sɪŋ/ /sɪt/ underlying representation
ʌ̃ — ũ — ɪ̃ — Vowel Nasalization
[bə̃n] [bəd] [dũm] [du] [sɪ ̃ŋ] [sɪt] surface representation
Examples are given here for three vowels only, but all the others would work the same.
Study Exercise #74
Demonstrate that Vowel Nasalization can apply to /ɚ/, using a close pair similar to bun / bud.
Include a derivation in the same format as above.
The case I could find are burn/bird, turn/turd, kern/curd, CERN/surd, Hearn/herd, stern/stirred,
spurn/spurred, kernel/curdle, and (in some dialects) earn/erred. If you don’t insist on near-
identity, there are many more. Derivations for burn/bird:
burn bird
/bɚn/ /bɚd/ underlying representation
ɚ̃ — Vowel Nasalization
[bɚ̃n] [bɚd] surface representation
____________________________________________________________________________
11. Rule ordering
We have seen earlier that different forms of linguistic rules differ in whether they “care” about
the order in which they apply. For instance, rules of inflectional morphology generally obey a very
strict order, which imposes a particular linear ordering on the prefixes and suffixes they attach. In
contrast, both word formation rules and phrase structure rules are “opportunistic,” applying freely,
or not, if they are applicable.
Phonological rules turn out to be like inflectional rules: it often matters what order they are
applied in, and in such cases the order is a strict, fixed one. Thus “ordering statements” must
therefore form part of the phonologies of human languages.
To develop our argument for ordering, we will need two phonological rules of American
English. Our first rule is based on the following data. / ̆/ is the IPA diacritic meaning “extra short”.
cat [kæ̆t] cad [ˈkæd]

pot [pɑ̆t] pod [ˈpd] Pa [ˈpd]
cop [kɑ̆p] cob [ˈkb]
cup [kʌ̆p] cub [ˈkb]
batch [bæ̆tʃ] badge [ˈbædʒ]
pasta [ˈpɑ̆stə] Mazda [ˈmɑzdə]
beat [ˈbĭt] bead [ˈbid] bee [ˈbi]
Bruce [ˈbŭs] bruise [ˈbuz] brew [ˈbɹu]
Rico [ˈɹĭko] Rio [ˈɹio]
spooky [ˈspŭki] boo [ˈbu] buoy [ˈbui]
goat [ˈgŏʊ̆t] goad [ˈgod] go [ˈgo] Goa [ˈgoə]
sake [ˈsĕɪ ̆k] sage [ˈsedʒ] say [ˈse] seance [ˈseɑns]
Normal-length and extra-short vowels in English are allophones of the same phoneme. In the
data above, extra-short vowels occur before voiceless consonants, and regular length vowels occur
before voiced consonants, before vowels, and finally. Thus the longer versions of the vowels are
the elsewhere allophones (as defined above; section 4), and we should set them up as the
underlying representations.
We also need a feature to write the rule with; for present purposes we can simply add the
feature [short].
With these assumptions, then, the rule of Vowel Shortening would be as follows.
Vowel Shortening
[+syllabic] → [+short] / ___ [−voice]
“Realize a vowel as extra short before a voiceless sound.”
Here are derivations:
Bruce bruise brew buoy

/ˈbus/ /ˈbuz/ /ˈbu/ /ˈbui/ underlying representation
ŭ — — — Vowel Shortening
[ˈbŭs] [ˈbuz] [ˈbu] [ˈbui] surface representation
The other rule we will need is the rule of Tapping already seen under (253) above and
repeated below. We will ignore here its (marginally) optional status.
Tapping
+alveolar +syllabic
+stop  →
E  / [+syllabic] ___ –stress  E
Here, it will be useful to use a fully formalized version of the rule, using features instead of
the symbol []. We need to know, then, just what features must be changed in order to turn both /t/
and /d/ into [].
First of all, a tap is voiced, so that the rule should add [+voiced] on the right side of the arrow.
This will correctly voice /t/, and it will do no harm for /d/. Tap also differs from the alveolar stops
in manner of articulation, being a liquid tap and not a stop. Using the features set up above, the
rule must be:
(265) Tapping (in features)
+alveolar
+voiced
–stop  +syllabic
+stop  →
E
+tap  E / [+syllabic] ___ –stress  E
+liquid 
With these two rules in hand, we can now see how they might interact. The crucial facts are:
• Tapping changes the voicing of a /t/.

• Vowel Shortening depends on the voicing of the following consonant.
Because of this, we will get different outputs depending on which order we apply the rules in.
@Now let us find some underlying representations that will help us detect the correct
ordering. For clarity, I will use underlying representations that are themselves derived
morphologically, since in these cases, the underlying form is crystal clear: in patting (roughly:
[ˈpæɾɪŋ]), the [ɾ] necessarily is derived from /t/, since we can hear the /t/ unaltered in pat. Likewise,
in padding, the [ɾ] necessarily is derived from /d/, since we can hear the /d/ unaltered in pad.
Here are derivations of these forms, using both orderings:
A. Tapping precedes Vowel Shortening
patting padding
/ˈpæt/ /ˈpæd/ underlying representations
ˈpæɾɪŋ ˈpæɾɪ Tapping
— — Vowel Shortening
[ˈpæɾɪ] [ˈpæ] surface representations
B. Vowel Shortening precedes Tapping
patting padding
/ˈpæt/ /ˈpæd/ underlying representations
æ̆ — Vowel Shortening
ɾ ɾ Tapping
[ˈpæ̆ɾɪ] [ˈpæ] surface representations
The predictions that the derivations make are clear: if Tapping precedes Vowel Shortening,
then patting and padding should be pronounced identically. If Vowel Shortening precedes
Tapping, then patting and padding should be pronounced differently; that is, padding should have
the longer vowel. The two words will be distinct (but in their vowels, not their taps). These
observations should hold true not just for these two words, but for all the words in which both rules
can apply (e.g. latter-ladder, writer-rider, Patty-Paddy, etc.).
What are the facts? There is actually no single outcome. Instead, different dialects of
American English use different orderings. Speakers from Michigan, Illinois, and Wisconsin
typically order Vowel Shortening before Tapping; thus they pronounce pairs like patting-padding
differently, with the length difference as shown in B above. Speakers from other areas tend to have
the opposite ordering, and the pronounce such pairs identically as in A above.
The essential point is not the details of the two dialects, but the very fact that they have the
same rules, yet differ. This implies that when one learns a language, and hence its phonology, part
of what one learns is ordering restrictions that must be imposed on its phonological rules.
Depending on what dialect of English you speak, you implicitly learned a particular ordering for
two of the phonological rules.
11.1 Analyzing rule order
To establish the ordering of two rules A and B, the simplest procedure is simply to find a
relevant form — a form where A and B are both applicable — and try both orders. Either you will
find that only the order A-B produces the right output, only B-A produces the right output. All
that’s really needed to do this test is to match up the rules with the forms with care, so you know
that you’ve found exactly what the rule predicts.
A slightly less mechanical skill is to explain clearly in words what you’ve found about
ordering. Here is an example description, for the example in the preceding section: “In the dialect
where patting is [ˈpæ̆ɾɪ] and padding is [ˈpæ], Vowel Shortening must be applied before
Tapping. The reason is that Vowel Shortening depends on the phonemic value of [voice] for the
following consonant, before that value is converted to [+voice] by Tapping.”
Here is a description of the ordering argument for the other dialect: “In the dialect where both
patting and padding are both pronounced [ˈpæ], Tapping must be applied before Vowel
Shortening. If we applied Tapping first, it would “see” the underlying /t/ of patting and wrongly
shorten the vowel.” This description is of the “counterfactual” type, which tells us what would go
wrong if we ordered the rules differently.
11.2 “Indifferent” cases
In a great number of cases, two rules simply don’t interact, and either ordering is compatible
with an adequate grammar. For instance, in pin, underlying /pɪn/, it simply doesn’t matter whether
we apply Initial Aspiration first and Vowel Nasalization second, or Vowel Nasalization first and
Initial Aspiration second; we will get [pʰɪ ̃n] no matter what. The necessity of ordering only arises
if one rule somehow is capable (in principle) to make a difference in the applicability of the other
rule.
12. Discovering phonemes with distributional analysis
I have so far given a fair number of allophones, together with rules for deriving them from
their underlying phonemes. All examples have been from English, and the analysis is given as a
sort of finish product. But consider a more difficult and realistic case: suppose we have only just
begun research on a particular language, never before studied. All we have to work with at this
earlier stage is a large notebook (or digital file), consisting of words taken down in IPA
transcription. How can we analyze such a data set to arrive at the right set of phonemes and
phonological rules?
There is a fairly standard and venerable technique (it dates from about 1940) for performing
this task. It has no standard name but might be called “distributional analysis.” The technique does
not even require that we know the structure of the words or what they mean — all that we need is
that knowledge of when two utterances are different words or the same word.
Distributional analysis for phonemes has two parts.
12.1 Location of minimal pairs
Remember what we said about phonemes earlier, as the “Phonemic Principle” (257): Every
language has a limited set of phonemes (= basic speech sounds); and every word in the language
consists solely of phonemes of that language. Now that we are including allophones in the system,
we must consider this a little more abstractly: the limited set of phonemes is actually a set of
abstract entities, from which the pronounced allophones are derived by rule. This leads us to a
more sophisticated characterization of the phoneme.
(266) Defn.: phoneme
The inventory of phonemes for a language is the smallest set of abstract sounds from
which all the (physically pronounced) sounds can be derived by rule.
So, for example, we set up /ɹ/ as the entity that underlies both surface [ɹ] and surface [ɹw],
along with a rule to derive the latter allophone. An adequate full phoneme inventory for English
would include enough phonemes to permit us to derive everything.
The definition in (266) immediately leads to one of the two principal techniques for figuring
out a phonemic system. Here is the rationale. If I give you the following set of paired words:
pin [pɪn] tin [tɪn]

pail [pel] tail [tel]
append [əˈpɛnd] attend [əˈtɛnd]
cap [kæp] cat [kæt]
aspen [ˈæspən] Aston [ˈæstən]
then I have firmly ruled out the possibility that [p] is an allophone of /t/ (or similarly that [t] is an
allophone of /p/). The reason is that there can be no environment for the claimed rules that derive
these allophones — no such environment could exist, given that they occur in exactly the same
locations in plainly different pairs of words. These locations are:
for pin and tin: [ ___ ɪn]

for pail and tail: [ ___ el]
for append and attend: [əˈ ___ ɛnd]
for cap and cat: [kæ ___ ]
for aspen and Aston: [ˈæs ___ ən]
Therefore, pairs like pin and tin are extremely informative about the phonemic system. Such
pairs are called minimal pairs.
(267) Defn.: Minimal pair
Two words form a minimal pair if they differ in just one sound, in the same location.
All the pairs given above are minimal pairs. Pin [pɪn] and Tim [tɪm] are not a minimal pair for /t/
and /p/ because they differ in more than one sound. Spin [spɪn] and pins [pɪnz] are not a minimal
pair for /s/ and /z/ because the [s] and [z] occur in different locations.
Linguists tend to be fond of minimal pairs; indeed long ago a linguist wrote “minimal pairs are
the analyst’s delight”. They instantly clarify a distinction; in the present case they show that two
sounds are separate phonemes. If you have a minimal pair, anywhere in linguistics,195 then you
know you have two structurally different things, and you know where the difference lies.
Even better than minimal pairs are minimal triplets, minimal quadruplets, minimal n-tuplets;
the more the better. A set like pin, tin, chin, kin, bin, gin, din, fin, thin, sin, shin, Zinn, Lynn, win
already establishes the phonemic status of a majority of English consonants.
12.2 Establishing complementary distribution
The other well-known method for figuring out a phonemic system is to locate pairs of sounds
that are in complementary distribution, defined as follows.
(268) Defn. Complementary distribution
Two sounds A and B are in complementary distribution if B never occurs in the

environments where A occurs, and vice versa.
Complementary distribution is a pattern that strongly suggests membership in the same

phoneme.196 It implies that there is an important phonological generalizaton present, which ought
to be accounted for by our system of rules. Usually, when A is in complementary distribution with
B, it is sensible either to derive surface [A] from underlying /B/ or derive surface [B] from
195
The minimal pair method is widely used in phonemicization, but in fact it is an important method of analysis
throughout linguistics. Thus, we have already seen minimal pairs in morphology (Turkish eli ‘hand-accusative’ / ele
‘hand-dative’), in syntax (“Fred stole/killed the chicken from Greeley”), and in semantics (“Alice congratulated
her/herself”). Throughout, the method used is to compare utterances that have just one single difference, in order to
learn the contribution made by that difference.
196
There are exceptions (complementary distribution, but separate phonemes), which you would have
to learn about in a more advanced treatment of phonology. See For Further Reading at the end of Chapter 12.
underlying /A/.197 In the discussions above, we established complementary distribution for various
sets of sounds: [ɹ] and [ɹʷ], short and normal-length vowels, oral and nasalized vowels, aspirated
and plain voiceless stops.
Sometimes, complementary distribution can be established just by staring at the data for a
while and letting the answer come to you. On the other hand, if you get stuck, there is a simple
procedure you can follow that increases your chances of success. This method has no official
name, but I will call it the method of local environments here.
Let us look at two sounds of English. The regular [l] we have already defined, as a lateral
approximant. The so-called dark l is transcribed [ɫ] (l with a tilde through it). It is made by
pushing the tongue body upward and backward at the same time the tongue blade makes the
appropriate movement for the l. I list below a bunch of words that have either the normal “light” [l]
or dark [ɫ]. Here are some data.
Words with light [l] Words with dark [ɫ]

lay [le] fall [fɑɫ]
lose [luz] tell [tɛɫ]
please [pliz] else [ɛɫs]
allow [əˈlaʊ] filter [fɪɫtɚ]
gremlin [ˈgɹɛmlən] milk [ˈmɪɫk]
freely [ˈfrili] apple [ˈæpəɫ]
In the method of local environments, you write down a dash, and before it whatever comes
before the target sound, and after it whatever comes after the target sound. When the sound comes
initially, we can use [ to mark a “left word boundary” and when it comes finally we can use ] to
mark a right word boundary. So, the local environments for the data just given are as follows:

lay [ ___ e fall ɑ ___ ]
lose [ ___ u tell ɛ ___ ]
please p ___ i else ɛ ___ s
allow ə ___ aʊ filter ɪ ___ t
gremlin m ___ ə milk ɪ ___ k]
freely i ___ i apple ə ___ ]
Looking at the data in this way often suffices to reveal the crucial environment. If not, you can
break the data down even further. Look just at the left side, like this:

lay [ ___ fall ɑ ___
197
The exception is when (for example) A is in complementary distribution with both B and C, but B and C
belong to separate phonemes. We must then evaluate a number of different analyses; one usually emerges as much
simpler.
lose [ ___ tell ɛ ___

please p ___ else ɛ ___
allow ə ___ filter ɪ ___
gremlin m ___ milk ɪ ___
freely i ___ apple ə ___
and also look just at the right side, like this:

lay ___ e fall ___ ]
lose ___ u tell ___ ]
please ___ i else ___ s
allow ___ aʊ filter ___ t
gremlin ___ ə milk ___ k]
freely ___ i apple ___ ]
In this case, the “right side only” display of the data yields the answer very clearly: light [l]
comes before a vowel. Nothing else appears to work. (In particular, we cannot use “after a vowel”,
since both [l] and [ɫ] occur after vowels; consider freely and allow.)
Once you have a description, the procedure is as before: select the elsewhere allophone as the
underlying representation, since it will generally make possible the simplest rule or set of rules. In
this case, we clearly should choose /ɫ/, since we can apply a single rule of Lightening to get the [l]
allophone. (Choosing /l/ is perhaps more intuitive, but would require two rules of Darkening, one
applying before a consonant, one at the ends of words. The diacritics of the IPA are a human-
created convention, and are not necessarily a good guide to the choice of underlying forms.)
The full analysis works as follows. For the description of the sounds, I employ the feature
[back]: dark [ɫ] is assumed to be [+back] and regular [l] is assumed to be [−back].
Phoneme: /ɫ/
Rule:
/l/ Lightening
ɫ → [−back] / ___ [+syllabic]
‘/l/ becomes light when it precedes a vowel’
Derivations:
lay please freely fall else

/ɫe/ /pɫiz/ /fɹiɫi/ /fɑɫ/ /ɛɫs/ underlying representations
l l l — — /l/ Lightening
[le] [pliz] [fɹili] [fɑɫ] [ɛɫs] surface representations
Summing up, the method of local environments is often an effective way to establish the
environments of particular allophones, and it worth trying if you get stuck.
Here are some limitations of the local-environment method. First, every once in a while, you
get a nonlocal environment, for instance a vowel allophone that depends on the vowel of the next
syllable. Second, sometimes the step of looking at the left side and the right side separately fails,
because the environment involves both sides. Most often, a two-sided environment is between
vowels (“intervocalic”), as in English Tapping.
12.3 Near-minimal pairs
From time to time, the language does not “cooperate” with the linguist in providing minimal
pairs for particular pairs of sounds. This can happen for several reasons. Sometimes, the words of
the language are long, so they are less likely to differ in just one place. Sometimes, minimal pairs
are lacking for sounds that happen to be rare in the language — as an example, you might try
finding English minimal pairs for /ʒ/ vs. [ð].198 Lastly, sometimes sounds tend (but only tend) to
occur in rather different environments; for instance: English [ʒ] mostly occurs after a stressed
vowels and before a stressless one (as in vision [ˈvɪʒən]), while [h] occurs initially or before a
stressed vowel (Horatio [həˈɹeʃo], apprehend [æpɹiˈhɛnd]). This makes it difficult or impossible to
find minimal pairs for the two.
In such cases, we often will end up concluding that, despite the lack of minimal pairs, the
sounds in question really are two separate phonemes. This is because the true criterion is not
whether a minimal pair exists, but rather whether a phonological rule predicting the difference is
feasible (a minimal pair instantly suffices to show that no such rule could exist). When we are
unable to derive a distinction between sounds by rule, we have no choice but to place the two
sounds in underlying representation, classifying them as separate phonemes.
In order to show that no rule could exist, the backup strategy normally followed is to find a set
of near-minimal pairs for the target sounds. These consist of pairs of words in which the two
target sounds occur in very similar, though not identical, environments. If there are enough such
pairs, it becomes plain that there could be no workable phonological rule to derive the distinction.
To give an example, consider a dialect of English that complete lacks minimal pairs for [ʒ] and
[ð] (i.e., where none of the possibilities in fn. 198 exist). Even in such a dialect, there could be a fair
number of near-minimal pairs for the distinction, which will differ from speaker to speaker. The list
in (269) is from my own speech; other speakers may pronounce some of the words differently.
198
Here are possibilities: leather [ˈlɛðɚ] vs. leisure [ˈlɛʒɚ], loathe [ˈloð] vs. loge [ˈloʒ], seethe [ˈsið] vs. [ˈsiʒ],
pleasure [ˈplɛʒɚ] vs. pleather [ˈplɛðɚ] ‘type of artificial leather made from plastic’. Not all speakers know these words,
or pronounce them as minimal pairs; so it seems pretty likely that at least some English speakers exist who have no
minimal pairs for [ð] and [ʒ].
(269) A set of near-minimal pairs for [ʒ] vs. [ð]
Words with [ʒ] Words with [ð]
adhesion [ædˈhiʒən] heathen [ˈhiðən]

azure [ˈæʒɚ] blather [ˈblæðɚ]
leisure [ˈliʒɚ] either [ˈiðɚ]
measure [ˈmɛʒɚ] nether [ˈnɛðɚ]
pleasure [ˈplɛʒɚ] feather [ˈfɛðɚ]
liege [ˈliʒ] breathe [ˈbɹið]
genre [ˈʒɑnɚ] though [ˈðo]199
The basis of the argument from near-minimal pairs becomes clear if you set yourself the
(hopeless) task of writing a rule to derive [ð] from /ʒ/, or [ʒ] from /ð/. Why should the far-away /m/
in measure induce a [ʒ], whereas the far-away /n/ in the similar word nether induces [ð]? Why
should a far-away /ɹ/ induce an [ð] in breathe whereas an /l/ in the same position induces a [ʒ] in
liege? There is no rhyme or reason to the data and thus there is no basis for a rule. The right
conclusion is to set up /ʒ/ and /ð/ as separate phonemes, and not try to posit “rules” that have no
validity.
12.4 Optional rules and the local-environments method
The existence of optional rules implies a slight change in how we determine the system of
phonemes: we need to look not just for cases of complementary distribution (defined above in
(202)) but also for cases of free variation. Free variation occurs whenever you have this situation:
in some particular context, wherever X occurs, so can Y, and vice versa. Thus in the English
example given earlier, in the context / ___ [+nasal], wherever [æ] can occur, so can [ɛə], and vice
versa. Two sounds occurring in free variation are treated as allophones of a single phoneme, and
the only difference is that the rule deriving the contextual allophone is optional.
The method of local environments can be adapted for free variation. The key is to make
separate columns for each variation pattern. Thus, for instance, if you were working on the data
for the rule of /æ/ Diphthongization (254) (p. 434), you would make a column headed “[æ] or
[ɛə]”, like this:
199
This last pair is not that close to minimal, I admit, but it seems significant: the frequency of this word has
risen tremendously since about 1960 (Google N-Gram Viewer), and by now it seems to be the first word of English to
have a generally agreed upon pronunciation with initial [ʒ]. Words with initial [ð] are mostly grammatical words; they
are individually frequent but collectively few in number.
[æ] or [ɛə] [æ] only

can / k ___ n cat / k ___ t
man / m ___ n pack / p ___ k
Spanish / p ___ n lap / l ___ p
dance / d ___ n lab / l ___ b
spam / p ___ m pal / p ___ l
tram / ɹ ___ m lass / l ___ s
From this, it would be straightforward to detect that the contextual allophone is [ɛə], and to
formulate /æ/ Diphthongization as applying before nasals.
One slight complication that arises is the need to limit oneself to systematic free variation.
There are plenty of words that have more than one pronunciation, but the two pronunciations are
phonemically different. Thus, in my own speech I find it perfectly acceptable to pronounce the
word envelope as either [ˈɛnvəlop] or [ˈɑnvəlop] — but this is an idiosyncracy of this particular
word, not a general pattern of free variation. Numerous minimal pairs like ken [ˈkɛn] vs. con
[ˈkɑn] already suffice to establish that /ɛ/ and /ɑ/ are phonemically distinct.
12.5 A full “flow chart” for distributional phonemic analysis
Integrating the method for free variation, we can describe the traditional method for
distributional phonemic analysis as a “flow chart” of options, as given below.
(270) A flow chart for phonemic analysis
START
Consider two
phonetically similar
sounds [x] and [y]
Are there any minimal

pairs (see (203)), or a
yes persuasive set of near- no
minimal pairs (sec.
12.3) for [x] and [y]?
Check for free

/x/ and /y/ belong to variation:
separate phonemes. yes Whenever [x] occurs, no
END is it also possible say
[y], and vice versa?
Check for
complementary
yes distribution:
wherever [x] occurs,
[y] does not, and vice
versa.
[x] and [y] are
allophones of the same
phoneme. Pick one no
(simplest choice) as the
underlying form and
write a rule or rules.
END You’re stuck.
See footnote.200
END
200
This is unlikely to occur, but I could imagine it happening. The scenario are that phonemes [x] and
[y] are so rare and/or so irregularly distributed that even near-minimal pairs are hard to find.
Study Exercise #75
Data from English. The target sounds are [s] and [t͡s]. Are these one phoneme or two?
(a) Collect local environments.

(b) Give the appropriate underlying forms and rules.
(c) Give underlying representations and derivations for dance, concert, and false.
Hints:
• In your local environments, put the stress mark before the vowel; rather than before the
syllable as IPA requires. For instance, for the [s] in [kənˈsid], write / n ___ ˈi.
Sorting local environments: make a list for “just [s]”, a list for “[s] and [t͡s] in free variation,
and for “just [t͡s]”.
dance [ˈdænt͡s]
Clarence [ˈklɛɹənt͡s]
mince [ˈmɪnt͡s]
hence [ˈhɛnt͡s]
concert [ˈkɑnsɚt], [ˈkɑnt͡sɚt],
cancer [ˈkænsɚ], [ˈkænt͡sɚ]
cancel [ˈkænsəl], [ˈkænt͡səl]
cancellation [kænsəˈleʃen], [kænt͡səˈleʃen]
tonsil [ˈtɑnsəl], [ˈtɑnt͡səl]
fancy [ˈfænsi], [ˈfænt͡si]
insert [ɪnˈsɚt]
concede [kənˈsid]
coincide [koɪnˈsaɪd]
soup [ˈsup]
false [ˈfɑls]
farce [ˈfɑɹs]
miss [ˈmɪs]
fussy [ˈfəsi]
13. Answer to Study Exercise #74
Local environments:
[t͡s] only
dance / n ___ ]
Clarence / n ___]
mince /n ___]
hence / n ___]
Environments for either [s] or [t͡s] in free variation:

concert / n ___ ɚ
cancer / n ___ ɚ
cancel / n ___ ə
cancel / n ___ ə
tonsil / n ___ ə
fancy / n ___ i
Environments with only [s]:

insert / n ___ ˈɚ
concede / n ___ˈi
coincide / n ___ ˈaɪ
soup / [ ___ ˈu
false / l ___ ]
farce / ɹ ___ ]
miss / ˈɪ ___ ]
fussy / ˈə ___ i
[s] and [t͡s] are sometimes in complementary distribution, sometimes in free variation.
[t͡s] only: word-finally after [n].

Free variation: between [n] and a stressless vowel.
[s] only: elsewhere
We set up /s/ as the underlying form (it would be quite a mess to try to state all the
environments for [s], but it works fine as the elsewhere allophone).
Rules:
Obligatory /s/ Affrication
–fricative 
s → +affricate / n ___ ]word
E
“An s becomes [t͡s] if it comes between [n] and the end of a word.”
Optional /s/ Affrication
–fricative  +syllabic
s → +affricate / n ___ –stress 
E E (optional)
“An s may become [t͡s] if it comes between [n] and a stressless vowel.”
Derivations:
dance concert false

/ˈdæns/ /ˈkɑnsɚt/ /fɑls/ underlying representations
t͡s — — Obligatory /s/ Affrication
— ˈkɑnt͡sɚt — — Optional /s/ Affrication

[ˈdænt͡s] [ˈkɑnt͡sɚt] [kɑnsɚt] [fɑls] phonetic forms
14. Phonemic analysis on novel languages
Phonemic analysis of English has a kind of trivial quality to it if you are an English speaker —
we intuitively sense our own phonemic system, and the rules are just adding the details. But this is
an English-internal perspective. The surprises happen when you do the same basic procedure on
other languages.
In particular, the sounds are often organized in a way quite different from how English works.
In English, // is an independent phoneme, supported by multiple minimal pairs. In Spanish, [] is
normally analyzed as an allophone of the phoneme /d/: wherever the /d/ phoneme would appear
between two vowels, the phonetically-similar allophone [] occurs instead. In English, the
aspirated stops [p t k] are treated as allophones, respectively, of /p t k/; whereas in Korean,
Mandarin Chinese, and many other languages, /p t k/ and /p t k/ form distinct series of
phonemes, supported by numerous minimal pairs. So the important idea is to work out of each
language’s phonemic system in its own terms.
Below I give the two principal errors that can arise during phonemicization.
14.1 Underdifferentiation
Sometimes linguists do not set up enough phonemes, so that pairs of words that have distinct
pronunciations fail to have distinct phonemic content — there is a “lost distinction”. This is
sometimes called underdifferentiation. The usual reason for underdifferentiation is that the
linguist cannot hear the lost distinction. Thus, for instance, for years the sounds [] and [] of the
West African language Okpe went missed by linguists, who heard them mistakenly as the [e] and
[o], which also occur as Okpe phonemes.
To repair underdifferention errors, it helps to bring more linguists onto the scene —
especially, native speaker linguists, who have the great advantage of having heard the distinction
from birth. Acoustic phonetic measurements also sometimes can be helpful.
14.2 Overdifferentiation
Linguists occasionally set up too many phonemes. This is the result of insufficient analysis:
the linguist fails to notice that two sounds are in complementary distribution. In this kind of error,
a generalization is missed, and we have a failure to note that the distributions of the allophones are
predictable by rule. The error is sometimes called overdifferentiation; two sounds are treated as
separate phonemes when they should be treated as the same phoneme.
An example of phonemic overdifferention occurred around the 16th century when the
invading Spaniards first encountered Tagalog in the Philippines. They assumed that all five vowels
of Spanish /i, e, a, o, u/ were vowel phonemes of Tagalog — and spelled them with their own five
Roman letters. But in the Tagalog of the time, [e] and [o] were allophones of the phonemes /i/ and
/u/, and indeed in the alphabet the Tagalog speakers were already using, [e] and [o] were spelled
with the same letters used for [i] and [u] (phonemic writing). The Spaniards’ error reflects the
natural but naïve expectation that a new language you encounter will be maximally similar to your
own.
15. Phonemic analysis of Yidiɲ
Let us work out some phonemes in a more detailed example, using data from an Australian
aboriginal language, Yidi ([] is IPA for the palatal nasal, like Spanish “ñ”). Yidi is no longer
spoken, though there may be a few aborigines alive today who remember a few words. Fieldwork
on Yidi was carried out in the 1960’s and 70’s by Prof. Robert M. W. Dixon of the Australian
National University, who also developed the phonemic analysis given here. The data below are
somewhat idealized, constructed from Dixon’s lexicon following his description of the facts.
First, here are the non-English IPA symbols needed.
• [] is a voiced palatal stop —same place of articulation as [j], but full stop closure.
• [] is a palatal nasal
• [] indicates that the preceding vowel is long
• [ɫ] is dark l, with the feature [+back]. As noted above in section 12.2, this sound occurs as
an allophone in English, though the environment is not the same as in Yidi.
• [r] is a trilled r.
• [ɻ] is a retroflex central approximant, with tongue tip curled up and back.201
201
Some English speakers use this kind of r, rather than the (more common) /ɹ/.
The following are consonant and vowel charts for Yidi. These are not just a casual review—
consulting the chart is actually a good procedure to follow when you are discovering the rule
environments.
(271) Phonetic charts: sounds of Yidiɲ
a. Consonants
Bilabial Alveolar Retroflex Palatal Velar
Stops (voiced) b d  g
Nasals m n  
Liquids nonlateral r ɻ
Lateral l
Lateral [+back] ɫ
Glides w j
b. Vowels
Front Back Back

High tense i, i u, u
High lax , ɪː , ː
Low a, aː
Here are the 54 words we will be working with.
(272) Data for Yidiɲ problem
1. [ɟʊmbaːgɪ] ‘tobacco’
2. [ŋawuːjʊ] ‘salt-water turtle’
3. [guɫaːɻ] ‘big-leafed fig tree’
4. [ŋuɲʊːr] ‘initiated man’
5. [duguːbil] ‘bark bag’
6. [muɲɟʊːɻ] ‘plenty’
7. [wigilwigil] ‘sweet’
8. [ɟambuːɫ] ‘two’
9. [ɟʊɫŋuːɫ] ‘waterfall’
10. [gabuːɫ] ‘stick for carrying fish’
11. [wurguɫ] ‘pelican’
12. [babuːɟʊ] (can’t find gloss)

13. [guɫgɪ] ‘sand, sugar’
14. [maguːɫ] ‘a root vegetable’
15. [muɫɲaːrɪ] ‘blanket’
16. [ŋumbuːbʊ] ‘new-born baby’
17. [jʊjʊɻuŋguɫ] ‘noise of snake sliding through grass’
18. [ɟʊgaːbal] ‘house frame’
19. [ɟʊwaːr] ‘wattle tree’
20. [ɟʊduːɫʊ] ‘brown pigeon’
21. [duɫnbiːlaj] ‘white cedar’
22. [ɟimuːr] ‘large house’
23. [gunbuːɫ] ‘billy-can’
24. [guɫaːn] ‘walnut tree’
25. [ɲʊŋguːɫ] ‘Torres Straits pigeon’
26. [mugaːɻʊ] ‘fish net’
27. [wiɻuːɫ] ‘shellfish species’
28. [wuɫmbuːɻ] ‘leafy broom’
29. [ɟʊriːn] ‘leech’
30. [ɲʊnduːba] ‘you-nom. sing’
31. [gujʊ] ‘tree vine species’
32. [wuɫmbuːɻ] ‘leafy broom’
33. [ɟʊɫugunʊ] ‘black myrtle tree’
34. [buɫuːɻ] ‘storytime person’
35. [ŋuɻuːɫ] ‘just now’
36. [guguːɫʊ] ‘recitative mourning style used by men’
37. [wubuːɫ] ‘lucky’
38. [dalŋudalŋʊ] ‘sound of bell ringing’
39. [dalʊ] ‘forehead’
40. [ɟʊmaːl] ‘straight spear thrower’
41. [ɟilŋgʊ] ‘down’
42. [baŋgaːmʊ] ‘English potato’
43. [muɫaːrɪ] ‘initiated man’
44. [ganguːɫ] ‘grey wallaby’
45. [guɟʊːn] ‘wind’
46. [gaɟʊːɫ] ‘dirty (e.g. water)’
47. [buɫguːɻ] ‘swamp’

48. [ɟʊŋguːm] ‘worm’
49. [ɟʊɻɪ] ‘sharp, pointed’
50. [gawuːɫ] ‘blue gum tree’
51. [bawʊː] ‘backbone’
52. [gawʊː] ‘tree species’
53. [galbɪː] ‘catfish’
54. [diwɪː] ‘small ground bee’
In doing phonemicization with raw data like this, the overall strategy one follows is to
consider pairs or small groups of sounds that are phonetically similar, under the hypothesis that
they are allophones of the same phoneme. There is no reliable principle to be followed here other
than general phonetic similarity; one must make guesses, some of which pay off in the discovery
of allophonic relationships. Usually in introductory textbooks, the author decides to send you in a
direction that actually turns out to work when you try it, and that will be generally true here.
We can start in on Yidiɲ by considering the two sounds [l] and [ɫ], which are indeed
phonetically similar (light vs. dark l). Following the method of section 12.2 above, we collect local
environments for these sounds by looking up each on in the data, and recording (a) the example
number; (b) the preceding sound; (c) the following sound. Here is such a chart for [l]:
(273) Local environments for Yidiɲ [l]
5. i___] 38. a___ŋ

7. i___w 38. a___ŋ
7. i___] 39. a___ʊ
18. a___] 40. aː___]
21. iː___a 41. i___
The first local environment on chart (273) was obtained by taking the l-containing form
5. [duˈguːbil]
from (272) and replacing l with ___, as follows:
5. [duˈgubi___ ]
and then removing all the material not next to the l:
5. [duˈgubi___]
The resulting entry in (273), 5. i ___ ], means “an [l] occurred after [i] and at the end of a word.”
All the other entries in (273) were obtained the same way.
The same procedure yields this list for dark [ɫ]:
(274) Local environments for Yidiɲ dark [ɫ]ː
3. u___a 20. u___ 34. u___u

8. u___] 21. u___n 35. u___]
9. ___n 23. u___] 36. u___
10. u___] 24. u___a 37. u___]
11. u___] 25. u___] 43. u___a
13. u___g 27. u___] 44. u___]
14. u___] 28. u___m 46. ___]
15. u___ 32. u___m
47. u___g
17. u___] 33. ___u
50. u___]
These lists are then inspected for pattern. It’s useful to look first at “right sides” alone, then at
“left sides” alone, and remember the phonetic character of the sounds in involved. In the present
case, the payoff comes from looking at the “left side” environments for [ɫ], which, shown alone,
look like this:
(275) Inspecting the local environments for Yidiɲ [ɫ]: left sides only
3. u___ 20. u___ 34. u___

8. u___ 21. u___ 35. u___
9. ___ 23. u___ 36. u___
10. u___ 24. u___ 37. u___
11. u___ 25. u___ 43. u___
13. u___ 27. u___ 44. u___
14. u___ 28. u___ 46. ___
15. u___ 32. u___ 47. u___
17. u___ 33. ___ 50. u___
which in turn reduces to the following four cases:
u ___, u___, ___, ___
These four cases occupy a specific region of the vowel chart, repeated below:
Front Back Back

High lax  , 
Low a, a
This can be characterized very simply as the round vowels.202 Thus, using our feature set, “in the
environment, after a round vowel” is stated:
+syllabic
/ +round  ___ E
This is clearly a meaningful discovery; there are enough data that this pattern is very unlikely to be
true by accident.
The next thing to check is: how does the distribution of the phonetically similar light [l]
sound relate to this environment? Combing through the list of local environments (273), we find
+syllabic
that there are no cases of light [l] in the environment +round  ___. Thus, we have established
E
complementary distribution.
The phonemic analysis would work like this. [l] is the “elsewhere” allophone, since there is
nothing particular that defines its distribution, other than not matching the environment for dark
[]. Therefore we set up the elsewhere allophone as the phoneme /l/, and write the following rule:
(276) /l/ Darkening (Yidiɲ)
+syllabic
l → [+back] / +round  ___ E
“Realize the /l/ phoneme as back (dark; velarized) when it follows a round vowel.”
This rule can be illustrated with derivations of words chosen from the original data in (272).
To make the illustration clear, we pick one form that is eligible for the rule and one that isn’t:
(277) Derivations to illustrate Yidiɲ /l/ Darkening
10. /gaˈbul/ 5. /duˈgubil/ underlying representation

ɫ — /l/ Darkening
[gaˈbuɫ] [duˈgubil] surface representation
202
It’s true that these four vowels are also [+back] and [+high]; we’re going for a terse
characterization here (see p. 414) and there’s no point in using more features than necessary.
15.1 Some minimal pairs in Yidiɲ
In many languages (for instance, Italian and Swahili), long vowels are allophones of the their
short counterparts. This might be true of Yidi — in principle — but the following minimal pair
data show that we needn’t pursue this hypothesis very far:
[malan] ‘flat rock’

[malan] ‘right hand’
[wuɻu] ‘spear handle’

[wuɻu] ‘river’ or ‘snake species’203
[guil] ‘smell-present tense’

[guil] ‘smell-past tense’
Plainly, the long and short vowel pairs must be counted as separate phonemes.
15.2 More Yidi allophones
The local-environment method for detection of allophones is applied below to [u], [], [u],
and []; again the data we are working from are from (272).
203
Two unrelated meanings, like English bank.
(278) Environments for [] in Yidiɲ
1. ___m 48. ___ 20. ɫ___]

9. ___ɫ 49. ___ɻ 26. ɻ___]
33. n___]
12 ___] 30. ___n 36. ɫ___]
18. ___g 2. j___]
38. ___]
19. ___w 17. j___j
39. l___]
17. j___ɻ
20. ___d 31. j___]
41. g___]
29. ___r 16. b___]
42. m___]
33. ___ɫ
40. ___m
(279) Environments for [u] in Yidiɲ
11. w___r 26. m___g 36. g___g

13. g___ɫ 28. w___ɫ 37. w___b
15. m___ɫ 3. g___ɫ 38. ___d
16. ___m 31. g___j 4. ___ˈ
17. ɻ___ 32. w___ɫ
43. m___ɫ
33. ɫ___g
17. g___ɫ 45 g___
33. g___n
21. d___ɫ 47. b___ɫ
34. b___ɫ
23. g___n 5. d___g
24 g___ɫ 35. ___ɻ
6. m___
This one is a bit harder: you have to notice that there are two environments for []: after a
palatal consonant, and at the end of a word. The [u] cases occur in neither environment, so we have
a more complex complementary distribution. This is something that we will have to accept; i.e. the
idea that the same allophone might be derived by two different rules.
We set up the basic phoneme (“elsewhere”) /u/, and write two rules. Both turn out to be
generalizable when we look at further data, so these are preliminary versions.
(280) Yidiɲ Postpalatal Laxing (preliminary)
–syllabic
u → [−tense] / +palatal  ___
E
“/u/ becomes lax when it follows a palatal consonant.”
(281) Yidiɲ Final Laxing (preliminary)
u → [−tense] / ___ ]word

“/u/ becomes lax at the end of a word.”
A note on terminology: “lax” is the opposite of “tense” in phonology, hence the rule names with
“Laxing”.
Some derivations of three sample forms are as follows. Note that these forms have /l/’s as
well, which redundantly illustrate /l/ Velarization.
1. /ilgu/ 9. /uɫngul/ 4. /buluɻ/ underlying form

—  — Postpalatal Laxing
 — — Final Laxing
— ɫ ɫ /l/ Velarization
[ilg] [ɫnguɫ] [buɫuɻ] phonetic form
There is no need to order the rules in any particular way (any ordering works), and the ordering
given is an arbitrary choice.
We turn next to the long vowels [ʊː] and [uː].
(282) Environments for [] in Yidiɲ
4. ___r
45 ___n
46. ___ɫ
6. ___ɻ
51. w___]
52. w___]
(283) Environments for [u] in Yidiɲ
10. b___ɫ 25. g___ɫ 37. b___ɫ

11. g___ɫ 27. ɻ___ɫ 44. g___ɫ
12 b___ 28. b___ɻ 47. g___ɻ
14. g___ɫ 30. d___b 48. g___m
16. b___b 32. b___ɻ 5. g___b
2. w___j 34. ɫ___ɻ 50. w___ɫ
20. d___ɫ 35. ɻ___ɫ 8. b___ɫ
22. m___r 36. g___ɫ 9. g___ɫ
23. b___ɫ
It should be clear that the situation is parallel to what we saw with short [u] and []: the laxed
vowel occurs finally and after a palatal consonant, whereas the tense vowel occurs elsewhere.
There are fewer data here, but our confidence should be increased by the fact that we’ve seen the
pattern before.
The analysis needs to be revised, not replaced, to handle these data: evidently the rules of
Postpalatal Laxing and Final Laxing must apply to the class of vowels [ u, u ]. We need, therefore,
to be explicit about the vowel features we are using, and then use features to handle the data. Here
are vowel features for Yidi:
i [+high, −low, −back, −round, +tense, −long]

i [+high, −low, −back, −round, +tense, +long]
 [+high, −low, −back, −round, −tense, −long]
u [+high, −low, +back, +round, +tense, −long]

u [+high, −low, +back, +round, +tense, +long]
 [+high, −low, +back, +round, −tense, −long]
 [+high, −low, +back, +round, −tense, +long]
a [−high, +low, +back, −round, −tense, −long]

a [−high, +low, +back, −round, −tense, +long]
With these features, we can we restate the rules as follows, capturing natural classes:
(284) Yidiɲ Postpalatal Laxing (final version)
+syllabic –syllabic
+round  → [−tense] / +palatal  ___
E E
“Any rounded vowel becomes lax after a palatal consonant.”
(285) Yidiɲ Final Laxing (still preliminary)
+syllabic
+round  → [−tense] / ___ ]word
E
“Any rounded vowel becomes lax in final position.”
+syllabic
The designation +round  suffices, in a language like Yidi with a small vowel inventory, to
E
designated all and only the vowels of the set { u, u }. As before, the idea behind the rule is that it
changes only the feature [tense], with all other features remaining the same. As a result, /u/
becomes [] and /u/ becomes [], each retaining their value of the feature [long] — features not
specified by the rule are assumed to remain unaltered.
Note that in this kind of analysis, part of the goal is to achieve as much generality as you can.
In principle, you could describe the language with a great number of detailed rules, each applying
to one sound in one environment. But aiming for more general rules gives a clearer picture of the
overall pattern.
Study Exercise #76
Find three appropriate forms from the list above and illustrate the revised versions of these
rules as they apply to long vowels. Use the derivations given in (277) above as your model.
52. /bawuː/ 6. /muɲɟuːɻ/ 11. /wurguːl/ underlying representation

—  — Postpalatal Laxing
 — — Final Laxing
— — ɫ /l/ Velarization
[bawː] [muɲɟʊːɻ] [wurguːɫ] phonetic form
_____________________________________________________________________
We are almost done sorting the data. Here is how the high front vowels [i] and [ɪ] are
distributed.
(286) Environments for Yidiɲ [i] and []
[]:
1. g___]
13. g___]
15. r___]
43. r___]
49. ɻ___]
[i]:
22. ___m 7. w___g

27. w___ɻ 7. g___l
41. ___l 7. w___g
7. g___l
5. b___l
These data also suggest complementary distribution: all of the []’s are final and no [i]’s are
final. Knowledge of phonetics helps here: [] is the lax partner of [i] just as [] is the lax partner of
[u], suggesting that our Final Laxing rule should be generalized even further, to include the front
vowels. However, Postpalatal Laxing should not be generalized further, since as examples 22 and
41 show, we get [i], not [], after palatals.
If Final Laxing applies to long /u/, to short /u/, and to short /i/, then it have better apply to
long /i/ as well. Data are few, but apparently conform to the prediction:
(287) Environments for Yidiɲ [i], []
[] [i]
53. b___] 21. b___l

54. w___] 29. r___n
Let us go out on a limb, assuming that collection of further data would continue to confirm the
overall pattern. Thus we will complete the fully-generalized rule. We want it to apply, in final
position, to { u, u, i, i }, but not [a, a]. This can be done if we formulate it to affect only non-low
vowels:
(288) Yidiɲ Final Laxing (final version)
+syllabic
–low  → [−tense] / ___ ]word
E
“Non-low vowels are made lax in word-final position.”204
——————————————————————————————————————————
Study Exercise #77
Review the completed phonemic analysis of the Yidiɲ vowels and specify all the natural
classes it uses that have more than one member. Describe each natural class according to (a) the
rule that uses it; (b) a list of sounds in { }, (c) a description in IPA terminology.
204
[+high] would work as well as [–low], since Yidi has no mid vowels.
Postpalatal Laxing is triggered by the natural class of palatal consonants, which in Yidiɲ is { ,
, j }. Postpalatal Laxing applies to the class of round vowels, which in Yidiɲ is { u, u }. Final
Laxing applies to the natural class of nonlow vowels, which in Yidiɲ is { i, i, u, u }.
______________________________________________________________________________
15.3 The full Yidi phoneme inventory
We’ve now succeeded in showing that several of the sounds of the Yidi phonetic chart in
(271) above are not independent phonemes, but merely allophones. These are placed in
parentheses in the revised charts below:
(289) The sounds of Yidiɲ with allophones shown in parentheses
a. Consonants
Bilabial Alveolar Retroflex Palatal Velar
Stops (voiced) b d  g
Nasals m n  
Liquids nonlateral r ɻ
Lateral l
Lateral velarized (ɫ)
Glides w j
b. Vowels
Front Back Back

High lax (), (ɪː) (), ()
Low a, a
This reduces the phoneme population to 19, which is a rather small phoneme inventory
compared to most languages.
16. When to use features in writing rules: normal descriptive practice
To repeat a point made earlier: the theory assumed here, in its pure form, includes no phonetic
symbols at all; rather, a speech sound is simply the matrix (in square brackets) of its feature values,
as for example in (245) above. So in a completely rigorous, theory-compliant world, the rules, the
representations, and the derivations would include no phonetic symbols at all. On the other hand,
there is a virtue of making our work reasonably terse and intelligible to others, and so often a
working phonological analysis is carried out with the occasional use of phonetic symbols. This is
done repeatedly above.
I feel that in semi-formal presentation, it is appropriate to use a mixed notation, using phonetic
symbols where they lead to no harm, and features where they contribute insight. Here are ways in
which rules benefit by writing them with features.
To capture a natural class. We’ve seen multiple examples of this: in A → B / ___ C, if

either A or C is a group of sounds describable with a set of features (natural class), it makes sense
to use the features for A or C.
To provide an insightful way of showing what features are changing. This relate to the
“→ B” expression in the rule schema A → B / ___ C. If, for example, if a rule changes [p t k] to
[b d g] one would want to express the change as “→ [+voice]” to capture the generalization of
voicing change across three places of articulation. Moreover, if we encountered a rule that changed
only /p/ to [b], leaving /t/ and /k/ unaltered, it would still make sense to express the change as “→
[+voice]” rather than “p → b”, because this characterizes the natural relatedness of [p] and [b] and
shows that the change induced by the rule is a minor one (one feature).
To capture an assimilation. An assimilation is a phonological process in which one sound

becomes more similar to a neighboring sound. In feature theory, assimilations show up when hteh
changing feature is that same as a feature present in the environment of the rule. It makes sense to
write the rule using features to make it clear that it is in fact an assimilation rule. For example, in
English, /k, g, / becomes fronted [k, g, ] before front vowels, as in keel [ˈkil], gale [ˈgel], or
dinghy [ˈdi]. This is an assimilation, which can be expressed by:
Velar Fronting
+syllabic
[+velar] → [−back] / ___ –back  E
“A velar consonant becomes fronted before a front vowel.”
When none of these factors is present, it seems sensible to use IPA symbols; this makes the
rule easier to read, and a reader equipped with your feature chart could probably work out a strict
featural version without much trouble if necessary.
For further reading

For works on phonology in general, see the end of the next chapter.
The discussion of Yidiɲ is based on an extensive reference grammar written by the primary
field worker for this language, R. M. W. Dixon: A Grammar of Yidiɲ, Cambridge University
Press, 1977.
Chapter 12: Alternation and Neutralization
1. Overview
This chapter will deal in more detail with how phonology interacts with other parts of the
grammar, focusing on two concepts that arise as a result of this interaction: alternation and
neutralization.
2. Alternation
Let us start with a definition:
(290) Defn.: alternation
Alternation is the appearance of a single morpheme in different phonetic forms in

different contexts.
Alternation is found in all languages of the world. It normally results from an interaction of
morphological and phonological rules. To show how alternation arises, we can consider an
example, which requires some background material on the morphology and phonology of
American English.
For morphology, we can briefly review a few rules of word formation rules, covered in
Chapter 2. In that chapter we worked out what we called the -able Rule, using English spelling.
We can now be a bit more realistic now that we have IPA; the rule now looks like this:
(291) -able Rule (updated from (52))
[ X ]Verb → [ [ X ]Verb əbəl ]Adj meaning: “able to be Verbed”
This derives, for instance, lovable and wearable from love and wear, respectively.
In Study Exercise #7 (Chapter 2) we established another word formation rule, the -ation Rule,
which, updated with IPA, looks like this:
(292) -ation Rule
[ X ]Verb → [ [ X ]Verb ˈeʃən ]Noun meaning: “process of Verbing”
This derives, for instance, accusation, improvisation, and indentation.
We will also use some phonological rules that interact with the morphological rules just given.
Of these, the following is new:
(293) Pre-Stress Aspiration
+stop  +syllabic 
–voice → [+aspirated] / [+syllabic] ___ +stressed
E E
This is part of family of rules assigning aspiration; see also Initial Aspiration in (263)
above.205 This one is needed to cover that cases of aspiration that occur other than at the beginning
of the word. Here are examples:
appeal [əˈpil]
attend [əˈtɛnd]
account [əˈkaʊnt]
In these examples the voiceless stop is between two syllabic sounds (vowel, diphthong, or syllabic
consonant), of which the second is stressed. Note further that when the second is not stressed, the
aspiration is absent (or at least quite weak):
caper [ˈkepɚ]
tickle [ˈtkəl]206
Another phonological rule we will need is given below:
(294) Preglottalization
+stop 
–voiced → [+glottal] / ___ ]word (optional)
E
This applies to the voiceless stops [p t k] and assigns them a second place of articulation, so that
[p], for instance, becomes [+bilabial, +glottal]. Examples include cap [ˈkæʔp], cat [ˈkæʔt], and lack
[ˈlæʔk]. In full detail, the glottal closure comes a little bit before the bilabial/alveolar/velar one;
hence the term “preglottalization” and the transcription with a preposed superscript glottal stop.
The remaining phonological rule we need has already been discussed but is repeated below for
convenience:
205
Not all phonologists have been happy with setting up two aspiration rules (one initial, the other pre-
stress), and various proposals have been made to unify them.
206
We can’t check /t/ here because it would undergo Tapping (253), which makes it not a stop at all
and hence ineligible for aspiration.
(295) Tapping
+alveolar
+voiced
+stop  →
E
+liquid 
This applies, almost obligatorily, in butter [ˈbəɾɚ], motto [ˈmɑɾo] , vanity [ˈvænəɾi], ready [ˈɹɛɾi],
and many other words.
Both Preglottalization and Tapping are optional, but for simplicity we will assume here that
they apply obligatorily. This simplifying assumption will not change the analysis in any crucial
way.
3. Alternation, and how it derives from morphology/phonology interaction
With all of these morphological and phonological rules in hand, we can now cover the crucial
data:
note notable notation

[ˈnot] [ˈnoəbəl] [noˈtheʃən]
quote quotable quotation

[ˈkwot] [ˈkwoəbl] [kwoˈtheʃən]
The first three forms are, or are derived from, the stem note and the last three from quote; the
relevant rules of word formation are the -able Rule and -ation Rule. If we cross out the phonetic
material that forms the affixes -able and -ation, then we can look at what is left over; that is, the
way that the stems appear in this context:

[ˈnot] [ˈnoəbəl] [noˈtheʃən]

[ˈkwot] [ˈkwoəbl] [kwoˈtheʃən]
or simply:

[ˈnot] [ˈno] [noˈth]

[ˈkwot] [ˈkwo] [kwoˈth]
These variant forms of the stems are called allomorphs, defined more carefully below:
(296) Definition: allomorph
When an morpheme X has more than one possible pronunciation in surface (phonetic)
representation, these different pronunciations are called allomorphs of X.
So, the morpheme that in spelling we denote as note has (at least) three allomorphs, [ˈnot], [ˈno],
and [noˈth].
The example also illustrates the the definition given above for “alternation” ((290):
“Alternation is the appearance of a single morpheme in different phonetic forms in different
contexts.”). Since both note and quote have more than one allomorph, they both alternate.
Most, but not all alternation, has a simple explanation, which we have already stated but is
worth repeating since it is crucial here.
Morphological rules precede phonological rules.
More specifically, as noted above the phonological component of grammar is assumed to occupy a
position in the grammar following the components for morphology. The crucial scenario that arises
is this: morphological rules, by adding prefixes, suffixes, etc., change the environments in which
the phonemes occur. Then, phonological rules sensitive to these environments apply differentially
to different instances of the same morpheme. This make the morphemes show up differently in
different contexts; that is to say, it causes the morphemes to alternate.
Here is a derivation showing how the scheme works for the words and rules given so far. It
will be a “bicomponential” derivation, with first word formation then phonology.
As noted above in chapter 2, section 16, rules of word formation apply freely; they represent a
choice made to derive a new word from an old one. Since this is essentially a form of optionality,
we can again use the branching derivation formalism to show the various possible routes:
[ˈnot]Verb stem
[[ˈnot]Verb əbəl]Adj -able Rule (291)
[[not]Verb eʃən ]Noun -ation Rule (292)
[[ˈnot]Verb əbəl]Adj [ˈnot]Verb [[(ˈ)not]Verb ˈeʃən ]Noun output of morphological

component
The resulting forms [[ˈnot]Verb əbəl]Adj, [ˈnot]Verb, and [[not]Verb ˈeʃən ]Noun are submitted to the
phonology, in order to convert the abstract schemata of phonemes to an overt, pronounceable
string of sounds. There are reasons to think that the bracketed structure of the morphology is
retained in the phonological component, but since this is not necessary here, and it is helpful to
keep the representations maximally legible, I will discard the brackets. The phonological
component thus starts with:
ˈnotəbəl ˈnot noˈteʃən207
These forms are indeed the phonological underlying representations for these words, and so it
it appropriate to surround them with slant brackets, like this:
/ˈnotəbəl/ /ˈnot/ /noˈteʃən/ Underlying representations
The representations are “underlying” for purposes of phonology, where they form the most
abstract level of representations, though they are output representations from the viewpoint of
morphology.
As before, the phonological derivation consists of applying the rules in order. In this
derivation the phonological rule ordering does not matter, and the order shown was arbitrary
chosen.
/ˈnotəbəl/ /ˈnot/ /noˈteʃən/ underlying representations

— — t Pre-Stress Aspiration (293)
 — — Tapping (295)
— t — Preglottalization (294)
[ˈnoəbəl] [ˈnot] [noˈteʃən] phonetic representations
We have now produced an explanation for alternation: the -able Rule placed the /t/ of /not/ in
an environment where Tapping could apply to it; the -ation Rule placed the /t/ of /not/ in an
environment where Pre-Stress Aspiration could apply to it; and the lack of any morphological
affixation left the /t/ in word-final position, where Preglottalization could apply to it. The end
result is three allomorphs, [no], [not], and [not].
This pattern is very general in languages: morphology changes environments, phonology

“sees” these environments and accommodates the phonemes to them by assigning the appropriate
allophone. The theoretical conclusion is that, at least in the normal case, morphology precedes
phonology.208
207
There is an additional change here, removing the stress on the stem before the stressed
suffix -ation. This can be done by rule ([+syllabic] → [−stress] / ___ X [+stress]), but we’ll not deal with this
here.
208
It is certainly a consensus among linguists that at least some phonology follows morphology.
Linguists have also experimented with theories in which some phonological rules are premorphological, some
postmorphological; we won’t try to cover such theories here in a first course.
4. The ordering of syntax and phonology
As already discussed, phonology follows not just morphology, but also syntax: syntactic rules
string words together, resulting in changed environments and differential applicability of
phonology — therefore, alternation. The example we covered earlier, Alveolar Fricative
Palatalization, demonstrates the ordering of phonology after syntax. To give a fully worked-out
example, we can derive a sentence like We will use Schuh’s book. Syntactic rules apply first,
creating an ordinary tree:
/wi wɪl juz ʃuz bʊk/
The tree, with its words (shown here both in orthography and in IPA), is then handed off to the
phonology, which “sees” the underlying sequence /zʃ/ in Schuh’s book and applies Alveolar
Fricative Palatalization (along with other rules, perhaps), yielding something like this:
[wiwɪljuʒʃuzbʊk]. A consequence of this is that the word use alternates: it has the normal
allomorph [juz] in isolation, and the allomorph [juʒ] when affected by Alveolar Fricative
Palatalization.
Here is another example. English has not one but two Tapping rules. The original Tapping
rule from (253) is repeated here with examples:
Tapping
+alveolar
+voiced
+stop  →
E
+liquid 
utter /ˈətɚ/ → [ˈɚ]
batting /ˈbætɪŋ/ → [ˈbæɾɪŋ]
It is crucial that the second vowel in the context be [−stress], otherwise we would get
aspiration instead of Tapping. However, there is a particular situation in English where we get
Tapping even when the second vowel is stressed; namely, when the second vowel is in a separate
word.
Here are Tapping examples across word boundary, shown here with the brackets ]w [
Phonemic Phonetic
at Ed [ æt ]w [ ˈɛd ] [ æ ]w [ ˈɛd ]
get Alice [ gɛt ]w [ ˈæls ]w [ gɛɾ ]w [ ˈæls ]w
not Adam [ nt ]w [ ˈædəm ]w [ n ]w [ ˈæɾəm ]w
To handle these facts, we need to adopt an additional Tapping rule, which could be written
like this:
(297) Phrasal Tapping
+alveolar
+voiced
–stop 
+stop  → E
+tap  E / [+syllabic] ___ ]w [+syllabic]

+liquid 
In words, this says “make /t/ a tap when it is immediately preceded by a vowel and immediately
followed by a vowel which is in the next word.” Note that as with ordinary Tapping, the rule
applies to the natural class of alveolar stops, as see in examples like They paid Adam
[ … peɾ]w [ ˈæɾəm ]w
Phrasal Tapping is a phonological rule that can be used to demonstrate the ordering between
syntactic and phonological rules. For syntax, we will use the following phrase structure rules taken
from (146) above.
PP → P NP
Art 
NP → NP   (AP)* N (PP)*(CP)
 
A A E
(plus: lexical insertion)
Consider now the pronunciation of the PP at Ed. If the syntactic rules apply first, then we will
derive the correct output as follows:
SYNTACTIC COMPONENT:
PP PP → P NP
Art 
NP → NP   (AP)* N (PP) (CP)
 
A A E
NP
|
P N
PP Lexical insertion
NP
|
P N
| |
æt ˈɛd
PHONOLOGICAL COMPONENT:
/[ æt ]word [ ˈɛd ]/ underlying representation (word boundaries made explicit)
[ æɾ ]word [ ˈɛd ] Phrasal Tapping (297)
[æɾˈɛd] surface representation
It is easy to see that, as before, phonology must apply after syntax, else we would never be
able to apply Phrasal Tapping at all.
What about phonological rules in general? Certainly it is very common for phonological rules
to be sensitive to phrasal environments, so at the very least we can say that some phonological
rules are postsyntactic. Linguists differ on the question of whether there exists in addition a class
of presyntactic phonological rules.
5. Neutralization
An important possibility that often arises in phonology is that some rule might make distinct
underlying forms show up as phonetically identical. This scenario is sufficiently important that it
has a name:
(298) Defn.: neutralization
Neutralization is the phonetically identical realization of distinct phonemic forms.
The rule of Tapping is, at least in many dialects, a neutralization rule. The following data
demonstrate that Tapping apply to /d/ as well as to /t/ (plus sign is a notation for morpheme break):
Cases of Tapping with /t/ Cases of Tapping with /d/

write /ˈat/ [ˈat] ride /ˈad/ [ˈad]
writer /ˈat+ɚ/ [ˈaɚ] rider /ˈad+ɚ/ [ˈaɚ]
white /ˈwat/ [ˈwat] wide /ˈwad/ [ˈwad]

whiter /ˈwat+ɚ/ [ˈwaɚ] wider /ˈwad+ɚ/ [ˈwaɚ]
pat /ˈpæt/ [ˈpæt] pad /ˈpæd/ [ˈpæd]

patting /ˈpæt/ [ˈpæɾɪŋ] padding /ˈpæd+/ [ˈpæɾɪŋ]
Reformalizing the rule with features, it now reads:
The neutralization produced by Tapping takes the form given in (299):
(299) Neutralization diagram for Tapping
/t/ /d/
[]
Over the course of this text, we have seen several sources of ambiguity in language, arising in
the various components of the grammars. These are found in the morphological component
(specifically, word formation), as in (undoable, p. 51); in syntax (They saw the man with the
telescope, p. 83) and in semantic interpretation, as in Many arrows hit every target, (p. 365).
Phonological neutralization is yet another source of ambiguity. In the dialect under description
here, the listener hearing [ˈaɚ] must infer from context — or simply guess — whether the
speaker meant /ˈɹaɪtɚ/ writer or /ˈɹaɪdɚ/ rider. Usually, context suffices, but in my own experience
the particular ambiguity [ˈkɪɾi] = kitty/kiddie really does seem to create confusion — many
households have both kitties and kiddies in them! The neutralization diagram for this word pair is
given below.
kitty kiddie
/ˈkɪt+i/ /ˈkɪd+i/
[ˈkɪi]
The case of Tapping is somewhat unusual in that two phonemes are realized identically by
converting them into an allophone that happens to be different from either of them. More typically,
the neutralized output is identical to one or the other phoneme. Here is an instance; consider the
following data:
We live in Pittsburgh [m ptsbɚg]

Wouldja hand me the phone book? [fom bk]
I gotta make a phone call. [foŋ kɔl]
It’s all a con game. [k gem]
If one says these casually enough, the /n/’s at the end of phone, in, and con turn into either [m]
or []. The patterning is as follows:
n → m / ___ p
n → m / ___ b
n →  / ___ k
n →  / ___ g
Informally, we can write the rule as follows:
Nasal Place Assimilation

n → [same place] / ___ [−syllabic]
where “same place” is shorthand for changing all of the features for place of articulation to match
those of the following sound.
Nasal Place Assimilation is clearly a neutralizing rule; it neutralizes the difference between /n/
and /m/ in some cases, and between /n/ and // in others. For example, the following sentence is
ambiguous:
“They were [s glæsəz]”
The readings are the sensible “They were sunglasses”, and the phonetically literal but nonsensical
“They were sung glasses”.
(300) Neutralization diagram for “sunglasses” and “sung glasses”
/sn glæsəz/ /s glæsəz/ phonemic forms
[s glæsəz] phonetic forms
Study Exercise #78
Look at chapter 10, section 13.2. Justify this claim: “Alveolar Fricative Palatalization is
neutralizing”. Make reference to the definition of neutralization in (298) and provide a
neutralization diagram analogous to (300).

“Neutralization is the phonetically identical realization of distinct phonemic forms. Russ
Schuh /ɹəs ʃu/ and Rush Schuh /ɹəʃ ʃu/ are distinct phonemic forms (as shown by the distinct
pronunciations of Russ [ɹəs] and rush [ɹəʃ]). They are realized identically when Russ Schuh
undergoes Alveolar Fricative Palatalization and becomes [ɹəʃ ʃu], just like Rush Schuh. The
neutralization diagram is:
/ɹəs ʃu/ /ɹəʃ ʃu/ phonemic forms
[ɹəʃ ʃu] phonetic forms
6. Methods for solving problems with alternation
As with phonemicization, it helps to have some practical methods of analysis in hand when
you take on problems of alternation in novel languages. As before, the tricky part is first to “run
the derivations backward” when you don’t actually know the rules yet. This is actually more
feasible than it might sound at first.
As an example, let us consider some data from Servigliano, a Romance language spoken in
the Marche region of Italy. This dialect has numerous vowel alternations in its paradigms. Here are
some data; note that different paradigms are mixed together here.
(301) Phonology of vowels in Servigliano — some paradigms
a. Verb agreement
[ˈkɾedo] ‘I believe’ [ˈkɾidu] ‘you believe’
[ˈmetto] ‘I put’ [ˈmittu] ‘you put’
[ˈmɔro] ‘I die’ [ˈmoru] ‘you die’
b. Number in one variety of masculine nouns

[ˈfjore] ‘flower-masc. sg.’ [ˈfjuri] ‘flower-masc. pl.’
[ˈpotʃe] ‘flea-masc. sg.’ [ˈputʃi] ‘flea-masc. plur.’
[ˈpatre] ‘father-masc. sg.’ [ˈpatri] ‘father-masc. plur.’
[ˈmure] ‘cloth-masc. sg.’ [ˈmuri] ‘cloth-masc. plur’
c. Gender in adjectives and nouns

[ˈpesa] ‘heavy-fem. sg.’ [ˈpisu] ‘heavy-masc. sg.’
[ʃiˈfosa] ‘picky eater-fem. sg.’ [ʃiˈfusu] ‘picky eater-masc. sg.
[ˈloŋga] ‘long-fem. sg.’ [ˈlungu] ‘long-masc. sg.’
[tʃiˈlɛstra] ‘heavenly-fem. sg.’ [tʃiˈlestru] ‘heavenly-masc. sg.’
[ˈsgwɛtsa] ‘suspicious-fem. sg.’ [ˈsgwetsu] ‘suspicious-masc. sg.’
[ˈʃʃwɛrta] ‘strange-fem. sg.’ [ˈʃʃwertu] ‘strange-masc. sg.’
[ˈsprɔta] ‘pedantic-fem. sg.’ [ˈsprotu] ‘pedantic-masc. sg.’
[ˈmɔʃa] ‘dejected-fem. sg.’ [ˈmoʃu] ‘dejected-masc. sg.’
[aˈmika] ‘friend-fem. sg.’ [aˈmiku] ‘friend-masc. sg.’

[ˈpattsa] ‘crazy-fem. sg.’ [ˈpattsu] ‘crazy-masc. sg.’
Before going on, note that there are two types of masculine noun in Servigliano which form
their singulars differently. This is because Servigliano (like closely related Italian) has “declension
classes”; arbitrary set of nouns that inflect in different ways. In a formal analysis this can be
treated by using an arbitrary inflectional feature like [DeclensionClass], borne arbitrarily by noun
stems and referenced by the rules of inflectional morphology.
Returning now to Servigliano phonology, the first step in solving an alternation problem is to
break up the words into their morphemes. You sometimes have to be a little bit brave about this
when the morphemes alternate. We have already done some cases (Chapter 2) of breaking up
words into morphemes when there are no alternations, and the right strategy seems to be to find
sequences of sounds that are invariant whenever a particular gloss (meaning) is present. When
there are alternations, we have to be more permissive, being willing to assume that two sequences
are the surface versions of the same underlying morpheme even when they are somewhat different.
For Servigliano, division of words into morphemes is not difficult: it should be clear that the
inflectional suffixes in the words of (301) all consist of a single vowel, and that the stems consist
of everything else in the word. The alternating material is a vowel internal to the stems. Therefore,
we can form a list of suffixes, noting that none of them alternate.
[-o] first singular present agreement for verbs

[-u] first singular present agreement for verbs
[-e] singular ending for a particular type of masculine nouns

[-i] plural ending for this type of masculine nouns
[-u] singular ending for another type of nouns or adjectives, in the masculine gender
[-a] singular ending for this type of nouns or adjectives, in the feminine gender
Were it our focus to cover the morphological system of Servigliano, we could express the
inflectional categories in features and write simple rules to attach these suffixes when the relevant
features are present in the morphosyntactic representation. But let us skip this step and move on
with phonology.
The next step is to collect full sets of allomorphs of all the morphemes. This is already done
for the suffixes, which do not alternate. For the stems, we collect the allomorphs as below; the tilde
symbol ~ is often used to mean “alternates with”:
(302) Stem allomorphs present in the data of (301)

[ˈkɾed] ~ [ˈkɾid] ‘believe’
[ˈmett] ~ [ˈmitt] ‘put’
[ˈmɔr] ~ [ˈmor] ‘you die’
[ˈfjor] ~ [ˈfjur] ‘flower’
[ˈpotʃ] ~ [ˈputʃ] ‘flea’
[ˈpatr] ‘father’ (does not alternate)

[ˈmur] ‘cloth’ (does not alternate)
[ˈpes] ~ [ˈpis] ‘heavy’
[ʃiˈfos] ~ [ʃiˈfus] ‘picky eater’
[ˈloŋg] ~ [ˈlung] ‘long’
[tʃiˈlɛstr] ~ [tʃiˈlestr] ‘heavenly’
[ˈsgwɛts] ~ [ˈsgwets] ‘suspicious’
[ˈʃʃwɛrt] ~ˈʃʃwert] ‘strange’
[ˈsprɔt] ~ [ˈsprot] ‘pedantic’
[ˈmɔʃ] ~ [ˈmoʃ] ‘dejected’
[aˈmik] ‘friend’ (does not alternate)
[ˈpatts] ‘crazy’ (does not alternate)
Once we have done this, the next step is to extract the phonological essence of the alternating
pairs: find the segments that alternate. In the data above, it turns out that there are only four
alternations; we use the tilde symbol again to depict alternation.
(303) Alternating segments present in the data of (301)

a. [e] ~ [i]
[ˈkɾed] ~ [ˈkɾid] ‘believe’
[ˈmett] ~ [ˈmitt] ‘put’
[ˈpes] ~ [ˈpis] ‘heavy’
b. [o] ~ [u]
[ˈfjor] ~ [ˈfjur] ‘flower’
[ˈpotʃ] ~ [ˈputʃ] ‘flea’
[ʃiˈfos] ~ [ʃiˈfus] ‘picky eater’
[ˈloŋg] ~ [ˈlung] ‘long’
c. [ɛ] ~ [e]
[tʃiˈlɛstr] ~ [tʃiˈlestr] ‘heavenly’
[ˈsgwɛts] ~ [ˈsgwets] ‘suspicious’
[ˈʃʃwɛrt] ~ [ˈʃʃwert] ‘strange’
d. [ɔ] ~ [o]
[ˈsprɔt] ~ [ˈsprot] ‘pedantic’
[ˈmɔʃ] ~ [ˈmoʃ] ‘dejected’
It is also worth recording the three patterns of non-alternation:
(304) [a], [i], [u] do not alternate

[ˈpatr] ‘father’
[ˈmur] ‘cloth’
[aˈmik] ‘friend’
[ˈpatts] ‘crazy’
It is usually sensible in the case of non-alternators to give them the simplest possible underlying
representation, respecting the principle “what you hear is what you get”: non-alternating [a] is
underlying /a/; non-alternating [i] and [u] are likewise /i/ and /u/.209
The existing alternations can be interpreted more carefully, by plotting them on a chart
containing the full seven-vowel inventory of Servigliano using arrows. I will also circle the non-
alternating vowels and include the feature values for vowel height that I am assuming.
(305) The pattern of vowel alternation in Servigliano, with features
i u [+high, −low, +tense]
e o [−high, −low, +tense]
ɛ ɔ [−high, −low, −tense]
a [−high, +low, −tense]
We can see that if a vowel alternates, it alternates with an immediately higher or lower vowel in
the chart.
To figure out the rest of the problem, we assume that there is something about the
environment of the alternating segments, which is our analytic task to find. Plainly, this
environment must be located in the suffixes, which provide the only environment that could
differentiate the two derivations. I suggest that at this point you look again at the data in (301),
asking the question, “What causes the higher of the two alternating vowels to appear?” The
answer appears following this page break.
209
The exception is if the phonemic analysis, carried out earlier, has given reason to assign these sounds the
underlying represention corresponding to an elsewhere allophone; see above.
Whenever two vowels in Servigliano alternate, the higher of the two vowels appears when the
next vowel in the word (the suffix vowel) is high. This makes intuitive sense; i.e. the highest sort
of vowels cause their preceding neighbors to become higher; an instance of assimilation. Three
suffixes that have high vowels are the plural suffix [-i], the 2nd sg. suffix [-i], and the masculine
singular suffix [-u].
Now that we have located the alternating segments, the next step is to explore alternatives
for the underlying representation. In particular, when [A] alternates with [B], it makes sense that
either /A/ is the underlying representation, with [B] derived from it, or vice versa, with [A] derived
from underlying /B/. So, if [i] alternates with [e], as in Servigliano, we must consider both
possibilities: set up underlying /e/ for the alternating forms, and derive [i] from it; or set up
underlying /i/ for the alternating forms, and derive [e] from it. The same holds true for the other
three alternations, [u] ~ [o], [e ~ ɛ], and [o ~ ɔ].
Let us pick one of these hypotheses and see if we can make it work. In particular, suppose that
whenever two vowels alternate, it is the lower of the two that is underlying. Thus:
(306) One hypothesis about underlying representations in Servigliano

• [i] ~ [e] is underlain by /e/.
• [e] ~ [ɛ] is underlain by /ɛ/.
• [u] ~ [o] is underlain by /o/.
• [o] ~ [ɔ] is underlain by /ɔ/.
With this in place, we do the next step, which is to reconstruct the underlying
representations, and see if we can find rules that convert them into correct surface
representations. By “reconstruct”, I mean that when a morpheme has one of the alternations of
(306), one picks the relevant UR for that sound, plugging in to the UR for the morpheme as a
whole. Thus, since [ˈkɾed-o] ‘believe-1sg.’ / [ˈkɾid-u] ‘believe-2sg.’ displays the [e] ~ [i]
alternation, following (306) we would set up /ˈkred/ as the underlying form for ‘believe’.
Also by “reconstruct”, I mean that all words are re-formed to consist of the concatenation of
the UR’s of their constituent morphemes. Thus, since we are assuming that the underlying form for
the 2sg. ending is /-u/, we are therefore assuming that the underlying form for the word [ˈkɾid-u]
‘believe-2sg.’ is /ˈkred-u/.
Following this method, we can obtain a possibly-workable hypothesis for the underlying form
of every word, putting in good position to hunt for the rules. It can be helpful to do a set of “sketch
derivations”, juxtaposing underlying and surface forms, and listing (without necessarily yet
understanding) the changes that are needed.
For Servigliano, we will need 14 example forms: two each for all seven underlying vowels,
occurring before a high-voweled suffix (which by hypothesis triggers raising), and before a
phonologically-inactive non-high voweled suffix. We set these up as in (307). It is essential to
observe that for every morpheme, its underlying representation is exactly the same throughout the
paradigm; this follows from the fundamental principle of the theory that morphological rules apply
before phonological ones, concatenating phonemic representations.
(307) Outline of the needed derivations, assuming lower vowels as UR’s

‘believe-1sg’ ‘believe-2sg.’ ‘strange-f.sg.’ ‘strange-m.sg.’
/ˈkɾed-o/ /ˈkɾed-u/ /ˈʃʃwɛrt-a/ /ˈʃʃwɛrt-u/ Supposed underlying forms
— i — e Phonological Rules
[ˈkɾedo] [ˈkɾidu] [ˈʃʃwɛrta] [ˈʃʃwertu] Surface forms
‘flower-m.sg.’‘flower-m.pl.’ ‘die-1sg.’ ‘die-2sg.’

/ˈfjor-e/ /ˈfjor-i/ /ˈmɔr-o/ /ˈmɔr-u/ Supposed underlying forms
— u — o Phonological Rules
[ˈfjore] [ˈfjuri] [ˈmɔro] [ˈmoru] Surface forms
‘friend-f.sg.’ ‘friend-m.sg.’ ‘cloth-m.sg.’ ‘cloth-m.pl.’

/aˈmik-a/ /aˈmik-u/ /ˈmur-e/ /ˈmur-i/ Supposed underlying forms
— — — — Phonological Rules
[aˈmika] [aˈmiku] [ˈmure] [ˈmuri] Surface forms
‘father-m.sg.’ ‘father-m.pl.’
/ˈpatr-e/ /ˈpatr-i/ Supposed underlying forms
— — Phonological Rules
[ˈpatre] [ˈpatri] Surface forms
Seeing the derivations in outline like this, we can now reconstruct the rules: [o e] need to turn
into [i u] when a high vowel follows, and [ɛ ɔ] need to turn into [e o] when a high vowel follows.
This can be done with two ordered rules: first we raise the [e o] to [i u], getting them “out of the
way”, after which it is safe to raise [ɛ ɔ] to [e o]. To be more precise: the change from [ɛ ɔ] to [e o]
is actually tensing, given the features of (305) we are assuming. I express the rules thus, using the
Kleene star notation (explained on p. 157 above).
Servigliano Raising
+syllabic +syllabic
+tense  → [+high] / ___ C* +high 
“Tense vowels must become high if the next vowel in the word is high.”210
210
In principle, I could have included [−high] on the left side of the arrow, but it is not necessary; high vowels
can be vacuously rendered [+high] with no harm to the analysis.
Servigliano Tensing
+syllabic +syllabic
−low  → [+tense] / ___ C* +high 
“Nonlow vowels must become high if the next vowel in the word is high.”211
We test the rules out by applying them in order in a full set of derivations.
(308) Complete working derivations for Serviliano

‘believe-1sg’ ‘believe-2sg.’ ‘strange-f.sg.’ ‘strange-m.sg.’
/ˈkɾed-o/ /ˈkɾed-u/ /ˈʃʃwɛrt-a/ /ˈʃʃwɛrt-u/ Underlying forms
— i — — Raising
— — — e Tensing
[ˈkɾedo] [ˈkɾidu] [ˈʃʃwɛrta] [ˈʃʃwertu] Surface forms
‘flower-m.sg.’‘flower-m.pl.’ ‘die-1sg.’ ‘die-2sg.’

/ˈfjor-e/ /ˈfjor-i/ /ˈmɔr-o/ /ˈmɔr-u/ Underlying forms
— u — — Raising
— — — o Tensing
[ˈfjore] [ˈfjuri] [ˈmɔro] [ˈmoru] Surface forms
‘friend-f.sg.’ ‘friend-m.sg.’ ‘cloth-m.sg.’ ‘cloth-m.pl.’

/aˈmik-a/ /aˈmik-u/ /ˈmur-e/ /ˈmur-i/ Underlying forms
— — — — Raising
— — — — Tensing
[aˈmika] [aˈmiku] [ˈmure] [ˈmuri] Surface forms
‘father-m.sg.’ ‘father-m.pl.’
/ˈpatr-e/ /ˈpatr-i/ Underlying forms
— — Raising
— — Tensing
[ˈpatre] [ˈpatri] Surface forms
This works, and so the analysis is complete.
6.1 Testing the wrong analysis
There is, however, a loose end. Recall what was said earlier: “if [A] alternates with [B], it
makes sense that either /A/ is the underlying representation, with [B] derived from it, or vice versa,
211
As before, I could have put [−high] on the left side of the arrow, but since there is no harm in vacuously
reassigning [+tense] to high vowels, I kept the rule simpler by one feature.
with A derived from underlying /B/.” We have only tried one direction. What if we had tried the
other, with the higher vowels as underlying, and lowering rules.
(309) An alternative hypothesis about underlying representations in Servigliano

• [i] ~ [e] is underlain by /i/.
• [e] ~ [ɛ] is underlain by /e/.
• [u] ~ [o] is underlain by /u/.
• [o] ~ [ɔ] is underlain by /o/.
I suggest you take a minute to ponder why this would lead to failure.
The reason that a Lowering analysis would fail is that it cannot handle the stems that have [i]
throughout their paradigms — if some Lowering environment actually existed, why would these
[i] lower as well? Here is the essential comparison:
(310) Outline of the needed derivations, (wrongly) assuming higher vowels as UR’s
‘believe-1sg’ ‘believe-2sg.’ ‘friend-f.sg.’ ‘friend-m.sg.’
/ˈkɾid-o/ /ˈkɾid-u/ /aˈmik-a/ /aˈmik-u/ Underlying forms
e — — — Phonological Rules
[ˈkɾedo] [ˈkɾidu] [aˈmika] [aˈmiku] Surface forms
Under this analysis, /i/ must serve as the underlying form for two distinct patterns: invariant [i],
and the [i] ~ [e] alternation. No general rule will be found that can lower some, but not all of the
claimed underlying /i/. In contrast, the Raising/Tensing analysis works perfectly, with a simple
phonological environment.
6.2 Neutralization and Servigliano
The failure of one analysis, and the success of the other, can be related to the concept of
neutralization (298). We have in Servigliano the following neutralization patterns:
/e/ /i/ /o/ /u/
[i] [u] when a high vowel follows
The failed analysis above fails because it tries to “undo a neutralization”; predicting the height
difference by (some nonexistent) context when in fact the distinction is underlying.
6.3 Summarizing the method
The method employed above, suitably adapted, can be used to address many problems of
phonological alternation. The method is summarized below.
(311) A method for solving problems of phonological alternation
• Break up the words into their morphemes.

• Collect full sets of allomorphs of all the morphemes.
• Find the segments that phonologically alternate.
• Assume that non-alternating segments are the same as their surface forms.212
• Explore alternatives for the underlying representation of alternating segments.
• For each such alternative, reconstruct the underlying representations of the morphemes, and
(by concatenating) the words. Seek rules, possibly ordered, that convert these UR’s into
correct surface representations.
212
More precisely, to whatever emerges from the basic process of phonemicization, treated earlier. Typically it
is best to study alternations by working with already-phonemicized data.
• The correct underlying representations are those compatible with a set of adequate rules.
Study Exercise #79
This involves the case forms of nouns in Hungarian. You should ignore the vowel changes in
suffixes, which are due to a phonological rule of Vowel Harmony.
Phonetic symbols:
[ɟ] is a voiced palatal stop.
[c] is a voiceless palatal stop.
[ɲ] is a voiced palatal nasal.
[ː] means that the preceding vowel is long.
[ø] mid front rounded, as in German Goethe or French Chartreuse.
[y] high front rounded, as in German Führer or French tu or Mandarin [ny̌] ‘female’213
a. What stems alternate, and what are their allomorphs?

b. State a phonological rule that correctly derives the alternation, in both formalism and
words. Give your rule a name.
c. Give underlying forms and derivations for pɔd, pɔdnɔk, and pɔttoːl.
d. Is your rule neutralizing? Explain your answer.
Hungarian Data
Nominative Dative Ablative214 Essive215 Allative216 Gloss

ɾøgbi ɾøgbinɔk ɾøgbitøːl ɾøgbikeːnt ɾøgbihɛz ‘rugby’
ipʃɛ ipʃɛnɛk ipʃɛtøːl ipʃɛkeːnt ipʃɛhɛz ‘fellow’
kɔlɔp kɔlɔpnɔk kɔlɔptoːl kɔlɔpkeːnt kɔlɔphoz ‘hat’
kuːt kuːtnɔk kuːttoːl kuːtkeːnt kuːthoz ‘well’
juk juknɔk juktoːl jukkeːnt jukhoz ‘hole’
sɛm sɛmnɛk sɛmtøːl sɛmkeːnt sɛmhɛz ‘eye’
rɔb rɔbnɔk rɔptoːl rɔpkeːnt rɔphoz ‘prisoner’
ʒɛb ʒɛbnɛk ʒɛptøːl ʒɛpkeːnt ʒɛphɛz ‘pocket’
213
Unchecked data taken from a textbook. If you are a native Mandarin speaker and can confirm,
please contact me.
214
Ablative case means, roughly, “from”.
215
More accurately: essive formal. Essive case means, roughly, “as”.
My Hungarian grammar says: “used with expressions of attaching something to, adding to, or
216
communicating to someone or something” (Carol Rounds, Hungarian: An Essential Grammar, p. 109).

kaːd kaːdnɔk kaːttoːl kaːtkeːnt kaːthoz ‘bathtub’

pɔd pɔdnɔk pɔttoːl pɔtkeːnt pɔthoz ‘bench’
aːɟ aːɟnɔk aːctoːl aːckeːnt aːchoz ‘bed’
mɛlɛg mɛlɛgnɛk mɛlɛktøːl mɛlɛkkeːnt mɛlɛkhɛz ‘heat’
hɛrt͡sɛg hɛrt͡sɛgnɛk hɛrt͡sɛktøːl hɛrt͡sɛkkeːnt hɛrt͡sɛkhɛz ‘duke’
tɛmplom tɛmplomnɔk tɛmplomtoːl tɛmplomkeːnt tɛmplomhoz ‘house of worship’
byːn byːnnɛk byːntøːl byːnkeːnt byːnhøz ‘sin’
toɾoɲ toɾoɲnɔk toɾoɲtoːl toɾoɲkeːnt toɾoɲhoz ‘tower’
fɔl fɔlnɔk fɔltoːl fɔlkeːnt fɔlhoz ‘wall’
øːɾ øːɾnɔk øːɾtøːl øːɾkeːnt øːɾhøz ‘guard’
e. Ponder next the paradigm of ‘emerald’ below and suggest a minimal change for your
analysis to derive it.
f. Give a derivation for [smɔɾɔktkeːnt].
Nominative Dative Ablative Essive Allative Gloss

smɔɾɔgd smɔɾɔgdnɔk smɔɾɔkttoːl smɔɾɔktkeːnt smɔɾɔkthoz ‘emerald’
a. What stems alternate, and what are their allomorphs?
Stem Allomorph 1 Allomorph 2

rɔb rɔb rɔp
ʒɛb ʒɛb ʒɛp
kaːd kaːd kaːt
pɔd pɔd pɔt
aːɟ aːɟ aːc
mɛlɛg mɛlɛg mɛlɛk
hɛrt͡sɛg hɛrt͡sɛg hɛrt͡sɛk
b. State a phonological rule that correctly derives the alternation, in both formalism and
words. Give your rule a name.
Voicing Assimilation
[+stop] → [−voice] / ___ [−voice]
‘A stop becomes voiceless when it precedes a voiceless sound.’
c. Give underlying forms and derivations for pɔd, pɔdnɔk, and pɔttoːl.
/pɔd/ /pɔdnɔk/ /pɔdtoːl/ underlying forms

— — t Voicing Assimilation
[pɔd] [pɔdnɔk] [pɔttoːl] surface forms
d. Is your rule neutralizing? Explain your answer.
Yes. Look at this quadruplet, focusing on the underlined sounds:
kuːt kuːttoːl
pɔd pɔttoːl
There is a /t/-/d/ distinction, but it gets wiped out before a voiceless sound.
e. Ponder next this paradigm and suggest a minimal change for your analysis to derive it.
The crucial forms are forms like [smɔɾɔktkeːnt]. It looks like Voicing Assimilation has to be
allowed to apply to its own output (the standard term for this is “iterative”). The rightmost /k/ turns
a /d/ into a [t], and then this [t] turns the preceding /g/ into a [k].
f. Give a derivation for [smɔɾɔktkeːnt].
/smɔrɔgdkeːnt/ underlying representation

t Voicing Assimilation: first time
k Voicing Assimilation: second time
[smɔrɔktkeːnt] surface form
__________________________________________________________________________
7. The organization of grammar
At this point in the text we have covered (however briefly) most of the central areas of
linguistic analysis: morphology, syntax, semantics, and phonology. To sum up, we can fill out the
the flow-chart diagram for the organization of grammar given earlier on p. 225. This will indicate
how all the theories in these areas fit together, and how information flows through the system. I
emphasize that these are open research questions that linguists continue to debate, and that I am
only giving one particular view here. In the chart, components (modules of the grammar) are
shown in dotted boxes; level of representation (linguistic forms) are shown in solid boxes.
S, etc. (initial symbol fed to the grammar)
(bare tree)
lexicon
Lexical insertion
Deep structure Derivational

morphology
Constraints (acting as filter)

Transformations
Surface structure Syntactic

component
Inflectional morphology Semantic component
Phonemic form Logical form
Phonological
component:
Rule 1
Rule 2
...
Rule n
[phonetic form]
The syntax is the primary generative component, creating an infinite number of possible
sentences.217 The number is infinite because the phrase structure rules can apply recursively, in
loops. Deep structure is created by filling the trees created by phrase structure rules with words
(lexical insertion). Deep structures are modified by transformations, which have the power to
copy and move, generating more elaborate structures that could not be formed by phrase structure
rules alone. Constraints on transformations sometimes filter out sentences that the syntactic
component would otherwise generate.
The words that undergo lexical insertion into the syntactic tree are sometimes single-
morpheme stems like cat, and sometimes the result of rules of word formation. Following the view
of many linguists, the morphology of word formation is shown as a kind of adjunct to the lexicon.
It extracts existing words from the lexicon and forms new words from them, which are added back
to the lexical stock. Word formation rules string together morphemes, which are assumed at this
stage to be composed of phonemes, since the rules of the phonology have yet to apply.
Particular syntactic rules (transformations of agreement, case marking, etc.) build up a

morphosyntactic representation for each inflected word, which specifies the values of features
like [Number], [Case], [Tense], and so on. In a postsyntactic component, the inflectional
morphology, the features of the morphosyntactic representation trigger rules of affixation, which
manifest the inflectional categories in their phonemic form.
At the end of the grammar, the rules of the phonology provide a phonetic realization for the
syntactic structure, relating it to the physical realities of articulation and acoustics. They are places
do that they apply after syntax and morphology. This ordering accounts for the fact that the
morphemes alternate according to the environments in which the morphology and phonology
place them. Phonological rules often neutralize distinctions, sometimes creating ambiguity.
Semantics is placed in a location such that the rules of interpretation apply to syntactic
surface structure. They create a level of logical form, in which the aspects of meaning most
closely related to syntax, such as predicate-argument structure, pronoun reference, and scope, are
derived.
217
Word formation is also generative, and in most languages can likewise create an infinite variety of
structures (recall (75) from Chapter 2: eggplant plant plant…), though the structures typically are far less
elaborate.
For further reading

Quite a few textbooks lay out the basics of phonological theory in more detail; one is my own
text Introductory Phonology (2008; Oxford: Wiley-Blackwell). Another, very data-rich
introductory text is David Odden’s Introducing Phonology (2005: Cambridge University Press). A
more advanced introduction with many beautiful problem sets is Michael Kenstowicz and Charles
Kisseberth Generative Phonology: Description and Theory (1979: Academic Press).
The theory of phonemes was first established in the first half of the 20th century, as linguists
first confronted the extensive data emerging from fieldwork. An acclaimed work from this period
is Leonard Bloomfield’s Language (1932), a book still worth reading.
The system of phonology in which the surface phonetic forms are derived from an underlying
representation using a series of ordered rules is perhaps the oldest part of linguistic theory still in
general use today; it was worked out by the grammarians of ancient India, whose leading figure
was Pāṇini (ca. 500 BCE). The modern revival of the Pāṇinian system began around Bloomfield’s
time, but achieved full development with Noam Chomsky and Morris Halle’s 1968 book The
Sound Pattern of English, a massive study of English phonology.
There is no up-to-date textbook covering the many further developments in phonology over
the past few decades but a useful survey is The Handbook of Phonological Theory, 2nd ed. (2011),
ed. by John A. Goldsmith, Jason Riggle, and Alan C. L. Yu (Wiley-Blackwell).
The discussion of Servigliano in this chapter is drawn from Rachel Walker’s book Vowel
Patterns in Language, Cambridge University Press, 2011.
Chapter 13: Historical Linguistics
1. Outline
Languages change over time, in an interesting and paradoxical way. The speakers of a
language usually communicate fairly well with their grandparents in childhood and with their
grandchildren in old age. This covers five generations. But consider a passage of prose from the
English of about 40 generations ago (Old English, about 1000 C.E.):
(312) A Biblical verse in Old English
urne gedæghwamlican hlaf syle us to dæg

[ urnɛ gɛ dæɣwamlikan ˈl ̥af sylɛ us to dæɣ]
our daily bread give us today
‘Give us this day our daily bread’
This would be unintelligible to a speaker of Modern English, and many of the morphemes have
evolved so as to be only faintly recognizable (e.g. [dæɣ] = day, [ˈl̥ af] = loaf, [lik] = -ly). Somehow,
a series of changes that were little noticed as they were happening have gradually converted
English into an entirely different language.
Just to show an intermediate stage, the following passage is a Middle English translation (ca.
1400 A.D.) of the same Biblical verse. Remember to read it phonetically, not according to spelling.
(This should give you a clue why letters have such different values in English than they have in
European languages.)
yeue to us today oure eche dayes bred.

[gevə to us todæɪ urə etʃə dæɪəs bred]
give to us today our each day’s bread
Historical linguistics attempt to understand the process of linguistic change. The two
fundamental questions in the field are: (a) How and why do languages change? (b) What is the
history of the languages of the world?
2. Descent; related languages
When linguists speak of the “ancestry” of a language, they have a specific meaning in mind. If
Language B is descended from Language A, it means that there has been a continuous
transmission of the language, from generation to generation, going from A to B (with gradual
changes over time). We can speak of this form of language transmission as descent. Modern
English is related to Old English by descent (is “descended from” Old English), as there is a
continuous link through 40 generations of speakers between the two.
We need to be careful about the term “descent”: it certainly does not imply an actual chain of
biological ancestors, because there are countless people who are native speakers of a language
whose parents are not. Such speakers are part of the chain of transmission just as much as children
of native speakers.
For linguists, descent is the gold standard for language identity — descent has a completely
clear meaning and can be diagnosed with near certainty if enough data are available. Descent is not
always used as the criterion of language identity in ordinary usage, however. For instance, in the
real world you will hear people say things like:
“Modern English is a mixture of Old English, French, and Latin.”
This statement is perfectly true as a description of the vocabulary of Modern English, since over
the centuries English has borrowed thousands of words from French and Latin. But English is
descended solely from Old English;218 there was no continuous transmission of language from
generation to generation that leads from French or Latin to English.
Two languages are said to be related if they descend from the same ancestor language. That
is, it is often the case that a single language comes to be spoken in two geographically isolated
areas, or over a very wide area. Given enough time, such a language is likely to develop more than
one descendent. Because of lack of intercommunication, different areas evolve their own
descendent languages, which eventually become mutually unintelligible. Exactly this happened in
the evolution of the modern Romance languages from Classical Latin. Thus, the Romance
languages are related to one another (in the technical sense) because they all descend from the
same ancestor.
Languages can thus be thought of as family groupings. We can use family tree notation to
represent the ancestry of languages, in which a line represents a relationship by descent.
Latin
French Italian Spanish Portuguese Rumanian etc.
Here are some other examples of language families. The Germanic languages are all closely
related. They descend from a common ancestor which was spoken roughly at the same time as
Latin. However, this ancestor was spoken by an illiterate people, so we have no records of it. The
name used for the common ancestor of the Germanic languages is Proto-Germanic.
218
… and, at a deeper level, the earlier languages from which Old English is descended; more on this
immediately below.
Proto-Germanic
Old English Frisian Dutch German Old Norse

|
Middle English
| Norwegian Swedish Danish Icelandic
Modern English219
Although we have no written records of Proto-Germanic, we nonetheless have a fairly good

idea of what it was like. Just how we know this will be a central topic later on.
Latin and Proto-Germanic are in fact related to each other. They are (roughly speaking)
sisters, and descend from an ancestor language called Proto-Indo-European. The Indo-European
language family is a large one, and almost half the population of the world speaks an Indo-
European language. Here is a very sketchy version of the Indo-European family tree:
Proto-Indo-European
Proto-Germanic Celtic Italic220 Balto-Slavic Greek Albanian Armenian Anatolian Indo-Iranian Tocharian
German English etc. Baltic Slavic Hittite Persian Hindi etc.
Latvian Lithuanian Russian Bulgarian etc.
“Relatedness” should not be confused with “similarity”. For example, Modern Persian is in a
sense far more similar to Arabic than to Modern English, at least in vocabulary; thousands of
words of Persian are borrowed from Arabic.
[domˈhur] ‘republic’
[ˈelm] ‘science’
[mohænˈdes] ‘engineer’
[velˈjæt] ‘province’
[rædd] ‘refutation’
If one’s goal is to learn Persian, it may well be more useful to start off knowing Arabic than
knowing English. Nevertheless, Persian is related to English (they are “cousins”, both
granddaughters of Proto-Indo-European); and Persian is not at all related to Arabic. One can see
this in some of the core vocabulary of English and Persian:
219
Only English is shown with stages (following the custom, three of them: Old, Middle, Modern).
But the same could be done for all of the languages shown.
220
The Italic family consists of Latin and a few poorly-attested sisters. As noted above, all of the
Romance languages (also including: Portuguese, Romanian, Catalan, and others) descend from Latin.
[peˈdær] ‘father’
[bærɑˈdær] ‘brother’
[setˈre] ‘star’
[gɑv] ‘cow’
[æst] ‘is’
[bor-d-æn] ‘carry-past stem-infinitive’ = “bear”
[bu-d] ‘be-past tense’
Notice that these words, which are authentic cognates (shared inheritances) in English and
Persian, are core, commonplace words—the kind that a language tends to hang on to. The words
shared by Persian and Arabic are mostly more sophicated ones: Persian typically has borrowed its
vocabulary for the spheres of higher learning from Arabic.
One result of looking at things in this way is that statements like (313):
(313) “Lithuanian is a very old language”
become meaningless. In fact, they are often just expressions of nationalistic sentiment. In truth,
virtually all languages are equally old, in the sense that they have an ancestry that goes back
farther than anyone can trace.221 There are only two ways that statement (313) could be given a
true interpretation. It could mean that we have written records of Lithuanian dating back to the
distant past; or it could mean that Lithuanian has changed very little over the centuries.
3. Sound change
Sound change is a fundamental mechanism of language change. That is, one of the principal
reasons that languages change is because their sounds change. For example, the voiceless [l̥ ] in Old
English [l̥ af] ‘bread’ (one of the words in example (312)) has become voiced [l] in Modern
English. This change happened to all the voiceless [l̥ ]’s of Old English; for example, the words the
words lady, lot, and lean originally began with voiceless [l̥ ]’s.222
As a language evolves, it is subjected to dozens or even hundreds of sound changes, until it

takes on a form that would be unintelligible to the original speakers.
Sound change is connected in a curious way to phonology. Basically, sound change results
from the fact that throughout its history, a language has a large number of phonological rules. The
rules are the seeds of sound change.
221
Exceptions: invented languages, such as Esperanto; and more significantly cases like Nicaraguan Sign
Language, which quickly emerged as a real language, with grammar, starting from gestural/mimetic raw material. This
occurred during the years after young Deaf people in Nicaragua first had the chance to form a speech community, with
the establishment of a national school for the deaf.
The Old English for “lady” was hlæfdige, literally “kneader of bread”. “Lot” was hlot, and “lean”
222
was hlǽne. All three forms are from the Oxford English Dictionary, available online from many university
computers (including UCLA’s) at http://dictionary.oed.com/.
However, it is important to see that sound change and phonological rules are not the same
thing. A sound change is a historical event. For example, if all the words that in 1300 were
pronounced with voiceless [l̥ ] are pronounced with voiced [l] in 1500, then we say that the
language has undergone a sound change taking [l̥ ] to [l]. A phonological rule, on the other hand, is
something in the mind of a native speaker; it is part of a speaker’s unconscious mental grammar.
3.1 Sound change and restructuring
The link between phonological rules and sound change is a phenomenon called restructuring.
To understand this concept, it will help to do an example in detail.
The sound change we will examine is a fairly recent one. American English can be divided
into a dialect that has an extra phoneme /ɔ/ and a dialect that lacks this phoneme. I will call the
dialect that has /ɔ/ “Dialect A”, and the dialect that lacks it “Dialect C” (why not “B” will become
clear shortly). In Dialect A, caught is pronounced [ˈkɔt] and cot is pronounced [ˈkt]; whereas in
Dialect C, both words are pronounced [ˈkt]. In fact, Dialect C has // in all words where Dialect A
has /ɔ/.
Dialect A Dialect C
cot [ˈkt] [ˈkt]
caught [ˈkɔt] [ˈkt]
la [ˈl] [ˈl]
law [ˈlɔ] [ˈl]
hock [ˈhk] [ˈhk]

hawk [ˈhɔk] [ˈhk]
generally: [] []

[ɔ] []
It can be argued that Dialect A represents the original state of the language, and that
Innovating American English has undergone a sound change: ɔ has become  in all environments.
There are two reasons to believe this.
First, there is the fact that, with just a few exceptions, speakers of Dialect A agree with each
other on which words have [ɔ] and which words have [ɑ]. This fact would be very difficult to
explain unless the distinction is inherited. There’s no official committee that decides to change the
pronunciation of words. Rather, children usually just adopt the pronunciation of the previous
generation.
The other reason to think that the [ɔ]-[] distinction reflects the earlier state of the language is
that all the old written documents through the centuries spell out the distinction.223 English spelling
was invented, probably by scribes who already know how to read and write Latin. There’s every
reason to think that the old scribes did their best to reflect in their spelling what they heard with
their ears.
Let us therefore adopt the assumption that Dialect C is the one that has innovated, and that it
has undergone a sound change. What was the mechanism of the change? The clue lies in what I
will call “Dialect B,” the crucial intermediate case.
Speakers of B have free variation in the caught class of words. Extending the data above to B,
we have:
Dialect A Dialect B Dialect C

cot [ˈkt] [ˈkt] [ˈkt]
caught [ˈkɔt] [ˈkɔt] ~ [ˈkt] [ˈkt]
la [ˈl] [ˈl] [ˈl]

law [ˈlɔ] [ˈlɔ] ~ [ˈl] [ˈl]
hock [ˈhk] [ˈhk] [ˈhk]

hawk [ˈhɔk] [ˈhɔk] ~ [ˈhɔk] [ˈhɔk]
generally: [] [] []

[ɔ] [] ~ [ɔ] []
It is in Dialect B that we can see sound change in progress. Evidently, B speakers have a rule
of neutralization, which applies optionally—in other words, they have a distinction, but sometimes
wipe it out phonologically. Here would be the phonological analysis of Dialect B:
(314) /ɔ/ Unrounding
ɔ → [−round] in all environments (optional)
A 1440 Latin glossary: “Hawke, falco”. A 1398 source: þe hocke is a nesche herbe (as in
223
hollyhock).
Example derivations:
hock hawk
/ˈhk/ /ˈhɔk/ underlying representation
— ˈhk — /ɔ/ Unrounding

[ˈhk] [ˈhk] [ˈhɔk] surface representation
It is claimed here that B represents the intermediate stage in the historical evolution from A to
C. When a language has an optional rule, it tends to be applied more and more often through time.
That is, people’s standards of what constitutes “careful speech” get lowered, and the casual-speech
rules get applied more frequently.
The next step involves the introduction of a new generation of speakers. As young children,
these speakers face the task of learning the phonemic system of their language. However, in the
present case, the task is a very difficult one. The older speakers, who supply the data, have in their
minds a phonemic distinction between /ɔ/ and //. However, in their actual pronunciations, /ɔ/ is
fairly rare, because most of the time these speakers apply the voicing rule. The new generation has
very little data that they could use to learn the /ɔ/ phoneme. The potential for acquisition error is
great.
It is easy to imagine how this situation will turn out. The younger generation is likely not to
notice the [ɔ]’s at all, and they will acquire a different phonological system, in which [ɔ] plays no
role at all. Here are the oldest, intermediate, and youngest phonological systems compared:
Dialect A
two phonemes, /ɔ/ and //
no applicable phonology
Dialect B
two phonemes, /ɔ/ and //
Phonological rule: ɔ → , optionally
Dialect C
one phoneme: //
no applicable phonology
The interesting point is this:
• The speech of “late decadent” Dialect B and Dialect C are almost identical; B speakers
pronounce the old [ɔ] words with [ɑ] (let us say) 95% percent of the time, whereas C
speakers pronounce them with [ɑ] 100% of the time.
• But the phonological systems of B and C are drastically different, due to the acquisition
error that created C—the children who brought C into existence failed to notice a phoneme,
and thus also failed to learn the rule.
The technical name for this phenomenon is restructuring:
(315) Defn.: Restructuring
a change in the phonological system of a language, induced by the phonological rules of

the older generation.
Dialect C probably arose from a restructuring by younger speakers of the unstable

phonological system of Dialect B. Dialect B in turn represents an innovation (through the
introduction of the rule of /ɔ/ Unrounding) in Dialect C. American English preserves all three
dialects today, though we can perhaps anticipate that in a couple centuries all Americans will
speak C.
To summarize, most sound changes are the result of the following process. (a) A new
phonological rule is introduced into a language. (b) The rule is applied with increasing frequency.
(c) A new generation restructures the system, getting rid of the rule.
It can now be seen why speakers don’t notice their language changing. The basic ingredient of
the change, the optional phonological rule, is an inherent, normal part of the language. The
restructuring by the next generation is phonetically very minor, even though it is a radical change
in the underlying system.
A bit of notation: when linguists write “→”, the arrow implies a phonological rule: part of
the tacit knowledge of a living speaker. When they write “>” instead, they mean a sound change—
a historical event that arose as a consequence of phonology. The fact that the material on either
side of these sides is the same should not blind us to the fact that a rule and a sound change are
logically very different things. Thus:
ɔ→ “/ɔ/ is realized as [ɑ]”, part of grammar

ɔ   “[ɔ] evolved into [ɑ]”, part of history
4. The regularity of sound change
It is true of most phonological rules that they apply regularly. For example, the rule of
Tapping in English is regular; there are no exceptions to it in the whole vocabulary. Now if sound
change is the result of phonological rules, then we would expect sound change to be regular as
well. In general, this turns out to be true. Thus, for instance, in Dialect C of American English, not
a single [ɔ] is left; they have all turned into /ɑ/.
To give another example: another recent, exceptionless sound change of American English
converted /æ/ to /e/ before //. Here again, the conservative dialect still exists alongside the
innovating dialect.
æ > e / ___ 
Conservative dialect: marry [ˈmæi] Mary [ˈmei]

carry [ˈkæi] Cary [ˈkei]
arable [ˈæəbəl] airable [ˈeəbəl]
Harry [ˈhæi] hairy [ˈhei]
Innovating dialect: marry [ˈmei] Mary [ˈmei]

carry [ˈkei] Cary [ˈkei]
arable [ˈeəbəl] airable [ˈeəbəl]
Harry [ˈhei] hairy [ˈhei]
This is intended as one further example of the exceptionlessness of sound change: if you
speak the innovating dialect, the odds are that you have no words whatever that still contain /æ/
before //; indeed, such pronunciations may seem outright unnatural.
For a sound change that had exceptions, we can consider *ʊ → ə, which occurred roughly
during the 1600’s and affected most dialects of English.224 This sound change had just a few
exceptions (for example, put), which means that /ʊ/ survived as a phoneme, but is rare in English
today.
224
The regional dialects of northern England were not affected by this change, and speakers of these
dialects use [ʊ] in many words where other speakers would use [ə], such as luck.
Study Exercise #80
Let’s add in “Dialect BR” — standard British English, and do some comparisons with American
Dialect A.
Word BR A
sore [sɔ] [sɔɹ]
saw [sɔ] [sɔ]
door [dɔ] [dɔɹ]
daw225 [dɔ] [dɔ]
lore [lɔ] [lɔɹ]
law [lɔ] [lɔ]
pore [pɔ] [pɔɹ]
paw [pɔ] [pɔ]
roar [ɹɔ] [ɹɔɹ]
raw [ɹɔ] [ɹɔ]
Using the same reasoning as given earlier, decide which dialect has changed, what the change
was, and what the original forms were.
225
A kind of bird.
The original forms were like dialect A, the American one. There are two reasons to believe
this is true. First, Americans agree with one another about which words should have an /ɹ/ in them.
This would be very hard to explain if the /ɹ/’s were innovated. Second, the spelling of the words,
established long before British and American English split, indicates the early presence of /ɹ/ in
sore, door, lore, pore, and raw.
————————————————————————————————————
The idea that American speech is, at least in this respect, a more accurate continuation of the
historical past of the English language is confusing, at least to some people. After all, the culture of
Britain is centuries older than that of America. But the documentary evidence that this aspect of
American speech is “older” is easy to find. Looking up the words of the Study Exercise in the
Oxford English Dictionary, I find that for all of these words, there are attestations that predate the
split of American and British English (no earlier than 1607, when English people first succeeded in
establishing a colony in America). These clearly show an r in exactly the words that most
Americans pronounce with /ɹ/ to this day.
1300: þe touche of senewes haþ no feling of soore and of smerte.

1340: File sawe and spindelle
1000: Seo duru wæs belocen 226
1600: The theevish Daw, and the dissembling Pye.
1502: Who wil not for shame a short tyme suffir lore and lerne.
1596: The Venetian Law cannot impugne you as you do proceed.227
1400: þat neiþer cold ne hoot ne schulde not sodeynli entre þe poris of the skyn.
1380: This egle..with hys grymme pawes stronge (Chaucer)
1391: With such a noise and such a rore
1325: þe deuel huem afretye, Rau oþer a-roste!228
One other fact: the geography of deleted /ɹ/ is what we would expect if it originated among
fashionable people in London sometime in the 1600’s or 1700’s. It spread outward from London,
reaching the large cities of Birmingham and Liverpool, but never reaching Scotland or Ireland or
indeed much of the rural territory of England. R-less pronunciation was exported from England by
emigration to Australia and New Zealand. To some degree it was exported to America and became
part of the dialects of coastal cities such as Boston, New York, and Charleston, South Carolina.
However, it arrived too late to affect the people who had already settled inland; hence the majority
dialect in America preserves historical /ɹ/.
226
“The door was shut”, from an English version of the Bible, Matthew 25:10.
227
Shakespeare, The Merchant of Venice IV.1.178.
228
I believe this is something like “The devil him affright, raw or roasted!”.
5. How do languages make up for lost phonemes?
Before going on, I will address a problem that is raised by the sound changes we have seen.
Notice that two of these sound changes eliminated phonemes from the language: ɔ > ɑ eliminated
the /ɔ/ phoneme, and æ > e / ___  eliminated the /æ/ - /e/ distinction before //. In fact, sound
changes do this fairly often. If this is so, why don’t the world’s languages gradually lose all their
distinctions, and become an incoherent stream of muttering, say [dədədədədədədə]?
One answer is that languages borrow phonemes from neighboring languages. This happens
frequently; for example, English borrowed the phoneme /v/ from French (it had a [v] before, but
only as an allophone of /f/, not as a separate phoneme.) Here are examples:
value from French vallue ca. 1300

vanquish from French vencus ca. 1330
view from French veue ca. 1415
Japanese long ago borrowed [tʃ] from Chinese ([tʃa] ‘tea’), and much later [f] from English.229
However, it is also possible for a language to create a new phoneme entirely on its own. Here
is an example of how this can happen, from the history of German. I will show how German
created a new phoneme, during the transition from Old High German (the ancestor of Modern
German, spoken around 1000 A.D.) to Middle High German (an intermediate stage, spoken around
1400 A.D.).
Here are the relevant facts. I give a partial paradigm for the adjective hox ‘high’ in both Old
High German and Middle High German. [x] stands for a voiceless velar fricative, and [ø] is a front
rounded vowel.
OHG MHG
‘high’ ˈhox ˈhox
‘higher’ ˈhox-iro ˈhøx-ərə
‘highest’ ˈhox-isto ˈhøx-əstə
‘high (adv.)’ ˈhox-o ˈhox-ə
You can see from the data that Middle High German has two sounds, /o/ and /ø/, where Old
High German has only /o/. Further, /o/ and /ø/ must be separate phonemes, because there is no
reasonable way to predict which one will occur in a given environment. Thus Middle High
German has created a new phoneme. How was this done? The mechanism was simply sound
change. The evolution of the forms above is the result of the following two sound changes,
applying in the (historical) order given:
229
More precisely, Japanese uses the sound [], a voiceless bilabial fricative, to render English the
phonetically similar /f/, as in [aito] ‘fight’; [esutibau] ‘festival’.
I. Umlaut: o > ø / ___ [−syllabic] i
+syllabic
II. Vowel Reduction: –stress  > ə
E
Umlaut turned /o/ into the corresponding front vowel [ø] when the vowel /i/ occurred in the
next syllable (this makes sense, since /i/ is itself a front vowel). Vowel Reduction converted all the
unstressed vowels into schwa. The stress in Old High German and Middle High German always
fell on the first syllable, so in effect Vowel Reduction applied to all vowels in non-initial syllables.
The following derivations show how Umlaut and Vowel Reduction jointly created a new
phoneme:
hox-iro hox-o
höx-iro hox-o Umlaut
höx-ərə hox-ə Vowel Reduction
By itself, Umlaut introduced only a new allophone. At the beginning of its existence, [ø] was
only a phonetic variant of /o/. The dirty work was done by Vowel Reduction: this sound change
obliterated the environment that had triggered Umlaut. The sound [ø] was “stranded”; it was no
longer predictable from the context, and thus came to be a phoneme on its own.
You can see, then, that it is possible for a language to acquire a new phoneme, strictly from its
own resources, without borrowing it. The general mechanism is this: a new rule created an
allophone, then a later sound change wipes out the conditioning environment for that allophone.
The allophone then stands alone as a new phoneme.
One further point: it’s clear that the “wiping out of conditioning environments” often will
happen, as it did in German, by removing phonemes—what averts the crisis of the language’s
words becoming so short that they get confused with each other? The answer appears to be that
morphology comes to the rescue. For instance, a striking aspect of Mandarin Chinese is that a great
fraction of its basic vocabulary consists of compound words. It is thought that this compounding
arose as a response to massive phonological erosion, the result of sequence of dramatic sound
changes in the earlier history of the language.230
The overall picture is that languages manage (probably through the efforts of innovating
children during the course of acquisition) to retain a kind of balance, in which there are sufficient
phonemic contrasts, and the words are sufficiently long, to keep the vocabulary items reasonably
distinct from one another.
230
A miniature example of the same kind, from the Web, where a Southerner reports: “The reason we
say straight pin is that, in many Southern dialects, pin and pen are homonyms. To ensure that the correct item
is fetched, one says Please fetch me a straight pin or Please fetch me an ink pen. (Source:
http://everything2.com/e2node/straight%2520pin) The sound change that took place in Southern dialects is
*ɛ > ɪ / ___ [+nasal], hence [pɪn] for both pin and pen.
6. The Comparative Method
I said earlier that the modern Germanic languages all descend from a single ancestor, called
Proto-Germanic. In addition, both Proto-Germanic and Latin descend from a common ancestor
called Proto-Indo-European. We do not have written records of either Proto-Germanic or Proto-
Indo-European. How do we know that these languages existed, and how do we know what they
looked like?
Our knowledge is the result of the Comparative Method. The Comparative Method is a way
of recovering information about a lost proto-language by comparing its known daughter languages.
This method was worked out over the course of the 19th century by a research community of
mostly European linguists.231
The basis of the Comparative Method is the fact that sound change is normally regular. It is
the regularity of sound change that permits us to prove that languages are related, and to recover
information about their lost ancestor.
In outline, the Comparative Method works like this:
a) Compare sister languages sound by sound.

b) Determine what regular sound changes could have given rise to the correspondence of
different sounds across languages.
c) “Reconstruct” the original language by undoing the various sound changes.
To illustrate the Comparative Method, I will apply it to the language of instruction in this
course, comparing it with its sister languages German and Swedish to recover information about
the hypothesized answer, namely Proto-Germanic. Here is the first batch of data:232
English German Swedish

A. good ˈgd ˈgut ˈgud
drive ˈdrav ˈtrab-ən ˈdriv-a233
ride ˈrad ˈrat-ən ˈrid-a
wide ˈwad ˈvat ˈvid
deed ˈdid ˈtat ˈdod
231
One of them, Jacob Grimm, was also a pioneer (in collaboration with his brother Wilhelm) in the
scholarly collection of folklore; hence “Grimm’s fairy tales”.
232
Swedish forms were converted to IPA using the rules given in Philip Holmes and Ian Hinchcliffe
(1997) Swedish: An Essential Grammar, Routledge. I have not yet checked my conversions with a Swedish
expert. For future reference, the Swedish words below are spelled: god, driva, rida, vid, dåd, grön, gå, giva,
gås, binda, rund, land, hund, lind, stol, sten, bäst, lista, vit, fot, söt, ut, tecken, salt, smärta, spinna, spade,
löpa, hop, pund, and pipa.
233
For simplicity I’ll ignore the phonetic differences between English, German, and Swedish r, which
are actually [], [] (voiced uvular approximant), and [].
The data have been chosen in the following way. They all contain a /d/ in English, and the
corresponding German and Swedish forms are phonetically similar and mean roughly the same
thing. (In the German forms, I have added a suffix in various places. This simplifies the problem,
without distorting it in any crucial way.)
The crucial observation to be made here is this: wherever English has /d/, Swedish also has /d/
in the same location of the word; but German has /t/. We can express this as a formula:

d t d
The formula holds true not just for these words, but for hundreds of words throughout vocabularies
of the three languages.
What could account for the d-t-d correspondence? The answer proposed here is:
(a) English, German, and Swedish all descend from the same proto-language. That is, at one
time they all were the same language, namely Proto-Germanic.
(b) Following the breakup of Proto-Germanic, German underwent a sound change that
changed /d/ to /t/ in all environments.
Because sound change is regular, this explanation accounts for the regularity of the t-d-t
correspondence.
There is a standard way of expressing our hypothesis in a compact form. We use an asterisk to
designate a hypothetical sound; thus if we assume that Proto-Germanic had a /d/, we designate the
/d/ as *d. (Thus in historical linguistics, asterisk means “hypothetical”, not “ungrammatical”.). We
can write the proposed sound changes with the same notation as phonological rules. Here, then, is
the analysis:
Correspondence:
English German Swedish Proto-Germanic

d t d *d
Sound change:
d  t in German
Notice that this is not the only possible analysis. It is conceivable that Proto-Germanic had *t,
and that English and Swedish changed; or even that Proto-Germanic had something completely
different, and all three daughters changed. What we say about the phonetic identity of the original
sound is a more or less educated guess; what we can be sure about is that there was some particular
sound in Proto-Germanic that gave rise to English /d/, German /t/, and Swedish /d/.
Note finally that the sound change is hypothesized to have once been a phonological rule; that
is, that the early speakers of German first optionally changed their /d/’s to [t]’s, and gradually
came to do this regularly, causing the next generation to restructure (see section 3.1 above).
Let us continue the reconstruction, with the following data:

B. good ˈgd ˈgut ˈgud
green ˈgrin ˈgryn ˈgrøn
go ˈgo ˈge-ən ˈgo
give ˈgv ˈgeb-ən ˈgiva
goose ˈgus ˈgans ˈgos
Here the focus is on /g/. Clearly, not much work is needed here, since all three language have
this sound. The most reasonable hypothesis is that Proto-Germanic had *g, and that it has evolved
unchanged in the daughter languages.
Correspondence:

g g g *g
The following examples look like they might be a problem. Where English and Swedish have
/d/, German has /d/, rather than the expected /t/:

C. bind ˈband ˈbind-ən ˈbind-a
round ˈrand ˈrund-ə ˈrund
land ˈlænd ˈland-əs ˈland
hound ˈhand ˈhund-ə ˈhund
linden ˈlndən ˈlind-ə ˈlind
The problem can be resolved if we carefully compare the data under A with the data under C.
In all the examples of C, the /d/ of German occurs after /n/. In the examples of A, the /t/ of German
never occurs after /n/. We know already that phonological rules have environments; so it is
reasonable to suppose that the *d  t change had one. In particular, it was blocked after /n/, so
that in this set of words German retains the Proto-Germanic /d/. The analysis, then, must be
something like this:
Correspondence:

d t d *d
d d d *d
Sound change:
d  t except / n ___ in German
Sometime the environments for a sound change are more complicated. In the following data,
we are looking at what corresponds to English /t/. In German, /ts/ stands for an alveolar affricate:

D(i) stool ˈstul ˈʃtul ˈstul
stone ˈston ˈʃtan ˈsten
best ˈbɛst ˈbɛst-ə ˈbɛst
list ˈlst ˈlst ˈlista
(ii) white ˈwat ˈvas ˈvit

foot ˈft ˈfus ˈfut
sweet ˈswit ˈzys ˈsøt
out ˈat ˈas ˈʉt234
(iii) token ˈtokən ˈtsaxən ˈtɛken

salt ˈsɔlt ˈzalts ˈsalt
smart235 ˈsmart ˈʃmɛrts-ən ˈsmerta
Lent ˈlɛnt ˈlents ?
The correspondences are as follows:
234
[ʉ] is a high central rounded vowel.
235
In the sense of pain, as in “that smarts”.

D. i. t t t
ii. t s t
iii. t ts t
It is possible to show that all three rows reflect *t in Proto-Germanic. German retains /t/ after a
fricative, shifts *t to /s/ after a vowel (including a diphthong), and shifts *t to the affricate /ts/ in
word initial position or after a non-fricative consonant. The analysis would be as follows:

D. i. t t t *t
ii. t s t *t
iii. t ts t *t
Sound Changes in German:
t  [+fricative] / [+syllabic] ___
–syllabic 
t  [+affricate] / –fricative ___
E
One more set of data suggests a slight revision of the analysis:

E. spin ˈspn ˈʃpn-ən ˈspna
spade ˈsped ˈʃpatən ˈspade
leap ˈlip ˈlaf-ən ˈløpa
heap ˈhip ˈhafən ˈhup
pound ˈpand ˈpfund-əs ˈpund
pipe ˈpap ˈpfaf ˈpipa
Here the correspondences are as follows:

p p p
p f p
p pf p
In analyzing these data, the trick is to ignore temporarily the minor difference between
bilabials and labio-dentals, and refer to them collectively as “labials”. If we do this, we find a close
similarity between the labials and the alveolars. That is, German has converted stops to fricatives
after a vowel or glide, and has converted stops to affricates after a non-fricative consonant. Thus to
handle the labials, we needn’t assume additional sound changes, but only generalize the previous
ones:

p p p *p
p f p *p
p pf p *p
Sound Changes in German:
+stop 
–voiced > [+fricative] / [+syllabic] ___
E
+stop  –syllabic 
–voiced  [+affricate] / –fricative ___
E E
There are a couple of loose ends to clear up. First, we have to add some detailed sound
changes to specify our rather vague “labials” of German as either bilabial or labiodental. This step
is not particular interesting, so I will skip it here. We also have to determine the facts for the third
voiceless stop of Proto-Germanic, namely *k.
For the first sound change (the one that created fricatives), we are on safe ground. Proto-
Germanic *k did indeed become a fricative (the velar one) in German, as is shown by cases like
token = tsaxən, seek = zux-ən, make = max-ən. The messy part concerns the expected velar
affricate /kx/. This does exist in Swiss German, and written records show that it once existed
through much of the German-speaking area. However, a later sound change caused /kx/ to revert
back to /k/ in most German dialects. Thus the system as it stands today is not as symmetrical as we
might expect.
At this point we have reconstructed several sounds of Proto-Germanic using the Comparative
Method:
*p *t *k
*d *g
We could go on, until we have reconstructed the entire inventory of Proto-Germanic

phonemes. When this is done, we can reconstruct entire words. For example, the Proto-Germanic
word for ‘good’ (English gd, German gut, Swedish gud) has been reconstructed as *gud. One
can use similar methods to reconstruct much of the morphological system, and, to a limited extent,
even the syntax.
I should admit that this exercise is artificial in an important way. No one seriously attempting
to reconstruct Proto-Germanic would use Modern English, Modern German, and Modern Swedish
as the basis of the reconstruction. Better results are obtained by using the oldest available written
records of these languages. Real reconstructions employ Old English (oldest records 800’s A.D.)
instead of Modern English, Old High German (800’s A.D.) instead of Modern German, and Old
Norse (800’s) instead of Swedish. Other languages are used as well. The oldest attested data from
a Germanic language is from Gothic, a language now extinct that was spoken by one of the tribes
that overran the Roman Empire. Parts of the Bible were translated into Gothic around 600 A.D.
To summarize: the comparative method involves (a) locating “sister words” from sister
languages; (b) determining the sound correspondences; (c) writing the sound changes in each
language; and (d) determining the original forms to which the sound changes applied.
Study Exercise #81: Proto-Germanic θ
The following correspondence sets illustrate the fate of Proto-Germanic * in English,

German, and Swedish. The German data are slightly fudged, as bad, klad, tod, and lad are
actually pronounced [bat], [klat], [tot], and [lat]. The pronunciations shown are used whenever an
ending is added to the word. For purposes of the problem, assume the pronunciations given.

1. bath [bæθ] bad bad
2. brother [brðɚ] bruder bruder
3. cloth [klɔθ] klad kled
4. death [dɛθ] tod dod
5. further [fɚðɚ] (be)fordərn (be)furdra
6. loath [loθ] lad led
7. thank [θɪŋk] dak takk
8. thing [θɪŋ] d ti
9. three [θri] dra tre
a. Arrange the data into phonetic correspondence sets, i.e. complete the chart that would begin
as follows:
Eng. Ger. Swed.

θ d d 1,3...
ð d d 2,5
b. Determine the sound changes that *θ has undergone in the three languages, and write them
in the format
X  Y / P___Q in Language L
c. The following cases seem to go against what you’ve seen before (cf. nos. 1, 3, 4, and 6).
How might they be explained?
10. bathe [beð] baden bada

11. clothe [kloð] kladen kleda

12. loathe [loð] laden leda
There are two clues to consider: first, the spelling of English was established long ago in the
history of the language, before a number of sound changes took place. Second, consider
differences in the corresponding German and Swedish forms.
a. English German Swedish
θ d d 1,3,4,6
ð d d 2,5
θ d t 7,8,9
b. English: θ > ð / [+syllabic] ___ [+syllabic]
German: θ > d (everywhere)
Swedish: θ > t / [word ___

θ > d / elsewhere
c. These forms have English /ð/ matching German and Swedish /d/, whereas the “normal”
forms of 1, 3, 4, and 6 have English /θ/ matching German and Swedish /d/.
We know that in English, *θ became [ð] just in case it was between two [+syllabic] segments.
A reasonable hypothesis would be that at the time of the θ > ð sound change, the *θ’s of bathe,
clothe, and loathe really were between two [+syllabic] sounds; in particular, that there was a final
vowel in these words that is no longer pronounced. The final vowel dropped out only after the θ >
ð change had already happened.
The history of bathe under this hypothesis would be as follows:
*beθə (assuming that the final vowel was schwa)

beðə θ > ð / [+syllabic] ___ [+syllabic]
beð dropping of final schwas
There are a number of facts supporting this hypothesis. First of all, the “missing vowel” really
is present in German and Swedish. That is, in those cases in which English has “mysterious ð”,
German and Swedish have an extra vowel that is missing in English; and in those cases in which
English has the normal final [θ], German and Swedish do not have an extra vowel. That is, we can
use German and Swedish to suggest what English originally looked like, and thus explain an
otherwise mysterious change.
In addition, notice that in just those cases where English has “mysterious ð”, the spelling puts
a “silent e” at the end of the word. This silent e is pointless from a modern point of view, but it
makes sense if the e was at one time pronounced. The spelling of these words remained the same,
even though one of the vowels was no longer present.
7. Study Exercise #82: Proto-ABC
This problem has made-up data, but the patterns are patterns seen in real language histories.
We assume a proto-language, called ABC, with three attested daughter languages, A, B, and
C. The goal is to characterize the sound system of Proto-ABC and all of the sound changes that
applied in its daughter languages.
The data below are alphabetized by Language A, but not otherwise organized.
No. Language A Language B Language C gloss236

1 kaku kaku kaku ‘snow’
2 kapo kabo kaba ‘daughter-in-law’
3 kawi gawi gawi ‘lamb’
4 kene gene dʒana ‘salmon’
5 kepi gebi dʒabi ‘victory’
6 keta geda dʒada ‘cow’
7 kiko giko dʒika ‘grass’
8 kiko giko dʒika ‘ancestor’
9 kita kida tʃida ‘rye’
10 kitu kitu tʃitu ‘weft’
11 kopa kopa kapa ‘loom’
12 kopu kobu kabu ‘harness’
13 kuki kugi kudʒi ‘reins’
14 mame mame mama ‘battle’
15 maru malu maru ‘goddess of the hearth’
16 mupu mubu mubu ‘bridle’
17 mura mula mura ‘sacrifice’
18 mura mula mura ‘wine’
19 naki naki natʃi ‘wheat’
20 nari nali nari ‘rain’
21 newi newi nawi ‘sheep’
22 niwo niwo niwa ‘elbow’
23 noto noto nata ‘goddess of wisdom’
24 pako pako paka ‘chicken’
25 peka peka paka ‘old’
26 peko bego baga ‘wool’
27 pika bika bika ‘wheel’
28 poke boke batʃa ‘brother-in-law’
29 poku pogu pagu ‘linen’
30 pomo pomo pama ‘god of thunder’
31 pono bono bana ‘sow (seeds)’
32 popa bopa bapa ‘young’
236
The glosses are meant to be words that could have occurred in Proto-Indo-European, a society that
(as we know from the actual reconstructed vocabulary) raised crops, milked cows, obtained wool from sheep,
spun and wove cloth, fought with chariots, and worshipped many gods.
33 puke puke putʃa ‘plow’

34 punu punu punu ‘goat’
35 putu budu budu ‘shield’
36 rike like ritʃa ‘household god’
37 riwu liwu riwu ‘blanket’
38 roti loti rati ‘sky’
39 ruti ludi rudi ‘chief’
40 taki tagi tadʒi ‘roast’
41 tawo dawo dawa ‘cloud’
42 tepo depo dapa ‘altar’
43 tewe tewe tawa ‘pray’
44 tinu dinu dinu ‘harvest’
45 tipi dibi dibi ‘god of war’
46 tipu dibu dibu ‘chariot’
47 tita dida dida ‘gizzard’
48 tito dito dita ‘amphora’
49 towu towu tawu ‘beget’
50 wute wute wuta ‘spear’
51 kupa kupa kupa ‘sword’
52 kuma guma guma ‘leather’
Here is the problem solved, through a series of Socratic questions.
Question: (a) Find the correspondence series for liquids (l, r), and conjecture what was the
ancestor sound.
Answer to Study Exercise #79, part (a):
Everywhere (for example: 15, 17, 18, 36-39), we find that A, B, C, have [r, l, r]. The simplest
guess is that *r in Proto-ABC evolved into [l] in B.
R to L
r>l in Language B
Study Exercise #79, Question (b)
Give the inventories of stops and affricates in A, B, and C, arranging them into one chart for each
language.
Answer to Study Exercise #79, question (b)
+palato- 
[+bilabial] [+alveolar]  alveolar E [+velar]
A [+stop] [−voice] p t k
B: [+stop] [−voice] p t k
[+voice] b d g
C: [+stop] [−voice] p t k
[+voice] b d g
[+affricate] [−voice] tʃ
[+voice] dʒ
——————————————————————————————————
Study Exercise #79, Question (c)
Find the correspondence series for bilabial stops and reconstruct the ancestor sounds.
Answer to Study Exercise #79, question (c)
There are two series:
A B C
p b b as in 2, 5, 12, 16, 26-28, etc.
p p p as in 11, 24, 25, 30, etc.
Two possibilities: one single proto-sound *p, with it changing to [b] in some context in
Languages B and C. Or, two proto-sounds *p and *b, with a merger to p in Language A.
It seems pretty hopeless to find a context into which *p could have evolved into b: look for
instance at
B 30 [pomo] vs. 31 [bono]
or at
B 25 [peka] vs. B 26 [bego].
So it seems more sensible to set up two proto-stops, as follows:
*p
*b
and assume
b > p in A.
This sound change works perfectly for the data, since there are no [b] in Language A.
—————————————————————————————————————
Study Exercise #79, Question (d)
Find the correspondence series for alveolar stops and reconstruct the ancestor sounds.
Answer to Study Exercise #79, question (d)
This works very similarly. There are two series.
tdd as in 6, 9, 41, 42, 44, 45, etc.

t t t as in 10, 23, 40, 43, etc.
It looks pretty hopeless to try to derive the modern [d] from *t—for instance, why would t become
d in Language B 42 [depo], but remain t in Language B 43 [tewe]? Better to assume that *t and *d
were proto-sounds, and that the distinction got wiped out everywhere in A.
*t
*d
*d > t in A
This is actually encouraging, because it’s entirely similar to the bilabials above. So it now
becomes sensible, indeed imperative, to look at the velars.
——————————————————————————————————————
Study Exercise #79, Question (e)
For the moment, ignore the palato-alveolars. Find the correspondence series for velar stops and
reconstruct the ancestor sounds.
Answer to Study Exercise #79, question (e)
With the hint that we ignore palato-alveolars, the data look very much like the data for the last
two cases.
kgg as in 3, 26, 29
kkk as in 1, 2, 11, 12, 13, 24, 25, 27, etc.
As before, it’s very unlikely that the voicing distinction arose by a sound change — compare
25 and 26 in Languages B and C. So we can set up:
*k
*g
*g > k in A
——————————————————————————————————————
Study Exercise #79, Question (f)
Generalize your findings of the last three sections using features.

Answer to Study Exercise #79, question (f)
Stop Devoicing
*[+stop] > [−voice] in A
——————————————————————————————————————
Study Exercise #79, Question (g)
What are the vowel inventories of A, B, and C? Form a chart listing the vowels by their features.
Answer to Study Exercise #79, question (g)
[ieaou], [ieaou], and [iau], respectively. Here is the chart:
[−back] [+back]
+high
–low  E i u
–high
–low  E e o
–high
+low  E a
——————————————————————————————————————
Study Exercise #79, Question (h)
Find the correspondence series for vowels and reconstruct. To save time, here is a hint: the
original system had five vowels.
Answer to Study Exercise #79, question (h)
aaa 1, 2, 10, etc.

eea 4, 14, 21, etc.
ooa 2, 7, 8, 11, 12, etc.
iii 3, 7, 8, 9, 10, etc.
uuu 1, 10, 12 etc.
Given the hint, it’s a fairly obvious move to set up this proto-vowel system:
*i, *e, *a, *o, *u
and then assume a massive wiping out of distinctions in C: all three of [e,a,o] emerged as [a].
Mid Vowel Lowering
*e, *o > a in Language C
In features, this would be:
+syllabic
*–high  > [+low, +back]
E
——————————————————————————————————————
Study Exercise #79, Question (i)
Collect local environments for [k], [tʃ], [g], [dʒ] in Language C. Retain the original data next to
them. What vowels can follow k, g in C?
Answer to Study Exercise #79, question (i)
[k]:
No. Language A Language B Language C environment

1 kaku kaku kaku [ ___a
1 kaku kaku kaku a___u
2 kapo kabo kaba [___a
11 kopa kopa kapa [___a
12 kopu kobu kabu [___a
24 pako pako paka a___a
25 peka peka paka a___a
27 pika bika bika i___a
51 kupa kupa kupa [___u
[g]:
3 kawi gawi gawi [ ___a
26 peko bego baga a___a
29 poku pogu pagu a___u
52 kuma guma guma [___u
[tʃ]:
9 kita kida tʃida [ ___i
10 kitu kitu tʃitu [ ___i
19 naki naki natʃi a___i
28 poke boke batʃa a___a
33 puke puke putʃa u___a
36 rike like ritʃa i___ a
[dʒ]:

4 kene gene dʒana [ ___a
5 kepi gebi dʒabi [ ___a
6 keta geda dʒada [ ___a
7 kiko giko dʒika [ ___i
8 kiko giko dʒika [ ___i
13 kuki kugi kudʒi u___i
40 taki tagi tadʒi a___i
In C:
[k] and [g] can be followed by [a] or [u].
[tʃ] and [dʒ] can be followed by [a] or [i].
——————————————————————————————————————
Study Exercise #79, Question (k)
See if you can find a solution in which there were only *k and *g in Proto-ABC, with all instances
of [tʃ] and [dʒ] resulting from sound change. The big challenge is that in C, both [k] and [tʃ] can
occur before [a], and likewise both [g] and [dʒ] can occur before [a]. Hint: look at the original
vowel of the [a]’s preceded by [tʃ] and [dʒ], versus the original form of the [a]’s preceded by [k]
and [g].)
Answer to Study Exercise #79, question (k)
Cases where *k evolved into [tʃ]:

Environment Example
• followed by an [i] 9 [tʃida]
• followed by an [a] that used to be an [e] 28, [batʃa]; compare B [boke]
Cases where *k stayed [k]:
• followed by an [u] 1 [kaku]

• followed by an [a] has always been [a] 1 [kaku]; compare B [kaku]
• followed by an [a] that used to be an [o] 11 [kapa]; compare B [kopa]
So it looks like *k evolved into [tʃ] just in case—at the time—it was followed by [i] or [e]. These
are the [−back] vowels.
We can confirm this with the voiced counterparts [g] and [dʒ].
Cases where *g evolved into [dʒ]:
• followed by an [i] 7, [dʒika]

• followed by an [a] that used to be an [e] 6, [dʒada]; compare B [geda]
Cases where *g stayed [g]:
• followed an [u] 29 [pagu]

• followed by an [a] that has always been [a] 3 [gawi]; compare B [gawi]
• followed by an [a] that used to be [o] 26, [baga]; compare B [bego]
So it looks like *g evolved into [dʒ] just in case—at the time—it was followed by [i] or [e], that is,
by a front vowel.
Here is everything set up by sound changes.

(316) Palatalization237
–velar
 
+palato-alveolar
 / ___ –back 
+velar +syllabic
*+stop  > –stop
E E E
+affricate 
“Velar stops evolved into palato-alveolar affricates when they preceded a front vowel.”
Historically, Palatalization must have taken place before Mid Vowel Lowering, since it was
triggered by proto-*e, before *e was converted to [a].
——————————————————————————————————————
Study Exercise #79, Question (l)
Provide historical derivations for 15, 3, 4, 8, 9, and 33, in each language.
237
This is the general term for any phonological rule or sound change that moves sounds into the
general territory of the hard palate (including not just the palatal place of articulation, but also the palato-
alveolar).
Answer to Study Exercise #79, question (l)
Language A:
15 3 4 8 9 33
*maru *gawi *gene *giko *kida *puke Proto-ABC
— k k k t — Stop Devoicing
maru kawi kene kiko kita puke Language A
Language B:
15 3 4 8 9 33
l R to L
malu gawi gene giko kida puke Language B
Language C:
15 3 4 8 9 33
— — dʒ dʒ tʃ tʃ Palatalization
— a a a — a Mid Vowel Lowering
maru gawi dʒana dʒika tʃida putʃa Language C
Thus, we see Proto-ABC as having had a fairly simple phonological system, with the six stops
[ptk bdg], various other consonants, and five vowels [ieaou]. The voicing contrast was wiped out
in A. C underwent a fairly complex chained development, first developing the palato-alveolars
from velars before front vowels, then radically simplifying the vowel system to just [iau]. In B, a
trivial change shifted *r to [l].
8. The reconstruction of Proto-Indo-European
The greatest achievement of the comparative method has been the reconstruction of Proto-
Indo-European. Indo-European is so-called because the Indo-European languages in their original
territory (before the age of Western expansion) stretched from Europe to India. Proto-Indo-
European was reconstructed over a long period of research that spanned most of the 19th and early
20th centuries; the details are still being worked out today. The field of historical linguistics in fact
was developed mostly as a result of the efforts to understand the relationships of the Indo-
European languages.
The Indo-European family was mentioned above in connection with the concept of descent.
Here is a more detailed family tree given in outline form. Extinct languages are shown in italics.
(317) The Indo-European family tree
Italic, comprising
Latin and its modern descendents, the Romance languages
various ill-attested ancient languages of Italy
Greek (Ancient Greek, Medieval Greek, Modern Greek)
Indo-Iranian, comprising
Indic (Sanskrit, Hindi, Bengali, Marathi, Sinhala, many others)
Iranian (Persian, Pashto, Kurdish, others)
Balto-Slavic, comprising
Baltic (Latvian, Lithuanian)
Slavic (Russian, Ukrainian, Polish, Czech, Serbo-Croatian, Slovenian, Bulgarian,
Macedonian)
Germanic (see above)
Celtic (ancestor of Irish, Scots Gaelic, Welsh, Breton, Gaulish, Cornish)
Albanian
Armenian (today attested in two main daughter languages, Eastern and Western Armenian)
Hittite (Turkey, earliest written records of any Indo-European language)
Tocharian (Central Asia)
The reconstruction of the family was made easier by the fact that so many branches of the
family are attested in very old written documents; roughly 1700 BCE for Hittite, 1500 BCE for
Sanskrit, 1200 BCE for Mycenaean Greek.
One can find numerous foreign words that descend from the same Proto-Indo-European root
as familiar English words. These words are familiar, because English has borrowed heavily from
Latin and Greek. The following table gives some examples.
PIE
English father *pəter

Latin pater (cf. paternal)
Greek pater (cf. patriarch)
Persian pedær
English heart *kerd

Greek kardia (cf. cardiac)
Latin kord (cf. cordial)
English bear
Latin fer (cf. transfer) *bher
Greek pherein (cf. amphora) ‘vessel to carry things in’
English two
Latin duo (cf. dual) *dwo
Greek dis (cf. disyllabic)
Armenian erku
Proto-Indo-European is believed to have been spoken about 6000 years ago, give or take a few
thousand years. The Armenian form erku in the table gives an idea of how far a word can evolve
through sound change in this amount of time.
9. Grimm’s Law
You’ll see in the examples above that the consonants of Germanic generally deviate from
those of the remaining Indo-European languages. This is due to what is probably the most famous
of all sound changes, Grimm’s Law. In very rough outline, Grimm’s Law looked like this:
Proto-Indo- Proto-
European Germanic
ptk > f  h238
bdg > ptk
b d g > bdg
Here are examples:
238
On grounds of phonetic symmetry we would expect a voiceless velar fricative [x]. This probably
was an intermediate stage on the way to [h]; for example, in Polish [x] can be optionally pronounced [h].
father, Latin pater three, Greek treis heart, Greek kardia
hemp, Greek kannabis two, Latin duo knee, Latin genu239
brother, Latin frater240 do, Sanskrit da-241 guest, Latin hostis242
The American Heritage Dictionary is to my knowledge the only dictionary that bothers to take
the etymologies all the way back to Proto-Indo-European. You can find the original roots for these
correspondences in their Indo-European appendix:
*pəter *trei *kerd

*kannabis *dwo *genu
*bhraːter *dheː *ghosti
10. The method of reconstructed environments
The most virtuosic application of the Comparative Method uses a technique that, oddly, has no
standard name. To fill this gap, I will call it here the method of reconstructed environments
here.
In the method of reconstructed environments, the environment for a sound change in

Language A, which is no longer present in A, is determined using data from sister language B.
The method was already illustrated in the Proto-ABC example above. We used the vowels of
A and B to solve the problem of the sound change k g > tʃ dʒ in C.
Proto-ABC is modeled on a real-life case, namely the history of Sanskrit, of which the
following data are representative.
Latin Old English Greek Sanskrit gloss

-kwe — -te -tʃa ‘and’
kwis hwa tis tʃid ‘who’
kwod hwæt — -kas ‘what’
— hwæer poteros kataras ‘which of the two’
kwando hwanne — kada ‘when’
The correspondence series here are these:
239
[kni], until about 1700
240
The Proto-Indo-European b, preserved in Sanskrit batar, became f in Latin.
241
Meaning “to set”.
242
The Proto-Indo-European g became h in Latin.
Latin Old English Greek Sanskrit

kw hw t tʃ (first two)
kw hw p k (last three)
Normally, these are attributed to Proto-Indo-European *kw, which survived intact in Latin and
became [hw] in Germanic by Grimm’s Law. In Greek, the fate of *kw depended on the following
vowel: if this vowel was front, *kw evolved into [t], as in the first two rows; otherwise *kw
evolved into [p].
It is the Sanskrit forms that are the puzzle: they show sometimes [tʃ], and sometimes [k], but
in exactly the same environment, namely before [a].
The solution to the problem is to use the method of reconstructed environments. The crucial
insight is that the Sanskrit vowel inventory is missing vowels found in its sister languages, namely
the mid vowels [e] and [o]. If we consider just Greek poteros vs. Sanskrit kataras, it is plausible
that the Sanskrit vowel were (at some pre-attested phase of Sanskrit) the same as the Greek ones,
and that there was a merger:
(318) Mid Vowel Lowering
*e, *o > a in Sanskrit
In other words, we use Greek and Latin as a guide to the former quality of the Sanskrit
vowels. This lets us explain the behavior of *kw, as follows:
*-kwe *kwid *-kwos *kwoteros Pre-Sanskrit

tʃe tʃid — — *kw > tʃ before front vowels
— — kos koteros *kw > k elsewhere
tʃa — kas kataras *e, *o > a
tʃa tʃid kas kataras attested Sanskrit
This account both rationalizes the gap in the Sanskrit vowel system, and explains the development
of [tʃ] from *k.243
The method of reconstructed environments was introduced as a technique by several scholars

more or less simultaneously during the 1870’s, and marked the maturity of reconstruction as a
method. Further developments have mostly followed developments in phonology: we can make
better guess about old phoneme inventories by the study of what are typically phoneme inventories
243
Curiously, the very same pattern appears in the history of Salishan languages (northwestern United
States). Nez Perce plays the role of Sanskrit here. The scholars who reconstructed proto-Salishan presumably
didn’t have as hard a time figuring this out, since they already had the Sanskrit example to work with.
today; and our increased knowledge of phonological rules in the world’s languages permits more
informed guesswork about old sound changes.
11. Validating the comparative method
The best way to evaluate the comparative method is to apply it to a language family whose
ancestor is known from written evidence. Plausible candidates:
• Apply method to Romance languages, compare result with Latin

• Apply method to Hindi, Bengali, etc., compare result with Sanskrit
• Apply method to Slavic languages, compare result with Old Church Slavonic
• Apply method to Mandarin, Cantonese, etc.; compare result with oldest written Chinese
The result is generally encouraging, but also shows the limitations. Thus, Proto-Romance, the
reconstructed answer of the modern Romance languages, is not unsimilar to Classical Latin, but
departs from it in many important ways. Similar conclusions follow, I believe, in the other
examples just given.
11.1 Confirming Proto-Germanic Reconstructions
Opportunities to confirm Proto-Germanic reconstructions directly are almost non-existent, but

a famous case of this sort is often mentioned. The reconstruction is of interest, because it shows
how knowledge of phonology and sound change in general guides reconstruction.
The following forms are the oldest attested versions in Germanic languages of the word
“guest”:
Gothic gasts
Old Norse gestr
Old High German gast
Old English gæst
Given this data, a historical linguist experienced in the typical sound changes found in
languages might reason as follows:
• The final consonant of Gothic and Old Norse is plausibly the result of a long-lost [z]—this
sound can become [r] by weakening from fricative to liquid, and [s] by assimilating the
voicing of a preceding [z].
• Long consonant clusters are historically usually the result of the loss of vowels; thus
*gVstVz.
• The absence of the *z in some of the daughter languages (Old High German, Old English)
is hardly surprising, given the tendency of languages to simplify their consonant clusters.
• Again on the basis of examples seen elsewhere, it is likely that the Gothic and Old High
German vowels ([a]) represent the original form, and that the front vowels of Old Norse
and Old English are the result of assimilation: the vowel of the stem becomes front under
the influence of a following front vowel.244 The mostly likely such vowel is [i]—it is the
most common trigger of this kind of process, and is also the most likely vowel to delete.
• Thus, the ancestor form was plausibly *gastiz, and the history of the descendent forms is
perhaps something like this:
Gothic Old Norse OHG Old English

*ˈgastiz *ˈgastiz *ˈgastiz *ˈgastiz Proto-Germanic
— ˈgestiz — ˈgæstiz vowel assimilation
— ˈgestir — — weakening of z to r
ˈgastz ˈgestr ˈgastz ˈgæstz loss of stressless vowel
— — ˈgast ˈgæst cluster simplification
ˈgasts — ˈgast ˈgæst voicing assimilation
ˈgasts ˈgestr ˈgast ˈgæst attested forms
This is going fairly far out on a limb, and can only be called informed conjecture. Yet in
this case the conjecture was pleasingly confirmed by an archaeological discovery; a horn found
in southern Denmark, dated to about 400 A.D—only shortly after the breakup of Proto-
Germanic. The runic inscription on the horn is transcribed thus:
Ek Hlewagastiz Holtijaz horna tawido

I, Hlewagastiz, son of Holti, made (this) horn.
From http://alcor.concordia.ca/~shannon/335PP/Lecture01Germania.ppt#270,11,Runes
e k h l e w a g a s t i z | h o l t i j a z | h o r n a | t a wid o
244
Old Norse also shows a partial height assimilation.
Following the general pattern of early Germanic names, “Hlewagastiz” is interpreted as

“fame-guest”—thus giving gastiz as confirmation of the reconstructed Proto-Germanic form.
Latin hostis ‘enemy’ is taken to be further confirmation; its [h] is the normal counterpart of
Germanic [g]; and the two are thought to descend from Proto-Indo-European *gostis
‘stranger’—with opposite semantic drift in the two daughter language families.
In a similar case, the reconstruction for “king”:
Old English cyning

Old Frisian koning
Old Saxon kuning
Old High German kuning
Old Norse konongr
is taken to be *kuningVz, where V is some vowel that didn’t cause the stem vowel to become
front—probably a non-front vowel. Conveniently, this word was borrowed very early into Finnish
(not an Indo-European language), which preserved it in the form kuningas, essentially unaltered
(save for the z > s; Finnish has no [z]) for 2000 years.
11.2 Why the Comparative Method is imperfect
In spite of such gratifying examples, the more general truth is that the Comparative Method
cannot in general recover the prior state of languages intact, but only bring us closer to it than any
other procedure could. The problem is gradual data loss over time. If any part of a word is lost in
all of the daughter languages, it will not be recoverable by the Comparative Method. In section this
week, you’ll see some examples of reconstructed Proto-Romance, and you’ll see that they involve
very considerable differences from Classical Latin.
It is not just the sound that get irrecoverably lost. Whole words get replaced over time,
gradually removing the historical linguist’s raw material entirely. Thus, English marginally
preserves the Proto-Germanic word *hundo-z in the form of hound, but in general to refer to dogs
we say dog, of which the Oxford English Dictionary says:
“Late Old English; previous history and origin unknown”
Many words do not have etymologies—the best-informed scholars just plain don’t know. (OED on
big: “its derivation is entirely unknown”; on boy “of obscure origin”; on tag: “origin obscure”; on
miffed “origin uncertain”.)
12. How far back can we go?
Given the gradual loss of data over time, most linguists have been reluctant to pursue the
deeper ancestry of the Indo-European languages (and similarly for very deep relationships around
the world). It is generally agreed that the data aren’t sufficient to relate Indo-European to any of
the neighboring language families245 using the Comparative Method, and the debate hinges on
whether we are entitled to use any other method less rigorous than the Comparative Method, such
as merely combing through the data for resemblances that may well be quite accidental.
I believe most linguists are skeptical of such efforts. The world abounds in false cognates,
that is to say, words that look like they come from the same proto-word, but can be shown through
reasoning and evidence that they are not. A classic case is the Persian word [bæd], which means,
of all things, “bad”, but (as careful study of the sound correspondences and ancient Persian
documents will show) is not etymologically related to English “bad” at all.246
Thus, scholars who try to demonstrate deep relationships (of which the logical extreme is the
hypothetical “Proto-World”) risk the scorn of their colleagues. Typically a scholar who uses
“trans-comparative” scholarly methods will be regarded by a few colleagues as a visionary, and by
others as exhibiting scholarly irresponsibility.
The failure of the Comparative Method to go “really deep” is perhaps a bit sad, since it would
be nice to know the language our remote ancestors spoke. A useful comparison here is a parallel
discipline—evolutionary biology—that likewise has established the family trees of things (species)
through careful and systematic comparison. Evolutionary biology has better data—such as DNA
sequences—that have enabled biologists to reconstruct the unitary Tree of Life almost to its origin.
Historical linguistics, alas, only has words, which gradually get replaced over the centuries. The
complete Tree of Languages may be valid as a concept, but it cannot be accessed with the methods
we have and is unlikely ever to be.
An even less likely prospect is pinpointing when and how language first came to be. It seems
essentially certain that this required advances in human evolution, and, as we saw in Chapter 7,
some of the adaptations involved may have involved linguistic ability itself. But barring the
invention of time travel, we are not likely to find out much about the early stages of human
language.
13. Borrowing
Sound change is not the only way in which languages can change. Another important
mechanism is borrowing, the adoption of words from other languages. Over time, languages can
borrow thousands of words; indeed, Albanian is an Indo-European language, but it is of little use
in reconstructing Indo-European, because it has borrowed so heavily from other languages that
there are only a few hundred native Albanian words left.
245
Candidates include Uralic (Finnish, Hungarian, Estonian, etc.), Altaic (Turkish, Mongolian, etc),
Basque, and others.
246
The Middle Persian form is recorded as vat, more distant already…
Study Exercise #83
a. Use your knowledge of the sound changes developed earlier to predict what will be the
German words for to and pepper.
b. Given this, what would you expect the German word for party (in the sense of ‘political
party’) to be?

a. The German for to is /tsu/, spelled zu. The German for pepper is /pfɛfər/. Obviously, this
procedure doesn’t work all the time, since many other sound changes separate German and
English.
b. /pfartsi/. This is actually not right; see immediately following discussion.
————————————————————————————————————
Borrowing makes trouble for the Comparative Method. The difficulty is that words that are
borrowed after a given sound change look like exceptions to that sound change. The German for
party is in fact not /pfartsi/ but rather /partaɪ/. The word was borrowed from French, long after the
sound change that converted *t and *p into affricates.
In this particular case, the difficulty is not great. We have extensive old records of both
German and French, and it is not difficult to trace the history of the word through both languages.
But in other cases there is no documentation.
The procedure used in such cases is more subtle. Usually, one does a tentative reconstruction
based only on basic, core vocabulary items that are not often borrowed—words like father, arm,
moon, three, water, etc. From these basic words, one can get a rough idea of the sound
correspondences.
Once this is done, the sound correspondences themselves can be used to check for borrowings.
That is, the words that violate known sound correspondences are likely to be the borrowed words.
Study Exercise #84
Consider the following correspondences:

fish fʃ fʃ *fisk
shoe ʃu ʃu *sku
flesh flɛʃ flaʃ *flesk
In these cases, we have [ʃ] in English matched with [ʃ] in German matched with [sk] in
Swedish.
The English words skirt and shirt are both descended from the same Proto-Germanic root.
One of them is a borrowing, the other is native. Which is which?
Skirt is borrowed from English in Old Norse around the time of the partial Danish conquest of
England. The Old Norse form was skyrta. The form is recognizable as a borrowing because all
native *sk clusters had been converted to [ʃ].
Shirt and skirt were the same word in Proto-Germanic, reconstructed by the Oxford
English Dictionary as *skurtjon.
———————————————————————————————————
Once one has filtered out the borrowings, one can use the words that remain to get a better
idea of the sound changes. With this done, one can make a more accurate judgment of which
words are borrowed, which then permits a through a series of gradual improvements.
14. Grammatical simplification
I will discuss one further mechanism of language change: grammatical simplification. The
basic picture is this: sound changes over time tend to make the grammar of a language,
particularly its morphological rules, very complicated. In compensation, languages often
spontaneously simplify their morphological rules.
I will first show how sound change complicates the morphological rules. An example of
complexity in morphology is the set of irregular plurals in English, such as foot-feet, mouse - mice.
These are exceptions to the normal pattern of plural formation in English, which would lead us to
expect foots and mouses.
In the theory of inflectional morphology given in the course, a form like feet must be listed in
the lexicon, with its phonological form and a sort of pre-formed morphosyntactic representation.
Here are sample lexical entries for foot and feet:
foot
/ft/
feet
/fit/[Number:plural]
‘appendage at end of leg’
The theory of lexical insertion must stated such that, whenever there is a special listed entry like
feet, that entry is lexically inserted, and the form that would be derived by the rules of the
inflectional morphology, namely foots [fts], is preempted.
14.1 The origin of irregular forms
The existence of irregular forms can, in most cases, be attributed to sound changes of long
ago. The plurals feet and mice are in fact the historical descendents, through sound change, of a
system that was quite regular thousands of years ago, in Proto-Germanic times. What made them
irregular was a lengthy sequence of sound changes. I will go over them briefly here.
Here are the reconstructed forms for foot, feet, mouse, and mice in Proto-Germanic (around
500 BCE):
foot feet mouse mice
*ˈfot *ˈfot-i *ˈmus *ˈmus-i
Notice that there is nothing particularly irregular about them. The plural is formed by
attaching a suffix of the form -i, which in fact was the regular plural suffix for this class of nouns.
In the system of inflectional morphology used in this course, the rule would have been
(approximately) the following:
Early English Plural Formation
Suffix -i when [Number:plural]
The first step towards irregularity for these words was an innocent-looking phonological rule,
which created front vowel allophones of the back vowels /o/ and /u/:
Early English Umlaut
+syllabic
*+round  > [−back] / ___ [−syllabic] i
E
This change produced the following forms:

ˈfot ˈføt-i ˈmus ˈmys-i
To understand the next change, you need to know that in Proto-Germanic, the first syllable of
a word (and only the first syllable) was stressed. The next sound change converted all the stressless
vowels into schwa:
Vowel Reduction
+syllabic
*–stress  > ə
E

ˈfot ˈføtə ˈmus ˈmysə
This is reminiscent of how German acquired the phoneme /ø/ (see section 5 of this chapter,
above). In fact, pretty much the same thing happened in early English: when the triggering
environment for an Umlaut rule was lost, the language acquired front rounded vowel phonemes.
The next step in English was to lose the schwas:
Final Schwa Drop

*ə >  / ___ ]word

ˈfot ˈføt ˈmus ˈmys
Then the vowel /ø/ lost its rounding, and became the corresponding unrounded vowel /e/:
ø Unrounding
ø  –round]

ˈfot ˈfet ˈmus ˈmys
Once we have reached this stage, we are no longer relying on reconstruction. The above forms
appear in the oldest written documents for Old English.
• Beowulf 745 Sona hæfde unlifiendes eal efeormod fet and folma
‘swiftly thus the lifeless corse was clear devoured, even feet and hands.’
• 1297 He vel of is palefrey, & brec is fot.
‘He fell off his horse and broke his foot’
• Late Old English: King Alfred’s translation of Boethius’s The Consolation of Philosophy:
Gif ge nu gesawan hwelce mus þæt wære hlaford ofer ore mys
‘If you saw in a community of mice, one mouse asserting his rights and his power over
the others’
Around 1050 to 1100, the front rounded vowel /y/ underwent the same fate that /ø/ had
undergone earlier: it lost its rounding, becoming the corresponding front vowel /i/:
y Unrounding
*y  –round]

ˈfot ˈfet ˈmus ˈmis
Around 1500, for reasons that are not known, the tense vowels of English suffered a
convulsive change, which sent them all over the phonetic chart. This change is called the Great
Vowel Shift, and it marks the boundary between Middle English and Early Modern English.
Great Vowel Shift

i  a u  a
e  i o  u
æ  e ɔ  o

ˈfut ˈfit ˈmaʊs ˈmaɪs
Our words are now in recognizably modern state. There was one more sound change: the
vowel /u/ became lax in certain environments, in a complex and somewhat irregular change:
/u/ Laxing
*u  [−tense] in certain environments

ˈfʊt ˈfit ˈmaʊs ˈmaɪs
This is the end of journey of these vowels, for now. It is interesting to plot their trajectories on
a phonetic chart, to see how far the vowels have migrated in 2500 years:
The vowel of mus:
a
The vowel of musi:
i y u
a
The point of this example is to show that 2500 years of sound change can make a very simple
morphological rule into a complex one. It would be very hard to write a general rule that predicts
mice as the plural of mouse and feet as the plural of foot.
14.2 Grammatical regularization as a source of change
In fact, the language didn’t really tolerate the situation. At some point in the history of
English, the old, increasingly irregular system of plural formation was discarded and replaced by a
simpler rule. Basically, in Modern English plurals are formed by suffixing -z.247
Modern English Plural Formation
Suffix -z when [Number:plural]
The plurals mice and feet are relic forms; they have managed to hang on as exceptions to the
general rule.
The change in the system of plural formation in English is a classical case of grammatical
simplification. The language changed not through sound change, but in response to sound change.
It created a new rule for plurals, and replaced most of the old irregular plurals with newly created
forms.
Who is responsible for grammatical simplification? The most likely answer is small children,
who are still acquiring language. It is not hard to see why: one constantly observes small children
oversimplifying the grammar of the language they are learning. In particular, they don’t know, or
neglect to use, the special lexical entry for forms like feet. Instead, they generate foots using the
regular grammatical system. In some cases, particularly with less common words, such regularized
forms can be adopted by the speech community as a whole.
An example: the plural of cow was once [ka], or something like it (note the archaic form
kine). [ka] is the plural inherited though sound change from Proto-Germanic; its history is
essentially the same as that of mice, with the same vowel. The plural we use today, cows, was the
invention of children. It differs from foots only in that it managed to get adopted for general use.
Quite a few forms in English today are creations of children, of this kind. Another plural form
of this type is brothers (formerly brethren) and the past tenses helped (formerly halp) and melted
(formerly malt).
The upshot of this is that language change can be thought of as an eternal struggle. Over the
centuries, sound change alters the morphological system, making it more complex and obscure.
Fighting on the other side are small children, who refuse to learn the irregular forms, and replace
them with regular forms, as generated by the rules of the language at the time they learn it. The
current state of a language is the result of a temporary balance between these opposing forces.
247
There is a bit of phonology going on: the underlying /-z/ becomes [-s] after voiceless consonants
(cats, with /kæt-z/ → [kæts]) and a schwa is inserted to break up clusters of the form [s, z, ʃ, , tʃ, d] + [z], as
in badges (/bæd-z/ → [bædəz].)
15. Summary of historical linguistics
At this point we have covered the basic mechanisms of language change. An outline of the
field is as follows:
First, all languages have phonological rules. Phonological rules are vulnerable to restructuring
by the next generation, which results in sound change. Sound change is normally regular. It is this
regularity that makes it possible to reconstruct lost proto-languages, using the Comparative
Method.
Borrowing is another major source of language change. Borrowed words make the
Comparative Method more difficult to apply, but they can often be detected because they are
exceptions to the sound correspondences.
A third major source of language change is grammatical simplification, the abandonment by

children of irregular forms resulting from sound change in favor of regular forms. Sound change
and grammatical simplification are in eternal conflict: sound change complicates the morphology,
and grammatical simplification “repairs” the damage.
The Comparative Method yields well-supported family trees and the changes that the
languages underwent during their descent. It cannot go back more than a few thousand years and
thus the deep history of languages, as well as the origin of language in general, is not accessible to
investigation by this method.
Study Exercise #85: Historical Linguistics
Here are matched sets from three dialects of English. Apply the Comparative Method, forming
correspondence sets and positing sound changes. Here, is it best to compare sequences rather than
sounds. Do: [juɹ, uɹ, oɹ].
Dialect A Dialect B Dialect C

1. Muir [ˈmju] [ˈmju] [ˈmjo]
2. moor [ˈmu] [ˈmu] [ˈmo]
3. more [ˈmo] [ˈmo] [ˈmo]
4. cure [ˈkju] [ˈkju] [ˈkjo]
5. Coors [ˈkuz] [ˈkuz] [ˈkoz]
6. core [ˈko] [ˈko] [ˈko]
7. Buhr [ˈbju] [ˈbju] [ˈbjo]
8. boor [ˈbu] [ˈbu] [ˈbo]
9. bore [ˈbo] [ˈbo] [ˈbo]
10. endure [ɛnˈdju] [ɛnˈdu] [ɛnˈdo]
11. dour [ˈdu] [ˈdu] [ˈdo]

12. door [ˈdo] [ˈdo] [ˈdo]
13. Turing [ˈtju] [ˈtu] [ˈto]
14. tour [ˈtu] [ˈtu] [ˈto]
15. tore [ˈto] [ˈto] [ˈto]
16. inure [ˈnju] [ˈnu] [ˈno]
248
17. Koh-i-noor [ˈkohnu] [ˈkohnu] [ˈkohno]
18. nor [ˈno] [ˈno] [ˈno]
248
A famous diamond, from the Persian for “mountain of light”.
Correspondence sets:
Proto A B C Examples
*ju ju ju jo 1, 4, 7
*ju ju u o 10, 13, 16
*u u u o 2, 5, 8, 11, 14, 17
*or o o o 3, 6, 9, 12, 15, 18
The proto-language is identical to Dialect A.
B and C have both undergone:
*j    +alveolar] ___
See endure, Turing, and inure
C has also undergone:
*u > o / ___ 
which has merged moor with more, boor with bore, and so on.
For Further Reading
Two textbooks in historical linguistics I have enjoyed are the following. Introduction to
Historical Linguistics by Anthony Arlotto (1981: University Press of America) is very brief and
quite clear; Historical Linguistics by Theodora Bynon (1979: Cambridge University Press) goes
into greater depth. Leonard Bloomfield’s Language (1933, still in print), cited above for
phonology, has a wealth of good material on historical linguistics.
Chapter 14: Applications and Outlook
1. This text and linguistics in general
In this text I’ve presented one theory for each area, and for almost every particular area of
data, one analysis. This has given us the tools to analyze a fair amount of data, and to illustrate
what it means to do linguistic analysis. For purposes of an introductory text I’ve mostly picked
theories on the basis that they can be taught in a short period of time, and specialist courses can be
more ambitious. If you study linguistics further, you will learn more elaborate theories.
In current research, linguists explore many different theories, and try to find evidence for
which one is right. As research proceeds, the theories have tended to become more subtle, more
ambitious, and more accurate.
2. Unsolved research problems in linguistics
One indication that linguistic theory is making progress is that descriptive grammars are
getting better. Grammar authors, equipped with better theories, and better knowledge of what
languages are like in general, seem to be able to lay out languages more completely and
systematically than their predecessors 50 or 100 years ago.
On the other hand, I feel that most progress in linguistics is yet to come, and the linguistics of
100 years from now may well be quite different from the linguistics of today. For purposes of
illustration, here are what I take to be three of leading unsolved research problems in linguistics.
2.1 The island problem
Our islands (Chapter 6) consisted of an arbitrary-seeming list of syntactic structures, some of

them perhaps universal, others language-particular. Many other islands exist as well. One area
where theorizing has been intensive is the attempt to unify and simplify the theory of islands. An
approach that is commonly taken is that it’s probably better to specify where Wh-phrases can be
extracted from rather than making a big list of where they can’t. No current theory has obtained the
agreement of all specialists.249
2.2 The acquisition problem
We linguists solve linguistics problems through patience, guile, and occasionally inspiration.
It is unlikely that children learn language this way, since they seem more reliable than we are —
they proceed steadily onward to become fluent native speakers. To solve the problem of how
children acquire language so well, three things will be needed. First, we need to develop adequate
grammars of individual languages, which characterize the native speaker’s knowledge and
249
A quick, pessimistic, overview may be read at https://muse.jhu.edu/article/539657/, accessible via the
UCLA library and many other university libraries.
intuitions accurately. We also need adequate general theories of language that say what grammars
can be like. Both of these issues have been taken on, at least in an elementary way, in this course.
The next step would be to start modeling the child’s behavior directly: linguistics is getting
started on the task of developing formal systems (probably implemented as computer programs)
that mimic the child, learning grammars when exposed to realistic data from languages.
One of the very simplest such problems to learn a grammar that can form the past tense of
English verbs, given the present stem. The rules of the game are that the system is given a set of
verbs (perhaps a couple thousand) with their past tense, learns a grammar, and then is tested on
new verbs. One system of this sort250 when asked for the past tense of “spling”, guesses as follows:
(319) Three machine-generated guesses for the past tense of “spling” [ˈsplɪŋ]
splung [ˈspləŋ] first choice

splinged [ˈsplɪŋd] close second
splang [ˈsplæŋ] third choice
These guesses roughly matches the preference of people, who vary in the same way. Many vastly
harder tasks in modeling learning have yet to be addressed, since we don’t yet know how.
An important element of future learning systems is that they should not necessarily be
maximally accurate! Human children learning a language sometimes get it wrong, and
occasionally get it spectacularly wrong, producing (when the mislearned system is adopted by the
speech community) a major change in the language across time.251 I suspect that the task of
modeling failure (in a scientifically useful way) is likely to fall primarily to linguists, since at least
at present computer science is focused on achieving accuracy, which of course is of great value in
the practical world. Ultimately, I think, linguistics should try to pass the “Turing test”, as it applies
for language—the creation of an artificial system that behaves identically in all respects to human
learners.
2.3 The parsing problem
A parser is a procedure (usually embodied as a computer program) that, given a grammar and
a sentence, can figure out the phrase structure tree that the grammar assigns to the sentence. One
problem in parsing is that sentences often have many more parses than we as linguists think they
do. To give one example, the sentence:
They are flying planes.
Adam Albright and Bruce Hayes (2003) “Rules vs. Analogy in English Past Tenses: A
250
Computational/Experimental Study,” Cognition 90: 119-161

[http://www.linguistics.ucla.edu/people/hayes/#acquisition]
251
Some examples are given at
http://linguistics.ucla.edu/people/hayes/papers/HayesLSAPlenaryTalkSlidesJan8_2015.pdf.
has two obvious parses:
(They are acting as pilots)
(Those things up in the air are planes that are flying)
But a complete and thorough search yields parses that are absurd but possible. Thus, consider
the following set-up:
Smoking kills.
What are the facts? The facts are, smoking kills.
They are, smoking kills.
They are, flying planes.
(I envision small bits of a large board being slowly removed by impact with the propellers.)
The absurdity, indeed the “cheapness” of this example is perhaps even irritating, but it
illustrates a general problem. Parsers implemented as computer programs arrive at a great number
of parses that would never occur to people. In contrast, people seem to be able to arrive at the
correct parses almost instantaneously, without distraction. Much current research is devoted to
inventing parsers that can mimic the high level of human performance—partly in the hope that this
will shed light on how people perform this task.
Parsing is not just a matter of syntax. In morphological parsing, we seek to recover the stem
and the features of the morphosyntactic representation from the phonological form of an inflected
word, which may often be completely novel. In “phonemic parsing” — better known as speech
recognition — we seek to find the phonemic representation (as well as the lexical items present)
from a raw acoustic signal. Like syntactic parsing, morphological and phonemic parsing are only
partly solved problems, the topic of current research.
3. Linguistics: what is it good for?
Enrollment in undergraduate majors in linguistics has tripled in the U.S. since 2000.252 This is
a good thing, more or less, for linguistics departments. Is it a good thing for society?
I actually think it is; that is, I feel society would be better off if more people had knowledge of
linguistics. Some specific areas where linguistics has made or could make a difference in real life
are as follows.
252
See http://www.linguisticsociety.org/files/Annual_Report_2013.pdf, p. 10.
3.1 Teaching reading and writing
Children learn to read an alphabetic language do so, in part, by establishing correspondences

between the phonemes they learned in infancy and the letters used to spell the language. The
spelling system is like a code (letters encode phonemes), and many children have the experience of
“breaking” the code and thus suddenly becoming able to read novel words by decoding them
phonemically. This produces a major jump in reading ability.253
“Phonics” is the standard term in the teaching profession for what a linguist might call “the
system of letter-phoneme correspondences.” Phonics as a method for teaching reading was
eclipsed for a number of decades in the United States by an alternative “whole word” or “whole
language” method, which became controversial. The Congress appealed to the National Institutes
of Health to make a scientifically-guided comparison of the two methods, and the NIH panel
(reporting in 2000) came out firmly in favor of phonics.254
Beyond a theoretical rationale for phonics, I think there could be some useful further
applications of linguistics in the teaching of reading. In particular, it would pay for teachers to
know the local dialect in the area of their schools and in particularly to understand the phonemic
systems of their students. Here is an example: if a student has no phonemic distinction between []
and [ɛ] before nasals (saying, as tens of millions of Americans do,255 both pin and pen as [pɪn]),
then a reading teacher should not correct the student who reads pen as [pɪn] —this can only
confuse the student and undermine her confidence, given that she correctly interpreted the letters
in the context of her own phonemic system. Indeed, I think that that same thing would be sensible
for features — such as absence of // — that clearly mark the student as a speaker of a non-
standard dialect. For such a student [mf] counts as success, at least in local terms, in reading the
word myth.
In the later school years, children are taught to write in a standardized, normatively-defined
style. We can debate the merits of having such a style (see Chapter 3), but let’s just assume for
purposes of argument that ability to write in the standard variety is of sufficient value to students’
future lives that they ought to be taught it. Here, having teachers who understand syntax can help
in making clear to children what the requirements of this style are. One common instance arises in
sentences like the following.
Being in a dilapidated condition, I was able to buy the house very cheap.
253
One author writes, “once children can map sounds to letters … their reading vocabulary suddenly jumps to
nearly the number of words they can comprehend orally.” See Snider, Vicki A. (1995) A primer on phonemic
awareness: What it is, why it's important, and how to teach it. School Psychology Review 24: 443-456.
254
You can read their basic recommendations at
https://www.nichd.nih.gov/publications/pubs/nrp/findings
255
Geographically, older [] evolved to become [] before nasals in dialects of the South, the Southwest, the
southernmost parts of the Midwest, inland California, and in African-American Vernacular English, which originated
in the South.
In many English dialects, this sentence can have a meaning in which it is the house that is
dilapidated. However, this reading is not possible in the written standard, where the only possible
reading is one in which the speaker is dilapidated. Since people who command the written standard
often hold strong normative views (chapter 3) against the non-standard pattern, teachers can
protect their students from future harm by teaching them the standard pattern.
To do this, it is necessary to have at least an elementary concept of syntax. For example, an

English teacher might in some way convey to his students: “the implicit subject of a preposed
clause must refer to the subject of the main clause”.256 The concepts of clause, implicit subject,
and coreference—all covered in this text — are clearly needed in some way to get the idea across.
3.2 Teaching foreign languages
Language instruction can be either intuitive or structural. The latter approach, one lays out the
grammar in a systematic way, much as a linguist tries to do. The teaching of pronunciation varies
perhaps most of all. Some language textbooks give the student nothing but orthography, along with
the advice that they should imitate native speakers. In contrast, some texts include training in basic
phonetics. Although not in the United States, many language textbooks around the world actually
use the IPA as a tool for making the correct target pronunciation as clear as possible.
In some cases, linguistic theorizing has produced better descriptions of how the language
works, notably in Japanese and other tonal languages. It remains to be seen whether such
developments will help in language instruction. Here again, the question is whether the students
should be told “Listen closely to native speakers and mimic their pitch patterns” or given a clear
description of how the system works phonologically, then try to make adherence to the
consciously-learned system an automatic and habitual pattern.
3.3 Alphabet design
The members of many of the world’s speech communities are unable to write in their native
language because it has not yet been given an orthography. As mentioned in Chapter 11, phonemic
analysis is commonly used to determine what sounds need to be symbolized by letters in a new
spelling system.
3.4 Human-machine interaction
It is of course a goal of many people and companies that we will someday engage in fluent
conversations with computers and other machines; presumably when this happens our interactions
with machines will be far more convenient and helpful to us. However, those who use machines
and software for synthesis and recognition will know that neither of these capacities has yet
reached perfection. We are still at point where they can cause considerable frustration, for instance
when the speech recognizer cannot understand our utterances; falling short of perfection also
256
Strunk and White’s book of normative grammar, The Elements of Style (from which the example
above derives) says “A participial phrase at the beginning of a sentence must refer to the grammatical
subject.” This is vague in using the term “refer to”, but seems clear enough to be useful. Link:
http://orwell.ru/library/others/style/english/estyle.
implies fatigue when we try to listen to the unrealistic productions of synthesizers. What is needed
to make things better?
Different people will give different answers to this question. Obviously, the answer I feel most
sympathetic to is, “more and better linguistics.” For instance, we cannot hope to have a good
speech synthesizer until we have exquisitely detailed — and generalizable — knowledge of the
rules for English allophones, both within the word and across word boundaries within the phrase.
Whether this knowledge will take the form of a traditional rule-based linguistic description or
something different is not firmly established.257 The problem of speech recognition may also
benefit from deeper and more detailed phonetic description and grammars.
Syntax and semantics must also be invoked to improve the abilities of computers to converse
with us. We can get an idea of the state of advancement achieved here by examining the behavior
of the grammar checker included in a leading word processor. Examples like the following indicate
that the busy crew at Microsoft has gotten strikingly good at parsing long noun phrases and
making sure that the verb agrees in number with their head (sequences underlined are those
identified as a problem by the grammar-checker in Word 2010):
(320) Some verb agreement errors detected by Microsoft Word

The turtles is green.
The turtles are green.
The turtles in the pond is green.

The turtles in the pond are green.
The turtles that many of us believe to be swimming in the pond is green.

The turtles that many of us believe to be swimming in the pond are green.
The turtles that the ducklings that the wolves ate believe to be swimming in the pond is
green.
The turtles that the ducklings that the wolves ate believe to be swimming in the pond are
green.
On the other hand, any student who has learned the content of this text could tell what is
wrong with the ungrammatical sentences below, which the Word grammar checker fails to detect:
(321) Some verb agreement errors not detected by Microsoft Word

Which books do you think are on the table?
*Which books do you think is on the table?
*Which book do you think are on the table?
Which book do you think is on the table?
257
Even the phonemes apparently need work: a system of letter-to-sound rules must come into play when a
speech synthesizer encounters a novel word. Currently, I startle whenever I hear my Apple smartphone pronouncing
my home street Calvin Avenue as *[ˈkælvaɪn] — even the most elementary set of letter-to-phoneme rules should be
able to avoid generating this output!
That is to say, verbs must agree with their subject NP when it is in situ, prior to the possible
leftward displacement of that NP by Wh- Movement.258
Not surprisingly, there are industrial syntacticians, who develop detailed grammars for various
languages, and use the grammars to assign parses to sentences (as in the grammar-checking
application above.) There are also industrial semanticists, who attempt to extract meanings from
sentences in the primitive mentalese of computers.
Quite a few students from UCLA (both undergraduate and graduate) have gone on to careers
in “industrial linguistics.” Often, though not always, they have expertise in both linguistics and
computing.
258
In 2014 I asked the language research staff at Microsoft about this, and they told me they do have
Wh-movement in their grammar checker, and that it ought to be catching the agreement error described here.
Perhaps there is a bug in the implementation. Thanks to Bill Dolan and Karen Jensen for their help.
As of 2021, the verb agreement errors in the longer examples of (320) are no longer detected in the
current version of Word.
Chapter 15: More review problems
These are given in the same order in which the topics appear in the text.
Study Exercise #86: Inflectional morphology in an imaginary language
a. Set up inflectional rules to derive these forms. Be sure to state your rules in the correct
order. Give your rules names. Assume features [PossessorPerson, Number, PossessorNumber].
kitab ‘book’
kitabam ‘my book’
kitabi ‘your book’
kitabeʃ ‘his/her book’
kitabilam ‘our book’
kitabili ‘you-all’s books’
kitabil ‘their book’ (not a typo)
kitabim ‘books’
kitabimam ‘my books’
kitabimi ‘your books’
kitabimeʃ ‘his/her books’
kitabimilam ‘our books’
kitabimili ‘you-all’s bookss’
kitabimil ‘their books’ (not a typo)
b. Provide a derivation for kitabimilam ‘our books’.

a. Rules in correct order:
Number Rule
X → Xim if [Number:plural]
Possessor Number Rule

X → Xil if [PossessorNumber:Plural]
Possessor Person Rule

X → Xam if [PossessorPerson:1]
X → Xi if [PossessorPerson:2]
X → Xeʃ if [PossessorPerson:3, PossessorNumber:Singular]
Note that the third part of the Possessor Person Rule must include the feature
[PossessorNumber:Singular], because otherwise it would attach the suffix -eʃ in plurals, deriving
*kitabileʃ rather than the correct kitabil for ‘their book’.
b. Provide a derivation for kitabimilam ‘our books’.
The morphosyntactic representation is:

[Number:Plural, PossessorNumber:Plural, PossessorPerson:3]
kitab stem
kitabim Number Rule (since [Number:Plural] is present)
kitabimil Possessor Number Rule (since [PossessorNumber:Plural] is present)
kitabimilam Possessor Person Rule (since [PossessorPerson:3] is present).
———————————————————————————————————
Study Exercise #87: Non-concatenative morphology
These forms follow the rules of a language game that was developed for the television show
The Simpsons; they are excerpted from an article by the linguist Alan Yu. These are only a small
part of the overall data, covering disyllabic words with initial stress. Formalize the rule for creating
the “disguised” form of the word. Treat diphthongs as single vowels, in spite of their sequential
IPA transcription.
Word Normal Game Word Normal Game

pronunciation pronunc. pronunciation pronunciation
oboe [obo] [obmbo] wonder [wnd] [wndmd]
washing [w] [wm] music [mjuzk] [mjuzmzk]
opus [ops] [opmps] scramble [skæmbl] [skæmbmbl]
water [w] [wm] Kieran [kin] [kimn]
party [pi] [pmi] stinky [stki] [stkmki]
piggy [pgi] [pgmgi] joking [dok] [dokmk]
aura [] [m] table [tebl] [tebmbl]
purple [pɚpl] [pɚpmpl] listen [lsn] [lsmsn]
tuba [tub] [tubmb]
Answer to Study Exercise #87, Non-Concatenative Morphology
Simpsons Language Game Rule

X C V C* → X C  m  C V C*
1 2 3 4 1 2 2 3 4
This answer uses Kleene star, which we used earlier for syntax, to mean “any number of
consonants”.
Study Exercise #88: Wh- Movement and islands
Show why
*What donor might Sue wonder what books donated to the library?
is ungrammatical, given the Wh- Island Constraint below. In particular, first extract what books to
the lower Comp, then extract what donor to the higher Comp, showing the island violation
graphically.
Wh- Island Constraint

CP whose Comp contains a wh- phrase.
CP
*
Comp S
|
wh-
Deep structure (all wh- phrases in situ), with lower instance of Wh-Movement; also Subject-
Aux Inversion in upper clause:
Resulting tree, with subsequent movement of what donor into the higher Comp. This violates
the Wh-Island Constraint; the island is enclosed in a dotted box:
Since a wh-phrase is moved out of the island, the resulting sentence is ungrammatical.
————————————————————————————————————
Study Exercise #89: Semantics; anaphora
The wizards believe that the witches turned the girls into copies of each other.
a. Produce the phrase structure tree.

b. Show clausemates with brackets, show c-command with arrows.
c. Explain with reference to rule given below the possible reference of each other.
Each Other Reference
Each other may refer only to a plural c-commanding clausemate.

The wizards believe that the witches turned the girls into copies of each other.
a. Diagram/b. Show clausemates with brackets, show c-command with arrows.
The witches, the girls, and each other are all clausemates, but the wizards is not clausemates
with any of them.
Looking at the tree and the crucial NPs, we see the following relations of c-command:259
the wizards c-commands the other three NPs
the witches c-command the girls and each other
the girls c-commands each other
259
Recall how this is determined: go up one node from any NP, and anything dominated by this node
is dominated by this NP.
c. Putting it all together, we see that:
• the girls c-commands and is a clausemate of each other, and so can be coreferent with each
other
Scenario: The wizards believe that the witches turned Sue into a copy of Ellen, and
turned Ellen into a copy of Sue.
• the witches c-commands and is a clausemate of each other, and so can be coreferent with
each other
Scenario: The wizards believe that Alice, a witch, turned the girls into copies of Miriam
(another witch), and that Miriam turned the girls into copies of Alice.
• While the wizards c-commands each other, it is not a clausemate of each other, and so it
cannot be coreferent with each other
Scenario: Bob, a wizard, believes the witches turned the girls into copies of Ted, another
wizard; and Ted believes the witches turned the girls into copies of Bob.
Logically possible, but evidently not available linguistically.
————————————————————————————————————
Study Exercise #90: Semantics, Scope I
This sentence has a scope-based ambiguity.
Many people visit two islands.
i. Describe clearly in words the two meanings of these sentence.

ii. Using Quantifier Translation and Quantifier Raising, derive the logical forms for each
meaning.
Quantifier Translation
Replace
…
instead; etc.
Quantifier Raising
Left-adjoin a quantified NP to S, leaving behind a variable in its original location.
Describe clearly in words the two meanings of this sentence. Give a scenario of which it could
hold true.
Many people visit two islands.
i. Describe clearly in words the two meanings of this sentence.
(a) It is true of many people that they visit two islands (not necessarily the same two).
(b) It is true of two islands that many people visit them (not necessarily the same people).
ii. Derivation of logical forms
Surface structure:
Quantifier Conversion:
At this point, the meanings depend on the order in which the quantifier operators are raised.
(a)
Quantifier Raising I
Quantifier Raising II
“It is true of many people that they visit two islands”
(b)
Quantifier Raising I
Quantifier Raising II
“It is true of two islands that many people visit them.”
——————————————————————————————————
Study Exercise #91: Semantics / raising quantifiers different distances
Sentence: John has a plan to visit every city
a. Explicate both meanings in philosopher’s language.

b. Apply Quantifier Translation and Quantifier Raising to derive both meanings.
See previous question for the rules you need.
To answer this question, you’ll need a bit of help with the syntax, there being material that this
text does not cover. We’ll assume that the clause to visit every city is an S, and it has an NP subject
that is empty (but is interpreted as being coreferent to John). This is the same sort of empty subject
discussed in section 4 of Chapter 1 of this text, under the name “implicit noun phrases”. We’ll use
the standard notation for this empty subject, which is: PRO (it is essentially a kind of pronoun).
If, further, we say that to is an Aux, the structure will be as follows:
It also seems appropriate to indicate that PRO refers to John; we can do this in the usual way with
indices, though we have no rules yet that can carry this out:
a. I. For every city, John has a plan to visit it.

II. John has a plan such that in it, he visits every city.
b. For both readings, we start with Quantifier Conversion, deriving:
To derive meaning I, we raise the quantifier to the highest S, adjoining it there, as follows:
To derive meaning II, we raise the quantifier to the lower S, adjoining it there, as follows:
——————————————————————————————————
Study Exercise #92: Phonetic Dictations
southern
myrrh
corpulent
whether
multiple
coinage
parameter
ostentatious
turmoil
trapezium
southern [ˈsɚn]
myrrh [ˈmɚ]
corpulent [ˈk{o,o,ɔ}pj{u,,ə}lənt]
whether [ˈwɛðɚ]
multiple [ˈmlt{, ə}p{əl, l̩ }]
coinage [ˈkɔn{ə,ɪ}d]
parameter [pəˈæməɚ] [ɚ] for first [ə] or [ə] ok
turmoil [ˈtɚmɔɪl]
ostentatious [ɔstɛnˈteʃəs]
trapezium [təˈpiziəm]
—————————————————————————————————————
Study Exercise #93: Phonology
This is an imaginary language but the rules it has are found in real languages.
[, , ] are voiced fricatives (bilabial, dental, velar). [t̪ , d̪, n̪] are dental.
‘Noun’ ‘the Noun’ ‘two Nouns’ ‘five Nouns’

1. [pama] [la bama] [d̪ue bama] [kwin̪d̪o bama] ‘tuna’
2. [peli] [la beli] [d̪ue beli] [kwin̪d̪o beli] ‘swordfish’
3. [t̪ ue] [la d̪ue] [d̪ue d̪ue] [kwin̪d̪o d̪ue] ‘mackerel’
4. [t̪ azo] [la d̪azo] [d̪ue d̪azo] [kwin̪d̪o d̪azo] ‘cod’
5. [kame] [la game] [d̪ue game] [kwin̪d̪o game] ‘mahi mahi’
6. [koli] [la goli] [d̪ue goli] [kwin̪d̪o goli] ‘carp’
7. [bafi] [la afi] [d̪ue afi] [kwin̪d̪o afi] ‘catfish’
8. [belu] [la elu] [d̪ue elu] [kwin̪d̪o elu] ‘pollock’
9. [d̪aa] [la aa] [d̪ue aa] [kwin̪d̪o aa] ‘yellowtail’
10. [d̪azo] [la azo] [d̪ue azo] [kwin̪d̪o azo] ‘sturgeon’
11. [gele] [la ele] [d̪ue ele] [kwin̪d̪o ele] ‘halibut’
12. [gova] [la ova] [d̪ue ova] [kwin̪d̪o ova] ‘salmon’
a) Produce consonant and vowel charts, labeling the rows and columns with features. You
may assume [+dental] is a feature.
b) Do the stems alternate? Explain
c) Give rules, naming them.
d) Is any rule ordering required?
e) Give right order/wrong order derivations for la dazo and la azo.
a) Produce consonant and vowel charts.
[+bilabial] [+labiodental] [+dental] [+alveolar] [+velar]

[+stop] [−voice] p t k
[+voice] b d g
[+fricative] [−voice] f s
[+voice]  v  z 
[+nasal] m n
[+liquid] l
[+glide] w
–back  +back  +back 

–round –round +round
E E E
+high
–low  E i u
–high
–low  E e o
–high
+low  E a
b) Do the stems alternate? Explain
Yes, for example the stem for “tuna” has the two allomorphs [pama] and [bama].
c) Give rules, naming them.
Intervocalic Voicing
[+stop] → [+voiced] / [+syllabic] ___ [+syllabic]
This voices any stop occurring between vowels. It can be applied harmlessly to [b, d, g],
since they are already voiced, so I left out [−voice] from the left side of the arrow.
Intervocalic Spirantization260
+stop  –stop 
+voice → +fricative / [+syllabic] ___ [+syllabic]
E E
This turns any voiced stop between vowels to its fricative counterpart, thus [b, d, g] →
[, , ].
260
Standard terminology for a rule that creates fricatives. “Spirant” is an old-fashioned synonym for
“fricative.”
d) Is any order required?
Intervocalic Spirantization must precede Intervocalic Voicing, to keep the voiced stops that
derive from voiceless from turning into fricatives—we want Intervocalic Frication to apply
“too late” to affect those stops.
e) Give right order/wrong order derivations for la dazo and la azo.
Correct:
/la tazo/ /la dazo/ underlying representation

—  Intervocalic Spirantization
d — Intervocalic Voicing
[la dazo] [la azo] surface representation
Incorrect:
/la tazo/ /la dazo/ underlying representation

d — Intervocalic Voicing
  Intervocalic Spirantization
*[la azo] [la azo] surface representation
————————————————————————————————————
Study Exercise #94: Phonemics of Mississippi English
This exercise is based on an unpublished article by Prof. Elliott Moreton, an eminent linguist
who teaches in the Linguistics Department at the University of North Carolina. The article is
posted on his professional web site:
https://users.castle.unc.edu/~moreton/Papers/RaiseAlphaNotes1999.pdf. The native speaker is
Prof. Moreton himself, who grew up in Oxford, Mississippi. He writes, “If you’re going to imitate
my accent, you might as well do it right.”
Data for Part I

gripe [gɹaɪp] bribe [bɹaːb] I [aː] bias [baːəs]
light [laɪt] slide [slaːd] thigh [θaː] biology [baːˈɑlədʒi]
spike [spaɪk] migrant [maːgɹənt] sigh [saː]
knife [naɪf] hive [haːv]
price [pɹaɪs] prize [pɹaːz]
line [laːn]
lime [laːm]
fire [faːɹ]
Directions. [aː] and [aɪ] are allophones of the same phoneme.
a. Decide what form the underlying phoneme should take. Justify your decision. As always,
you should select the simplest analysis.
b. Write a rule to derive the contextual allophone.
c. Give derivations for price, prize, sigh, and bias.
Data for Part II.

write [ɹaɪt]
writer [ˈɹaɪɾɚ]
ride [ɹaːd]
rider [ɹaːɾɚ]
d. How should the rule you wrote be ordered with respect to the rule of Tapping? The rule is
restated here for convenience.
Tapping
+alveolar
+voiced
+stop  →
E
+liquid 
e. Justify your answer with right and wrong derivations for the four words just given.
a. We should pick underlying /aː/. This is the “elsewhere” allophone, and we can get a nice
clean analysis picking it as the phonemic representation. All we have to do is set up a rule turning
it into [aɪ] before a voiceless consonant. If we set up underlying /aɪ/ and tried to turn it into [aː] as
an allophone, the rule needed would be very complicated, since you need three environments
(voiced consonant, vowel, end of word).
b. The rule that is needed is this:
Diphthongization
aː → aɪ / ___ [−voice]
A lesson that emerges (if your own English happens not to be Mississippian) is: don’t assume
that another person’s phoneme is necessarily the way you say a sound! Each phonemic pattern
must be analyzed in its own terms, dialect by dialect.
c. Derivations
/ˈpɹaːs/ /ˈpɹaːz/ /ˈsaː/ /ˈbaːəs/ underlying representations

aɪ — — — Diphthongization
[ˈpɹaɪs] [ˈpɹaːz] [ˈsaː] [ˈbaːəs] surface representations
d. Diphthongization must preceding Tapping, because it applies based on the underlying, not
derived, voicing value of the tap.
e. Good derivations:
write writer ride rider
/ˈɹaːt/ /ˈɹaːt-ɚ/ /ˈɹaːd/ /ˈɹaːd-ɚ/ underlying representation
aɪ ˈɹaɪtɚ — — Diphthongization
— ˈɹaɪɾɚ — ˈɹaːɾ-ɚ Tapping
[ˈɹaːt] [ˈɹaɪɾɚ] [ˈɹaːd] [ˈɹaːɾɚ] phonetic representation
Bad derivations:
write writer ride rider
/ˈɹaːt/ /ˈɹaːt-ɚ/ /ˈɹaːd/ /ˈɹaːd-ɚ/ underlying representation
— ˈɹaːɾɚ — ˈɹaːɾ-ɚ Tapping
aɪ — — — Diphthongization
[ˈɹaːt] *[ˈɹaːɾɚ] [ˈɹaːd] [ˈɹaːɾɚ] phonetic representation

Hayes Introductory Linguistics 2021

Uploaded by

Copyright:

Available Formats

Hayes Introductory Linguistics 2021

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hayes Introductory Linguistics 2021

Uploaded by

Copyright:

Available Formats

Introductory Linguistics

Copyright 2021 by Bruce P. Hayes

Chapter 1: What is Linguistics?

1. What this book will be like

2. Implicit and explicit knowledge of language; working with consultants

(1) “What is the word for “duck” in your language?”

(2) “How would you say “two ducks”?

3. The reference of each other

(3) We like each other.

(5) *I like each other.

(6) Restriction on reciprocals

(8) *Alice and Sue think I like each other.

(9) We believe they like each other.

Again, reason it through for yourself then read the footnote.3

(10) Defn.: clause

A clause is either a whole sentence or a sentence within a sentence

(11) [ We like each other. ]S

We also need to define noun phrases.

(16) Defn.: noun phrase

(17) Each Other Reference Rule

(18) [We believe [they like each other.]S]S

(19) *[John and Bill think [I like each other.]S]S

3.1 Ambiguous sentences with each other

(20) [We wrapped the ropes around each other.]S

3.2 Chains of reference

(21) [John and Bill say [they like each other.]S]S

We will cover the behavior of ordinary pronouns like they in Chapter 9.

4. The role of implicit noun phrases

(23) [They seem to us [to like each other]S]S

(25) [They seem to us [(they) to like each other]S]S

(26) Each Other Reference Rule (revised)

5. Each other: the bottom line

6. The field of linguistics

Syntax study of rules for forming sentences

Phonetics study of how sounds are produced and perceived

7. What’s this about “native speakers”?

Study Exercises for “each other”

Answer to Study Exercise #1

a. [ I believe that [ turtles can swim. ]S ]S

c. [ Bill said that [ [ Jane sang ]S and [ Fred danced. ]S ]S ]S

d. [ I persuaded Fred [ (Fred) to buy a telescope. ]S ]S

This sentence is ambiguous:

My sister and I gave our parents books about each other.

Answer to Study Exercise #2

My sister and I gave our parents books about each other.

[ My sister and I gave our parents books about each other. ]S

This sentence is ambiguous:

Answer to Study Exercise #3:

For this sentence:

Answer to Study Exercise #4

(28) Defn.: morpheme

A morpheme is the smallest linguistic unit that bears a meaning.

2. Two kinds of morphology

Inflectional morphology is grammatical morphology, normally closely related to the syntax.

• tense on verbs (present tense jumps, past tense jumped)

(29) Turkish morphology data

1. el ‘hand’ ev ‘house’ zil ‘bell’

• “plur.” or “pl.” abbreviate “plural”.

4. Breaking up the words into morphemes