Cherokee Papers From UCLA
Cherokee Papers From UCLA
Cherokee Papers From UCLA
Number 16
by
Filippo Beghelli, Barbara Blankenship, Michael Dukes, Edward S. Flemming,
Pamela Munro, Brian Potter, Robert S, Williams, and Richard Wright
edited by
Pamela Munro
Department of Linguistics
University of California, Los Angeles
1996
Copyright 1996 by the Regents of the Universityof California
All rights reserved
Printed in the United :Statesof America
Cherokee Papers from UCLA
Table of Contents
1
2
Cherokee Papers from UCLA: An Introduction
Pamela Munro
This volume is a collection of papers begun during a UCLA graduate class in linguistic field
methods on Cherokee during 1992-93. Together these eight papers constitute a survey of many
important aspects of the grammar of Cherokee, based on work with our consultant, Mrs. Virginia
Carey, who is originally from Tahlequah, Oklahoma. We hope you enjoy learning about Cherokee
from them.
Cherokee is a language of the Southern branch of the Iroquoian family of American Indian
languages, currently spoken primarily in North Carolina and Oklahoma (for a good survey of
Iroquoian and Cherokee dialectology, see Scancarelli 1987: 7-14). Although the Cherokee people
and their language fascinate many Americans,1 there has been relatively little linguistic study of the
language, perhaps because of its considerable morphological complexity and irregularity. Still,
Cherokee is the subject of four fairly recent dissertations (King 1975, Cook 1979, Foley 1980,
and Scancarelli 1987) and one excellent though brief dictionary and grammatical sketch (Feeling
1975 and Pulte and Feeling 1975, in one volume),2 as well as a number of shorter works.
The history of Cherokee studies at UCLA goes back at least to 1984, when I taught a field
methods class on Cherokee during the UCLA Linguistic Institute, with students from all over the
world, again with Mrs. Carey as consultant. Normally, field methods at UCLA is an intensive
two-quarter (six month) class which students take after several years of graduate study of linguistic
theory in order to develop techniques for learning about a language by questioning a native speaker
(in a mock "field" situation), rather than studying from books or tapes. My selection of Cherokee
as the target language for a brief six-week course during the Institute, many of whose students
have little background in linguistics, was quite naive, since the phonetic, phonological, and
morphological complexity of the language made it hard for students to get to the point where they
could understand and appreciate its syntactic structure. But my practice is to try to begin a field
methods class as ignorant of the target language as my students, so I was unaware of what a
difficult task I had chosen for us! Despite the problems, the class was successful, due largely to
Mrs. Carey's talents as a consultant and the enthusiasm of the students. Two of them, Janine
Scancarelli and Geoffrey Lindsey, were UCLA graduate students who continued working on
Cherokee (with Mrs. Carey and later with other speakers) after the conclusion of the Institute.
Scancarelli went on to complete a dissertation on the language (1987, cited in almost every paper in
this volume), and Lindsey made use of extensive Cherokee phonological data in his own
dissertation, a comparative study of intonation (1985).
I had never repeated a field methods language before, but the speaker of another language
who had agreed to serve as consultant for my Winter-Spring 1993 class backed out two weeks
before the class was to begin. Panicked, I called Mrs. Carey, who very kindly agreed to
substitute.3 Despite some health problems, she served once again with patience and good humor,
continuing to work with many students on their papers after the class was over. We are extremely
grateful to Mrs. Carey, to her husband Levi Carey (who worked briefly with some class
3
members), and to Janine Scancarelli (who made a brief but very helpful visit to our class) for all
their assistance. Janine Scancarelli and Geoffrey Lindsey also provided us with copies of
unpublished papers and advice.
Three of the papers in this volume address well-known problems in Cherokee grammar.
Edward Flemming's and my papers concern the related morphophonological rules of laryngeal
metathesis and alternation, while Michael Dukes's paper analyzes the unusual system of Cherokee
agreement. The topics of the other five range from Richard Wright's study of Cherokee tonal
phonetics and phonology, to Barbara Blankenship's description of the derivational morphology of
Cherokee classificatory verbs, to Robert Williams's account of Cherokee possession, to Filippo
Beghelli's investigation of Cherokee clause structure, to Brian Potter's paper on the syntax of
Cherokee nominalizations. 4
In the remainder of this Introduction, I will survey Cherokee phonology and the orthography
used to write Cherokee in this volume and introduce some features of Cherokee morphology. 5
Following the Introduction is a list of the abbreviations we use in our papers.
4Three additionalpaperspreparedfor the field methodsclass were unfortunatelynot submittedfor publicationin this
volume.These were studiesof the phoneticsof Cherokeefinal vowelsand clitics by VictoriaAnderson,of Cherokee
reflexivizationby NhlanhlaThwala, and of the phoneticsof Cherokeevowel length by KimberlyThomas.
51 am grateful to EdwardFlemmingfor commentson this Introduction.
4
The apostrophe ' represents the glottal stop. DI and ti are unaspirated and aspirated lateral
affricates; j and ch are alveolar or alveopalatal affricates. In earlier linguistic work j has been treated
as a ts cluster; see Scancarelli (1987: 24-25) for arguments that it is a unit. 6 DI and gw may be
units as well or may arise as d + I and g + w clusters (see Scancarelli 1987: 23, 47-48). Like stops
and affricates, all sonorants may also cluster with h. We again follow Feeling and Pulte ( 1975) and
native speaker intuition in writing these as clusters.7 (For more on all these alternations, see
Flemming, this volume, and Munro, this volume.) Mis rare, and band p occur almost exclusively
in loanwords.
Cherokee has a six-vowel system, plus length 8 and tone: in addition to the usual five vowel
qualities, there is a low mid central vowel which is always nasalized. In all romanizations, this
sixth vowel is written v.9 Romanizations for Cherokee differ in how to represent the long vowels
and the complex tonal system (called "pitch" by most previous investigators; cf. Wright, this
volume, who analyzes Cherokee as having "lexically marked tone as well a metrically determined
stress accent superimposed upon the normal tonal melody in certain forms"). Our group's work on
Cherokee tonal phonology draws heavily on an important unpublished paper by Lindsey (n.d.).10
Scancarelli (1987: 28-29) writes a colon after long vowels, and represents pitch (or tone)
with a system of (handwritten) iconic diacritics developed by Geoffrey Lindsey (though not used
in Lindsey 1985). Feeling and Pulte (1975) write a dot under vowels whose shortness cannot be
predicted by context, 11 and represent pitch with superscript numbers or number sequences. 12 We
write long vowels doubled, in part to facilitate the representation of contour tones:
ii ee aa vv 00 uu
There are two level tones, high and low, which may occur on either long or short vowels, and four
contour tones, which occur only on long vowels. The superscript tone numbers are a fully
adequate way to represent Cherokee's tonal contrasts (see Feeling 1975: xi-xii; Lindsey 1985
discusses some phonetic problems with this description), but they are difficult to type and invite
omission. Lindsey and Scancarelli's system is iconically attractive, but untypable. To overcome
these drawbacks, our group developed a system of typable tone diacritics. 13 We represent Feeling
61naddition to these arguments, there also appear to be phonetic contrasts between unit ch and cluster t + s.
7ourorthography is thus less consistent in this regard than the standard linguistic orthographies discussed earlier.
8Cherokee vowel length seems to be predictable in certain cases (cf. Foley 1980, Scancarelli 1987: 22, and
suggestionsby Kimberly Thomas). However, since it is contrastiveon the surface and since phonological rules such
as Laryngeal Metathesis (cf. Cook 1979 and Flemming, this volume) depend crucially upon it, we write it
consistently, following the practice of King 1975, Cook 1979, Lindsey 1985, and Scancarelli 1987.
In addition to the long and short vowels, Cook {1979)also mentions a class of "extrashort" vowels. We have
nothing to say about these. (I thank William Pulte for discussion of them, however.)
91:bechoice of v follows the general use of this letter in 19th Century missionary orthographies in Oklahoma to
represent a deviant a sound - in Choctaw and Creek (Muskogeanlanguages unrelated to Cherokee, but like
Cherokee members of the group of Five Civilized Tribes), for example, it represents the often schwa-like short a.
1Ortis indeed disappointing that this important paper has not been published. I hope that the citations we provide
here will make clear to the reader how seminal it is.
11Feeling and Pulte assume that vowels in closed syllables are always short, and that vowels in open syllables are
generally long, but that short open syllable vowels can be indicated by underdotting. One way in which this system
seems to be inadequate (in addition to its untypability and the ease with which underdots can be lost or ignored by
the user) is that there are rare cases of long vowels in closed syllables.
12Toe authors (1975: iii) acknowledge Eunice Pike's help in developing this system.
13Allmembers of the 1993 UCLA Cherokee group participated in working out this tonal orthography, but the basic
concept is due to Richard Wright. I have since taught this system to two non-linguist UCLA students learning
Cherokee (Carol Buswell and Ayisha Jamal, whom I thank for their patience); my observation is that it was easy for
them to learn and use.
5
and Pulte's level pitch 2 with a grave (low) tone mark and level pitch 3 with an acute (high) tone
mark. Unmarked vowels indicate peripheral tones: thus, descent from low to superlow (contour
pitch 1) is represented with a grave on the first vowel and an unmarked second vowel, while rise
from high to superhigh (contour pitch 4) 14 is repriesented with an acute on the first vowel and an
unmarked second vowel. Our tone marks, with corresponding Feeling-Pulte pitch numbers, are
exemplified below on the vowel a.
Cherokee intransitive verbs used in the present tense may take one of two sets of pronominal
subject agreement markers, called "A" and "B". \1{hile the distribution of these sets is somewhat
reminiscent of intransitive agreement in an "active-stative" system, it is considerably less
semantically regular than is usual with such systems, 16 and clearly must be lexically specified.
First and second person distinguish three numbers, singular, dual, and plual, while third person
distinguishes only singular and dual-plural; first person further distinguishes non-singular
inclusive (includes second person) from exclusive (excludes second person). The intransitive
subject agreement prefixes are listed on the next page: 17
6
A B
1 singular *ji-, g- agi-, agw-
2 singular hi- ja-
3 singular a-, ga- uu-
1 dual exclusive oosdii- ooginii-
1 dual inclusive linii- ginii-
2dual sdii- sdii-
1 plural exclusive oojii- oogii-
1 plural inclusive lidii- ligii-
2 plural iijii- lijii-
3 dual-plural anii- uunii-
Cherokee verbs have five separate basic stems: the present, imperfective, punctual,
perfective, and infinitive. (All other forms are based upon these.) The choice of A or B prefixes on
a given verb depends not only on that verbs lexical classification as an "A-verb" or a "B-verb", but
also on the stem choice. B-verbs always use B prefixes. A-verbs use A prefixes with their present,
imperfective, and punctual stems, and B prefixes with their perfective and infinitive stems. Below
are some third-person singular examples (adapted from Scancarelli 1987: 66-67), with 'dance' (an
A-verb) and 'laugh' (a B-verb):
Cherokee transitive verbs with third-person singular inanimate objects mark their subjects
with the appropriate A and B prefixes (depending on the stem), as explained above. 18 All other
combinations of subjects and objects are marked by prefix combinations, which are used with all
transitive stems.
18Thereis also a rare group of transitiveB verbs, which use only B prefixes with singular inanimateobjects.
7
8
Abbreviations and Glossing Conventions
Because of the complexity of the data we are describing, we made the decision to use a
uniform set of abbreviations and glossing conventions in our examples. In this section I describe
the way we have glossed the agreement morphology described above, and provide a list of the
abbreviations used in this volume.
The order used in our glosses is person, number/animacy information, agreement set (A or B, for
intransitives). Thus for instance lsA = first-person singular A, ldexB = first-person dual exclusive
B, and so on. Transitives are specified with the subject element followed by> followed by the
object element, e.g. ls>3san = first-person singular subject, third-person singular animate object.
In cases other than these, however, if two words are run together in a gloss, they are
separated either by a period, a colon, or a semicolon (usage unfortunately varies), e.g. tie.up
expresses the single morpheme "tie up". The rule we try to follow is that the number of hyphens
(morpheme breaks) and spaces must match exactly in the data and the gloss underneath it.
9
Tone and Accent in Oklahoma Cherokee
Richard Wright
Introduction
Oklahoma Cherokee has six surface tones in non-final position. Tone is thoroughly
marked and there is a brief description of the tonal system in Pulte and Feeling (1975). Lindsey
(1985, 1987) provides a full analysis of the tonal system in which Cherokee has a complex pitch
accent system. This paper will present an analysis in which Oklahoma Cherokee is a language
with lexically marked tone as well a metrically detennined stress accent superimposed upon the
normal tonal melody in certain forms. The data for this study is drawn primarily from the speech
of Mrs. Virginia Carey, a native speaker of Oklahoma Cherokee. Examples will also be drawn
from Pulte and Feeling (1975) and Lindsey (1985, 1987).
1. Description of tones
1.1 Transcribing tone. There is general agreement in transcriptions of four of the six
surface tones of Oklahoma Cherokee. Pulte and Feeling use Eunice Pike's number system for
marking tone: low level (2), high level (3), rising (23) and falling (32). Examples are given here
with Pike's numbers for reference but in the rest of the study tones will be marked using Land H
in formulating rules and using accent marks in orthographic representations (examples from
Lindsey, 1985).
The remaining two surface tones are a lowfall (1), sometimes referred to as a superlow,
and a highrise (4), sometimes referred to as a superhigh. All contour tones are limited to long
vowels. The lowf all is characterized by a declining pitch starting from the normal low tone range
and by glottalization at the lowest part of the fall. Lindsey (1985) gives evidence that in many
cases the low fall is derived from the loss of a glottal stop in coda position through an active
process of vocalization. It also appears necessary to posit underlying laryngealization in some
cases that appear to be derived historically from the loss of a glottal stop in coda position.
lowfall (1)
Example 2: nvvya 'rock'
The highrise is marked by a rise that begins at a variable pitch and rises above the normal high
tone register. It is also characterized by an amplitude peak that corresponds to the peak of the
pitch excursion. The rise in amplitude is not a characteristic of the regular high tone. Pulte and
Feeling describe the highrise as a level superhigh, but Lindsey found that measures from narrow
band spectrograms indicated that it rises. Measures taken from narrow band spectrograms of
twenty tokens read in isolation show a clear pitch rise to above the normal high tone range in
every case, confirming Lindsey's earlier measures measurements. The highrise is limited to
certain syntactically detennined forms: verbs in subordinate clauses, deverbal nouns, locatives
and most adjectives. Its distribution is discussed further below.
highrise (4)
Example 3: n6oya 'sand'
11
The convention for marking tone arrived at for this volume is as follows:
low = a,aa high = a,da rising = aa falling = ad lowf all = aa highrise =da
1.2 Word final syllable tone. Feeling amdPulte leave the final vowel's tone unmarked
because it is predictable. They describe it as a fal1ing superhigh (4) except when it is preceded by
a penultimate superhigh, in which case it is realiz:ed as a normal high. Lindsey on the other hand
describes it as a boundary high tone (H%) with optional upstep. For this study measurements
were made using narrow band spectrograms of tokens read in isolation. The data set consisted of
five tokens each with the following penultimate tones: low, high and highrise. Results indicate
that the final syllable is realized at or above the word internal high tone range. Again this agrees
with Lindsey's (1985) measures. Whether this should be analyzed as a boundary tone that
associates late in the derivation, as is normally the case (e.g. Goldsmith, 1990), or an
underlyingly specified high tone is another quei;tion. For the purposes of this study it will be
treated as an underlying high tone with phonetic upstep occurring.
First of all, in pitch accent systems there is generally at most one accent per word, although
there may be none. On the other hand, Cherok◄:e is more like tone systems in that the OCP
allows only one underlyingly specified tone per morpheme (Goldsmith, 1989). In
morphologically complex languages like Cherokee this results in the possibility of several
underlyingly specified tones per word, although again there may be no underlyingly specified
tone. Second, the rise in pitch and the catethesis following the accent usually seen in pitch accent
languages (Pierrehumbert and Beckman, 1988) is not present in Cherokee, even within a
morpheme. Finally, there is evidence for a floating high tone, an exclusive characteristic of tone
languages. While these features could fit into a pitch accent system, it seems more
straightforward to analyze them as a tone system.
2.1 Tone in accentless forms. Level high and level low tones are found in syllables
with both long and short vowels but the rising (LH) and falling (HL) tones are found exclusively
in syllables with long vowels. Given the usual restrictions of having one tone per tone bearing
unit (TBU) and given that the mora is a unit of vowel length, this indicates that the mora is the
TBU in Cherokee.
High tones are usually found in pairs except at the right edge of a word. That is, they are
found as level high tones on long vowels (example 4), or they are found as a rising tone (LH) on
a long vowel followed by a falling tone (HL) on a long vowel or by a high tone on a short vowel
(example 5), or they are found as a high tone on a short vowel followed by a falling tone (HL) on
a long vowel or by a high tone on a short vowel (example 6). Exceptions to this generalization
are discussed below.
12
Example4 Example 5 Example 6
g66dvsga gwgiilidiya aagvvhalliiiysga
'I'm making a fire' 'he's sifting it' 'he's chopping it up'
There is no restriction on the number of low tones that can occur consecutively. The only
restriction is related to the distribution of high tones: a low tone never appears following a rising
tone or a high tone on a short vowel ( i.e. *LHL). Lindsey notes that Pulte and Feeling give many
apparent counterexamples to this last generalization, but he shows convincingly that all of these
cases involve a mistranscribed highrise accent tone. Subjective judgments by several listeners
support Lindsey's findings and are reflected in the transcription of tone in this volume.
The distributional evidence for high and low tones points to a process of high tone spread
limited to one TBU. If no spreading applied, then sequences of odd numbers of high tones should
be attested in the data. Since only two consecutive moras may bear high tone, the domain of the
spread is one mora. Lindsey (1985) presents further evidence from morphological alternation that
high tone spreads rightwards. The following examples are from Lindsey (1985, 130-131):
H H H
I I I
Example 7: ka + 1+ iisk + oo'i aa + ktoost + iisk + oo'i
3sA + take out + imp + hab 3sA + look at + imp + hab
kaaliisk6o 'i aakto6stiisk6o 'i
In the first form there is one underlying high tone that results in two surface highs and in the
second example there are two underlying highs that result in four surface high tones. In the first
example -iisk, which is marked for a high tone on the second mora of its long vowel ends up with
a rising tone. In the second example it ends up with a level high due to the high tone spread from
the preceding syllable.
Although Lindsey's later analysis (1987) has the mora as the TBU, he formulates his rule
using V to represent moras. This leads to difficulty in his analysis in separating laryngeal features
that link to the Root node of the vowel and tonal features that link directly to the mora. The high
spread rule is reformulated here using moras.
Figure 1
Hi~h Spread
Spread H rightwards noniteratively.
H
µ
f',,µ
While High Spread correctly predicts that high tones can spread across any number of
onset consonants and that it will spread only one mora to the right, it incorrectly predicts that it
can spread to a mora dominating a coda consonant. In Oklahoma Cherokee there is evidence that
only/// and /h/ may appear in coda position underlyingly and that they can bear a mora. The
strongest evidence is that vowels tend to be short before /// or /h/ preceding a consonant. In
addition, deletion of a glottal stop in coda position leads to compensatory lengthening of the
vowel. Therefore it is necessary to reformulate High Spread so that there is no possibility of tone
spreading onto a glottal stop or glottal fricative. This restriction turns out to have predictions that
are borne out in the case of the non-derived falling tone discussed below.
13
High Spread reformulated
Given this rule, long level high, HH, is the result of the high tone specified for the first
mora of a long vowel spreading onto the second mora. It can also result from high tone spread
from a preceding syllable onto the first mora of a long vowel that is specified on its second mora
for high tone as in. the example above. Rising tone, LH, is the result of high tone specification for
the second mora of a long vowel when there is no spread from a preceding syllable. The majority
of examples that appear in Feeling and Pulte are di~rivedfrom the spread of a preceding high tone
onto the first mora of a long vowel. The nondelived forms are the result of laryngealization,
underlying or derived, on the vowel blocking High Spread. Below are three examples illustrating
the most common tonal patterns seen in Oklahoma Cherokee.
VY
It would be nice to leave all the moras that:receive no high tone unspecified at the output
of the phonology to avoid fill in rules. However, the phonetic data do not support this type of
analysis. As is noted in Pierrehumbert and Beckman (1988), the pitch of vowels that are
unspecified for tone is dramatically affected by the~tone of the adjacent tonally specified vowels.
Interpolation between the word edge and the specified tone results in a contour. Measurements
from narrow band spectrograms indicate that Cherokee vowels with low tone are level, even
when adjacent to vowels bearing high tone. An example spectrogram is given in Fig. 3 below.
Since the low tone vowels do not pitch interpolation low tone is analyzed as filled in.
• -
..-..~
....._.
--,
...... - -
• -
-.
500
..
---
Hz 0__,'1,.1 L L
..._-_-_-_-_-_-_-_-_-_-_-"'T-
0
H
•
--_-_-_-_-_-_-_-_-_-
.25
~
-
_-.---
H L
.. _-_-_-_-_-_....
.50
L
_-_-_- -_-.-.._-_-_-_-_-_-_-_-_-_-_-_-_----=-
.75 1 sec.
Low tone fill in is ordered after High Spread to avoid blocking the spread of high tones.
14
Fig. 4 Low tone fill in
r~
H H H High Spread
t,
uuneesdala agwvvsa
t\ V\'i i g a h V sg a
l-, H L
I. ~
I\ I.
\ H ¼.,,,_
Low Fill in
agwvvsa w\)ga~h~~-
uuneedala agwvvsa wiigaakah°vsga
'ice' 'myself' 'he's sending it'
2.2 Laryngealization and tone. In addition to the familiar looking tone rules seen
above, there are some interesting interactions between tone and laryngealization that result in the
lowfall tone. As was noted above the lowfall is characterized by fall in pitch that starts at and
ends below the low tone region. It is also characterized by creaky voice in the second half of the
vowel where the pitch excursion reaches its lowest point. In Oklahoma Cherokee a coda glottal
stop lenites. This results in the lengthening of preceding short vowels. It also results in
laryngealization of the second half of the preceding long vowel. This process is termed
vocalization.This process is analyzed here as spreading of the vowel's Place node onto the Root
node of the coda glottal stop. When no high tone is specified for a mora within the syllable, the
result is the lowfall tone described above (see example 8 below). When high tone is underlyingly
specified for a mora within the syllable the characteristic glottalization is lost and the resultant
tone is falling (HL) tone. This can be seen as a constraint that prohibits tautosyllabic high tone
and constricted glottis from appearing in the same syllable.
geal
Evidence for laryngealization resulting from the lenition of the glottal stop comes from a
process termed Laryngeal Alternation (LA). In Cherokee there is pronominal inflection, resulting
in forms termed glottal grade by Scancarelli (1987), that turns the first /h/ of the verb stem into a
glottal stop. This process is described in more detail in Munro (this volume) and so will be
discussed in no further detail now. What is of interest here is that it provides evidence that much
15
of the laryngealization that results in the lowfall and the falling tones that aren't the result of high
spread can be derived from the lenition of coda glottal stops. One piece of evidence is that when
LA results in a glottal stop that is intervocalic, neither the laryngealization nor the resultant
lowfall occur (see example 9 below). There is also reason to posit underlying laryngealization
since some morphemes have the characteristic falling and lowfall tones but never surface with a
glottal stop. Lindsey (1985) gives evidence that underlying laryngealization is historically
derived from lenition of coda glottal stops. The 3rd singular present prefix in example 8 below is
an case of underlying laryngealization.
aateethoska
I gateethoska
I
I
[e.g.]
H H Laryngeal Alternation
aateethoska
I gatee?toska
I
I
[cg.]
I
[e.g.]
H H Vocalization
aateethlska gateetoska
I
I
[e.g.]
I
[e.g.]
:,:,,,
L
IH J.. H Low Fill In
.' ',',
aat 'e ·e tho ska I
•"\..I
I,._
''•
gateetoska
I
[e.g.]
r
[e.g.]
/aateeth6ska/ /gateet6ska/
aadeet6ska gadeed6ska
16
Example 9: Laryngeal Alternation resulting in an intervocalic glottal stop
guuh
1i1
t sg a
f
g u ? v sk a
Laryngeal
alternation
I
[e.g.]
H H
Vocalization
g u uh
I
v sg a
I
gu?vska (can't apply)
I
[e.g.]
L H L H
:-- I Low fill in
gu·uhvsga
\ I
gu?vska
I
[e.g.]
/guiihvska/ /gu?vska/
In the formulation of High Spread above, high tone is prohibited from spreading onto a
mora that dominates contrastive laryngeal features such as [c. g.] or [s.g.]. This is motivated by
the need to block high tone from spreading onto coda /h/ and///, but it is also useful in analyzing
the high-low fall; high tone is blocked from spreading onto the second mora of the long vowel.
There must also be a rule to delink the laryngeal features from the vowel since they don't surface
in the phonetics. The rule, called Laryngeal Delinking, delinks any laryngeal feature that is
linked to a vowel that is also specified for high tone.
17
Lazynt:;ealdelinkint:;
Delink [constricted glottis] in the presence of tautosyllabic high tone.
There is phonetic motivation for avoiding combining laryngealization and high tone on
the same vowel: glottalization will tend to wipe out high tone. Some languages preserve
underlying tone by tone rescue rules that relink stranded tones. Laryngeal delinking is part of a
pattern of high tone preservation in Cherokee th:at, instead of rescuing, prevents processes that
would strand tone. A second process that would obliterate high tone, were it not blocked, is
vowel deletion due to spreading of [spread glottis] onto vowels (see Flemming this volume).
H H H
I 1. d .I.. r-,oye
tar 1 eeJ11no
I
Hspread
2. 3 Floating tones. There is a complex set of affixes in Cherokee some of which affect
the tonal pattern of the verb or noun to which they are adjoined. One such morpheme is the
distributive prefix dee-. It is used to mark multiple acts of the same action or, in some cases, to
indicate plural objects. When it prefixes to the verb a high tone appears on the first syllable of the
verb. There are several allomorphs of this prefix that depend on the shape of the first vowel of
the verb. In vowel initial verbs, the distributive ajppears as d- only and is not accompanied by a
high tone.
18
Example 13: Association of floating high tone
H ® H
Floating high
gvvyillv'iha \,
d ee gvvy1
.
vv
11 I.
1
h
a
association
H
High spread
J\'.
gvvy1 · 1vv 1 ha dee glyilv"iha
~',,tI\ 1r
j
g vv y·i I v v ' 1 ha
I ' '
k
dee
1 \
I \
g
Ht\
~H~ r
I
I
yilvv'iha
I
I
Low fill in
/gvvtilvv'1ha/ /deegvvy1lvv'1ha/
19
Example 14: Floating high association after laryngeal metathesis and laryngeal alternation
(8) H (B)H
I
dee gahnooyee'a Laryngeal alternation deejynboyee'a
[e.g.]
(8) H (BH
deekanooyee'a
I Laryngeal metathesis
... I
deeJ1jnooyee a
I
[e.g.]
~H
\. ~ I
Floating High deeJnrooyee a
Association
[e.g.]
HH tJH .
lh
deekanooyee'a High Spread .l. h
dee,Jimooyeea
I
I
[e.g.]
HH
deekLb,yee'a Laryngeal Delinking
L HH L
~
I\ t,
,1!,.~k
lI I\
I,.,
:rm
\"'
II
I I
I I
-I
I
•1•
~' LI\
..
I \
~ nooyeca Low Tone Fill in dee Jnn yee'a
/deekan66yee' a/ /deejiinooyee' a/
3. Accent in Cherokee
This last section analyzes the one remainiing tone of Cherokee. Cherokee is said to have
two accent systems, tonic and atonic. Unlike the other tones of Oklahoma Cherokee, the high
rise, referred to in the literature 'atonic accent' is :syntactically restricted: it appears only on verbs
in subordinate clauses, on deverbal nouns, and on adjectives. Although it has been referred to as
'atonic accent' in the literature, for clarity it will be referred to simply as accent.
The phonological characteristics of acce:mt also set it apart from the tonal system of
Oklahoma Cherokee. Its distribution is restricted not by the presence or absence of other tones
but rather is determined by factors that are normally associated with prosody. Only one accent
may appear per word, and in the syntactically determined forms listed above at least one accent
20
must appear per word. Accent falls on the rightmost long vowel of a word or on the first syllable.
It is characterized by a gradual rise in pitch that begins at a variable level and rises to a point
above the normal high tone register and by a rise in amplitude. There is no parallel rise in
amplitude for the normal high tone. Lastly it is imposed over the existing tonal pattern,
overriding the melody of the syllable on which it occurs. All of these factors point to a metrically
determined stress accent rather than pitch accent system.
Given the distributional data, it is proposed that Cherokee has quantity sensitive
unbounded feet and assigns prominence with End Rule right (ER) (Prince 1983). The pattern of
assigning accent to the first syllable in the absence of a heavy syllable is typical of many
languages with unbounded quantity sensitive stress systems (Prince, 1983). The edge prominence
rules can be broken down into two stages: edge prominence right EPR and edge prominence left
EPL. EPR applies first and is quantity sensitive, EPL is quantity insensitive and applies only if
EPR fails to assign accent.
EPR
Assign prominence to the syllable dominated by the rightmost grid peak
*
X
X X
'good' 00 Sta /6osta/
*
X X
X X X X
'sweeper' h vv n oo s a sgi /hvvn6osagi/
2nd sg+deverbal stem
The highly restricted quantity sensitive stress accent in Oklahoma Cherokee is probably a
remnant of an accent system that had a wider distribution in Iroquoian languages. Interestingly
some of the Northern Iroquoian languages, such as Cayuga, still maintain an active metrical
stress system (Prince, 1983). It may have survived in the forms that it has because Cherokee has
an extremely free word order (Scancarelli, 1987) and little or no case marking on nouns.
Furthermore, the only difference between many of the derived nouns and the habitual form of the
corresponding verb is the presence of accent. While the classes of stress bearing words are
restricted, the lexical items found in these classes are used frequently. For example, the
derivational processes resulting in deverbal nouns and in adjectives is highly productive. Thus,
maintaining the stress system can be seen as surviving in these forms as a necessary way of
differentiating these highly similar forms. Such a system would also benefit from maintaining a
phonological way of marking the difference between various syntactic structures such as matrix
and embedded clauses.
4.0 Conclusion
Cherokee has been analyzed as a pitch accent language with two accent systems, tonic
and atonic, that operate independently. This paper has proposed that what has traditionally been
termed tonic accent, is in fact a tone system. This proposal is motivated by the fact that pitch in
Oklahoma Cherokee has features that are more characteristic of tone languages than pitch accent
languages. Additionally a metrically determined stress accent is superimposed over the existing
tonal melody in certain forms. Further evidence for such a system comes from comparative data.
21
Prince (1983) analyzed Cayuga as having both tone and a metrically determined stress accent.
The distribution of tones and their interaction with laryngealization form an interesting pattern
that suggests that Oklahoma Cherokee has recently changed from being a predominantly stress
accent language to being a tone language with stress accent restricted to certain forms. As such, it
is an example of a language in which tone and stress systems interact.
22
Laryng,eal Metathesis and Vowel Deletion in Cherokee
Edward S. Flemming
1 . Introduction
'Laryngeall metathesis' is the label assigned to a complex set of metathesis and deletion
processes conditioned by laryngeal features in Cherokee (Cook 1979, Scancarelli 1987).
(2a) shows the loss of underlying /o/, rendering /gw/ adjacent to /h/, which results in an
aspirated labialize:d stop. (2b) shows the source of evidence for root-internal deletions of this
kind: the 'glottal grade' of the stem. Certain agreement prefixes, including 1sA select the so-
called glottal grade of a stem. The process of glottal grade formation is discussed in detail by
Munro (this volume), but the crucial aspect here is that it typically involves replacing the first /h/
in the stem by a glottal stop, and since /h/ conditions metathesis and deletion, this change can
remove the environment for these processes. In the dialect of our consultant, pre-consonantal
glottal stops surface as glottalization and lengthening of the preceding vowel, so no actual glottal
stop is transcribed in (2b), however the h-£ alternation can be seen clearly in forms such as those
in (4) In any case,. no /h/ is present so underlying /o/ surfaces undeleted. Comparison of glottal
grade forms with their 'h-grade' counterparts thus provides useful evidence for the occurence of
deletion.
23
(6) Metathesis: CVhX ➔ ChVX
Deletion: CVhX ➔ ChX
Both processes are subject to complex conditioning factors outlined in (7).
2. Optimality theory
The analysis will be formulated in terms of Optimality Theory (Prince and Smolensky
1993). Important features of this approach adopted here are as follows:
2. The output form of a given input UR is that representation which best satisfies the
constraints. Thus constraint violation can be forced by a conflicting higher ranked constraint. For
example, if *[-back, +round] is ranked above 'Spread [+round]' then spreading onto a front
vowel will be blocked.
3. A crucial set of constraints are the 'faithfulness' constraints (cf. Prince and Smolensky
1993) which require the output to be similar to the input. Without constraints of this kind, the
input could be replaced by a dramatically different, but otherwise extremely well-formed, output.
The key families of faitllfulness constraints adopted here, from Kirchner 1993, are ParseF and
*Insert F, which can be formulated as follows:
Optimality theory is adopted here as a tool which allows us to bridge the gap between the
work of phoneticians such as Ohala (1983, 1992), Lindblom (1990), Stevens and Keyser (1989)
and Kawasaki (1982) on providing substantive phonetic explanations for tendencies in
phonological patterning, and the analysis of the phonologies of individual languages. For
example, the tendency for front vowels to be unrounded, and back vowels to be rounded has been
explained in terms of the acoustic effects of rounding. Back vowels differ from front vowels in
having a smaller difference between the frequencies of the first and second formants. Rounding
lowers the second formant, thus rounding a back vowel makes it more distilnct from a front
vowel. However, phonologists must analyze individual languages, not cross-linguistic
tendencies, and some languages contravene the tendencies. For example Turkish has front
rounded and back unrounded vowels.
24
order of priority. If we regard the phonetic desiderata adduced to explain phonological tendencies
as constraints, we can account for the failure of specific languages to observe them in terms of
the ranking order. A superordinate constraint overrides the tendency to observe a lower-ranked
constraint, as in the case of a language which has front rounded vowels, where constraints that
favor making a large number of vowel contrasts available override the dispreference for front
rounded vowels.
The analysis of metathesis in terms of optimality theory raises interesting issues. It is not
difficult to see motivations for metathesis. For example the pattern of historical metathesis
shown in (9) can be understood as being motivated by the well-established preference for open
syllables (Ultan 1978). The question is what constraints restrict the application of metathesis.
(10) Parse Association: A feature in the input should appear associated to the same position
in the ouput.
(11) F F F F
I I I I
X X X ➔ X X X or X X X ➔ X X X
I I I I
G G G G
Reversing the order of two features constitutes a violation of Ordering:
25
(12) F1F2 F1 F2
I I I I
X X X ➔ X X X
I I
G G
A final type of temporal sequencing constraint relates to epenthesis. Inserting a segment
into an input does not alter any precedence relationships, but it does violate the contiguity of the
sequence. Thus we will adopt the following cons1traint(cf. Kenstowicz 1994):
(13) Contiguity: Do not insert segments that are not present in the input 1
Before we turn to the analysis of metathesis and deletion it is necessary to clarify the
phonetic and phonological nature of the Cherokee sounds with particular relevance to these
processes, those which involve the feature [spread glottis] ([s.g.]). Table one shows the
representation of Cherokee sounds central to the present paper in orthography and in IP A,
together with brief phonetic descriptions.
It is essential to realize that while the orthography includes sequences of sonorants and h,
these are in fact conventional linearization of essentially simultaneous events. For example hw
represents a breathy labio-velar, not a sequence of /h/ followed by a labio-velar. Thus I assume
that the breathy glides are represented phonologically as identical to voiced glides apart from the
addition of the [s.g.] specification.
Note that breathy sounds become partially devoiced before voiceless obstruents and are
then spelled wh, yh, nh respectively in the orthography. However, I assume that this devoicing is
a gradient phonetic effect, and that their phonological representations are essentially unchanged.
Realizing that [s.g.] occurs associated to sonorants eliminates some apparent cases of
1This constraint could be alternativelybe formulated as 'do not alter immediateprecedence
relations', where
immediate precedence is a relation defined only between ad~acentelements.
26
metathesis suggested by the orthography. For example, the derivation in (14a) essentially
involves vowel deletion. However, in the orthographic transcriptions given, deleting i leaves the
sequence lh, whereas the actual result is written hi. So, orthographically, this derivation does
involve metathesis of h across l. but in fact what is occurring is the merger of /h/ and /1/ to
produce a [s.g.] lateral.
More generally, the sequences [Ch] and [hC] are not well-formed unless C is a stop
closure. Thus the segment /h/ appears only adjacent to a stop closure or intervocalically, as in
(15).
nmi 'you'
/h/ preceding a stop:
hmdnya 'you're sharpening it'
sahkoonge 'blue'
iilaasmdeeni 'his foot'
Any underlying sequence consisting of /h/ and a sonorant will surface as a single segment with a
[s.g.] specification, as discussed above. This distribution is analyzed in the following section.
Given the fact that the orthography obscures some important aspects of the phonological
representation of words, it is important to keep orthographic and phonological representations
distinct. To this end, orthographic transcriptions will be presented in italics, e.g. aahltawo, while
phonetic and phonemic transcriptions will be indicated by square brackets, e.g. [aaitawo], and
slant brackets, e.g. /aafihtawo/, respectively.
27
( 16) unaspirated b. aspirated p: plain nasal n: breathy nasal hn:
[s.g.] [s.g.]
I I
AoAmax AoAmax AoAmax AoAmax
I I I I
place place [nasal] [nasal]
I I
labial labial
We turn now to the analysis of the generalization that the sequences [Ch] and [hC] are not
well-formed unless C is a stop closure. Presumably general constraints on syllabification or
phonotactics rule out tautosyllabic [Ch] and [hC] clusters, other than aspirated stops (18), since
such clusters are cross-linguistically very rare, but for present purposes we will formulate a
single constraint, *Ch, barring these clusters.
28
(21) /hl/ *Ch Parse [s.g.] Parse Assoc
hl *! *
[s.g.] *
> I
I
1 *!
The two remaining possibilities, the heterosyllabic clusters C.h and h.C are ruled out by a
dispreference for codas, formulated as the constraint *Coda:
This constraint is also ranked above Parse Association, so merging Ch and hC clusters is
preferrable to creating a coda, even at the cost of violating Parse Association:
Thus whenever a cluster of a sonorant and /h/ arises, it will surface as a breathy sonorant.
But in the case of clusters of /h/ and an oral stop closure, there is an additional higher ranked
constraint against merging /h/ with the closure (24). Effectively, this constraint states that there
are no preaspirated stops in Cherokee.
So, as shown in (26), in this case hC will be syllabified with /h/ in coda.
Thus the only cases in which the sequences /Ch/ and /hC/ will surf ace is when C is a stop
closure, and thus forms an aspirated stop in Ch, and is unable to merge with /h/ in /hC/.
29
following sound, X, is a stop, sonorant, or vowell (the fricative /s/ is discussed below). Examples
instantiating each cell in the table are given in (28)-(39).
(27) Incidence of metathesis and deletion in the configuration: CVhX
C\X
plosive Iplosive
del
son~_- ....
son
met
vowel
de...,1....--+-----+-----t
del
30
deletion', in anticipation of the proposed analysis. The second process applies when C is a
plosive, and may result in deletion or metathesis. This process accounts for the observations in
the two rightmost cells in the top row of (27), and will be labelled 'metathesis', even though it
results in deletion in some cases.
Evidence that there are two distinct processes operative here is provided by the fact that
/s/ also conditions deletion of a preceding vowel, without regard to the nature of the preceding
segment (40). I.e. breathy vowel deletion is triggered by /sf as well as /h/ preceding a stop,
whereas metathesis is triggered by /h./or a breathy sonorant only.
(40)
TVs ➔ Ts aagi-sgaasdaaneelv ➔ aagsgaasdaaneelv 'he scared me'
1sB-scare;pst
The basis for the difference between the two processes can be analyzed in terms of the
constraints that motivate them. Metathesis results from the attempt to associate [s.g.] to the
optimal position, whereas breathy vowel deletion is motivated by a requirement to spread [s.g.]
from a voiceless sound. In the remainder of this section we ·will consider first metathesis then
breathy vowel deletion, accounting for the properties particular to each phenomenon, then we
will turn to the factors that they have in common. We expect shared properties because both
processes involve the feature [s.g.] and both involve deletion so constraints on [s.g.] and
constraints on deletion will apply equally to both.
6.1. Metathesis. The first process can be understood as resulting from the attempt to
associate [s.g.] to the optimal position. The best docking site for [s.g.] is on the release of a stop,
where the high rate of air-flow allows the realization of salient cues to the state of the glottis
(Kingston 1990). The alternative positions in which [s.g.] can appear are in isolation as [h], or
associated to a sonorant. The feature [s.g.] is undesirable on a sonorant because it conflicts with
the realization of sonorancy and voicing (Stevens and Keyser 1989). The segment [h] is also
problematic. Post-vocalic [s.g.] preceding a stop (e.g. -iht-) has essentially the same spectral
shape as the preceding vowel, and thus will be substantially masked by it (Bladon 1986). Inter-
vocalic [h] also does not produce any major transitions in spectral shape or amplitude, especially
if voiced (as in Cherokee), although there will be some widening of formant bandwidths, and
loss of higher frequency energy. Thus it is unsurprising that the distribution of /hi is often
restricted (for instance, in English, /hi may only occur as the onset of a word or stressed syllable).
6.1.1. Deletion in TVhV sequences. Conceptually, deletion and metathesis can result
from the requirement that [s.g.] associate to the optimal position, the release of a stop, even at the
cost of deleting an intervening vowel. The constraints required to formalize this analysis
correspond to the considerations adduced above regarding the optimal placement of [s.g.]. So
*[s.g., son] represents the fact that [s.g.] and [+sonorant] are antagonistic features, and *h
represents the fact that [h] is also dispreferred. By contrast, [s.g.] on a stop release does not
violate any constraints. Two 'faithfulness' constraints against deletion of vowels and [s.g.] are
also required.
31
(41) *[s.g., son]: Avoid the feature combination [+s.g., +sonorant].
*h: Avoid [h].
Parse V: Do not delete vowels.
Parse [s.g.]: Do not delete [s.g.]
In Cherokee, these constraints are ranked as shown in (42). In a /l'Vh V/ sequence this
results in deletion of the first vowel, leaving /h/ in the release position following the plosive (43).
Deletion does not apply with a preceding sonorant (44). This is because deletion in this
situation would result in a breathy sonorant, which is as marked as [h] (45).
Simply linearizing [s.g.] to precede a sonorant, then deleting the intervening vowel, as in
the case of intervocalic /h/ above (4.1.1. ), would not yield the correct output:
32
stop, this is also the position to which [s.g.] is associated, so a sonorant following an aspirated
stop will be [s.g.]. Thus the ultimate output of (48) would in fact contain a breathy sonorant, as in
(49), and thus represents no improvement over the input.
There are two reasons to prefer the analysis in terms of deletion and epenthesis. Firstly,
metathesis is not generally possible in Cherokee. Although the optimal position for [s.g.] is the
release of a stop, [h] does not metathesize with a following stop to associate to this position:
sahkoonge 'blue'
Given the current formulation of Ordering, permitting metathesis of [s.g.] with a vowel
would also permit metathesis with a consonant, because Ordering constrains reversal of ordering
relationships without regard to the type of segments involved. It would be possible to replace
Ordering with a set of constraints such as 'Preserve CV Ordering' and 'Preserve CC ordering'
with the latter being ranked higher than the former in Cherokee, permitting CV metathesis but
not CC metathesis. However this sacrifices the simplicity of the Ordering constraint adopted
here.
The more important reason for preferring the deletion/epenthesis analysis of laryngeal
metathesis is that it allows a unified analysis of parallels between metathesis and deletion. Both
33
metathesis and deletion apply only with short vowels (6.3.1). If both processes involve deletion,
this fact can be accounted for in terms of a single constraint against deleting long vowels, as
proposed below. If metathesis is a distinct operation, then separate constraints will have to be
invoked to block the two processes under the same conditions.
Obviously, metathesis does not apply where the preceding consonant is a sonorant (53)
because it would result in an breathy sonorant, which is precisely the segment that metathesis
applies to eliminate.
Metathesis also does not apply in the absence of any preceding consonant (54).
Metathesis in this case would violate several constraints to create an /h/, which is as undesirable
as a breathy sonorant (55).
34
Metathesisonlyiwpliesto breathysonorants
There are two other configurations in which metathesis might apply: /fVh V/ and /fVhT/
(the case of [s.g.] on a fricative is discussed in 6.1.7). In the case of /fVhV/, discussed in 4.1.1.
above, the result of deletion is a well-formed configuration, [ThVJ, in which [s.g.] is associated
to the release of a stop, so there is no motivation to violate further constraints to epenthesize a
vowel as in metathesils. In fact the result of epenthesis, [ThV 1V 2], would be ill-formed because
vowel sequences are not permitted in Cherokee and are resolved by deletion of V l •
Deletion of the vowel in a ffVhT/ sequence also produces a well-formed output: [ThT]:
In any case, deletion in this environment is also motivated by the constraints responsible for
breathy vowel deletion (6.2, below).
Metathesis applies only to breathy sonorants because it is in this case that the output of
deletion is a configuration in which epenthesis is motivated, i.e. an aspirated stop-sonorant
sequence.
Metathesisoperatesleftwards
Note that it is not stipulated anywhere that metathesis should operate leftwards. This is
related to the observa1ion that /h/ does not metathesize with a following stop (56).
(56) hT -x➔ Th :
This shows that Ordering, the constraint against reversing precedence relations, must be
ranked above *h, so metathesizing in this context is a greater violation than allowing /h/ to
surf ace. In fact precedence relations are never violated in Cherokee, so Ordering is an
undominated constrai1t1t.
This ranking also yields the result that shifting [s.g.] to the right cannot result in an
improvement (57). For an input containing a breathy sonorant, the only output preferrable to
leaving [s.g.] in the same position is one in which [s.g.] is associated to the release of a stop. This
can only be achieved by moving [s.g.] rightwards if that movement is across a stop and this
would require reversing the precedence relation between [s.g.] and the following stop. That is,
given that metathesis 1cannotapply across an adjacent stop, it clearly cannot apply across a vowel
and a stop.
6.1.5. The b•ehavior of labialized velars. It is interesting to note that [s.g.] does not
metathesize onto a labialized velar stop (58).
35
(58) aagwa-hnuuwa -X➔ *aakwanuuwa 'I'm wearing a shirt'
lsB-wear.shirt;prs
This is expected given the phonological representations assumed here according to which a
labialized stop consists of two positions; a stop closure and a release which is identical to the
sonorant [w] (59). Thus metathesis onto a labialized stop is blocked for exactly the same reason
that [s.g.] does not metathesize onto [w]: it would produce a dispreferred breathy sonorant.
(59) Ao Amax
I I
k w
This analysis implies that aspirated labial velars are dispreferred because they involve the
marked feature combination [s.g., +sonorant]. Thus we might expect that the vowel epenthesis
that applies in metathesis would apply erroneously to break up this sequence.
(60) kw ➔ *khiw
The crucial obsiervation in explaining why this does not occur is that we have not
analyzed metathesis as involving epenthesis of a vowel per se, rather it involves insertion of a
vowel position. The features associated to that position are present in the underlying
representation. Thus we can prevent epenthesis breaking up-labialized velars by giving a high
ranking to *Insert Feature, a constraint against inserting features (61). So while inserting a vowel
segment in violation of contiguity is less violation than creating a breathy sonorant, inserting
features on that vowel is a greater violation.
6 .1 .6, The behaviour of /1/. The laterals exhibit apparently exceptional behaviour
with regard to metathesis, in that [s.g.] does not metathesize off /hl/, with one apparent
exception:
exception:
d. dee-ga-lihgwadeega➔ deeka.llhgwa.deega 'he's turning him over'
dis-3sA-turn.over;prs
Thus hl is not behaving like a sonorant, in that it seems to be a satisfactory docking site
for [s.g.], and thus does not trigger metathesis. This is unsurprising if we take into account the
fact that hi is phonetically a lateral fricative [i], not a breathy approximant, and thus does not
36
violate *[s.g.• son].
Note that the lateral fricative is in a sense the [s.g.] counterpart of the lateral approximant,
in that an underlying sequence of /h/ adjacent to N does surface as [l]. For example, in (63)
deletion of the vowel between /h/ and N results in [l] (orthographically hl):
(63) aa-lihtawo ➔ aahltawo 'he's combing his hair'
3sA-comb.hair;prs
Presumably the contrast between a breathy lateral and a lateral fricative would be too slight to be
reliably perceived. Certainly Maddieson's (1984) survey identified no languages that contrasted
voiceless lateral approximants and fricatives. This potential distinction is thus neutralized in
favour of the lateral fricative in Cherokee, perhaps because of the greater salience of voiceless
fricatives compared to voiceless or breathy approximants (cf. Maddieson 1980).
Although the fact that hi does not trigger metathesis is in fact unexceptional, given its
phonetic nature, we now have to explain why [s.g.] does not metathesize Qll1Q N, if it is an
acceptable docking site. We can account for this fact by ranking Parse [sonorant], a constraint
against changing the value of [sonorant], higher than *[s.g., son]. Changing a lateral approximant
into a fricative, i.e. changing it from [+sonorant] to [-sonorant], is then a greater violation than
creating a breathy sonorant :5
6.2. Breathy vowel deletion. We now turn to the analysis of breathy vowel deletion.
This process is triggered by /h/ followed by a stop, or by /s/ (64-68). Note that
5The lateral fricativecan also conditionvowel deletion, howeverthis deletionis dsitinct from the processes
discussedhere. It is optional and only appliesto the vowel [i].
37
vowels are deleted. 6 Spreading of [s.g.] is simply a matter of the preferred timing of the [s.g.]
gesture with respect to a preceding vowel, possibly to allow more time to achieve a fully
abducted glottis. The motivation for deletion is presumably the elimination of a marked sound
type, breathy vowels. We shall see further evidence that breathy vowel deletion involves
spreading of [s.g.] when we consider the circumstances under which this process is blocked
(below).
If breathy vowel deletion is conditioned by [s.g.], the question remains as to why [s.g.]
only spreads from /s/ and /h/ preceding a stop, but not from breathy sonorants. I suggest that the
reason lies in the fact that these latter sounds are breathy voiced, and hence involve only partial
abduction of the vocal folds, allowing vibration to persist. Preceding a stop, /h/ is fully voiceless
(although it is breathy intervocalically), and /s/ is always fully voiceless. So these sounds involve
greater abduction of the vocal folds than breathy sonorants, and hence might might be more
prone to extending the duration of the abduction gesture. 7 The spreading constraint can be
formulated as in (69).
Deletion results if we rank this constraint above Parse V. There are then two conflicting
demands on a vowel preceding voiceless [h] or [s]: it must be breathy to satisfy Extend [s.g.], but
if it is breathy it violates *[s.g., son]. Given that we have already motivated ranking *[s.g., son]
above Parse V, and that Extend [s.g.] is ranked above Parse V, the best resolution of this conflict
is to delete the vowel so neither higher-ranked constraint is violated (70).
(70) *[s.g., son], *h >>Extend [s.g.] >> Parse V
(71)
/nas/ *[s.g., son] *h Extend [s.g.] Parse V
nas *!
nj:!.S *!
> n<a>s *
Breathy vowel deletion is blocked by high tone
Breathy vowel deletion does not apply to a vowel that bears high tone:
(72) aa-gohvvsda -x➔ *aakvvsda 'he burned it'
3s>3s-burn;prs
This blocking effect can be explained in terms of a constraint against the co-occurrence of
[s.g.] and high tone:
38
(73) *H
I
~
[s.g.]
This constraint is phonetically motivated because spreading the glottis has a lowering effect on
fundamental frequency, which would disrupt the realization of the high tone. The constraint is
also independently required in the analysis of Cherokee tonal phonology (Wright this volume).
Note that this account of the blocking effect of high tone depends on the assumption that
breathy vowel deletion involves the spread of [s.g.] onto the deleted vowel, because it explain the
blocking in terms of a constraint on this spreading, rather than on deletion per se.
Word-initial vowels are not subject to breathy vowel deletion before /s/:
(74) ' '
asgaya 'man'
'
usga 'head'
The status of deletion of initial vowels before /h/ is unclear because there do not appear to be any
words in Cherokee which begin with a sequence of the form NhT-/.
It is not clear what the best analysis of these facts is. Possibly there is a constraint against
deleting word-inital vowels. This analysis seems stipulative, but such a constraint might have a
basis in the importance of word onsets to lexical access (Marslen-Wilson and Zwitserlood 1989).
Another possible basis for an explanation for this phenomenon is the fact that words which are
underlyingly vowel-initial are typically produced with an initial glottal stop. There might be a
constraint against adjacent [constricted glottis] and [s.g.] features, since these features involve
contrary movements of the vocal folds, ..and this constraint would thus prevent [s.g.] .from
spreading onto a vowel preceded by a glottal stop. Some support for this analysis is provided by
the following form in which a vowel preceded by a glottal stop is not deleted by a following /s/:
6.3. Shared properties of metathesis and breathy vowel deletion. There are
two properties common to both metathesis and breathy vowel deletion: both are blocked by long
vowels, and neither applies to a vowel preceded by a [s.g.] consonant. As mentioned above, we
expect these processes to exhibit commonalities, because although they differ in their
motivations, both involve [s.g.] and both involve vowel deletion, so any constraints relating to
[s.g.] or vowel deletion are likely to affect both processes. We shall see that in each case, a
single, well-motivated constraint accounts for the shared property of the two processes.
6.3.1. Metathesis and deletion are blocked by long vowels. Metathesis does not
apply across a long vowel, and long vowels are never deleted (76-78).
39
(76) Deletion does not apply to /fVVh V/:
9ee
91nr1
9oohvv'i 'he saw us'
3s> 1p-see;pst
a.a.-goohwa.htii
'a 'he sees him'
3s> 3s-see;prs
Sequences of the form NVhT/ do not occur in Cherokee, so it is not possible to show that
breathy vowel deletion triggered by pre-consonantal /h/ does not apply to long vowels.
The non-application of both processes can be accounted for jointly, since we have
analyzed metathesis as crucially involving deletion. That is, metathesis across a long vowel
implies deletion of a long vowel (79), and if this deletion is impossible, then so is metathesis.
w-uuw-a.a.ka.hnvv'i-x➔ *wuuwa.akanvv'
i 'he's placing it'
way-3sB-place;prs
40
(83) Breathy vowel deletion does not apply to /[s.g.]VhT/:
Both effects can be analyzed in terms of a single 'OCP' -type constraint forbidding
adjacent [s.g.] specifications:
(85) *[s.g.][s.g.]
I I
r r
This constraint is undominated, so a vowel between two [s.g.] segments cannot be deleted
because the output would then contain adjacent [s.g.] segments in violation of the constraint (86).
Metathesis is blocked in the same way because it also involves deletion of a vowel preceding a
[s.g.] segment (87).
6.3.3. Deletion cannot feed metathesis. The analysis developed so far correctly
predicts that deletion cannot feed metathesis. For example, breathy vowel deletion applies to the
form shown in (88), resulting in a configuration (ganh) that appears to be an appropriate input for
metathesis, but metathesis does not apply.9
41
been reversed in violation of the undominated constraint Ordering, so this output is not possible.
A combination of deletion and metathesis will always violate Ordering in this way.
6. 4. Further issues. There are a number of exceptions to the rules proposed here.
Some are probably idiosyncratic lexical exceptions, but there are also some apparently
systematic exceptions that suggest further embellishments of the analysis.
6.4.1. Deletion before /t/. In the forms in (89),deletion applies before /ti, apparently
without any /h/ present.
(89) aag1-tahlawoosga ➔ aaktahlawoosga 'I'm getting angry'
1sB-get.angry;prs
A possible analysis of these data is to propose that these stems are h-initial underlyingly.
The /h/ doesn't surface after a long-vowelled prefix like /uu-/ because it cannot be syllabified: as
noted above, Cherokee does not permit sequences of the form [VVhC]. With a short vowelled
prefix, like /aagi-/, the /h/ causes vowel deletion.
Thus the forms in (91) could be blocked by syllabification constraints, but substantiating this
claim would require a full analysis of Cherokee syllable structure, which is beyond the scope of
this paper.
42
6.4.2. Other exceptions. The following are exceptions about which I have little to
say at present:
(93) .'
1-usga -x➔ *tsga 'heads'
pl-head
Feeling and Pulte (1975) show a number of exceptions to deletion involving the sequence
dohd (94). According to Pulte, this sequence has two sources: it can be part of the 'unintentional'
suffix /dohd(an)/ on verbs, or an instrumental suffix /dohdi/, forming nouns. In many cases the
vowel bears a high tone (94b), and thus is not expected to delete, but even when it bears a low
tone, deletion does not apply (94a), so perhaps these morphemes are lexical exceptions to
deletion.
7. Conclusions
The laryngeal metathesis and deletion processes in Cherokee are highly complex, being
conditioned by a wide range of factors. Furthermore, while they appear to be independent
processes in some respects, they share a number of common properties. We have seen that, in
spite of their complexity, these processes can be analyzed in terms of a ranked set of simple
constraints, most of which have clear phonetic motivation. In particular, many of the properties
of metathesis follow from an understanding of the process as involving the optimal placement of
the feature [s.g.].
43
phonology. This issue is raised most directly by metathesis phenomena, but is relevant to a wide
range of processes including epenthesis (cf. Kenstowicz 1994). We tentatively adopted three
constraints: Parse Association, requiring a feature to remain assocaited to its underlying position,
Parse Ordering, a constraint against reversing precedence relations, and Contiguity, which
requires that adjacent segments remain adjacent.
Acknowledgement
I am grateful to Pam Munro, Janine Scancarelliand members of the Cherokee Field Methods group for comments on
earlier versions of this paper. Many thanks are due to Virginia Carey for endless patience and good humour.
44
The Cherokee Laryngeal Alternation Rule
Pamela Munro
Cherokee, an Iroquoian language spoken in Oklahoma and North Carolina, has an unusual
morphophonological rule that renders many paradigmatic alternations in the verbal system
extremely opaque. The rule is unusual in several respects. It is morphologically triggered: no
specific phonological environment for its operation can be stated. While the rule operates strictly
left-to-right, there may be a considerable distance between the trigger morpheme and the affected
element. The effect of the rule is complex: any of a number of segmental or prosodic changes may
be said to follow from it. Finally, it interacts in complex ways with other prosodic phenomena of
the language, and is generally blocked from occurring at all in the environment of certain prosodic
patterns. Even the basic facts of this rule, which I will refer to (following Lindsey n.d.) as
Laryngeal Alternation, have been incompletely described in the literature, and no complete analysis
has been offered for it. In this paper, I will present a clearer description of how the rule of
Laryngeal Alternation works and suggest conclusions that follow from the data.1 An idea I will
consider in conclusion is that the current notion of the underlying phonological contrasts among
Cherokee consonants should be revised
Cherokee verb paradigms are difficult for the language learner to master because of
complex agreement morphology and morphophonemics. Consider the paradigmatic relationship
between the third-person singular and first-person singular forms of a Cherokee "active" verb in
the present tense. There are several morphophonemic alternants of the pronominal agreement
affixes for these two types of subjects. Thus, third-person singular subject verbs of this type may
show either a ga- or an a- prefix, as in the (a) examples below, and first-person singular subject
verbs of this type may show either aji- or a g- prefix:
I want to thank all the membersof thie1993 UCLA Cherokee group, without whose help the data presented
1
could not have been analyzedand understood,especially(of course) Mrs. VirginiaCarey, formerlyof Tahlequah,
Oklahoma,whose dialect is describedhere. Mrs. Carey's good humor and patience made investigatingthis topic a
pleasure. (Our group includedVictoriaAnderson,Filippo Beghelli,Barbara Blankenship,Mike Dukes,Edward
Flemming, Brian Potter, Nhlanhla Thwala, Kimberly Thomas, Bob Williams, and Richard Wright) Thanks also go
to GeoffreyLindsey and especiallyJanine Scancarellifor their help and inspiration,to Donca Steriadefor useful
ideas, and to Janine Scancarellifor supportivecomments on an earlier version of this paper. She has told me that
some ideas in this paper are reminiscentof analysesproposedindependently(and I am sure earlier)by Geoffrey
Lindsey, of which unfortunatelyI have seen no written version.
I should note that several dialectsof Cherokeehave been described,and I have only scatteredinformationon the
exact realization of the Laryngeal Alternationrule in most of them. Mrs. Carey's dialect agrees quite well with the
dialect (or mixtureof dialects,accordingto Scancarelli)presentedin Feeling (1975)and describedby Pulte and
Feeling (1975), and her speech is the major source for work by Scancarelliand Lindsey. Specificfacts may well
differ between dialects, but Scancarelli(1987: 20-21) suggests that the amount of interdialectalvariation is not great.
45
Cherokee has a complicated pronominal agreement system, with two series of prefixes (A and B),
each used for certain inn-ansitive constructions and for certain transitive constructions one of whose
arguments is third person, and a separate set of transitive agreement prefixes that mark
combinations of non-third-person subjects and objects (see Scancarelli 1987 for a complete
analysis of the allomorphy and occurrence of the different prefixes). 2 Each verb has different stems
used in different tense/aspect/mode configurations. (1-2) include the present stems of 'speak' and
'act silly', which remain constant throughout their paradigms. However, two rules operate to
greatly obscure the relationship between members of many Cherokee paradigms. One of these is a
complex of metathesis and associated vowel deletion (cf. Flemming, this volume), which we'll see
more examples of shortly. The second is Laryngeal Alternation, the subject of this paper. Here is a
simple example of this process:
(3a) is closer to an underlying form: the stem contains a hw preceded by a low-tone long oovowel.
(3b ), in contrast, shows the characteristic effects of the Laryngeal Alternation rule. The h has been
lost before thew, and the low-tone oohas been replaced by the low falling tone oo.These are the
two characteristic changes associated with Laryngeal Alternation: the first h in a stem is lost, and
there may be tonal alternations, typically with the introduction of a low falling tone. This
application of the rule is conditioned by the first-person singular A prefixji- or g-, which triggers
Laryngeal Alternation in all appropriate cases; the corresponding third-person singular A prefix a-
in (3a), on the other hand, does not trigger the rule's application.
I will not be concerned further here with describing the paradigmatic verb forms in which
the Laryngeal Alternation rule operates. King (1975), Cook (1979), and Scancarelli (1987) each
provide lists of the specific agreement prefixes that condition the alternation-the first-person
singular A prefix in (3b) is the only one that occurs on intransitive verbs, but roughly half of the
transitive prefixes also condition it; in some cases, two segmentally identical prefixes may differ in
terms of whether they condition Laryngeal Alternation or not Scancarelli (1987: 101-02) compares
the slightly different groups of prefixes that condition the change in different Cherokee dialects,
and provides a historical account of the spread of the rule. To facilitate comparison of the forms in
this paper, I have restricted most of my examples to present-stem verbs with third-person singular
versus first-person singular subjects, just as in (3). A prefixes also are used to mark subjects on
predicative adjectives and nouns and possessors on certain possessed nouns. In these cases, the
Laryngeal Alternation rule also operates, but I will not exemplify these here.
Cherokee has the following surface consonantal segments, in the orthography I will use for
most purposes in this paper. 3 This orthography is just about the same as that used in the popular
Cherokee dictionary by Feeling (1975; cf. Pulte and Feeling 1975).
2
To minimize the complexity of the examples in this paper, I present underlying as well as surface fonns in
only a few cases. In some cases, then, my segmentationsmay be somewhat impressionistic.A variety of
phonological rules that I will not treat here {butsee Scancarelli 1987: ch. 2) affect the form of the pronominal
prefixes. Abbreviationsand glossing conventionsare listed earlier in this volume. I generally follow Scancarelli's
conventions (1987) for glossing verbs, with no morphologicalsign of the present stem segmented.
3
Thanks to Janine Scancarellifor her advice on orthographicmatters. Scancarelli(1992) provides a
comprehensivediscussion of Cherokeeorthographies,includingSequoyah'ssyllabary.
46
(4) Cherokee surface consonants
(b) d dl j g gw
t tl ch k kw
s h
m n
hn
hl
w y
hw hy
T, tl, ch, k, and kw are aspirated and voiceless; b, d, dl,j, g, and gw are unaspirated and may be
weakly voiced in some environments. Ch andj are alveopalatal affricates; tl and dl are lateral
affricates; kw and gw are labiovelar or labialized velar stops. 4 The apostrophe(') represents the
glottal stop (which occurs as a surface segment only intervocalically in some dialects). Mand
especially b (which only appears in loanwords) are rare. Hn, hl, hw, and hy are generally
perceived as voiceless sonorants 5 with voicelessness maintained throughout, but sometimes they
seem more like clusters of h and the corresponding voiced sonorant. The orthographic
representation of these complex segments--standard in all recent sources-thus appears to accept
an analysis of these as phonologically derived.
t tl C k kw ?
s h
m n
I
w y
In (5), t, tl, c, k, and kw represent essentially the same sounds written as d, dl,j, g, and gw in the
practical orthography (4). (I ignore b. This could be added (asp, in this system), but it is
4
There is some controversy in the literature about whether the simplest forms of the affricates (j and di) and the
labiovelar stop (gw) should be analyzed as units, as argued for by Scancarelli (1987). I will assume a unit analysis of
these simplest forms here without further discussion. However, I report below a number of ways in which the
affricates pattern very differently from stops. (Cf. also fns. 12, 24, and 27 below.)
5
I intend the term "voiceless sonorant" only as a convenient label here. Flemming (this volume) shows that
surface hi patterns as an obstruent with regard to vowel deletion. I return to this topic below.
47
irrelevant.) In the analysis of (5), it is not only the voiceless sonorants that are seen as clusters of h
plus an underlying sound. The aspirated stops are considered to be similarly derived: thus aspirated
t is seen in the standard linguistic analysis as a cluster of underlying phonemic t plus h, and so on.
Thus, when t in this system occurs other than immediately before h, it has the same phonetic
realization as din (4). I will assume this linguistic analysis for the moment, but I will continue to
use the orthography in (4). Thus, when I want to refer to the underlying simple stop series given
above, I will use the "voiced" symbols d, dl, j, g, and gw.
Cherokee has a six-vowel system, the familiar five vowels plus a sixth vowel, generally
transcribed as a nasalized carat, which is written v in all sources. Nasalization is distinctive only
for this vowel, although other vowels are nondistinctively nasalized in certainly environments ..(i
Vowels may occur short and long; 7 we represent long vowels with doubled vowel symbols.
Cherokee has a complex prosodic system (variously described as a tone or pitch-accent system)
with seven different tones or pitches; I will follow Wright (this volume) in using the term "tone"
qereafter. (6) presents a comparison of how we represent these tones orthographically on syllables
containing the vowel a, keyed to the Feeling-Pulte system of tone numbers (written as superscripts
on vowels), which most recent sources often use for reference, with descriptive labels based on
Scancarelli (1987: 29):8
Primarily finally, following a nasal consonant,or separatedfrom another nasalized vowel by a glide.
6
7
It may be that some vowel length is predictable (cf. Foley 1980 and work by Kimberly Thomas). Certainly,
there are more long vowels than short ones. Note that Feeling (1975) and Pulte and Feeling (1975) do not represent
vowel length completelyconsistently(or perhapsthere are considerabledialect differencesin this regard):all vowels
written as underdottedare short, and all vowelsin syllables indicatedas open are long. But this is not an accurate
representationof the phonetic facts for all dialects, since vowels in closed syllables,though generally not underdotted
by Pulte and Feeling, may be either short or {rarely,but contrastively)long for Mrs. Carey.
8
The editors of Feeling (1975: iii) note that this system was analyzed with the help of Eunice Pike. I do not
present here the correspondencebetween our system of tonal representationand those used at various times by
Lindsey (1985, n.d.) and Scancarelli {1987,1992),which are not easily typable.
9
See Lindsey (1985) for more discussionof the realization of this tone and problems with its treatment in
Feeling (1975), and see Wright (this volume) for more infonnation on the system. (Victoria Anderson has also
studied related matters.)According to Lindsey (1985), some speakersmay produce tone 4 on true shon vowels; we
have not heard this from Mrs. Carey. Some contour syllables do appear to have vowels whose duration is somewhat
shorter than other long vowels, but we have analyzed them as underlyinglylong in every case.
48
Cherokee has a fairly extensive range of medial and initial consonant clusters, 10 but many
of these are derived by some form of the vowel deletion rule described by Flemming (this
volume).The main surface onset clusters are clusters of hors with another sound. H may cluster
with any other sound except 'or m, both of which occur only medially in the swface phonology of
the dialect I am describing. The case of sis special: Feeling (1975: x-xi) writes, "whenever a short
vowel is followed by ans, a faint his always present between the vowel and the s"; while we do
not always record these h's in Mrs. Carey's speech, they are often audible, particularly after a, and
they have also been noted by other scholars, as I report below. The other obstruents, or at least d,
g, and gw, work differently from the sonorants l, n, w, and y. The sonorants may occur in
underlying forms in clusters with h, but there is no contrast between preaspiration and
postaspiration in these forms, though conventionally such clusters are written preconsonantally
with the h before the sonorant. 11 However, medial stops may occur plain (d, g, gw), preaspirated
(hd, hg, hgw), postaspirated (dh, gh, gwh 12--0r, in our orthography, t, k, kw), and both pre- and
postaspirated (hdh, hgh, hgwh-or, in our orthography, ht, hk, hkw), as exemplified in (7).
Cherokee's morphological complexity makes it hard to find even near minimal pairs, but I have
controlled for environment by putting the stop after a low-tone short vowel in all but one case: 13
(7a) medial plain stops: wado 'thank you', Jalegi 'Cherokee', agwvvsa 'me alone'
(7b) medial preaspirated stops: galehdi 'I stand it up', wahga 'cow', guhgwe 'quail'
(7c) medial postaspirated stops: jitaaga 'chicken', akiyuusgeeni 'elbow', aakwiyvv'ee'a 'he's
paying'
(7d) medial pre- and postaspira1ted stops: aagoohwahtii'a 'he sees it', hlhko6de'a 'you're
shoveling it', sihkwa 'pig'
My recording of such examples follows the range of possibilities presented in Feeling (1975), but
is somewhat at odds with the description in such sources as Bender and Harris (1946: 18), who
10
My descriptionof and thinkingon questionsof Cherokee syllable structureowe a lot to input by every
memberof the UCLA Cherokeegroup.
11
Conventionally,in the Feeling-Pulteorthography(and, as far as I can tell, standard linguistic analyses),all h-
sonorantclustersare written as hS (whereSis any Cherokeesonorant)when prevocalic. The same clusters are
written as Sh when preconsonantal(as a result of the processesdescribedby Flemming this volume),except in the
case of I, which is written hi in all environments(cf. also Feeling 1975: x). These differences in the treatment of hn,
hw, and hy clustersreflect some perceptualdifferences:observedvoicelessnessdoes often seem to be greatestat the
beginningof an intervocaliccluster but at the end of a preconsonantalcluster. But there is no contrast, and there
seems to be no differencebetween 1and the other sonorantsin this regard. In this paper, I will follow the standard
orthographyfor all but preconsonantalhl, which I will write, followingthe treatment of the other voiceless
sonorants,as lh. (But see also fn. 5 above.)The differenttraditionalrepresentationsof how h seems to be sequenced
with a sonoranthas influencedanalysesof the Cherokeelaryngealmetathesisrule (in terms of whether h needs to be
viewed as metathesizingwith a precedingresonant);cf. Flemming (this volume).
12
Gwh (or kwh) representsariaspiratedunit labiovelarstop. Scancarelli( 1987:47-48 and personal
communication;cf. also fns. 24 and 27 below) cites argumentsfrom Lindsey that there is a contrast between a unit
gw (kw, in her orthography)and a cluster of k plus w, as well as similar unit-cluster contrasts between di (ti, in her
orthography)and d plus I. I regret that I do not yet understandthese arguments,and will assume for now that there
are no such contrasts.However,all the affricates(j. di, and aspiratedch, tl) have far more reslricteddistributionthan
than the labiovelarstops, as the discussionin the text and the followingfootnote should make clear.
13 I have not
yet elicted specificallyfor words illustratingthese contrasts.I'm reasonably confident that I'll be
able to present better examples in a later version of this paper. It should be noted, thought, that there are a number of
importantdistributionalobservationswhich may later prove to be important, such as the fact that in most (though
not all) cases where pre- and postaspiratedstops like those in (7d) occur in my data, the immediatelyprecedingsound
is what Scancarelli(1992) would call [+h], as discussedin the text following(38) below.
49
state "it is possible to consider each voiceless consonant as being phonemically h plus the
homorganic voiced counterpart, since h consonant does not otherwise occur", and, as we will see
below, in Scancarelli (1987).
Thus, Cherokee presents a contrast in how aspiration may combine with stops and non-
stops (the classes either of sand sonorants or of n and continuants) that provides nice support for
recent proposals about consonant structure by Steriade (1992b). She argues that "plosives (stops
and affricates) have representations in which their closure and release appear as distinct positions,
capable of independently anchoring distinctive features ... plosives have more clustering
possibilities than continuants, because plosives are bipositional" (2). Cherokee stops compared to
non-plosives (in Steriade's sense) seem to clearly validate this claim. The case of the affricates·,
however, is not yet settled. I can document a three-way contrast, between plain (j, di), preaspirated
(hj, hdl), and postaspirated (jh, dlh-or, in our orthography, ch, ti), exemplified in (8).
Conservatively speaking, it may be that there are no coda consonants at all in Cherokee
underlying forms. The main candidate for a coda consonant is h-Feeling (1975) regularly writes
h and h-sonorant clusters as codas, when they precede a following non-sonorant consonant, for
example. 15 And while h-sonorant clusters may occur initially (some at least of them derived by
vowel deletion), h-stop clusters never do. Further, it is likely that only short vowels may occur
before h-stop clusters (cf. Scancarelli 1987: 27). But whatever the proper analysis of possible coda
h, I will assume that, at some level, no other consonants occur underlyingly in Cherokee cooas.16
Let us return now to the Laryngeal Alternation rule. Here is a second example:
The two examples in (9) use similar allomorphs of the agreement prefixes, but these prefixes are
differentiated, once more, by their morphological effect on the following verb stem. Once again,
the h in the stem of 'put in water' (9b), like that in the stem of 'see' (3b), is lost. But in (9b) the
intervocalic h of (9a) is not simply deleted, but replaced by glottal stop. Alternations like this led
Scancarelli (1987) and Lindsey (n.d.) to refer to the (a) forms of such verbs as "h-grades" and the
14
There are other such gaps. I know of no instances of sj, sch, sdl, or stl clusters in Mrs. Carey's speech.
Feeling (1975) transcribes at least one stem, dasdlusga 'he's splitting it' (I omit vowel underdots and tone numbers),
with an sdl cluster. Mrs. Carey pronounces this word with sl, itself an unusual cluster.
15
This is clear, apparently, from the fact that superscriptedtone numbers follow syllables, as implied by
Feeling (1975: i:x-x).However, because the facts of vowel length do not systematicallyline up with what is claimed
in this explanation (cf. fn. 7), I assume this is (at least in part) just another conventionalizati.on.
16
As hinted here, initial clusters generally agree with what are assumed here to be possible medial onsets.
50
(b) forms, regardless of whether they contain a surface glottal stop, as "?- grades" or, as I will call
them, glottal grades. This terminology is reflected by my use of ";h" and";"' in glosses for the (a)
and (b) forms in paradigms cited here.
On the basis of the transparent hi' alternation in verbs like (9), King (1975) proposed that
the process we are concerned with should be generally seen as a morphologically conditioned
replacement of the first h in the stem of a verb with '; this same rule statement is retained by later
sources. Thus we would expect -goo'w- in the glottal grade of 'see' (3b). Lindsey (1985, n.d.)
accounts for the cases where the inserted glottal stop is preconsonantal, proposing that
preconsonantal glottal stop is realized as the low falling 1 tone on the preceding vowel, as in (3b ).
Cook (1979: 6), who reports that "in some [North Carolina?] dialects/?/ is realized post-vocalically
as a full glottal closure [?], while in others it is realized as glottalization of and fall in pitch of the
preceding vowel"; some overt preconsonantal glottal stops in the glottal grade are written
occasionally in Feeling (1975).17Lindsey's proposal explains the structure of (3b). We do not see a
low falling 1 contour in (9b), however, since the glottal stop is not preconsonantal and because, as
Scancarelli observes (1987: 56), low-tone ("unaccented") vowels are always short before
intervocalic glottal stop. As many examples in this paper demonstrate, the low falling 1 tone
develops not only as a result of Laryngeal Alternation; in addition, it appears on many pronominal
prefixes as a result of a regular morphological process of Tonic Glottal Insertion (Lindsey 1985:
136; Scancarelli 1987: 64-65).
We have now accounted for many of the consonantal and tonal alternations produced by
Laryngeal Alternation. The same process relates the examples in (10)-(12), which show a similar
loss of hand accompanying tonal change in the environment ofy, n, and I respectively:
In each case, the preconsonantal h of the stem (the form we see in the third-person subject (a)
form) is lost in the first-person subject (b) form, and the tone change is as predicted, with the
development of tone 1 in the first-person subject glottal grade. (The situation with other hf s can be
more complicated. We return to this matter below.)
11
Feeling (1975) consistentlytreats such preconsonantalglottal stops differentlyfrom prestop h's, generallynot
treating them as syllable-final(cf. fn. 15). I don't know the significanceof this is for Feeling's analaysis, however.
A good survey of dialectal variationinvolvingglottal stop is given by Lindsey (1985: 137-38).
18
The prefixjiiy- on this verb is a transitive prefix. In some examplesMrs. Carey simplifies the expected form
of this prefix to resemblethe intransitiveprefixji(i)- seen in previous examples;I give such shortenedprefixes an
intransitivegloss. Some tonal irregularitiesmay develop as a result of this morphological simplification.
51
Similar changes may occur in combination with metathesis of h (Flemming this volume):
Here, the h of the underlying hn cluster metathesizes over short unaccented a to produce an
aspirated initial k (13a) .. This process does not occur in (13b), since Laryngeal Alternation has
removed the underlying h in the verb stem; instead, the inserted glottal stop causes the development
of low falling tone 1. In addition to demonstrating a new type of Laryngeal Alternation pattern,
(13) shows that some Cherokee aspirated k's are indeed derived from g (the same pronominal
prefix seen in (12a), for iexample) plush.
Other processes may obscure the expected tonal alternations. For instance, as Lindsey
(n.d.) observes, when a distributive prefix (whether meaningfully indicating a plural object or, as
in (14), lexically required)precedes a verb in the glottal grade, following tones are raised.
The aspirated k of (14a) again is derived by metathesis. (14b) is structurally parallel to (13b), but
the presence of the distributive prefix causes the expected low falling 1 tone to surface as a high
falling 32. 19
We have not yet observed what happens when the first h in a stem is clustered with an
obstruent rather than a sonorant These cases are much trickier to describe and analyz.e, and are not
usually given as examples in discussions of the Laryngeal Alternation process. Although I question
parts of Scancarelli's analysis (1987, 1992) below, it should be remembered that she is the only
person to attempt a full description of the data.
Let's first consider what happens in stems containing s. Scancarelli (1987: 26-27) suggests
that in underlying forms all Cherokee s's are preceded by h, but that these h's do not surface after
long vowels. If underlying Cherokee s is always preceded by h, we would expect Laryngeal
Alternation to occur in verb stems containing s. The tonal alternations in paradigms like (15)
suggest that this is true: we assume an underlying stem loohsga in (15a); in the Laryngeal
Alternation form (15b), the his replaced by', leading to the development of the characteristic low
falling 1 tone.
19
There are some additionalcomplexities.First, other prefixes, may have a similar effect Second,and most
mysteriously,the raising process often appears to occur only in the glottal grade form. Note that it doesn't affect the
second-syllable2 tone in (14a). Lindsey makes other importantpoints about tonal alternationsthat are confirmedin
my data but which I will not discuss in the text, such as the fact that the LaryngealAlternationform of a 23 tone is
high level 3, as exemplified,for example,in aadlago6sga'he's scratching'/ gadlag66sgaTm scratching'(see the text
for a discussionof Laryngeal Alternationbefore s).
52
(15a) galoosga 'he is passing it'
ga-loohsga
3sA-pass;prs;h
In (15), the underlying hs is preceded by a long vowel; when the preceding vowel is short, as in
(16)-(18), vowel deletion (Flemming this volume) occurs in the third-person singular subject form,
while the first person singular form develops the same low falling 1 tone as in (15b):
Then and l in verbs like (16a) and 1(17a)devoice before (h)s, so they are comparable to the hn and
hi in examples we've seen earlier. This makes pairs like (16) and (17) initially look similar to cases
like (11) and (12) in which Laryngeal Alternation results in paradigmatic alternation between an
underlying h-sonorant cluster and !thecorresponding plain sonorant. However, as the underlying
forms here show, the h involved in Laryngeal Alternation in these forms is not underlyingly
adjacent to the sonorant. The location of the derived low falling 1 tone is crucial- in (16b) and
(17b), it appears in the syllable before s, while in examples like (11) and (12) it appears in the
syllable before the sonorant. (Example (16) demonstrates that the locus of the Laryngeal
Alternation rule may be separated by two underlying syllables from the prefix that triggers the rule.
Such cases are common, and in theory the distance could certainly be greater.)
(18) and (19) show comparable cases where an obstruent (g or J) underlyingly appears in
the syllable before (h)s. Vowel deletion produces a Chs sequence, which is realized as a voiceless
and possibly lightly aspirated C before the s:20
20
There is some dispute about this. The UCLA Cherokeegroup elected to write some Cs clusters resulting from
this deletionas unaspirated(d, g, etc.), departingfrom the standardFeeling (1975) orthographictradition.I retain the
traditionalorthography here (using t, k, etc.). Crucially, there is no contrast in stop aspiration in this position.
53
(19a) aayeetsdffi 'he is.laughing at him'
a-yeejahsdii'i
3sA-laugh.at;prs;h
(19b) ji-yeejaasdii'i 'I am laughing at him'
lsA-laugh.at;prs;'
(19) is only of only a few available Laryngeal Alternation examples involving affricates.
Underlyingly, the stem of 'laugh at' includes the sequence -jahs-. Vowel deletion prcxiuces -jhs-,
which we would expect to surface as -chs-; however, the affricate is realized preconsonantally as ts
(Scancarelli 1987: 25). The expected ts-s-d cluster in (19a) is simplified to tsd.
If an underlying stem contains a cluster of h and a stop or affricate, several different forms
of Laryngeal Alternation can result
When the underlying form of the verb includes an hC cluster, where C is a stop or
affricate, aspirated or unaspirated (in the available examples, t (i.e., dh), d, g, or di), the third-
person (a) subject form shows deletion of the vowel before the h, and in the first-person (b) forms,
Laryngeal Alternation replaces the h, there is no vowel deletion, and the expected tonal alternation
occurs. This happens both when the original vowel-h-consonant sequence was preceded by an
obstruent, such as the third-person prefix ga- in (20a)-(22a}-
21
I have no explanationat presentfor the tonal alternationon the oo stemvowelshere.
54
(24a) aadaanh1te'he's thinking'
a-adaanvhte
3sA-think;prs;h
If the first h in the underlying stem appears immediately to the right of a stop, however,
results appear to vary lexically.
We would expect that if an aspirated stop was analyzeable as a stop-h sequence (as we
have seen in the aspirated stops derived by metathesis, as in (13a) and (14a)) thatremoval of the h
during Laryngeal Alternation would produce an unaspirated stop in the first person form, and this
happens in about half the cases I have examined, as in (25)-(26) below (and also (27)-(28)). On
the basis of what we have seen so far, however, we can have no expectation about what should
happen tonally in such cases. Glot1talstop cannot simply replace the postconsonantal h in a verb
like (25a) and (26a), since Cherokee ' does not occur postconsonantally. A new low falling tone 1
may develop before the affected stop, as in (25). (In cases like (26) the development of first-
person singular tone 1 appears vacuous.)
Just about as often, however, the vowel before the unaspirated stop in the first-person singular is
short and unaccented, and there is no tone 1:
55
(29b) g-uutee'a 'I am picking it up'
1sA-pick.up;prs;'
Thus, in (29-31) the only sign of Laryngeal Alternation is in the tone, and sometimes the
tone "change" in the first-person singular forms appears vacuous in comparison with the third-
person forms. When I began studying the Laryngeal Alternation rule, I originally interpreted these
cases as rule failures, since the rule appears to have no effect in such forms. These patterns were
matched by other cases like (32), in which, again, there is no visible alternation:
(32) g-uuhlv'sga 'he's putting a lid on it', 'I'm putting a lid on it'
3sNlsA-put.lid.on;prs
In such examples, as Lindsey (n.d.: 5) has observed, the presence of a preceding high 3 tone
blocks the application of Laryngeal Alternation. 22 Once principled restrictions like this one are
realized, there are only a handful of problem cases among the hundreds of verb paradigms in
Feeling (1975). Laryngeal Alternation is thus a very regular (though complex and still not well
understood) rule. 23
Aspirated stops can thus be involved in three types of Laryngeal Alternation rule
alternations. First, like unaspirated stops and affricates (21-22), they can be preceded by h in the h-
grade and appear intervocalic after a 1 tone in the glottal grade (20, 24). Next, they can become
unaspirated in the glottal grade (again often with development of a 1 tone) (25-28). Finally, they
can appear unchanged in both forms, with the only difference being the 1 tone in the glottal grade
(29-31).
56
may be analyzed as always preceded by an /h/. This /h/...alternates with fl/ in the h-grade ~ ?-grade
alternation ....An /h/ which precedes an obstruent is dropped by a low-level rule if it in tum is
preceded by a long vowel" (1987: 26-27). By this analysis, we would assume that all three types
of aspirated stop paradigms are underlyingly e.g. hdh. The pre-stop h's surface only in forms of
the paradigms illustrated in (20) and (24) because only in those forms are they preceded by short
vowels. Cases like (29-31) have long vowels before the unchanging aspirated stops; these long
vowels block the appearance of the prestop h's in the h-grade. But there seem to be problems with
this account. (25-28) also show long vowels before aspirated stops in the h-grade. We would have
to assume that these stops also were preceded by underlying h. But in these glottal grades, it is the
h after the stop that is lost, and there are frequent tonal anomalies as well. 25 Another problem with
this proposal is that it is simply not true that all surface aspirated stops are preceded by h when they
follow a short vowel. (7c) and (8c) above include words containing aspirated stops and affricates
preceded by short vowels; there are a number of others with t, k, and ch in this environment
(though I am currently lacking any such examples of kw or tl). So this proposal will need
amendment if it is to handle all the data.
An alternative is to assume that--contrary to the linguistic analysis that has been assumed
since the work of King (1975) at least-Cherokee has a contrast between two types of aspirated
stops: aspirated stops derived by coalescence of a sequence of stop plus h, and underlying
aspirated stops. Under this analysis, when any stop is preceded by h in the underlying form of a
verb, that h will simply be replaced by ' in the glottal grade. When the Laryngeal Alternation
trigger is an aspirated stop or the h of a Ch cluster,' will be inserted before the whole cluster, and
h, if there is one, will be lost. This suggests that Laryngeal Alternation is not a simple replacement
of one sound by another, but rather a two-step process by which, first, ' is inserted to the left of
any consonant or consonant cluster that includes h or an aspirated element, and next, the first h in
that sequence, if there is one, is deleted, as outlined in (33):
(33) Assume an underlying contrast between unit unaspirated d, g and aspirated t, k, along
with derived (cluster) dh, gh (with tldh and klgh phonetically neutralized). Laryngeal
Alternation first inserts ' before the first single consonant or consonant cluster that includes
h or the feature [+aspirated]. Then Laryngeal Alternation deletes the first h, if there is one,
of the resulting '-initial sequence.
Thus Laryngeal Alternation operates differently on the following strings:
/ ... dh .. ./ > ...'dh ....>.. .'d ...
/ ... t. . ./ > ...'t ....
Although this solution appears to account for the facts, it violates naive notions of underlying
economy and seems a bit add hoc. I am hopeful that additional evidence from the paradigms above
will help further this investigation. In particular, I plan to survey second-person singular forms of
these verbs, which often seem to be closer to underlying stem forms. For instance, we have
assumed that the underlying form of the present stem of 'use' (21) includes an hd sequence, even
though no h appears in either the third- or first-person subject forms. The second-person form,
however, does include this stem h:
Similarly, we have assumed following (33) that the underlying form of 'swallow' (28) must
include a derived aspirated stop not preceded by h (i.e., gh or, in our system, k), and the second-
person subject form confirms this:
containsa cluster,but I wouldargue it is a two-membercluster,of unit gw plus h. As far as I can tell, gw and kw
behavelike all the otherunaspriatedandaspiratedstops.
25 Scancarelli(1992:138)
appearsto acknowledgethat caseslike theseconstitutea problem.
57
(35) h-lklsga 'you're swallowing it'
2sA-swallow;prs;h
More such evidence will hopefully help either to validate or to disprove the suggestions advanced
here. The final analysis may invoke additional contrasts, perhaps among medial derived and
underlying aspirated stops preceded and not preceded by h.
Finally, there is one more possible obstruent to consider, hl, which displays two types of
exceptional behavior. Although I included this sound in the class of "voiceless sonorants", it is
phonetically a voiceless lateral fricative and thus an obstruent, in contrast to hn, hw, and hy, which
may be described as breathy sonorants (Flemming, this volume). Unlike the other members of the
h-sonorant class, hl participates in paradigmatic alternations in which its h is not lost in the glottal
grade:
The pair in (36) is extremely similar to that in (12). But in (12), the characteristic tonal change is
accompanied by a change of h-grade hl to glottal grade l, which we have analyzed as usual as
showing the loss of h. Why, then, do we not see a change of hl to l in (36b)? In fact, (36) looks
like (29-31), other cases in which, I have suggested, underlyingly aspirated obstruents identify a
locus for Laryngeal Alternation, but do not lose an h, because they have no h to lose. One analysis
of cases like (36), along the lines of (33), might be that the hl in such cases is an underlying hl.
Flemming considered that an underlying hi might explain the failure of h to metathesize in cases
like (37a) (in contrast to apparently similar forms like (13a) and (14a) above). However, as he
observed, the glottal grade form in (37b) shows that the hl in 'sleep', like that in 'get full', but
unlike that in 'make', is decomposable into h plus l.
Another surprising thing about hl is that sometimes this sound appears to reflect a lateral
affricate. Consider alternations like
Since (38b) is a glottal grade form containing dl, we would expect that (38a) would contain either
hdl or ti: either of these h-grade forms should yield dl in the first-person glottal grade. But instead
(38a) includes hl. Feeling (1975: xviii) reports that many Oklahoma Cherokees are "converting tl to
hl in many words", 26 suggesting that we should assume that conservative pronunciations of (38a)
26
Mrs. Carey alternates hi and ti in the negativeparticle hla /tla, for instance.Scancarelli(1987:47-48) reports
that some OklahomaCherokee speakerspronouncethe word 'he is having a nightmare',whichFeeling (1975) writes
58
would include tl rather than hdl. Thus, (38) reflects an original alternation much like those in (25)-
(28).
This suggests another, potentially more plausible account of the non-alternating hfs in
verbs like (36). I noted above that we might account for such cases by assuming that Cherokee has
an underlying hi in addition to the hi derived from clusters of h and I. However, there is a second
possibility: that a verb like (36) contains not underived hi, but rather a ti changed to hi by the same
process seen in (38). Then the similarity between (36) and cases like (29)-(31) is even closer, since
(36), like them, may be analyzed not just as an obstruent, but as an original "plosive".Xl
Now let's return to the interaction of hand stops versus hands. Scancarelli (1992: 139)
(following suggestions by Lindsey)211 proposes that laryngeal features [±h] and [±?] may be
associated with Cherokee consonants:
Both the aspiration and glottalization features may be associated with the same
consonant ...../h/ typically affects sounds before and after the consonant with which it is
primarily associated whereas/?/ affects only sounds before the consonant with which it is
primarily associated.
This description recognizes an important distinction between aspiration and glottalization (or h and
') in Cherokee: aspiration (h) may either precede or follow another consonant, but glottalization (')
may only precede a consonant. 29 It is true that glottalization interacts in the same way with every
class of Cherokee consonants. However, different classes of consonants interact with h in different
ways. While stops and possibly also affricates may be either preceded or followed by h, or both,
sonorants and s are different. Sonorants coalesce with h; while there seem to be perceptual
differences in different environmen1tsas to whether the voicelessness of the h comes before or after
any voiced part of the sonorant, there are no contrasts: in any given environment, a sonorant may
combine with h in only one way. There is some analytical confusion abouts. Many s's appear to
be preceded by h; other s's show the characteristic tonal affects associated with preconsonantal
glottal stop. Are these mutually exclusive, or can it be true, as Scancarelli suggests (1992: 150 and
personal communication), that "/s/ may be regarded as always being [+h]; it is sometimes also
[+?]" when it undergoes Laryngeal Alternation? Such sources as Bender and Harris (1946: 18),
Reyburn (1953: 175), and Foley (1980: 107) agree on transcribing h before sin h-grades but not
in glottal grades. However, these sources are not trustworthy on other matters, especially
as ahligi'a (tone and underdotssu~). with an aspirated ti. Mrs. Carey, as Scancarelliconfirms,has reanalyzed
the hi (from older t{) in this word as an h-l cluster, as shown by her forms aahligit'a 'he's having a nighbnare' /
galigif'a 'I'm having a nighbnare'. Cf. also Scancarelli (1992: 143).
27
Scancarelli(1987: 48) provides more support for this idea, suggestingthat some OklahomaCherokee speakers
pronounce verbs like 'place (something)on (something)',in which both Feeling (1975) and Mrs. Carey have non-
alternating hl, as in (37), with aspirated tl. (Mrs. Carey's forms are aahlahvsga 'he's putting it on it'/ jiilhahvsga
'I'm putting it on it'.)
Scancarelliproposes (again citing Lindsey; cf. fn. 12) that the contrastbetween 'have a nightmare' (fn. 26) and
verbs like 'place (something)on (something)'is between a cluster d-l (Scancarelli'st-1,itself clustered with h) in
'have a nighbnare' and a unit di (Scancarelli'sti, plus h). It's clear that the facts regarding all the the affricates are
complicated.The alternativepronunciationsof 'split' reported in fn. 14 above may be relevant for the question of
whether there are unit versus cluster dfs.
28
Scancarellicites a talk given by Lindsey in 1987, an earlier version of Lindsey (n.d.). Unfortunately,many of
the most provocativeclaims in the talk (which I heard myself, but do not remember accurately!)do not appear in the
written version.
29
Cook (1979: 6) appears to disagree with this statement: "Post-consonantallynt is always[?]." However, he
gives no examples, and goes on to say, "Words written with an initial vowel have a/!/ automaticallybefore the
vowel in several contexts. Since the occurrencesof this /! I are predictablefrom phonetic context it is not written".
Thus, perhaps he is not claiming that word-internalpostconsonantal' exists.
59
involving suprasegmentals, so we should not rely on them too far. Possibly further instrumental
studies will prove valuable here.30
As noted earlier, the popular orthography for Cherokee (as in Feeling 1975, or essentially
that used in this paper) is very different from the standard linguistic orthography. The linguistic
analyses that the linguistic orthography reflects are supported by unaspirated-aspirated alternations
resulting from Laryngeal Alternation and, much more extensively, metathesis. But in fact most
Cherokee stops in most words do not participate in such alternations, which arise only in certain
specific environments. Cherokee may exhibit a process of gradual lexicalization of originally more
productive contrasts in aspiration, by which only a few aspirated stops in the lexicon (crucially,
those that occur as the fust aspirated element in stems that take A prefixes) must be specified as as
clusters or units, while most other aspirated stops are unspecified for this history. Scancarelli's
survey of the different representations of aspiration in different ways of writing aspiration in
Cherokee (1992) confirms that this is an ongoing concern for speakers.
More evidence must be considered before reaching final conclusions about the consonantal
alternations we have observed in the Cherokee h- and glottal grades. Steriade's claims concerning
the structure of different types of consonants provide a motivated account of why stops (and to
some degree affricates) behave differently from sonorants ands. At present, the evidence suggests
that we should reevaluate the standard linguistic analysis of Cherokee underlying segments, since
adding an aspirated stop series at some underlying level may facilitate the description of the very
productive process of Laryngeal Alternation.
30
Thomas'sfindings(cf. fn. 7) may confum the reports in the text; I am not sme if her recordingsare
specificallygrade fonns, however.
60
Classificatory Verbs in Cherokeei
Barbara Blankenship
Cherokee is among the American Indian languages that have classificatory verbs, a system
where the choice of verb is determined by the shape or some other quality of the item that serves as
subject (of an intransitive verb) or object (of a transitive verb). There are at least 40 Cherokee verbs
that have classificatory variants (King 1978).
The C form is also used in questions where the identity of the object is not yet known.
Previous research
Haas (1948) has the earliest modem linguistic reference to Cherokee classificatory verbs.
Since the article focuses on Muskogee, the Cherokee examples are given without analysis, except
for a description of the five categories of physical objects that the verbs refer to.
61
zero-morphemes and suppletive forms.
King (1978) gives five category-stems for each of 40 Cherokee classificatory verbs-.
Several possible classificatory morphemes are identified, but there is no attempt to group them
systematically. On the basis of deficient verb sets (to be described below), King theorizes that the
categories originally distinguished items on two axes: rigid/flexible and solid/liquid.
The section on classificatory verbs in Cook's (1979) grammar of North Carolina Cherokee
is based on Reybum's (1954) conclusions, but carries the phonological analysis further. His
discussion of four of the category-stems for each of 16 verbs traces the surf ace morphemes back
through phonological rules to determine their underlying forms. There is no attempt to relate the
forms systematically to each other. Instead, they are used to display the phonological mechanisms
of verb construction.
Since there has been no published study analyzing the Cherokee classificatory morphemes
in terms of an overall semantic system as Davidson et al "(1963) have done for Athapaskan
languages, this paper will begin to develop such a system, and discuss its relation to Mithun's
(1984) theory of noun incorporation processes.
Stem structure
Each verb stem in examples (1-5) comprises a class morpheme (-kaa-, -neeh-, -nw-,etc.)
and a verb root (-nee- ).2 (The root in (4) is reduced to -ee. In Appendix A, groups la, 1b, and 4
present other examples where the root for the L class verbs is different from the root for the other
classes of the same verb. Why this is so is a topic for further study.)
The usual position of the class morpheme is after the prefixes and immediately before the
verb root, as illustrated in (1) through (5). There are exceptions where the class morpheme follows
the root, as in (8) through (10).
2Throughoutthe paper, "root"refers to the verbroot withouta class morpheme,and "stem"refers to the
combinationof root and class morpheme.
62
Semantic range of classes
Since other Native American languages exhibit verb class distinctions for as many as 13
classes, further classes were sought in Cherokee as well. Following the method described in Basso
(1986), verbs were elicited in the frame 'She is holding X', with about 100 different nouns in the
position of X. The semantic possibilities, adapted from the Davidson et al (1963) description of
Navajo, included small grains (sand, sugar), a soft plastic mass (mud, dough), a fluffy mass
(wool), ropelike objects, fabric, parallel long items (wood in a woodpile), bulky items (crate,
barrel), animals too large to carry, and an aggregate of small items (a handful of coins).
None of these items elicited separate classifiers in Cherokee, nor were other classes
discovered accidentally in the course of the elicitations. Thus we can be reasonably certain that
there are only five classes.
It is interesting to observe the semantic ranges of the Cherokee classes in detail. The
animate (N) class includes all animals and plants, but not their parts. Thus a tree is N, but its fruit
could be C, L, or F depending on its shape and rigidity. Animals are N, but their meat is usually
C, (hotdogs are L).
The animate class need not be rigidly adhered to, however. Depending on which quality the
speaker wishes to emphasize, a cat can be N or F, a tree can be N or L. When the verb does not
have an N stem (see discussion of deficient verb sets below), all animals are classified as F, even
those that seem rigid, such as bugs and turtles.
For materials that can assume the shape of their container, the quality of the container often
determines the class. Sugar in a bowl is C, but sugar in a bag is F. Honey in a comb is L, but
honey in a jar is C. Liquids are also classed by the shape of the container when the verb does not
offer a Q stem.
Regardless of their shape, furniture and other items made of wood belong to the L class,
presumably because lumber is long. Tools are classed not by overall shape, but by the shape of
their most salient part. Thus a hammer is L because the head is long, while an axe is C because the
head is compact.
63
When there is no N form, N nouns take th1eF verb, i.e., a verb containing what is usually
the F morpheme for that kind of verb. Thus, for example, 'he is finding N' and 'he is finding F'
both are expressed with ganvvhwtf'a. We know that the morpheme -nv- in this verb usually refers
to F objects, as illustrated by the pair uunv'a 'he has F' and uuwaakdha'he has N' (-ka- is the N
morpheme.) Other such examples may be found in Appendix A.
When there is no Q form, Q nouns take the L or C verb, depending on the shape of the
container in which the liquid resides. There is always a C form (unless it is semantically
impossible, as in 'fall over'), which acts as the default verb in the absence of other forms.
The majority of classificatory verbs have to do with handling physical objects: 'have',
'hold', 'handle', 'break', 'drop', 'give'/'get', 'carry'/'leave behind', 'hang up'/'take down',
'hide' /'find', 'pick up' /'set down', 'put into water'/'take out of water', 'put into fire' /'take out of
fire', 'put into a container' /'take out of a container'. Some less concrete ways of handling objects,
such as 'send', 'eat', and 'wash', are also classificatory. There are a few intransitive verbs as well,
usually related semantically and morphologically to a transitive verb in the set:
Some transitive verbs that seem equally concrete are not classificatory. The family does not
include 'catch', 'touch', 'move', 'lift', or 'cut'.
The choice of set is determined by the meaning of the verb. Sets la and 1b are by far the
most frequent, and appear to the be default sets for verbs whose meaning does not place them in
one of the specialized categories that follow; la and lb differ only in the morpheme for L objects.
Set 2 is used in verbs that relate to objects in the hand: 'hold', 'handle', 'take somewhere by
hand'. Set 3 appears in verbs relating to containers: 'contain', 'put into a container', 'take out of a
container', 'hide', as well as the nouns for containers. Set 4 occurs with verbs having to do with
hanging: 'hang' (intransitive), 'hang up', 'take down'. Set 5 is found only in 'fall' and 'drop'. It is
of interest that Davidson et al (1963) found sets semantically similar to 2 and 5 in Athapaskan
64
languages.
For the most part, the class morphemes do not appear related to modern Cherokee nouns or
adjectives. The morpheme -neh- is cognate with the northern Iroquoian noun stem *-hnek- 'liquid'
(Cook 1978:42, Mithun 1984: 884). The morpheme -yvh- may be related to the Cherokee noun
yvwi 'person'. 'Person' is associated with the L class in some languages (in the Yuman family for
example), and may have been Lin Cherokee at an earlier phase. According to Mithun (1984: 865),
a study of incorporation not limited to American languages, it is common for a noun with a narrow
meaning to be incorporated into a verb as the classifier for a broad class of meanings. Her example
is the Caddo noun root -'ic'ah- 'eye', which when incorporated stands for any small, round object.
By the same process, 'person' could come to stand for long objects in Cherokee.
Mithun (1984:865) theorizes that since the C morpheme in Cherokee usually has the
shortest form, it probably evolved from a plain verb root with no incorporation. This theory also
accords with the fact that even defective sets always have a C form. If the C form is derived from
the naked verb root, then of course there would be no modern Cherokee noun related to the C
forms.
65
1975:284). Thus these nouns are probably derived from verbs.
The verb 'contain' distinguishes the classes of .bQ1hthe container and the thing contained. A
flexible container like a sack (F) requires a different set of verbs from those of a rigid container like
a box (L) or basket (C). These forms are shown in examples (16) and (17). It will be seen that the
set for C or L containers is closely related to the noun forms in (12-15). The set for F containers
may have originated as a non-classificatory verb gaanhda,which borrowed the L, Q, and F forms
from the set shown in (16).
C L Q F N
(16) 'C or L contains' gahldi galdfisdi aadlfisdi gruvvdi gatw
( 17) 'F contains' gaanhda galdi1sdi aadlfisdi gruvvdi gaanhda
English constructions like 'a bag of sticks' are expressed in Cherokee by 'a bag containing
sticks', leading to complex verb selection requirements. Examples ( 18) through (21) illustrate
some of the possible combinations. In each of the examples, the participial verb 'containing' is
selected to agree with 1theshape of the container (C/L or F, as in 16-17), and incorporates a
morpheme indicating the shape of the object contained. The main verb 'hold' incorporates a
morpheme indicating the shape of the container being held. (See Cook 1979:130 for a description
of participial forms ending in -vv'i.)
66
lncorpor ation
The presence of class markers in the classificatory verbs hints that the incorporation of nouns
into verbs may once have been a characteristic of Cherokee, as it is today in northern Iroquoian
languages. (It should be noted, however, that Cherokee is the only Iroquian language that has
classificatory verbs (King 1978).) Mithun (1984) delineates four types of incorporation. Each type
develops from and co-exists with the previous type. That is, if a language has productive Type III
incorporation, it must necessarily have Types I and II as well, although they may or may not be
productive. We will outline the hierarchy of types, and then see where Cherokee fits into the
system.
Type I. A verb stem and a noun stem are combined to form an intransitive predicate. The
noun loses its individual semantic salience; instead of referring to a specific entity, it simply
narrows the scope of the verb. An example is:
(22) I am coconut-grinding.3
The incorporated noun also loses its syntactic role in the sentence; it cannot be marked for
definiteness, number, or case, nor modified by demonstratives or adjectives. Particles that
normally cliticize to the verb will cliticize to the noun-verb combination. The process of
incorporation lowers the valence of the verb by deriving an intransitive predicate from a transitive
one.
Type II. This process is identical to Type I, with the· addition that an oblique argument
advances into the case position vacated by the incorporated noun. Thus when 'face' in (23) is
incorporated, 'his' becomes the direct object, (24).
Type III. Incorporation is used to manipulate the discourse structure. New noun items are
presented in full case positions, but old items are put into the background by incorporation. 'Meat'
is a new item in (25), but old in (26).
Example 28 illustrates a Type IV paragraph in Cherokee. Thus we know that Cherokee achieved
67
productive Type IV incorporation at some point in its history, and, according to Mithun's theory,
must also have had Types I through III. It does not, however, give evidence whether incorporation
is currently productive, a question to be addressed later in this discussion.
'John's mother gave him a shirt. He hung it up in his room. The next day
he couldn't find it. His sister had hidden it.'
Mithun further stattes that productive use of incorporation may reach any stage (Type I-IV)
and then decay, leaving lexical fossils of the incorporation process. The Cherokee classificatory
verbs appear to be fossils of Type IV incorporation. The use of incorporated NV combinations is
lexicalized for certain verbs, but is never extended. innovatively to other verbs. This point will be
discussed further below.
Morphemes of sets 2 through 5 were not yet full Type IV classificatory morphemes. Their
application to a limited semantic range of verbs (e.g., 'containing', 'hanging') suggests that they
had retained their original meanings denoting a specific item in the physical world, and thus their
incorporation was limited to the small group of verbs that express possible actions upon that item.
Example (30) illustrates Type I incorporation, where the direct object 'shirt' has been
incorporated into the verb, turning it into an intransitive predicate. Example (31) shows Type II
incorporation, where the oblique argument 'baby' has advanced to the direct object position
vacated by 'shirt', triggering a transitive agreement prefix on the verb.
68
When the incorporated element is modified by an adjective, the incorporated noun can
optionally be stated as a separate argument. fu (32) the noun diktinutdi is omitted, in (33) it is
present. (The examples are from Feeling 1975: 87 and 71; no tones were provided.)
(32) Agalisgvm y-edo julsihgi d-u-ktj.nvto'i.
sunshine cnd-3sA;walk.around dark dis-3sB-glasses. wear;prs
'When she's out in the sun, she wears dark glasses.'
(33) Igada aagalisgv y-anedoha juls_ihgi diktinvtdi
some sunshine cnd-3pA;walk.around dark glasses
d-ana-ktinvtvsgo'i.
dis-3p> 3p-glasses.put.on;prs
'When some people are in the sun, they put on dark glasses.'
But it is not clear whether these examples are lexical forms or show productive use of
incorporation. Attempts to elicit incorporation of a less standard item of clothing yielded:
(34) 'I am putting on a watch.' Waaji nigvvnhdi'a.
watch 1sA;put.on;prs
(35) 'I am putting a watch on the boy.' Aachuuja waaji nijiyvvnhdaanee'a.
boy watch 3s>3s;put.on;prs
There is no verb incorporating 'watch' in (34) and (35). Nlgvvnhdi'a may function as a general
verb for articles of clothing that don't have incorporated verb forms in the lexicon. Like the
clothing verbs, nigvvnhdi'a has related forms for 'have on, wear' (nu 1wanh3di) and 'take off'
(nj2gvnh3de3'a),(examples fromFeeling1975: 148 and 147, transcription unmodified). Note also
that if nigvvnhd- is assumed to be an incorporated morpheme, the remaining markers for 'put on'
(-i') and 'take off' (-e') are the same as those for the clothing verbs, (see appendix B). Unlike the
clothing verbs, however, it is used for non-clothing items ('put a decal on the car', 'put a lock on
the door', 'put blinds on the window'). Our consultant says the verb nlgvvnhdi'a is not used for
clothing at all. 'Watch' is not viewed as an article of clothing in (34-35). But it is impossible to
determine whether this point of view is due to the fact that there is no lexical verb form for 'put on
a watch'.
(37) 'I am putting swim fins on the boy.' Aachuuja aadla dee-J1laasuhlvvsga.
boy rubber dis- ls> 3s;shoe.put.on;prs
Our consultant says that (36) and (37) would also do for 'put on galoshes', or any other footgear
made of rubber. Thus the verb 'put on shoes' is being extended to other kinds of footwear by use
of an adjective to specify the object This construction is analogous to the 'dark glasses' example in
(32-33). There is no way to incorporate 'galoshes' or 'swim fins' into a verb. 4
4our consultantsays that (37) does not allow a Type II form with 'boy' in object position. It has not been
determinedwhetherthis is an idiosyncracyof the particularverb, or whetherit always applies to incorporatedverbs
with adjectivalmodifiers.
69
Thus the evidence is fairly certain that incorporation is nonproductive with items of
clothing. When it was productive, it was of Type IV, where a fairly general noun stem ('shoe')
was incorporated to narrow the scope of the verb, while a more specific lexical item ('rubber')
identified the argument of the verb. This evidence supports Mithun's (1984:883-884) conclusion
about Cherokee incorporation.
Summary
This paper has confirmed that Cherokee classificatory verbs distinguish five classes of
objects in the physical world and established that there are at least six sets of morphemes used to
mark the differentiation. The incorporation process that derived the class morphemes is no longer
productive, nor is incorporation productive in verbs applying to articles of clothing. (Verbs
applying to body parts, not attended to in this study, may offer further insights on the question of
productivity.) Despite its lexicalized status, the classificatory system can be remarkably complex in
actual use.
70
Appendix A
Words in parentheses are borrowed from elsewhere in the paradigm to fill a gap. This
phenomenon is discussed in the section on deficient verb sets.
Those marked with * have the class morpheme after the root; the rest are in the more standard
order with the class morpheme before the root
Gloss Root C L Q F N
71
4 -t- -too(s)- -d-
hang (intrans) -v- gatv ktoo'a 6 gadv (gadv)
hang up (trans) -vvs- gatvvga kt66sadf'a5 gadvvsga (gadvvsga)
take down -ee'- aahlee'a atoosii'a (aahlee'a) gadee'a
72
Appendix B
Note the similarity between the stems for 'put on an item' and 'hang up' (appendix A, set
4); and for 'take off an item' and 'take down' (appendix A, set 4). These forms appear to
incorporate the noun signifying the article of clothing into the verbs 'hang up' and 'take down'.
For each article of clothing, the stem for 'wear an item' is related to the remote past stem of 'put on
an item'.
Gloss Item 'put on item' 'wear item' 'change item' 'take off item'
hat a12 sgwe 1nlwo al 1sgwe 3 tv2hy 3 s ga ul 1sgwe3tuh 2 ga al 2 sgwe 3 tuh 2 gi3'a
2 wo
2hn11, alhn:;i.2 wo 3 •a u 1hnu 32 wa
shirt i;,. alhn11,2w11,32,i2yv3•a a 1hni;,.2 we 3 •a
shoes di2la 1su231o dal 1a32su2hly3sga du 11a32su2hla dal la32su2t:;i.3 •i2 yv3•a da 1Ia 32su21e3•a
73
74
Animacy and Agreement in Cherokeei
Michael Dukes
O, Introduction
Agreement in natural language has usually been treated as a purely syntactic phenomenon
in generative grammar. Within R[elational] G[rammar] and Transformational Grammar
(including G[overnment] B[inding] theory), agreement marking has consistently been analysed
as the 'spelling out' of some structural relationship between two elements in a syntactie
representation, either by copying over syntactic features from one category onto another or by
moving an agreeing element from one position to another.2 Despite a large (and growing) bcxiy
of evidence that this view of agreement is insufficiently general to account for many properties
of the phenomenon, the tree-theoretic approach to agreement has continued to predominate.
Recently however, some generativists have acknowledged that agreement is probably not a
purely syntactic matter but requires some direct semantic and pragmatic input. The interaction of
agreement with reference transfer (Nunberg 1977), polite plurals (as in German and French) and
honorific agreement in Japanese and Korean all point to the conclusion that a correct statement
of constraints on agreement processes frequently requires reference to semantic and pragmatic
factors. 3
1I would like to thank our principal Cherokeeconsultant,,Mrs. VirginiaCarey, for sharing her knowledgeof the
Cherokeelanguageand for her patienceand good hwnour in dealingwith our (sometimes)peculiar questions. I
would also like to thank Mr. Levi Carey for his help as a consultant.Thanks are also due to all the other participants
in the 1993Field Methodsclass for discussionof various aspects of Cherokeegrammar,particularlyFilippo
Beghelli and Pam Munro who discussedwith me some of the issuesaddressedin this paper. Pam also provided
extensivecommentson an earlier draft which have helped to gre.atlyimprovethe paper. Finally, I thank Janine
Scancarellifor her commentson an earlier draft and for her generousencouragement I hope the analysis presented
here does justice to the accmmtof Cherokeeagreement in her dissertation.
2An extreme versionof this approachis currentlyin vogue in GB, in which essentially all agreement is claimed to be
the result of movement(either overt or post-syntactic)of agreementtriggers to the 'Spec' positions of various
functionalprojections(as has been suggestedfor Cherokee by FilippoBeghelli).
3 A concise summaryof much of this data can befound in Pollard and Sag (1993: chapter 2).
75
The contents of this paper will be organised as follows. In section 1, I will provide a
general overview of the Cherokee agreement data and the constraints upon it, essentially
reviewing the descriptions found in Pulte and Feeling 1975 and Scancarelli 1987. Section 2
outlines a lexically based account of the agreement data from intransitive and transitive clauses
within a version of HPSG. Section 3 attempts to extend the analysis to the treatment of
ditransitive clauses, which exhibit a grammaticalized restriction on the personhood of 'demoted'
direct objects. Section 4 provides a brief review of some cross-linguistic issues raised by the
Cherokee agreement data and considers the relative plausibility of two competing approaches to
the treatment of animacy restrictions on ditransitive clauses.
When attached to verbs, A and B prefixes encode the person and number of one of the
arguments of the verb (as well as inclusivity/exclusivity in the case of first person arguments).
On intransitive verbs, an A or B prefix agrees with the subject (i.e. the sole argument of the
verb). On transitive verbs, an A prefix agrees with the subject while a B prefix may agree with
the subject or the object depending on certain factors to be outlined below. 5 In certain parts of the
paradigm, transitive verbs also bear portmanteau agreement prefixes which simultaneously
provide information about subject and object. All these cases will be discussed in more detail
below. It is important to note that Cherokee verbs always bear agreement prefixes, even on
verbforms that are referred to in the Cherokee literature as 'infinitival'. Thus all verbs bear either
an A prefix, a B prefix or a portmanteau prefix. In addition to these A/B and portmanteau
prefixes, transitive verbs with a third person plural object obligatorily bear a prefix called the
'distributive' in certain parts of the transitive paradigm. The conditions governing the appearance
of the distributive are partly lexicalized and somewhat complicated but the inflectional properties
of this morpheme are relevant to the treatment of Cherokee grammatical relations to be outlined
in a later section of this paper. A few examples of Cherokee verbforms bearing the A/Bprefixes,
portmanteau and distributive morphology are provided here for illustration.
(1) Aagi-y66siha
1sB-hungry;prs
'I'm hungry'
(2) Hi-w66niha
2sA-speak;prs
'You're speaking'
76
(3) Aagw-v'v'hnflv'v"i
lsB-hit;pst
'I hit it'
(4) Dee-j-v'v'hnflv'v''i
dis-2sB-hit;pst
'You hit them (inn)'
77
In a clause headed by a transitive A verb, the constraints on agreement marking become
somewhat more complicated. As mentioned earlier, two arguments are sometimes marked
simultaneously on the verb by portmanteau morphemes. Thus, in a sentence such as (5) below,
the verb is marked for a ls subject and a 2s object, whereas in (6), the verb is marked for a 2s
subject and a ls object:
(5) Gv'vy-v'v'nfha
ls>2s-hit;prs
'I'm hitting you'
(6) Sgw-v'v'hnfha
2s> ls-hit;prs
'You're hitting me'
Portmanteau fonns like those above are largely restricted to cases in which both arguments of the
verb are non-third person. On verbforms agreeing with at least one third person argument, the
transitive agreement prefixes are mostly identical to one or other of the intransitive A or B
prefixes. The A vs B distinction is not relevant in the cases involving portmanteaus since both A
and B verbs employ the same portmanteau forms. The distinction is thus neutralized. A full
listing of the transitive agreement prefixes is given in Scancarelli (1987: 71). I will not repeat all
the forms here. However, Table 2 below illustrates some generalizations regarding the form of
agreement morphemes that follow from Scancarelli's table, most of which are noted by
Scancarelli herself.
Table 2: Agreement Morphemes For A Stem Verbforms
0 ls ld Ip ld Ip 2s 2d 2p 3s 3ns 3s 3ns
s exc exc inc inc an an inn inn
ls RFL PM PM PM PM PM As DAs
Id RFL PM PM PM As DAs As DAs
exc
lp RFL PM PM PM As DAs As DAs
exc
Id RFL eeAs DAs As DAs
inc
Ip RFL eeAs DAs As DAs
inc
2s PM PM PM RFL PM DAs As DAs
2d PM PM PM RFL eeAs DAs As DAs
2p PM PM PM RFL eeAs DAs As DAs
3s Bo Bo Bo Bo Bo Bo Bo Bo As/ Asf Asf DAs/
Bo Bo Bo Bo
3p gBo gBo gBo gee gee gee gee gee As/ As/ As/ DAs/
Bo Bo Bo Bo Bo Io Ionii Io J.onii
(adapted from Scancarelli 1987: 71)
Table 2 summarizes the type of prefixes that are found on A stem transitive verbf orms. The
subject is given vertically, the object horizontally. 'PM' denotes a field in the table in which a
portmanteau prefix is found which agrees with both subject and object. 'D' denotes an obligatory
occurrence of the distributive prefix. Italicized forms are actual Cherokee morphology found in
those fields of the table. 'As' indicates that an A intransitive prefix form marks agreement with
78
the subject, 'Bo' indicates that a B intransitive prefix form marks agreement with the object. The
'As/Bo' and 'As/ Io' forms in the bottom righthand corner of the table denote cases where the
choice of agreement prefix is determined by the relative animacy of the two third person
arguments. The verb may agree with the subject or the object (but not both). The inverse prefix
gvvwa- (denoted I in table 2) is formally distinct from the B agreement forms but marks
agreement with a third person object when the subject is third person plural and less animate than
the object. Since it fulfills the role of a B prefix in a restricted part of the transitive paradigm I
will treat it here as a suppletive form of the third person singular B prefix that must be
substituted for the usual B form by an adhoc morphological rule.
The table for B stem verbforms would differ from table 2 only in the following respects
- the 'As/Bo' and 'AsfJ.0 ' forms in table 2 would be replaced by B and I forms respectively. The
verb would agree either with the subject or object, depending on animacy. The final two columns
(i.e. the columns denoted by 3sinn and 3nsinn) would also be replaced by B forms.
Several interesting generalizations emerge from table 2 regarding the A/B split. Firstly, in
certain parts of the transitive paradigm, the A/B split is neutralized. Thus, the forms marked in
boldface do not change to B forms even if the verb is put in a B stem form. I will refer to the
boldfaced forms, including the portmanteaus, as the 'non-alternating prefixes'. The existence of
these nonaltemating forms indicates further 1hat transitive agreement must be learnt, at least
partially, as a distinct system from intransitive agreement. The same conclusion is supported by
the fact that the portmanteau morphemes which appear on transitive verbs are not simply
sequences of A+B morphemes (see Scancarelli (1987: 71) for a full summary of the forms)
they are essentially distinct forms. Finally, transitive agreement prefixes display certain
obligatory additional markers indicating plurality (e.g. the g- prefix in the bottom row of the
table and the distributive prefix in the 3nsan and 3nsinn columns) or inclusion of the hearer
within the reference of the subject or object (i.e. the ee- prefix found in most of the forms
involving a second person) which the Cherokee speaker must learn to prefix to an A or B
morpheme specifically in transitive sentences. I will refer to these additional morphemes as 'non-
core' agreement prefixes to distinguish them from the 'core' A/Band non-alternating forms that
seem to mark person, number, inclusivity as well as animacy straightforwardly. 10
It is clear then that the transitive paradigm is considerably more than just a combination
of the A and B intransitive forms. On the other hand, it seems reasonable to assume, following
Scancarelli (1987: chapter 4), that in the cases where transitive agreement prefixes are identical
to intransitive A and B prefixes, the transitive verb actually agrees with only one argument,
either the subject or the object. The alternative is to assume that in these cases there is a 'zero
morpheme' which covertly marks the second transitive argument (i.e. either the subject or the
object). However, I think there are good reasons for rejecting this idea. Firstly, from a
metatheoretical point of view, it is desirable to keep otherwise unmotivated unobservables to a
minimum. The assumption that there are zero agreement morphemes is clearly unmotivated
except to the extent that it 'regularizes' the transitive agreement paradigms. Secondly, from a
Cherokee-internal point of view, we would be forced to say that every third person agreement
morpheme (both A and B forms) has at least two distinct forms, an overt form and a covert form.
But the distribution of the covert form (unlike the overt form, which appears in both transitive
and intransitive clauses) is not predictable from any grammatical properties independent of the
overt form. We would simply assume that it has to appear on every transitive verb that doesn't
bear the corresponding overt form. Since we know that Cherokee verbs do overtly mark both
79
subject and object information simultaneously (as in the portmanteau forms in table 2) there is no
obvious reason why null morphemes should exist in such a large chunk of the transitive
paradigm nor why null morphemes do not occur at all in the intransitive paradigms. The more
natural assumption is that transitive verbs sometimes agree with only one argument and not both.
The motivation for this phenomenon and its treatment in a formal grammar are the main
concerns of this paper.
The generalizations given above can be seen to be conditioned by the Cherokee Animacy
Hierarchy (Cherokee AH), as given below in (7):
(7) CherokeeAnimacy Hierarchy (Scancarelli 1987: 126)
1st person > 3 human > 3non-human > 3 inanimate
2nd person animate ·
Arguments at the top end of the Cherokee AH always trigger core agreement while those in the
middle or at the bottom of the hierarchy usually only trigger core agreement if there are no more
animate arguments in the clause. The examples in (8) illustrate some of the possibilities:
(8) a. Dugi jii-goo'v sv'v'hi
Richard ls>3san-see;pst yesterday
'I saw Richard yesterday'
c. Sginii-goohv'v"'i
2s> ldexc-see;pst
'You saw us'
80
motivating a formal account of the Cherokee AH. Since HPSG structures provide a
representation for all kinds of information relevant to the utterance of a sentence (crucially,
discourse information in this case) it is possible to formulate constraints on Cherokee grammar
that are based on structures independently needed within the theory. The theory thus allows for
the direct influence of nonsyntactic information on grammatical processes such as agreement.
The formal constraints proposed within RG can thus be naturally motivated by basing them on
some externally driven requirement - a possibility which is not available in theories that isolate
morphosyntactic systems from 'nongrammatical' information.
Representations of sentences within HPSG do not differ in kind from the representations
of lexical items. Sentences are structured matrices constructed via monotonic combination of
constituent elements according to various syntactic, phonological, semantic and pragmatic
constraints. Some constraints may be universal, others may be parochial. The overriding
condition on the combination of elements into larger structures is stated in terms of 'unification'
(Pollard and Sag 1987) - any two elements of the appropriate sort may be combined into larger
objects provided that they do not contain conflicting information. Let us consider a couple of
sentences to illustrate how the framework operates:
Oversimplifying a little in the meantime, a verbform such as aandaw66'a would have a lexical
representation like (11):
81
CXT [C-INDS SPEAKER ....]
ADDRESSEE ....]
[BGRD .... [1] ....]
The lexical entry gives the pronunciation of the verb, its morphosyntactic category and its
subcategorization frame (called SUBCA T) which gives an ordered list of elements selected by
the head. 11 CNT specifies that there is one (plural) referent that fulfills the single semantic role
associated with the denotation of the verb aandaw66'a.1 2 Note that the subject need not be
specified morphosyntactically as a plural because no plurality constraint is imposed on the
subcategorized NP in CAT. CNT ensures however that the subject denotes a plural entity (the
distinction will be relevant to the treatment below of wahya in (10)). CXT specifies that the
index of the swimmer(s) must be one which has been introduced in the discourse as part of the
BACKGROUND (BGRD). No constraints are placed on the indexing of the speaker or hearer by
this particular verbform since it is not a first or second person verbform. If relevant, the
referential indices of the speaker and hearer (along with locational and temporal information
about the utterance situation) would be indicated in the CXT submatrix denoted
CONTEXTUAL-INDICES (C-INDS). If the verbform is combined with the NP dinry6otli via
unification to form sentence (9), the NP must bear the index [1] to satisfy the subcategorization
requirements of the verb. Since dinry6otli satisfies the constraints imposed by the verb it may
unify with the verb to give sentence (9). 13
A verbform such as duuhlv 'v "i would have a lexical representation resembling that given
in (12):14 ·
82
indicates that wahya has a plural referent in (10). Since Baab and wahya both satisfy the
constraints imposed by the verb, they may unify with the verb to give the well-formed sentence
(10), where Baab will be associated with the 'agentive' semantic role and wahya with the 'patient'
role via coindexation.
(13) FCR 1:
VFORM [TYPE A]
[STEM +PROO] [AGRSET A]
SUBCAT (XP)
which requires that any progressive A verb bear set A prefixes. Following Gazdar et al 1985, I
will refer to such conditional restrictions on the featural content of categories as Feature Co-
occurrence Restrictions (FCRs). Hence, (13) is labelled 'FCR 1'. Implicit in this analysis is the
assumption that set B prefixes are the default form of agreement. Again following Gazdar et al
1985, I will assume that default feature values are filled in via Feature Specification Defaults
(FSDs). The default AGRSET value is determined via FSD 1:
Thus, unless some constraint like (x) comes into play, an intransitive verb will always be marked
with B prefixes. The verb will mark the (person, number and inclusivity) features corresponding
to the referent of its argument, which will be denoted in the HEAD features by the submatrix
AGR[EEMENT]. The value of AGR for intransitives is defined via a set of lexical constraints
which provide a morphosyntactic 'translation' of the kinds of situations that the verbforms may
be used in. These translations correspond to the Cherokee speaker's knowledge that, for example,
use of a verb marked with first person singular morphology (whether it is an A or B form)
indicates that the utterer of the sentence is the one carrying out the activity described by the
83
sentence. Such constraints can be described as FCRs of the form given in (15)a-c, which
determine the value of AOR for singular arguments in the first, second and third person
respectively and correspond to standard 'subject/ verb agreement': 15
84
First and Second Person Arguments-Absolute Animacy. One of the key
generalizations regarding animacy from section 1.3.3 was that a first or second person argument
always play some role in the determination of the form of agreement on a verb. This property of
Cherokee morphosyntax can be described within HPSG by a condition on the translation
function that defines the value of AOR for any given verbform. As shown earlier, CXT matrices
are broken down into two main submatrices, C-INDS and BORD. Since first and second person
arguments are, by definition, coindexed with SPEAKER and ADDRESSEE, they are always
coindexed with obligatory elements of any discourse, which are represented in the C-INDS file.
Third person arguments by contrast are not coindexed with discourse participants, and their
referents must be introduced as additional assumptions in the BORD file. The relevant condition
on the translation function can therefore be stated as 'Translation Condition 1' in (17):
(17) TC 1:
SUBCAT ( ..•NP[x]•··>
C-INDS ...... [x] :::, [AOR f(F[x], ... )]
(17) requires that the value of AOR for some verbform with an argument whose index is found in
the C-INDS file (i.e. a first or second person index) must be (at least partially) determined by the
location of that index (i.e. the file it appears in). 16 No such condition holds of third person
arguments since the form of agreement on a transitive verb with a third person argument is
frequently determined solely by the index of the other argument (which may be first, second or
third person). The index of a third person argument may be 'ignored' by agreement processes in
the language. Thus the fact that first and second person arguments form a natural class with
respect to agreement in Cherokee can be straightforwardly captured via the HPSO representation
of context.
The condition given in (17) accounts for some of the absolute animacy effects involving
first and second person arguments. However, there are two other problems which must be
accounted for; the relationship between A and B prefixes in transitive clauses and the
representation of non-alternating agreement prefixes (including the portmanteau prefixes). I will
start by considering the second problem.
There are at least two possible approaches to the treatment of the non-alternating prefixes.
One approach would involve treating them as a paradigm distinct from the A and B prefixes.
Under this view, AOR would have a third possible value, which we might call C, that would
neutralize the distinction between A and B and require that AORSET be transitively valued (i.e.
be specified for agreement with two (ordered) arguments). A second approach would involve the
assumption that non-alternating morphemes are also defined in terms of the AGRSET values A
and B but that the shape of the morphemes happens to be the same in parts of the paradigms. In
this case it will be necessary to assume that both A and B AGR prefixes can sometimes be
transitively valued. There is a considerable body of evidence favouring the first approach over
the second. The second approach requires a large coincidental overlap of forms between the two
paradigms that makes the A/B distinction meaningless in such cases. It also requires that A and B
prefixes sometimes be intransitively valued and sometimes transitively valued. On the first
approach however, A and B forms will always be intransitively valued for AOR even when they
are marked on transitive verbs. 17 Furthermore, the lack of alternation of the boldfaced forms in
table 2 between A and B agreement is immediately accounted for by the fact that they comprise a
16nie expression '[AGR f(F[x], ...)]' should be read: The value of AGR is a function of the file in which [x] is
located and possibly other files'.
17Conditionsthat fix this requirement will be stated below.
85
distinct inflectional set. The only disadvantage of employing a distinct C inflectional set is that
there is some overlap in the shape of the A and C forms when objects are animate third person.
However, since these A-like forms do not alternate with B forms they clearly cannot be
considered typical A forms and the overlap must be considered accidental.
Verbs specified with the C value for AGRSET must have their AGR value determined by
information about the location of both argument indices. This datum can be represented as a
condition on the translation function for transitive verbs analogous to the condition given earlier
in ( 17) that applied to both transitive and intransitive verbs:
(19) TC 2:
[AGRSET C] [AGR f(F[x], F[y])]
SUBCAT (NP[x], NP[y])
86
Thus for any transitive verbform whose agreement marker is selected from the C agreement
paradigm the form of the agreement marker will be determined as a function of the location of
the indices of both the subject and the object.
The above analysis accounts for the nonaltemating forms within the Cherokee verb
agreement paradigm by treating them as a distinct inflectional subparadigm labelled C. But in the
remaining parts of the paradigm given in table 2 there are transitive agreement markers that are
clearly members of either the A or B paradigm. When one argument of the verb is first or second
person and the other is an inanimate third person, the verb always agrees with the first or second
person NP. Furthermore, the choice of A or B agreement form for progressive A verbs is
determined according to whether the triggering NP is subject (A agreement) or object (B
agreement). Assuming again that Bis the default value for AGRSET, these agreement facts can
be derived by imposing the FCR given in (20):
(20) FCR4:
HEAD VFORM TYPE A
SIBM +PROO
SUBCAT (NP[x], NP[y]) [AGRSET A]
(21) TC 3:
[AGRSET -,C] [AGR f(F[x])]
Finally, it is necessary to provide a condition to ensure that a transitive progressive A verb agrees
with its object when it is marked with B agreement:
(22) TC 4:
[TYPE A]
[SIBM +PROO] [AGR f(F[y])]
[AGRSET B]
SUBCAT (NP[x], NP[y])
The conditions outlined above are sufficient to account for all the animacy effects
involving first and second person arguments. It just remains to account for 'relative animacy'
effects with two third person arguments before turning to the relationship between the Cherokee
AH and ditransitive verbs.
Third Person Arguments-Relative Animacy. When a transitive verb has two third
person arguments the choice of A or B agreement (as well as the choice of agreement trigger)
87
hinges on the relative animacy of the two arguments. A verb agrees with a human argument iri
preference to a nonhuman one and with an animate argument in preference to an inanimate one.
An adequate account of this relative animacy can be given by making use of the idea introduced
earlier that BGRD is broken into three subfiles, HUM, ANIM, and INAN. Unless further
technical machinery is introduced however, it will be necessary to make use of disjunctive
statements to describe these relative animacy conditions:
(23) TC 5:
VFORM HEAD SUBCAT (NP, NP)
CNT ...[x]
... [y]
BGRD HUM ...[x]
ANIM ...[y] v INAN ... [y] [AGR f(F[x])]
V
BGRD ANIM ...[x]
INAN ...[y]
TC 5 requires that any transitive verb must have its AGR value determined by the more animate
argument whether it be subject or object. It is also necessary to ensure that a transitive
progressive A verb agrees with its subject when it is marked with A agreement:
(24) TC 6:
HEAD VFORM [TYPE A]
[STEM +PROO]
[AGRSET A] ::, [AGR f(F[x])]
SUBCAT (NP[x], NP[y])
The inverse condition matching TC 6 (i.e. that lB agreement is object agreement on transitive
progressive A verbs) is already covered by TC 4 and applies to verbs with two third person
arguments in the same manner as to verbs with a first or second person argument. Thus cases of
'inverse' agreement l(Scancarelli 1987: chapter 2) with third person objects are treated by exactly
the same machinery as that which accounts for object agreement with first and second person
arguments. The default AGRSET value for a transitive verb with two third person arguments will
be set as B by FSD ll when TC 6 fails to apply.
In cases where two third person arguments do not outrank each other in animacy,
additional discourse factors decide which argument will trigger agreement on the verb
(Scancarelli 1987:131). Such factors could also be formalized within the the kind of framework
employed here though there is presumably variation from speaker to speaker as to precisely how
these factors might be weighted, as there is in the determination of relative animacy (Scancarelli
1987: 128-33). I will not be addressing these issue:s in this paper.
88
Ditransitive verbs in Cherokee exhibit a number of interesting properties with respect to
agreement. In many languages, including English, such verbs often display two alternative
syntactic patterns; one in which the a goal or benefactive argument is presented as a prepositional
phrase and another in which the goaVbenefactive (henceforth 'dative') argument is presented as a
'second object' of the verb. Typically, in this second case, the dative appears to assume the
syntactic properties of the 'original' direct object. The rule by which this state of affairs comes
about has been described as 'Dative Shift' in the transformational literature (e.g. Oehrle 1976)
and as '3 -> 2 Advancement' (32A) in the RG literature (Blake 1990). Under these kinds of rules,
the original direct object is demoted to some kind of marginal status where it no longer has the
syntactic properties of a direct object. In RG terms it becomes a 'chomeur'. 19 In Cherokee, there
is no alternation between two ditransitive structures. Ditransitive verbs obligatorily treat the
dative argument as the object for core agreement purposes and cannot agree with the theme
argument:
Within RG, this state of affairs corresponds to the obligatory application of 32A in clauses
headed by ditransitive verbs (typically marked in Cherokee by the presence of the dative suffix
-eelon the verb (boldfaced in the examples above)). Such obligatory 32A has been motivated in
a number of other languages. 20 Ditransitive benefactive verbforms are also (usually) marked with
the same suffix:
If the analysis of core agreement on transitive verbs developed above is to be extended to deal
with ditransitive verbs it will be necessary to simulate the RG analysis of obligatory 32A in order
to explain the fact that such verbs agree with the dative argument rather than with the theme. The
191tshould be noted however that analyses of 32A in RG have also been associated with a concomitant application
of '2 -> 3 Retreat' (Blake 1990). In such cases, hypothesised for Kinyarwandafor example, the direct object does not
become a chomeur but maintains a grammaticalrelation to the predicate (i.e. 3).
20BJake(1990: 6) cites Manam, Blackfoot, Mohawk, Tzotzil and Huichol as examples of languages with obligatory
32A. Rude 1982 claims that the same is true in Nez Perce.
89
alternative would be to develop a completely new set of constraints to account for ditransitive
agreement which would nevertheless be identical to the transitive ones developed already except
insofar as referencing the goal argument in place of the theme. Such a redundant analysis seems
highly undesirable. On the other hand, there are reasons for thinking that the theme direct object
may not be a chomeur. Firstly, there is the relationship between the theme argument and the
distributive morpheme. As we saw in table 2, plural direct objects obligatorily trigger the
appearance of the distributive prefix in certain parts of the transitive paradigm. In ditransitive
clauses (as Scancarelli (1987: 223) notes), the theme still triggers the appearance of the
distributive prefix even though it cannot trigger the appearance of core agreement prefixes (see
also (28) above):
(31) Eedi goohweeli dee-ga-nv;v;ne Baab
Ed paper (book) dis-3sA-give;F;dat;prs Bob
'Edward is giving the books to Bob'
However, plural dative arguments also trigger the appearance of the distributive prefix, as one
would expect if they are indeed objects:
(33) Eedi giihli dee-gaa-kaanee'a cfini-y6otli
Ed dog dis-3sA-give;N;dat;pres p-child ·
'Edward is giving the dog to the children'
Thus, with respect to the marking of distributiv1emorphology, Cherokee verbs appear to have
two simultaneous objects. The treatment of the distributive morpheme is in general somewhat
mysterious in Cherokee since it appears to have both inflectional and derivational uses
(Scancarelli 1987: 216). But with respect to the obligatory inflectional uses of the prefix in the
verbal paradigm it seems that both the theme and the goal must be treated a.sobjects.
The second piece of data suggesting that themes in ditransitives ar1etrue arguments is an
animacy-based restriction on the types of arguments that may be themes in ditransitive clauses.
Cherokee does not allow first or second person themes when the verb also has a dative or
benefactive argument. The examples given above in (25) - (33) with third person themes are all
grammatical, but corresponding examples with first or second person themes are not. Attempts to
elicit such examples led my consultants to restructure the subject maner of the target sentence in
such a way as to avoid the use of a ditransitive with a first or second person theme. (34 )a is a
grammatical sentence with a third person theme but when the theme is switched to first or second
person, my consultants omit the benefactive argument and the dative suffix, as in (34)b:
(34)a. Baab wi-gv'v'y-anv;v;'ee'i
Bob way-ls>2s-call;N;dat;prs
'I'm calling Bob for you'
b. Baab wi-cha-yanll'a
Bob away-2sB-call;prs
'Bob is calling you' (for me) (no benefactive encoded)
In such cases. the intended first person benefac1tive argument apparently must be understood
inferentially. (35) iUus1tratesa parallel case with a goal argument in which a free first person
pronoun was suggested as a possible theme argument. Note that the well-fmmedness of examples
like (35)a shows that the restriction is not simply one that requires themes to be inanimate or
90
nonhuman:
If the theme were not selected as an argument of the verb (i.e. if it were not an element of the
verb's SUBCAT list), it would be difficult to relate this animacy restriction to the previously
stated Cherokee AH conditions, because these conditions were almost all stated in terms of
relationships between files of indices and the SUB CAT list.
One other piece of data provides further evidence for the view that theme arguments in
ditransitive clauses are in fact selected by the verb in some manner. Scancarelli (1987: 123)
briefly discusses a process of quantifier float (QF) in which the plural universal quantifier
nigdada may be floated away from the noun phrase with which it is construed to attach to the left
of the verb. (When unfloated, nigaada precedes the nominal it modifies.) Scancarelli suggests
that the process may be restricted to absolutives (i.e. transitive objects and intransitive subjects)
but points out that the data is inconclusive. In my own attempts at elicitation, I have found that
Mrs Carey accepts QF from transitive subjects and objects:
These data suggest that goal and theme objects may both be targetted by the rule of QF, along
with subjects. But if theme objects in ditransitive clauses are 'demoted' from their usual
grammatical function it is difficult to account for the fact that they behave like subjects and
objects for the purposes of QF. It should be borne in mind however that examples involving
91
floated nigaada were never freely volunteered (with the exception of (38)b) but had to be
checked as possible alternatives to volunteered examples in which nigaada was not floated. The
data clearly require further investigation.21
I propose to treat the ditransitive data in the following manner. I will first provide a
lexical rule that relates a transitive lexical entry to a ditransitive one and promotes the dative
argument to object for the purposes of agreement. Secondly, I will discuss the status of the
animacy restriction on theme arguments of ditransitive verbs and finally I will discuss the
grammatical status of the theme argument demoted by the lexical rule with respect to distributive
marking and QF.
Let us assume that the dative suffix -eelderives a lexical category in which a dative
argument is represented as direct object. Since there is no 'underlying level' at which the dative is
something other than an object, the analysis does not involve an advancement of the dative,
unlike the RG 32A proposal. Nevertheless, the analysis shares with 32A the assumption that the
dative is an object at surface. I will assume for the time being that the former direct object is
demoted to 3 (i.e. third element on the SUBCAT list) and is thus still an argument of the verb.
The lexical rule therefore adds an extra element to the SUBCAT list of a transitive verb that it
attaches to and adds an extra semantic role to COl~NT:
92
considerations, the lexical rule given in (39) can be motivated in similar terms. Goal and
benefactive arguments almost always denote animate and typically human referents; indeed, the
use of most ditransitives verbs entails or at least strongly implies that the dative argument is
sentient. By contrast, theme arguments of ditransitive verbs usually denote inanimate objects and
even when they do not, their referent displays none of the prototypical properties of a sentient
individual. A rule like (39) can thus be seen as a grammaticalization of the prototypically greater
animacy of the goal argument with respect to the theme. Crucially, this kind of conventionalized
animacy restriction must be distinguished from the Cherokee AH discussed earlier because the
dative lexical rule is not sensitive to the relative animacy of the arguments in a particular
sentence. It simply applies obligatorily whether or not the AH would determine the goal to be
more animate than the theme.
The constraint which blocks first and second person theme arguments with ditransitives
can be stipulated straightforwardly as the following FCR (assuming that the demoted theme is
treated as a 3):
(40) FCRS: SUBCAT (NP, NP, NP[2J) :.:> [BGRD .... [2] ..]
FCR 5 requires that any 3 must have a referent drawn from the set of background indices, thus
ruling out the possibility that the theme would be first or second person. However, it is arguable
that the effects of FCR 5 are independently derived as a consequence of the interaction of two
constraints introduced earlier and need not be stated as an additional constraint. The condition
described by this constraint can be seen as the result of competition between the theme and the
dative for the grammatical role of object arising from the conflicting requirements imposed by
the Cherokee AH and the dative lexical rule. Consider what happens if a Cherokee speaker wants
to use a ditransitive verb with a first or second person theme. By the stipulated lexical rule in
(39), the verb must agree with the dative if it outranks the subject on the Cherokee AH or if it
otherwise qualifies as an agreement trigger according to the translation conditions given earlier
that apply to the objects of transitive verbs. But TC 1, the translation condition given in (17), will
demand that the verb agrees with the theme if it is a first or second person argument. Therefore
the result in such cases is a conflict between (17) and (39) because only one argument can act as
the trigger for core object agreement. The grammaticalized animacy requirements imposed by
the dative lexical rule override the pragmatically determined requirements of the Cherokee AH,
blocking the possibility of agreement between the theme and the verb. The speaker must
restructure the information conveyed by such a sentence in order to resolve the conflict between
the theme and the dative. There are at least two options available in such cases. As we saw in
(34)b, the speaker may simply omit the dative argument and rely on the inferential abilities of the
hearer in comprehending the intended proposition. A further example of this strategy is given in
(41):
Alternatively, with certain predicates like 'sell', which lacks a benefactive form in Cherokee
(Scancarelli 1987: 105), the speaker may break down a proposition expressed by one ditransitive
clause in English into two or more related transitive clauses, as in (42):
Finally, the speaker may employ some construction or other with a meaning closely related to the
93
English dative equivalent but in which there is no dative suffix and no true dative argument:
94
Since 32A is obligatory and since demoted themes are apparently chomeurs in Tzotzil (Aissen
1983), (46) has the effect of blocking ditransitive clauses with first or second person themes.
Allen et al (1990: 330) state the condition in (47) as a constraint on clause structure in Southern
Tiwa (Tanoan):
(47) The Participant Chomeur Ban: A relational network in which a first or second person
nominal heads a chomeur arc is ill-formed.
Ditransitive clauses in Southern Tiwa that are derived by 32A are ruled out by (47) if they have a
first or second person theme because themes are demoted to chomeur in such cases. 23
There are two fundamental problems with the chomeur constraints stated in (46) and (47).
Although they suffice to describe the fact that chomeurs in Tzotzil and Southern Tiwa may not
be first or second person they fail to provide any motivation for the contrasting behaviour of
those arguments as compared to third person ones. The grammatical framework provides no
mechanism by which first and second person arguments form a natural class to the exclusion of
third person arguments. The second problem with these constraints is that they stipulate
conditions which may not need to be stipulated at all provided that discourse information can be
mentioned in grammatical representations. The claim was put forward in section 3 that the
nonexistence of first and second person ditransitive themes in Cherokee follows from the
interaction of the Cherokee AH and the dative lexical rule, which corresponds to 32A within the
RG framework. This analysis makes the prediction that any language with an animacy hierarchy
similar to the one in Cherokee will lack first and second person ditransitive themes in structures
derived by 32A. Although any conslusions are highly tentative at this stage, this prediction is
largely borne out to the extent that I have been able to investigate it. Aissen ( 1987: 40) notes that
first and second person arguments in Tzotzil "always determine number agreement somewhere
[Aissen's emphasis; M.D.] in the sentence, while third person nominals do not". This datum is
extremely reminiscent of the animacy based conditions on agreement in Cherokee noted earlier
in this paper (i.e. TC 1). Taken together with the obligatory nature of 32A in Tzotzil, the
restriction on dittansitive themes in this language appears to follow from essentially the same
constraints as in Cherokee. Allen et al (1990) provide numerous examples and constraints
indicating the pervasive influence of an animacy hierarchy on the determination of grammatical
relations in Southern Tiwa. 24 In cases where 32A applies, the unacceptability of first and second
person ditransitive themes will follow for the same reasons noted for Cherokee and Tzotzil. 25
231 do not address the issue here of whetherconstraintsof this sort, includingthe one given for Cherokeein (40),
shouldbe stated in tenns of chomeurhoodor demotionor some other grammaticalconceptwithin the HPSG
framework.The analysisof ditransitivethemesin Cherokeeas 3s could perhaps be replaced by a multistratal
analysisin which they are treated as chomeurs,but developingsuch an analysis is not practical here for reasons of
space.
24see Rosen 1990for an accountof SouthernTiwa agreementbasedon a mixed grammatical/ pragmatichierarchy
that incorporatesa personhoodhierarchyof the kind proposed for Cherokee.
25Notethat when 32A (which is not obligatory)does not apply in SouthernTiwa, first and secondperson themesare
acceptable(Allen et al (1990: 333)).
95
has, in RG terms, obligatory 32A but I have been unable to find any reference to an animacy
hierarchy in my Huichol sources (Comrie 1982, Grimes 1964).26
It seems clear that a motivated account of the relationship between agreement and
grammatical relations in Cherokee (as well as other languages) requires reference to contextual
information that is not usually made available in generative frameworks. The animacy
hierarchies which partially determine this relationship directly model the relative prominence of
information in discourse situations. I hope to have shown that an HPSG account of such data
goes some way towards accounting for these correlations between grammar and discourse.
26Pam Munro (p.c.) points out that there may be other lang1J1ages
whichban first and second personditransitive
themes even though they lack evidence of animacy hierarchies.If it turns out that such languagesreally do lack
animacy hierarchiesit may be necessaryto employ stipulationsof the kind discussedhere after all. Perhaps such
cases can be motivatedin diachronic tenns if it can be shown that related languagesstill have 'active'animacy
hierarchies.
96
Cherokee Possession and the Status of -jeelP
Robert S. Williams
1. Introduction
This paper examines the system of possession marking in Cherokee, an Iroquoian
language spoken mainly in Oklahoma and North Carolina, and focuses in particular on certain
syntactic properties of -jeeli, the Cherokee lexical possessive marker. Possession is expressed in
Cherokee using both lexical items and morphosyntactic devices.
Cherokee, as is the case with many other American Indian languages, often indicates
possession by prefixing a possessive marker to the possessed noun. The Cherokee system uses
two sets of markers to indicate person and number agreement with the possessor which are
allomorphs of pronominal agreement markers. These are identical to the two sets of pronominal
agreement markers, discussed in detail elsewhere in this volume, which are prefixed to
intransitive verb stems and are referred to here and elsewhere in the literature as A and B
markers. 2 Examples of A and B possession constructions are as follows:
(1) jii-nhgo
lsA-tongue
'my tongue'
(2) ag-sgwoohli
lsB-stomach
'my stomach'
Where possession is not marked by affixation, it is indicated by a separate word, -jeeli, which is
always prefixed with one of a set of agreement markers similar, but not identical to, B markers.
I refer to such markersas J markers since they only occur when prefixed to -jeeli.
(5) J + jeeli
ls agwa-jeeli 2s ja-jeeli
ldin giin-jeeli 2dl sda-jeeli
ldex oogiin-jeeli 2pl iijii-jeeli
lpin iigii-jeeli 3s uu-jeeli
lpex oogii-jeeli 3ns uunii-jeeli
1I am very much indebted to Mrs. VirginiaCarey, who is my Cherokeeteacher and who providedall of the original
data used in this paper. Thanks are also due to ProfessorPamela Munroand all of my fellow students in the 1994
UCLA Field Methodsclass, as well as to CynthiaWalker and Karn King for their thoughtfulcomments.
2For a more detailedaccount of the pronominalagreementsystem,see Dukes (this volume).
97
J markers are identical to B markers in all but three forms. Both first person dual fonns lack the
final /-ii/ found in theiir B marker counterparts, while the second person dual form ends in /-a/
rather than the /-ii/ of its B counterpart.
With the exception of kinship terms, the specification of possession marker type for
inalienable nouns is specified for either A or B markers, lexically, rather than semantically. This
can be seen in the arbitrariness of possession marking for the following nouns.
(9) a-hnawo
3sA-shirt
'shirt'
98
(10) uu-hnawo
3sB-shirt
'his/her shirt'
According to Holmes and Smith (1977), articles of clothing which are being worn are treated as
inalienables and thus are marked for possession with either A or B prefixes. When they are not
being worn, they take J+jeeli.
(11) agwa-hnawo
3sB-shirt
'my shirt (the one I'm wearing)'
(12) agwa-jeeli uu-hnawo
lsB-jeeli 3sB-shirt
'my shirt (the one I'm not wearing)'
Holmes and Smith also state that body parts, which are normally considered to be
inalienable, take +jeeli when they are detached from the body, though the noun must still be
marked with a possessive marker.
(13) ja-jeeli ga-nvvsgeeni
2sB-jeeli 3sA-leg
'your (severed) leg'
For my consultant, Mrs. Virginia Carey, however, marking body parts with J+jeeli, be they
attached or not, is unacceptable.
Kinship terms are the only class of inalienables which appear to be semantically specified
for possessive marker type, always taking B markers. They can never occur without possessive
markers.
(14) oogii-do
1pexB-sister
'our sister'
(15) j-uu-duuji
pl-3sB-uncle
'his uncles'
3. Possessor Raising
Cherokee appears to have possessor raising, a syntactic process which makes the
possessor, rather than the possessed NP, a central argument of the clause. In the case of
Cherokee, possessor raising is indicated by verbal agreement with the possessor instead of the
possessed NP. In examples (16) and (17) the possessoris raised; in (18) and (19) it is not.
(16) Hi-nvvsgeen ja-sdaane-'a.
2sA-leg 2sB-hurt-prs
'Your leg is hurting'
(17) Jii-nhgo ji-leeyvvh-i.
lsA-tongue lsA-bum-prf
'I burned my tongue'
99
(18) Vvge.e-do uu-dlaagaa'i.
lsB-sister 3sB-sick
'My sister is sick'
Note that the verbal agreement marking must agree only in person and number with the
possessor and not in agreement marker class, as in example (16). As stated previously, the
assignment either A or B possession marker class is arbitrary for all inalienables except kinship
terms and the assignment of either A or B verbal agreement depends upon factors independent of
possession markers. 3
b. *agwa-laasihdeen a-joocle
lsB-foot 3sB-itch
'My foot itches'
Example (20b) shows the nonoptionality of possessor raising. In all other cases with body parts
as subjects, including verbs of temperature, possessor raising is prohibited.
b *agwa-laasihdeen agwa-dihleega
lsB-foot lsB-be hot
'My foot is hot'
b. *agwa-laasihdeen agwa-hyvvdla
lsB-foot lsB-be cold
'My foot is cold'
b. *ag-sdiihgv agwa-dlisgwaaneeda
lsB-hair lsB-be curly
'lVIyhair is curly'
100
(24) a. Jii-nvvsgeen ga-liijoohiida.
lsA-leg 3sA-be fat
b. *jii-nvvsgeen jii-liijoohiida
lsA-leg lsA-be fat
'My leg is fat'
In one instance Mrs. Carey accepted both raised and non-raised versions of a possessed
sentence but noted aspectual difference between the two. A raised possessor with the verb
wotiisa'to swell' (25), denotes progressive aspect for Mrs. Carey and an unraised possessor (26)
perfective aspect.
No other such instnaces of possessor raising carrying aspectual or any other grammatical
information were found in the data.
Pulte and Feeling (1975: 352) state that -jeeli can either precede or follow the possessed
101
noun. Mrs. Carey agrees with this but usually places -jeeli after the possessed noun when asked
to translate a possessive structure into Cherokee. In instances such as (29) where J+jeeli is
predicative, and for intransitive and transitive sentences with only one overt NP, such as (30),
(31 ), and (32), Mrs. Carey accepts all possible word orders, even those involving split
constituents.
102
(32) Giihli ja-jeeli uu-sgaala
dog 2sB-pss 3s> 3san-bite:prf
'Your dog bit him'
However, when a sentence with two overt arguments contains a possessed NP as either a
subject or object, discontinuous constituents are not allowed. This is because word orders
resulting in an NP intervening between -jeeli and the possessed noun appear to interrupt the
anaphoric relationship between the two.
Partly because of Cherokee's relatively free word order within the possessed NP, the
grammatical status of -jeeli is problematic. Scancarelli (1987: 289) glosses -jeeli as a 'possessive'
but does not speculate as to its grammatical status. Pulte and Feeling (1975: 328-29) cite -jeeli'i
as the stem of this word and state that it is a possessive pronoun which also functions as an
adjective, as in their examples (43) and (45), 4 reproduced here as (34) and (35) .
That J+jeeli functions like a possessive pronoun is clear, though there is no other empty
pronominal form which inflects for person and number. The fact that -jeeli inflects for person
and number does make it similar to an adjective. It is also a fact that J +jeeli can in certain
circumstances function as a predicate, but that it might actually be a verb of some kind is
problematic, since it is unable, like other verbs, to inflect for aspect and mode. For example,
J+jeeli can only be used as a predicate in the present aspect, as in the following example:
(36) Agwa-jeeli.
lsB-pss
'It is mine'
In expressing other temporal aspects, such as perfective or future, a verb must also be used,
indicating that perhaps -jeeli is a noun or an adjective.
103
(37) Uulsdana agwa-jeeli.
be:prf lsB-pss
'It was mine'
(38) Agwa-jeeli geesesti.
lsR-pss be:fut
'It will be mine'
6. Conclusion
Cherokee also has possessor raising, which occurs only with body parts and is specified
by the class of the verb. Possessor raising is obligatory for a subclass of verbs of physical
sensation and prohibited for all other verbs. While possessor raising confers argument status on
the possessor NP, it is not clear whether subjecthood is also conferred.
Word order within possessive NPs is constrained only in that quantifiers must precede all
other NP constituents, a constraint which also holds for non-possessive NPs. However,
discontinuous possessive NPs are allowed only in single argument clauses. Though both SOV
and OSV word orders aireallowed, Mrs. Carey did not readily accept postposing of the subject in
two argument clauses, either in the immediate post-verbal or end positions.
104
Cherokee Clause Structure
Filippo Beghelli
1. Introduction
Cherokee syntax is largely unexplored tenitory. Cook (1979}-which focuses of the North
Carolina dialect-and Pulte and Feeling's (1975) brief grammar contain articulated descriptive
sketches. More recently, Scancarelli (1987) offers in-depth analyses of a number of phenomena.
especially inverse constructions and pronominal agreement. Yet some of the basic issues in clause
structure remain to this day unaddressed in the literature.
This paper is an attempt at addressing some of these issues. 1 In particular, the status and
position of argument NPs will be discussed. Of special relevance in this respect is the work of
Mark Baker on a related language, Mohawk (cf. Baker 1992, 1993). The conclusion that will be
reached is that Baker's account of Mohawk non-configurationality seems largely applicable to
Cherokee.
Given that our current knowledge of Cherokee syntax is generally quite::poor, and-
especially-given the very limited data at my disposal (all the data come essentially from one native
speaker),2 the claims in this paper should be taken as tentative, pending verification with further
data, and rather than conclusive, as suggestive of relevant areas for future research.
The issue, then, is how the syntactic structure of Cherokee allows for such freedom. This
is all the more significant considering that the usual constituency tests which show that in
languages like English the subject asymmetrically c-commands the object, do not seem to yield
comparable results in Cherokee. A syntactic definition of 'subject' seems elusive. Scancarelli
admits to not knowing of any subject properties in Cherokee.
Let's consider this point in greater detail. There is evidence that in Cherokee subjects and
1 The original version of this paper contained a section on Cherokeepronominal agreement This:part has been
omitted in the present version. I wish to thank Pamela Munro for helpful discussion of a number of issues of
relevance to this paper. All errors areof course my own.
2 My Cherokee consultant for this paper has been Mrs. VirginiaCarey, a native Cherokee speaker from
Tahlequah, Oklahoma.I am very grateful to Mrs. Carey for her patience and for sharing with me heirinsights on her
lan~uage.
out by
In fact, a number of word order restrictions exist in Cherokee within phrasal constituents, as JX>inted
Scancarelliherself. Observationslike the following suggest a head-finalpropensity:
1. Only postpositions (and not prepositions) exist;
2. Adverbialmodifiersmust precede adjectives they apply to;
3. Inflection (lnfl) is suffixal: tense and aspect, as well as causatives,are suffixes;
4. It is far more common for adjectives to precede, rather than follow, nominals; it is also more ,commonfor
genitives to precede nominals.
105
objects c-command each other. This evidence is provided by Scancarelli's observation that WEAK
CROSS-OVER (WCO) effects are absent in this language.
As first observed by Reinhart (1976), for a pronominal to acquire a bound reading, it must
be c-commanded by the thematic position of its binder (a quantified NP or a Wh-operator). Yet in
Cherokee a bound reading is possible even when the thematic position of the binder is the object
position, and the pronominal is in subject position. Consider (1) below: speakers can interpret the
pronominal as being part of the subject NP, and the thematic position of the Wh-element to be the
object. Alternatively, they can interpret the pronominal to be inside the object, and the Wh-element
to fill the role of subject.
In English, on the contrary, a bound readiing is possible only with Who loves his mother.
A sentence like Who does his mother love does not have a bound variable reading (unless focus is
applied to his mother), and the pronominal must be interpreted as referring to a unique third person
previously introduced in the context. Example (1) indicates, then, that subject and object position
mutually c-command each other in Cherokee.
The next examples (2-3) show similar effects with Quantifier Phrases (the examples
illustrate both verbs marked with set A prefixes and verbs marked with set B prefixes--cf. Part
These data offer a less icompelling argument, however, since it is not clear whether we have true
m.
bound variable readings. Bound pronouns are typically singular; and Cherokee does not seem to
have a universal quantifier comparable to English every, which binds a singular pronoun. The
universal quantifier in Cherokee rather resembles English all, a determiner that supports collective
as well as distributive readings, and which shows less pronounced WCO effects (if at all).
The lack of WCO effects in sentences like (1) offers the strongest argument for the claim
that subjects and objects symmetrically c-command each other in Cherokee. Another reliable test
can be provided by negative polarity licensing: the licensor of a negative polarity item must
c-command the polarity item. Unfonunately, this test does not seem to be available, as I haven't
been able to find negative polarity NPs in this language. 4
4
Another test of c-command may be provided by quantifier scope construals. This test is, however, less
significant owing to the subtlety of judgments involved. Cross-linguistically, it is generally the case that if a
quantifier phrase Q1 c-commands a quantifier phrase Q2, then Q t.alcesscope over Q2 (subject QPs generally can be
construed as wide scope oveirobject QPs). The inverse configuration, when Q2 t.alcesscope over Q even though Q
1 1
a-symmetrically c-commands Q2 • is not always available cross-linguistically. When inverse scope is possible,.
however, as it is in Gennanic and Romance, it is only selectively available with certain quantifier types {cf.Liu
1990, Beghelli 1993, Ben-Shalom 1993). Indefinites like 'some' in object position may t.alcescope over a
c-commanding quantified subject, but other indefinites, like 'no', 'few', etc. cannot. The examples below illustrate:
106
Scancarelli's conclusion is that Cherokee has a 'flat' clausal structure. For her (1987:122),
the clausal structure of Cherokee would look like (4b), instead of (4a), the commonly assumed
clausal structure of English:
(4) a. b.
IP
~
NP1 VP
~
V NP2
As suggested by Jelinek (1984) and Baker (1992), this conclusion is not inescapable. Lack
of asymmetrical c-command between subjects and objects does not in principl,e exclude the
possibility that there may be a VP constituent.
3. VP Constituency Tests
There is in fact no clear evidence for the lack of a VP constituent. Evidence: against a VP
node appears inconclusive. VP pronominalization and deletion in Cherokee are at present too
poorly understood to provide data crucial to the issue. The same can be said about position of
adverbs: Scancarelli's observation that adverbs may be placed either between subject and verb or
between object and verb (ibid.:121-2) does not offer a strong argument for the lack of a VP
constituent.
Arguments for the opposite claim, that VlPexists, seem more cogent. Noun incorporation is
a well-known characteristic of Iroquoian languages, and is present in Cherokee as well, though to
a much more limited extent than for example in Mohawk. As related by Scancarelli (1987:38) it
involves only body parts and items of personal clothing. The existence of object incorporation, but
significantly, not of subject incorporation, suggests that objects have structurally a closer
relationship with verbs than subjects do, consistently with the hypothesis that thc:re exist a VP
constituent including the verb and its complement(s). An example of object incorjporation is the
following (curiously complex) verb form, provided by Scancarelli (1987:36):
107
(5) yi-w-agw-adaa-sk-gwaloo-sd-a'n-ido-'li
i1T-transl-lsB-rfl-head-bump-cs-ambul-prf
'if I go about bumping my head at a distant place'
(6) a. h-vvhniha
2sA-hit:prs
'you are hitting it'
b. h-aadaad-vvhniha
2sA-rfl-hit:prs
'you are hitting yourself'
b. aadaa-hiih-i
aca-kill-nom
'poison' or 'killer'
Since subjects and objects can be freely ordered with respect to each other, the lack of
subject incorporation would be quite surprising if Cherokee's clause structure were indeed a 'flat'
one, as in (4b ). If data from incorporation are consistent with the hypothesis of a VP constituent,
these don't provide a.strong argument, as it could be claimed that incorporation facts have really a
morphosyntactic and not strictly syntactic nature. In the next sections, facts of the kind discussed
in Baker (1992) will be considered. These data provide stronger syntactic evidence for a VP
constituent, and suggest the hypothesis that Cherokee is, at an abstract level, closer to the basic
clause structure of English than suggested by the flat structure hypothesis.
Before reviewing other evidence for VP, we need to consider the obvious question: If a VP
constituent does exist in Cherokee, how can we have symmetrical c-command between overt
subject and object NPs? The apparent paradox is solved by assuming that these overt NPs are not
the actual arguments of the verb, but occur in adjunct positions. This is a solution along the lines of
Jelinek (1994), who extends this treatment to 'pronominal argument' languages. As we shall see
later, this profile seems to fit Cherokee quite well. Baker (1992), who follows the spirit of
Jelinek's proposal, argues that in Mohawk the basic clause structure has the same hierarchical
organization as in English, except that the actual arguments of the V are pronominals, with which
the overt NPs (in adjunc1tpositions) are co-indexed.
The underlying structure of a simple transitive sentence like 'John loves Mary' in Cherokee
would, then, be paraphrasable as 'John, Mary, he loves her'. The tree diagram below illustrates
this:
108
(8) IP
~
NP1 IP
~
NP2 IP
~
pro1 VP
~
V pr02
I
loves
In this and the next section, I will argue that this proposal is consistent with Cherokee data.
I will assume, accordingly, that in Cherokee the following condition holds, as proposed in Baker
1992 for Mohawk;Baker assumes that it holds at S-Structure:
This clause structure account is consistent with the data reviewed so far: in fact it suggests
some explanation for the facts that we have observed, as well for other features of Cherokee
grammar:5
the verb of the second conjunct can be understood as having either John or Mary as its subject.
This contrasts with pro-drop languages like Italian, where sentences similar to the one above are
obtained by VP conjunction; this forces the subject of the first verb to be the same as that of the
second:
If the arguments of the verb are pronominals, the Cherokee sentence in (10) above is not an
example of VP conjunction, but of clausal conjunction, and the subject ambiguity derives simply
from the fact that gender is not marked on the verb (uuhloohyilvv'i).
109
5. Wh-Movement and the Status of ClausalComplements
Baker's proposal is that overt NPs that are thematically arguments are, syntactically, in
A'-positions. This is a strong claim. Fortunately, it is possible to check this prediction, since
Cherokee appears to have Wh-movement.
Evidence for Wh-movement in Cherokee comes first from examples like the following,
where the wh-word is extracted out of a complement clause and placed in sentence-initial position:6
Next, Cherokee shows the familiar distinction between Wh-extraction from complement
clauses, which is fine as we just saw, and extraction from adjunct clauses, which is typically
deviant (by Subjacency). For example, the following extractions are unacceptable:
It could be objected, at this point, that the above examples, despite the distinction between
complement/adjunct clauses, do not really involve wh-movement, but result from the widespread
topicalization processes that operate in Cherokee (and are perhaps in part responsible for the free
6
As observedby Scancarelli,Wh-wordsare typicallypJa1~ in sentenceinitial positionor immediatelyfollowing
a to.r,icalizedargument
For 'after' clauses, cf. Feeling and Pulte (1975:351)
8Toea. and b. sentencesare apparentlyvariantsof each other in Mrs. Carey's speech.
9
For 'when' clauses, cf. F1eelingand Pulte (1975:351)
110
word order). In other words, it could be argued that in (12-13) the wh-word is placed in
sentence-initial position simply because of its discourse-prominence. Of course, the 'topicalization'
theory will also have to explain why (14, 15, 16) are bad, if wh-movement is not involved. It is
possible, however, to provide evidence that topicalization is not involved in (12-13).
Topicalization (which does not involve extraction) and wh-movement can be distringuished
because they ot,ey different restrictions. A minimal pair like the following indicates that (14, 15,
16) are-very likely-bad because they feature impossible extractions. (17a), which is similar to
(14b) except for the negation in the matrix clause, is ungrammatical, but (17b), with no fronted
wh-element, is much better. Negation has been inserted in (17a,b) to insure that the sentence-initial
word is outside of the embedded clause: negation in Cherokee seems to act as a boundary to free
re-ordering within the sentence (B. Potter, p.c.).
Another minimal pair is given: (16b), a degraded sentence featuring extraction out of a
relative clause, contrasts in acceptability with (18):
Assuming, as the above data seems to indicate, that Cherokee has a syntactic process
comparable to Wh-extraction in English, we are faced with a seeming contradiction: since there is a
difference between extraction out of complement clauses and adjunct clauses, these must be
distinguished structurally-recall that we have assumed that lexically realized verbal complements
are really adjuncts. The most plausible conclusion, as suggested by Baker (1992), is that adjunct
status must be limited to nominal arguments: clausal arguments will not be adjuncts, but sit in true
argument positions.
I will henceforth assume that clausal arguments are in A-positions at D-Structure. This
follows Baker's (1992) account of Mohawk: nominal arguments are in A'-positions, but clausal
arguements are in A-positions.
One further issue needs to be considered, as pointed out by Baker. Assuming (with Jelinek
and Baker) that lexical arguments are adjuncts does not force the conclusion that extraction of
thematic arguments is actually extraction of adjuncts. Baker assumes that the licensing condition in
(9) strictly holds only of overt NPs. The crux of the matter is the position of the trace left by
extraction: is the trace in an A or A' position? Baker proposes that when we extract the whole NP.
and this NP is thematically an argument, we are indeed extracting an argument and not an adjunct:
i.e., the trace of the movement is in an A-position. This is because Condition (9) is taken to affect
only overt NPs, and not traces. Baker assumes that this is because the condition holds at
S-Structure: if an NP moves out of its original A-position by S-Structure, the trace it leaves behind
is not subject to Condition (9).
111
In this section I will consider the followil!lg question: how can we support the hypothesis
that the NPs that occur in argument positions in Cherokee are (covert) pronominals. The tests are
essentially those in Baker 1992, and involve the use of Principle C of the Binding Theory, with its
corollary below (sometimes referred to as 'Principle D'):
b. Digi a-gowhti'a
Dick 3sA-see:prs
'Dick 1 sees him 2 ' (l 1t 2)
Let's now review some data that illustrates: the application of Principle C in Cherokee. The
following examples show that in Cherokee, like in English, a pronominal subject cannot be
construed as coreferent with an NP in a complement clause: i.e., within the pronoun's c-command
domain. Recall that unlike NP complements, clausal complement are in A-positions.
If however the positions of the pronominal and the NP are switched, coreference is
possible:
112
(23) Maaygi jiiya-hnihv hla y-uu-tadlawoosvv'i
Mike lsA-hit:prf neg irr-3sB-anger:prf
'That I hit Mike 1 didn't make hiin 1 angry' [Coreference OK]
All these data further support the hypothesis that complement clauses in Cherokee are
arguments of V, and hence structurally distinct from adjunct clauses.
Coreference is also possible in the following example, where the overt NP ('Mike')in the
matrix clause is not c-commanded by the object pronoun in the adjunct clause:
Having established that Principle C does seem to apply to Cherokee, we can now show that
Principle C effects are consistent with the clause structure hypothesis developed so far, lending
support to it.
On the other hand, if the overt NP sits in a complement clause, a pronominal argument in
the matrix c-commands it, and we would expect-again, given Principle C-that the pronominal
argument could not corefer with the NP. The examples below confirm this prediction:
Naturally, for the above tests to be valid, we need to argue that the overt NP is in fact
inside the adjunct/complement clause, and not part of the matrix. The point is not obvious, owing
to freedom in word order. To control for scrambling of the NP out of the adjunct in (25a), the NP
has been placed to the right of uuneelagi:this connective seems to work as a complementizer, given
that the copula igi is prefixed with the subordinating marker j-. Generally in Cherokee a
complementizer defines a boundary to scrambling of NPs. In (25c ), we control for scrambling by
113
positioning the ove1t NP (an object) between negation and V. As noted, word order restrictions in
Cherokee are such 1thatin a negative clause the object cannot be placed to the left of the negation
word.
Principle C effects do not only provide support for the assumption that there is a structural
diff ere nee between complement and adjunct clauses; they also give us some positive evidence that
A-positions are occupied by small pros and that overt NP arguments are in A'-positions. The
relevant data involve pronominal reference with possessor phrases.
Consider the case where the overt refer1ential expression is a possessor NP embedded
inside another lexical NP. If this overt argument is in fact an adjunct, it would be outside the
c-connnand domain of a pronominal in argument position, and coreference should be possible. The
underlying structural configuration of a sentence like 'He 1 cut Mike 1's hair' (whose interrogative
counterpart is given in (28) below) would be, in other words, as follows:
(27)
Mike1
~
--------
NPj
N
II~
IP
NP
~
IP
VP
hair I NPj V
pro1 I I
proi cut
In the above example, the position of the question clitic ('-s') ensures that 'Mike's hair' is a
constituent.
A structural configuration akin to the one considered above with possessor phrases is given
by sentences witllt locative phrases, The example below displays the same coreference effect:
7. Conclusion
This paper has been an attempt at addressing some issues in Cherokee syntax. The status and
position of argument NPs has been discussed, and I concluded that Baker's account of Mohawk
non-configurationality (1992, 1993) seems largely applicable to Cherokee.
114
Cherokee Agentive Nominalizations
Brian Potter
Holmes and Smith (1976) also gloss -i forms such as (lb) with the English suffix -er but
suggest that the distinction between Cherokee -i and -oo'i is one of meaning, with the former
translated as "it is made to happen repeatedly through personal effort" and the latter as "it happens
habitually". Holmes and Smith would gloss (lb) as the complete sentence "he is a speaker"
(Holmes and Smith 1976: 130).
Cherokee constructions involving the -i suffix can also include an overt object and subject
argument. Examples (2b) and (3b) demonstrate the basic possibilities. (2a) and (3a), which I
assume without argument are standard Cherokee sentences, are given for comparison.
Interestingly, my consultant for this paper translates the -i constructions as either nominalized
expressions similar to English agentive nominalizations, or, as complete sentences equivalent to the
habitual forms using the -oo'i suffix.2
11 am grateful to my consultant, Mrs. Virginia Carey, whose mastery of the Cherokee language and willingness to
teach made this paper possible, and to the other members of the field methodsclass.
All data in this paper is taken either from the UCLA 1993field methods class data set, or from Pulte and
Feeling's grammar (1975). In the latter case, the page number of the grammar is noted for each example.
2Cherokeenouns referring to humans may be used as complete sentences(Pulte and Feeling 1975:308). This does
not, however, eliminate the need to verify the categorial status of the Cherokee nominalizations.In fact, it introduces
the additional question,which will not be consideredin this paper, of the proper treatment of nouns referring to
humans.
115
(3b) Asgaya dloo,geesigaa-daluugiis-g-i
man field 3sA-plow-imp-nom
i) 'The man is a field plower' ii) 'The man plows the field'
The categorial status of these Cherokee constructions cannot be decided on the basis of
gloss alone. However, if the -i forms in the (b) examples above are in fact nominalized
expressions, syntactically distinct from the habitual (a) constructions, distributional asymmetries
between the two types of expression should exist
In the discussion below, I examine such asymmetries between Cherokee verbs with the
habitual -oo'i suffix and with the -i suffix and demonstrate that Pulte and Feeling are correct in
classifying the -i forms as nominalized expressions. I then discuss the Cherokee nominalizations
from a typological perspective with similar constructions in English and Navajo. Although the
Cherokee forms exhibit a number of properties common to both English agentive compounds and
Navajo agentive nominalizations, certain distinctions among these languages suggest that the
Cherokee data cannot be the result of a single, stmcture-based constraint that theoretically accounts
for the English and Navajo forms.
As example (5) demonstrates, although a constituent headed by a verb with the -i suffix can
be glossed as a sentence, that constituent cannot serve syntactically as a sentential complement to
the verb 'know'. To make (5) a grammatical construction, the addition of the copular verb after the
-i constituent, as in (6), is necessary.
116
(7) Maaygi hla ga-wooniisgi yigi.
Mike neg 3sA-speak-nom neg.be
'Mike is not a speaker'
The syntactic asymmetries in (4)-(8) are unexpected if verbs with the -i suffix only vary in
meaning from verbs with habitual or present tense suffixes. The data are clearly more consistent
with a structural distinction between the -i constructions in (5)-(7) and the standard sentences in (4)
and (8). Examples (9)-(20) further support a structural distinction between Cherokee -i and -oo'i
constructions. Cherokee sentences can be conjoined via the clitic -Imo as in example (9).
When -hno is used to conjoin a construction headed by a verb with the -i suffix and a
standard sentence, however, the resulting sentence does not have an interpretation of two
conjoined clauses. In (10), the -i construction follows the conjunction clitic and is interpreted as the
object of the initial verb in the sentence. In (11), the -i construction precedes the clitic and is
interpreted as the subject argument of the sentence.
While it is unclear what function the clitic -hno serves in these latter two examples, the
distinct interpretation of these sentences from that of (9) clearly motivates a structural distinction
between -i constructions and standard Cherokee sentences.
Another form of conjunction in Cherokee also points toward a syntactic distinction between
-i constructions and sentences. Sentences in Cherokee can be conjoined via the word haleeh as in
(12).
Example (13) illustrates that two -i constructions can be conjoined in this manner.
As examples (14) and (15) demonstrate, however, haleeh cannot conjoin a Cherokee
sentence with a -i construction.
117
(14) *Dloogeesi gaa-daluugiisg-oo'i haleeh wahya dii-hiih-i
field 3sA-plow-hab also coyote dis-kill-nom
'He plows fields and is also a coyote killer'
The ungrammaticality of (14) and (15) thus motivate a structural distinction between
Cherokee sentences and -i constructions. Note that the ungrammaticality of these examples cannot
be due to a requirement that two conjoined clauses exhibit similar tense and aspect morphology. In
(16) for example, a present tense habitual clause is conjoined with a future tense perfective.
In constructions based on verbs with the -i suffix, however, such freedom of word order is
either constrained or results in distinct interpretation. Except for the suffixal ending on the verb,
example (19) is equivalent to (17). In (19), however, the adjective is interpreted as modifying the
derived nominaliz.a.tion,which denotes the non-overt third person subject of the verb, rather than as
modifying the overt object.
Examples (9)-(20) suggest that a Cherokee verb with the -i suffix forms a syntactic unit
with an argument that is distinct from the structural configuration of a verb and argument in the
absence of this suffix. The necessity of an auxiliary verb in (6) and (7) further suggests that this
structural unit does not bear the verbal features necessary to form a sentential complement or
participate in sentential niegation.
118
In fact, there are good reasons to accept Pulte and Feeling's characterization of these -i
forms as nominalizations. Cherokee postpositions, which take nominal arguments, cannot take
sentential complements. (21) illustrates the use of the postposition 'toward' with the nominal
complement 'house'.
Example (23), however, which is distinct from (22) only in the presence of the suffix -i, is
grammatical. This case thus suggests that Cherokee constructions headed by a verb with the -i
suffix are categorically appropriate as nominal complements to postpositions.
Finally, example (24) demonstrates that Cherokee -i constructions can be conjoined with
canonical nominal expressions.
U9
(25) NO
I \
N N
fish / \
V nom
eat -er
One straightforward hypothesis regarding Cherokee nominalizations is that they are
structurally and derivationally equivalent to their English compound counterparts. If so, taking into
account independent differences between Cherokee and English syntax, correlations should exist
across nominalizations for the two languages. In the following sections, I demonstrate that
Cherokee nominalizatio1t1sexhibit two properties characteristic of English synthetic compounds. In
each case, I review accounts of the English data which rely crucially on the word level status of
compounds.
Under the assumption that the prominence theory is sensitive to the argument structure of the
deverbal head of a synthetic compound, and not simply the arguments overtly realized,
Grimshaw's account might be extended to include the examples in (28). In (28), a theme argument
of a dative verb is acceptable in a compound construction while a goal argument is not.4
With respect to dative verbs, Cherokee -i constructions pattern similarly to the English
agentive synthetic compounds. For example, (29) illustrates the Cherokee word for 'giver' and
(30) demonstrates that it is possible to include an overt theme argument of the nominalized form of
'give'.
120
Example (30) might be uttered in a context where there is a particular person who regularly
hands out apples to someone or some group of people. Assuming a context in which children are
the recipients of the apples, however, 'child' is unacceptable as the sole argument of the
nominalized verb (31). Note that the intended reading of (31) is with 'child' as the goal argument
of the nominalized verb.
Adopting for the moment Grimshaw's analysis of English dative compounds, the
unacceptability of goal arguments in Cherokee nominalizations supports an analysis of the
Cherokee -i forms as word level constructions. The presence of the morpheme aadaa in (31 ),
however, introduces a complication into the dative paradigm. Pulte and Feeling note that aadaa is
placed before a consonant initial verb stem to indicate a covert animate direct object of the verb. For
example, the nominalized form of 'kill' is realized as in (32) if there is no explicitly stated direct
object, but as in (33) if there is an overt direct object.
Pulte and Feeling state that aadaa will only surface if there is no overtly expressed direct
object (1975: 298). With the dative verb 'give' in example (30), however, aadaa is present despite
the overt realization of the direct object argument. (31 ), in which the presence of an overt indirect
object is unacceptable, thus suggests that with dative verbs aadaa marks a covert animate indirect
object. 5
That aadaa denotes an indirect object argument for dative verbs can also be demonstrated by
a comparison with dative stems without the aadaa morpheme. Example (34) illustrates a
grammatical sentence including a dative verb and three overt arguments.
In sentences without aadaa on the verb stem and with only one overt argument (35), the
overt argument can be interpreted as subject, direct object, 6 or indirect object. With aadaa present,
however, a lone argument cannot be interpreted as indirect object (36).
121
On the basis of the discussion above, it seems clear that aadaa marks a covert animate
indirect object of a dative verb. The ungrammaticality of (31), which includes both aadaa and an
overt indirect object, is therefore predicted independently of the prominence theory of theta
assignment and does not by itself constitute a correlation between the Cherokee nominalizations
and their English counterparts. Additional data, however, do point to such a correlation.
Although aadaa, as evidenced by (34) and (35), is not an obligatory element in Cherokee
dative sentences, nominalization constructions based on dative verbs without the aadaa morpheme,
such as (37) and (38), are unacceptable.7
Still, it is not obvious that Grimshaw's prominence theory 8 can explain the Cherokee data.
Under the assumption that aadaa receives the goal theta role assigned by a dative verb, Grimshaw's
prominence theory approach requires that the the:ta role assigned to aadaa be assigned after the
theme role is assigned to the direct object svvhkta.Exactly the opposite ordering of assignment is
expected, however, if as Grimshaw suggests, structural proximity to the theta assigning stem
directly correlates with early assignment. In summary, the correlation between Cherokee
constructions involving dative verbs with the -i suffix and equivalent nominalizations in English
and Navajo does not offer independent motivation for the prominence theory of theta assignment.
Therefore, although a cmsslinguistic account of the dative paradigm might prove consistent with a
word level analysis of the Cherokee nominalizations, the data do not in and of themselves
constitute evidence for such an analysis.
Navajo tense is realized via morphology on the verb stem. As with English, verbs marked
for tense cannot be nominalized. (41) and (42) offer a nice minimal contrast between a tensed verb
with the nominalizing suffix and a tensed verb with the relativizing suffix respectively.
122
(41) Navajo 9 *ma'ff neid66itsit-i' (42) Navajo ma'ff nefd66itsil-i'gff
coyote kill.fut-nom coyote kill.fut-rel
'coyote killer (future)' 'the one that will kill coyotes'
Cherokee constructions involving th1e suffix -i and a verb stem marked for tense are
similarly ungrammatical. As (44) indicates, the future tense form of the verb 'speak' cannot be
nominalized. To indicate past or future time for a nominalized verb, an auxiliary must be used as in
(45).
Note that (45) demonstrates that the ungrammaticality of (44) is not a result of the 1 person
subject morphology on the nominalized verb stem. First and second person marking on
nominalized verbs will be discussed in more detail in the next section.
DiSciullo and Williams (1987) formulate a theory of syntactic atomicity which attempts to
account for the lack of referential and non-generic material within word level categories of a
language. Briefly, DiSciullo and Williams suggest that :xO level categories are the smallest units
visible to syntactic rules and principles and argue that no linguistic information important to
sentence or discourse level syntax and semantics can be wholly contained below word level. I will
return to this theory below in a discussion of the referentiality of arguments of nominalized verbs
in Navajo and Cherokee. For now, note that syntactic atomicity is applicable to the lack of tense in
English compounds if, as commonly assumed, these compounds are word level categories. To
account for the lack of tense in the Navajo and Cherokee examples along similar lines it is therefore
necessary to tentatively assume that the nominalization constructions, or at least their deverbal
heads, are word level categories in these languages as well.
91 would like to thank Ms. Lilly Lane of the Navajo Nation, who providedthe Navajo examplesgiven in this t.ext.
123
English, simply nominalized verbs which retain th1etamarking properties.
4. 1. Phrasal Arguments. The argument noun of a nominalized transitive verb can be modified
by an adjective in both Cherokee and Navajo. In Cherokee, adjectives can occur either to the left or
right of the noun they mcxlify.10Examples (46) and (47) illustrate that either ordering is possible
between an adjective and a noun argument of a nominalized verb.
Adjectives can also modify the arguments of nominalized verbs in Navajo. Example (48)
demonstrates that modification of the nominal by multiple adjectives is acceptable.
Within analyses in which adjectives projiect adjectiv~ phrases which either adjoin to or
dominate modified noun phrases, the facts in (46) - (48) suggest that the arguments of nominalized
verbs in Navajo and Cherokee are phrasal. It is also possible, however, that the adjectival
modification in these examples represents adjective-noun compounding. Thus, on the basis of
these cases alone it cannot be concluded that the argument of a nominalized Cherokee or Navajo
verb may be phrasal The examples below, however, do indicate phrasal status for the arguments
of nominalized verbs.
In the examples below, Cherokee nominalized verbs have syntactically complex phrasal
arguments. The argument in (51), for example, includes a canonical instance of a Cherokee relative
clause.
Example (52) indicates that the argument of a nominalized Navajo verb may also be a
complex phrasal constituent
124
(52) Navajo hast6f shi-da'a'k'eeh nidefnHheesh-i'gff nefitseed-f
men I-field irrigate-rel kill-nom
'[men that irrigate my field] killer'
The presence of phrasal arguments with the Cherokee and Navajo nominalized verbs poses
a serious problem for any account of these constructions which suggests that the nominalized
expression is a word level category. (53) once again gives the structural analysis commonly
applied to English synthetic compounds. Assuming that the Cherokee and Navajo nominalizations
are as minimally distinct as possible from this construction, the phrasal arguments of the
nominalized verbs in these languages requires the structure given in (54).
I \ I \
N N NP/CP/AdjP N
argument verb+ er argument verb+ i
The structure in (54), however, is incompatible with Xbar theory (cf., Stowell 1981). Xbar
theory holds that all heads or word level categories project phrases. But in (54) the noun that
results from affixation of the nominalizing suffix to a verb stem does not project any higher
structure. Furthermore, although in compounding constructions such as (53) all heads need not
project higher order structure, (54) cannot be an example of compounding as the argument of the
nominalized verb is phrasal. DiSciullo and Williams do note a few cases in French that seem to
require that a phrasal constituent be dominated directly by a word level category, but require that
such structural configurations be limited to idiomatic and/or lexicalized expressions. For both
Cherokee and Navajo, however, nominalized verbs productively take phrasal arguments.
Within an Xbar theoretic framework, the phrasal arguments of Navajo and Cherokee
nominalized verbs must be realized external to any word level nominal category. But the question
remains as to the surf ace syntactic location of these arguments as well as the nominalization clitic
itself. The rough format for two candidate structures are given in (55) and (56).
(55) NP (56) NP
I \ I \
NP N VP N
argument verb -i argument+verb -1
In (55), the -i suffix is treated similar to the English agentive suffix -er and affixes to a verb
to form a noun. In this case, the phrasal arguments in the Cherokee and Navajo nominalizations are
realized in some adjunct position within the noun phrase projected by the nominalized verb. In
(56), the nominalization suffix is treated as a noun which takes as argument some verbal projection
and projects a noun phrase. In the next section, I provide evidence from Cherokee and Navajo
which clearly supports the structure given in (56).
12Potter(1992) argues that this 3rd person marking is not a default case. The Navajo 3rd person objectyi surfaces
only if there is an object referent For an unspecifiedobject the morphemea'i is used.
125
(57) dloogeesi gaa-daluugiisg-i (58) Navajo ma'ff neiitseed-f
field 3sA-plow-nom ma'ii na-yi-f-tseed-i
'field plower' coyote iter-3.obj-cl-kill-nom
'coyote killer'
There are two general possibile analyses of agreement morphology in Cherokee and Navajo
nominalizations. First, along the lines originally proposed in Pollock (1989) and Chomsky (1989)
on the basis of IndoEuropean languages, agreement morphology in examples (57) and (58)
surfaces as a result of a specifier-head relationship in syntax between the verb and its noun phrase
argument. Such a relationship necessarily entails the existence of a verb based, phrasal constituent.
Figure (59) illustrates thi~basic syntactic structure involved for a Cherokee nominalization.13
AGRP NOM
I \ I
NP AGR' i
dloogeesi I \
AGR VP
gaa + daluugiisgi
Alternatively, agreement markers in some languages might actually be the overt realization
of verbal arguments. Baker (1992) argues that thi~ overt noun phrases in Mohawk sentences are
actually adjuncts adjoined to the sentence. Baker suggests that the actually arguments in Mohawk
sentences are the agreement markers which surface on the verb stem. In fact, Beghelli (1993) and
Jelinek (1989) argue that Baker's analysis of Mohawk is applicable to Cherokee and Navajo
respectively. These accounts suggest that the third person agreement morphology in (57) and (58)
represents incorporated third person arguments. Since these arguments are the complement or
object arguments of the given verbs, they must originate in a complement position to these verbs.
But the presence of a complement position to a verb necessarily implies a verb phrase constituent.
Figure (60) provides a simplified illustration of the proposed structure.
In summary, within any Principles and Parameters based approach to Cherokee and Navajo
agreement morphology, the presence of agreemen1t marking on the nominalized verb stem clearly
indicates a phrasal projection of the verb within the nominalization construction. Rather than
transform a verb into a noun, as argued for English -er, the Cherokee and Navajo nominalization
suffixes transform verbal projections into nominals.
Of course this argument is based upon the assumptions that phrasal constituents cannot be
dominated by word level categories and that the presence of agreement morphology entails a
phrasal projection. Without these assumptions, the possibility that the Cherokee and Navajo
nominalizations consist of word level nominalized verbs to which phrasal arguments adjoin might
still be considered. Potentially such an account could make extensive use of the word level status
13Speas (1990)argues that the agreement
affixesare attachedto the verb stem via a processof lowering.Although
Speas(pc) no longerfavorsa loweringapproach,the alternativesare unclear. Cruciallyfor this paper,any account
fonnulated withincurrentPrinciples& Parameterstheorymakesuse of the specifier~headrelationship.
126
of these nominalizations to explain the properties characteristic of the construction cross-
linguistically. Additional examples discussed below, however, demonstrate that Cherokee and
Navajo nominalizations are distinct from each other as well as from English. This three-way
distinction among the English, Navajo and Cherokee constructions strongly suggests that the
power of several independent principles, as opposed to the elegance of a single constraint on word
level categories, is necessary for a comprehensive theory of nominalization.
5.1. Specific NP in Nominalizations. Specific determiners in both Cherokee and Navajo can
occur either to the left or right of the noun they modify. As (61) and (62) demonstrate, a specific
determiner can modify a noun argument of a Cherokee nominalized verb.
(63) Navajo dff da'a'k'eeh neinitheesh-i' (64) Navajo *da'a'k'eeh dii neinitheesh-i'
this field imgate-nom field this irrigate-nom
'this [field irrigator]' ' [this field] irrigator'
Cherokee permits other specific and/or inherently referential noun phrases as arguments of
nominalized verbs. As with the specific determiner examples above, the corresponding English and
Navajo constructions are ungrammatical. Example (65) demonstrates that the name of an individual
can occur as an object argument in a Cherokee nominalization.
In the corresponding Navajo example (66), however, the name cannot refer to a specific
127
individual but may only represent the class of people named 'John'. The English example in (67)
has a similar interpretation.
As the examples. above illustrate, Cherokee nominalizations do not exhibit the referential
constraint on nominalization arguments that is prevalent in the English and Navajo examples. If the
absence of tense on Cherokee nominalized verlbs is due to such a constraint, these data are
surprising. However, it is again possible to appeal to the adjunct theory of Cherokee lexical
phrases and suggest that the object NP in the Cherokee examples above are not included within the
nominalization construction and are thus not subject to opacity constraints. Still, the nominalized
verb stem, which under any account must be included within the nominalization structure, also
exhibits violations of opacity constraints.
Once again, the corresponding Navajo and English examples are ungrammatical.
Since the nominalizationsdiscussedin this paper refer to humanbeings, ie. 'coyotekiller', 'field irrigator',
etc..., the presence of subject agreement morphologyin (70) may not be a result of a syntacticrelationshipbetween
the verb stem 'speak' and a coivert1st personsubject.Crucially,however,the presenceof object agreement
morphology,as with the 1st person markingin (71) and the 3rd person markingexhibited in the examples
throughoutthis paper, clearly suggest a syntacticrelationshipbetweenthe verb stem and its thematicobject
128
(72) Navajo *shi-difid-'i (73) *me-burner
1-burn-nom 'one who burns me repeatedly'
'me-burner'
Clearly. Cherokee nominalizations do not exhibit the opacity constraint against agreement
and specific NP that is operative in English and Navajo nominalizations. Despite this fact,
Cherokee still patterns with English and Navajo in the exclusion of tense from nominalization
constructions. The Cherokee data thus present a paradox for the theory of referential opacity. If
tense, agreement and specific NP are excluded from nominalization constructions in English and
Navajo because of a general opacity constraint that applies to word level categories, then this
constraint should either apply in whole or not at all to Cherokee. As noted, however, only tense is
excluded from the Cherokee forms. The Cherokee data thus offer no support in favor of a single
word based account of referential opacity. Furthermore, since the arguments of Cherokee
nominalized verbs can be phrasal as well as specific and/or referential, there is no reason to
hypothesize that these arguments are dominated by a word level nominal category as in the case of
English synthetic compounds. These facts in conjunction with the presence of agreement
morphology on the nominalized verb stem strongly suggest that the Cherokee nominalizing suffix
transforms a verbal projection into a nominal.
(74) NP
I \
VP N
argument+verb -i
The internal properties of the Cherokee nominalizations, such as the possibility of phrasal
arguments and agreement morphology, suggest that the nominalization construction contains a
phrasal projection of the verb stem. As noted, the presence of phrasal projections under an :x:O
level category. the situation if Cherokee nominalizations are word level categories, is incompatible
with Xbar theory. From a theoretical perspective then, there is reason to believe that the Cherokee
nominalizations are phrasal. In this section I discuss one potential data based argument that the
result of Cherokee -i suffixation is phrasal.
There are several suffixes which Pulte and Feeling note are restricted to nouns. One such
suffix is the -'i morpheme which can be loosely translated as 'place of. In (75), this suffix
attaches to the Cherokee noun for 'crow' and results literally in 'the place of the crow'. The
resulting expression is currently the name of a market in Oklahoma. (76) gives another example of
the -'i suffix. In this case, affixation results literally in 'the place of the women' which might be
used in the context of an all women's seminar.
129
Another suffix restricted to nouns in Pulte and Feeling is the suffix -yaa'i which means
'real' or 'pure'. In (79), this suffix attaches to the noun for 'people' and the literal result 'pure
people' signifies a Native American. Attached to the noun for 'pig' in (80), this suffix results in the
meaning 'pure bred pig'.
Pulte and Feeling also note a number of suffixes that can attach to either nouns or verbs.
(83) and (84) illustrate the use of the suffixes -ju and -dv which correspond to interrogative and
focus markers respectively. Note that these suffixes are compatible with the nominalization
construction. (85) demonstrates affixation of the interrogative morpheme to a verb.
It is somewhat surprising that the noun suffixes in (75)-(82) cannot attach to the -i
constructions if these constructions are indeed nominal. One possible explanation for this fact,
however, is that these suffixes attach to a word level category, NO, but that the nominalization
construction is phrasal. Assuming that the suffixes in (83)-(85) attach to any phrasal constituent of
a sentenc~. the suffixation of these morphemes to the nominalization construction is then
straightforward.
There is, however, another possible explanation for the data above. Fabb (1988) explicitly
suggests that some English suffixes simply will not attach to an already suffixed word. Such an
analysis of the Cherokee 'place of and 'pure' suffixes would readily account for the lack of
suffixation to nominalized verbs. Alternatively, Richard Wright (pc) suggests that adjunction of
these suffixes to a nominalized verb might result in some form of unresolvable intonation clash.
Since I have largely ignored the intonational contours of these nominalization constructions, I
cannot at this point comment on the potential of such an account.
7. Summary Discussion
130
grammar. Unlike English and Navajo, however, the Cherokee nominalizations do not exhibit
referential opacity effects with regard to agreement and specific NP.
Word based accounts of English compounds exist for each of the properties common to the
nominalizations of all three languages. If the noiminalization forms of these three languages are to
be grouped together and treated similarly, however, one universally applied, structurally based
constraint on word level categories cannot explain both the lack of tense, agreement and specific
NP in English and Navajo constructions and the lack of tense but presence of agreement and
specific NP in the Cherokee forms.
(i) NP
I \
AGROP N
I
VP
Under this approach, the ungrammaticalityof specific NP arguments within the nominalization
construction follows from the fact lhat the positions above AGROP necessary for the licensing of specific NP (cf.
Sportiche 1992and Diesing 1990) are not available.The Cherokeedata might be derived in a similar fashion with
the nominalizingsuffix selecting a verbal projection higher than AGROP as complement.
131
132
References
Aissen, Judith. 1983. Indirect Object Advancement in Tzotzil. In David Perlmutter, ed., Studies in
RelationalGrammar1. Chicago: University of Chicago Press.
Aissen, Judith. 1987. Tzotzil ClauseStructure.Dordrecht: Reidel.
Allen, Barbara, and Donald Frantz. 1983. Advancements and Verb Agreement in Southern Tiwa.
In David Perlmutter, ed., Studies in Relational Grammar 1. Chicago: University of Chicago
Press.
Allen, Barbara, Donald Frantz, Donna Gardiner, and David Perlmutter. 1990. Verb Agreement,
Possessor Ascension and Multistratal Representation in Southern Tiwa. In Paul Postal and
Brian Joseph, eds., Studies in RelationalGrammar3. Chicago: University of Chicago Press.
Barwise, Jon, and John Perry. 1983. Situations and Attitudes. Cambridge: MIT Press.
Basso, Keith H. 1968. The Western Apache Classificatory Verb System: A Formal Analysis.
SouthwestJournal of Anthropology24:3, 252-266.
Beghelli, Filippo. 1993. A Minimalist Approach to Quantifier Scope. In Proceedingsof NELS 23.
Bender, Ernest, and Zellig S. Harris. 1946. The Phonemes of North Carolina Cherokee. /JAL 12:
14-21.
Ben-Shalom, Dorit. 1993. Object Wide Scope and Semantic Trees. In Proceedingsof SALT 3.
Bladon, Anthony. 1986. Phonetics for Hearers. In Graham McGregor, ed., Language for
Hearers.Oxford: Pergamon.
Blake, Barry. 1990. Relational Grammar.New York: Routledge.
Chung, Sandra. 1981. Transitivity and Surface Filters in Chamorro. In J. Hollyman and A.
Pawley, eds., Studies in Pacific Languages and Cultures in Honour of Bruce Biggs, 311-332.
Auckland: Linguistic Society of New Zealand.
133
Collier, Rene, Leigh Lisker, Hajime Hirose, and Tatsujiro Ishijima. 1979. Voicing in Intervocalic
Stops and Fricatives in Dutch. Journal of Plwnetics 7: 357-373.
Comrie, Bernard. 1981. Language Universals and Linguistic Typology. Chicago: University of
Chicago Press.
Comrie, Bernard. 1982. Grammatical Relations in Huichol. In Paul Hopper and Sandra A.
Thompson, eds., Studies in Transitivity, Syntax and Semantics 15. New York: Academic
Press.
Cook, William Hinton. 1979. A Grammar of North Carolina. Cherokee. Yale University Ph.D.
dissertation.
Davidson William, L.Vv. Elford, and Harry Hoijer. 1963. Athapaskan Classificatory Verbs. In
Henry Hoijer, et al., Studies in the Athapaskan Languages. University of California
Publications in Linguistics, Vol. 39. Berkeley: University of California Press.
Diesing, Molly. 1990. The Syntactic Roots of Semantic Partition. University of Massachusetts
Ph.D. dissertation.
Dowty, David. 1982. Grammatical Relations and Montague Grammar. In Polly Jacobson and
Geoffrey K. Pullum, eds., The Nature of Syntactic Representation. Dordrecht: Reidel.
Feeling, Durbin, and "rilliam Pulte. 1975. Includes Feeling 1975 and Pulte and Feeling 1975.
Tahlequah, OK: Cherokee Nation of Oklahoma
Frishberg, Nancy. 1972. Navaho Object Markers and the Great Chain of Being. John Kimball,
ed., Syntax and Semantics 1. New York: Seminar Press.
Gaz.dar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag. 1985. Generalized Phrase Structure
Grammar. Cambridge, MA.: Harvard University Press.
Goldsmith, John. A. 1990. Autosegmantal and Metrical Plwnology. Oxford: Basil Blackwell.
Grimshaw, Jane, and Armin Mester. 1988. Light Verbs and Theta-Marking. Linguistic Inquiry 19:
205-232.
134
Hale, Ken. 1972. A Note on Subject-Object Inversion in Navajo. In Braj Kachru et al, eds.,
Papers in Linguistics in Honor of Henry and Renee Kahane. Urbana: University of Illinois
Press.
Hale, Ken. 1983. Warlpiri and the Grammar of Non-configurational Languages. Natural Language
and LinguisticTheory 1: 5-49.
Holmes, Ruth Bradley, and Betty Sharp Smith. 1977. Beginning Cherokee.2nd ed. Norman, OK:
University of Oklahoma Press.
Jelinek, Eloise. 1984. Empty Categories, Case, and Configurationality. Natural Language and
Linguistic Theory 2: 39-76.
Jelinek, Eloise. 1989. Headless Relatives and Pronominal Arguments: A Typological Perspective.
CLS Parasession on Native American Languages, Indiana University Press.
Jelinek, Eloise. 1991. Grammatical Relations and Coindexing in Inverse Systems. In Katarzyna
Dziwirek et al, eds., GrammaticalRelations.A Cross-theoreticalPerspective.Stanford: CSLI.
Kawasaki, Haruko. 1982. An Acoustical Basis for Universal Constraints on Sound Sequences.
University of California, Berkeley, Ph.D. dissertation.
Kingston, John. 1990. Articulatory Binding. In John Kingston and Mary E. Beckman, eds.,
Papers in Laboratory Phonology 1: Between the Grammarand Physics of Speech, 406-434.
Cambridge: Cambridge University Press.
Kirchner, Robert. 1993. Round and Back Vowel Harmony and Disharmony: An Optimality
Theoretic Account Ms, UCLA.
Larson, Richard K. 1987. On the Double-Object Construction. Center for Cognitive Science,
Lexicon Project WorkingPapers. Cambridge, MA.: MIT.
Lichtenberk, Frantisek. 1982. Individuation Hierarchies in Manam. In Paul Hopper and Sandra A.
Thompson, eds., Studies in Transitivity,Syntax and Semantics15. New York: Academic
Press.
135
Lieber, Rochelle. 1992. Deconstructing Morphology. Chicago: University of Chicago Press.
Lindblom, Bjorn. 1990. Phonetic Content in Phonology. PERILUS 11: IOJ:-118. Stockholm:
University of Stockholm.
Lindsey, Geoffrey Alan. 1985. Intonation and Interrogation: Tonal Structure and the Expression of
a Pragmatic Function in English and Other Languages. UCLA Ph.D. dissertation.
Lindsey, Geoffrey. n.d. [c. 1987]. Cherokee Pitch Phonology. Ms., University College, London.
Liu, F.-H. 1990. Scope Dependency in English and Chinese. UCLA Ph.D. Dissertation.
Maddieson, Ian. 1980. A Survey of Liquids. UCLA Working Papers in Phonetics 50: 93-112.
Los Angeles: UCLA.
Marslen-Wilson, W. D.,, and P. Zwitserlood. 1989. Accessing Spoken Words: The Importance of
Word Onsets. Journal of Experimental Psychology: Human Perception and Peiformance 15:
576-85.
Martin, Jack. 1991. The Determination of Grammatical Relations in Syntax. UCLA Ph.D.
dissertation.
Munro, Pamela, and Lynn Gordon. 1982. Syntactic Relations in Western Muskogean: A
Typological Perspective. Language 58: 81-115.
Nunberg, Geoffrey. 1977. The Pragmatics of Reference. City University of New York Ph.D
Dissertation.
Oehrle, Richard. 1976. The Grammatical Status of the English Dative Alternation. MIT Ph.D
Dissertation.
Ohala, John J. 1983. The Origin of Sound Patterns in Vocal Tract Constraints. In Peter F.
MacNeilage, ed., The Production of Speech, 189-216. New York: Springer-Verlag.
Ohala, John J. 1992. Alternatives to the Sonority Hierarchy for Explaining Segmental Sequential
Constraints. Papers from the Parasession on the Syllable. 319-338. Chicago, IL: Chicago
Linguistics Society.
Pierrehumbert, Janet, and Mary Beckman. 1988. Japanese Tone Structure. Cambridge, MA.:MIT
Press.
Pollard, Carl, and Ivan Sag. 1987. Information-Based Syntax and Semantics, volume 1. Stanford:
CSLI.
Pollard, Carl, and Ivan Sag. 1994. Head-Driven Phrase Structure Grammar. Chicago: University
of Chicago Press.
Pollock, Jean-Yves. 1989. Verb Movement, Universal Grammar and the Structure of IP.
Linguistic Inquiry 20.
136
Potter, Brian. 1991. Dative Compounding and the Prominence Theory of Theta Assignment.
Proceedingsof the EasternStates Conferenceon Linguistics,289-300.
Potter, Brian. 1992. Nominalization and Compounding in Navajo. Talk presented at the
Athabaskan Language Conference, Flagstaff, Arizona.
Prince, Alan, and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative
Grammar. Ms., Rutgers University and University of Colorado at Boulder.
Pulte, William, and Durbin Feeling. 1975. Outline of Cherokee Grammar. In Feeling and Pulte
1975, pp. 235-355.
Reyburn, William D. 1954. Cherokee verb morphology III. /JAL 20: 44--64.
Rosen, Carol. 1990. Rethinking Southern Tiwa: The geometry of a triple-agreement language.
Language 66.669-713.
Rude, Noel. 1982. Promotion and Topicality of Nez Perce Objects. In BLS 8.
Silverstein, Michael. 1976. Hierarchy of Features and Ergativity. In R.M.W. Dixon, ed.,
GrammaticalCategoriesin AustralianLanguages.Canberra:Australian Institute of Aboriginal
Studies.
Speas, Margaret. 1990. Phrase Structure in Natural Language.Studies in Natural Language and
LinguisticTheory21. Dordrecht: Kluwer Academic Publishers.
Sportiche, Dominique. 1992. Clitic Constructions.Ms., UCLA.
Steriade, Donca. 1987. Locality Conditions and Feature Geometry. Proceedings of the North
EasternLinguisticSociety 17: 595-618.
Steriade, Donca. 1992a. Complex Onsets as Single Segments: The Mazateco Pattern. Ms., UCLA.
Steriade, Donca. 1992b. Mazateco Onsets. To appear in Jennifer Cole and Charles Kisseberth,
eds., Proceedingsof the Universityof Illinois PhonologyConference.Stanford, CA: CSU.
137
Steriade, Donca. 1993.. Closure, Release, and Nasal Contours. In Marie Huffman and Rena
Krakow, eds., Nasals, Nasalization, and the Velum, Phonetics and Phonology 5. New York:
Academic Press.
Stevens, Kenneth N., and Samuel J. Keyser. 1989. Primary Features and Their Enhancement in
Consonants. Laguage 65. 81-106.
Stowell, Tim. 1981. Origins of Path Structure. MIT PhD dissertation.
Ultan, Russell. 1978. A Typological View of Metathesis. In Joseph H. Greenberg, ed.,
Universals of Human Language, vol 2: Phonology, 367-402. Stanford, CA: Stanford
University Press.
Wallace, Karen Kay. 1993. Verb Incorporation and Agreement in Crow. UCLA Ph.D.
dissertation.
Yoshioka, Hirohide; Anders Lofqvist; and Hajime Hirose. 1982. Laryngeal Adjustments in
Japanese Voiceless Sound Production. Journal of Phonetics 10: 1-10.
138