44 Robert Beard

2 Derivation
ROBERT BEARD

1 Derivation versus inflection

Unlike inflectional morphology, which specifies the grammatical functions
of words in phrases without altering their meaning, derivational morphology or
word formation is so named because it usually results in the derivation of a new
word with new meaning. This traditional definition, however, has failed to
secure a distinction between the two types of morphology, and the reasons for
this failure have become matters of considerable discussion. Before proceeding
to the question of what is derivational morphology, therefore, it makes sense
to first attempt to locate the inflection–derivation interface.

2 The derivation–inflection interface

Chomsky (1970) proposed a sharp modular distinction between lexical and
syntactic processes, known widely under the rubric of Lexicalism.1 According
to the Lexicalist position, words are derived in the lexicon and emerge with an
internal structure to which syntax has no access (Lexical Integrity Hypothesis,
Postal 1969). Sentences like I speak Russiain though I’ve never been therei are
thereby ruled out, since the pronoun there is syntactically coindexed with a
lexeme-internal morpheme, Russia, which has no independent status in the
syntax. Sentences, on the other hand, are generated by the principles of syntax,
to which lexical operations have no access. This rules out phrasally based
lexical items such as over-the-counter in over-the-counter sales, widely held to be
extragrammatically generated.2
Lexicalism entails a set of diagnostics which distinguish derivation from
inflection. First, if inflection is relevant only to syntax, the output of inflectional
rules cannot be listed lexically. Derivation, on the other hand, is purely lexical,
so the output of a derivation rule is a new word which is subject to lexical
 Derivation 45

listing. Listing allows lexical but not inflectional derivates to semantically idio-
matize or lexicalize. Even though went has been phonologically lexicalized for
centuries, semantically it has remained no more than the past tense of go.
Terrific, on the other hand, has lost all semantic contact with its derivational
origins in terror and terrify, despite its residual phonological similarity.
Second, if lexical operations precede syntactic ones, and if derivational opera-
tions map isomorphically onto marking operations (see section 6 for alternat-
ives), inflectional markers will always occur outside derivational markers, as
in Russian lët-cik-a fly-agent-gen ‘the flyer’s (pilot’s)’, where the derivational
agentive marker -(s)cik precedes the inflectional case marker -a. Third, since
inflection is purely syntactic, it cannot change the lexical category of a word;
derivation can. The agentive suffix in this example changes the verbal base to
a noun, but the case ending does not affect that nominal status.
Finally, since inflection specifies syntactic relations rather than names
semantic categories, it should be fully productive. If an inflectional stem is
susceptible to one function of a paradigm, it is susceptible to them all. No
verb, for example, should conjugate in the singular but not the plural, or in the
present but not the past tense. The productivity of derivation, however, is
determined by semantic categories, and we would expect derivation to be
constrained by less predictable lexical conditions.
Unfortunately, each of the Lexicalist diagnostics is vexed by some aspect
of the data. Derivation does change the meanings of words so as to allow
the derivate to become a lexical entry in the lexicon. Case functions, however,
also lexicalize. In Russian, for example, the Instrumental never marks punctual
time with the odd exception of instances involving temporal nouns which form
natural quadruplets – for example, utr-om ‘in the morning’, dn-em ‘in the after-
noon’, vecer-om ‘in the evening’, and noc ’-ju ‘at night’. There is simply no way
to derive punctuality from the major or minor functions of the Instrumental:
that is, manner, means, vialic, essive. Punctuality is productively marked by
v ‘in’ + ACC in Russian, e.g. v to vremya ‘at that time’. The instrumental time
nouns apparently must be lexically marked, even though punctuality is a case
function.
Under most current grammatical theories, lexical selection occurs prior to
agreement operations and the amalgamation of functional categories under
INFL. If derivation is a lexical process, inflectional operations must apply
subsequent to lexical ones. Assuming again an isomorphic relation between
form and function, it follows that inflectional markers will emerge in sur-
face structure outside all derivational markers. However, inflectional markers
occur widely inside derivational markers. For example, the derivation of verbs
by preverbs, prefixes which often share the form of an adverb or adposition,
is considered derivational, since these derivates often lexicalize semantically.
In English these derivations are marked with discontinuous morphemes: for
example, bring (someone) around. In Sanskrit, however, similar derivations
prefix the base: for example, pari=wayat, literally ‘around he.leads’, the present
active for ‘he marries’. The imperfect is derived by inserting a marker between
46 Robert Beard

the idiomatized prefix and stem: that is, pary = a-wayat. Georgian exhibits a
similar tendency: for example, mo=g-k.lav-s Preverb=2Obj-KILL-3Sub ‘He will
kill you’.
The third entailment of lexicalism, that derivation changes the category of
a stem while inflection does not, also faces a variety of problems. The first is
a practical one: a dearth of research on lexical and grammatical categories.
Whether N, V, A, for example, are lexical or syntactic categories has never
been resolved. It has been common to presume that they are both and to
ignore the fact that this presumption violates the strict modularity of lexicalism.
Assuming that these categories are lexical, they are not changed by deriva-
tions like violin : violinist, cream : creamery, zip : unzip. A diminutive does not
alter the referential category of its base, even though it changes its sense, very
much as does inflection. Thus Russian dozd’ ‘rain’ : dozd-ik ‘a little rain’ : dozd-
ic-ek ‘a tiny little rain’ – all refer to rain, even though they might express
varying judgments and attitudes of the speaker towards a particular instance
of rain.
There are also ostensible inflectional functions which belong to categories
other than that of the base. Participles like English talking and raked, for in-
stance, freely reflect the inflectional categories of aspect, tense, and voice, as
in John is talking and the leaves have been raked. They also serve the relational
adjectival function of attribution – for example, the talking boy, the raked leaves
– and agree adjectivally in languages requiring agreement – for example,
Russian govorjasc-ij mal’cik ‘talking boy’, but govorjasc-aja devuska ‘talking girl’.
The diagnostics of lexicalism, therefore, remain fragile until contradictions like
these are resolved. Nonetheless, an intelligible picture of derivation emerges
from the data underlying them.

3 The nature of derivation

Three accounts of derivation have emerged in the recent literature.3 The first
considers derivation simply a matter of lexical selection, the selection of an
affix and copying it into a word-level structure. Others see derivation as an
operation or set of operations in the same sense that Matthews and Anderson
see inflection. A derivational morpheme on this view is not an object selected,
but the processes of inserting or reduplicating affixes, vocalic apophony, etc.
Finally, Jackendoff and Bybee argue that derivation is a set of static paradig-
matic lexical relations. In light of the lack of agreement on the subject, a brief
examination of each of these three accounts would seem appropriate.
It is common to assume that the lexical entries (lexemes) upon which deriva-
tional rules operate comprise at least three types of features: a phonological
matrix, a grammatical subcategorization frame, and a semantic interpretation,
all mutually implied. For future reference, let us illustrate these relations with
the hypothetical entry for English health in (1).
 Derivation 47

(1) /hgl0/ Phonological representation

↑
↓

 +N 
 –Animate  Grammatical representation
 –sg., –pl. 
↑
↓
[HEALTH(X)] Semantic representation

There is general agreement on these three constituents of a lexical representa-

tion, and that they mutually imply each other in the Saussurean sense; that is,
no one such representation occurs without the other two, as indicated by the
double-headed arrows in (1). Current disagreement centers on whether lexemes
comprise only open open-class morphemes (N, A, V stems) or whether they
include grammatical (functional) morphemes as well. We will return to this
issue further on.

3.1 Derivation as lexical selection

Advocates of Word Syntax, including Selkirk (1982), Lieber (1981, 1992), Scalise
(1984), and Sproat (1985), reduce derivation to the selection of an affix from
the lexicon (see Toman, Word Syntax). This particular view of derivation is
dependent upon the existence of word-internal hierarchical structure: that is,
below the X 0 level. Lieber (1992) claims that this structure in no way differs
from syntactic structure, so that words contain specifiers, heads, and comple-
ments, just as do clauses. If words contain their own structure, and if affixes
are regular lexical entries like stems, then derivation, compounding, and regu-
lar lexical selection may all be accomplished by a single process: lexical selec-
tion. (2) illustrates how compounds and derivations might share the same
structure.

(2) (a) N (b) A (c) N

N A

bread winn er un health y draw bridge

+N +V +N +A +N +A +V +N

Derivational affixes are not distinguished from stems, but share the same
classification, morpheme, defined as a classical linguistic sign. That is, derivational
48 Robert Beard

morphemes have the same mutually implied phonological, grammatical, and

semantic representations as do lexemes. According to Lieber, the grammat-
ical representation contains the category and subcategorization of the affix,
plus any diacritics, such as its Level Order, the level at which an affix applies
under Lexical Phonology (Kiparsky 1982b). The semantic representations of
the stems and affixes in (2), for example, compose under the scope conditions
provided by the structural hierarchy and the head-dominance principles. In
(2), the rightmost lexical item dominates and assigns the grammatical and
semantic categories to the derivate or compound, as indicated by the boldface
branches. The simplicity of the Word Syntax theory of derivation is achieved
by the assumption that affixes are regular lexical items, and as such may serve
as heads of derivates. However, morphology involves far more types of mark-
ing than simple affixation, and most of these types represent problems for
Word Syntax.

3.2 Derivation as morphological operations

Anderson (1992), Aronoff (1976, 1994), and Beard (1981) have extended the
notion of grammatical morphemes as operations developed in Matthews’s
Word-and-Paradigm (see Stump, Inflection) theory to derivation. Process
morphology addresses first and foremost those types of morphology other
than external affixation. For example, both inflectional and derivational mor-
phology are characterized by reduplication. Reduplication is a process which
copies all or part of the phonological representation of a stem as an affix: for
example, the Dakota de-adjectival verbalization: puza ‘dry’ : puspuza ‘be dry’,
c hepa ‘fat’ : c hepc hepa ‘be fat’. Notice that reduplication presupposes the prior
existence of some lexeme, making it difficult to classify this process as a lexical
item as Marantz (1982) proposes. Whatever reduplication is, it must take place
subsequent to lexical selection, and hence cannot be accounted for by lexical
selection itself, unless that process is enhanced in an ad hoc fashion.
In addition to external affixation, languages also widely exhibit infixation.
The inchoative de-adjectival verb in Tagalog infixes the base; for example, ganda
‘beautiful’ : gumanda ‘become beautiful’, gising ‘awake’ : gumising ‘awaken’. Pro-
cessual morphology handles infixation with the same sort of rules employed in
accounting for external affixation. Structures like (2) cannot adequately explain
infixation without special phonological rules which determine the position
of infixes but not prefixes and suffixes. The issue between Word Syntax and
process morphology then reduces to the question of whether such special
operations differ qualitatively from other phonological operations.
Whether affixes are copied from stems to which they are attached, or whether
they are written external or internal to the lexical base, are matters of indif-
ference if affixation is a process, rather than the selection of a lexeme. This
interpretation of derivation distinguishes operations on the grammatical rep-
resentation of the lexical base from phonological modifications of the base
 Derivation 49

such as affixation. (3) illustrates how affixation is realized on the derived base
for unhealthy on this hypothesis.

(3) un health y Phonological representation

 +Neg 
 +Adj   Realization Rules
 +Poss(Y)  
 +N  Grammatical representation

Affixation applies after morpholexical and morphosyntactic rules have provided

the base with derivational features. Since no grammatical or semantic operations
are involved, affixation becomes a set of purely phonological modifications of
the phonological representation of the base conditioned by the grammatical
features. The head of such derivations is the lexical base. The crucial factor
determining the order of affixes is not structural relations, but the order in which
they are attached. Scope relations are determined by autonomous semantic
operations which follow the order of grammatical features in the base.

3.3 Derivation as lexical relations

Jackendoff (1975) and Bybee (1988) have argued that derivation is simply a
static set of lexical relations. Jackendoff argued that all derivates must be listed
in the lexicon since they are subject to lexicalization. Derivational rules are
redundancy rules, rules which state the single redundant relation “is lexically
related.” The nominalization rule for assigning -ion to Latinate verbs would
then have the form (4):

(4)  /y+ ion/  ↔  /y/ 

 +N   +V 
Separate semantic rules are similar in that they express the same redundancy
relation between the meanings of the base and the affix.

(5)  +N 
 +[NP1’s ((P)NP2)]   +V 
 abstract result of  ↔  +[NP1 ((P)NP2)] 
 act of NP1’s Z -ing   NP1 Z NP2 
 NP2 
Jackendoff proposed that such rules as (4) and (5) could be applied generatively
in speech to create neologisms; however, generation is not their purpose in the
competence model. Jackendoff also left open the question of how such regu-
larities arise in the lexicon in the first place if they are lexically superfluous.
Bybee offers a psychological answer to that question.
50 Robert Beard

Bybee argues for a connectionist theory of morphology, inflectional and

derivational, based on the theory of parallel distributed processing by Rumelhart
and McClelland (1986). In her view, lexical rules have no status “independ-
ent of the lexical items to which they are applicable. Rather, rules are highly
reinforced representational patterns or schemas.” Schemas are abstractions
from memorized lexical items which share semantic or phonological propert-
ies. One such schema results from the association of verb pairs like cling : clung,
sling : slung, sting : stung.4 A derivation rule on Bybee’s account is simply a
relationship which is more strongly represented, where “strongly” refers to the
number of representations a pattern has in long-term memory. In the instance
just cited, the phonological relation /iN/ : /vN/ is more strongly represented
than /kl/ : /sl/ or /sl/ : /st/. The more recurrent phonological relation is
therefore more likely to be associated with the past tense than the less fre-
quent ones.
When speakers add the past tense innovatively, they simply search their
memories for phonological relations associated with the past tense and choose
one analogically. Following recent connectionist theories, the most highly re-
inforced relation is most likely to be selected for the neologism. The relation
/iN/ : /vN/, for instance, will not be as strongly represented in the semantic
schema for past tense as ø : /d/. Speakers are therefore more likely to add
/d/ to a neologism than to replace a stem vowel /i/ with /v/. If the neo-
logism ends in /N/, however, the probability that this method will be selected
increases.
Bybee’s suggestion has the advantage of conflating derivation and deriva-
tional acquisition. A derivational rule reduces to the arrangement of memor-
ized items in mental storage. Without derivation rules, all morphology may be
confined to the lexicon as in Word Syntax, and only one rule, lexical selection,
is required to account for morphology in syntax. Moreover, morphological
creativity reduces to the general cognitive process of analogy which is com-
monly used in categorization. So far, however, many of the processes vital to
Bybee’s model remain undefined, so it is not currently possible to determine
this theory’s efficacy in accounting for the derivational data.

4 Derivational heads

If affixes are regular lexical items which may be selected for word structures
as fully derived words are selected for phrase structures, they should be able
to serve as heads, as do fully derived words. If affixes are the results of pro-
cesses, however, they cannot be lexical heads, and the traditional assumption
that stems represent morphological heads regains credibility. This issue has
been a central concern of recent morphological research, so is next on the
agenda.
 Derivation 51

4.1 Affixes as heads

If derived words are structured, the question naturally arises as to whether

word structure is the same as syntactic structure. Lieber and Sproat claim
that not only are the two types of structure identical, but the principles for
composing words are precisely those of X-bar syntax. It follows that mor-
phology may be dispensed with altogether, resulting in yet another major
theoretical economy under Word Syntax. A major contention of modern X-bar
theory is that the head of a phrase (X) determines the category of the whole
phrase (XP). A sound test of Word Syntax, therefore, is whether the head of
a derived word determines the category of the whole word. Since the outer-
most affix of a word is often associated with the category of the whole word,
it might be possible to mount a case for affixes serving as the heads of derived
words.
E. Williams (1981b) advanced the simplest account of affixes as heads of
words: the head of a word is its rightmost element. Thus the head of bread-
winner in (2a) would be -er which, under the premise that affixes are lexical
items, is a noun in the same sense that bridge in drawbridge (2c) is a noun. Both
-er and bridge are nouns which determine the category of bread-winner and
drawbridge. The heads of redraw and unhealthy (2b), on the other hand, are the
bases draw and healthy, since prefixes in IE languages tend not to change the
category of the derivates to which they adhere.
Some features, however, must be raised from nonheads. Diminutives, for
example, usually bear the features of the base rather than the affix. In Russian,
for example, both sobaka ‘dog’ and its diminutive, sobac-k-a, are feminine; jazyk
‘tongue’ and its diminutive, jazyc-ok, are both masculine. This contrasts with
German diminutives, which are all consistently neuter: for example, der Brief :
das Briefchen ‘letter’, die Lampe : das Lämpchen ‘lamp’. To redress this problem,
Di Sciullo and Williams (1987) proposed that feature inheritance relativizes
the head; that is, features of categories present in the stem but not in the affix
determine the lexical categorization of the final derived word. This new vari-
ation presumes that affixes, like Russian diminutive suffixes, are unmarked
for certain features such as gender; this allows gender features from the next
highest node to be inherited by the derivate. The German suffix -chen, on the
other hand, does bear an inheritable gender valuation, neuter, and so passes
this feature on to the derivate.
Unfortunately, relativizing morphological heads renders them radically dif-
ferent from phrasal heads, which are always absolute and never relative.
Derived words differ greatly from derived phrases, where face is just as good
a noun phrase as a strange face peering through the door. *Ist is not just as good
a noun as violinist. Relativizing morphological heads then defeats the original
purpose of postulating affixal heads. This difference between word and phrase
heads nonetheless must be characterized in an adequate model of grammar,
even though it impedes the reduction of morphology to syntax.
52 Robert Beard

4.2 Head operations

There is another clue to the question of morphological heads. The phonolo-

gical structures of a wide range of derivations do not isomorphically parallel
their semantic structures. English, for example, restricts the comparative suf-
fix -er to monosyllabic adjectives or disyllabic stems ending on a weak vowel:
for example, quick : quicker, hateful : *hatefuller but happy : happier (see Sproat,
Morphology as Component or Module, for further details). Trisyllabic stems
are wholly excluded from the distribution of this suffix, with one exception:
disyllabic stems prefixed with un-: unhappier. This exception is obviated on the
assumption that -er attaches to happy before un-; however, the semantic read-
ing of such terms is not ‘not happier’ but ‘more unhappy’. The morphological
and semantic structures of such forms are hence “mismatched.”
To circumvent exceptional treatment of such morphosemantic mismatches,
Hoeksema (1985) proposed that every rule of derivation has a correlate that
applies specifically to heads, but is in all other respects a context-free rewrite
rule. Stump (1991) argues that this correlate is the default. English derived verbs
exhibit the effect of a head operation in maintaining their conjugations even
when serving as a base in a derivate. The past tense of understand is understood,
and that of overdrive is overdrove. This seems to indicate that although past tense
has scope over the entire derivation (or compound) in these instances, mor-
phology applies strictly to stand and drive, respectively; otherwise, we might
expect the past tense to be the productive *understanded and *overdrived.
Morphosemantic mismatches like unhappier are susceptible to the same
interpretation; the morphology of the negative adjective applies to the head of
the derivate, happy, even though the scope of comparison extends to the entire
word unhappy. Head operations may also be extended to instances of inflec-
tion occurring inside derivation, such as the Sanskrit perfect mentioned above,
pari=wayat ‘he marries’ : pary=a-wayat ‘he married’, and to diminutives like the
Hebrew loan xaxom-im-l-ex ‘smart little people’ in Yiddish and Breton bag-où-
ig-où ‘little boats’ (Stump 1991), assuming that diminutive suffixes are gram-
matically empty and that the stem is the head. On this account the Sanskrit
perfect inflection is added to the head (stem) inside the preverb, because the
preverb is a phrasal head clitic or semi-discontinuous morpheme. The Hebrew
and Breton plural mark both their scope and the head of the derivation, for
reasons undetermined.
Head operations remain exceptional so long as affixes may be heads, since
semantic evidence indicates that affixes are never themselves affixed. No lan-
guage exhibits scope ambiguities such that the plural of a locative nominal-
ization like bakeries would refer either to an aggregate of places, only some
of which are devoted to baking (head marking, the affix the head), or to an
aggregate of baking places (scope marking). The scope of all derivational func-
tions is the entire word to which it is added, derived or underived; the only
variation is in the placement of affixes marking them. This situation, combined
 Derivation 53

with the failure of theories of affixal heads, endorses the traditional assump-
tion that the morphological head of a word is its root or stem. Morpholexical
and morphosyntactic feature operations seem to apply concatenatively to the
base lexeme; the distribution of affixes, on the other hand, seems to be deter-
mined by language-dependent rules of spellout.

5 Synthetic compounds and derivation

If affixes are not morphological heads, the question arises as to whether

compounds and derivations are at all related as (2) implies. If they are, it is
doubtful that their relation is structural. It is common to distinguish analytic
from synthetic compounds by the presence of affixation. Drawbridge, redhead,
houseboat are thereby analytic compounds, while truck-driver, truck-driving, red-
headed are all synthetic. There is little evidence that most analytical compounds
are related to derivation. Rather, other analytic compounding appears to be a
simple process of combining lexemes. The head of those compounds composed
of constituents belonging to different categories determines the category of
the compound. The right constituent determines the category of English com-
pounds, so that a houseboat is a boat while a boathouse is a house. However,
this description excludes prepositions, since compounds with prepositional
modifiers are often adjectives (inland, between-class, outboard), and those with
prepositional heads may be anything but prepositions (sit-in, hold-out, run-
away). Even most P + P compounds fail the head test: without does not imply
out, and in and on would seem to be the heads of into and onto, respectively.
The evidence from compounding hence suggests that adpositions are not
lexemes in the sense that N, V, A stems are.
The distinction between analytic and synthetic compounds is nevertheless
imperfect at best. Synthetic compounds do resemble simple derivations in
several respects. For example, they share the same derivational categories
often marked by the same affixes: bearded : gray-bearded, driver : truck-driver,
driving : truck-driving. Analytic Bahuvrihi compounds, like redhead, long-hair,
hardhead, for instance, share their derivational function with possessional adject-
ives like red-headed, long-haired, hardheaded. Indeed, the same possessional func-
tion (“having N”) emerges in simple derivations like headed and hairy. Parallels
like these suggest that synthetic compounding is derivation which allows an
optional modifier. Affixation, however, is not a reliable indicator of the distinc-
tion between compounds and derivations, since the zero morphology, which is
available to simple derivations, is also available to compounds. Indeed, Booij
(1988) has shown that synthetic compounds, whose structure is presumed to
be [truck-driv]ing, may be explained with equal cogency as analytic compounds
with the structure [truck][driving], given a semantic level capable of resolving
morphosemantic mismatches.
54 Robert Beard

Until research better clarifies the subject, it is best to assume that analytic
compounds represent an independent lexical means of derivation; however,
it is doubtful that those bearing adpositions are compounds (see also Fabb,
Compounding). Analytic compounds in this sense should not be confused
with zero-marked Bahuvrihi compounds. Like synthetic compounds, bahuvrihis
may be interpreted as derivations with optional modifiers. This area of research
is very fluid, however, and Booij has shown how all compounds may be reduced
to analytic concatenation.
Morphosemantic mismatches raise another important issue in morpho-
logy: the fact that derivational meaning and the affixation marking it are not
always isomorphic. Karcevskij (1929) called this phenomenon morphological
asymmetry. It is an attribute of morphology whose importance is only now
being appreciated.

6 Morphological asymmetry

Karcevskij noted that while several endings mark the genitive in Russian –
-i, -a, -u – each of these endings also has multiple functions. The ending -a, for
example, also marks feminine nominative singular and neuter plural. The end-
ing -i marks feminine and masculine nominative plural, as well as genitive,
dative, and locative singular in declension III. In other words, it is common for
grammatical morphemes to be cofunctional (-i, -a, -u above) and multifunctional
(-i), to use the terms of Szymanek (1989). In addition to cofunctionality and
multifunctionality, Matthews (1972) identified extended and cumulative exponence
as morphological asymmetries. In the Latin word rCxistD [re:k-s-is-ti:] ‘you (sg.)
ruled’, for example, the suffix -ti: cumulatively (simultaneously) marks second
person, singular, and perfective. The remaining markers, -s and -is are empty
extensions of -ti, redundantly marking the perfective, too. The same phe-
nomena characterize derivation. In the adjective dram-at-ic-al, -at and -al are
empty extensions of -ic; cf. theatr-ic. The German suffix -er in Lehr-er ‘teacher’
cumulatively marks [+subjective], [+masculine], and [declension I]. Finally,
zero (null) morphology reflects morphological asymmetry. While most non-
count modalic (instrumental) nominals require either the suffix -er (conditioner,
softener) or -fnt (stimulant, relaxant), many require no suffix at all: for example,
a rinse, a wash, a spray. Again, the relationship between the grammatical and
phonological levels is nonisomorphic.
Bazell (1949, 1952) argued that these phenomena collectively indicate a fault
in structuralist morphology, which he dubbed the Correspondence Fallacy, the
assumption that an analysis at one linguistic level will isomorphically map
onto analyses at other levels. Bazell argued that the phonological analysis of a
word need not correspond to its semantic analysis; however, it does not follow
from this that no analysis is possible. It is quite conceivable that each level is
defined in its own terms, and that mapping from one level to another involves
 Derivation 55

more sophisticated relations than the isomorphic relation of the classical lin-
guistic sign.
To obviate the correspondence fallacy, Beard (1966, 1976), Kiefer (1970), and
Leitner (1973) proposed what was subsequently called the Separation Hypothesis,
the claim that the functional and spelling operations of derivation are dis-
crete and autonomous. The Separation Hypothesis assumes that lexical items
are restricted to N, V, and A stems, all of which are perfect signs comprising
mutually implied phonological, grammatical, and semantic representations, as
in (1). It then provides a set of abstract lexical operations on the grammatical
representation of a lexical item discrete from operations on the phonological
and semantic representations. Algorithms in an autonomous morphological
spelling component like those proposed by Matthews (1972) then modify the
phonological representation of grammatically and semantically derived stems.
By the same token, compounding operations which combine words like truck
and driving mentioned above need not establish the semantic scope of com-
pound constituents. This can be accomplished by autonomous principles of
composition based on the argument structure of the phrasal head, in this case,
drive.
The separation of grammatical and phonological operations allows for a
simple account of all morphological asymmetry. Cumulative exponence results
from a single-stem modification conditioned by several grammatical features,
while extended exponence is the collective marking by several stem modifica-
tions of a single feature. Cofunctionality and multifunctionality are explained
similarly. Finally, zero morphology is simply derivation without affixation,
while empty morphemes result from affixation without derivation.
Morphosemantic mismatches like those in unhappier, those in compounds
like truck-driving on Booij’s interpretation, and those in head operation con-
structions may be resolved by a similar separation of derivation and semantic
composition. The asymmetry explored by Karcevskij and Matthews, on the
other hand, is more a morphophonological mismatch between derivation and
phonological realization. The ultimate implication of asymmetry, therefore, is
that semantics, derivation, and affixation represent three distinct levels of
morphological operations, which require two distinct mapping systems.

7 Types of derivation

We have surveyed the general attributes of derivation and the major accounts
of them. We may now turn to the particular properties of derivation: the types
of derivation and the types of affixation marking them. In its broadest sense,
derivation refers to any process which results in the creation of a new word.
However, the output of some morphological operations is far more principled
than the output of others. The derivations in (6), for example, form a sort of
lexical paradigm which holds for many other bases:
56 Robert Beard

(6)  re-laser, out-laser, over-laser, . . .

 laser-er → laser-er-s, laser-er-’s
Laser  laser-ing → laser-ing-s, laser-ing-’s
 (un)laser-able → (un)laser-abil-ity

Some types of derivation do not fit into derivational paradigms like (6). It
is well known that words may be misanalyzed when a phonological sequence
identical with that of an affix is misperceived as that affix. The result is that
a previously nonexistent underlying base is extracted and added to the perman-
ent lexical store via a process known as back formation. Sculptor, for example,
was borrowed as an integral base into English. However, because the final
phoneme cluster /fr/ is identical with an agentive marker in English, and since
sculptor is an agentive noun, a verbal base, to sculpt, has been extracted and
added to the stock of English verbs. Consequently, sculptor changes from a
lexical base to a derivate.
Several facts obstruct the conclusion that back formation is a derivational
process. First, in order to use back-formed words, we must be familiar with
them. While some potentially back-formed words are used, far more may not
be. It is not possible, for instance, to say that a butcher *butches or that a barber
*barbs, even though these verbs are potential back derivates as legitimate as
sculpt. There is no grammatically definable constraint preventing this; it simply
is not acceptable to do so. Second, we do not find positions for back derivates
in lexical paradigms like (6). Take the back derivate of laser itself: to lase, for
example. It generates exactly the same paradigm as (6). Thus, in those dialects
which use lase, one may say relase, outlase, overlase, laser, lasing, (un)lasable,
(un)lasability – all with the same sense as the corresponding zero-derived verb
in (6). In other words, rather than forming a derivational relationship with a
lexical base, back-formed words create a new base, expanding the underived
lexical stock in a way that regular derivations do not. This characterization
partially fits several other types of word formation which need to be distin-
guished from regular, grammatically determined derivation.

7.1 Lexical stock expansion

Clipping (telephone : phone), blends (smoke + fog = smog), acronymization (aids), and
analogical formation (workaholic) all conform to the description of back forma-
tion in significant ways. Back formation generates a base which the lexicon
lacks. Clipping, on the other hand, produces a redundant base, but a new one
all the same. With rare exceptions (e.g. caravan : van), the input and output of
clipping rules are semantically identical, and both remain active in the lexicon.
Both telephone and phone have the same range of grammatical derivations, all
with the same meaning: (tele)phoner, (tele)phoning, etc. Notice, too, the irregu-
larity of clipping. It usually reduces a polysyllabic word to a monosyllabic
one; however, this may be accomplished by removing the initial syllables
(phone), the final syllables (rep), or the initial and final syllables ( flu).
 Derivation 57

Blending, acronymization, and analogical formation also tend to be con-

scious operations, unlike grammatical derivation. Words like smog, motel, and
tangelo are created intentionally by a logical rather than grammatical process:
if the reference is part A and part B, then the word referring to it should com-
prise parts of the words for A and B. Acronyms like laser, scuba, aids, have
been converted from phrases to the initial letters of the words in those phrases,
which are not part of grammar, then the initials have been phonologically
interpreted. Acquired immune deficiency syndrome, for example, provides aids,
which is rendered pronounceable by applying English spelling rules in reverse.
The process hence requires considerable conscious activity outside the bounds
of grammar. As in the case of clipping, the phrase and the acronym are syn-
onymous, and both remain in the language.
Analogical forms like workaholic, chocaholic and cheeseburger, fishburger, chicken-
burger differ from regular derivations in that they require prosodic identity.
Genuine suffixes like -ing may be added to stems of any length or prosodic
structure. Pseudo-derivates like chocaholic, however, must additionally fit the
prosodic template of their analog, in this case, alcoholic: the output must con-
tain four syllables with penultimate accent. Thus chocolaholic, shoppingaholic, and
handiworkaholic do not work as well as chocaholic, shopaholic, and workaholic.
When we begin to find acceptable violations of this extragrammatical prin-
ciple like chickenburger, we usually find that the remainder, in this case burger,
has become an independent back-formed word capable of undergoing regular
compounding.
This does not exhaust the catalogue of lexical stock expansion processes. That
catalogue also contains borrowing (troika, detente, thug), commonization (quisling,
aspirin), semantic narrowing (percolator, escalator), loan translation (German Einfluß
‘influence’, Mitleid ‘compassion’), folk etymology (craw[l]fish from Old French
crevise), and perhaps others. The point is that these processes tend to be con-
scious, extragrammatical, and hence grammatically irregular. Rather than filling
a position in some lexical paradigm, they create new lexical bases which then
generate their own paradigms. To better understand the difference, let us now
examine the regular derivation types.

7.2 Lexical derivation

Four distinct types of regular grammatical derivation have been described in
the literature; featural derivation, functional derivation, transposition, and express-
ive derivation. While all the details of the properties of these types of derivation
and their interrelations have not been refined, their basic nature and functions
may be broadly described.

7.2.1 Featural derivation Featural derivation does not change the category
of the underlying base, but operates on the values of inherent features. An
obvious candidate for such a rule is natural gender, as described by Jakobson
58 Robert Beard

(1932, 1939) in connection with his concept markedness.5 In most languages

which support natural gender, the default or unmarked form is masculine. A
convenient technical notation of the fact that unmarked masculine nouns may
refer to males or females is [+Feminine, +Masculine]. This requires a Jakobsonian
principle of markedness whereby in cases of conflict, the surface realization
will default to that of the unmarked category, masculine. Thus the Russian noun
student ‘student’ may refer to females or males, but all grammatical agreement
will be the same as purely masculine nouns like brat ‘brother’, otec ‘father’,
which cannot refer to females.
Default masculines like student differ from pure masculines in that they are
susceptible to feminization. This requires some rule on the order of student(ø)
→ student-k(a) which converts the default masculine noun into a purely femin-
ine one and marks this fact by transferring the base from declension I, marked
in the nominative by -ø, to declension II, marked in the nominative by -a. All
that is required grammatically and semantically of this rule is the toggling of
the masculine feature from positive to negative: that is:

(7) [+Feminine, +Masculine] → [+Feminine, −Masculine]

(7) converts the lexical description of the base from unmarked masculine to
marked feminine like the purely feminine nouns sestra ‘sister’ and mat’ ‘mother’,
which may refer only to females and not to males. The addition of any feature
[+Feminine] would be inappropriate since (7) applies only to nouns with nat-
ural gender: that is, those which inherently (lexically) possess lexical gender
features.

7.2.2 Functional derivation Kury4owicz (1936) first distinguished rules

which add features to the underlying base from those which merely change its
category. Consider (8), for example:

(8) (a) recruit : recruit-er

(b) recruit : recruit-ee
(c) bake : bak-ery

Kury4owicz referred to this type of derivation as “dérivation lexicale,” because

the derivate differed semantically from its base. In the middle of this century,
several European linguists, among them Belić (1958: 140–1, 148–50), noted
a similarity between the functions of these derivations and those of the case
system. The traditional names of nominals like recruiter and recruitee, “agentive”
and “patientive,” suggest that the functions of this type of derivation are
semantic. However, many “agentive” forms are not animate as the name implies
(breaker, floater, sparkler), and many are not even active (riser, marker, divider).
The same is true of patientives: alongside employee, recruitee, draftee, we find
inanimate “patientives” of resultative verbs like painting, carving, writing. The
suffixal distinction does not change the fact that a painting is an object painted
 Derivation 59

in just the same sense that an employee is an object employed. It therefore

seems more likely that this type of derivation is based on case functions: for
example (nominative of) subject, (accusative of) object, (locative of) place
(bakery, fishery), (genitive of) possession (dirty, forested) and material (oaken,
woolen), (ablative of) origin (American, Brazilian), (dative of) purpose (roofing,
siding), (instrumental of) means (cutter, defoliant).
Languages with rich morphologies have dozens of such derivations, includ-
ing those just mentioned. Even in Serbian and Polish all these derivations are
still quite productive, and all their functions serve as pure case functions ex-
pressed without adpositions in some language. Basque has a locative of locus,
mendi-an = mountain-Loc ‘on the mountain’; Serbian exhibits the possessional
(qualitative) genitive: covek plav-ih oci-ju ‘a man of (with) blue eyes’; Turkish
marks origin and material with the ablative, Ankara-dan = Ankara-Abl ‘from
Ankara’, ta/-tan bir ev = stone-Abl one house ‘a house of stone’; and Latin has
a dative of purpose: castr-Ds locum = camp-DatSg place-AccSg ‘a place for a
camp’. All languages express these functions with case endings, adpositions,
or a combination of both. Few verbs and no nouns are subcategorized for
these argument relations, yet they are widely available to functional lexical
derivation in languages with rich morphological systems like Serbian, Inuit,
and Chukchee.6
If the ultimate constraint on functional derivations is the set of case func-
tions, the question becomes why some functions seem to be missing and why
subject and object relations are more productive and diachronically stable than
others. Some omissions are obvious: the (ablative) absolute relation is missing
because it is purely a syntactic relation, that of sentential adverbs; the same
applies to the (ablative of) distinction, used to mark comparatives in many
languages: for example, Turkish Halil’-den tembel = Halil-Abl lazy ‘lazier than
Halil’. The reason why we find more subjective and objective nominalizations
than others is, no doubt, high pragmatic demand. This is an area which has
received little attention historically, and thus no definitive answers to these
questions are available.7 However, it is clear that functional derivations involve
far more functions than the argument functions found in the base, yet few if
any productive derivational functions fall outside those found in the inflec-
tional system.

7.2.3 Transposition Another type of derivation which reflects a simple

change of category without any functional change is transposition, illustrated
in (9):

(9) (a) walk : walk-ing (V → N)

(b) new : new-ness (A → N)
(c) budget : budget-ary (N → A)

Kury4owicz called derivations like those of (9) “dérivation syntaxique,” but

Marchand (1967) used the more distinctive term, “transposition.”
60 Robert Beard

Transposition introduces no argument structure, but simply shifts a stem

from one category to another, sometimes marking the fact affixally, sometimes
not. The definition of dryness must coincide with that of dry in all essential
respects, since, unlike bake and baker, its reference is identical to that of its base.
The same is true of all the relations represented in (9). Whether transpositions
are marked by real or zero affixation is a separate issue, bound up with the
general issue of the nature of zero morphology.

7.2.4 Expressive derivation Expressive derivation does not change the

referential scope of its input; however, expressive derivation also does not
change the lexical category of the base. As mentioned in section 2, the refer-
ence of the three grades of the Russian word for ‘rain’, dozd’, dozdik, and dozd-
ic-ek, all refer to the same conceptual category. The formal variation reflects
subjective perceptions of the speaker: whether he perceives the rain to be
relatively light, beneficial, or pleasant. For this reason, expressive derivation
may be recursive, applying to its own output as in the Russian example. In
addition to diminutive and augmentative expressive derivation, pejorative and
affectionate forms also occur: for example, Russian kniga : knizonka ‘damned
book’ (cf. knizka ‘little book’) and papa ‘daddy’ : papocka.
There is no obvious means of relegating expressive derivation to any of the
other three types. The categories involved are not found elsewhere in gram-
mar as are functional categories, nor are they inherent lexical categories like
gender. Since expressive derivation does not involve a category change, it
cannot be a form of transposition. It therefore remains mysterious in many
respects.

8 Realization and productivity

The types of phonological realization (stem modification) which express deriva-

tion are by and large the same as those which express inflection. The glaring
exception seems to be that derivation is not expressed by free morphemes: those
which are not modifications of stems, but which stand alone. This would
follow from the assumption that only inflection is syntactic. Since free mor-
phemes require a structural position, this type of realization would be ruled
out for lexical derivation by the Lexicalist Hypothesis.
Evidence indicates, however, that the bound phonological realization of
derivational and inflectional morphology is provided by a single component
(see also the discussion of the Split Morphology Hypothesis in Stump, Inflec-
tion). The English suffix -ing, for example, may be attached to verb stems to
generate inflectional forms like the progressive (is painting), the present parti-
ciple (painting machine), as well as derivational forms like the objective nominal-
ization (a painting). The same is true of -ed, which productively marks the past
 Derivation 61

tense and participles (John (has) annoyed Mary), as well as derivations like the
possessional adjective: for example, two-headed, forested. The important point
is that derivation seems to be an abstract process independent of the various
means of phonological realization and of the means of semantic interpretation.
Two specific types of marking, subtraction and metathesis, weakly represented
in inflection, apparently do not mark derivation. Papago, for example, seems
to derive perfective verbs from imperfective ones by deleting the final con-
sonant if there is one, him ‘walking’ : hi: ‘walked (sg.)’, hihim ‘walking (pl.)’ : hihi
‘walked (pl.)’ (Anderson 1992). However, aspect is probably inflectional, though
the matter remains unclear. Metathesis for the most part is an allomorphic
change effected by affixation, as in the case of the Hebrew reflexive prefix,
hit-, whose final segment metathesizes with initial voiceless stridents: for
example, xipes ‘seek’ : hitxapes, but silek ‘remove’ histalek.

8.1 Discontinuous morphemes

Evidence suggests that one type of verbal derivation is marked by discontinuous

morphemes, morphemes which may be loosened or removed from their base.
The English correlate of preverbs, a type of verbal prefix expressing a closed
set of adverbal functions, is a particle which is written separately and may
appear either immediately following the verb or the VP. Consider the Russian
examples and their English counterparts in (10) for example:

(10) (a) Ivan vy-vel sobaku ‘John brought [out] his dog [out]’
(b) Ivan v-vel sobaku ‘John brought [in] his dog [in]’
(c) Ivan so-stavil plan ‘John put [together] a plan [together]’

Because verbs with preverbs form notoriously irregular patterns and are equally
notorious for idiomatizing (e.g. pri ‘to’ + pisat’ ‘write’ = pripisat’ ‘attribute’), they
are considered lexical derivates. How, then, may their markers appear a phrase
away from the stem which they mark?
Preverbs are in fact often loosely attached to their stem as the examples
above from Sanskrit (pari=a-wayat = Around=Imp-lead ‘he married’) and Geor-
gian (mo=g-k.lav-s Prvb=2Obj-KILL-3Sub ‘He will kill you’) illustrate. These
preverbs attach to the outside of the fully inflected verb, the head of the VP.
One possible account of these morphemes is that they are clitics, defined in
terms of attachment to either the phrasal head or periphery, depending on the
morphological conditions of specific languages. The important point is that
their position is morphologically predictable by Anderson’s general theory of
affixation (see affixation in section 8.2.1 below), and requires no syntactic pro-
jection as do lexemes and free morphemes. Hence it is possible to explain these
derivations without violating the Lexicalist Hypothesis, given the Separation
Hypothesis.
62 Robert Beard

8.2 Other types of stem modification

8.2.1 Affixation Affixation (prefixation, suffixation, and infixation) is the
most productive means of marking derivation. Anderson (1992: ch. 8) points
out that affixation may be defined in the same terms as cliticization, assuming
that the peripheral element of a word is its initial or final segment or syllable
and its head is the accented syllable. That is, affixes may attach only to the
inside or outside of the initial or final phoneme or syllable, or to either side
of the head, the accented syllable.8 This purely morphological definition of
affixation is far more accurate than structural descriptions, and does not
require word structure or any sort of affix movement. Circumfixation, such as
Indonesian ke . . . -an as besar ‘big’ : ke-besar-an ‘bigness, greatness’, is merely
extended exponence involving a prefix and a suffix simultaneously.

8.2.2 Apophony (stem mutation, revoweling) This type of stem modifica-

tion is well attested in Semitic languages. Lexical items in those languages com-
prise consonants only, and vowels are used to mark morphological functions.
The (Algerian) Arabic stem for ‘write’ is *ktb-, and the derivate for ‘book’ is
ktaab, while that for ‘writer’ is kaatfb. This type of morphological modification,
like subtraction and metathesis, raises the question of the limits on modification
of the phonological representation of the base: to what extent may the base be
corrupted before it becomes unrecognizable? This is another open question in
morphology.

8.2.3 Conversion Transposing a lexeme from one category to another

without affixation is sometimes called conversion. The evidence weighs against
a separate operation of conversion, however, for we find precisely the same
semantic relations between conversional pairs as between derivational pairs.
Thus for every conversion to dry, to wet, to empty we find at least an equal
number of affixed derivates with the same relation: to shorten, to normalize, to
domesticate. Moreover, precisely those stems which affix are precluded from con-
version (to *short, *normal, *domestic), and precisely those which convert are
precluded from affixation: to *endry, *wetten, *emptify.9 The simpler account of
such forms is that those without affixation are null marked variants of the
same derivation which is otherwise marked by a variety of affixes.

8.2.4 Paradigmatic derivation A common means of marking lexical deriva-

tion is shifting the base from one nominal declension class to another, with or
without a derivational marker. Thus, in Swahili, diminutives are formed by
shifting nouns to noun class 3: for example, m-lango ‘door’ (class 2) : ki-lango
‘little door’ (class 3), m-lima ‘mountain’ : ki-lima ‘hill’. Feminine agentives in
Russian are usually derived from masculines of declension I (= noun class 1)
by adding a declension II (= noun class 2) suffix: for example, ucitel’ ‘teacher’ :
ucitel’-nic-a, where the final -a indicates declension II. However, the processes of
adding the suffix and changing the declensional paradigm must be independent,
 Derivation 63

since the latter may apply without the former: rab (masc., declension I) :
rab-a (fem., declension II) ‘slave’, suprug (masc., declension I) : suprug-a (fem.,
declension II) ‘spouse’.

8.2.5 Prosodic modification A derivational function may be marked by

simply shifting the accent of a word or modifying the intonation, perhaps a
variant of apophony. Thus, in English, it is common to indicate the objective
(resultative) nominalization by shifting accent from the stem to the prefix: for
example, survéy : súrvey, suspéct : súspect. The process is productive with verbs
prefixed by re-: rewríte : réwrite, remáke : rémake. The morpheme here seems to
be the process of shifting the accent from one syllable to another.

8.2.6 Reduplication In addition to attaching phonologically specified affixes

to a stem, derivation is often marked by the full or partial reduplication of a
part of the stem attached to it. Indonesian forms adverbs from all categories
by completely reduplicating them: kira ‘guess’ : kira-kira ‘approximately’, pagi
‘morning’ : pagi-pagi ‘in the morning’. Reduplication may be combined with vari-
ous types of affixation as in Indonesian anak ‘child’ : ke-anak-anak-an ‘childish’.

8.3 Productivity and allomorphic variation

Proponents of Natural Morphology (NM) have long noted that not all the modes
of stem modification surveyed in the previous section are equally productive
(Dressler et al. 1987); some means of morphological marking are more pro-
ductive than others. Aronoff (1976) first noted that affixes such as the English
suffixes -ing and -ness (e.g. deriding, kindness), which are transparent, in that
they involve no allomorphy, tend to be more productive and more predictable
than those which do induce allomorphy: for example, -ion and -ity (e.g. deride :
derision, but ride : *rision; curious : curiosity, but spurious : *spuriosity).
NM argues that the isomorphic linguistic sign is the linguistic ideal, and
that the further a morpheme deviates from this ideal, the more difficult it is
for languages to sustain it. If the subjective nominalization changes or adds
semantic material to the underlying base, it should add phonological material
to the stem iconically and transparently. English derivates like bak-er, resid-ent,
escap-ee then are more natural, and thus more likely to be productive, than
unmarkered derivates like (a) cook, guide, bore. Opaque affixes which cause or
require phonological adjustments such as the Latinate suffixes mentioned above
should be less productive inter- and intralinguistically. Zero and empty mor-
phology should be rare, and subtractive morphology nonexistent for the same
reasons.
NM offers a means of uniting Word Syntax and processual morphology.
Notice that while NM offers the isomorphic morpheme as the ideal, it impli-
citly admits the sorts of asymmetrical variations that processual morphology
is designed to explain. Processual morphology, however, holds that this ideal
is restricted to lexemes in the lexicon. Moreover, it has no inherent account of
64 Robert Beard

why transparent, symmetrical markers seem to be more productive than asym-

metrical ones. If the predictions of NM hold, they could make a major contri-
bution toward unifying the two major approaches to derivational morphology
discussed in this chapter.

9 Conclusion

Derivational morphology differs from inflectional morphology in that it pro-

vides new lexical names for objects, relations, and properties in the world.
Lexical names may be combined in syntactic constructions to generate descrip-
tions of the real world in speech, or may be used to label objects in the real
world: for example, bakery, careful, occupied, slippery, gentlemen. The grammat-
ical relations upon which derivation operates seem to be the same as those
found in inflection; the difference is that these relations hold between the
derivate and its base in derivation, rather than between two different phrases
in syntax. Consider the cookie-baker bakes cookies, for instance. The noun cookie-
baker exhibits internal subject and object relations between the derivate and its
constituent parts. The same relations hold in the sentence between the subject,
cookie-baker, the object, cookies, and the verb. The semantic difference between
the two uses is that cookie-baker is the name of a semantic category while the
sentence is a description of a specific event. We would expect inflectional means
of marking the subject and object relations in the sentence but derivational
means of marking it in the derived compound.
Evidence supports the conclusion that morpholexical derivation and
phonological realization are two discrete processes executed by autonomous
components. Lexical rules apparently provide for derivation, and either the
phonological component or a separate morphological component is respons-
ible for phonological realization. The Split Morphology Hypothesis notwith-
standing, it is possible that a single autonomous spellout component accounts
for inflectional and derivational morphological realization. If the lexicon and
syntax appropriately distinguish morpholexical features and morphosyntactic
features, derivation and inflection may be distinguished despite the fact that
both operate over the same morphological categories, realized phonologically
by the same spellout component.

NOTES

1 Bybee (1985) is one of several processes form a scale along which

morphologists to take issue with the rules are more or less derivational
Lexicalist position itself. She argues and inflectional. A rule’s position
that derivational and inflectional along this scale is determined by its
 Derivation 65

generality and relevance. Tense, for is merely the other end of the
example, is more general than the same continuum (see n. 1).
functions of prefixes like trans- and 5 Note that the following discussion
re- since it applies to all verbs. Tense, does not apply to grammatical gender,
on the other hand, is more relevant which amounts to no more than
to verbs than is person, since it lexical class (Halle 1989).
directly modifies the meaning of the 6 Only two accounts of the parallel
verb, while person simply denotes between inflectional and derivational
an argument of the verb. The less categories have been suggested.
general and more relevant the According to Botha’s Base Rule
meaning of a morphological Theory (Botha 1981, Beard 1981), both
operation, the more “derivational” lexical and syntactic rules operate on
it is; the more general and less deep structures, which must contain
relevant, the more “inflectional” these functions. Borer (1988) argues
it is. Tense by this measure is less for a single word formation
derivational than verbal prefix component which operates at two
functions, but more so than person. levels, deep and surface structure.
No strict division between the two At present it is not clear whether
may be made, however, according these two approaches differ in any
to Bybee. essentials.
2 Beard (1995: 286–9) offers a 7 Beard (1993) argues that higher
contemporary account of phrasal productivity is also facilitated by
adjectives as an extragrammatical “dual origin.” Subjective and
phenomenon. objective nominalization e.g. may
3 Chomsky’s Minimalist Program be derived either by functional
(Chomsky 1995b) as of the moment derivation or by transposition. The
provides no theory of derivation, semantic interpretation is identical in
hence any comment would be either case; however, if the desired
premature. However, since words output is blocked for any reason via
are copied from the lexicon “fully one derivational type, another means
inflected,” it would seem that of derivation, which may not be
Minimalism currently makes no blocked, is available in these cases.
distinction between either derivation 8 Clitics may be attached at correlate
and inflection or lexemes and points of a phrase, i.e. either side
grammatical morphemes. To the of the phrasal head or either side
extent that this observation is of a peripheral word or constituent
accurate, Minimalism is susceptible (see Halpern, Clitics).
to the problems with these 9 A few sporadic examples of
assumptions discussed here. derivation–“conversion” pairs may
4 Although Bybee exemplifies her be found, e.g. to clear : clarify, winter :
hypothesis with inflectional winterize. However, such pairs are
categories, she makes it clear rare and semantically unpredictable,
that she intends it to extend to and may be explained as easily in
derivation, which, in her view, terms of zero morphology.