CHAPTER4:

THE LEXICAL I DERIV ATIONAL I INFLECTIONAL CONTINUUM

1. Distinguishing derivation from inflection

One of the most persistent undefinables in morphology is the distinction

between derivational and inflectional morphology. Wh ile Iinguists seem to
have an intuitive l. nderstanding of the distinction , the objective criteria
behind this intuition have proved difficult to find. The most su∞essful criter-
ion is obligatoriness , applied to the definition of derivation and inflection by
Greenberg 1954. Obligatory categories force certain choices upon the
speaker. In English every noun phrase must be either Definite or Indefin饨，
eve可 finite c1 ause either Past or Present. Derivational morphemes are not
obligatory in this sense. Derivatiooalmo甲hemes. are , ac∞rding to Green-
berg:

Greenberg goes on to say that the noun duckling may be substituted for
monomo叩hemic nouns such as turkey or goose or even duck , without chang-
ing the construction. Th at is , no construction requires 由e -ling suffix. An
)ntlectional morpheme , then , is a b~und nonroot morpheme whose a皿ar­
m aUn a partkular t>QsitiQDJs∞mD些旦旦. Th us English progressive -饥g

appears obligatorily in a sentence such as The duckling w田 swimming. There

is 00 monomorphemic word that can be substituted for swimming without
changing the construction entirely.
Matthews' (1974: 48) definition and S. Anderson's (1982) proposal 缸毡
similar to Greenberg's in that they both saggest that inflectional mo甲hemes
are those which are required by the syntax of the sentt"nω. But there are
also quite different approaches to the distinction. Kuηfowicz 1964 propo咽S
that derivational processes create new lexical items , while inflectional proces-
ses do no t. Th is , unfortunately , is a theory-intemal definition , sir=e it is an
82 MORPHOLOGY

open question what constitutes a separate lexical item. Bloomfield 1933

observes that inflection is charal: terized by a "rigid para l1 elism of underlying
and resultant forms" (p. 223). He means by this that inflectional paradigms
are highly stmctured sets of words with regular patterns. So , for example ,
nearly a l1 English nouns underly a derived plural. Bloomfield , Nida 1946
and many others have observed that derivational morphemes o~cur c1 0ser
mth t"~叶 t1l a o jnflectional morphemes ,, It is also a common observation
(Nida 1946 , for example) that a language has more.!!el ivational mQmh.emes
tlilln inflectional ones , and this is backed by Greenberg's 1963 finding that
the existence of inflection in a lan~uage implies 吧=二二与ζ日e of delÎ vati、JI1.
Some authors attempt to define inflection by the meanings represented by
inflectional mo叩hemes. Nida points out 由 t ha
川t exoress
刽，ions of gI冒.~，
relations , if morphological , are infl~c:tional ， ~nd Cowgill 1963 supplies a list
of thecategories expressed inflectionally in Indo-European as his characteri-
zation of inflection. Fina Ily , a frequently-cited criterion is that d匹且到皿m.l
lllQ rphemes may change the svntactic categorvof the resulting word. while
in.Jl.e.flional morphemes never ~o.
Noneofthesecritepaj?xmptRe 巾严 the obligatoriness c附 rion ， actu-
a l1 y provides a discrete division between derivational and inflcctional proces-
ses. Nonetheless , they all point to some interesting differences among non-
root morphemes. In Chapter 2 1 introduced principles that determine what
can be expressed lexica l1 y and inflectionally. Using these same principles , it
can be shown that derivational morphology is transitional between lexical
and infledional expression , and that !he differences that çanhe ohseryed
between inflectional and derjyatjooal expressjoo are ju~ 舍~二 mtnent
instances of the differences identifiable amon2 inflectionaLcategories.
1.1. Relevance and two ηpes o[ derivational morphology
In order to apply the relevance criterion to derivation , we must first
recognize two typcs of derivational morpheme: those that change the syntac-
tic category of the word to which they apply , and those that do no t. Let us
consider the latter type first , for here we can apply the relevance criterion
in much the way we have been applying it to inflectional categρries ， by
considering the extent to which the meaning of the affix alters or affects the
meaning o~ the ste
 LEXICAL I DERIV ATIONAL IINFLEC了1l 0NAL ('ON Tl N lJlJ M H.\

Large meaning changes are characteristic of derivational processcs which

do not change synt?ctic categories. In Chapter 2 it was pointed out that
valence-changing categories produce large meaning changes in verbs , since
an event can be changed substantially if the number of participants and the
nature of their roles change. Thus kill differs from die , and send differs from
go in the events being desεribcd. So it is not surprising that in the cross- Ii n-
guistic survey , valence was found to be frequently mentioned as a derivational
categ'时y for verbs. Large meaning changes are al50 evident in other deriva-
tional categories. The event describcd by un- plus a verb in English , e.g.
untie, unhook , unzip , is a cI osely relatc d but distinct event from that described
by the verb alone. Similarly with nouns , a derivational process affecting a
noun produces a noun with an entirely different referent , e.g. garden , gar-
dener; auction , auctioneer; or Spanish higo 'fig' , higuera 'fig tree'; durazno
‘ peach' , duraznero ‘ peach trce' (Malkiel 1978).
Applying the relevance criterion to dcrivational morphcmes that changc
the syntactic category of the stem they modify is somewhat differen t. Here
we must consider the result of the derivational process and apply the notion
of semantic differentiation that 1 argued in Chapter 2 was a consequence of
rclcvance. Th e qucstions then becoane: does the derived word differ substan-
tially in mcaning from the basic one , and does the derived word represent
a coherent concept? Two general points suggest affirmative answers to these
questions. First , the ~yntac1jc ca1egory of 组 worrl 飞 an jnherent part of its
semantic represeutation , so an indicator of s}'ntactic category will always be
highly relevant to the word as a whole S巳 cond ， a pojot whjch app!i es 10 all
types of derivational morphology is that the results of derivational proc吗岳仲
usuallv have a few lexical counterparts , that is , lexical items in whi~
same con、 binat;oos of mcaoiogs arc ç"p士r\ sed mpngmorphρmir兔1ly. For
example , sad for unhappy , pilot for flyer , size for largeness. 1 Th is i咀主ates
that words derived by derivational morphol02Y do represen1 con俨俨P1.WIJly
coherent..unit‘ =unitsthat ∞ uldaswcllbc 町 nresscd b" sinole mornhemes
Oerivations that change the syntactic categρry of a word make varying
amounts of semantic change , depending on how much semantic content they
contribute along with the category changc. Somc morphemes that mak
84 MORPHOLOGY

read in bed. The suffix -ly that is added ωadjectives to produce adverbs
does not change the quality described by the adje创刊， although it does add
the sense that the word desribes the manner in which the event took place.
Cf. Sara is intelligent vs. Sara answered intelligently. Th e agentive suffix -er
that is added to verbs , as in rider, baker, player , etc. , dοes more than changβ
the verb to a noun , since it specifies that the noun is the agent of the activity
named by the verb.
Th us while the amount of meaning change Qroduced by a morpheme is
an important difference between dcrivatu)nal and inflectional morph~，
luoes not produce any discrete division b~tween the two beca!!叫lere are
also wide differcnccs among dCBvational morohcmcs in the amount of
semantic
coe:::. cbange t l:t ev effect
--0- ----.J

1.2. Lexical generality

Derivational processes are more likely than inflectional processes to
have lexical restrictions on their applicability. Of ∞urse ， nQ立旦旦旦ctive
rocess~s will have arbitrary lexical restrictions , but even productive deriva-
tional processes may be apphcable only 1ft a very 且 ωtrie凡、1 selllandc , syntachc
or phonological dQmai!!. For examplê , trle verbal prefix un- applies only to
verbs that are inherently reversible , even though it is productive in this
domain (Thompson 1975). The suffix -ize applies productively only to words
of more than one syllable , e.g. , idolize, magnetize, fossilize , traumatize. On
the other hand , inflectional categories must have fulllexical generality: if a
category is required by the syntactic structure , then there must be an exponent
of that category for any lexical item that fits that slot in the syntactic structur巳.
Th is does not mean that all expressions of an inflectional category must be
regular or productive 一 it does not matter if an English verb forms its Past
Tense by suffixation or vowel change - it just means that there must be
some way to form the Past Tense of every English verb.
Th些mOt:e-Sρn町aLa mn吃、hologie二l-p wcess ，也;且e-m6fo二t.t w jJJ resea叶、le
an inflectional orocess.., For instance , the -ly suffix in English that produces
an adverb from an adjective has fulllexical generality in the sense that any
member of the category a叫jective can take this suffix. The few exceptions
resemble the exceptions to inflectional processes: good. and well may be
thought of as suppletive forms , and fast may be thought of as having a zero
allomorph. It could even be argued in this case that the morpheme -ly is
required by the synt
 LEXICAL / DERIV ATIONAL / INFLECfIONAL CONTINUUM 85

is required in the senten∞ Sara gave a though扩'ul answer but an adverb is

required in Sara answered though仙lly. In fact~ i! is possible that the principle
lkat morphology that changes the syntactic categ。可创 a worC! IS 创wavs
也riy~tional is a false principle , and that -ly , gerundlaI -zng and ∞mp缸able
rnomhemes in otber la Dl!U al!es should be consid问。rJ in 伺a户tinn~1
The criterion of generality does not ab咀olutely distinguish derivational
and inflectional morpbology , since inflectional processes are not always
totally general cithc r. 2 For cxample , Beard 1981 points out that in English
for the category of number in nouns, which is ordinarily thought of as an
inflectional category , there are a numbcr of defective paradigms. These
in cJ ude plurals that do not have singulars (p luralis tantum) , such as pants,
oats , measles , and pliers , and singulars that do not have plurals (singularis
tantum) , such as air , fleece , peace, contemplation , c/eanliness , and so on.
While Beard argues from these and other facts that Indo-E.uopean nominal
number is a derivational category, in the present framework these facts can
be interpreted as supporting the proposal that all morphological categories
belong on a continuum that ranges from lexical to inflectional. Noun number
lies closer to the lexical end of this continuum than case (with which Beard
compares number) , or definiteness. Th is follows from the fact that a change
in number produc臼 a change in the entity or entities being referred to , while
case signals the relation of the noun to other constituent~ in the senten白，
and definiteness shows the place of the noun in the discourse. Case and
色坐监旦旦坦旦旦旦ect on the inherent aualities of the entit百二币2IeS
H iJlg referred to , while numb~does.
'‘ fher examples of inflection lhat is not totally general are frequently
encountered. For example , in Acoma , negation and other moods are indi-
cated by the obligatory choice of one of a set pronominal prefixes for the
verb. In the case of negation, however, there is a distinct pronoun only in
the 1st person. The other persons use an unmarked pronominal prefix and
a negative adverb to indicate negation (Miller 1965). ln Serbo-Croatian ,
there is a very general distinction between the perfective and imperfective
aspects , but only some verbs further distinguish the iterative and durative in
the imperfective (Partridge 1964).
Even where inflection is not totally general , it must nonetheless have
at 1east one type of forrnation ffiafì
 86 MORPHOLOGY

the defective paradigms we discussed above do exist , and even though other
means of plural formation , such as vowel change , do exist , because new
nouns automatically take this inflection. Productivity is not nec巳ssarily
affected by lapses in generality , by irregularity , nor by the derìvational -
inflectional continuum , since derivational processes may or may not be pro-
ductive.
13CausesoflackoflexicalgeMrglity
Th ere are several reasons why a morphological process may lack lexical
generality. A brief consideration of these reasons will help us understand
the difference between derivation and inflection. W
First , consider the fact that because of their high relevance , dprh叶 tinn~1
nrnrp~~e L.CIi ftpn~
.μ且与单~
卢.....n.....:_~~..:......…
‘---… .~~.....ó 、...V J. IIVJ 且思 ........--..ν
..L......
~........‘
__.... Ull.....UU~pl"'.:n...
UIC
....1_........ ...1... ___tforl
l1 L"'U

kxi且II事. 1 have already mentioned examples such as the agentive forrnation

flyer for pilot , and the negative adjective forrnation of unhappy for sad.
Th ese cases are unusual because both a derìved and a nonderived form exis t.
As Clark and Clark 1979 have observed , it is usual for the derived form to
be rejected if the semantic combination is already represented lexically. Th us
unp 陀tty is rejected because of ugly , cutter cannot be used to mean scissors ,
and so on. Valence-changing processes are very good cxamples of this phc-
nomenon , because they are very frequently representcd morphologically in
the languages of the world , yet they are often not general , b包旦旦旦旦旦l旦旦旦
s an inherent p~t of the meaning of a verb , and differences in valence are
到旦旦旦Poresented lexicallv , Even though , for example , transitivizing mor-
phology occurs in a large percentage of languages , there are probably not
any languages in which alllexical verbs are intransitive , and all transitives
are forrned by affixation. The reason is that there exist in the world as we
experience it certain evcnts that are inheren tI y transitive , and not necessarily
divisible into an intransitive event plus a transitivize r. As mentioned above ,
because derivational meaning is highly relevant , it is often the case that
l~cal expression exists for meanings similar to the de巾 ational one.
~ Another cause of the 1ack of lexical generality is the specificity of deri-
旦坐旦出旦旦!~ compared to jn f1 ectjonal meaniig. Th e greater specificity
in derivational meaning restricts the applicability of derivational pro臼豁出­
For instance , a morpheme meaning "enter into a state" wil\ be applicable
only to stative verbs. For instance , Maasai has an Inceptive suffix for verbs
of state: á-íbór "1 am white" , á-íbórr心 "1 become white"; á- r5k "1 am black" ,
房时k.ù "1 become black"; 在 náná "1 am soft\ á-nána-u "1 become soft" ,
 LEXICALI DERIVATIONAL IINFLECTIONA I. ('ON Tl NIIIIM M1

but this suffix does not apply to active verbs. 3 The spccificity of its mcanill).\
can be seen best by comparing inceptive to a related but in f1 ectional aspccl.
the perfective. Th e meaning of the perfective is more general , for cxamplc.
in Spanish , where the Preterite may apply to verbs of any type. Its gencral
meaning is "bounded event". When it applies to active verbs , it givcs a
punctual or completive meaning , but as we saw earlier , when it applics to
stative verbs such as saber "to know" an inceptive reading of "found out"
results. In th南 case of in f1 ection , then , the lack of lexical restrictions coincides
with extreme semantic generality. When derivational categories lack seman-
tic generality , then one can expect lexical restrictions. 1 wil\ return in section
5 of this chapter to a discussion of thc difference bctwccn in f1 ectional meaning
and derivational meaning.
Th e examination of derivational and in f1 ectional categories shows , then ,
that the formal differences between these two expression types are closely
related to properties of the mcanings of these categories. Th e parameters
along which meanings differ , r!levance and generality , are the same paramet-
ers that are applicable to differences among in f1 ectional categories. 1 四匹!ude
, tben..1hat therc is not ncccssar i\ v ~discrete distinction between in f1 ection
and derivation. but 巾 t ha
创t the properties of the meanings expressed by
兰a4巳士get'己ie
击= ζιL
‘on川
νhtα ‘e-t川
E二i邑刮
吨 hι '"ιJ
二~伫巳 1仇 凹 、 in..which the cat忧
h~ fn.rn
m e~or
町 y沁
i s expressed.
Thc ncxt two scctions examine the causes of lexical and in f1 ectional splits ,
and show that the factors affecting splits between derivationally-related
~ords are the same as those affecting splits among in f1 ectionally-related
words , and that the differences between the two types of morphology are
just a matter of degree.
Some linguists would be unhappy with the conclusion that the distinction
between derivational and in f1 ectional morphology is not discrete , but rather
a gradient phenomenon. Th is is especially true if one espouses a formal
theory in which derivation and in f1 ection have differ毡 nt types of representa-
tion in the grammar. For instance , in f1 ection , which is required by the syntax ,
is derived by rule , while derivation is represented or derived in the lexicon.
Themod时 that results from the theoretical principles developed in this book ,
however , recognizes gradient phenomena of various so
xx MORPHOLOGY

2. Lexical split

One of the properties that characteriz启s derivationally-related pairs of

words most ∞nspicuously is their tendency to split up , to move away 齿。m
one another both in meaning and in form. For example , even where the
phonological relations are quite c1 ear , there are often semantic di任erences
among derivationally-related words. Thus , something can be dirty without
involving real dirt at all , but rather from having , for example , maple syrup
srilled on it. Even zero-derivations work this way: someone can soil an item
without being anywhcrc near real soil. Similarly , something that is awful
does not inspire awe anymore , nor do you have ωbe terrified to think some-
thing is terrific. There are also cases where phonological distance contributes
to the dissociation: despair, desperate; and cases where tl.e lack of productiv-
ity of the relating process aids in the demise of the relation: bake, batch;
shade, shadow. Le xical split may be described in terms of the theory devel-
oped in Chapter 3, as dl!e to the increasin l!. autonomv of the derived memb~r
gi..t比一pai r. The factors that determine autonomy also determine the likeli-
hood of Icxical split:
(1) Th e more frequent a derived form is , the more likely it is to become
autonomous.
(2) The greater its phonological distance from the basic form , the more
likely a derived form is to be autonomous.
(3) Th e greater the original semantic change made between the basic
and derived form , the more likely the derived form is to become autonomous.
While lexical splits are extreme.y c('mmon in derivational morphology ,
they are not unknown in inflectional morphology. As predicted by the lower
autonomy of inflected forms , inflectional splits are relatively uncommon ,
but they do occur. Evidence for them is found in suppletive paradigms , such
as that for the verbs be and go in English , and in former singular I plural
pairs such as brother I brethren , and cloth I clothes. In this section and the
next , it wi\l be shown that the factors that determine autonomy can be used
to predict the likelihood of lexical splits in both derivational and inflectional
morphology.
Two absolute determinants of autonomy are semautic and mo甲ho­
phonemic unpredictability. If a word is not derivable by general semantic ,
morphological and phonological rules from some other word or stem , it must
have its own lexical entry and be autonomous. However , even a word that
 LEXICAL I DERIVATIONAL I INFLECf IONAL CONTINUUM 89

is non-autonomous may gradually develop into an autonomous word. In this

diachronic development several factors are importan t. First , there is the
original degree of sUlllllUiι.ch.ange between basic and derived forms. In Chap-
ter 3 it was argued that the degree of semantic difference between basic and
derived forms dctcrmines the degree of relatedness among forms , and that
the degree of rclatedness among members of categories follows the relevance
hierarchy established in Chapter 2. For derivational processes it is much
more difficult to establish degrees of semantic relatedness among pairs of
words , so this notion will remain an intuitive one in subsequent discussion.
The general proposal is , however , t ha.L1bρ 81'士乙二the-ot仨imri 技maMi c
4ifferepce i5 betw乙二R-a巳asic aAd 士 rlA吨，ρrI fnr Tn thρ n1nr.. lilr ..lv th..r.. i" to
be a 中lit be1we主且主þem IÞat is , the more li卫士 Iy tbe r.. js to be a further
输血豆豆 tie lliU4tF1.Atiatiop d"ve1oped Þetween them ð. second catalyst foflex-
ical split is the phonological distance between forrns. Th e most important
factor in the development of autonomy in derived forrns , however , is the
卢'equency of use of the derived forrn. High frequency correlates with both
semantic and phonological differentiation , as can be seen from the following
case study.
Pagliuca 1976 studied the 323 extant words with the prefix pre- listed in
the Shorter Oxford English Dictionary. Th is prefix is listed as occurring with
four di征erent vowel qualities , represented as [iy] , [i ], [吁， and [E]. For each
prefix , Pagliuca recorded inforrnation about the vowel quality , the frequency
of use in texts of each word with the prefix (as reported in Thorndike and
Lorge 1944) , and whether or not the meaning of the word with the prefix
was a predictable sum of the meaning of the base plus the meaning of the
prefix. 4 With the aid of a computer Pagliuca sought correlations of the vowel
quality with the other variables. Of particular interest to us here is the fact
that there is 挡坦n& [elatinnship betlal问R-tl咱 YQwel aualitv .arulthe freq哩!1cy
of the word ---.an d the vowel aualitv apd th.. semantic predjctabilitv of the

nt ,
se
 90 MORPHOLOGY

Vowel quality Frequency % 01 words with Examples

predictable meaning
[iy] 05.74 74.76 predecease
[i] 02.54 59.52 predestine
[i-] 49.80 03.30 prediction
[E] 81.32 02.89 preface

Table 5: Correlation of vowel quality , frequency and semantic predictability

in pre- words.

Pagliuca's int f' rprctation of these facts is that high frequency leads to
se时mtic differentiation of derived from basic forms. A high-frequency
derived word can be learned by rote , without an a l' alysis into constituent
parts , and without relating it to other words. A high-frequency derived word
may develop contexts of use that are independent of the contexts in which
阳 related base word is used~This is also possible for lower-frequency
words. but not as likely.) Thus ;ì'ô use an example cited earlier. uses of dirty ,
as in "the dirty c1 0thes hamper" or "the pile of dirty c\ othes" may occur
partially independcntly of rcfcrcnccs to actual dirt. Similarly. rcfcrcnccs to
an interstate hì坡hway may occur in contexts other than thosc in which the
notion of actually traveling between states is importan t. Thus the term may
come to mean any four-Iane highway. On the other hand , a term such as
pre-wash is currently restricted to contexts in which the term wash and its
associated notion alsοoccu r. This helps to maintain a transparcnt scmantic
relation between pre-wash and wash. Th è important point is that the higher
也e frequency of a derived word. the more likely jt js to occur in a vàñw
ofcontexts , including soæe in which its related base word... 且 nd the semantic
-llQtions expressed by it. do not occur
The data on the prefix pre- also show a correlation between the loss of
semantic transparency and the change of the vowel away from the canonical
[priy]. It is not certain whether original phonological differen臼s between
derived and basic forms en∞urage lexical split (see the discussion of split in
intlectional paradigms below) , but it is certain that the semantic changes that
时在ur in lexical splits are accompanied by phonological differentiation of
derived and basic form. Phonological changes affecting the derived form as
its relation to its base slackens are reductive changes such as loss of secondary
stress , with concommitant vowe i' and consonant reductions. and greater
 LEXICAL / DERIVATIONAL / INFLECI1 0NAL CONTINUUM l} 1

fusion of the originally separate units of the word. Cornpare for exarnple thc
word highness , as in the phrase Your Highness , to the word slyness. Thc
forrner rhyrnes with sinus , having a;horter vowel in the first syllable , and
no secondary stress on the second syllable , and the vowel of this second
syllable is also rnore reduced in sinus than in slyness. Reductions of this sort
are well-known frorn cornpounds in English , such asforehead and breakfast ,
which have highly reduced second syllables in rnost dialects.

3. lnflectional split

Splits in inflectional paradigrns are conditioned by precisely the sarne

factors as splits between derivationally-related words , i.e. frequency , sernan-
tic relatedness , and , to a lesser extent , rnorpho-phonernic unpredictability.
Any situation in which two inflectionally-related words lose their in f) ectionaL
relation can be characterized as a SD~sucb solits mav come 油out in
ightly dift町:ent.wa户， and ba"e diffpHlR w:o.n seauences. Consider the case
of the split between the verb work and its forrner past participle , wrought.
In this case , a new regular past participle was forrned , which was in cornpet-
ition with wrought for a time. The new past participle then cornpletely
rcplaccd wrought in all uses that wcre transparently relatable to work , whilc
wrought rcrnained in ccrtain fixcd phrases , such as wrought-iron , over-
wrought or highly wrο ught ， and ccrtain spccializcd language , such as poetic
and biblicallanguagc. Thlls work now has a reglllar paradigrn , and wrought
is an isolatcd , largcly archaic word. Thc split hctween cloth and its forrncr
plural c!othcs is sorncwhat diffcrcn t. Thc plural c!othes becamc spccialized
to refer only to garrnents , while the singular continlles to refer to woven
rnaterial in genera l.
Sornetirnes splits in inflectionally-related forrns result in suppletive
paradigrns 一 infl.!ctional paradigrns that have forrns built on two or rnore
sterns that are etyrnologically frorn different sources. For instance , the Eng-
lish verb 10 go has Pr毡 sent ， Infinitive and Participle forrns built on go , but a
Past Tense forrn , wenl , which is historically the Past Tense of the verb 10
wend. Sirnilarly , the English verb to be has stems frorn three sources: be,
being and been corne frorn one verb , am , is and are come frorn another , and
was and were frorn a third source. Let us take the exarnple of to go to see
how suppletive paradigrns develop. Prior to the 15th century , the verb 10 go
already had a suppletive Past Tense (eode) , and the verb to wend had the
Past Tense went. This latter verb had a mcaning that was more restricted
92 MORI'HOLOGY

than 10 go , including in its mcaning a notion of turning or winding. Th e Past

Tense forms of wend gradually dissociated themselves from the other forms
of the verb. These forms must have gradually increased in frequency , coming
to have the more general meaning of go , while the Present Tense forms
retained the more restricted meaning. Thus a lexical splitωcurred between
wend and wenl. At the same time thc oldcr past forms for 10 go became less
and less frequent , and were eventually 10喝t. (See Rudes 1980 for a description
of an ongoing change in Rumanian similar to this.) Since suppletive
paradigms result from splits in inflectional paradigms , the characteristics of
suppletive paradigms provide evidence for the factors that govern inflectional
splits. Th e interaction of semantic r官latedness with frequency of usage is
important here just as it is in lexical splits among derivationally-related forms.
First of all , ~uppletive paradi2m这盯ρrestricted to the most bigbly-frequent
lexical items. Second. suppletjye paradj雪ms are divided alon l! the cate l!o'1Y
lines that involve tbe I!reatest chan l!e in meanin l!. i.e. in verb‘ l I Jnno lI S I:lect
and tense lines..1Ju:.豆豆豆。!1 d principle ca l!be overridden in cases of extrem~ly
high freQuen~~ach of these principles will be discussed in the follo叫wing.
It is well known that suppletive paradigms for verbs in Indo-European
languages always involve the most frequent verbs. Common examples are
the verbs 10 be and 10 go which are suppletive in many branches of the family.
This phenomenon is not restricted to Indo-European. In the sample of lan-
guages described in Chapter 2, there were nine for which suppletive verb唱
were listed: four had suppletive stems for a verb meaning 10 be , six had
suppletive stcms for a verb meaning 10 go , and six had suppletive stems for
a verb meaning 10 come. Other meanings for verbs with suppletion occurring
in more than one 1anguage were 10 give , 10 bring , and 10 sit or 10 be localed.
The phencmenon is a1so observable in lndo-European adjectives , where the
most common ones , good and bad, have suppletive comparative and super-
lative stems , e.g. English good , better, besl , and bad, worse, worsl , and
Spanish bueno, mejor and malo, peor.
Th e seωnd important factor that determines where splits occur among
inflectional forms is the degree of relateoness of the forms. Closely related
forms are less Iikely to split apart , thus inflectional splits are most Iikely to
coincide wit
 LEXICAL / DERIVATIONAL / INFLECTIONAL CONTINUUM 93

verbs from my own survey , indicate that the most ccmmon place for a
阳radi阴阳刚it is alω瑞丽而~n盹 Splits also OCCl' r along tense lines ,
,

less commonly along moòã而否?而y survey also turn.:d up a number of

cases of fuppletion governed by the number of the subject or 0均e吼These
cases are discussed in detail in section 5.2 of this chapter. A much smaller
number of cases show suppletion along person agreemem lines , but these
occur only in the spccial circumstances to be described below.
It is my impression that suppletion in noun paradigms is somewhat less
common than suppletion in verbal paradigms. This follows in fact from the
principle that the original semantic differentiation determines the likelihood
of a split. For instance , we would not expect splits to occur among nouns
distinguished only for case , since case does not a任'ect the meaning of the
noun stem , but only signals its relation to other constituents in a particular
sentence. Nor would we expect ~plits to be especially ∞mmon between sin-
gular and plural nouns , since ordinarily the occurren臼 of one versus 仇e
occurrence of many does not change the il1herent qualities of 由e entity. Nor
would we ordinarily find many cases where a plural would be used in a
context that was independent of the ∞ntexts in which the singular was used.
Tþere are ca函es ， however , where a group of entities isωnceptualized 部
being inherently di侄'erent from an individual instance of the entity , and in
these cases we do often find different stems being used. But we also find the
meaning difference between the stems in such cases to be distinct enough to
refer to the non-singular forms as collectives rather than plurals. Thus we
do not usually think of pairs such as cow, cattle , and person , people as sup-
pletive singular / plural pai邸， but rather as singulars and ωllectives. When
suppletion between singular and plural nouns does occur , it 创兀urs in predict-
able lexical items for the mo喝t part. For example , Maasai has suppletive
plurals for the nouns meaning cow, OX, tree and somewhat less predictably ,
girl.
The second principle , which predicts that inflectional splits are less likely
among the most closely-related forms , may be violated in cases of extreme
high frequency. Rudes 1980 points out that suppletion that divides the person
/ number forms of a tense , aspect or mood occurs only in the present tense ,
the most frequent tense in the languages he examined. Furthermore , this
type of suppletion is extremely rare , and tends to occur in the mωt fr ,
94 MORPHOLOGY

the lrish verb bheirim '1 give , bring' , which uses three different stems , one
for the Present and Irnperfect Indicative , another for the Past , Dependent
and Subjunctive forms , and a third for the Dependent Future and Condi-
tional , has a 2s Imperative that is different frorn all the other Imperative
forms. This is explainable in terms of the extreme high frequency of the 2s
of the lrnperative ∞ mpared to other Imperative forms. Indeed , rnost lan-
guages do not even have first and third person forms in the Irnperative. It
should also be noted that a second person imperative form would be especially
frequent for a '1 erb such as ‘ to give'.
Theextreme high frequency and wide variety of f'l ACtiQAS fer 二飞'sfb
su(".h as ‘ to be' orobablv leads to tbe a l.ltQ亡 emy 士f-all Of its fc J;Ø亡;--ëA'~ a
paradjgm stDlctured somewbat d.i.ff士士~二1ft tkat ef QtA8f 山由~ Consider
the fact that not only does the verb to be in English serve a number of
functions , i. t:., as the copula , a5 part ofthe progressive and passive construc-
tions , as well as the pos在essive ， the cleft and p咀eudo- c1 eft constructions , it
also has a larger number of forrns than any othcr English verb. There are
two sources for these forms. The forms that differentiate person and number
in the Present (am , is and are) and nurnber in the Past (was and were) are
forms that have survived long after all othcr vcrbs havc ccased to mark these
distinctions. And each of these variants has proliferated into multiple forms
through contraction with nouns , pronouns and with the n巳 gative elemen t.
Many ..Qf the contracted formζ 兔Ie not syncb a:.g吃 16二 II~ preèi二止!二句-ιζ Heral
rule and are thus autonomous二 Moreover ， forrns that are predictable by rule
may also be -autonornous , and evcn forms that are phonologically "identical"
may have multiple lexical representations given their extrerne high frequency ,
and their occurrence in constructions of diffe 陀 nt types (Pagliuca 19但).
The fact that a split rnay occur betwecn two diffcrcnt uses of an identical
form necessitates dual or multiple represcntations for single forms. Consider
the development of the rnodal use of supposed in English. The verb suppose
in its meaning ‘ to hold as opinion or belie f' has hccn in usc since the 14th
century. This use continues in sentences such as "They supposed him to be
qualified" , "He supposed that she would be home". or "Lct us suppose
tha1... ". At the same time , however , a newer use has dcveloped from a
passive construction with this verb. Thus "They suppos
 LEXICAL I DERIV ATl ONAL IINFLECf IONAL CONTINUUM 95

in meaning is accompanied by a gradual differentiation in phonological form.

The modal use of supposed is undergoing phonological reduction at a faster
rate than the more conservative and now less frequent use of supposed. In
the modal use , the first syllable may be reduced considerably more than in
the other use , and the final consonants ar巳 usually voiceless , due to the
following to , whereas in the "belief" use of suppose the final consonants are
never devoiced , even when a voiceless consonant follows , as in "Th ey sup-
po画ed Tom to be the culprit".
Th is type of diachronic development shows that a single form with a
unified meaning can split into two or more forms. ~ince a single form with
aILoriginally unified meaning can spljt jnto two or mou 吧'…士raR士了达1兰al
Remantic differentiation. whkhJ组rI'll ed is a factor in deter…~ρ liIc eli­
hQod of infleÇ!ional split. cannot be consjdered a necessary condjti on for sp!i t
A final factor to be considered is phonological differentiation between
the two related forms. In the cases of inflectional split discussed early in this
section , work , wrought; c/oth , c/othes; and other cases such as brother, bret-
hren , or get, got , there is a greater phonological distance between the mem-
bers of the pair than is predicted by the regular inflectional rules of the
languagβ. Th ese differences were prcsent before the split occurred , and could
have been instrumental in producing the split , especially since the m.QIllho-
phonemic unpredictability wouJd Jead tg a兰~工tm!ζ 三slexical represent吃ÍQAS
fllUhe derived forrn!..(see Chapter 5). In fact. in the case of wrought and
brethren the main impetus for the split was the creation of a new regular
form that gradually replaced the irregular one. In the case of c/othes and got
the di任'erentiation in meaning may well have been aided by the formal differ-
ences in the related pairs. It is questionable whether the split between wend
and went could have been aided by the small phonological difference between
these two forms , but it is po~sible~ρ ÌUêg巳 larity gf 由 ρp ，，~曹 formation
could have encouraged the development of a new regular Past for wend which
wõiilOTeave went available~呐 t什h
仙、W
俨
factors having to do with the formal relation between the inflectional Q!创ir臼l's:
{Ithe phonologlcal di鸣<;tan】 c臼e between them. and momho-phonemic irregula!:Ï ty.
Neither of these is a necessary condition for split in general , however , since
lãentical forrns can al
96 MORPHOLOGY

any real differences in quality in splits between derivational and inflectional

forms. 丁be fact that splits are much more common wber.deriyational rela喃
tions are involved is due to the fact that there is

4. Degree of fusion

It is often observed that derivational affixes occur closer to the root or

base than inflectional affixes do. Th is ob咀ervation also holds for the languag剧
reported on in Greenberg 1966 (see Universal #28) , and it was not ∞n­
tradicted by any of the languages in the survey reported on here. As we saw
in Chapter 2 , there were also some significant generalizations ∞ncerning
the ordering of inflectional affixes with respect to the verb base. In particular ,
the more relevant the meaning of the affix to the verb , and the greater the
semantic chan2e involved in their combjnatjoo 仙 p rlnc:.pr 阳仇~巾 b空军芒
the affix wi11 appear. It can now be seen that this principle also predicts that
derivational affixes , to the extent that they are highly relevant , will occur
closer to the verb base than inflectional ones. The observed ordering regular-
ity of derivational and inflectional mo甲hemes does not , then , yield a discrete
division between derivation and inflection , but rather follows from the
hypothes that a/l morphological proc臼ses can be ordered on a scale such as
the one proposed in Chapter 2 for inflectional processes.
One language-specific example demonstrates nicely the diagrammatic
relation of proximity to the stem and greater semantic fusion. 7 1n the dialect
of Eskimo reported on in Sadock ard Olsen 1976 , there is a root for "person" ,
a suffix meaning "big" , and another meaning "little". 币lese three units can
be ∞mbined in two different orders. When "big" (-rssu-) 0∞urs closer to the
root (ino-) than "little" (-ang时， as in inorssuanguag , the meaning is "little
giant". When "little" occurs closer to the root , as in inunguarssuag , the
meaning is "big midget". Th e suffix that is clo咀er to the root affects the
inherent meaning of the root , while the outer suffix functions more like an
attributive adjective. Note that both of these suffixes would probably be
consider叫 derivational. This means that the proposed principle accounting
for the ordering of affixes also governs the ordering among derivational
a征ixes.
Some examples from English i11ustrate a similar point: 1 mentioned
above two suffixes that change the syntactic catego可 of the words they attach
 LEXICAL I DERIVATl ONAL IINFLF. CTIONAL CONTINUUM 97

to, but otherwise produce only a small meaning change , and are very general
and regular in application , adverb-forming -钞， and gerundial -ing. Th ese
suffixes are always the last in the word if there are other derivational suffixes.
Consider the ordcr of suffixcs in the following words with averbial -Iy: mer-
c矿'ul秒， regre扩'ul秒， assertively , grammatical:纱; and words \\ ith the gerundial:
.I~ystematizing， whitening and so on.
- - Statements by Burling 1961 about Garo suppo口 the same poin t. In Gar。
而ere are a number of productive affixes that cannot be ∞nsidered inflec-
tional because they are not obligatory. Burling calls them ‘ aùverbial affixes"
and describes their distribution as follows:
Every verb may have from none to se明 ral adverbial affixe3 , then , and there
are n' l formal limits on the number or the order in whi阻h t卜 ese occur.
However , statistically some are more likely to occur first and others last.
ln general the most common ones ( =泪ost general, JLB) last, occur the
rarer ones first , the rarer ones being more firmly attached to the 、 :rbb部队
while the more common ones are more readily sübstitutable affixes. (Burlin鸣
1961:17)
1 equate "general" with "common" in this case because Burling is referring to
how commonly the suffixes occur in differcilt combinations , i. e. , their type
frequency. A su日ix might have a high token frequency because itα;cun. in
one or two very frequent words, and in this case it might not be general at
all. (See Chapter 5.12 for a discussion of the effect of type frequency on
affixes.)
Other measures of the degree of fusion discussed in Chapter 2 also show
derivational pro臼 sses to fit along the same scale propωed for inflectional
pr.ωesses. For example, derivational processes tend to have a greater effect
on the root than inflectiona1 processes do. Th ey condition and are sometimes
signalled solely by root changes. Reduplication is more common among deri-
vational processes than among inflectional. Indeed, in the survey described
in Chapter 2, on1y two languages were found to have reduplication as an
inflectiona1 process for verb咀， while there were numerous instances of redup-
Iication mentioned for derivational processes. Even more striking , however ,
is the fact th血d旦坐旦旦as not foun但业旦旦dωm81 pmcess in a!!Y
s examined.whíl
如m坤而已e 盹也
The other measure Qf fusion is the morpho二Qbooem;c effect tMιthe
pQt has on the af[x. Th is is manìfested both in greater fusion in the sense
oi.jn tem垒1 sandhi proce~且tcllan2e~affix. aru口画 lexicallv-determined
98 MORPHOLOGY

a 1I0morphy where the root determines the choice of morphological process

used to express the çategor~Examples are numerous: Eng \i sh unable but
mcapable , and unhappy but discontent , as well as deverbal nomina \i zations ,
whkh seem to be partially dependent on the final segments of the verb , but
are not totally predictable:
prevent pr~vcntion *preventment
resent resentment *resention
present presentation .presention
satisfy satisfaction .satisfiance
comply comp \i ance .complaction
apply application
hesitate hesitation
orient orientation
decide decision
abide
A variety of lexically-determined affixes and processcs arc characteristic
of derivational morphology , and are found in inflectional morphology to a
greater or lesser extent , depending upon the degree of scmantic relevance
of the inflectional category to the stem. This measurc of thc dcgrcc of fusion ,
like all the othcr criteria that distinguish derivational "rom inflectional mor-
phology , also distinguishes categories within the inflcctional range.

5. The difference between derivational and inflecticmal It/ {'(lning

Th e major proposal developed in this chaptcr is thal Ihc COpt8A专 of

morphological categorie~ßetermines whether they will appcar jn jnflectjoDaJ
or derivational expression< Other Ii nguists who havc studicd this question
have not arrived at such a firm conclusion. Sapir proposcs hoth a classification
of concepts , and a classification of expression typcs and su~~csts a correlation
between them. Howevcr , hc cvcntually concludcs hy sayill~:
We must dispense ... wilh a wcll-urdcrcd classificali川101 山 Ic~orics. What
boots il 10 pUI tcnsc and modc hcrc lI nd numhcr thcrc wlll'lI thc ncxl IlI nguugc
one hanØles puts tcnsc a pc 且 "'owcrdown" ， modc and nUlllhcr a pcg "highcr
up"? (Sapir 1921: 107)
From a much different point ofview Kurylowicz 19M argued that some
inflectional categories are closcly related to some dcrivational categories.
He gives the examples , among others , of Aktionsart as related to inflectional
 LEXICAL I DERIVATIONAL I INFLECTIONAL CONTINUUM 99

spect; collectives (derivational) as related to plurals (inflectional); and gen-

jer in nouns (derivational) as related to gender in adjectives (inflectional).
This would suggest that the same meanings can be expressed in different
ways. Much more recently the point has been made again by S. Anderson 1982 ,
who argues explicitly that the inflectionall derivational distinction cannot
be made on the basis of meaning. Anderson points to the examples of an
ordinarily derivational category , diminutive , which in Fula is inflectional ,
and the ordinarily inflectional category of tense , which in Kwakiutl , Anderson
c1aims , is derivational. 9 In view of these opinions , it is worthwhile to c1arify
what exactly can be predicted abcut tne relation of meaning to derivational
or_ inflectional expression.
The relevance principle predicts that certain categories may have deriva-
tional expression , in particular , valence-changing categories , voice and aspect
for verbs. Categories that have propositional s∞ pe ， such as tense and mood ,
are not derivational , nor are agreement categories , which index the arguments
of the verb. For nouns , we can predict that gender and number may have
derivational expression , because they modify inherent characteristics of the
referent , but case and definiteness will not , because they se 凹e to relate the
noun to other elements in the sentence or discourse.
These predictions bring us c10ser to our goal ,1 n'tt they say nothing at
all about some categories , and allow others to have either derivational or
inflectional expression. For thes c:: categories it mus tJlf: ShOWD tbat 舍 h闻ρis
some difference between the meanilH!. exoressed a~ g@ri哩 ratiGRa] aRd 巾。
币乓ing expressed as inflectional. Here the cri时OA ef geneFfÛi
furJi1 cases where similar co~tual content is expresse~

often[)rOd uccs jdjosyncratic meanings in comhination with diffcrcnt Icxical

五三自剖In thc following two scctions thcsc points will hc illustrated in discus-
sions df aspect as a lexical , derivational and inOcctional category , and the
so-called "plural" of verbs as a Icxical and derivational category
100 MORPHOLOGY

5. 1. The expression of aspect

It is particularly appropriate to begin this discussion with the set of
related notions labe \l ed as aspect because this set of concepts a\l ows every
type of expression from lexical to periphrastic. Consider first lexical expres-
sion: verhs tcnd to havc inhcrent aspectual meaning because the activities ,
cvents and situations dcscribcd by vcrbs tcnd to havc inhcrcnt tcmporal
properties. Some verbs describe events that take place in a very short period
of timc , such as cough and blink. Thcsc arc refcrrcd to as "四且且也'
verbs. Some verbs are stative in meaning since the situations they describe
are continuous and unchanging ovcr time , for example to know , to have; in
some languages , notions such as to be lall , 10 be black are expressed lexica \l y
as verbs. Some verbs , "atelic" verbs , describe activities that do not have
inherent endpoints , such as to eat or to sing , while others , "telic" verbs ,
describe events with internal structurc , such as a necessary beginning and /
or end point: to dive , to devour.
Some of these lexical distinctions may also be expressed as derivational
distinctions: In Quileu忧， the substitution of the suffix -(a)ts instead of the
normal-l 吁 ndicates a more rapid or energetic action" (Andrade 1933: 227) ,
which appears to be semelfactive: ceq + 0 + l ‘ to pu \l', and ceq + w + ats
‘ to jerk'; k'ix + a + l ‘ to lift gradually' , and k'ix + a + ts ‘ to lift suddenly';
wa:x + i + 1 ‘ to stop' , and wa:x + a + ts ‘ to stop suddenly'; k'wadaq + a +
l ‘ to tear (cloth of any fabric or textile)' , and k'wadaq + a + ts ‘ to tear with
a jerk\Comrie 1976 mentions the example of a set of Russian verbs , all with
a sequence -nu- , a \l of which describe semelfactive events: kasljanut' ‘ to
cough' , blesnut' ‘ to flash'. Th e identification of the -nu- sequence as a suffiτ
does not seem justified since the stems do not apparently occur without this
sequence. Rather semelfactive in this case has a mode of expression that is
half-way between lexical and derivationa l.
As mentioned in section 1.3 of this chapter , inceptives tend to be
restricted to stative verbs , so in languages with a morphological inceptive
(or inchoative) marker , this marker is derivationa l. A similar observation
may be made about iterative markers. While it is common for languages to
have morphological iterative markers , th巳se are usually derivational also ,
because they are applicable primarily to semelfactive verbs (although they
may apply to activity ver
 LEXICAL / DERIV ATIONAL / INFLECìl0NAL CONTINUUM 101

time" , or punctual: irimaoor-ai-wado ‘ shriek many times'; oriodor-ai-wado

‘ go astem frequen tI y'; iaeed-ai-wado ‘ pu l1 the bowstring many times (without
shooting); idi-ai斗也'ado ‘come up from below often'. 10
Th e preceding are examples of derivatiυnal aspec且 markers that are
restricted by their M t< aning to a semantjc dasi of V8rbs. Another way.. that
gvational meanin2 differs from in f\ ectional is that it is not 50 "阴阳".
derivational markers that may beprimarily aspectual might contain other
components of meaning as well. For instancc , in Slavic languages. the perfec-
tivizing prefixes oft巳 n have more specific aspectual meanings , or meaning
components that are not rea l1y aspectual (Forsyth 1970 for Russian). P缸­
tridge (1964 :94) gives these examples for Serbo-Croatian: iéi ‘ to go' , and
proéi ‘ to go through , go past'; teéi ‘ to f\ ow' and proteéi ‘ to f\ ow past'; or
consider the various perfective derivative 'i of písati ‘ to write': potpisati ‘ to
write under , to sign'; natp{sati ‘ to write over'; nap{sati ‘ to write down'; pre-
pfsati ‘ to re-write'.
Chapter 6 discusses in detail the in f\ ectional expression of aspect in the
50-1anguage sample. Th e major cross-linguistic pattem found for inf\ ectional
aspect consists of a perfective / impe收ctive distillction sometimes
supplemeated by a habitual / continuous distinction (meaning "always do it ,
used ωdo it" versus 气s doing it , was doir.g it") in the impe1如tive. In some
languages the per.如tive / imperfective distinction is clearly inf\ ectional , fvr
example in Spanish , where eve町 verb has both a Preterite and a1'1 Imperfec-
tive form. In other languages , it may be a pervasive distinction , but a deriva-
tional or lexical one , as in Serbo-Croatian. On the other hand , th巳坦坦al
/ ca ntinuouy distin ctinnJs neyer denyatioAal 0~二ieal-;己ttt-WJ且M 岳it-oeeurs ，
itjs..either in f\ ectional or marked with frcc grammatical mO!Q坠皿~uch
as auxiliaries). That is to say , wbether an actlo filSßabitual 0r not is not a
distinction that is made lexically or derivationally. Thc:.，r姐姐旦&-t二……bi邸'
-alitv does not combine wit~semantic CQID.DIDlents denotin2 e~‘ nT
l'Il'"ti创刊芒iß...S二 ehω w-ay ö i) LU P且 G 兰兰叩~仗?厅空吧~引叶与1-，四hi l'"h is
to sav that habitualitv is n口t l'I S TeJevant to the meanin2 of a verb as some of
the other concepts that are classified a~~spec~s. l1
An examination of the different expression possibilities for aspect shows ,
then , that while related aspectual notions may have more than one expression
type , there is by no means complete free
lUL MORPIIOLOGY

lexical dcrivational inflectional

telidatelic X X
semelfactive X X
stative/active X X
mceptíve X X
lteratlve X
perflimperf X
habituallcont X

Table 6: Expression types for aspectual notions.

Not only do generality and predictabi1i ty ùf meaning condition possible mor-

phological expression types , b吨1t even within the conceptual domain of aspect ,
there are differences in the relevance of aspectual notions to the verb , in
that there are differences in the extenLt!L which an aSDe~lual modification
旦旦旦tivity or event prodllce:> a_ distinct activity or event~

5.2. Number in verbs

As mentioned several times , agreement categories have less relevance
for a verb stem than any other inflectional categorie~ ， and person-agreement
categories do not have lexical ùr derivational expression , cxcept in the rarest
of cases. However , agreement for number of the subjcct or object appears
to present quite a different ca!-e. 12 Th ere are some languagcs in which number
distinctions are lexicalized in verb stems. A very cI ear example is !Kung , a
language ofSouthern Africa , which has no inflectional agrc巳 ment categories ,
but does have a small but central group of verbs which has one form in a
sentence in which the absolutive noun phrasc (the object of a transitive or
subject of an intransitive) is singular and anothcr form when that noun phras巳
is plural. When the verb has a plural form , the plural suffix -si on the noun
is optional (Snyman 1 叨0: 124 , 131-132): ([!], [1] , [丰] and [111 are ingressive
consonants)
migun!ao ‘ 1 take the bow'
mi n I'hwi n!aosi ‘ 1 take the bows'
n!eu !o'ag!heï ‘ The elder breaks the stick'
n!eu kx'oma g!heisi ‘The elder breaks the sticks'
kx'aollul阳na ‘ Kx'ao hangs up the shirt'
kx'ao g 11 ao hemasi ‘ Kx'ao hangs up the shirts'
niwa !ei ‘Th e cat dies'
 LEXICAL / DERIV ATIONAL / INFLEC寸 IONA I. ('ON 门 Nll lJ M 10\

nlwa !ao ‘ The cats die'

n!eugli ‘ The elder goes ou t'
n!eug!e'i ‘ The elders go ou t'
xeisu ‘ The loaf of bread Ii es fla t'
xeig丰 a ‘ The loaves of bread lie flat'
Another language with a similar phenomenon is Ainu , which , like
!Kung , döés nothave inflectional expression of agreement on verbs , but
does have lexical and derivational expression for some verbs of the number
of the absolutive noun phrase. Th e following lexical pairs of verb forms are
given in Batchelor 1938: 121-122:
singular plural
a at ‘ tobe ‘
a rok ‘ to si t'
am amba ‘ to carry'
ralge ronnu ‘ to kill'
ek araki ‘ tocom巳'
arapa paye ‘ togo'

Th e majority of plural verbs are formed , however , by the addition of the

suffix -pa to the singular form:
ama amapa ‘ to put or place'
aship ashippa ‘ toflower'
heashi heashpa ‘ to begin'
hekatu hekatpa ‘ tobeborn'
oboso oboshpa ‘ to pass througtí'
rai raipa ‘ todie'

In Diegueño and Kwakiutlplural vcrbs arc derivationally rclated to

singulars by a variety of irregular reduplicative , stem-changing and affixation
processes. For these languages , the analysis of meaning makes it cI ear how
plurality of subject or objcct can affcct thc mcaning of thc vcrb stem. Boas
(1947: 246) explains that in Kwakiutl there are three distinct types of plurality
for verbs: "0ne indicating sev旦丛~拎二二;二二二 SQRG iDdicatiDg aD asti包
也ωrring at the same time in different parts of a unit: and a third , expressi[l g
repeated action."
mede'lqwela ‘ it is boiling'
meηnede'lqwela ‘ many are boiling'
ma气 'mdelqwda ‘ is boiling in all of its parts'
 104 MORPHOLOGY

mEdE'lxumEdE'lqw El a ‘ it is boiling repeatedly'

tE'nk' El a ‘ it is sizziing'
饨 'ηE'nk' El a ‘ many are sizzling'
tE'ntwk' Ela ‘ it is sizzling in all its parts'
twx' twk' El a ‘ it is sizzling repeatedly'
E主~ examplcs show that p1ma1i 勺I in v rh 1Tto勺I expJ"eilì ~Ol'@ than the
-:lI ('O.

anumber of the subject or objec t. Plurality of action may involve either dis-
tribution or iteration of the action. These various plural notions may be
expressed by a single morpheme , as in Pawnee , where the prefix wa: may
signal a distributed activity or state , an iterative action , a distributive plural
object , or a dual an d/or plural subject (Parks 1976: 279). Consider these
examples:
/wa: + wiu:s/ wa:wiua ‘ to defecate here and there'
Ira + wa: + hak/ rawa:hat 'to pass to (various people)'
/wa: + u/ wa:?u ‘ to give (various things)'

Freeland 1951 observes a similar phenomenon in Sierra Miwok. She says

(Freeland 1951: 112):
Ordinarily , in Miwok there is no expression of plurality in the verb apart
from person. Th e transition between the idea of discontinuous iteration and
th剖 of plurality of subject or object , however , is very easy. ~哩卫且血，ese
iterative verbs that a陀 transitiv 、 in meaning convey quite definitely th晴 idea
、巨王汪二句叫J

Her examples show the distributive suffix + i: +:

po?a:1 +‘ toslitopen'; po?:al + i: +‘ to slit open severa l'
ma?ta +‘ to kill'; ma?:at + i: +‘ to kill severa l'
ha?ta +‘ totoss'; ha?:at+i:+ ‘ to toss away repeatedly or several'
Detailed semantic information about !Kung and Ainu are not available ,
but the Kwakiutl , Diegueño , Pawnee and Miwok data show that number
can be much more than just an agreement categ。可 when it is expressed
derivationaI1 y. In f~fh.!his type of verbal plurality is more Ii ke an a旦旦!.J旦
that it has an effect on the inherent meaning of the verb. It invólVes meaning
components other than aspecturuones , however , implýìììg as it does , particu-
lar arrangements of the absolutive argumen t. T吧is i旦旦旦旦th旦旦些gWhpn ， ht
shows clearlv the difference between lexical and derivational meanin
the Qne luuuoi. íiRQ jQflect10nal meãïiírll! on the ot~OLWhen numl
 LEXICAL / DERIV ATIONAL / INFLECfIONAL CO~INUUM 105

Observe also that inflectional number agreement is often redundant

eve币面he ∞ntexfOnñë αause ， SIO臼 rneveroa1 毡四~t is requiredeven
咀且旦旦叩Dei' ts given in ttîenoun phrases. Th is is not so with derivational
or lexical num5er. Snyman notestor !Kung that the plurahzmg suffbt for
nouns , -Sl , IS not necessary in the examples given above , since the verb
indicates plurality. Boas says about Kwakiutl , that plural verbs are not used
if plurality of noun phrases is indicated by a numeral 0 1" other quantifier.
Unlike inflectional number , lexical and derivational plurality is not useù
redundantly. Th is indicates that the amount of meaning supplied in this wa}
is much greater than that supplied by inflections. 1 conclude that even though
、、 there are cases of what seem to be th~ame conceots havin e: derivational
and inflectional expression , it tums out that they are not precisely the same ,
Jb,!t onlv related CQ，!lcepts. 飞

6. Compounding and incorporation

Two morphological expression types related to derivation are com-

pounding and incorporation. Morphologically-complex words created by
these processes differ from those created by derivation and inflection in that
such words cannot be analyzed as consisting of a stem or T0 0t plus affixes ,
rather they contain more than one stem or roo t. Th at is , the elements com-
bined in these formations are not lexica! plus grammatical , but rather two
or more le吨ical elements. Compounding can be easily ilIustrated in English ,
where noun - noun , adjective - noun and noun - verb combinations occur.
For example:school bus, black boud， babJVisit-NEP旦-noun and adjective
·咱刨n combination r. differ from regula r R9 1:1 R ph.rases bUl -res旦旦ble
monomo巾 hf'mlc 吧巳出U皿血四把旦Fss is on the fi
t~compound ， while for phrase这 th.. m~<.:.t1"P四 on the last lexical el叫ent.
Noun - verb compounds differ from verb - object phrases in the position of
the noun.
The term "inco甲oration" has been used to ∞ver a variety of verb-for-
mation phenomena ranging from processes very similar to compounding to
processes that are much like derivational morphology (Sapir 1911 , Woodbury
1975 , Sadock 1980 , Rudes 1984). Typically , "incorporation" refers to thc
fusion of the nominal patient of the verb with the verb , but often two verb
stems can be fused as well. Both of these situations can be ilI ustrated with
examples from Tiwi , an Australian language (Osbome 1974: 46-48). Consider
飞hese examples:
106 MORPHOLOGY

ji + m;;ln + ta lJ kina = he + me + steal ‘ He stole it from me'

ji + m;;ln + alipi + a lJ kina = he + me + meat + steal ‘ He stole my meat'
ji + m;;l ci + lJ ilimpa lJ;;l + ra lJ kina
hc + me + sleeping + steal ‘ He stole it from me while 1 was asleep'
ji + m;;lni + lJ ilimpa lJ + alipi + alJ kina
he + me + sleeping + meat + steal ‘ He stole my meat while 1was asleep'
Morpheme strillgS such as these are considered to be instances of noun-incor-
poration because of the position of the noun between the pronominal prefixes
and the verb. A non-incorporated noun object would follow the verb , and
the pronominal prefixes would be attached directly ωthe verb. 1!.i旦啤旦l
for an incorporated noun , like a derivational morpheme , to occur closer to
I!比 verb than the inflectional moroheInes.
Compounding is distinct from all other combinatory processes of lan-
guage in that it has the characteristics of both syntactic and lexica1expression.
Compounding resembles syntactic express防π1ñ that thellnits combined also
always exist independently as words - that is , they are complete both
phonologícally and semantically. Compounding resembles lexical expression
in that the resulting unit is a word , and the meaning of this word is not
prectictable from a summation of the meaning of its parts. Even though
compounding may be productive in the sense that new compound,; are freely
created , the results of the compounding process ar~ lexicalized 2 n d ut勺
tend g望自~Iy lose their seman t. ic and phonological transp~~y.
•Th ere is a diachronic relation between compounding and derivational
morphology , in that one elemcnt of a compound may become a derivational
affix if it occurs in a large numb~τof combinations. For instance , the Modern
English adjective and adverb-forming suffix -ly , as fo'md infriendly , manly,
king纱， and so on , developed from an earlier compound similar to the modern
compounds with like , such as child-like , god-like and phantom-like. Ac币ord­
ing to the Oxford English Dictionary , its meaning generalized from that of
"having the appearance of' (the cntity described by the noun-stem)ω"hav­
ing the qualities appropriate to" (the entity described by the noun-stem).
Compouoding is oot a proαss which alllaoguages use. Some languages , e.g.
Fr毡 och and Spanish have little or no compouoding , while others , e.g. German
and Dutch , make extensive use of compounding.
二lrlsome languages~e incorooratjoo of a patient noun into t~bal
40mplex resembles compO Jl oding , wh jJ e in others it appears to be
 LEXICAL I DERIV ATIONAL I INFLEcrIONAL CONTINUUM 107

and verb that are fused both also exist as independent words in other construc-
tions. In other languages , such as Quileute and Tiwi , the incorporated ele-
mer.ts do not always have cognate free morphemes. For instance , about Tiwi ,
Osborne (1974: 48) says that the "incorporated and free forms are generally
not cognate". Andrade 1933 estimates that about 60 pcrcent of the incorpo-
rated forms in Quileute are not ∞gnate with free forms. In such languages
there is a closed stock of st~ms that can appear incorporated噜 al1d many of
tñe百ëãiinQt occur independently. In this respect , then , they resem 1:>1e i1eIi-
vational morphemes.
ln addition , there are often mOfl~ho-Dhonological differences between
a free and mcOiporated form when they are co~nate.....FOr example , Tiwi
body parts , in incorporated and free forms:
mumu- mumuta ‘ back'
agk:)li- jiIJkala ‘ upperleg'
pula- jimpula ‘ knee'
m:)l:)- malampwara ‘ foot'
m:)r:)- murupuaka ‘ ankle'
Th e semantic domain of the incorporated noun is usually restricted. For
instance , in Pawnee , nouns referring to body parts , natural phenomena ,
foods and cultural products are regularly incorporated. In fact , terms for
body parts are almost always incorporated. On thc other hand , personal
names (of individuals or tribes) , kinship tcrms , pcrsonal 110uns (man , child ,
etc.) , names of animals , and names of particular specics (of tree , for example)
are not usua l\ y incorporated (Parks 1976: 251-252).13 Further , in some
languages there are restrictions on the verbs which may take certain incorpo-
rated nouns. For instanc哩， in Tiwi , body part names can be incorporatcd
into only three verbs , -ni ‘ hi t', -na ‘ grab' , and -kuwirani ‘ burn\Languagcs
with compounding do not havc rcstrictions of this sor t. In languages where
incorporation has fewer restrictions it uears a stronger resemblance to com-
pounding. In languages where there are more restrictions , it resembles deri-
vational morphology. Indeed , for Greenlandic , both Sapir (1911) and Sadock
(1980) liken verbs which take incorporated nouns to verbalizing suffixes.
Co~ounding ， incorporation and derivation are 00 a CQAliA\l巳....，.均
:在号公æ-g士 A8Fality of meaning. The elements that enter into a compound
do not differ in gβnerality from the same elements when they occur free ,
except that nouns in noun - verb compounds havc a generic function rather
'.han a refercntial one , e.g. baby in babysit docs not rcfcr to any particular
 -Anuoo
MORPHOLOGY

baby. Th e same applies to incorporation 一 the iAGQ t:porated AO'l A ba~a

generic rather than a referential functÎon and moreov问， oftep implies
catego可 or ltem rather than a particular item~such as liquid versus m叩h
syrup) (Woodbury 197可. Conslder as an example , the commentary by
Andrade (1933: 181):
Th邸， for example , our word hat can be expressed by the free morpheme
tsiya.pus or by the post-positive (=incorporated form , JLB) -dist'c , but the
latter can also refer to a cap or to some kinds of head-dress , if no specific
hat has been mentioned in the contex t.
Recall also Parks' (1 976: 251) comment on Pawnee , that incotpota与A "-"\0'1四S
denote basic-level classification. Th at is. a~ 组 mρfor 组 nl'l rti~lI l l'1 r !:Df'!de5l. Æ
Ww JlI not be mcorporated.
飞 While ::ompoundmg is free of kxical restrictions , the ability of a 0011""
or verb to enter into an incoæorated copstruction dej>ends in part on the
, relevance of tbe meaping gf tbρnnllntn 舍b l'l t of tbe verb.Thus Sapir suggests
that
...what may be called typical or characteristic activities , that is ，也a丛jn
ybjcb actiyity and nhJ ect 8 rp fonnd regu)arly conjojpo叫树叩回回锦 (e.g.
rabbit-killing , looking f时 a trail , setting a net) , tend ωb王三皿四皿.ö by
verbs with incorporated objects , whereas "accidental" or indifferent
activities (e. g. seeing a house , finding~a stone) are rendered by verbs with
independent , syntactically determined nouns.
(Sapir 1911:264). Th us tbe meaning of the noun and that of tbe yerb ~ombine
in such a wav~ 句 describe a sinele di 5l.crete and cllltural1 v salient actiyi ty.
As Parks (1976) describes derivational in∞rporation in Pawnee , tbc..且Q.un
builds un tbe meanine of tbe verb Tb e reslllt is a new 叩开''11.''由旬，1;仔erent
~ing.14
ICompounding , incorporation , derivation and inflectiQn are on a cq11-
ttinuur而ñ.lnwhich
臼 com、p、011
川'町冒阳1iDg..iS
‘ t巾
b、le 什
fr哇巳凹虫st...&n川.士~史~士=兰~士兰士ed..an
dφpeen)class of ite
怡em町
m!:旦 with the...ri心训
;t
归虹皿叩鹏ε蹦蝴时亏A 且￥四叫 01四s-ttb
忡、回ρm
…兔州
llesεm
创 s剑t 也蹬旦旦旦tem凹监血e
E坐1旦~~~!!act and gen_eral of meanings) Languages di他r with respect to the
extent to 面hìêh they make use of ttrese different methods of combining
mo甲 hemes. Languages which a l10w in∞甲oration are not common , but they
are of considerable theoretical interest , since they allow the fused expression
of a greater variety of semantic notions than other languages. In Part 11 ,
where we examine the semantic notions that may be expressed morpholog-
ically on verbs. we will see tbat wbile certain concepts , such as imperative
 LEXICAL / DERIV ATIONAL / INFLECTIONAL CONTINUUM 1ω

are extremely frequent as verbal morphology , other concepts , such as deontic

modality , are expressed morphologically only in languages with extremely
high degree of morphological fusion.

7. Con c/usion

Th is chapter considers the derivation / inflection distinction as a ∞ntìnu­

ous scale , rather than a discrete division of expression types. 1 have ar.伊ed
that there is a correlation between the type of meaning expressed by a mor-
phological category and the form its expression takes , and that these ∞rre­
lations are predictable from the semantic parameters of relevan臼 and gen-
erality. 1have also argued that inflectionally- and derivationally-related forms
split apart from one another under exactly the same conditions , but these
conditions are met more often for derivation than for inflection. ln addition ,
we have seen that compounding and incorporation are related to derivational
morphology. The comments offered here are intended as useful ways of
viewing the distinction between derivation and inflection , and also as a way
of understanding differences betwen related concepts in general categories ,
such as aspect. In the next chapter 1 will present a model of lexical represen-
tation that can accomodate some of the observations made here.

NOTES

1. Clark and Clark 1979 point out that ordinarily a derived innc '1ation that is preciscly
synonymous to an established word will beωnsidered unacceptable and not survÌ\'e This is 由e
m酬I there are not more 阴irs of the type unhappy / sad , i. e. unpretty. ung/'QC，φd， unbri.彷tare
pre-empted by established lexical items: ug.纱， clumsy, dull.
2. Generality is probably what Bloomfield meant by "rigid parallelism" of underlying and
derived form. He also re∞gnized that even inflectional categories are not always totally genetal.
3. With a few verbs , 'wish' , 'wan t' ,‘ hate' and ‘ bleed' , this same suffi 、 givr-s a future meanlOg.
Its use with 'bleed' is an exception to the restriction that it be uscd onl: , with stative vcrbs.
4. Other variables coded by Pagliuca were stress placement and the lengt" of ti01e the word
has been uscd in English.τbese variables will not be discussed here.
5. Th e change of [i) to [E) occurred early , affecting the wO l"ds that had a high frequency , and
is independent of the [iy) > [i) > [i) change.
6. Th e c1 ass of verbs characterized by the devoicing of a final/d/, i. e. , bend, bent; spend,
spent; build , built has been shrinking faster than any other English irregular ve由c1ass. In the
experiments reported on in Bybee and Slobin 1982 , we found that this c1a皿 provoked 由.e highest
rate of regularization of any English verb c1 ass.
110 MO t{ PHOLOGY

7. David Za ~er brought this cxample to my attention. The mo叩heme segmentation is only
approximate bçcau回 the fusional morphophonemics of Greenlandic are highly ωmplex and
need not concern us.
8. Of eourse there are languages with with inflectional infixation , for instan白， thelanguages
of the Philippines , rut the fact that no examples occurred in our 50 language sample shows that
inflectional infixatio咽 is very rare
9. The diminutive category that Anderson refers to is part of a noun classification system in
which almost e四ηi noun can be used with a class suffix denoting diminution (Arnott 1970). 'fh e
diminutive .:ategory here does not necessarily~ρactual .ize nf the entitv. which wn咀Id
be 主!l jnherent 俨h 组lli'teristic or the entity , but is used also as a pejorative which indicates the
s吃faker's attitud陆 about the value of the cntitv ioτ…...:叫
NOn=obligaωry affixes expressing temporal notions , and applicable both to nouns and verb喝
(ωmparable perhaps to English terms such as "former" or "ex-") occur in Kwakiut l. As might
be expected , their s.:ope and function is quite different from that of tense inflections. Such exam-
ples seem to be infrequent cross-linguisticálly
10. 1 am not certain wh晴th晴 r iteration and inception are frequent as components of lexical
meanmg.
11. Habituality is apparently more often conceived of as an inherent characteristic of agents
than of activiti睛， SIO∞ people are often referred to with terms denoting habitual activities , such
as jogger , smoker , etc.
12. The example of number in verbs is also discussed in Bybee 1984
13. Parks \976 describes two kinds of incorporation fúr Pawnee: one in which the inco甲orated
noun serves as the subject or object of the clause , and one (termed "derivational") in which the
noun builds up the meaning of the verb , but does not really serve as an argument in the c1 ause.
Both ofthese types of incorporation are subject to similar semantic restrictions on the ineorporated
noun.
14τne meaningofverb-adjunct incorporation was not discussed in the Iiterature 1consulted.