?tekauer Pavol, Lieber Rochelle. Handbook of Word-Formation

Подождите немного. Документ загружается.

290

PETER A

KEMA

D NEELEMAN

.2 Haplology

hereas the most basic cases of competition between morphemes are governed

the Elsewhere Principle onl

, there are various t

pes

f com

etition that involve

ther conditions. Once such condition is what Menn and MacWhinne

(1984) call

eated Mor

h Constrain

a condition disfavourin

acent morphemes that

ave an identical (or very similar) fo

m. Suppose that there are two adjacent

ositions P

in the morphosyntactic structure of some word. Suppose,

furthermore, that if we look at the s

ecifications of the mor

hemes in the lexicon of

he language and simply apply the Elsewhere Principle, we would expect

obe

pelled out by

while P

would be spelled out by

dentical form

or if

ends in a string identical to m

languages may choose a

pell-out different from

-m

in order to avoid a violation of the Re

eated Mor

onstraint.

There are four strate

ies in which lan

es deal with violations of the Repeated

Morph Constraint. The first is to simply tolerate the violation, as happens in the

English comparative of

cleve

which is

cleve

r. The second is to rule out s

ell-out

f the morphosyntactic construction in question altogether. This applies to a case

ike English

uglil

for which a circumscription is required, such as

n an ugly wa

In addition to this, one of the offending morphemes can fail to be spelled out

eparately, or it can be spelled out by a form which is not normally the optimal spell-

ut for the feature combination in question. We will now discuss some examples of

the latter two strate

ies.

A simple case of non-spell-out is presented by the English genitive of plural

ouns. Since both the genitive and the plural are marked by -

the genitive of a

lural noun should end in

-s

But in fact, such ex

ressions end in

-s

(

see

(

))

ote that there is no problem in the genitive -

attaching to irregular plurals (see

(

7d), so that we indeed seem to be dealing with a case of haplology, rather than with

orphological incompatibility of plural and genitive. Note, moreover, that the

enitive -

can be attached to certain underived words endin

in /s/, showin

that we

are not dealin

with a purel

phonolo

ical phenomenon either (see (7e)). (The issues

nvolved are discussed in more detail in Yip 1998)

(7) a. The girl’s house

b. *The girls’s house

c. The girls’ house

ewo

n’

ouse

e. Professor

.’s lectures

A similar pattern is found in Spanish c

itic clusters. Grimshaw (1997) points out

that at least in some dialects a sequence of a reflexive and an impersonal clitic,

ected to surface as a

se se

sequence, surfaces as a single clitic instead:

(

)

(

*se

)

lava

one oneself washes

‘

ne washes oneself’

RMATI

N IN

PTIMALITY THE

The fourth strategy to deal with repeated morph cases, which consists of spelling

ut one of the offending morphemes by using an ‘unexpected’ candidate, can also be

llustrated by clitic clusters. In certain variants of Italian, in structures comparable to

(8) one of the clitics is realized by a clitic (

)

that is used otherwise for the first

person plural (see Bonet 1995 and Grimshaw 1997):

(9) a. Lo si sveglia Impersonal

MPER

wakes.up

‘One wakes him u

’

b. Se lo com

ra Reflexivese/si

REFL

ACC

buys

‘S/he buys it for himself/herself’

c. Ci/*Se si lava Impersonal plus reflexive

MPERS REF

washes

‘

es o

ese

lf’

similar case from Spanish is the phenomenon known as ‘spurious

’

(

see

Perlmutter 1971 and Bonet 1995

)

. Where one would ex

ect to find a se

uence of the

hird

erson dative clitic

and the third

erson accusative clitic

the dative is

eplaced by

a clitic that is otherwise used in various different structures

(

such as

mpersonal, reflexive and unaccusative struc

ures). Note, by the way, that this

xample demonstrates that the Repeated Mo

ph Constraint is violated by two forms

hat are phonolo

icall

similar, but not absolutel

identical, somethin

we cannot

(10) a. El premio, lo dieron a Pedro aye

the

rize 3

ACC

gave-3

to Pedro yesterda

‘The prize, they gave it to Pedro yesterday’

b. A Pedro, le dieron el premio aye

to Pedro

DAT

gave-3

the prize yesterda

‘Pedro, they gave the prize to him yesterday’

c. A Pedro, el premio, se/*le lo dieron a

to Pedro, the prize,

ACC

gave-3

esterda

ote that, although they both involve suppletion, there is a difference between

he cases in

(

)

and

(

)

. In the for

er, a violation of the Re

eated Mor

onstraint is avoided by using a clitic that is more richly specified than the clitic it

eplaces (the first person feature of

is not present in morphosyntax). In the latte

ase, a clitic is used that spells out less features than are p

esent in morphosyntax

(ar

uabl

, the clitic

is hi

underspecified; it certainl

lacks the number and

ase features present in

In the rest of this section we will discuss wh

an anal

sis in terms of

ompetition, and more specifically an OT a

ccou

nt in

h f

eev

against a set of ranked, violable constraints, may be the best way to deal with some

292

PETER A

KEMA

AD NEELEMAN

properties of repeated morph effects (in particular the cross-linguistic variation we

ee in the way the problem is dealt with).

n structures that potentially violate the Repeated Morph Constraint, various

factors come into play. The first is, of c

urse, the Repeated Morph Constraint itself

(

see (11a). We have already encountered

iti

sec

n 2

(

where we treated them as unviolable

For a start

each feature bundle in the

orphos

ntax should receive a realiza

ion in the morphophonol

ical output. We

will split this condition into

wo constraints. The first requires a transparent match

between morphos

ntactic structure and morphophonolo

: it is violated if there is a

ack of one-to-one mapping between the two (see (11b)). The second requires that

phi-features are realized by phonological material that is specified for the right

features (see (11c)). Finally, no features may be spelled out that are absent in the

orphosyntax. Thus, affixes that are lexically specified for some feature F may not

be used for inputs that lack F (see (11d)). These constraints are independently

otivated, in that they play an essential role in the analysis of various othe

uistic phenomena (note that the Repeated Morph Constraint can be seen as a

subc

se o

f th

bli

ator

Contour Principl

)

(

)

a. Re

eated Mor

h Constrain

if M

Iconicit

One element in the morphophonological structure is the realization

f one element in the morphosyntactic structure

Assi

n to each feature in the

orphos

ntax a properl

specified

orpho-phonolo

ical realization

Faithfulness

The morphophonology does not realize features absent in the

orphosyntax

n the OT conception of grammar, (11 a-d) must be violable constraints that are

anked in a language-particular order with r

pect to each other and with respect to

traint

Let us therefore consider the patterns of constraint violation induced b

the

icall

possible strate

ies to deal with repeated morphs. Suppletion with an

vers

ecified form involves the use of a mor

heme that s

ells out more features

han are

resent in the mo

phosyntactic input. This satisfies all conditions except

aithf

ess.

Suppletion with an underspecified form satisfies Faithfulness, but violates Parse.

There are two related strategies that underparse the input. The first is not to realize

ne of the offendin

morphs. This violates Parse, but also Iconicit

. The second is to

ot realize the morphos

ntactic input at all; t

at is, to use the so-called null parse. Of

ourse this violates Parse, but it ar

uabl

does not violate Iconicit

: since there is no

orphophonological structure, the morphophonology cannot be non-iconic either.

further strategy is to associate the two morphosyntactic feature bundles to a

ingle phoneme (whose form will of course b

suitable to spell out both, given that

ORMATION IN OPTIMALITY THEORY

293

we are dealing with repeated morphs). Thi

coalescence strategy satisfies Parse, as

well as Faithfulness

but it violates Ic

nicity, as it involves two-to-one mapping

between morphosyntax and morphophonology.

Finall

, repeated morphs can be tolerated, somethin

that obviousl

violates the

epeated Morph Constraint, but none of th

iti

ari

ous co

traint

iolation patterns are

iven in Table 1. (The asterisk between brackets in the column

r Par

that th

lati

f thi

sco

traint that i

duced

by the null parse depends on the number of features that are present in the

orphosyntactic structure).

The ranking of the four constraints determines which strategy is employed. The

rucial factor is which constraint is ranked lowest.

(

)

If this is Faithfulness, we will

et suppletion with an overspecified form. (ii) If it is Parse, there are two

possibilities, namel

suppletion with an und

ecified form and avoidance. Which

f these is chosen de

ends on the lexical

nventor

of the lan

e. Given that Parse

refers the s

ell-out of some features

ver the s

ell-out of none, su

letion will

block avoidance whenever there is a

honeme that can realize a subset of the

features in the morphosyntactic input. In

he absence of such a phoneme, we will get

avoidance, that is, the repeated morph con

ruction is not allowed to surface.

(

iii

)

Iconicity is the lowest ranked constraint, the best solution is to link both

orphosyntactic feature bundles to a single phoneme. Finally, (iv) if the Repeated

Morph Constraint itself is ranked lowest, the result is tolerance of repeated morphs.

iven that the four strate

ies result from the low rankin

of four different

onstraints, an OT-account alon

the lines

ust sketched would appear to be purel

descriptive. However, such an analysis has two potentially attractive properties. The

fir

t i

that it r

es ou

n a

trat

y. This is because it incurs violations on

both Iconicity and Parse. Since there are strategies that violate only Iconicity

(namely coalescence) or only Parse (namely avoidance and suppletion with an

nderspecified form), deletion will not be

he optimal strategy under any ranking of

he constraints (in the terminolo

of OT, the candidate involvin

deletion is

Iconicit

Par

Faithf

ess

letion

(

overs

ecified

form

)

Suppletion

(

underspecified

form

)

(

null parse)

(

…)

oalescenc

n * *

ran

PETER A

KEMA

D NEELEMAN

armonically bounded by the candidates involving the other strategies jus

entioned

)

. This im

lies that in all cases where repeated morphs are spelled out by a

single phoneme, this phoneme must be associated with both morphemes, rather than

with

ust one of them. It might seem that this is a difficult prediction to test, but

ecently De Lacy (1999) has provided empi

that

that th

ant

ses

deed

ve co

esce

rath

r than

second potentiall

correct prediction is that suppletion strate

ies can onl

appl

to forms that are part of a paradi

m, and not to derivational affixes,

ompounds and the like. Suppletion onl

make sense if there are morphemes whose

feature s

ecification is either a su

erset or a subset of one of the feature bundles

present in morphosyntax. Such elsewhere relations typically hold of functional

hemes

(

see section 2.1

)

, but not of lexical ones. Indeed, as far as we know,

epeated morph constructions involvi

g derivational morphology or compounding

are either tolerated (as in English ex-ex-presiden

Afrikaan

oon-tjie-tjie ‘

an-

DIM

’

) or avoided (as in English

uglil

kop

je-je

‘cup-

DIM

’)

but they never involve suppletion. (Strictly

speaking, we could expect to find cases

f coalescence with lexical mor

hemes, but in order to test this one has to find

sequences of semanticall

different but phonolo

icall

identical derivational affixes

hat are in principle grammatical. We have not been able to do so.

)

ithin a single language, not every repeated mor

h context will be dealt with in

he same way (as will be clear from the English data mentioned in the discussion

above). One might hope that this variation is partially due to the fact that lexical and

functional morphemes will behave differently in repeated morph contexts, as

ust

explained. In the worst case, the Parse and Faithfulness constraints mi

ht have to be

split into more specific constraints that mention subcate

ories of features, o

ndividual features in some extreme cases. It would take us too far afield to explore

his issue here (but see below for more discussion on splitting constraints in this

way).

2.3

kedness

e now turn to another type of competition between a null form and an overt

ealization of an affix. In the relevant cases, the opposition between the two forms is

sed to mark certain properties of the s

ntax, in particular the markedness of

particular phi-features in an ob

ect and

r sub

ect. The phenomenon can be observed

with both case and agreement.

subs

tanti

lit

rat

hat counts as a marked subject o

bject. In a seminal paper, Silverstein (1976) argues for a universal markedness

ierarchy along the following lines:

(12) 1 > 2 > 3/proper noun > 3/human > 3/animate > 3/inanimate

subject

marked the lower its properties

n this hierarch

. For example,

any third person subject is more marked than a second or first person subject. In

ORMATION IN OPTIMALITY THEORY

295

ontrast, the lower the properties of an objec

the

less

marked it is. Thus

a second

person ob

ect is more marked than any third person one.

n some languages, morphological case is s

sitive to the status of the sub

ect or

ect with respect to the markedness hiera

in (12). In particular, overt cases

seem to be preferred for more marked ar

uments. In an a

olutive-er

ative case

stem, er

ative tends to be overt; in a

ominative-accusative case s

stem it is the

bject case, accusative, that tends to be

vert. In certain split case systems, then,

arked subjects distinguish themselves from unmarked ones by carrying ergative

ase (rather than nominative, which does not show up morphologically). Similarly,

arked ob

ects carry accusative (rathe

han absolutive, which again has no

orphological correlate). The answer to the question of what kind of sub

ect is

arked enough to warrant ergative case m

king differs from language to language,

as does the cut-off point for accusative markin

on ob

ects.

This variation amon

st lan

es with a

plit-case s

stem can be anal

zed as

nvolvin

competin

forms, one of which is selected on the basis of a set of

onflicting constraints – as in OT-grammar, that is. A proposal along these lines is

eveloped by Aissen (1999), who translates Silverstein’s hierarchy into a set of

onstraints that require overt case marking for particular types of arguments. The

mark

a f

eco

inati

n f

r a particular type of argument, the more

prominent the constraint requiring overt case for this argument. Thus, the following

wo constraint hierarchies obtain

(

where CM stands for ‘case mark’

(13) a. CM [Sub

, 3/inanimate] > CM [Sub

, 3/animate] > CM [Sub

3/human] > CM [Subj, 3/proper noun] > CM [Subj, 2] > CM

[

Subj, 1]

b. CM [Obj, 1] > CM [Obj, 2] > CM [Obj, 3/proper noun] > CM

[

, 3/human] > CM [Ob

, 3/animate] > CM [Ob

, 3/inanimate]

rucially, it must be assumed that the con

traints in

(

)

cannot be reranked with

espect to each other, which

ould

ive rise to lan

e-particular rankin

s of them,

since the essence of Silverstein’s markedn

ss hierarch

is that it is universal. The

onstraints can be reranked, however, with respect to a constraint that militates

against the morphological realization of case. To this end, Aissen adopts a very

general constraint that penalizes structure (*Struc). The position of *Struc in the

traint hi

rar

es de

rmin

ecu

-off

oint between case-marked and case-

ess sub

ects and between case-marked and case-less ob

ects.

Note that in this system the marki

g of case for sub

ects and ob

ects is in

principle independent. That is to say, the ordering of the constraints in the hierarchy

n (13a) with respect to the constraints in the hierarch

in (13b) has no effects. This

ndependence means that the s

stem ma

e rise to sentences with an Er

ative-

ccusative case pattern, namely when both

he subject and the object classify as

arked (the respective CM constraints mentioning their features both being ranked

We simplify the details of Aissen’s proposal somewhat. She generates the constraints in (13) using a

echnique called

ocal con

unction (due to Smolensk

1995). This does not affect the ar

umentation

296

PETER A

KEMA

D NEELEMAN

above *Struc). Languages with such patterns do indeed occur (see for instance

oolford 1997), but there are also languages in which such a case pattern seems to

be disfavoured. Let us assume that there is a constraint which has the effect that only

ne argument in a transitive clause can be case-marked (OneCase). If such

onstraint is sufficiently highly ranked, c

nflicts arise in case the sub

ect and the

ect both have properties that would normally require case marking. In that case,

he mutual rankin

of the ob

ect and sub

ect constraints becomes crucial. Suppose,

for example, that the followin

rankin

obtains:

(14) One-Case > CM[Obj, 3/human] > CM[Subj, 3/animate] > *Struc >

M[Obj, 3/animate] > CM[Subj, 3/human]

Given this constraint ranking, a third person animate sub

ect will usually contrast

with a third person human sub

ect in being case-marked. However, when a third

person human ob

ect is also present, this will require case marking as well, and

iven that CM[Ob

, 3/human] outranks CM[Sub

, 3/animate] while both are

dominated b

OneCase, this precl

des case markin

of the sub

ect.

Further research is required to explo

e whether there are case s

stems that

display these kinds of interactions. However, Trommer (2004) discusses an example

f an agreement system in which subjects and objects compete for a single

agreement slot on the verb in this way. The language in question, Dumi, favours

agreement with arguments that have features that are higher on the following two

rar

es:

(

)

a. 1 > 2 > 3

b. Plural > dual > sin

ular

Dumi does not seem to care whether agree

nt is with the ob

ect or the subject,

although object agreement in certain circumstances re

uires that an additional

arker be added (glossed as MS for ‘ma

ed scenario’). The effects of the person

ierarchy in (15a) are illustrated in (16). The example in (16a) shows that a first

person dual sub

ect beats a second pers

n dual ob

ect in the competition fo

reement, where as (16b) shows that a first person dual ob

ect beats a second

person dual sub

ect.

(

)

a. du:khuts-i

see

DUA

‘We (dual) saw you (dual)’

du:

-i

-see-1.

DUA

‘You

(

dual

)

saw us

(

dual

)

The exam

les in

(

)

illustrate the work

s of the number hierarch

in (15b).

Irrespective of grammatical function, a plu

al argument beats a dual argument in the

battle for agreement.

ORMATION IN OPTIMALITY THEORY

(

)

a. do:khot-t-ini

see-

NONPAST

-3

They (plural) see them (dual)’

b. do:

t-t-ini

see-

NONPAST

-3

‘The

(dual) see them (plural)

situation can occur in which one argument qualifies better for agreement on

ne hierarch

, while the other is to be preferred on the basis of the other hierarch

for example if one ar

ument is first sin

ular, while the other is third plural. We

ht expect that in such circumstances either the person hierarch

outranks the

number hierarchy, or vice versa. However, as Trommer notes, the situation is more

lex.

or a start, it often depends on the exact feature content of the arguments which

ierarchy carries the most weight. In the case of a second person singular subject

and an object that is third person dual or plural, it is the number hierarchy that

prevails: agreement is with the object. On the other hand, if one argument is second

person dual and the other third person plural, it the person hierarch

that is decisive:

he chosen a

reement marker is specified as second person dual. There is a wa

which this pattern can be described usin

he kind of constraints proposed b

Aissen

(see above). The idea would be to formulate a separate agreement-demanding

onstraint for every possible combination

erson and number features, and to rank

all these constraints in the appropriate order under the constraint that rules out

double agreement (call it OneAgr).

Trommer shows, however, that there is a phenomenon in Dumi that excludes

such an account. As it turns out, there is one case in which conflictin

demands

arisin

from the person and number hierarchies are reconciled b

havin

more than

ne a

reement marker after all. The crucial example involves a first person sin

ula

argument and an argument specified as sec

nd or third

erson and as dual or

lural:

(

)

a. do:khot-t-e-ni

see-

NONPAST

-1

‘I see them

lural

)

’

du:

-t-

-ni

-see-

NONPAST

-1

‘They (plural) see me’

This situation cannot be described in terms of reranking OneAgr with respect to

onstraints that require the spell-out o

certain feature combinations. One might

hink that (18) can be accounted for b

ankin

both R(ealize)[1s

] and R[3pl] above

neA

r. This will lead to a rankin

paradox, however

since there are contexts in

hich at least third person plural does not

ive rise to a

reement, apparentl

as a

onsequence of OneAgr. In particular, con

der the situation in which a third

erson

plural argument competes with a second person dual argument. As noted above,

here is only one agreement marker in this case, for the second person argument.

298

PETER A

KEMA

D NEELEMAN

This implies that R[3pl] must be ranked below OneAgr, in direct contradiction to the

nitial suggestion.

Trommer shows that the agreement patter

s of Dumi can be captured by an OT-

analysis, but that they require context-sensitive constraints of the type “Realize

agreement

eature F

in the presence of F

”

(

where

is more prominent than

n the same markedness hierarchy). Ranking

uch constraints (with respect to each

ther and with respect to a constraint like OneA

r) does

ive rise to a consistent

rammar for Dumi. We refer to Trommer’s wo

k for details

but it is not difficult to

see wh

this works: (18) indicates that both ‘Realize 1 in the

resence of 3’ and

‘Realize plural in the presence of singular’ are ranked above OneAgr. The

ression of

plural in the presence of second person dual indicates that the

grammar must also have a partial constraint ranking such that ‘Realize plural in the

presence of dual’ is ranked below both ‘Realize 2 in the presence of 3’ and OneAgr.

These two partial constraint rankings can be combined into a single ranking without

his leading to a ranking paradox.

MPETITI

N BET

EEN

NENT

.1 E

lsewhe

ecases

s we have seen

the basic case o

competition in morphology can be

haracterized by the Elsewher

Principle: a more specifi

form is preferred over a

ore general one where both are in principle grammatical. By definition,

ompetitors are those forms that can be used to express the same concepts. It is

possible, therefore, that competin

structures are

enerated in different components,

n particular morpholo

and s

ntax.

well-known example involves the En

lish comparative affix -

which must

attach to short (maximally bisyllabic) ad

ectives (see (19a,b)). This morpheme is in

ompetition with the syntactic modifie

which can in principle attach to both

hort and long ad

ectives, and is therefore

he more general form. In the context of

hort adjectives, the Elsewhe

Principle dictates that -

blocks

(

see

(

19c,d

))

(

We add

(

19e

)

to show that in circumsta

es where the Elsewhere Principle does not

apply

can indeed modify short adjectives.)

(

19) a. Bi

b. *Intelli

ente

c. *More big

d. More intelligent

e. Bigger means ‘more big’

his classical application of the Els

where Principle demonstrates that a

orphological complex can be in competition with a syntactic phrase. However, the

effects of the Elsewhere Principle are not limited to morpholo

blockin

ntax. As

ORD

FORMATION IN OPTIMALITY THEORY

299

(

Chomsky 1995) can be seen as an insta

e of blocking. Because a lower landing

an attra

t a

subse

f th

that a higher landing site can attract, it is, in

his sense, more specific than the higher one. Consequently, movement to a higher

andin

site is blocked where movement to the lower landin

site is possible. More

elevant in the present discussion are cases in which, as opposed to the one in (19),

he specific form is s

ntactic and the

eneral form morpholo

ical

The En

lish

imple past, for instance, is morphological. Yet, in the perfect, it is blocked by a

yntactic periphrastic construction, which is more specific as it roughly expresses

ast with

resent relevance.

nother case of competition in which a more specific syntactic construction

blocks a more general morphological for

oncerns the negated form of the first

person singular of the verb

to be

as discussed by Bresnan

(

1999

)(

the account below

s somewhat simplified and involves a sli

htl

different interpretation of the data as

ompared to Bresnan’s account). Normall

, a sentence with a finite form of

to be

an be ne

ated b

morpholo

ical means, namel

addin

n’t

to the verb (see

wicky & Pullum 1983 for arguments that

n’t

is an affix). There is a gap in the

paradigm of these negative forms, however:

n’t

cannot be added to first person

ingula

f the Elsewhere Principle could only compare morphological forms, we may

expect that, given the abs

nce of the s

ecific for

amn

’

, the more general for

’

is used. In inversion contexts, this is indeed the form that occurs:

(21) a. *Amn’t I working

b. Aren’t I working

It is important to realize that, in addition to the forms in (20) and (21), En

lish

allows a s

ntactic realization of ne

ation that

compatible with (a

)

(

22) I’m not working

This syntactic combination of

not

expresses the concept “negation of the

first person singular of

” more accurately than the more general

en’t

and will

ence block the latter if the Elsewhere Principle applies across components. We see

his happenin

in sentences without inversion: (23) is blocked b

(22).

(23) *I aren’t working

The question, then, is why inversion should have an effect on the realization of

he negated first person of

Since inversion is an operation of head movement

(Aux-to-C movement), it must strand nonaffixal negation. This rules out (24a) (in

OT terms, GEN cannot generate (24a),

thi

ill n

eve

(

20) a. *I amn’t workin

d. we aren’t workin

ou aren’t workin

c. s/he isn’t workin

f. the

aren’t workin