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

Подождите немного. Документ загружается.

ELLEN M

KAI

(

)

Taiwán+ése

→

Tài

ése

Jap

→

pan

The result violates a general family of constraints called paradigm uniformit

onstraints. The iambic stress pattern of the base is not maintained in the derivative.

ut at least the morphology is perm

tted to create the new word.

ithin

Raffelsiefen’s analysis, it is better to move th

ess

than t

low ad

acent stresses.

H dominates PARADI

NIF

RMI

)

[STRESS]. The constraint

hat forces the phonolo

and morpholo

to compromise and create some form, M-

PARSE (Prince and Smolensk

1993), also dominates PU [STRESS]

(

22) Ranking fo

ése

*CLASH, M-PARSE >> *PU

[

STRESS

]

For

ée

however, the result is a

ap. No ad

ustment is made b

the phonolo

and the morphology suffers by having no word

formed at all. Raffelsiefen uses the

ool of Google searches to discover what kinds of new words English speakers can

form productively. Her searches turned up virtually no words of the form

baton-

eer

hat is

ée

ati

ves

ed o

ses w

ith final

ess.

Raff

udes

hat different suffixes in English i

voke different constraint rankings. Fo

ée

M-

PARSE is dominated both by the phonologically motivated *CLASH and by the

orphologically motivated PU, so no comp

omise is reached and no pronounceable

form emer

es at all.

The factorial t

polo

would predict we mi

ht also find cases where PU [Stress]

ins out and *CLASH is violated, and Raffelsiefen ar

ues that this rankin

tantiat

ée.

Finally-stressed bases are permitted to combine with this suffix,

esulting in outputs with clashing stresses but where the derived form maintains the

stress of the base. She finds many coinages like

elèctée

retì

ée.

n these cases

he morphology is not inhibited by the phonology, with the result that outputs which

would not occur in morphologically simple words are readily found in derivatives.

. LEXI

AL PH

Y AND M

RPH

Y AND IT

ILL

One of the most attractive features of

the LPM model was its unification of

apparent generalizations a

bout the phonology of cohering suffixes, their interaction

with morphology (especially their triggering of cyclic rul

lication

)

, and thei

inear order. LPM ado

ts and elaborates the

evel ordering hypothesis

(

also known

as the affix ordering generalization) of Siegel (1974). This hypothesis claims that

ohering, stress-affecting (+

boundary) affixes (called level 1 affixes in LPM) will

ccur close to the root

while non-coheri

g, stress neutral (#-boundary) affixes,

One goal of Raffelsiefen’s (2004) article is to support the original M-PARSE explanation for

phonolo

icall

induced

aps over the revision to

T proposed b

un and Sprouse (1999). These

authors have proposed that the EVAL(UATION) module of OT be supplemented with a CONTROL

odule

which contains inviolable constraints. I

there is no way to satisfy the constraints in

ONTROL, the ‘null parse’ results. This means that the morpholo

fails to parse the morphemes

nto a coherent,

ronounceable word and a

ap appears in word formation.

ORMATION AND PHONOLOGY

alled level 2 affixes, will always occur outside the cohering ones. In other words,

evel 1 affixes cannot attach to a word to which a level 2 affix has already been

attached. We can illustrate this claim with the schematization in

(

)

, where

stands for a coherin

affix and A

for a non-coherin

affix.

(

)

# … # A

…

+[root

]

…#

…

Thus, according to Siegel, Kiparsky

(1982) and other LPM work, the non-

ce o

lik

i#ness+al

er+ous

m th

t that

contain a word formed with a non-coherin

affix (

ness, #er

)

to w

h a

oherin

affix (

al, +ous

)

has been affixed. There is nothin

wron

with these

ords as far as the

art of s

eech of the ba

e to which the final affix is added

since

ous

do attach to nouns

(

person+al, danger+ous).

(

ffix ordering permits

lik

person+al+it

since both

+al

and

eve

l 1

ffix

es.

ewo

anger+ous#ness

s also fine, since it contains a level 1 suffix followed by a level 2

suffix. And words with strings of level 2 suffixes are also fine:

eamlessly,

seamlessness

. The correlation with phonology is that all the affixes starting from the

stem outward to the first non-coherin

affix should form part of the visible input to

evel 1 lexical phonolo

ical rules, while all of the affixes startin

from the first non-

oherin

affix outward will not be the tri

er or tar

et of an

such rule, and will

nly undergo postlexical rules, the ones that apply between words

n (24) I present a diagram of the workings of the LPM model, modified

somewhat from the one found in Kiparsky (1982: 133).

(

)

ntax postlexical phonolo

n this model, morphological operations occur one affix at a time. Each time an

affix is added at level 1, the form is passed to the level 1 phonology, which applies

o the string as it is currently concatenated. The form is then

assed back to the

orphology at that same level, over to the phonology again, and so on until all level

1 affixes for that word are added.

(

Hence the

⇔

symbol between the level 1

Borowsky (1993) achieves this result by ordering all phonological rules at level 2 (the word level)

ore all morpholo

ical operations at level 2.

RPH

Y PH

eve

l 1

(

ste

evel

)

boundary inflection and

derivation

;

zero-derivation

f n

⇔

tress rules, trisyllabic

axing, velar softening,

onorant s

llabification.

eve

l 2

(

word

evel

)

boundary derivation and

ompounding, zero-

ati

nouns

most inflection

ound stress

ELLEN M

KAI

orphology and phonology.) The cyclic application first introduced in

resu

from this interleaving of phonological and morphological operations at level 1. Then

he form passes on to level 2. In English, it appears that all the cyclic phonological

ules apply at level 1.

To illustrate the segregation of morphol

gical operations into levels and the

oncomitant applicability of certain phonological rules only to strings created at that

evel, consider a point about the zero-deri

ati

f n

bs ve

sus

that

erbs from nouns made in Kiparsk

(1982). As is well known, when nouns are

derived from verbs with final stress, the

preserve that stress as a secondar

prominence but add a penultimate, primary

tress of their own

as befits a noun.

Thus we find noun/verb

airs such as

cónvìct

ct,

pé

pè

rt,

tó

mènt

tò

mént.

In a derivational theory such as LPM, one would say that

nouns are zero-derived from verbs at level one, and are thus sub

ect to another cycle

f the stress rules after that affixation occurs. However, Kiparsky argues, the zero-

derivation of verbs from nouns takes place at

level 2. Since the stress rules of

lish do not appl

at level 2, the str

ss does not chan

e in denominal verbs like

ttern, even thou

h, as we noted above, pri

verbs endin

in two consonants

rece

final

ess.

us we

could say that deverbal

∅

is a level 1, cohering suffix

hil

ede

minal

∅

is a level 2, non-cohering suffix.

lassical LPM was probabl

the last model of phonolo

-morpholo

nteraction to en

a wide consensus (No

er 2004). However, it has been

ecognized for some time that it embodies at least one strong claim about word

formation that is probably not correct, certainly not for English, namely the affix

rdering hypothesis. Problems with any theory of the level ordering of affixes had

been recognized as early as Aronoff (1976). Most well-known, at least among

phonologists, are cases where affix ordering proves to be too strong a theory, ruling

inati

that actuall

occur such as

ment+a

(

governmental)

(

iz+ati

(

eutralization

)

swe

nti

ment

are stress-neutral

witnessed b

a form like

gó

vernment

with stress in the same position as its source

erb and no stress on its heavy

enult

(

com

are the underived noun a

éndi

)

, and

márginaliz

with stress four syllables from the end of the word. Bu

and

ation

are stress-affecting. The unpredicted existence of syntactic phrases inside of

ompounds, and the existence words with sub-compounds inside co-compounds and

f words with co-compounds inside sub-compounds, discussed for Malayalam by

Mohanan

(

1982

)

, form anothe

ss o

iffi

lti

r th

affix ordering

ypothesis,

as do bracketin

paradoxes like un#

rammatical+it

, an

aniz+ation

ese

and man

other cases have demonstrated that the affix orderin

pothesis

nder

enerates

Equally interesting and less often disc

sed is Fabb’s

(

1988

)

demonstration that

he affix ordering hypothesis is too weak and overgenerates. Fabb notes that affix

ithin LPM, loops permitting a return to an earlie

evel of affixation were one major proposal fo

accountin

for unexpected orderin

s and compound t

; also helpful was reduction of the numbe

f levels posited for a lan

e, so that man

affixation and compoundin

proc

sses we

aila

ver

level. However, none of these LPM proposals wa

reall

satisfactor

, and as far as I know, none

ould deal with the results of Fabb

(

1988

)

disc

ssed shortl

. See Spencer (1991: 397-420) for a

discussion of the man

proposals for d

alin

with bracketin

paradoxes.

ORMATION AND PHONOLOGY

rdering restrictions account for only a small percentage of the sequences of affixes

hat simply don’t occur in English. He lists 43 common affixes in English. If there

ere no affix ordering hypothesis, w

might expect around 600 grammatical

ombinations of these affixes, if we simpl

de su

that affix

that

r a

particular part of speech were combined onl

y with that part of

speech, and that the

particular stress requirements of the affixes like deverbal -

were met. With affix

rdering, we can pare that number down to an expected 459 combinations. But

English words actually contain only about 50 pairs of suffixes! The main reason fo

that 28

f th

eco

ffix

– m

than half – n

eve

ith

another suffix. Six suffixes combine with only one suffix; for instance -

only

attaches to unsuffixed stems or words

(

comic, metallic

)

r t

esu

ffix -

ist

(

mode

nistic

). Six other suffixes are semi-productive: noun-selecting -al

for

nstance, combines with three coherin

and non-coherin

affixes -ion

-ment

or.

able

deverbal -

and -

ness

show no selectional restrictions be

ond part

f speech. I am unaware of a response t

’

swo

ithin th

LPM m

ode

The inevitable conclusion seems to be that the affix ordering hypothesis must be

ejected, at least for English,

and with it, that part of Lexical Phonology and

Morphology that rests upon it. I suspect that

the enduring contributions of LPM will

ie rather in its recognition of a set of fundamental characteristics of lexical rules and

a largely complementary set that inhere in postlexical rules.

n contrast to their re

ection of the affix orderin

pothesis, both Fabb (1988)

and Aronoff and Sridhar

(

1987

)

continued to

eve

in an

r t

f l

eve

rderin

, namel

that the word-boundar

(

)

and morpheme boundar

(

)

affixes

ould be sorted into two coherent groups on the basis of their phonological behavior,

just as had been claimed in

and LPM. The +boundary suffixes were available to

ules of stress assignment – both those assigning stress, that is creating foot

structure; and those choosing which of these would be the primary stress, that is, the

ead foot); they were available for syllabification so that the vowel-initial ones

cou

eed

onorant Syllabification

(

inder, hinder#ing

but hindrance

)

;

llabification rendered their content visible to Tris

llabic Laxin

so that the

could

provide sufficient material to place a vowel

hree s

llables from the end of the word.

am not aware of man

discussions that place this claim under the scrutin

that the

affix ordering gener

ization has received. However

Raffelsiefen

(

2004

resents

evidence that the division of affixes into two clear groups is too strong a claim. We

eturn to her

osal in section 7.

do not know if arguments like Fabb’s go through fo

languages with highly productive agglutinative

uffixation

such as Turkish.

Kaisse and Har

us (1993) summarizes the findin

s of the contributors to the volume

tudies in lexical

phonolog

and contains a discussion of the counterexamples to the predictions of LPM that had been

amassed through 1990, when the conference on wh

ch the volume was based took place. They

onclude that thou

h almost ever

claim of LPM runs into counterexamples, the overall predictions of

he theory continu

obev

ELLEN M

KAI

. MORE RECENT DEVELOPMENTS OF LEXICAL PHONOLOGY AND

RPH

ith the arrival o

Optimality Theor

much work in phonology in the last

decade has simply turned to matters which are ancillary to the concerns of Lexical

Phonology and Morphology

As Hammond

(

2000

oints out, few students in the

nited States are exposed to more than a cursory introduction to the results of that

theory, and the question of whether Lexical Phonology is defunct is certainly worth

asking. However, the answer seems to be ‘no,’ or at least ‘not exactly’. In one

esponse to earlier difficulties, Gie

erich (1999) proposes to keep the basics of LPM

while replacin

its a

ix-driven strati

ication with a stratal or

anization that relies on

the base to which affixation applies. Roots, which do not belon

to an

lexical

ategory, are listed in the lexicon along with a list of the

oot-level affixes that

attach to each one. This listing account

for the relative non-productivity and non-

ompositional semantics of such morphology. Once a root is converted to a word by

aving a lexical category label assigned to it, it enters the word-level morpholog

and can receive the more productive, semantically compositional affixes. Since

affix

t r

tri

a single stratum, they can show both stem- and word-

evel properties and orderin

. Hammond co

ments that the critical process of

onversion to word is reall

not well explained in Gie

erich’s model – how, wh

and when does this happen, and what accounts for the fact that some bases become

ouns while others become verbs or adjectives? Nonetheless, the idea of

base

iven

tratification

ay be worth pursuing.

Lexical Phonology and Morphology also survives in newer versions where it is

married with Optimality Theory. Work by

Kiparsky (2000), Bermúdez-Otero (1999,

forthcoming,) and Rubach (2000), among

thers

uses ranked and violable

onstraints in con

unction with a division amon

stem-level, word level, and

postlexical strata. Constraints can be ranked differentl

at each stratum

and the

utput of each stratum is used as the inpu

to the next. Kiparsky (2000) goes so far

as to say that on the stem level, every stem is a cyclic domain. Thus, presumably,

ecou

dbe

a n

ew ev

ati

ates every time a stem-level suffix was

added. However, an extended treatment of English morphology-phonology

ithin

Stratal Optimality Theory

has not yet appeared, to my

nowledge.

er (2004), points out that various critical parts

f the LPM theor

are i

ompatible not onl

with

lassical monostratal Optimality Theory but also w

th the theory of Distributed Morphology (DM; see

for instance Embick and Noyer 2001). He explains

hat LPM’s inclusion of a lexical morphological

and phonological module, which operates before syntactic structure is available

makes no sense

within DM. In DM, the inputs to the s

ntax are not full

formed words but abstract morphemes whose

assembly into both words and phrases is perf

ormed by syntax and post-syntactic morphology.

onetheless, Noyer regrets the loss

f the ability to

hara

riz

lexical syndrom

that is the

ation of characteristics of lexical

vs.

post-lexical rules. I do not know of an

treatments of

lish morpholo

-phonolo

interactions in DM.

ORMATION AND PHONOLOGY

. HOW DO RELATED WORDS AFFECT EACH OTHER? THE CYCLE

TRANSDERIVATIONAL EFFECTS

PARADIGM UNIFORMITY AND THE

ne of the major preoccupations of phonologists at least since

SPE

has been the

uestion of how to ca

ture the influence that the

ronunciation of one word can have

n other words related to or derived from it via word formation or inflection. Indeed

some form of this preoccupation goes back to the Neogrammarians, under the rubric

analogical sound chang

Perhaps the earliest influential modern treatment is

hat of Kurylowicz (1949). The topic returns us to wondering in what ways

orpholo

– in this case, the relation of one word to another – can intervene in the

therwise purel

sound-oriented basis on which phonolo

would prefer to operate .

The c

cle of

especiall

when put to

ether with Brame’s (1974) restrictions

n when new word-internal cycles can be motivated, provided a constrained method

f capturing some of these interactions. To oversimplify a bit, Brame argued that a

orphologically complex word can undergo a second cycle of rule application only

f it contains an independently occurring word that contributes its full meaning to the

arger word. In Brame’s most celebrated example, Arabic, /fihim#na/ ‘he understood

us’ contains the third person verb /fihim/ (with a zero 3p. masc. sg. affix)) and thus

aintain

tra

ces o

f th

ewo

-i

tial

ess o

f th

esu

rfa

íhim, in this case b

preservin

the initial vowel rather than deletin

it, emer

ihímna

However

/fiihm+na/ ‘we understood’

does not contai

the inde

endent word ‘he understood’;

t does not contain any independent word but only the bound stem /fihim-/.

Therefore, it undergoes only one application of stress assignment, receiving only

penultimate stress, and its initial vowel must therefore be elided, yielding fhímna.

The cases under which one word can influence the pronunciation of another, then,

are relatively constrained and formally easy to state within a cyclic, derivational

heor

Cases in which a base-derivative relationshi

exerts an influence on

pronunciation are solel

ascribed to the c

cle.

owever, startin

in the earl

1990’s, for reasons lar

ortho

onal to

orphology-phonology concerns, many phonologists turned away from rules and

derivations to Optimality Theory. Most ve

sions of

T involv

only one evaluative

step – potential output candidates are evaluated simultaneously for their satisfaction

f constraints on pronunciation or perceptibility (markedness constraints) and fo

heir satisfaction of the requirement to r

emble their underlying representations as

losely as possible (faithfulness cons

raints.

)

There are no intermediate

epresentations which can form the input to a second round of c

clic rule

application. Thou

h, as we have noted, there are versions of Stratal OT that maintain

ross architecture of the derivational

PM theor

and can th

refore recapitulate

ome results of a level 1/level 2/

ostlexical division, most a

roaches to base-

derivative resemblances in

T have eit

er relied u

on out

ut-out

ut constraints that

See Lahiri

(

2003

)

for a recent introduction to the concept of analo

in lin

uistics.

However, as we shall see shortly, not every case where we might expect cyclic effects necessarily

xhibits those effects. Sometimes complex wo

ds are stressed as if the

were simplex.

ELLEN M

KAI

nforce resemblances among the pronunciation of related words; or they have

mployed alignment constraints that enforce a match-up of phonological boundaries

(such as feet and syllables) with morphological boundaries. An early instantiation of

he latter approach can be found in Kenstowicz (1995). His analysis encodes the

ffect of the stem-suffix boundary on stress in Indonesian with a constraint requiring

he right edge of a stem to coincide with the right edge of a foot. The high ranking of

his constraint results in different stress patterns for morpholo

icall

simple and

orpholo

icall

complex words.

enua (1995, 1997) pioneered the use of output-output correspondence

onstraints to capture cyclic phenomena with a

non-stratal OT

In B

a’

Transderivational Correspondence Theor

a morphologically derived surface for

suc

h a

còndènsátion

s more faithful if it closely resembles the base on which it is

formed

in this case

condénse

which has a full vowel and a stress on the syllable

dense

One of her most accessible examples deals with the nickname pair

Larry/Lar. The truncated form,

ronounced

[

ær

]

in dialects where the long, source

form is

ronounced [

læ

]

, violates an otherwise

eneral restriction a

ainst

autosyllabic

[

]

The

ressure for the truncated f

rm t

ese

utweighs the phonological markedness constraint. McCarthy (in press) contains a

ucid comparison of various OT treatments of the analogical influences one form can

ave on another. He points out that Benua’s Base Priority Principl

all

ows

nfluences from derived forms to base forms,

ust as the phonological cycle did. On

he other hand

Uniform Exponenc

(

Kenstowicz 1996

)

and

nti-Allomorph

(

Burzio 1996), which require consistent realization of morphemes in all thei

phonolo

ical properties, allow influence in both directions. McCarth

ues tha

both kinds of correspondence constraints are needed.

Steriade (1999) introduces another allomorphy-minimizing, paradigm uniformity

rinci

le, Lexical

onservatism

“Newly coined forms are penalized if they do not

losely resemble already existing forms.” She points out that the English level 2

affixes generally obey lexical conservatis

uch better than level 1 affixes, though

they often do so at the expense of phonol

gical well-formedness. Thus we recognize

ínvalidism

as being related to

ínvalid v

a th

iti

f a l

eve

l 2

ffix

bec

use

aintains the stress of the stem in isolation. The strin

of four unstressed s

llables

that results is the price the phonolo

s for the success of the morpholo

aintaining identity between base and derivative. The apparent level 1/level 2

distinction, she argues, is an artefact: so-called level 2 forms are just forms based on

mpoverished paradigms

where there is no phonologically preferable form on which

a n

ew de

ati

ve.

How does one handle the somewhat opposite fact that cyclic effects do not

always occur even where one might expect them to? Pater (2000) is one of the most

developed accounts of En

lish stress

nd c

clicit

written recentl

within

Optimalit

Theor

. He

uxtaposes the followin

cases, where the examples in (25a)

and

(

)

show an effect of the base on the derived form, while those in

(

25c

)

and

(

)

osuc

Pater bases his transcription of non-reduced (secondarily stressed)

vs .

educed vowels on Kenyon and

Knott (1953) and, where there is disagreement, on Webster (1981) as well. For a few of the second

ORMATION AND PHONOLOGY

(

)

condénse

còndènsátion

órm

ormátion

exhórt èxhòrtátion trànspórt trànsportátion

contést

còntèstátion consúlt

cònsultátion

imp

tion s

nt s

gment

tion

àugmént àugmèntátion trànsfórm trànsformátion

àuthéntic àuthèntícit

n the (b) cases, preserving som

stress on the syllable before -

átion

resu

in a

tress clash; we find ad

acent stressed s

llables. We know that phonolo

ies,

ncludin

that of En

lish, prefer to alternate stresses rather than toleratin

or creatin

lashes. Pater speculates that as words become more lexicalized, familiar, an

established, the pressure for the base they contain to influence their pronunciation

educes and they become more likely to be treated as the phonology would have

iked to treat them all along. Thus he asks us to compare the more frequent,

everyday word

nformation

ith

exho

tation.

nformation

the second syllable

educes

so that

nformation

is s

essed

lik

ved wo

Pennsylvania

gorgonzola. The foot structure [ìnfor][má]tion is more optimal, phonologically

peakin

, than in [ìn][fòr][má]tion would be

But

[

èx

][

hòr

][

tá

]

tion is chosen ove

[èxhor][tá]tion because onl

exho

tation

s lexicall

marked as sub

ect to a special

igh-ranked constraint calle

Ident(ity)-Stress-S

Words which show the cyclic

effect of stress preservation are lexically marked with a diacritic that makes the

subject to this output-output constraint, while those, like

nformation

which do not

preserve stress, have no marking and thu

do not override the constraint against

lash in order to maintain the stress of their bases. A high-ranked identity constraint

ike Ident-

tress-

is cloned from the general Ident-Stress constraint of the

e. But the

eneral constraint i

ranked below the phonolo

ical markedness

onstraint *Clash

so that

ormátion emer

es as the optimal form.

. DO THE COHERING AFFIXES FORM A COHERENT SET? SPLIT BASES

AND PH

NETI

IN M

RPH

t the end of section 4

we asked whether the traditional division of suffixes into

roups could be maintained. We are

ow read

to understand a recent challen

o this claim raised in Raffelsiefen

(

2004

)

Bec

use s

ewo

ithin a

ptimality Theory in which different

affixes invoke different constraint rankings,

affelsiefen’s e

uivalent to a coherent

grouping of stem-level versus word-level affixes would be a group of affixes which

ause the forms to which they attach to b

evaluated by one ranked set of constraints

yllables of the words in (25d), Webster permits a secondary stress to appear in an alternate

pronunciation, while Kenyon and Knott cit

only a reduced vowel pronunciation.

ado

t here the common metrical notation that uses square brackets to show the grouping of syllables

nto feet. The brackets do not indicate phonetic transcription.

2 ELLEN M

KAI

ersus a second set which call up another ranking of those constraints. But, as we

arli

r in th

scuss

-ése, -éer

and

she argues that in fact every affix

as its own individual constraint ranking

d that there are no coherent sets.

ompares at some length ranking of the constraints which control -

affixati

ith

the ranking controlling the affixation of another verb-forming suffix -if

suffix is usually classified as stress neutral, the latter as stress-shifting. (

flúid,

(

lúidize

luídi

onsider first the verb-formin

suffix

-i

, which can attach productivel

ouns

(

Clintoniz

skeletonize

)

and ad

ectives (

randomi

mar

inaliz

)

.Th

productivity and semantic compositionality of -

formations suggest it is word-

evel (level 2) as does the fact that it can be added to antepenultimately stresse

ses

lik

skéleton

márginal

without inducing a stress-shift to repair the

equence of unstressed syllables. On the other hand, -

is able to select pre-existing

tems

rather than words

as a base, a clear stem-level

(

vel 1

)

characteristic, if this

will allow it to avoid stress clash

(

26a

)

nat

phonological results

uch as re

eated identical onset consonants

(

26b

)

(

)

súblimì

sed on

sùblimátion

ath

r than

sublíme

immunì

sed on

ìmmunólog

rath

r than

immúne

máximi

máxi

cf.

rádiumi

)

ápp

etiz

áppe

pá

rasitiz

)

f no source which can avoid clash is available

no word is coined. Thus

úshì

is all but unattested in Google searches, while

líntonìz

gets over 100 hits

–

this despite the fact that the base

Bush

is much more commonly encountered in

ontemporary searches than the base

linton

ow co

r th

ebe

f -i

tandardl

seen as a stress-affectin

boundar

affix (

Raffelsiefen ar

ues that -i

does not

ratuitousl

attach to

stems an

more than -

does.

It does so onl

to avoid other ill-formed phonolo

ical

esults, and only when a suitable stem already exists in a related word in the

peaker’s lexicon. Thus



pify



with the lax initial vowel preferred b

Tris

llabic

Laxing can be coined because



pical



existed first within the paradigm. fluídify,

ith

a stress configuration that avoids a *L

E violation could be coined in 1857 due

o the

rior existence of fluídity, first attested according to the

n 1603

when there is no plausible source for a fo

m that satisfies Trisyllabic Laxing,

eakers coin forms with tense vowels:

ste

elify, sto

nify, gro

ssify

etc. And when

there is no source with a stress shift that avoids a violation of *LAPSE, a gap results:

ándom+ify, prívat +ify tunnel+ify,

yield no output whatsoever, since there is no

andómit

or other form with stress on the second s

llabl

to suppl

a stem. Thus,

ues Raffelsiefen, neither -i

are

enuine stress-shifters. Both require a

Raff

n f

rmaliz

thi

lity as the domination of SUBCAT

, a constraint that requires word

formation to follow Aronoff’s (1976: 21) injunction that

affixes should attach to words, not stems. For

affelsiefen

a stem used as a base of word for

ation is a surface form, critically a form with stress

assigned to it.

ORD

FORMATION AND PHONOLOGY

source to which to match their stressed sylla

bles. Therefore they are stress-neutral.

et they are also cohering, if

udged by other criteria Furthermore, both are

phonologically interactive with their bases in different ways. -

anks *LAP

E as

elativel

unimportant and thus does not seek out bases with stress near the end of

he word, but it ranks *CLASH so hi

that it attaches to bound stems or leaves

aps in word formation where no suitable stem exists. It ranks TRISYLLABIC

AXING low

(

diumize)

affix -i

n the other hand

ranks TRISYLLABIC

AXING fairly high (



pify



)

and cannot violate *LAPSE at all. Raffelsiefen’s claim

which may or may not turn out to be right, is that virtually any ranking of constraints

upplied b

the factorial t

polo

be found for an affix. Of course, in order for

this ar

ument to

o throu

h, one must acce

t inde

endent rankin

s called up b

different affixes, a controversial proposal. If she is correct, though, the second ma

laim of level ordered phonology will also lo

e its force. Not only will level ordering

fail to predict the linear order of affixes, it will also fail to predict their phonological

Raffelsiefen is not the only author to note that the pre-existence of a

phonologically suitable word may be necessar

in order to permit a new coinage.

Steriade

(

1999

)

introduces a similar

nomenon which she calls the s

lit-base

ffect and which is closel

related to the principle of lexical conservatism which we

scussed

bove.

hile similar in conce

t, the s

lit-base effect differs from Raffelsiefen’s

proposal in that Steriade thinks a word can have distinct semantic, morphological

and phonological bases. Thus in the triad

émed

(

vb.

)

, r

emédial

emédiable

Steriade would argue that the verb

émed

is the morphological base of

remédiable

able

equires a verb as its base, but that

remédial

s the phonological base of

remédiable

ince an output based on the phonology of

remédial

better satisfies

*LAPSE. Raffelsiefen

in contrast

does not think that -

able

can take an

thin

but a

erb as its base, and ar

ues instead that both the morpholo

ical and phonolo

ical

eve

remédiate

ith

remédiable

formed b

violatin

SUBCA

hoose the already-stressed ste

remédi

as the base. This disagreement indicates the

difficulty of determining just what can form a base exerting influence on a

derivative, given our current state of understanding. Both authors are clearly on the

ight track, as they p

t that th

párodiabl

will have to suffer the lapse of

four unstressed syllables because there is no existing word, be it

paródial

paródiat

, which could form a model for the phonolo

icall

more pleasin

paródiable.

Steriade (2000) proposes a more radical influence of paradi

m uniformit

phonology than what we have considered so far

laim i

that

ati

nal

orphology is powerful enough to coerce a derived word to agree even in phonetic

details with other members of its paradigm

Her English example involves the rule

Steriade notes that the split base effect was independently proposed by Burzio (1997).

Steriade’s definition of a paradigm is “a

set of words sharing a morpheme (e.g. {

bomb

bombing

bombard

…

}

)

r a set of phrases sharing a word (e.g. {

bomb

the bomb

…}).” The reader may notice

that this definition is considerabl

more flexibl

than that

used

tra

iti

nall

. For instance Spence

(1991: 11-12) defines a paradigm

as “the set of all the

lected [emphasis mine – EMK] forms which