Journal of Memory and Language 44,644–665 (2001)doi:10.1006/jmla.2000.2760, available online at http://www.academicpress.com on

Phonological Constraints on Reading:Evidence from the Obligatory Contour Principle

Iris Berent

Florida Atlantic University

and

Joseph Shimron and Vered Vaknin

School of Education, University of Haifa

Hebrew frequently manifests gemination in its roots, but strictly constrains its position: Root-final gemination isextremely frequent (e.g., bbd), whereas root-initial gemination is rare (e.g., bdd). This asymmetry is explained bya universal constraint on phonological representations, the Obligatory Contour Principle (McCarthy, 1986). Threeexperiments examined whether this phonological constraint affects performance in a lexical decision task. The re-jection of nonwords generated from novel roots with root-initial gemination (e.g., Ki-KuS) was significantly fasterthan roots with final gemination controls (e.g., Si-KuK). The emergence of this asymmetry regardless of the posi-tion of geminates in the word implicates a constraint on root, rather than simply word structure. Our results furtherindicate that speakers are sensitive to the structure of geminate bigrams, i.e., their identity. Nonwords formed fromroots with final gemination (e.g., Si-KuK) were significantly more difficult to reject than foils generated fromfrequency-matched no gemination controls (e.g., Ni-KuS). Speakers are thus sensitive to the identity of geminatesand constrain their location in the root. These findings suggest that the representations assembled in reading a deeporthography are structured linguistic entities, constrained by phonological competence.© 2001 Academic Press

Key Words:phonology; morphology; reading; Hebrew; OCP.

The role of phonology in the recognition ofi

r

b

e m

ers are sensitive to the division of the syllablex,

ssi,s’y

onryote-m-aythlylsti,

printed words has been the subject of a contous research effort. Most research has focusewhether phonological representations constreading: Are words’ meanings activated direcfrom print or is reading also constrained phonological representations, assembled by mping graphemes to phonemes? Less is knohowever, about the structure of phonological rresentations assembled in reading. Do theseresentations consist of a linear chain of phoneor are they multilinear linguistic entities?

There is some evidence suggesting the etence of linguistic organization in assemblphonological representations. For instance, re

This research was supported by NIH FIRST award 1R

0749-596X/01 $35.00Copyright © 2001 by Academic PressAll rights of reproduction in any form reserved.

64

gh9;

&

DC03277-017 to Iris Berent.Address correspondence and reprint requests to Iris Be

Department of Psychology, Florida Atlantic University, 77Glades Road, P.O. Box 3091, Boca Raton, FL 33431-09E-mail: iberent@fau.edu.

4

nu-d onaintlyyap-wn,p-

rep-es

xis-edad-

into onset and rime (e.g., Treiman, MullenniBijeljac-Babic, & Richmond-Welty, 1995; Trei-man & Zukowski, 1988) and to the word’s strepattern (Colombo, 1992; Colombo & Taboss1992; Miceli & Caramazza, 1993). Readersensitivity to the structure of printed words maindicate the presence of linguistic constraints reading. If reading a word entails the recoveof its linguistic structure, then reading may nbe fully explicable by simple associations btween letters and sounds. The ability to asseble adequate phonological representations mrequire intact linguistic competence. Thus, boskilled and impaired reading must be ultimateexplained by the interaction of decoding skiland linguistic knowledge (e.g., Berent & Perfet1995; Farah, Stowe, & Levinson, 1996; Gou& Tummer, 1986; Harm & Seidenberg, 199I. Liberman & A. Liberman, 1990; I. Liberman& Shankweiler, 1991; Patterson, Suzuki,

29

rent,791.

I

9e

letacioi

d

thjeif

nc8e

aCh

toa8n

neious

eforgesdic

ent

n.P

us

inr-ren.so-

eheot

af-

rbed

,i-

e

n-erna-ni-

.,on.c-

as-arei-and

iers.ton,

an abstract set of timing units holding distinct

1 To clarify the morphological structure of the word, we in-dicate the root consonants using uppercase letters. Syllableboundary is indicated by a hyphen. However, the orthogra-

OCP CONSTRA

Nakayama-Wydell, 1996; Pennington, 19Perfetti, 1985, 1992; Shaywitz, 1996; StuddKennedy & Mody, 1995).

The present investigation addresses the rophonological constraints on reading using Obligatory Contour Principle (OCP) as a cstudy. The OCP is a universal ban on adjaidentical elements in phonological representat(McCarthy, 1979, 1986). OCP effects concernsegments, features, and tones have been mented in a variety of languages (e.g., Goldsm1991, Kenstowicz, 1994; Leben, 1973; McCar1979, 1986; Yip, 1988, 1995). If reading is subto linguistic constraints, then readers may exhsensitivity to the OCP in the performance osilent reading task, lexical decision.

The Nature of OCP Effects: Lexical andDerivational Manifestations

Linguistic theory has documented two mafestations of identity avoidance in phonologirepresentations (McCarthy, 1986; Yip, 198One manifestation of the OCP is found in the divation of phonological representations. Morphlogical and phonological processes occasionresult in adjacent identical elements. The Ointervenes to prevent such outputs by eitblocking the operation of rules whose output islicit or triggering phonological rules designedrepair such outputs. A familiar manifestationthe derivational OCP in English concerns thejacency of coronals (Russel, 1997; Yip, 198Many English suffixes begin with a coronal cosonant. The concatenation of such suffixes aroot ending with a coronal results in adjacent sments with an identical place of articulation, vlating the OCP. The OCP repairs such illicit oputs in several ways, including vowel epenthe(e.g.,buss+ s→busses; want+ d→wanted, add+ion→addition) and segment deletion (e.g.,com-plete+ tion→completion; ad + sist→assist). Thehypothesis that repair rules are triggered by idtity avoidance offers a simple explanationthese otherwise arbitrary phonological chanapplying during derivation (Yip, 1988). A seconmanifestation of the OCP is as a morphemstructure constraint: The OCP bans adjac

identical elements in the phonological represetation of a morpheme in the lexicon. It is the lex

NT ON READING 645

0;rt-

ofheseentnsngocu-ith,y,ctbit a

i-al).r-

o-llyPeril-ofd-).-

d ag--t-is

n-

cal OCP that is the center of our investigatioOne of the best known cases of the lexical OCconcerns the structure of Semitic roots. We thassess its presence in Hebrew.

OCP Effects in Hebrew Root Structure

To understand the nature of OCP effectsHebrew, we must first briefly consider word fomation in this language. Hebrew words aformed by inserting a root into a word patterThe root is a sequence of typically three connants. For instance, the root bdd indicates thecore meaning of isolation. To form a word, throot must be inserted into a word pattern. Tword pattern includes place holders for the roconsonants. It also supplies the vowels andfixes. For instance, the root bddmay be insertedin the word pattern CiCeC, resulting in the veBi-DeD1, he isolated. The insertion of the samroot in the word patterns CiCuC, CCiCut, anhitCoCeC yields the nouns Bi-DuD (isolation),BDi-Dut (loneliness), and the reflexive verb hit-Bo-DeD(I isolated myself), respectively. Thusa single root may yield a family of morphologcally related words.

Our main interest is in the structure of throot. Consider again the root bdd. This root con-tains the geminates dd. Hebrew frequently mani-fests gemination in its roots, but strictly costrains its position (a property found also in othSemitic languages, Greenberg, 1950). Gemition is highly frequent at the root’s final positio(e.g., bdd, gll, kff, sbb, smm). In contrast, gemnation is extremely rare root initially (e.g*ssm). This asymmetry requires an explanatiMcCarthy (1979, 1986) provided an elegant acount for this fact. His account assumes by suming that phonological representations multidimensional entities: Different phonologcal constituents, such as segments, tones, syllables, are each arranged on separate tThese different tiers are anchored to a skele

n-i-phy makes no distinction between the representation of rootand nonroot graphemes nor does it mark syllable structure.

O

n

h

e

h

eyw,

ng-tn-edht-ow ofl-r-t,

mi- ofeds

the,

but

re-hel

n,rthe

sei-

rdsngnenbi-

bided and hitboded. The root consonants are represente

slots for consonants and vowels. Because themitic root morpheme is a phonological costituent, it is represented on its own tier, seggated from intermediate vowels and affixes. Trepresentation nicely captures the integrity root consonants, regardless of their surface acency. For instance, consider again the wordsBi-DeD and hit-Bo-DeD(see Fig. 1). The representations of these verbs encode the root bdd on aseparate tier. Thus, despite the presence of inmediate vowels and affixes, root radicals are pchologically adjacent elements in a coherphonological constituent.

As a phonological constituent, however, troot consonants are subject to phonological cstraints. According to the OCP, lexical phon

a separate tier, segregated from vowels and affixes.

logical representations ban adjacent identip

from left to right, the rightmost consonant slot remaphoneme into the free slot. The doubly linked phone

N, AND VAKNIN

Se--

re-isofdja- -

ter-sy-nt

eon-o-

dd in bdd are adjacent identical elements, thmay not be stored in the lexicon. On this viethe lexical representation of bdd is bd. Gemi-nates may only emerge productively, duriword formation. Specifically, during word formation, the root bd is aligned with the three rooslots in the word pattern (see Fig. 2). The aligment proceeds from left to right, leaving thrightmost root slot empty. This slot is next filleby the spreading of the second root radical rigward. Consequently, a single radical, d, is nlinked to two slots. The surface manifestationthis double linking is geminates, dd. In what folows, we refer to this process of geminate fomation as reduplication (Everett & Beren1998; Gafos, 1998). Importantly, because genates are the result of a productive processreduplication, and because this process procerightward, geminates can emerge only at root’s final position, not its initial position. Thusgeminates are well formed at the root’s end ill formed at its beginning.

Our previous research investigated this pdiction. One set of experiments examined tacceptability of novel words formed from noveroots including geminates (Berent & Shimro1998; Berent, Everett, & Shimron, 2001). Ouresults demonstrated that speakers constrainlocation of geminates in the root. Words whoroots manifest root final geminates (e.g., SKeK) were rated as more acceptable than wowhose roots exhibit geminates at their beginni(e.g., Ki-KeS) Similar findings were obtained ia production task (Berent, et al., 2001). WhHebrew speakers produce words from novel

rbsd on

646 BERENT, SHIMR

FIG. 1. The autosegmental representation of the ve

calconsonantal roots (e.g., SK), they frequently), but

illus-eds

honological elements. Because the geminatesreduplicate the final radical (e.g., Si-Kek

FIG. 2. The formation of root-final gemination from its underlying biconsonantal form. The left figure trates the alignment of the biconsonantal root bd with the word pattern CiCeC. Because the alignment proce

ins unfilled. The right figure describes the spreading of theme surfaces as root final geminate, bdd.

I

dtaapecnsi-k.

aecdrraenthn

osean i. cro,

.d bselssfun

ogthnnetie

ersei-

x-g

herdpat-stednotots-d-

ve topre- the

ofofareellheoot butdasenda-uldot-

ld

c-e

bil-la-

No gemination KDM NKS

OCP CONSTRA

they practically never reduplicate the initial raical (e.g., Si-SeK). These findings agree withview that the OCP forms part of Hebrew speers’ linguistic competence. The questiondressed here is whether this linguistic princiconstrains reading. If the representation assbled in reading is constrained by phonologicompetence, then readers should exhibit setivity to the OCP in a silent on-line reading tas

Does the OCP Constrain Reading?

To test the hypothesis that the OCP constrreading, we examined readers’ sensitivity to idtical consonants within the root using a lexidecision task. Participants were simply askedetermine whether a printed stimulus cosponds to an existing Hebrew word. This tdoes not explicitly require that participants attto the phonology of the word. The depth of Hebrew orthography renders the computatiophonology particularly challenging. Hebrew isconsonantal writing system that omits mvowel information from the orthography. Conquently, the pronunciation of a Hebrew word cnot be predicted from its orthographic represetion. For instance, the word KeLeB, dog,written as a string of three consonants, KLBthe absence of knowledge regarding its spepronunciation, this string of letters may be pnounced as KiLiB, KuLuB, KaLaB, KaLiBKoLoB, etc. If a word’s phonology is assembledfrom left to right, then the absence of vowel in-formation may block its computation altogetherConversely, if the representations assemblereading are multilinear entities, constrainedphonological competence, then consonant asbly may proceed despite the absence of vowethe orthography (Berent & Perfetti, 1995). Asembled phonological representations may ther exhibit the effects of phonological costraints, such as the OCP.

Our present experiments gauge for phonolcal constraints on reading by examining wheperformance in the lexical decision task is setive to phonological well formedness, as defiby the OCP. To this end, we presented parpants with words and nonwords that includ

gemination. Table 1 summarizes the root struture of the target words and foils used in our e

insn-al toe-skdeofast-

n-ta-sInific-

inym- in-r--

i-ersi-d

ci-d

periments. The target words exhibited eithroot-final gemination or no gemination. BecauHebrew has only two productive roots with intial gemination, it was impossible to fully eplore the effect of root structure for existinroots. Our principal interest was thus in tstructure of the nonword foils. The nonwofoils were generated using the same word terns as the target words, but their roots consiof three-consonant combinations that did correspond to existing Hebrew roots. These roexhibited root-initial gemination, root-final gemination, or no gemination. Our experiments adress three questions: (a) Are readers sensitithe location of geminates? (b) Do readers resent the structure of geminates? (c) What isdomain of the constraint on gemination?

a. Readers’ sensitivity to the location geminates. The OCP constrains the location geminates in the root: Root initial geminates ill formed, whereas root final geminates are wformed. To examine readers’ sensitivity to tlocation of geminates, we compared two rtypes. These roots both contain geminates,differ in their location. One root type includegeminates in its beginning (e.g., KKS) wherea second root type included geminates at its (e.g., SKK). If participants constrain the loction of geminates in the root, then they shodiscriminate between these two root types. Roinitial gemination is ill formed hence it shoube easier to reject than root-final gemination.

b. The representation of geminates’ struture. A logical prerequisite for constraining thlocation of identical root consonants is the aity to represent identity. Identity is a formal re

NT ON READING 647

-hek-d-lem-al

TABLE 1

Structure of the Roots Employed for Targets and Foilsin Experiments 1–3

Word roots Nonword roots

Initial gemination — KKSFinal gemination KDD SKK

c-x-tionship. Using the variable X to refer to anyconsonant, the structure of identical consonants

O

nisse uiru na

nitthwofonobnenb

bicbrditc

on.n r

eeineo tos4the tldug-

eno-

ofh a ofch ton- aMl

n-rd

gi-r-).

i-ni-x-

a-re

.g.,

ot-ts. of a

faster rejection of root-initial gemination than of

n-isbyatnd

theewm,Ifoot

beeasier to reject than final gemination controls. If

648 BERENT, SHIMR

is defined as XX. Although a distinction betweeroot-initial and root-final geminates is constent with the proposal that readers repreidentity, these findings may also be explainedthe statistical properties of these items. Becaroot-initial gemination is ill formed, such gemnates correspond to rare bigrams. The easiejection of roots with initial geminates may thbe due to the rarity of the geminate bigram,its structure. However, not all forms of gemintion are rare in Hebrew. Recall that root-figemination is extremely frequent. Roots wfinal geminates thus permit investigating representation of identity. To this end, matched the foil roots with final geminati(e.g., SKK) and no gemination (e.g., NKS) their root bigram frequency. If readers do represent the identity of geminates, then rfinal gemination should be indistinguishafrom frequency-matched roots with no gemition. Conversely, if identity is represented, threaders should be able to discriminate root-figemination from no gemination controls virtue of their structure. Specifically, if the OCis active, then root-final gemination must formed by a grammatical process of redupltion. The formation of root-final gemination the grammar may provide evidence for “wohood.” Thus, novel words formed from roots wfinal gemination may be more difficult to rejethan words whose roots include no geminati

c. The domain of the constraint on gemition. The lexical OCP constrains the locationgeminates within the root morpheme. To repsent the OCP, readers must be able to reprits domain, a morphological constituent. This a growing body of evidence demonstratthat Hebrew readers decompose the root in ring. For instance, the priming of a Hebrew wby its root facilitates its identification relativephonological, semantic, and orthographic ctrols (Bentin & Feldman, 1990; Deutsch, Fro& Forster, 1998; Feldman & Bentin, 199Frost, Forster, & Deutsch, 1997). Likewise, productivity of Hebrew roots facilitates their dcomposition from their word patterns relativemorphologically unproductive controls (Feman, Frost, & Pnini, 1995). These findings s

gest that Hebrew speakers may routinely deco

N, AND VAKNIN

-nt

byse

- re-sot-

alheenr

otot-lea-nalyPea-y-

ht

pose the root from the word pattern. If so, ththey may also exhibit sensitivity to the phonlogical characteristics of the root morpheme.

The predicted sensitivity to the presence geminates and their location is consistent witconstraint on root structure. In the absenceproper experimental controls, however, sufindings may also be explained by appealingword structure. For instance, if our stimuli cosisted entirely of the CiCuC word pattern, thenrejection of forms such as SiSuM and PiPumight reflect the rejection of the word initiaunit CxiCx rather than the rejection of CxCxCy

roots. To establish that the domain of this costraint concerns root, rather than merely wostructure, our experiments vary the morpholocal structure of nonwords. In Experiment 1, tagets and foils are all unaffixed (e.g., Ka-KoSConsequently, root-initial gemination is invarably word initial whereas root-final geminatiois always word final. In contrast, in Experments 2 and 3 the root is affixed. Roots in Eperiment 2 are followed by a suffix (e.g., KKaS-tem), whereas in Experiment 3 they asandwiched between a prefix and a suffix (ehit-Ka-KaS-ti). A sensitivity to word structureshould not result in a consistent rejection of roinitial gemination across our three experimenConversely, if readers constrain the structurethe root, then each experiment should result in

a-ofe-sentregad-rdon-t,;e-o-

final-gemination controls.

EXPERIMENT 1

Experiment 1 examines the rejection of noword foils whose morphological structure transparent. These nonwords were formed inserting the root in nominal word patterns thinclude no affixes (see Table 2). The words anonwords were matched for the structure of word pattern, but differed in their roots. Thtarget words all consisted of existing Hebreroots. In contrast, nonwords were formed fronovel roots exhibiting root-initial geminationroot-final gemination, or no gemination. speakers are sensitive to the constraint on rstructure, then root-initial gemination should

m-the constraint in question specifically appeals to

N

a

v

hg

tr

h

a

nr

ta

Aen

.e

bi-ny

bi-al.pro-n-thisrials.re-re-nd

al.int

-oc-on-oe,alal

-heat-er

r

i-ot-

rdsofes,unte-s.of

n-

sre-20an-redntal in

No gemination Di-∫uN Ri-GuS

the structure of geminates, then the rejectionroot-final gemination may also differ from thof frequency-matched no gemination controls

Method

Participants. Twenty University of Haifa stu-dents who were native Hebrew speakers seras participants. They were paid to take partthe experiment.

Materials. The materials consisted of 9words and 90 nonwords (see Appendix A). Twords and nonwords were formed by insertinroot in one of several nominal word patternCiCuC, CaCiC, CoCeC, and CaCuC. Thus, word patterns of the targets and foils corsponded to existing Hebrew nouns. The targand foils shared the same word patterns. Tdiffered, however, in their roots. Targets weformed from existing triliteral roots such ththe combination of the root and word pattecorresponded to an existing Hebrew noun.contrast, the roots of the nonword foils cosisted of three consonants that do not cospond to an existing Hebrew root.

Nonword foil roots. The 90 nonword foilswere generated from 30 trios of novel rooEach trio included three root types: root-initigemination (e.g., ggs) root-final geminatio(e.g., sgg), and no gemination (e.g., rgs). Tthree members of the root were inserted in pcisely the same word pattern such that thesulting words differed only in root structure. comparison of performance with words genated form root initial vs. root-final geminatiothus permits the assessment of participants’ ssitivity to the location of geminates in the root

To examine whether participants are also s

sitive to the structure of the geminate bigram these roots, we matched the roots with final ge

T ON READING 649

oft.

edin

0e as:hee-etsey

retrnIn-

re-

s.lnhere-re-

r-

en-

n-

ination and no gemination for their summed gram frequency. Hebrew does not have acounts of root frequency. To estimate the typegram frequency of Hebrew roots, Berent et (2001) generated a database including all the ductive triconsonantal roots from the EveShoshan (1993) Hebrew dictionary. We used database to assess the frequency of our mateWe calculated the type summed bigram fquency of the root foils by adding the bigram fquency of the first two radicals, the second athe third radical, and the first and third radicThe consideration of both adjacent and disjobigrams was motivated by the presence of cocurrence restrictions on adjacent as well as nadjacent root consonants in Arabic (Frisch, Br& Pierrehumbert, 1995). The mean positionsummed bigram frequency of our roots with fingemination and no gemination was 12.27 (SD =3.72) and 11.83 (SD = 3.93), respectively.

Word target roots. The 90 words were generated from 45 pairs of existing Hebrew roots. Tmembers of the pair shared the same word ptern, but differed in root structure. One membhad root-final geminates (e.g.,dimum, bleeding,from the rootdmm) whereas the other membehad a root with no geminates (e.g.,di∫un, fertil-ization, from the rootd∫n). There are only twoproductive Hebrew roots with root-initial gemnation; hence, existing Hebrew words do nallow readers’ sensitivity to the location of geminates in the root to be tested. The target woalso do not allow for an accurate inspectionreaders’ sensitivity to the presence of geminatsince, in the absence of a word frequency cofor Hebrew, it was impossible to match the frquency of words with final vs. no geminateOur main interest thus concerns the effectroot structure on the identification of our noword foils.

Practice trials. To familiarize the participantwith the experimental task, they were first psented with practice trials, consisting of words and 20 nonwords, presented twice in rdom order. These words and nonwords shathe same word patterns as the experimestimuli. None of the practice stimuli appeared

OCP CONSTRAI

TABLE 2

Structure of the Targets and Foils in Experiment 1

Words Nonwords

Root-initial gemination — Gi-GuSRoot-final gemination Di-MuM Si-GuG

inm-the experimental trials. All stimuli in the prac-tice and experimental trials were presented in

a

rewi

k

n

rTl

i

x

de.

rn

t

-

y

n

thhen.racyc-n-

ot

y

intari-ordthei-

edont-er

lyaln-

p-eira-c-i-

eseurre

i-

onn

No gemination 697 94.2

650 BERENT, SHIMR

‘uppercase’ (dfus) Hebrew script without any diacritic marks.

Procedure. At the beginning of each trial,fixation point consisting of four dots appearedthe center of the screen. Participants initiatedtrial by pressing the space bar. They were thpresented with a string of letters at the centerthe computer screen, displayed until participaresponded or a maximum of 2 s elapsed. Paripants were asked to determine whether the stof letters corresponded to an existing Hebrword. Word responses were given by pressthe 1 key. Nonword responses were givenpressing the 2 key. These two keys were marby “yes” and “no” labels and positioned sucthat participants used their preferred hand to pvide both responses. Slow responses (resposlower than 1500 ms) and inaccurate responreceived negative feedback from the computethe form of a tone and a computer message.experiment initiated with the practice stimuli folowed by the experimental trials. Participanwere tested individually. The order of the trialsthe experiment was random.

Results and Discussion

To eliminate the effect of outliers, we ecluded correct responses falling 2.5 standardviations beyond the mean response latencieswords and nonwords. This procedure resultethe elimination of 2.97 and 2.57% of the corrword and nonword responses, respectivelythis and all subsequent experiments we ad.05 as the level of statistical significance.

Root structure did not significantly affect peformance for our target words. Response latefor words with root-final gemination (M = 666ms) did not differ from responses to words wino gemination roots (M = 665 ms,F1(1,19)< 1,MSE= 331.98;F2(1,44) < 1, MSE= 2369.74).There were also no differences in response acracy to root final (M = 94.3%) vs. no geminationtargets (M = 93.9%,F1(1,19)< 1, MSE= .002;F2(1,44)< 1,MSE= .005). However, response latency (M = 665 ms) to words was significantlfaster (F1(1,19)= 9.05,MSE= 1125.88;F2(1,73)= 4.70,MSE= 3597.83) than responses to nowords (M = 697 ms). Response accuracy

words (M = 94%) and nonwords (M = 94%)

ngbyedhro-ses

sesinhe-tsn

-de- for inct

Inopt

-cy

h

cu-

-

-to

did not differ (F1(1,19)< 1,MSE= .002;F2(1,73)< 1, MSE= .006).

Our principal interest is in performance winonwords. Specifically, we wish to examine teffect of the foils’ root structure on their rejectioTable 3 shows the response latency and accufor nonword foils as a function of their root struture. One-way ANOVAs on the responses to noword foils yielded significant main effects of rotype in latency (F1(2,38)= 12.73, MSE= 914.87;F2(2,58) = 5.74, MSE = 3584.43) and accurac(F1(2,38)= 13.18, MSE= .002; F2(2,58)= 4.96,MSE= .007). We next investigated the constraon root geminates by means of planned compson. If speakers constrain root structure in accwith the OCP, then they should be sensitive to location of geminates in the root. Roots with intial gemination should be perceived as ill formand hence less word-like than final-geminaticontrols. In accord with this prediction, rooinitial gemination was rejected significantly fastthan root-final gemination (t1(38) = 4.43 t2(58)=3.08). Likewise, participants were significantmore accurate in the rejection of root-initigemination than of root-final gemination cotrols (t1(38)= 4.81; t2(58) = 2.96).

We next examined whether participants reresent the structure of geminates, namely, thidentity. Our design equated root-final gemintion and no gemination for their statistical struture. If readers represent the identity of gemnates, then they may discriminate between thtwo roots despite their matched frequency. Oresults support this prediction. Participants wesignificantly slower at rejecting root-final gem

ON, AND VAKNIN

-

attheenof

ntstic-ing

TABLE 3

Response Latency and Accuracy in the Rejectiof Nonword Foils in Experiment 1 as a Functioof Root Structure

Response Response accuracy

latency (ms) (% correct)

Root-initial gemination 696 95.4Root-final gemination 738 88.9

nation than no gemination controls (t1(38) =

I

- t 1

OCP CONSTRA

4.31; t2(58) = 2.77). Likewise, root-final gemi

nation resulted in lower accuracy than no gem

e

tic

b

i

hlco

ctnty

fihSia

auonse

the ate

tesnot)to-rnsalal-ur tootsn-nsi-ot,)i-

eir

ty of

t 1.t.setB).ingrdim

lat-

tedec-

nation controls (t1(38)= 3.95; t2(58)= 2.41).

EXPERIMENT 2

The findings of Experiment 1 demonstrathat Hebrew readers are sensitive to the presof geminates and strongly constrain their potion. However, our present results cannot demine unequivocally the domain of the constraon the position of geminates. According to MCarthy (1986), this constraint concernsrootstructure. Although our results are compatiwith this view, they cannot rule out an explantion that merely appeals to the surface positof geminates in theword. The roots included inour first experiment were not affixed. Thus, tlocation of geminates in the root generaagrees with their word position. For instanthe root-initial gemination in Ka-KoS is alsword initial. Our next two experiments dissoate root and word structure by presenting paripants with roots that were affixed. In additiothese experiments demonstrate the generaliour findings for a new set of roots.

Experiment 2 examines lexical decision target words and foils that were formed by serting the root in a verbal word pattern tcontains a suffix. For instance, the foil Ki-Katem was formed by inserting a root with initgemination, KKS, in the word pattern CCaCtem indicating a verbal form in the ptense in the second person masculine plThis nonword was compared to two contrgenerated from roots with either final gemition or no gemination, SKK and NKS. Theroots were inserted in the same word patt

yielding the words Si-KaK-tem and Ni-KaStem (see Table 4).

ewted nototsersrn,berashe

ni-

No gemination Si-MaN-tem Ni-KaS-tem

NT ON READING 651

i-

tencesi-er-nt-

lea-on

elye,

i-ic-,of

orn-at-l

i-stral.lsa-ern,-

If the constraint we observed in Experimenmerely concerns the location of geminates inword, then the distinction between geminatesroot initial and root-final position should beliminated when the word-location of geminais altered. Thus, words manifesting root, but word, final gemination (e.g., Si-KaK-temshould be no more difficult to reject relative root-initial gemination (Ki-KaS-tem). In contrast, if the constraint on gemination conceroot structure, then the rejection of root-initigemination should remain easier than fingemination controls. As in Experiment 1, odesign also probes for participants’ sensitivitythe structure of geminates by equating rowith final gemination and no gemination cotrols for bigram frequency. If speakers are setive to the presence of geminates in the rothen root-final gemination (e.g., Si-KaK-temshould be more difficult to reject than no gemnation controls (e.g., Ni-KaS-tem) despite thopaque morphological structure.

Method

Participants. Participants were 20 Universiof Haifa students who were native speakersHebrew and did not participate in ExperimenThey were paid to take part in the experimen

Materials. The materials consisted of a new of 90 words and 90 nonwords (see Appendix Target words and foils were formed by inserta root in an affixed verbal word pattern. The wopatterns were past tense forms in binyanPaʔaL and PiʔeL (ʔrefers to a glottal stop). Alnonwords were generated from the verbal ptern of PiʔeL in the past tense. Words consisof real roots whose conjugation in their resptive word patterns yielded an existing Hebrverb, whereas nonword foils were generafrom a sequence of three consonants that didcorrespond to an existing Hebrew root. The roof target words consisted of 45 pairs. Membof a pair were matched on their word pattebut differed on their root structure. One memof the pair had root-final gemination wherethe other member exhibited no gemination. Tfoil roots consisted of 30 novel root trios, ma

TABLE 4

Structure of the Targets and Foils in Experiment 2

Words Nonwords

Root initial gemination — Ki-KaS-temRoot final gemination Si-NaN-tem Si-KaK-tem

festing root-initial gemination, root-final gemi-

nation, or no gemination. The root foils with

o

t

im-talr

loun

emrenr ofo

n

e

ro-

,

es

on-ncyonotoils

la-uc-

se-a-

ftly

slower than the rejection of no gemination con-

atatesbe-g-t toalt-rer-entintareendtheg-

onn

No gemination 724 95.6

652 BERENT, SHIMR

final and no gemination were equated for bigramfrequency. The mean summed positional bigramfrequency of root final and no gemination rowere 10.93 (SD= 5.58) and 10.63 (SD= 5.24),respectively.

Participants were first presented with practrials, including 20 words and 20 nonwords. Twords and nonwords shared the same wordterns as the experimental materials. None ofpractice stimuli matched the experimental stuli. The stimuli in the practice and experimentrials were presented in an upper case Hebscript (dfus) without any diacritic marks.

Procedure. The procedure was as describedExperiment 1.

Results and Discussion

An error in the construction of the materiaresulted in the inclusion of one foil whose rocorresponded to an existing (albeit nonprodtive) root, shpd. This item indeed resulted ivery high error rate (shipadti, M = 85% error).This root and its two matched trio membwere thus excluded from all analyses. To elinate the effect of outliers, we excluded sponses falling 2.5 SD beyond the grand meafor words and nonwords. This procedure sulted in the elimination of 2.76 and 2.85%the total correct responses for words and nwords, respectively.

Root structure did not significantly affect peformance with target words. Response lateto words with root-final gemination (M = 682ms) did not significantly differ from responsto roots with no gemination (M = 677 ms,F1(1,19) < 1, MSE = 622.402; F2(1,44) < 1,MSE= 2723.93). Likewise, there were no diffeences in responses accuracy to existing rwith final gemination (M = 93.1%) and no gemination (M = 93.4%, F1(1,19) < 1, MSE= .001;F2(1,44) < 1, MSE= .006). As in Experiment 1however, responses to words (M = 679 ms) weresignificantly faster (F1(1,19) = 27.70, MSE =846.78; F2(1,72)= 29.22, MSE= 1403.80) thanresponses to nonwords (M = 727 ms). Responsaccuracy to words (M = 93.3%) and nonword(M = 94.5%) did not differ significantly (F1(1,19)

= 2.196, MSE= .001, p = .1584; F2(1,72) < 1,MSE= .003).

ew

in

stc- a

rsi--

e-

Our primary interest is in responses to nword foils. Table 5 shows the response lateand accuracy for nonword foils as a functiof their root structure. Root structure did nmodulate response accuracy to nonword fF1(2,38) < 1, MSE= 0.002; F2(2,56) < 1, MSE=0.007). However, the ANOVAs on response tency revealed a significant effect of root strture (F1(2,38)= 10.77, MSE= 827.30; F2(2,56)=8.49, MSE= 1552.18). Replicating the findingof Experiment 1, root-initial gemination was rjected significantly faster than root-final gemintion (t1(38)= 4.58; t2(56)= 4.09). The rejection oroot-final gemination was also significan

ON, AND VAKNIN

ts

icehepat- the

TABLE 5

Response Latency and Accuracy in the Rejectiof Nonword Foils in Experiment 2 as a Functioof Root Structure

Response Response accuracy

latency (ms) (% correct)

Root-initial gemination 709 94.8Root-final gemination 751 93.7

n-

r-cy

s

-ots

trols (t1(38)= 2.92; t2(56)= 2.48).

EXPERIMENT 3

The findings of Experiment 2 demonstrate threaders are sensitive to the presence of geminand constrain their location. The convergence tween the findings of Experiments 1 and 2 sugests that this constraint is general with respecthe location of geminates in the word: Root-fingemination is more difficult to reject than rooinitial gemination whether the geminates aword final, in Experiment 1 (e.g., Sa-KoK), oword internal, in Experiment 2 (e.g., Si-KaKtem). The consistent performance across differword structures is incompatible with a constraon word structure. In contrast, these findings easily explained by assuming that speakers attto the structure of the root: They decompose root from the word pattern, treat it as a phonolo

ical constituent, and attend to the location ofgeminates within this abstract domain.

N

foceoda

te-

dhjetah

ldn

ftt

dretd

re-nds

heasof ex-ts,an-onotted noof inithne ineyg

ices,

rdsrnses-heeup-ia-

r-

iso anof

iNo gemination hit-Ba-LaT-ti hit-Na-KaS-ti

Experiment 3 provides an additional test this hypothesis by creating a stronger dissotion between root and word structure. In this periment, participants were presented with rothat were sandwiched between a prefix ansuffix (see Table 6). Thus, root-initial gemintion is never word initial. Likewise, root-finagemination is never word final. For instance, root trios KKS, SKK, and NKS were presentas hit-Ka-KaS-ti, hiS-ta-KaK-ti, and hit-NaKaS-ti2. If the constraint on the location of geminates appeals to the root as its domain, anreaders attend to root structure despite the opacity of these forms, then they should rehit-Ka-KaS-ti more easily compared to hiS-KaK-ti. If the constraint further appeals to tstructure of geminates, then the presenceroot-final gemination in hiS-ta-KaK-ti shoualso impair its rejection compared to the gemination control, hit-Na-KaS-ti.

Method

Participants. A new group of 20 University oHaifa native Hebrew speakers served as parpants. Participants were paid for taking parthe experiment.

Materials. The materials consisted of 90 worand 90 nonwords (see Appendix C). The woand nonwords shared the same word pattThey were both generated by inserting a roothe word pattern of binyan hitpa?el and the a

tion of an inflectional suffix. For instance, thword hitbasasti (I established myself) was

n-herect

ide-i-

2 The location of prefix t and the root-initial consonant is switched due to a metathesis rule applying to root initsibilants. This presents an extreme case of morphologopacity, as the root consonants are not only highly affixbut are linearly discontinuous. We assess the implicationsthis process in the General Discussion.

T ON READING 653

ria-x-ts a-

lhed

- if

ighct-eof

o

ici-in

sdsrn: indi-e

formed by inserting the root bssin the word pat-tern hitCaCaCti. Thus, the roots were always pceded by a prefix and a suffix. Word targets afoils differed, however, in their roots. The wordwere all formed from existing roots such that tresulting word is a familiar Hebrew verb wherenonwords were formed from combinations three consonants that do not correspond to anisting Hebrew root. As in previous experimentarget roots consisted of 45 pairs of triconsontal roots. Members of the pair were matched their word pattern and differed only on their rostructure. One member of the pair manifesroot-final gemination (e.g., bss) and one hadgemination (e.g., blt). The foil roots consisted the 30 trios of novel triconsonantal roots usedExperiment 2. These trios included a root winitial gemination (e.g., KKS), final geminatio(e.g., SKK), or no gemination (e.g., NKS). Ththree members of each trio were conjugatedprecisely the same word pattern, so that thmatched in all prefixes and suffixes, differinonly in their root structure.

The participants were presented with practtrials consisting of 20 words and 20 nonwordpresented twice in a random order. The woand nonwords exhibited the same word patteas the materials used in the experimental ssion. None of the practice stimuli matched texperimental stimuli. The stimuli in the practicand experimental trials were presented in an percase Hebrew script (dfus) without any dcritic marks.

The procedure was identical to that in Expeiments 1 and 2.

Results and Discussion

One foil trio was excluded from all analysbecause one of its members corresponded texisting root (shpd). To eliminate the effect outliers, we excluded responses falling 2.5 SDbeyond the grand mean for words and nowords, resulting in the exclusion of 3.29% of tcorrect word responses and 2.32% of the cornonword responses.

As in previous experiments, root structure dnot affect performance with target words. Rsponse latency for words with root-final gem

Sialicaled

OCP CONSTRAI

TABLE 6

Structure of the Targets and Foils in Experiment 3

Words Nonwords

Root-initial gemination — hit-Ka-KaS-tiRoot-final gemination hit-Ba-SaS-ti hiS-ta-KaK-t

nation (M = 690 ms) did not significantly differ of

O

na

r

te

-

o

tigt

y

t

nionl

ericls

readers extract the root and constrain its phono-

nceog-centre-cann.em-al

ri-er-s ahey

-dsnwither-

t inot,ca-, aasS-

x-s in ofa-

ithon

654 BERENT, SHIMR

from words with no gemination roots (M = 683ms, F1(1,19) < 1, MSE= 429.52; F2(1,44) < 1,MSE= 4010.90). There were also no significadifferences in accuracy to words with root fin(M = 91.0%) vs. no gemination roots (M =91.1%, F1(1,19) < 1, MSE= .002; F2(1,44) < 1,MSE = .019). However, word responses (M =686.72 ms, M = 94.8%) were significantly faste(F1(1,19) = 28.09, MSE = 985.88; F2(1,72) =34.77, MSE = 1412.01) and more accura(F1(1,19)= 5.64, MSE= 0.002; F2(1,72)= 2.96,MSE = 0.006. p = .0895) than nonword responses (M = 739.34 ms, M = 90.9%). Table 7lists response latency and accuracy for nonwfoils as a function of their root structure.

An ANOVA conducted on response latencynonword foils yielded an effect of root type, snificant in the latency data by participan(F1(2,38)= 3.45, MSE= 956.07, and marginallysignificant by items (F2(2,56) = 2.62, MSE =2269.80, p = .08). Root type did not significantlaffect response accuracy (F1(2,38)= 1.26, MSE= .002, p = .2947; F2(2,56) < 1, MSE= 0.009).Planned comparisons indicated significanfaster responses to roots with initial comparedfinal gemination (t1(38) = 2.21; t2(56) = 2.15. Inaddition, response latencies to roots with figemination were slower than to no geminatcontrols, a trend significant by participants o(t1(38)= 2.33; t2(56)= 1.75, p = .08).

The results of Experiment 3 present a conctual replication of our previous two expements. Participants strongly constrain the lotion of geminates in the root and they a

manifest some sensitivity to their presence. E

l

Root-final gemination 756 95.3No gemination 734 93.3

N, AND VAKNIN

vs p-

useed,sctofifi-a-thr-

t-

o

dently, despite the rich affixation of our materia

TABLE 7

Response Latency and Accuracy in the Rejection Nonword Foils in Experiment 3 as a Function of RoStructure

Response Response accuracy

latency (ms) (% correct)

Root-initial gemination 729 94.1

y-

tl

rd

o-s

lyto

alny

p--a-oi-,

logical structure.

GENERAL DISCUSSION

This research examined whether performain a silent reading task is subject to a phonolical constraint, the OCP. The OCP bans adjaidentical elements in lexical phonological repsentations. Consequently, root gemination only be formed productively, by reduplicatioBecause reduplication proceeds rightward, ginates are well formed only at the root’s finposition, not at root initial position. Our expements tested this prediction for nonwords genated from novel roots. If speakers possesmental constraint such as the OCP, then tshould consider nonwords with root-initial gemination as ill formed. The rejection of nonworwith root-initial gemination in a lexical decisiotask should thus be easier than that of roots final gemination. The results of our three expiments support this prediction.

To secure the conclusion that the constrainquestion concerns the structure of the rorather than merely the word, we varied the lotion of the root within the word. For instanceroot with initial gemination, such as KKS, wpresented as Ka-KoS in Experiment 1, Ka-Katem in Experiment 2, and hit-Ka-KaS-ti in Eperiment 3. Despite these marked differenceword structure, participants rejected eachthese stimuli more easily than its final gemintion control. The consistent performance wdifferent words indicates that the constraint geminates concerns the root.

Our experiments further investigated the reresentation of geminates’ structure. Becaroots with initial geminates are ill formed, thfrequency of their constituent bigrams, anspecifically, their initial geminate bigram, tendto be low. Their easier rejection may thus refleknowledge regarding the statistical frequencyconsonant combinations, rather than, speccally, their identity. To probe for the representtion of identity, we examined performance wiroot-final gemination. In each experiment, paticipants had difficulty in the rejection of roofinal gemination compared to their frequenc

oft

matched no gemination controls. The distinction

IN

asaledx-i-y tootssetes,

toltsootaloentllurisno-7;is-a-

a--alter-n-m

vi-ofw-s-ei-

s-lr-tly,nsith

ri-n

oh

sisnssr

n

a

o

ti

dci

ec

:

ioVthoaa

h

, nemth

OCP CONSTRA

between geminates and frequency-matched ngeminates clearly indicates sensitivity to tstructure of geminates, their identity.

Why are participants inhibited by the presenof identity at the root’s end? The OCP providepossible explanation for this finding: If identity erased from the lexicon, then root-final gemition must be formed by a grammatical procesreduplication. The attribution of novel wordwith root final geminates to a grammatical opetion may render them more wordlike, impairitheir discrimination from existing words. On aalternative account, however, the deleteriousfect of gemination reflects a task-specific stregy, rather than grammatical knowledge. Bcause the proportion of root final geminatiamong our experimental targets (M = 50%) ishigher than among the nonword foils (M = 33%),participants could have acquired the associaof root-final gemination with “wordhood” fromexposure to the experimental list. Our findingsnot support this view. An impairment in the rejetion of root-final gemination relative to no gemnation roots is evident early on in the experimtal session, at the first third of the trials in eaof our experiments (Experiment 1: M = 79 ms,F1(1,19)= 11.81, MSE= 5355, F2(1, 29)= 11.11,MSE= 11594; M = 7.1%, F1(1,19)= 5.49, MSE=.009, F2(1,29)= 5.40, MSE= .011; Experiment 2M = 11 ms, F1(1,19) < 1, F2(1,28)= 1.16; Exper-iment 3: M = 22.85 ms, F1(1,19)= 4.26, MSE=1225, F2(1,28)= 4.0, MSE= 1743). There is alsono significant increase in the effect of geminatthroughout the experimental session. An ANO(2 root × 2 block) comparing performance wiroot final gemination and no gemination contrpresented at the first and last third of trials in eof our three experiments revealed no significinteraction of root type and block (all p > .10). Infact, the deleterious effect of gemination in tfirst third of the trials in Experiment 1 (M = 79ms) was numerically larger relative to the la(M = 29 ms). The association of root-final gemnation with wordhood thus seems to indicateleast in part, some knowledge that is indepeent of the experimental manipulation. The viof root-final geminates as products of the gramar, as required by the OCP, may capture

knowledge.

T ON READING 655

The proposal that gemination is perceivedindicating wordhood can also explain additionaspects of our results. Our investigation probfor readers’ knowledge of root structure by eamining their sensitivity to the location of gemnates in the root. We assessed the sensitivitthe location of geminates by comparing rowith initial and final geminates. Because theroots are matched for the presence of geminatheir discrimination must indicate sensitivity the location of geminates in the root. Our resushow significant differences between these rtypes. However, performance with root-initigemination did not differ from roots with ngeminates. This finding appears inconsistwith the view of root-initial gemination as iformed, a conclusion firmly supported by oprevious finding that root-initial gemination rated as significantly less acceptable than gemination controls (Berent & Shimron, 199Berent et al., 2001). If root-initial gemination ill formed, why aren’t roots with initial gemination easier to reject than roots with no gemintion in the lexical decision task?

This puzzle is solved upon a closer examintion of the lexical decision task. Unlike our previous rating experiments, participants in lexicdecision experiments were not asked to demine the acceptability of novel letter strings. Istead, they were required to discriminate thefrom familiar Hebrew words. Discriminabilityis clearly affected by well formedness, as edent in readers’ sensitivity to the locationgeminates in the root. Well formedness, hoever, may not be the only factor affecting dicrimination. There is ample evidence for thsensitivity of speeded, forced-choice discrimnation to the global familiarity with the stimulu(e.g., Balota & Chumbly, 1984), its wordlike appearance. The difficulty in rejecting root-finagemination indicates that gemination is peceived as evidence for wordhood. Consequengemination impairs foil rejection. A comparisoof initial-gemination with no-gemination rootthus confounds the presence of geminates wtheir location. This comparison pits the deleteous effect of gemination against the facilitatioresulting from their illicit location. The similar-

n-e

ce a

a- of

a-gnef-t-

e-n

on

o--n-h

nA

lschnt

e

sti-atd-w

-is

ity in performance with root-initial gemination

O

n

lcets

e

cn

ing etetultsicalof

c-i-

nk-oy,ot

mes.po-achonftalr

656 BERENT, SHIMR

and no-gemination roots may reflect the cacellation of these conflicting forces. In cotrast, the comparison of root-initial geminatiowith root-final gemination adequately controfor the deleterious effect of gemination. Sucomparison yields strong evidence for an effof geminates’ location. These results reflecphonological constraint in a silent reading taWe now examine in greater detail the naturethe representation implicated by readers’ pformance.

The Multilinear Representation of PrintedWords: Evidence for PhonologicalConstraints in Silent Reading

The observation of OCP effects in the lexidecision task suggests that the representatiosembled in silent reading encodes several aspof the morphophonological structure of spokHebrew words. First, readers’ sensitivity to ro

structure indicates that their representation seg

contrast, the consonantal Hebrew orthography rendbe captured within a linear representation (3c).

N, AND VAKNIN

n--nshcta

k.ofr-

al as-ectsenot

previous research on morphological processin Hebrew (Bentin & Feldman, 1990; Deutschal., 1998; Feldman & Bentin, 1994; Feldman al., 1995; Frost et al., 1997). Second, our resindicate that readers encode the phonologstructure of the root, specifically, the identity adjacent root radicals and their location.

Readers’ sensitivity to the location and struture of geminates agrees with the view of gemnates as represented by means of multiple liing of a single phonological element onto twskeletal slots (e.g., Goldsmith, 1990; McCarth1986), as described in Fig. 3a. Interestingly, rogeminates are not necessarily adjacent phoneFor instance, the root final geminates in the sken word Sa-LaL (he paved) are separated by vowel phoneme. Capturing the structure of sudisjoint geminates within a linear representatiwould result in line-crossing, a fatal violation othe association convention in autosegmenphonology (Goldsmith, 1990). An account fo

re-such geminates thus requires that root conso-

egated

presen- and 3c

ation ofnel). In

gates the root morpheme and affixes, as shown bynants are encoded on their own tier, segr

FIG. 3. The representation of geminates and its effect on tier segregation. Figure 3a illustrates the retation of the geminates 11 by means of double linking of a single element to two skeletal slots. Figures 3bdemonstrate the phonological and graphemic representation of a word containing these geminates, salal (hepaved). The separation of geminates by a vowel in phonological representation (3b) requires the segregroot consonants from vowels (see right panel), as the failure to do so results in line crossing (see left pa

a. The geminates ll

b. Phonological representation (Salal)

c. A linear grapheme representation

ers these geminates adjacent graphemes, whose structure may

IN

xi

ars

oe

lu

td

nu

ntmaao

aame

ord1yad

a

eib-t7 na

eialtly,

es

re-tes ofsi-nta-in-ownur-on-

theped

es-re ofP,

thurlso is isnaln-u- in-

gslin-CPticintsy,ond

he byas-po-ep-odeandver,of

notions

OCP CONSTRA

from intermediate vowels and consonant affi(see Fig. 3b). Indeed, the existence of disjogeminates is a prima facieevidence for the viewof phonological representations as multilineAlthough root radicals in spoken Hebrew alinearly disjoint, this is not necessarily the cafor written words. Because the Hebrew orthoraphy omits most vowels from the script, rogeminates often correspond to linearly adjacletters. The structure of such geminates maycaptured by a linear graphemic account, as iltrated in Fig. 3c.

To assess the multilinearity of the represention assembled by readers, we attempted to sociate the perceived adjacency of geminafrom their linear position. To this end, we ispected performance with words that interrthe integrity of geminates by an intermedialetter. Consider first root-final geminatioTwenty-two of the 30 novel words with roofinal gemination in Experiment 1 had their geinates interrupted by a vowel letter. A separanalysis of these items revealed a robust impment in their rejection relative to no-geminaticontrols (M = 51 ms, F1(1,19)= 18.59, MSE=1387; F2(1,21)= 7.71, MSE= 3776; M = 5.4%,F1(1,19)= 14.89, MSE= .002), F2(1,21)= 4.40,MSe= .012). Thus, the disruption in the lineadjacency of geminates does not eliminate reers’ sensitivity to the presence of root-final geination. The lack of linear adjacency also donot appear to abolish the ill-formedness of rinitial geminates. Seventeen of the 30 wowith root-initial gemination in Experiment had their root-initial geminates separated bvowel letter. Despite the disruption in the lineadjacency of geminates, these items exhibitesizable facilitation relative to their final gemintion controls (M = 39 ms, F1(1,19)= 7.28, MSe=2063.01; F2(1,16)= 4.17, MSe= 3799, p = .06).Likewise, 7 of the 30 items in Experimentmanifested a disruption of root-initial geminatby an intermediate consonant prefix due to slant metathesis in this word pattern (e.g., hiSSaK-ti, from the root SSK; see Bolozky, 198The rejection of such roots was neverthelessmerically faster compared to root-final gemintion (M = 27 ms, F (1,19) = 2.63, MSe =

1

2879.81, p = .121, n.s; F2(1,6) = 1.32, MSe=

T ON READING 657

esnt

r.ee

g-tntbes-

a-is-

tes-ptte.--teir-n

rd--s

ots

ar a

-

3si-

a-).u--

3723, p = .3, n.s.). Finally, practically all thwords used in Experiment 2 had their root-initgeminates separated by a vowel letter. Evidenthe sensitivity to the location of geminatemerged despite the lack of linear proximity.

The robustness of gemination effects with spect to the linear adjacency of geminademonstrates that the perceived adjacencyroot radicals is independent of their linear potion. These findings suggest that the represetion assembled in reading Hebrew is multilear: It segregates root consonants on their tier, separate from vowels and affixes, and it fther distinguishes between geminates and nidentical root radicals. Thus, the structure of representation assembled in reading is shaby linguistic competence.

Our results leave several unanswered qutions. One question concerns the precise natuthe constraint on identity. According to the OCthe lexical representation of triliterial roots wifinal gemination is biconsonantal. Although ofindings agree with this account, they are acompatible with the view that root geminationstored in the mental lexicon, but its locationconstrained by the grammar, preferring root-fiover root-initial gemination. A related issue cocerns OCP violations. McCarthy’s (1986) atosegmental analysis portrays the OCP as anviolable constraint. Our experimental findin(Berent et al., 2001), however, converge with guistic accounts (e.g., Odden, 1986) of the Oas violable. Such facts favor the view of linguiscompetence as governed by violable constra(e.g., Optimality theory; Prince & Smolensk1997). The resolution of these issues falls beythe scope of our present analysis.

Our results also cannot fully determine tcontents of the representation assembledreaders. The similarity of the representation sembled in silent reading to the structure of sken language agrees with the view that the rresentations assembled in silent reading encphonological elements, such as phonemes features. On an alternative account, howethe terminal nodes in the representation printed words correspond to graphemes, phonemes. Indeed, graphemic representat

manifest multilinear structure that is quite simi-

ts t

c

ei

aa

t

p o

o

n

iona

e

t

rc

e in

g is.

led

d-one-ono-innts,92;i,e-y,-n-

m-ntsnu- of&-

&en,4,ac-ic

foro-rs’o-ionalr-

kl,;g,then-orn,&r-e,

1988; Rumelhart & McClelland, 1986). This

658 BERENT, SHIMR

lar to that implicated by our results. In particlar, breakdowns to the graphemic system in Iian (Caramazza & Miceli, 1990) and Engli(McCloskey, Badecker, Goodman-Shulman,Aliminosa, 1994) have been shown to resulspecific impairment to the representation of geinate letters, including erroneous shift in the lotion of geminates (e.g., pebble→peeble) andgeminate deletion (e.g., cigarette→cigarete).The fact that gemination in the English orthoraphy exclusively concerns adjacent letters cresponding to a single phoneme, and the setivity of gemination errors in Italian to thco-occurrence of letters, violating the integrof phonemes and phonological constraints,implicate a graphemic, rather than a phonolocal, representation. In contrast to the Italian English findings, root gemination in Hebrew ways concerns two phonemes, geminatesnot necessarily captured by adjacent letters,their location is subject to a phonological costraint. Our results thus bear none of the hmarks of a graphemic representation. On grounds of parsimony, we prefer an account attributes the effects of the OCP in written aspoken language to a single phonological resentation over an account that assumes twomorphic sets of constraints and representatione in phonology and one in orthography. Rgardless of the contents of the representatiterminal nodes as graphemes or phonemesstructure clearly reflects phonological constrai

The documentation of phonological costraints in reading Hebrew is of particular snificance. Hebrew is a deep orthography whvowel phonology is unpredictable from pri(Berent & Frost, 1997). The fact that silent reing in this orthography nevertheless encodesphonological structure of the language agrwith the fundamental role of phonological costraints in reading (e.g., Van Orden, Penning& Stone, 1990). Previous research on readHebrew demonstrated the role of phonologythe pronunciation of Hebrew words by obseing the contribution of vowels to word pronunation (e.g., Frost, 1994, 1995; Koriat, 198Navon & Shimron, 1981). Our results suggthat phonological constraints play a role even

the silent reading of a consonantal script. Th

ON, AND VAKNIN

u-al-h&

inm-a-

g-or-nsi-

tyallgi-ndl-areandn-all-thehatndre-iso-ns,e-n’s, itsts.

n-g-setd-

thees

n-on,ing inv-i-4,st

the representation assembled in silent readinshaped by readers’ phonological competence

The Nature of Subword Constituents Assembin Reading: The Representation of MentalVariables

Our conclusions regarding the effect of reaers’ linguistic knowledge on the representatiof printed words converge with previous rsearch demonstrating that readers encode phlogical and morphological units at various grasizes, such as the syllable and its constitueand the onset and rime (e.g., Rapp, 19Treiman et al., 1995; Treiman & Zukowsk1988), as well as morphological units (for rviews, see Marslen-Wilson, KomisarjevskWaksler, & Older, 1994; Pinker, 1999). An important question raised by these findings cocerns the nature of the linguistic units assebled in reading. Symbolic accounts of cognitioview morphemic and phonological constituenas variables, abstract place holders that can emerate a large number of tokens, regardlesstheir idiosyncratic properties (Berent, Pinker, Shimron, 1999; Kim, Marcus, Pinker, Hollander, & Coppola, 1994; Kim, Pinker, Prince, Prasada, 1991; Marcus, Brinkmann, ClahsWiese, & Pinker, 1995; Pinker, 1991, 1991999; Rapp, 1992). Conversely, associative counts consider phonological and morphemconstituents as descriptive labels, standing the coalition of orthographic, semantic, and phnological features of specific tokens. Readeapparent sensitivity to morphemic and phonlogical units does not indicate the representatof variables. Instead, it reflects the statisticstructure of phonological and morphemic corelates (e.g., Hare & Elman, 1995; RuecMikolinski, Raveh, Miner, & Mars, 1997Rumelhart & McClelland, 1986; Seidenber1987, 1997). The fierce debate surrounding nature of morphemic and phonological costituents stems from its direct implications funderstanding how the mind works (ElmaBates, Johnson, Karmiloff-Smith, Parisi, Plunkett, 1996; Fodor & Pylyshyn, 1988; Macus, 1998, 2001; Pinker, 1997; Pinker & Princ

us,debate also has some implications for

IN

cosey am

icaantivdar

t iicartiitxu

thtihectits nreifyngt, ont ic.rilyti

ngeprtiolsl inc

topm

culiz

rin-ts ofut)ese

con-In- the

et-01;llynist net-tityt toao,th-

hatse- se-

-en- oflesring

ose bytis-ort of

dis-o-ta-

usit-ta- afre-andicale-

ioncalar- the

OCP CONSTRA

accounts of reading. If readers encode the stituent structure of variables, then the repretation of variables is central for the adequacreading models. OCP effects allow us to exine this issue.

The OCP constrains the location of identelements in the root morpheme. The root identity are both mental variables. Associaaccounts eliminate variables; thus they renthe OCP unpresentable. The elimination of vables, however, does not necessarily erasedistinction among root types. If the constrainquestion could be inferred from the statiststructure of specific word instances, then paipants’ behavior could mimic the OCP desptheir inability to represent it. Specifically, to ehibit OCP effects, an associative system machieve two goals: It must identify the root asdomain of the constraint and encode the locaof identical elements within this domain. Tsuccess of associative accounts with respethe first goal, the abstraction of root-like unis presently uncertain. Hebrew roots sharephonological orthographic or semantic featuThe ability of associative models to identthese dissimilar objects as members of a siclass remains to be demonstrated. In contrasrepresentation of identity in contemporary cnectionist models has been recently subjecclose scrutiny. These conclusions have impltions for associative accounts of OCP effects

Identity is a formal relationship between vaables, namely, XX. Marcus (1998) systematicalinvestigated the representation of an idenfunction by multilayer perceptrons, includifeedforward networks and a simple recurrnetwork. Such networks are capable of resenting specific instances of an identity func(e.g., a rose is a rose). However, these modeare unable to generalize this function to novestances that fall outside the model’s represetional space of training items, its training spaThis failure does not stem from the inabilityrepresent the novel items, as the problem sisted despite the acquisition of the novel itein the context of an unrelated function. Mar(1998) further demonstrated that this genera

tion failure is not specific to the training algorithm or the representation format (localist vs

T ON READING 659

n-n-

of-

ldeeri-thenl

c-e-ste

on

to,o

s.

lethe-toa-

-

ty

nte-n

n-ta-e.

er-ssa-

distributed representations). The problem is pcipled: the independence in learning the weighconnections on any given unit (input or outpfrom connections on other units. To be sure, thconclusions do not challenge the adequacy of nectionism as a computational framework. deed, there are several proposals concerningrepresentation of identity in connectionist nworks that include variables (e.g., Marcus, 20Shastri, 1999). Marcus’ conclusions specificachallenge a well-defined subset of connectionetworks, those that eliminate variables. Suchworks are unable to freely generalize the idenfunction. This failure stands in marked contrashuman behavior. Marcus, Vijayan, Bandi Rand Vishton (1999) demonstrated that 7-monold infants can acquire an artificial grammar tconstrains the location of identity in short quences. For instance, infants trained on ABBquences (e.g., galala) discriminated novel ABBsequences (e.g., wofefe) from novel AAB se-quences (e.g., wowofe). Marcus et al. (1999) further showed that the acquired constraint on idtity cannot be due to the statistical propertiestheir materials. Thus, the elimination of variabprevents associative systems from captuhuman behavior.

The constraint on root identity has some clsimilarities to the artificial languages studiedMarcus et al. (1999). Our investigation of statical properties of Hebrew roots lends no suppto a statistical account for the representationidentity. Readers in our present experiments criminate roots with final geminates from ngemination roots despite their equation for stistical properties. Likewise, in a previoinvestigation examining the formation of trileral roots from their biconsonatal representions (Berent et al., 2001), we observedmarked contrast between the observed quency of root final gemination responses its expected probability based on the statiststructure of Hebrew roots. The distinction btween root final gemination and no geminatcontrols is thus inexplicable by the statististructure of Hebrew roots. In fact, Berent, Mcus, and Shimron (2001) recently observed

-.constraint on root structure even when speakershave no knowledge regarding the distribution of

O

efe a

sd spic

en-reri-

dingatenta- ofgi-

660 BERENT, SHIMR

root-radicals, for roots containing novphonemes composed of novel phonological tures. Such results challenge associativecounts of OCP effects.

The OCP effects in the lexical decision tasuggest that the representation assemblereaders encodes mental variables. Thus, dereaders’ undeniable sensitivity to the statist

structure of printed words, associative accoun

nocec nodep

N, AND VAKNIN

la-c-

kbyite

al

land, Seidenberg, & Patterson, 1996; Seidberg & McClelland, 1989) may not fully captutheir performance. The representation of vaables appears central to the adequacy of reamodels. Taken as a whole, our results indicthat readers assemble multilinear represetions that encode the constituent structuremental variables in accord with their phonolo

tscal competence. These results indicate the pres-

lic po-n does

(e.g., Harm & Seidenberg, 1999; Plaut, McClel-ence of linguistic constraints on reading.

APPENDIX A

Materials Employed in Experiment 1

Note: Hebrew nonback stop consonants p, b, and k undergo spirantization to p, v, and x, respectively, in post vocasitions (Bolozky, 1978). To capture phonemic structure, a structure reflected also in the orthography, the transcriptionot specify spirantization. In addition, note that most vowels provided here are not encoded in the orthography.

Target words Nonword foils

ʔitut ʔitur babiz zabib razib

dimum di∫un momel lomem rodebhisus himur pipuz zipup zerubzikuk zeruz baboʔ ʔabob padolxidud xidu∫ bibuʔ ʔibub ʔidukxitut xitul babic cabib ra∫ibximum xikuy poped dopep dope∫kinun kivun gigun nigug pigudlikuk likuy kakos sakok nakosʔerur ʔerub kokeʔ ʔokek ∫omelcitut cilum gagib bagig bagi∫kidud ki∫ut gigus sigug riguskerur kerub goger rogeg bogenʔidud ʔimut kakod dakok radokbasis bariʔ gagil lagig tagibgalil ka∫ir xixub bixux niʔubdakik dabik xaxiz zaxix baʔinhagig habil ʔoʔe∫ ∫oʔeʔ noʔekdalil dalik xaxod daxox daxogxalil xaric xixut tixux tixukxa∫i∫ xasin reruz zerur ceruksabib sabil roret torer morestamim tapil nanob banon bano∫tanin palit nanid danin gapi∫borer boxer zizur rizuz ∫ibungozez gozem tatul latut razumgoses goneb sosel loses conebgorer gomer ∫a∫iʔ ʔa∫i∫ lagi∫zolel zorem zizun nizuz nizuxzomem ∫oken toten notet notemkolel koʔesmocec moxecnoded no∫eb

OCP CONSTRAINT ON READING 661

Target words Nonword foils

solel soxercolel corem∫okek rokelmaror marompa∫o∫ paʔot∫anun ∫aburtalul taburta∫u∫ kaxu∫cnon cxokcror bkor

dror thom

APPENDIX B

Materials Employed in Experiment 2

Target words Nonword foils

ʔotatnu ʔitarnu bibaʔti ʔibabti pidaltibodadtem bidartem bibagnu zibabnu rizabnubisasti bi∫alti bibamtem mibabtem cibagtemgazazti gazarti sisakti xisasti xilartigasasnu gabarnu didaxno lidadnu dilabnugarartem gazamtem didaztem pidadtem pidaktemdimamti dibarti gigatnuti gagnuti gaznuhisasnu hidarnu gigacti cigagti ciga∫tizalaltem zakartem lilartem rilaltem bilartemzamamti zaxalti lilatnu nilalnu rilamnuzikaknu zimarnu mimabti bimamti gimabtixibabtem xibartem mimaptem pimamtem ximaptemxagagti xagarti ninabti dinanti dinaktixidadnu xida∫nu ninattem tinantem ∫inattemximamtem xima∫tem pipabnu bipapnu nipabnuxapapti xaparti pipadti zipapti ∫ipadtixa∫a∫nu xara∫nu kikabnu mikaknu pikabnuxacactem xacabtem kikastem sikaktem nikastemyilalnu yi∫abnu kikaʔti ʔikakti dipaktilikaktem limadtem kikaznu tikaknu bikatnumadadti makarti riralnu lirarnu miralnumacacnu malaknu ∫i∫azti zi∫a∫ti zi∫attinadadtem nadartem titagnu gitatnu ∫itagnunopapti nitarti titantem nitattem nitamtemsobabti sibakti cicagti gicacti bicagtisalalnu saparnu cica∫nu ∫icacnu bica∫nusinantem simantem xixaztem zixaxtem bixaztemʔodadti ʔibadti zizanti nizazti nizaxtiʔorarnu ʔerabnu zizarnu rizaznu dizarnupocactem picaltem ziza∫tem ∫izaztem biza∫temporarti perakticidadnu cilamnucalaltem cabartemcinanti cimakticarartem caramtem

kidadnu kidamnu

662 BERENT, SHIMRON, AND VAKNIN

Target words Nonword foils

komamti kibaltikacactem ka∫artemkerarnu kerabnuromamti rokantirisasnu ripaʔnuro∫a∫tem ritaktem∫adadti ∫abarti∫otatnu ∫idarnu

um

∫alaltem ∫amartem

APPENDIX C

Materials Employed in Experiment 3

Target words Nonword foils

hitʔonantem hitʔabaltem hitbabaʔti hitʔababti hitpadaltihitʔo∫a∫ti hitʔabanti hitbabagnu hizdababnu hitrazabnuhitbodadnu hitʔadamnu hitbabamtem hitmababtem hictagabtemhitbasasti hitbalatti histasakti hitxasasti hitxalartihitbonantem hitgalaxtem hitdadaxnu hitladadnu hitdalabnuhitbolalnu hitbacarnu hitdadaztem hitpadadtem hitpadaktemhitgonantem hitganabtem hitgagatti hittagagti hittagaztihitbo∫a∫ti hitgala∫ti hitgagacnu hictagagnu hictaga∫nuhitgodadnu hitgamadnu hitlalartem hitralaltem hibalartemhitgorarnu hitgaʔa∫nu hitlalatti hitnalalti hitralamtihitgo∫a∫ti hitgaradti hitmamabtem hitbamamtem hitgamabtemhizdakaktem hizdahabtem hitmamapnu hitpamamnu hitxamapnhitxababnu hizdahamnu hitnanabtem hitdanantem hitdanaktehitxolaltem hitxabattem hitnanatti hittananti hi∫tanattihitxamamti hitxabakti hitpapabti hitbapapti hitnapabtihitxanantem hitxada∫tem hitpapadtem hizdapaptem hi∫tapadtemhityadadnu hitxazaknu hitkakabnu hitmakaknu hitpakabnuhitronanti hitkabadti hitkakasti histakakti hitnakastihitkopapnu hitkabasnu hitkakaʔtem hitʔakaktem hitdakaptemhitlonanti hitkaxa∫ti hitkakaznu hittakaknu hitbakatnuhitlocactem hitlaba∫tem hitraralti hitlararti hitmaraltihitlakaknu hitlahatnu hi∫ta∫azti hizda∫a∫ti hizda∫attihitnosastem hitlaxa∫tem hittatagnu hitgatatnu hi∫tatagnuhitnocacti hitlamadti hittatantem hitnatattem hitnatamtemhistobabtem histabantem hictacagti hitgacacti hitbagactihistodadnu histadarnu hictaca∫nu hi∫tacacnu hitbaca∫nuhistananti histaʔarti hitxaxaztem hizdaxaxtem hitbaxaztemhitʔodadti hitʔatapti hizdazanti hitnazazti hitnazaxtihitʔalaltem hitʔatartem hizdazarnu hitrazaznu hitdazarnuhictanantem hictaxaktem hizdaza∫tem hi∫tazaztem hitbaza∫temhitʔorarnu hitʔata∫nuhictopapti hictalamtihitromamnu hitragalnuhitronanti hitraga∫tihitrocactem hitraxabtemhitro∫a∫nu hitraxaknuhi∫tobabtem hitratabtem

hi∫tomamti hitrasakti

OCP CONSTRAINT ON READING 663

Target words Nonword foils

hitgolalnu hitrapaknuhitgapaptem hitra∫amtemhitxababti hi∫tadaltihitxadadnu hi∫tatapnuhi∫tomamtem hitgalamtemhi∫tolalnu hi∫talatnu

hi∫tokakti hi∫taparti

f

go

ier

c

1

n

s

u

e

-

o

i

ndthe

g-

.,

nt

in

alhe

is

ngce

ni-

n,

inra-,

ing

y,

ed--

tions. Journal of Experimental Psychology: Learning,

REFERENCES

Balota, D., & Chumbly, J. (1984). Are lexical decisionsgood measure of lexical access? The role of word quency in the neglected decision stage. Journal of Ex-perimental Psychology: Human Perception and Perfomance, 10,340–357.

Bentin, S., & Feldman, L. B. (1990). The contribution of mophological and semantic relatedness. Repetition primat short and long lags: Evidence from Hebrew. QuarterlyJournal of Experimental Psychology, 42,693.

Berent, I., Everett, D., & Shimron, J. (2001). Do phonolocal representations specify variables? Evidence frthe Obligatory Contour Principle. Cognitive Psychol-ogy, 42,1–60.

Berent, I., & Frost, R. (1997). The inhibition of polygraphconsonants in spelling Hebrew: Evidence for a recurrassembly of spelling and phonology in visual worecognition. In C. Perfetti, M. Fayol, and L. Riebe(Eds.), Learning to spell: Research, theory and practiacross languages(pp. 195–219), Hillsdale, NJ: Erlbaum

Berent, I., Marcus, G. F., Shimron, J. & Gafos, D. (200The scope of linguistic generalizations: Evidence froHebrew word-formation. Manuscript submitted forpreparation.

Berent, I., & Perfetti, C. A. (1995). A rose is a REEZ: Thtwo cycles model of phonology assembly in readiEnglish. Psychological Review, 102,146–184.

Berent, I., Pinker, S., & Shimron, J. (1999). Default nomininflection in Hebrew: Evidence for mental variableCognition, 72,1–44.

Berent, I., & Shimron, J. (1997). The representation of Hbrew words: Evidence from the Obligatory ContoPrinciple. Cognition, 64,39–72.

Bolozky, S. (1987). Some aspects of Modern Hebrphonology. In R. Aronson Berman (Ed.), Modern He-brew structure(pp. 11–67), Tel Aviv: Universities Publishing Projects.

Caramazza, A., & Miceli, G. (1990). The structure graphemic representations. Cognition, 37,243–297.

Colombo, L. (1992). Lexical stress effect and its interactwith frequency in word pronunciation. Journal of Ex-perimental Psychology: Human Perception and Perfomance, 18,987–1003.

Colombo, L., & Tabossi, P. (1992). Strategies and stresssignment: Evidence from a shallow orthography. Ad-

vances in Psychology, 94,319–342.

are-

r-

r-ing

i-m

cntdne.).m

eg

al.

e-r

w

f

on

r-

as-

Deutsch, A., Frost, R., & Forster, K. I. (1998). Verbs anouns are organized and accessed differently in mental lexicon: Evidence from Hebrew. Journal of Ex-perimental Psychology: Learning, Memory, and Conition, 24,1238.

Elman, J., Bates, E., Johnson, M., Karmiloff-Smith, AParisi, D., & Plunkett, K. (1996). Rethinking innate-ness: A connectionist perspective on developme.Cambridge, MA: MIT Press.

Even-Shoshan, A. (1993). Ha’milon ha’xadash[The newdictionary] Jerusalem: Kiryat-Sefer.

Everett, D. L., & Berent, I. (1998). An experimental approachto the OCP: Evidence for violable identity constraints Hebrew roots. Rutgers Optimality Archive, Available:(ROA–235).

Farah, M., Stowe R., & Levinson, K. (1996). Phonologicdyslexia: loss of a reading-specific component of tcognitive architecture? Cognitive Neuropsychology,13,849–868.

Feldman, L. B., & Bentin, S. (1994). Morphological analysof disrupted morphemes: Evidence from Hebrew. Quar-terly Journal of Experimental Psychology, 47,407–435.

Feldman, L. B., Frost, R., & Pnini, T. (1995). Decomposiwords into their constituent morphemes: Evidenfrom English and Hebrew. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 21,947.

Fodor, J., & Pylyshyn, Z. (1988). Connectionism and cogtive architecture: A critical analysis. Cognition, 28,3–71.

Frisch, S., Broe, M., & Pierrehumbert. J. (1995). The role of similarity in phonotactic constraints. Unpublishedmanuscript, Northwestern University at EvanstoIL.

Frost, R. (1994). Prelexical and postlexical strategiesreading: Evidence from a deep and a shallow orthogphy. Journal of Experimental Psychology: LearningMemory, and Cognition, 20,116–129.

Frost, R. (1995). Phonological computation and missvowels: Mapping lexical involvement in reading. Jour-nal of Experimental Psychology: Learning, Memorand Cognition, 21,398–408.

Frost, R., Forster, K., & Deutsch, A. (1997). What can wlearn from the morphology of Hebrew? A maskepriming investigation of morphological representa

Memory, and Cognition, 23,829–856.

O

ny

a

e

ic

ae

r

l

y

s

s

,

v

h g

smp

-

kh

M

&n

y--

8

, &p-ired

ofxia.

in-ncel

le.

6).xia:

of

ing),

-

n-

ec-g

K.rdo-

ral

or-ed.

F.n,

.ngral-E.

i-i-:

D.

664 BERENT, SHIMR

Gafos, D. (1998). Eliminating long-distance consonaspreading. Natural Language and Linguistic Theor,16,223.

Goldsmith, J. A. (1990). Autosegmental and metricphonology. Cambridge, MA: Basil Blackwell.

Gough, P., & Tummer, W. (1986). Decoding, reading and ring ability. Remedial and Special Education, 7,6–10.

Greenberg, J. (1950). The patterning of morphemes inmitic. Word, 4, 196–208.

Hare, M., & Elman, J. (1995). Learning and morphologchange. Cognition, 56,61–98.

Harm, M. W., & Seidenberg, M. S. (1999). Phonology, reing acquisitions, and dyslexia: Insights from conntionist models. Psychological Review, 106,491–528.

Kenstowicz, M. (1994). Phonology in Generative Gramma.Cambridge, MA: Blackwell.

Kim, J., Marcus, G., Pinker, S., Hollander, M., & CoppoM. (1994). Sensitivity of children’s inflection to grammatical structure. Journal of Child Language, 21,179–209.

Kim, J., Pinker, S., Prince, A., & Prasada, S. (1991). Whmere mortal has ever flown out to center field. Cogni-tive Science, 15,173–218.

Koriat, A. (1984). Reading without vowels: Lexical accesHebrew. In H. Bouma, D. & G. Bouwhuis (Eds.), Atten-tion and performance: Control of language proces(pp. 227–242). Hillsdale, NJ: Erlbaum.

Leben, L. (1973). Suprasegmental phonology, dissertationCambridge, MA: MIT Press.

Liberman, I., & Liberman, A. (1990). Whole language code emphasis: Underlying assumptions and theirplications for reading instruction. Annals of Dyslexia,40,51–76.

Liberman, I., & Shankweiler, D. (1991). Speech, the alpbet and teaching to read. In L. B. Resnick, & P.Weaver (Eds.), Theory and practice of early readin(Vol. 2, pp. 103–132), Hillsdale, NJ: Erlbaum.

Marcus, G. (1998). Rethinking eliminative connectioniCognitive Psychology, 37,243–282, doi:10.1006/cog1998.0694.

Marcus, G. (in press). The algebraic mind: Integrating connectionism and cognitive science. Cambridge, MA:MIT Press.

Marcus, G., Brinkmann, U., Clahsen, H., Wiese, R., & PinS. (1995). German inflection: The exception tproves the rule. Cognitive Psychology, 29, 189–256,doi:10.1006/cogp.1995.1015.

Marcus, G. F., Vijayan, S., Bandi Rao, S., & Vishton, P.(1999). Rule learning by seven-month-old infants. Sci-ence, 283,77–80.

Marslen-Wilson, W., Komisarjevsky, L., Waksler, R., Older, L. (1994). Morphology and meaning in the Elish mental lexicon. Psychological Review, 101,3–33.

McCarthy, J. (1979).Formal problems in Semitic phonologand morphology. Doctoral dissertation, MIT, Cambridge, MA. Distributed by Indiana University Linguistic Club. Published by Garland Press, New York, 19

McCarthy, J. (1986). OCP effects: Gemination and antigeination. Linguistic Inquiry, 17,207–263.

N, AND VAKNIN

tal

l

ad-

Se-

al

d-c-

a,-

no

in

es

s.im-

a-A.

..

er,at

.

g-

5.m-

McCloskey, M., Badecker, W., Goodman-Shulman, R. A.Aliminosa, D. (1994). The structure of graphemic reresentations in spelling: Evidence from a case acqudysgraphia. Cognitive Neuropsychology, 11,341–392.

Miceli, G., & Caramazza, A. (1993). The assignmentword stress: Evidence from a case acquired dysleCognitive Neuropsychology, 10,273–296.

Navon, D., & Shimron, J. (1981). Does word naming volve grapheme- to-phoneme translation? Evidefrom Hebrew. Journal of Verbal Learning and VerbaBehavior, 20,97–109.

Odden, D. (1986). On the Obligatory Contour PrincipLanguage, 62,352–383.

Patterson, K., Suzuki, T., & Nakayama-Wydell, T. (199Interpreting a case of Japanese phonological aleThe key is in phonology. Cognitive Neuropsychology,13,803–822.

Pennington, B. (1990). Annotation: The genetics Dyslexia. Journal of Child Psychology, 31,193–201.

Perfetti, C. A. (1985). Reading ability. New York: OxfordUniversity Press.

Perfetti, C. A. (1992). The representation problem in readacquisition. In P. Gough, L. Ehri, & R. Treiman, (EdsReading acquisition(pp. 145–174), Hillsdale, NJ: Erlbaum.

Pinker, S. (1991). Rules of language. Science, 253,530–535.Pinker, S. (1994). The language instinct. New York: Morrow.Pinker, S. (1997). Words and rules in the human brain. Na-

ture, 387,547–548.Pinker, S. (1999). Words and rules: The ingredients of la

guage. New York: Basic Books.Pinker, S., & Prince, A. (1988). On language and conn

tionism: Analysis of parallel distributed processinmodel of language acquisition. Cognition, 28,73–193.

Plaut, D., McClelland, J., Seidenberg, M., & Patterson,(1996). Understanding normal and impaired woreading: Computational principles in quasi-regular dmains. Psychological Review, 103,56–115.

Prince, A., & Smolensky, P. (1997). Optimality: From neunetworks to Universal Grammar. Science, 275, 1604–1610.

Rapp, B. (1992). The nature of sublexical orthographicganization: The bigram trough hypothesis examinJournal of Memory and Language, 31,33–53.

Rueckl, J., Mikolinski, M., Raveh, M., Miner, C., & Mars, (1997). Morphological priming, fragment completioand connectionist networks. Journal of Memory andLanguage, 36,382–405, doi:10.1006/jmla.1996.2489

Rumelhart, D. E., & McClelland, J. L. (1986). On learnithe past tense of English verbs: Implicit rules or palel distributed processing? In J. L. McClelland, D. Rumelhart, & The PDP Research Group (Eds.), Paral-lel distributed processing: Explorations in the mcrostructure of cognition: Vol. 2: Psychological and bological models (pp. 216–271), Cambridge, MABransford Books/ MIT Press.

Russel, K. (1997). Optimality theory and morphology. In Archangeli, & L. Langendoen (Eds.), Optimality The-

ory (pp. 102–133), Oxford: Blackwell.

OCP CONSTRAINT ON READING 665

Seidenberg, M. (1987). Sublexical structures in visual wordrecognition: Access units of orthographic redundancy?

I:

e

deng

-: A

& e

in the description, use, and acquisition of English orthog-raphy. Journal of Experimental Psychology: General,

d

0).

d

r-

In M. Coltheart, (Ed.), Attention and performance XIReading(pp. 245–263), Hillsdale, NJ: Erlbaum.

Seidenberg, M. (1997). Language acquisition and use: Ling and applying probabilistic contraints. Science, 275,1599–1603.

Seidenberg, M., & McClelland, J. (1989). A distributed velopmental model of word recognition and namiPsychological Review, 96,523–568.

Shastri, L. (1999). Infants learning algebraic rules. Science,285,1673–1674.

Shaywitz, S. (1996). Dyslexia. Scientific America, 275,98–104.

Studdert-Kennedy, M., & Mody, M. (1995). Auditory temporal perception deficits in the reading impairedcritical review of the evidence. Psychonomic Bulletinand Review, 2, 508–514.

Treiman, R., Mullennix, J. W., Bijeljac-Babic, R., Richmond-Welty, E. D. (1995). The special role of rim

an-

-.

s

124,107–136.Treiman, R., & Zukowski, A. (1988). Units in reading an

spelling. Journal of Memory and Language, 27,466–477.

Van Orden, G. C., Pennington, B. F., & Stone, G. O. (199Word identification in reading proceeds fromspelling to sound to meaning. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 14,371–386.

Yip, M. (1988). The Obligatory Contour Principle anphonological rules: A loss of identity. Linguistic In-quiry, 19,65–100.

Yip, M. (1995). Identity avoidance in phonology and mophology. In Rutgers Optimality Archive. Available:(http://ruccs.rutgers.edu/roz.html).

(Received March 28, 2000)(Revision received August 24, 2000)