Memory & Cognition1987, 15(1), 1-12
The inflected noun system in Serbo-Croatian:Lexical representation ofmorphological structure
LAURIE B. FELDMANUniversity of Delaware, Newark, Delaware
and Haskins Laboratories, New Haven, Connecticut
and
CAROLA. FOWLERDartmouth College, Hanover, New Hampshire
and Haskins Laboratories, New Haven, Connecticut
Repetition priming is examined for alternating and nonalternating morphologically relatedinflected nouns. In Experiments 1 and 2, latencies to targets in nominative and dativellocativecases, respectively, were invariant over case of prime. In Experiment 3, latencies to nominativecase nouns were the same whether the nouns were primed by forms in which the spelling andpronunciation of the common stem were shared (nonalternating) or not (alternating) with thenominative form. Results are interpreted 88 reflecting lexical organization among the membersof a noun system. In Experiments 1 and 2, the pattern of latencies to primes suggests a satelliteorganization in which nominative forms are more strongly linked to oblique forms than obliqueforms are to each other. In Experiment 3, atypical cases of alternating forms showed a differentpattern of prime latencies, suggesting .that the organization within a noun system may differfor alternating and nonalternating forms.
In these studies, we examined the role of morphologyin the reading lexicon of speakers of Serbo-Croatian, thedominant language of Yugoslavia. The morphology ofSerbo-Croatian is particularlyinteresting to studybecauseit is substantially richer than that of English. Generally,in Serbo-Croatian, inflectional affixes are appended tonounsand adjectives, with the particular termination varying accordingto case, gender, and number. Analogously,for verbs, inflectional suffixes and sometimes infixesmayvary with tense, aspect, person, number, and sometimesgenderof thesubject. The formation of diminutives, agentives, and other derivations-which are characteristic ofSlavic languages-is similarly complex. Consequently,each Serbo-Croatianbase word has many variants, yielding extensive familiesof morphologically related words.
In the presentseriesof experiments, we exploredinparticular how the singular-ease inflected forms of a wordare related in the internal lexicon of adult readers who
We wish to thank the following students for collecting data: JasminaCesic, Sanda Parezanovic, Oara Andelkovic, and Teodora Vujin. Experiment 3 was suggested by Suzanne Boyce and Louis Goldstein. Inaddition, we thank Vicki Hanson and Jasmina Moskovljevic for manyhelpful comments on the manuscript. This research was supported byfundsfrom the National Academy of Sciences and the Serbian Academyof Sciences to Laurie B. Feldman; by NICHO Grant HO-01994 toHaskins Laboratories; and by NlCHO Grant HO-08495 to the University of Belgrade. Portions of this paper were presented at meetings ofthe Psychonomic Society in San Diego, CA, in 1983 and in SanAntonio,TX, in 1984. Reprint requests should be sent to the first author at HaskinsLaboratories, 270 Crown St., New Haven, CT 06511.
are native speakers of Serbo-Croatian. The experimentsrepresent an extension of earlier work by Lukatela andhis colleagues(Lukatela, Gligorijevic, A. Kostic, & Turvey, 1980; Lukatelaet al., 1978), who investigatedhowindividual inflected forms are recognized.
There are sevencasesof inflectednoun forms in SerboCroatian, which differ in their frequency of occurrencein printedtext (OJ. Kostic, 1965).Whensingular inflectedcases were presented in a lexical decision task, decisiontimes for the nominative singular form of a noun wereshorter than decision times for the same noun in(1) dative/locative and instrumental singular cases(Lukatelaet al., 1978)and (2) genitive and instrumentalcases (Lukatela et al., 1980). The decision times for allnon-nominative (viz., oblique) forms were equivalent.Lukatela and his colleagues (Lukatela et al., 1980;Lukatela et al., 1978) proposed that in the lexicon, thesingular cases of a noun make up a satellite-like systemin whichthe nominative singularof the nounor base formhas a special status in that it provides a nucleus aroundwhichthe obliquecasescluster in a uniformfashion. Thisorganization applies for inflected forms of both familiarand less familiar base words. That is, frequency of thenominativebase word, but not frequency of inflectionalcase, governs reaction time.
The satellite-entriesmodel reflects a position on a debated issue in the literature on how morphologicalstructure may influence word recognition (see Caramazza,Miceli, Silveri, & Laudanna, 1985). In that literature, thelexical entries are considered to consist of stem mor-
Copyright 1987 Psychonomic Society, Inc.
2 FELDMAN AND FOWLER
phemes or, alternatively, of whole words. In the formercase, polymorphemic words are decomposed into stemand affixprior to lexical access (Taft, 1979; Taft & Forster, 1975); in the latter, they are not. Instead, the lexicon may comprise a morphological principle of organization so that morphologically related words are nearneighbors (Stanners, Neiser, Hernon, & Hall, 1979)andlexical entries are accessed from whole words. In thestudies by Lukatela and his colleagues (Lukatela et al.,1980; Lukatelaet al., 1978), the same general pattern ofdecision latencies was obtained for masculine and feminine nouns in nominative and oblique singular cases,despitedifferences in the numbersof morphological transformations between nominative and oblique cases. (Specifically, in masculine words, the nominative singular isuninflected and therefore serves as the base morphemefor inflected forms. In feminine words, the nominativesingularis inflected; eachhasan A affix,whichis replacedto form other inflected forms.) This finding suggested toLukatela and his colleagues that entries for each case ina noun system are represented completely; that is, theyare not decomposed into a shared base morpheme plusan affix. It should be pointed out that Lukatela and hiscolleagues reported no direct comparisons of gender,although failure to find evidence of a case x gender interaction is critical to support the nondecompositioncharacterization of satellite entries.
The results of a second study also suggest that morpheme bases do not constitute the units of access to thenoun entries in a Serbo-Croatian lexicon. In an experiment designed to evaluate BOSS structure (Taft, 1979)as a unit of lexical access in Serbo-Croatian (Feldman,A. Kostic, Lukatela,& Turvey, 1983),BOSSunits(whichincluded the first unprefixed syllable as well as the longest sequence of consonants that can legally occur insyllable-final position) and base morphemes were fullyredundant. This was due in part to general constraints onorthographic structure for Serbo-Croatian and in part tothe criteria for selectingstimulus materials. The outcomeof the experiment was that where two different phonological interpretations of a letter string were equally possible, such that letter strings were bivalent, latencies inlexical decision were retarded as long as the entire word(i.e., the base morpheme) and the inflectionalaffix werebivalent.Whenonly the base morphemewas bivalent,decision latencieswere not changed relative to those for unequivocal controls. Feldmanet al. (1983) arguedthatmostvarieties of models that entail decomposition to a basemorphemeas the unit for lexicalaccess in Serbo-Croatianwouldpredictthat every word that includes a bivalentbasemorpheme should be affected.
These outcomes have served as the basis of argumentsagainstdecomposition of isolatedinflectednounsto a basemorphemein order to access their lexical entry. It shouldbe noted, however, that an interpretationof some of theseoutcomes as evidence for or against a morphemic representation for access may be inconclusive, in part becausea distinction between morphological processes arising
prior to and thosearisingsubsequentto lexicalaccessmaynot be possiblein a lexicaldecisiontask (Burani,Salmaso,& Caramazza, 1984;Henderson,Wallis, & Knight, 1984;Seidenberg & Tanenhaus, 1986).
The present series of experiments extends the satelliteentries account along two lines of inquiry. (1) We askwhether decision latencies to inflected forms of a nouncorrelate strongly. If members of a noun system are associated in the lexicon, then, nonlexical factors beingequal, decision latencies to inflected forms of a wordwould tend to be correlated. (2) We ask whether thenominative singular can prime and be primed by itsoblique-case satellitesas effectivelyas can an obliquecaseby other oblique cases or by a nominative. Reductionsin decision latencies to words in appropriate contexts orfacilitation by priming is sometimes explained in termsof activation among entries in the lexicon and is assumedto reflect, at least in part, lexical organization (e.g.,Seidenberg & Tanenhaus, 1986). Magnitude of facilitation, then, can provide an index of the cohesion amonglexical entries in a noun system. A variation of the lexical decision procedure, repetition priming, permits extensive investigation of the organization among regularand alternating inflected forms in the Serbo-Croatianlexicon.
In the repetitionpriming procedure (Forbach, Stanners,& Hochhaus, 1974; D. L. Scarborough, Cortese, & H.Scarborough, 1977;Stannerset al., 1979),each word andpseudoword is presented twice (with a lag of interveningitems) for a lexical decision judgment, and the facilitation to decision latency or priming due to repetition ismeasured. (The first presentationof the item is the prime.The second presentation is the target.t With Englishmaterials, it is not necessary that the identical word berepeated as prime and target for facilitation to occur.Generally, morphologically related words, including inflections and derivations, also reduce target decision latency, sometimes as fully as an identical repetition(Fowler, Napps,& Feldman, 1985; Stanners et al., 1979).For example, both the inflected form manages and thederived form management can facilitate a subsequentpresentation of manage. Sometimes, the effect is equivalent to an identical presentation of manage. (When thefacilitation with morphological relatives as primes isstatisticallyequivalent to the facilitationwith an identicalrepetition [following Fowler et al., 1985], the outcomeis full repetition priming. Priming with morphologicalrelativesthatis significant,but significantly less than withan identical prime, is partial.)
Repetitionpriming does not occur among orthographically similar but morphologicallyunrelated words, suchas ribbon and rib (Hanson & Wilkenfeld, 1986; Murrell& Morton, 1974; Napps & Fowler, in press), but it doesoccur when morphologically related primes and targetshave discrepant pronunciations and/or spellings (e.g.,health and heal) (Fowler et al., 1985;Hanson& Wilkenfeld, 1986; Napps& Fowler, 1983). Resultssuchas thesesupport an interpretation of repetition priming effects as
primarily lexical in origin (Fowler et al., 1985; Stannerset al., 1979), although there may also be a nonlexical orepisodic component (Feustel, Shiffrin, & Salasoo, 1983).Episodic contributions to repetition priming based on anexamination of derivational forms in Serbo-Croatian areconsidered elsewhere (Feldman, 1984; Feldman, in press;Feldman & Moskovljevic, in press). Currently, it appearsthat facilitation due to presentation of morphological relatives reflects lexical organization, but the differencebetween numerically full and partial priming may beat least in part episodic (Fowler et al., 1985). The longevity of the effect with morphologically related wordshas been offered as evidence that repetition priming maybe distinct from semantic or associative priming (Dannenbring & Briand, 1982; Henderson et al., 1984; Napps,1985). One way to capture this distinction is by proposing that morphological relatives activate the same lexicalentry but semantically associated words activate different entries.
Recent research has also identified a strategic contribution to the repetition priming effect (Forster & Davis,1984; Oliphant, 1983). As anticipated by Fowler et al.(1985), the large proportion of affixed primes followedafter a lag by their base forms may have permitted subjects to predict future targets from the prime. However,they found priming at long lags between prime and target (48-item lags). Finally, Napps (1985) demonstratedsignificant facilitation by morphological relatives, evenwhen only a very small proportion of morphemes wasrepeated. In light of these findings, the facilitation evidenced in repetition priming cannot be predominantly strategic in origin. Nevertheless, the experimental design introduced in that study and used in the present study doesnot prevent adoption of such a strategy by the subject,especially when base words serve as targets and inflections and derivations serve as primes.
The present series of experiments employs the repetition priming paradigm to investigate the lexical organization of Serbo-Croatian inflected noun systems in adults.In Experiment 1, nominative-case words served as targets, and we sought to learn whether, for real words, repetition priming was full such that primes morphologicallyrelated to their targets were as effective as identity primes.A by-product of this procedure permitted a replication ofthe original study on the satellite--entries account; specifically, it allowed an examination of the pattern of decision latencies for nominative and non-nominative formsof many words as it reflects the structure of the noun system. In addition, word gender was treated as a variableto ascertain that it did not interact with other effects asa decomposition account might predict. Finally, the pattern of correlations among pairs of satellite entries wasexamined. As discussed above, according to the satelliteentries account (Lukatela et al., 1980; Lukatela er al.,1978), the nominative singular case of both masculine andfeminine words enjoys a privileged status in the satelliteconfiguration. Taken in isolation, therefore, the outcomeof Experiment 1 is ambiguous. Plausibly, it reflects the
MORPHOLOGICAL ORGANIZAnON 3
coherence of the noun system. Alternatively, it reflectsthe special role of the nominative case. In Experiment 2,the pattern of facilitation for an oblique-case (viz., dative/locative-case) target was investigated. Once again,we examined the pattern of facilitation by various primesto learn about the lexical organization of satellite entriesand specifically about whether the nominative singularcase has a special status relative to oblique cases. In Experiment 3, the lexical organization for nouns that undergo sound and spelling changes in at least one of theirinflected case forms was investigated. Accordingly, thesimilarity of form between prime and target was reduced.Generally, decision latencies to primes and the pattern ofintercorrelations were interpreted with respect to the structure of the satellite system, and the pattern of facilitationin repetition priming was interpreted to reflect the coherence or organization within the noun system. Together,Experiments I, 2, and 3 provide an elaborated accountof the structure and coherence of the noun system of themature reader of Serbo-Croatian, thereby characterizingthe skilled reader's sensitivity to aspects of morphological structure.
EXPERIMENT 1
Experiment I examined priming of nominative-casenouns by identical and morphologically related words, Weaddressed three questions: (1) Does the presentation ofan inflected form of a noun facilitate lexical decision toa subsequently presented nominative form of the samenoun? Evidence suggests that the skilled reader of SerboCroatian is sensitive to morphological relatedness amongwords in that accessing one form necessarily accesses itsmorphological relatives. (2) Do differences in decisionlatencies for prime presentations of masculine and feminine words indicate different priming patterns? If not, thenas Lukatela and his colleagues (Lukatela et al., 1980;Lukatela et al., 1978) found, inflected nouns do not appear to be accessed from a base morpheme and then transformed or checked (in a fashion that affects reaction time)for the appropriateness of its affix. (3) Do decision latencies for inflected forms of a noun correlate? A positivecorrelation in conjunction with significant facilitation dueto repetition suggests that all inflected forms of a nounaccess the same lexical entry.
MethodSubjects. Forty-two students from the Department of Psychol
ogy at the University of Belgrade participated in this experiment.All were native speakers of Serbo-Croatian, and all had vision thatwas normal or corrected-to-normal. They participated in this studyin partial fulfillment of course requirements.
Stimulus materials. Twenty-four Serbo-Croatian words and 24pseudowords were included in the experiment. Words containedfour or five letters in their nominative form, and all were judgedby four independent rates to be very familiar. Half were feminineand half were masculine in gender; words in the two genders werematched according to length. No words were included that contained sequences of more than two consonants. Pseudowords weregenerated by changing one or two letters (vowel with vowel or con-
4 FELDMAN AND FOWLER
Table 1Examples of Regular Masculine and Feminine SiDguIar
Inflected NOUDS and Their FrequeDdes
sonant with consonant) in other real words with the same orthographic structure as the real words in the experiment. All materialswere printed in roman characters.
Each word appeared in three different singular cases: nominative, dativellocative, and instrumental. Each pseudoword also appeared with affixes for masculine or feminine words in the sameinflectional cases. Words were chosen so that inflectional suffixation did not alter the spelling of the baseform. Examples of regular masculineandfeminine words in theirseven intlected-case formsappear in Table I.
Procedure. Subjects individually performed a lexical decisiontask: As each letter string appeared, the subject hit a telegraph keywith both hands to indicate whether or not the string was a word.He/she hit the farther key (with index fingers) to signal "yes" andthe closer key (with thumbs) to signal "no." All letter strings weretyped in roman script, photographed, and mountedas slides. Stimuliwere projected from a carousel projector that was equipped witha modified camera lens as a shutter and were displayed on a screenuntil after the subject responded (approximately 750 msec). Thesubject viewed the screen from a distance of I m, and letter stringssubtended a visual angle between 2.6 0 and 3.9 0
• A dark field immediatelypreceded and followedeach display. The intervalbetweenexperimental trials was controlled by the experimenter and lastedabout 2,000 msec. Reaction times were measured from the onsetof the stimulus display.
Design. Threetest orders were created. Each one included threepriming conditions distinguished by the inflectional case of theprime, that is, nominative singular, dative/locative singular, or instrumental singular. (Case of prime was indicated as NI, 01, orII, respectively.) All targets were in the nominativecase. Half weremasculine gender and half were feminine. (The conditions ofnominativetargets preceded by nominative, dativellocative, and instrumental singular primes were indicated as NN, ON, and IN,respectively.) Words appeared in the same serial position acrossall test orders, although the inflectional form of the prime varied.For example, the word RUPA (meaning hole)was presented in itsnominative form as the target in the same serial position in all threetest orders,but it was preceded in thesame positionby either RUPA,RUPI, or RUPOM as a prime.
Each subject viewed one test order. Therefore, subjects saw eachmorpheme twice, once in a prime and once in a target. The average lag between the presentation of the prime and the target was10items, andlags ranged from 7 to 13. Filler items were introducedto maintain appropriate lags, and a practice list of 10items precededthe test list.
To summarize the experimental design, across test orders eachtarget word in nominative case was precededby its prime in nominative, dative/locative, and instrumental form. Within each order, abasemorpheme occurred once in a target and once in a prime, andcase of prime varied with item. Stated alternatively, all subjectsviewed thethree casesof prime on differenttarget items, and, acrosstest orders, each word was preceded by each case of prime.
PseudowordsNI 682 NN 704 Nl 729 NN 671 Nl 705 NN 68701 723 ON 695 Dl 721 ON 683 01 722 ON 68811 768 IN 700 11 773 IN 708 Il 770 IN 703
Note-NN, ON, and IN represent nominative targets preceded bynominative, dativellocative, and instrumental singularprimes, respectively.
NI 588 NN 534Dl 668 ON 541Il 670 IN 545
NI 600 NN 533 NI01 665 ON 539 0111 680 IN 543 11
Table 2Mean Reaction TImes (In MiIIiIecoDds) to NomiDative Targets
(NN, DN, IN) and 1'bek'RespectiveNomiDative-, D8tivelLocative-,and~ (Nl, Dl, 11) Primes In Experiment 1
MIIlICIJline Feminine Combined
Prime Target Prime Target Prime Target
Words576 NN 536672 ON 544661 IN 548
ResultsErrors and extreme reaction times (greater than
1,200 msec or less than 350 msec) were excluded fromall analyses. This procedure eliminated fewer than 4%of all responses. In addition, when a subject respondedincorrectly to one member of a prime-target pair, bothresponseswere excludedfrom subsequent analyses. Theerror-pairing procedure eliminatedan additional 3% ofall responses.
Mean reaction times for correct responses to nominative forms (Conditions Nl, NN, ON, and IN) of masculine and feminine words were calculated and subjectedto analyses of variance. Each comparison included ananalysis for subjects, averagingover items (PI), and foritems, averaging over subjects (P2). Means for Experiment 1 are summarized in Table 2.
For words, the effect of condition(Nl, NN, ON, IN)wassignifica..11t[FI (3,114) = 26.53,MSe = 1,759,p <.001;F2(3,66) = 11.41, MSe = 1,244,p < .001]. Theeffect of gender was not significant, although the interaction of condition x gender approached significance inthe subjects analysis but not in the items analysis[Fi(3,144) = 2.38, MSe = 1,651, p < .07; F2(3,66)= .26, MSe = 1,058, P < .85]. A secondanalysis thatincludedonly nominative targets (NN, ON, IN) revealedno significant differences among targets as a function ofcase of primeand no interaction involving gender.Therefore, the significant effectof conditionin the earlier analyses is due to the difference between the Nl conditionon the one hand and the three target conditions on theother; thus priming was full.
For pseudowords, neither the effect of condition northat of gender was significant, although their interactionwas significant only by a subjects analysis [FI(3,114) =3.98, MSe = 2,032, P < .01; F2(3,66) = 1.33,MSe = 1,864, P < .27]. Inspection of pseudowordmeans indicates that familiarity with pseudoword targetsslowed rejection latencies in the case of pseudo masculine nounformsand speeded rejection latencies in thecase
MIIlICIJline Freq. Feminine Freq.
OINAR 13 RUPA 9OINARA 9 RUPE 8OINARU I RUPI < IOINAR 6 RUPU 6OINAROM 2 RUPOM 2OINARU 4 RUPI 2OINARE <I RUPO <I
Case
NominativeGenitiveDativeAccusativeInstrumentalLocativeVocative
of pseudo feminine nounforms. Because the effect of condition on pseudowords was not significant, no analysiscombining words and pseudowords is included.
An analysis of variance on meanreactiontimesfor correct responses to word primes (Conditions N1, D1, 11)revealed a significant effect of case [Ft(2,76) = 40.22,MSe = 4,269, p < .001; F2(2,44) = 25.95,MSe = 2,036,p < .001].Therewasno effectof genderand, importantly for the satelliteinterpretation, no interaction of case x gender[Ft(2,76) = 1.89,MSe = 2,750,p < .16; F2(2,44) = .78, MSe = 2,036, p < 047]. Inspection of word means shows that for both masculineandfeminine words, the nominative case was recognizedfaster than (he obliquecases and recognition of obliquecases did not differ significantly.
An analogous analysis on pseudoword primes showeda significant effect of case [Ft(2,76) = 20.76, MSe =4,300, P < .001; F2(2,44) = 3.92, MSe = 7,006,p < .03] and an interaction of case x gender that wassignificant by the subjects analysis only[Ft(2,76) = 4.90,MSe = 2,800,p < .01; F2(2,44) = .60, MSe = 7,006,P < .56]. The pattern of pseudoword means revealedlongerrejectionlatencies for instrumental formsthan fornominative forms. For pseudofeminine nounforms, dative/locative latencies were similar to nominative latencies. For pseudo masculine noun forms, however, dative/locative latencies were intermediate between nominativeand instrumental latencies and significantly differentfrom each. All contrasts were significant at p < .01.
Noanalyses wereperformed on the error data, becausesomesubjects madeno errors and all subjects were veryaccurate. Outof 8 possible errors per condition, the meannumberof errors in Conditions N1, D1, andII for wordsand pseudowords, respectively, were 047, 049, and .18and .62, .69, and .64. The meannumberof errors on targets in eachcondition (NN, DN, IN), computed independently of the error-pairing procedure, was less than .20for both words and pseudowords.
Finally, mean reaction times for each prime word initsnominative (N1),dative/locative (01), andinstrumental(Il) forms were computed, and inflected forms of eachwordwerecorrelated. To theextentthatthevarious members of a nounsystemsharea lexicalentry or are equivalenton factors that contributeto reactiontime in a lexicaldecision task (Balota & Chumbley, 1984), correlationsbetween latencies for any pair of inflected forms will besignificant and all pair-wise correlations will be equal.Thecorrelations of nominative withdative(N1 and D1),nominative withinstrumental (N1 and11), anddative withinstrumental (D1 and 11) were r = .57, r = 049, andr = .67, respectively. (For correlations based on 24items, wheredf = 22, valuesof r greater than lAOIaresignificant at the .05 level.) Analogous correlations computedon latencies to pseudonominatives, pseudodatives,and pseudoinstrumentals did not approach significance.
DiscussionSignificant priming of nominative targets occurred when
real wordswere presentedfor lexicaldecisionin a repe-
MORPHOLOGICAL ORGANIZATION 5
tition priming procedure. The effect was obtained withboth identity primes (NN) and inflected relatives (i.e.,morphological primes: DN and IN). The means of thethree target conditions did not differ significantly, andtheir numerical valuesdifferedoverallonly by 10 msec.This outcome (viz., statistically full priming with smallnumerical differences between means) replicated resultsreported previously withEnglish materials (Fowleret al.,1985).One account providedby Fowler et al. is that thesmallnumerical differences in priming mayreflect anepisodic component that augments the lexical effects of repetition priming by selectively inflating the identity primecondition. However, as arguedelsewhere, thiseffectcannot be visual in nature (Feldman, 1984; Feldman &Moskovljevic, in press) becausethe magnitude of facilitation is as large when prime and target are printed indifferentalphabets as when they are printed in the samealphabet. Evidently, in the presentexperiment, presentation of related inflected-ease forms of a word facilitatedthe subsequent lexical decision about thatwordin nominative case, and both identical and morphological formsprimed fully. This outcome can be explained in terms ofa full spreading of activation amongindividual inflectedforms of a noun system (i.e., satellite entries) and itsnominative nucleus.
The suggestion of an interaction of condition x genderfor words indicated that the magnitude of the facilitationdue to repetition was larger for masculine nounsthan forfeminine nouns. However, inspection of meansrevealedthat the effect was carried by a difference betweenmasculine and feminine Dominative primes (N1) rather thanby targets (NN, DN, IN), andthe outcome of an analysisrestrictedto targetlatencies supported this interpretation.In summary, decision latencies to masculine and feminine target words were equallyfast whenan identical ormorphologically related prime preceded it.
Among pseudowords, evidence of a condition x genderinteraction made the absence of any overall facilitationwith repetitionequivocal. Inspection of means suggestedthatdecisions about masculine-gender targets wereslowedby a previous presentation of the identical prime, whereasdecisions aboutfeminine-gender targetswere facilitated.(Collapsing over gender, therefore, gave no evidence offacilitation withrepetition.) This effect is curious, becauseneithergendernor the interaction of condition X genderwassignificant for realwordtargets, and because theonlydifference betweenmasculine and feminine nominativecasepseudowords wasthe addition of an A suffix on feminineforms. In allotherrespects, theassignment of genderand, consequently, inflectional affixesto the two groupsof pseudowords was essentially arbitrary. At this point,we can suggest no explanation as to whyrepetition sometimes facilitated and sometimes impeded decisionlatencies for pseudowords.
The primary outcome of Experiment 1, based on thepatternof facilitation usingthe repetition primingprocedure, was that both nominative- and oblique-ease formscan prime a nominative target. Both identityprimes andmorphologically relatedprimesexhibited statistically full
6 FELDMAN AND FOWLER
priming with nominative targets. Following Stanners et aI.(1979) and Fowler et aI. (1985), we interpret repetitionpriming as an index of the interrelation among forms ofa noun in the internal lexicon. By this convention, alloblique-ease forms were tightly linked to their nominative nucleus. The facilitation evidenced in the repetitionpriming procedure with inflected nouns ofSerbo-Croatiancan be conceptualized to mean that once a satellite entryis accessed, the nominative nucleus of the noun systemis aIso activated.
The latency data for word primes provided a replication of previous results on inflected' forms in SerboCroatian (Lukatela et aI., 1980; Lukatela et aI., 1978).Nominatives were recognized faster than other cases, andrecognition of the oblique cases did not differ significantly.This outcome suggested that nominative forms are mostaccessible in the internaIlexicon. Importantly, there wasno interaction with gender. Masculine and feminine wordsdisplayed the same pattern of latencies among inflectedforms, despite differences in the complexity of derivinginflected forms from a nominative form. Equally strongcorrelations between mean latencies of two oblique casesof a word (D1 and 11) or of a nominative and one obliquecase (Nl and Dl, or Nl and 11) support this interpretation.
In conclusion, the outcome of the present experimentbuttresses the interpretation of Lukatela et aI. (1980) inthat it provided no evidence that the morphologicaI relatedness among inflected forms ofa noun was representedin the lexicon by a shared base morpheme and a set oftransformations whose complexity governs recognitionlatency. It appears that masculine nouns, for which thenominative singular and base morpheme are isomorphic,and feminine nouns, fer which the nominative singularincludes an A affixed to a base morpheme, are representedlexically in the same manner.
In the pseudoword prime data, decision latencies variedwith number of letters. For pseudo feminine items,nominative and dative/locative forms had the same number of letters, and they resulted in similar reaction times.Both differed from instrumental forms, which were oneletter longer. For pseudo masculine items, by contrast,nominative forms, which had the fewest letters, wererecognized significantly faster than dativellocatives, whichwere one letter longer than nominatives. Reaction timesto both of these were faster than those to instrumentals,which were two letters longer than nominatives. In lexical decision, length effects for orthographically regularbut meaningless letter strings have been reported previously in English and in other languages (e.g., Feldman& Turvey, 1983; Hudson & Bergman, 1985).
As reviewed above, the satellite-entries account positsa separate and complete entry for each affixed word andgrants a special status to the nominative case. Becausenominative-ease forms served as targets in Experiment 1,the outcome of the experiment (i.e., full priming withnominative targets) is inconclusive with respect to lexical organization within the noun system. The present out-
come may reflect the alleged privileged position of thenominative case in the satellite configuration. Alternatively, the same result could also arise if the nominativesingular case of a noun did not possess a special statuswithin the noun system (i.e., if the principle oforganization were uniform among all inflected forms). According to the homogeneous interpretation, however, the samepattern of full priming effects would emerge with anyoblique-ease target. In Experiment 2, we continued to explore the characteristics of the noun system. We used thepattern of facilitation in repetition priming to look for inhomogeneities in organization among entries. As in Experiment 1, it was our intention to ascertain how the principle of morphological relatedness operates within thenoun system, specifically whether, as predicted by thesatellite-entries account, there exist some inflected-easeforms that retain a privileged status when the oblique formof a noun must be activated.
EXPERIMENT 2
In Experiment 2, we asked whether primes that aremorphologically related to their targets facilitate recognition ofoblique-ease targets as effectively as they facilitate recognition of nominative-ease targets. Priming ofdative/locative-ease targets by nominative, dative/locative, and instrumental cases was examined. As in Experiment I, an identity prime condition served as the criterionfor determining full repetition priming. If inflected formsare defined relative only to the nominative singular, asposited in the satellite-entries account, then the instrumental singular case of a noun may facilitate lexical decisionon the dativellocative singular case of a noun less thanwould the dative/locative case itself. That is, the primingof dativellocative target by instrumental-ease forms maybe partial. Alternatively, if the organization among casesof a noun is homogeneous, then priming for oblique targets would be comparable to priming with nominativetargets.
MethodSubjects. Thirty-nine first-year students from the Department of
Psychology at the University of Belgrade participated in Experiment 2. None had participated in Experiment I. All were nativespeakers of Serbo-Croatian, hadnormal or corrected-to-normal vision, and never had participated previously in a psycholinguisticexperiment.
Stimulus materials. Thesame words and pseudowords presentedin Experiment I were used in Experiment 2. Moreover, the original order of presentation was preserved with one exception. In thetest list for Experiment 2, the dative/locative form, rather than thenominative form, appeared as the target. In Experiment 2, as inExperiment I, all letter strings were printed in roman characters.
Procedure. The procedure in Experiment 2 was identical to thatof the previous experiment.
ResultsErrors and extreme response times were eliminated
from the present analyses, according to the same criteriaused in Experiment 1. Fewer than 4 % of all responses
MORPHOLOGICAL ORGANIZATION 7
Table 3Mean Reaction Times(In Millisecoods) to DativetLocative Targets
(NO, DO, 10) and Their NClIIIiDatit'e-, DadvelLocative-, andInstrmnental-e- Primes (NI, 01, 11) In E¥perimellt 1
ND, DD, 10) were entered into one analysis, the interactionof condition x lexicality was significant [FI(3,114)= 15.72, MSe = 1,938, P < .001; F2(3,132) = 4.81,MSe = 1,950, P < .003]. Words were facilitatedmoreby repetition than were pseudowords.
An analysis of word primes revealed a significant effectofcase [FI(2,76) = 27.49, MSe = 2,762,p < .09;F2(2,44) = 22.14, MSe = 1,055, P < .001]. Neithertheeffect of gender nor the interactionof case x gender approached significance. An analogous analysis of pseudoword prime latencies revealed a significant effect ofcase [FI(2,76) = 19.32, MSe = 2,826, P < .001;F2(2,44) = 7.02, MSe = 2,393, p < .002] and a significant effectof gender [FI(l,38) = 9.18, MSe = 1,913,P < .01; F2(1,22) = .75, MSe = 7,156,p < .40]. Theinteraction of case x gender was significantby the subjects analysis only [FI(2,76) = 3.68, MSe = 3,046,p < .03; F2(2,44) = 1.44, MSe = 2,393, p < .25].
No analysiscould be performed on the error data. Outof 8 possible errors per condition, the mean number oferrors on ConditionsNI, D1, and Il for words and pseudowords, respectively, were .27, .29, and .42 and .34,.24, and .51. The mean number of errors on targets ineach condition (NO, DD, ill) computed independentlyof the error pairing procedure was less than .30 for bothwords and pseudowords.
Finally, meanrecognitionlatenciesfor prime words intheir nominative, dativellocative, and instrumental formswere computed and correlated for each word pair. Fornominative with dative (Nl, DI), nominative with instrumental (NI, Il), and dative with instrumental (Dl ,Il),thecorrelationswerer = .69,r = .66,andr =.71,respectively. Thesecorrelations, withdf = 22, are all significant at thep < .05 level. No pseudoword correlationswere significant.
DiscussionOverall, decision latencies were prolonged in the sec
ond ex.periment relative to those in the first. In light ofthe claim by Forster and Davis (1984) that magnitudeoffacilitation varies with word frequency (and hence reaction time) in unmasked presentations, no comparisonsacross ex.periments are offered. Inspection of decisionlatencies for word and pseudoword primes revealed adeviationfromthecharacteristic satelliteentriesoutcome.For words, dative/locative-case primes were respondedto faster than were instrumental-ease primes. Moreover,for pseudowords of both genders, responses to dative/locative- and nominative-case primes were nearlyequivalent. It appears that the preponderance of dativellocative-case targetwordsand pseudowords mayhavefacilitated all dativellocative forms. This findingdoes notinvalidate theanalysis of repetition priming,however,because all comparisons are on dativellocative-ease targets.
The strategy for interpreting repetitionpriming effectsadoptedin the present studyhas been to compare identityprime and morphemeprime conditionsand to define full
N\ 603 NO 563Dl 642 DO 552
11 665 ill 573
Masculine Feminine Combined
Prime Target Prime Target Prime Target
Words
593 NO 551649 DO 542655 ill 566
Nl 614 NO 576 NlDl 636 DO 563 Dl11 675 ill 580 11
Pseudowords
N\ 7\2 NO 691 Nl 715 NO 686 Nl 7\4 ND 68801 722 OD 688 01 710 OD 679 Dl 716 OD 68411 782 ill 710 II 739 ill 699 11 76\ ill 705
Note-NO, DD, and ill represent dativellocative targets preceded bynominative, dativellocative, and instrumental singular primes, respectively.
wereeliminatedaccordingto these criteria. An additional2%of all responseswere eliminatedby the error-pairingprocedure. Table 3 summarizes the mean recognitiontimes for dative/locative target words and pseudowordsin Experiment 2.
Analyses of variance, with condition (D1, ND, DD,10) and gender as independent variables,were performedusing subjects and items as random variables. As inExperiment 1, the effect of condition was significantforreal words [FI(3,1l4) = 59.48, MSe = 2,158,P < .001;F2(3,66) = 27.54,MSe = 1,435,p < .001].The effect of gender and the interaction of condition xgender were significantin the subjectsanalysis [FI(1,38)
= 6.27, MSe = 1,728, p < .02; F2(l,22) = .93,MSe = 3,589, p < .35] but not in the items analysis[FI(3,1l4) = 2.98, MSe = 1,913, p < .04; F2(3,66) =1.20, MSe = 1,435, p < .32].
A subsequent set of analyses, including only dative/locative-target latencies (conditions ND, DD, 10),revealeda significanteffect of prime condition [FI(2,76)= 4.02, MSe = 2,028, p < .02; F2(2,44) = 3.17,MSe = 790, P < .05] such that identity primes weremore effective than instrumentalprimes. There was alsoa significant effect of gender by the subjects analysis(FI(l,38) = 20.77,MSe = 1,125,p < .OOIJbutnotbythe items analysis [F2(l,22) = 3.07, MSe = 2,340,P < .09]. The interaction of condition x genderwas notsignificant.
An analogous analysisof pseudoword latencies showeda significant effectof primecondition [F1(3,114) = 6.77,MSe = 2,582, p < .001; F2(3~66) == 2.75,MSe = 1,952, P < .05]; however, no effect of genderand no interactionof condition x gender were found. Asubsequent analysis of pseudoword targets indicated a significant effect of condition such that instrumental-easeprimes facilitated less than did dative/locative- ornominative-ease primes [FI(2,76) = 3.37, MSe = 2,848,P < .04]. This effect was not significant in the stimulusanalysis, however [F2(2,44) = 2.06, MSe = 1,430,P < .14]. When word and pseudoword latencies (Dl ,
8 FELDMAN AND FOWLER
facilitation as effects that are not different from the identity prime condition. Consistentwith Experiment 1, Experiment 2 showed that lexical decision to nouns in thedative/locative case was facilitated by prior presentationof a morphologically related form. In contrast to Experiment 1, Experiment2 showedthat the instrumental singular primes produced only partial facilitation of dative/locative targets. Assuming that degreeof facilitationindexes closeness of relation or extentof activation spreadamong morphological relatives, it appears that connections within a noun system are not uniform. In Experiment 2, obliquecases were primed more fully by themselves than by other oblique cases. This effect wasdemonstrated both for masculine nounswhosebase morphemeand nominative were fully repeated in all obliqueforms and for feminine nounswhosenominative was notcompletely reiteratedin any obliqueform. Evidently, thelexical organization for a system of inflected nouns includes connections that vary in strength. Moreover, appreciationof morphological relatedness does not dependon a full overlap of the letters that constitute thenominative-ease form.
A comparisonof the patternof full and partial primingeffects in Experiments I and2 revealed someasymmetriesin organization for inflected forms that argue against ahomogeneous organization of morphological relatives. Bythe satellite-entries alternative, however, asymmetries areeasily accommodated because the nominative form functions as the nucleus of an inflected-noun system. Specifically, the relationship between nominative and obliquecaseswasas strong as the relationship between oblique andnominative casesin thatneitherwassignificantly differentfrom theidentity prime condition. Because facilitation withinstrumental primes was significantly different from thatwith identity primes, the relationship between two different oblique casesappears to be relatively attenuated. If inflected casesof a nounformed a homogeneous stnlctureeither as fully represented but independent lexicalentriesor as entriessharing a base morpheme, a claimsometimesmadefor English (e.g., Kempley & Morton, 1982)-thenpriming should havebeenequal among all inflected forms.Counter to the claim of a homogeneous representation,identity primesand morphologically relatedprimes werenot equally effective for all targets. In summary, the pattern of partial facilitation obtained in Experiment 2 arguesagainst a uniformly coherentnounsystem. Moreover, theobserved asymmetry in the facilitation among entries ofan inflected nounsystemcanbe interpreted to supporttheallegedspecialstatusof the nominative singularcase proposed by the satellite-entries account.
Theeffectof presenting a morphologically relatedwordprior to the presentation of a targetwordwassignificantlygreaterthan the analogous manipulation on pseudowords.However, the small, but nevertheless significant, effectof repetition on inflected pseudowords in Experiment2implicates a nonlexical contributionto facilitation in therepetition priming paradigm. The nature of inflectional
processes in Serbo-Croatian guaranteesthat membersofa satellitesystemgenerallywill be both orthographicallyand phonologically very similar. Consequently, all morphologically relatedprime-target pairs were visually andphonologically similar in their initial portion. The thirdand final experimentwas designedto examineappreciation of morphological relatedness in word pairs, withdiminished orthographic and phonological similarity.
EXPERIMENT 3
In Experiment 3, we askedwhethernounsthat includesound and spelling changes in some of their inflectedforms are represented in the lexicon by a satellite constellation. The present experiment included nouns withtwo types of sound and spelling changes: (1) femininewords with palatalization in their dative/locative formsand (2) masculine wordswithchanged nominative/accusative forms that includeeither (a) a movableA or (b) ano which elsewhereappears as L. We will refer to morphemesthat occur in more than one form as alternating.It is important to note that by linguistic accounts, thesealternations are regular and can be described by rules,although they are no longer productive. The repetitionprimingparadigm wasagainusedwithnominative targetspreceded by an identical prime and by two morphologicalprimes. For half of the items presented (i.e., masculinealternatingnouns such as PETAK), both morphologicalprimes differed in spelling and pronunciation fromthe target forms (i.e., PETKU, PETKOM). For the other halfof the items (i.e., feminine alternating nouns such asNOGA), half of the morphological primes differed inspelling and pronunciation from the target (i.e., NOZI)and half were identical in spelling and pronunciation ofthe stem morphemeto that of the target (i.e., NOGOM).As in previousexperiments, decisionlatenciesto targetsas a function of type of prime addresses the issue of cohesion among inflectedmembersof a noun system, andthe patternof decisionlatencies (andcorrelations) amongprimes hints at the structure of the noun system.
MethodSubjects. Forty-two first-year students from the Department of
Psychology at the University of Belgrade participated in Experiment 3. All hadparticipated in either Experiment I or Experiment 2approximately 6-8 weeks earlier.Stimulus~.Twenty-one alternating masculine words and
21 alternating feminine words were included in Experiment 3. Allof the masculine words had changed spellings in the nominative/accusative singular case;thiscaseconstituted theatypical form.For most masculine items, the alternation took the form of the addition of a vowel before the last consonant of the base form, thuseliminating certainconsonant sequences in word-final position ~toccurred as a consequence of the disappearance of a weak semivowel in word-final position (e.g., PETAK vs. PETKU [nominative singular vs. dative singular)). For other masculine forms, thealternation involved the deletion of L and its replacement by 0 insyllable- and word-final position (e.g., PETAOvs. PETLU [nominative/accusative vs. dativellocative singular)); this development
MORPHOLOGICAL ORGANIZATION 9
ResultsErrors and extreme response times were eliminated
from the present analyses according to the same criteriaapplied in the two previous experiments. Fewer than 3%of all responses were eliminated according to these criteria. An additional 2% of all responses was eliminated bythe error-pairing procedure. Table 5 summarizes the mean
Table 5Mean Reaction Times (in Milliseconds) to Nominative Targets
(NN, DN, IN) With Sound and Spelling Alternations and toTheir Nominative, DativelLocative, and Instrumental
Primes (Nt, 01, 11) in Experiment 3
Masculine Feminine Combined
Prime Target Prime Target Prime Target
WordsNI 664* NN 631 NI 706 NN 622 NI 685 NN 62701 728 ON 641 01 785* ON 634 01 757 ON 638II 726 IN 633 II 739 IN 631 IJ 732 IN 632
Pseudowords
NI 771* NN 744 NI 773 NN 765 Nt 772 NN 75401 758 ON 741 01 796* ON 777 01 777 ON 759It 806 IN 750 II 816 IN 757 II 811 IN 753
*Form that undergoes sound and spelling change.
occurredin 14th-eentury Serbo-Croatianand, again, it was relatedto the disappearanceof a weak semivowelfollowingsyllable-finalL (Belie, 1976). In each case, nominative/accusative and dative/locative forms contained the same number of letters.
All of the feminine wordshadchanged spellings in thedativellocative form wherethe alternationentailedpalatalizationof velar consonants (i.e., the consonants K, G, H change to C, Z, S whenfollowedby I derivedfrom "0: or the letterjot [second palatalization]"[Belie, 1976]). By comparison, the instrumental singularforms forboth masculine and feminine words were typical in construction.One consequence of the locus of the changed case form was thatfor masculine words the dativellocative and instrumental formssharedspellingand pronunciation, whereasfor feminine words thenominative and instrumental forms were similar. Masculine andfeminine pseudowords were constructedto include the same styleof spellingandsound changes that occurred in words. Examplesof alternatingmasculine and femininewords in their inflectedcaseforms are presented in Table 4.
Thetest order and composition of the list(s)were analogous withthoseof Experiment 1. In the presentexperiment, targetwords werepresented in nominative case and all items were printed in romanscript. As in previousexperiments, lags betweentarget and primeaveraged to items with a range of 7 to 13. With the exceptionofthe numberof wordsin a testorder, the testingprocedurewasidentical to that described above.
Masculine
recognition times for nominative targets of alternatingwords and pseudowords.
Analyses of variance with prime condition (N1, NN,DN, IN) and gender as independent variables were performed on real-word latencies using subjects and itemsas random variables. Consistent with the outcome forrepetition priming ofnominative targets for regular words,there was a significant effect of prime condition [FI(3,123)= 37.30,MSe = 1,630,p <.001;F2(3,120) = 19.57,MSe = l,553,p <.001]. The interaction of gender xprime condition was also significant [FI(3,123) = 10.38,MSe = 1,191, P < .001; F2(3,120) = 3.98, MSe =1,553,P < .01]. All feminine targets showed more facilitation relative to unprimed nominatives (N1) than did masculine targets. In subanalyses including only target wordlatencies (viz., NN, DN, IN), neither the effect ofgendernor the effect of prime condition approached significance.
An analogous analysis of pseudoword latencies indicated a significant effect of prime condition [FI(3,123)= 3.44, MSe = 1,775, P < .02], a significant effect ofgender [FI(l,41) = 8.98, MSe = 2,481, P < .005], andan interaction of condition x gender [FI(3,123) = 3.74,MSe = 1,338, p < .01]. None of these was significantby the items analysis, however [F2(3,120) = 1.76, MSe= 1,739, P < .16; F2(l,40) = 1.50, MSe = 7,425,P <.23;andF2(3,120) = l.44,MSe = 1,739,p <.23,respectively] .
Inspection of the latency data for word primes suggestedan interesting deviation from the familiar equivalenceamong oblique-ease latencies predicted by the satelliteentries account. Results of analyses of variance indicateda significant effect of case [FI(2,82) = 41.76, MSe =2,654, P < .001; F2(2,80) = 17.98, MSe = 3,082,p < .001], a significant effect of gender [FI(l,41) =55.60, MSe = 1,602, P <.001; F2(l,40) = 2.77, MSe= 16,051, P < .01], and an interaction ofcase x gender[F i(2,82) = 4.97, MSe = 2,079, p < .009] that was notsignificant by stimulus analysis [F2(2,80) = 1.68, MSe= 3,082, P < .19]. For both genders, nominative formswere recognized most quickly. For masculine forms,oblique cases, neither of which had changed spellings,were equivalent. By contrast, for feminine forms, instrumentals, whose stem morphemes were identical insound and spelling to those of their nominative forms,were significantly faster than dative/locative forms inwhich the stem morpheme was not identical [t(41) = 4.57,P < .01].
An analogous analysis of alternating pseudoword primesindicated that the effect of case was significant [FI(2,82)= 17.36, MSe = 2,147,p < .001; F2(2,80) = 5.45, MSe= 3,418, P < .01], as was the effect of gender [FI(l,41)= 11.57, MSe = 1,489, P <.002; F2(l,41) = 11.57,MSe = 1,489, P <.002]. The interaction of case xgender was also significant [FI(2,82) = 3.12, MSe =2,400, p < .05; F2(2,82) = 3.12, MSe = 2,400,p <.05].
No analyses were performed on the error data becausesome subjects made no errors and all SUbjects tended to
Feminine
PETAK* PETAO*PETKA PETLAPETKU PETLUPETAK* PETAO*PETKOM PETLOMPETKU PETLUPETCE PETLE
Table 4Examples of AltematiDg Masculine and
Feminine Singular IDfIected NOUDS
Case
NominativeGenitiveDativeAccusativeInstrumentalLocativeVocative
*Atypical form.
10 FELDMAN AND FOWLER
be extremelyaccurate. Out of 14possibleerrors per condition, the mean number of errors in ConditionsNl , D1,and 11 for words and pseudowords, respectively, were.63, 1.04, and .85 and .65, .42, and .69. The mean number of errors on targets in each condition (NN, DN, IN),computed independently of the error-pairing procedure,was less than .20 for both words and pseudowords.
Finally, means for prime words in each inflected casewere computed and correlated, with nominative-dative,nominative-instrumental, and dative-instrumental latencies groupedas pairs. Because there was a case x genderinteraction among primes, separate correlations weremade for feminine and for masculine word pairs. Thecorrelations ofNl,Dl, Nl,Il, and Dl,11 were r = .38,r = .77, and r =.22, respectively, for feminine wordsand r = .69, r = .77, and r = .82, respectively, formasculine words. No analogous pseudowordcorrelationsapproached significance. With 21 words (and df = 19)correlations of r = 1.441 are significantat the .05 level.In summary, with the exception of correlations involving femininedativellocativecase, correlations among allinflected forms of a noun were significant.
DiscussionDifferences between prime and target in spelling and
pronunciationof the shared morpheme did not eliminatethe effect of repetition. Facilitation with repetition obtained both when target and prime maintained a commonspelling and pronunciation and when they did not. Thisoutcome is consistent with that obtained by Fowler et al.(1985), which showed statisticallyfull priming for alternating English words, and also with many of the resultsreported by Stanners et al. (1979). It is not the same,however, as the outcome of an experiment by Kempleyand Morton (1982) in which irregular morphologicallyrelated words were presented auditorily for recognitionin noise, and in which no priming obtained between irregular and regular forms. Evidently, the outcomeof thepresent study indicates that regular alternations in soundand spelling do not mask morphological relationships.Treating the identityprime condition (NN) as a baseline,there was no significant reduction in facilitation due torepetition when morphological primes differed from targets in spellingand pronunciation (viz., dativeand instrumental masculine primes and dative feminine primes).Statistically, priming was full in all instances. Secondarily, and as describedabove, Fowleret al. (1985) reportedthat a nonsignificant numerical loss in priming typicallyoccurs when affixesof primeand target arenot identical.Results of an analysis of target latencies alonein the presentexperimentreplicatesthe outcomeof Fowler and her colleagues in a studyof English. There is a tendency for primetarget pairs with nonidenticalaffixes to show very smalland nonsignificant reductions in the magnitude of facilitation. Based on these data, overlap in sound and spelling betweentarget and prime (interpreted as a nonlexicalor an episodiccontribution) did not systematically modifythe facilitation that occurs in the repetition priming task.
The coherenceamong satelliteentriesof alternating nounsappears not to differ from that of nonalternating nouns.
Among pseudowords, inspection of means suggestedthat the magnitude of facilitation averaged over genderwas 18 msec when prime and target differed (Experiment 3) and was 58 msec in one condition when primeand target remained the same (i.e., for feminine pseudowordsin Experiment 1). In Experiment 3, the analysesof variancewere significantonly by the subjectsanalysis,and in Experiment 1, there was no facilitationwith repetition for masculinepseudowords. Nevertheless, it is important to point out that the differences among latenciesto pseudowordtargets cannot readily be ascribed to overlap of surfacecharacteristics for target and prime. Inspection of meanssuggested that, irrespectiveof caseof primeand in contrast to the outcomeof Experiment 1, alternating masculine pseudowordtargetswere primedmore consistently than were alternating femininepseudoword targets. However, morphologicalprimes were consistentlyless similar to their targets for masculine pseudowords(whose nominative/accusative was different from alloblique forms) than for feminine pseudowords (whosenominative overlapped formally with instrumental morphological primes but not with dative morphologicalprimes). In summary, the magnitude of facilitation wassignificantly reduced in alternating pseudoword targetsrelativeto that in regularpseudoword targets, but similaritiesof surfacecharacteristics do not accountsatisfactorilyfor the pattern.
For alternating primes, the interactionof case x genderand the pattern of correlations among recognition latencies indicated that the structure of the noun system formasculine and for feminine nounscontrasts. Latencies formasculine nouns supported the usual primacy for thenominative and the equivalence among oblique casesdescribedby a satellite-entries account, whereaslatenciesfor feminine nouns suggested that recognition of the dativellocative was impeded because its spelling and pronunciation were different from its nominative and otherobliquecases. This outcome suggests that at leastfor feminine alternating nouns the structure of the noun systemmay differ from the typical satellite configuration. Pairwise correlations between mean latencies for each wordin its nominative, dativellocative, and instrumental formssupported this interpretation. For masculine nouns, allcases were strongly correlated, whereas for femininenouns, the changed dativellocative form did not correlate significantly with its more regular forms, althoughthe regular cases did correlate with each other.
In summary, deviations in spelling and pronunciationaffect the structure of the inflected noun system as evidenced by latencies for changed dative/locativeforms offeminine alternating nouns that served as primes. Thefailure to demonstrate an analogous effect in masculinenounswas ambiguous, however. It mightreflecta qualitative differencein the irregular spellings. The phonetic environment for the application of the movable A rule orthe O-to-L alternation is perhaps less simply described
than is the environment for palatalization. Alternatively,this failure may provide further evidence for the primacyof the nominativecase. If typicality within a satellite system is defined relative to the nominative form, thenchangednominativeforms of alternatingmasculinenounsmay not, in effect, be deviant. The pattern of correlationssupports the latter interpretation.
In conclusion, the latency data for changed primes suggested that deviation in spelling and pronunciation alterinitial accessibility of inflected forms and the structureof the noun system, whereas the repetition priming dataon target words suggestedthat once an entry has been activated, the nominative nucleus of its noun system is activated as well. Deviations in spelling and pronunciationmay affect the structure of the noun system; it appears,however, that once the satellite entry of either a regularor an alternating noun system has been accessed, the entire noun system is activated.
GENERAL DISCUSSION
In Experiment 1, nouns in the nominative case wereprimed by identical or morphologically related forms,namely, dativellocative and instrumental cases. The outcome was statistically full facilitation by repetition in allprime conditions. This outcome is consistent with theclaim that inflected-noun forms in Serbo-Croatian arestrongly cohesive in the lexicon. The pattern of latenciesfor the primes replicated the pattern from which thesatellite-entries accountoriginated (Lukatelaet al., 1980;Lukatela et al., 1978). Moreover, the latenciesof primeswere significantly correlated. A critical characteristic ofthe satellite-entries account is that the nominative singular case has a special status in the lexical organization.One consequence of its privileged position might be thatthe nominative can prime and be primed more fully bynon-nominative cases thancan any oblique case. The outcomes of Experiments 1and 2 support this interpretation.In Experiment 1, we found full facilitationof nominativetargetsby both identical and morphological primes. In Experiment 2, lexical decision latency to nouns in the dative/locative case was facilitated by a prior presentationof a morphologically related inflectedform. However, instrumental singular primes produced only partial facilitation of dativellocative targets. The statistically significant pattern of full and partial priming was interpretedas evidence that the lexical organization among inflectedcases of a noun is not homogeneous; that is, connectionsamong inflected nouns are not uniformly represented inthe lexicon. In particular, the connection between twosatellites of an entry appears to be weak relative to theconnection between a satellite and the nucleus. Insofaras inhomogeneities in organization are evident, it isdifficult to conceive of a representation in which all inflected forms of a noun either share a base morpheme orare fully independent lexical entries.
In Experiment 3, nounsthat undergo regular sound andspelling changes in at least one of their inflected-case
MORPHOLOGICAL ORGANIZAnON 11
forms were presented as targets in the nominative case.Decisionlatencywas equally facilitated by a prior presentationof all morphologically relatedprimes. Thus, the pattern of facilitation observed does not depend on maintaining phonological and orthographic similarity betweenprime and target: The same outcome obtained with pairsincluding a sound and spelling change and pairs including no change. Likewise, the pattern of facilitation withpseudowordscould not be accountedfor entirely by soundand spellingoverlap. Collectively, the results suggest thatthe representation that underlies repetition priming mustbe sufficiently abstract to accommodate changes in thebase morpheme of morphologically related words.
The effect of repetition priming was consistently morerobust with words than with pseudoword targets and thisoutcome is interpretedas implicating,at least in part, lexical processes. Insofar as facilitation reflects activationamong lexical entries, results indicate that in addition tocapturing inflectional rules that are productive, theserepresentationsalso encompassalternations among formsthat are probably no longer productive.
Finally, the pattern of decision latencies for regularnoun primes and the correlation among forms indicatesthat inflected forms of a noun are associated. This outcome is interpreted as reflecting the structure of the nounsystem. For feminine alternating nouns, however, latencies were associated only when both words had identicalbase morphemes. Failure to observe a significant correlation between atypical and typical forms of alternatingnouns lends support to the assumption that the pattern ofcorrelations reflects, at least in part, lexical factors. Thepattern of decision latencies and correlations for masculine alternating nouns that had a changed nominative/accusativecase indicated that they were handled likeregular nouns: All cases were associated. This outcomepermits two interpretations: Either the nominative caseis special such that alternation is defined relative to thenominative or, alternatively, that the particular sound andspelling changes that appear in the present set of masculine words are different from the changes that occur infeminine words. Discussion of the specifics by which alternating inflectionalforms are represented and their rolein definingthe satelliteorganizationamong entries shouldnot be allowed to obscure the basic result. The outcomeof the present series of experiments is consistent with theclaim that inflected cases of a noun are represented fullybut not independently and that morphological relatednessprovides a principleof organization in the lexicon. In thisrespect, the present experiments conducted in the highlyinflected language of Serbo-Croatian are consistent withresults of repetition priming studies conducted with English materials (Fowler et al., 1985).
In summary, the present study extends the satelliteentries account of Lukatela and colleagues (Lukatelaet al., 1980;Lukatelaet al., 1978)in the following ways:The equivalence of decisionlatenciesfor all obliqueformsobservedwith nonaltemating nouns was not observedwithfeminine alternating nouns. These data, in conjunction
12 FELDMAN AND FOWLER
with the correlationsbetweenlatenciesfor intlected-easeforms, support theclaimthatalternating nouns do notconfigure in the typical satellite fashion. In the presentstudy,thepattern of full and partialfacilitation in repetition priming was deployedto probe the organization among satellite entries as a further extension of Lukatela's work.Amongregularnounsystems, the facilitation was alwaysfull for nominative targets, whereas facilitation was significantly diminished when an oblique-ease target wasprecededby a differentoblique-ease prime. Ifmagnitudeof facilitation can be interpreted as an index of the organization within the inflected noun system, then theseresults reveal inhomogeneities in the coherence of thesatellite system. Specifically, theconnections between twosatellite entries that represent different intlected-easeforms are weaker than the connectionbetween an entryand its nucleus. In contrast, the connections betweenthenominative nucleusand all of its intlected-ease satellitesare equallystrong. The latter outcome can be interpretedas further evidence for the primacy of the nominative.Finally, when typical and atypical forms of alternatingnouns were presented as primes, decision latencies tonominative targets revealed a pattern of facilitation thatwas comparable to that reported with nonalternatingnouns. This outcome, namely full facilitation, suggeststhat oncea satellite entry is activated, thenall componentsof its noun systemare accessed,and that this is true bothfor alternatingand nonalternating nouns. In conclusion,althoughthe noun systemof alternatingand nonalternating may differ, once access to an entry occurs, it necessarily entails the activation of its entire noun system.
REFERENCES
BALOTA, D. A.,.t: CHUMBUlY, J. I. (1984). Are lexical decisions a goodmeasure of lexkal access? The roleof word frequency in theneglecteddecisionstage. JoumDlofExperimental Psychology: Human Perception de PerfoT71lll1/Ce, 10, 340-357.
BEue, A. (1976). Fonetika. Universityof Belgrade: NaucnaKnijiga.BUIlANI, C., SALMASO, D.,.t: CARAMAZZA, A. (1984). Morphological
structureand lexical access. Visible Languag«, 18, 348-358.CAllAMAZZA, A., M1cEu, G., SILVERI, M. c., '" LAUDANNA, A. (1985).
Reading mechanisms and the organizationof the lexicon: Evidencefrom acquired dyslexia. CognitiveNeuropsychology, 2, 81-114.
DANNENBRING, G. L., '" BRIAND, K. (1982). Semantic priming andthe word repetitioneffect in a lexical decision task. Canadian Journal of Psychology, 36, 435-444.
FELDMAN, L. B. (1984, November).Lexical' organization ofmorphowgical relatives. Paper presentedat thePsychonomic Societymeeting, San Antonio, TX.
FELDMAN, L. B. (in press). Phonological and morphological analysisby skilled readers of SerOO-Croatian. In A. Allport, D. Mackay,W. Prinz, & E. Scheerer (Eds.), Umguage perceptionand production. London: Academic Press.
FELDMAN, L. B., KOS'I'IC, A., LUXATELA, G., '" TuRVEY, M. T. (1983).An evaluationof the "Basic OrthographicSyllabic Structure" in aphonologically sballow orthography. Psychological Research, 45,55-72.
FELDMAN, L. B., .t:MOSKOVUEVlC, J. (in press). Repetition primingis notepisodicin origin. JoumDl ofExperimental Psychology: Learning, Memory, de Cognition.
FELDMAN, L. B., .t:TURVEY, M. T. (1983, November). Morphologicolprocesses in wordrecognition. Paper presented at thePsychonomicSociety meeting, San Diego, CA.
FEUSTEL, T. C., SHIFFRIN, R. M., '" SALASOO, A. (1983). Episodicand lexical contributionsto the repetition effect in word identification. Journal of Experimental Psychology: General, lU, 309-346.
FORBACH, G. B., STANNERS, R. F., '" HOCHHAUS, L. (1974). Repetition and practiceeffects in a lexicaldecisiontask. Memoryde Cognition, 2, 337-339.
FORSTER, K., '" DAVIS, C. (1984). Repetitionprimingand a frequencyattenuation in lexical access. Joumal of Experimental Psychology:Learning, Memory, de Cognition, 10, 680-698.
FOWUlR, C. A., NAPPS, S. E., '" FELDMAN, L. (1985). Relations amongregular and irregular morphologically related words in the lexiconas revealed by repetitioo priming. Memory de Cognition, 13,241-255.
HANSON, V. L., '" WILKENFELD, D. (in press).Morphopbonology andlexcialorganizatioo indeafreaders. Umguagede Speech, 28, 269-279.
HI!NDERSON, L., WALUS, J., ",KNIGHT, D. (1986). MorphemicStnlClure and lexicalaccess. In H. Bouma& D. Bouhuis (Eds.), Attentionand performance X. London: Erlbaum.
HUDSON, P. T. W., '" BI!RGMAN, M. W. (1985). Lexical knowledgein word recognition:Word lengthand word frequencyin naming andlexical decision tasks.Joumal ofMemory de Umguage, 24, 46-58.
KEMPLEY, S. T., '" MORTON, J. (1982). Theeffects of priming withregularlyand irregularlyrelatedwords in auditoryword recognition.British Journalof Psychology, 73, 441-454.
KOS'I'IC, OJ. (1965). Frequency ofoccurrence ofwords in Serbo-CroaJion.Unpublished manuscript, Institute of Experimental Phonetics andSpeech Pathology, University of Belgrade.
LUKATELA, G., GUOORDEVIC, B., KOS'I'IC, A., '" TuRVEY, M. T. (1980).Representation of inflectednouns in the internallexicon. MemorydeCognition, 8, 415-423.
LUKATELA, G., MANDIC, Z., GUOORDEVIC, B., K<JSTJC, A., SAVIC, M.,'" TURVEY, M. T. (1978). Lexicaldecisionfor inflectednouns. Ianguage de Speech, 21, 166-173.
MURRELL, G. A., '" MORTON, J. (1974). Word recognitionand morphemic structure. JoumDl ofExperimental Psychology, 102,963-968.
NAPPS, S. (1985). Morphologicol, semantic, andformal relatioos amongwordsand rlu! organization ofrlu! mental lexicon. Unpublished doctoral dissertation, Dartmouth College, Hanover, New Hampshire.
NAPPS, S., '" FOWLER, C. A. (1983, November). Orthographic organization oflexical forms. Paper presentedat the Eastern PsychologicalAssociation Meeting, Philadelphia, PA.
NAPPS, S., .t:FOWLER, C. A. (in press). The effect of orthography onthe organizationof the mental lexicon. Journal of PsycholinguisticResearch.
OUPHANT, G. (1983). Repetition and recency effectsin lexical memory.Australian Journal of Psychology, 35, 393-403.
ScARBOROUGH, D. L., CoRTESE, C., '" ScARBOROUGH, H. (1971). Frequency and repctitioo effects in lexical memory. JoumDl ofexperimental Psychology: Human Perception de Performance, 3, 1-17.
SEIDENBERG, M. S., '" TANENHAUS, M. K. (1986). Modularity andlexical access. In I. Gopnik& M. Gopnik(Eds.), Frommodelsto modules: Studies in cognitivesciencefrom the McGill Workshops. Norwood, NJ: Ablex.
STANNERS, R. F., NEISER, J. J., HERNON, W. P., '" HALL, R. (1979).Memory representationfor morphologically related words. Journalof VerbalLearning de VerbalBehavior, 18, 399-412.
TAFT, M. (1979). Lexical accessvia an orthographiccode: The BasicOrthographicSyllable Stnleture (BOSS). Journal of VerbalLearning de Verbal Behavior, 18, 21-40.
TAFT, M., '" FORSTER, K. I. (1975). Lexical storage and retrieval ofprefixedwords. JournalofVerbal Learning de Verbal Behavior, 14,638-647.
(Manuscript received January 2, 1986;revision accepted for publicationJune 23, 1986.)