Haskins Laboratories Status Report on Speech Research1992, SR·1111112, 181·192

Morphological Analysis in Word Recognition*

Laurie B. Feldmant and Darinka AndjelkoviC+

One of the most refined techniques forinvestigating morphological processing in wordrecognition is the variant of the lexical decisiontask known as repetition priming (Stanners,Neiser, Hernon, & Hall 1979). It provides aprimary source of evidence, according toHenderson (1989), of facilitation between wordsformed from the same morpheme (i.e.,morphological relatives). Generally, target (secondpresentation) decision latencies and error ratesare reduced in the context of morphologically­related primes (first presentation). Words relatedto the target (e.g., HEALS) can be forms that areunaffixed (e.g., HEAL), inflected (e.g., HEALED)or derived (e.g., HEALER) in either the same ordifferent modalities (e.g., print or speech) and theycan be separated by as many as fifty interveningitems. Effects of morphological relatedness havebeen observed in the lexical decision task across avariety of languages including Serbo-Croatian(Feldman & Fowler, 1987) and Hebrew (Bentin &Feldman, 1990), English (Fowler, Napps, &Feldman, 1985; Feldman, 1991) and AmericanSign Language (Hanson & Feldman, 1991; seealso Emmorey, 1989).

In this chapter, we review evidence of morpho­logical processing in word recognition. In the firsttwo sections, studies of morphology that employthe repetition priming techniques are described.In section one, morphological and orthographicsimilarity effects are contrasted because alterna­tive accounts of morphological effects in wordrecognition often minimize the role of the mor­pheme unit and focus on orthographic and phono­logical similarity of morphologically-related

The research reported here was conducted at the Laboratoryof Experimental Psychology at the University of Belgrade andat Hebrew University in Jerusalem and was supported byfunds from National Institute Of Child Health andDevelopment Grant HD·01994 to Haskins Laboratories and byfunds from the University of Belgrade.

181

words. In section two, morphological effects aredifferentiated from effects due to semanticassociation, although both are based on wordmeaning. The repetition priming procedure is aviable tool for studying how morphologicalrelationships among words are represented in thelexicon; however, a confounding episodiccontribution to the pattern of facilitation can alsooccur (e.g., Feustel, Shriffrin, & Salasoo, 1983). Itis important, therefore, to differentiatemorphological effects from episodic and othertypes of facilitation and to provide convergingevidence of morphological analysis from otherword recognition tasks. Morphological effects havealso been observed in sentence verification andoral reading tasks. For example, facilitation due toshared morphemes (and/or shared syntacticstructure based on the ordering of subject, objectand verb constituents) in prime and target

.sentences have been obtained. In section three,morphological and syntactic facilitation effects areexamined in a sentence comprehension task.Finally, it is useful to ask whether analysis of aword's constituent structure at the level of themorpheme is linked to analysis of that word atother linguistic levels. For example, theassociation between morphological and phonologi­cal processing in beginning readers has been ex­amined by comparing performances on such tasks.In section four, a common underlying skill formorphological and phonological analysis is re­vealed.

1. Distinguishing between orthographic andmorphological effects

The standard morphological formation processesin English typically entail prefixation andsuffixation to a base morpheme. As a consequence,forms with a common base morpheme generallyshare orthographic and phonological as well asmorphological structure. For regular forms, thereis structural transparency (Henderson, 1989) in

182 Feldman and Andjelkovic

that related forms are structured around the samebase morpheme (e.g., COMPUTE- COMPUTER).Moreover, to the extent that compositionality ispresent, related words will commonly also havesimilar meanings. Covariation of morphologicaland orthographic structure in related formsinvites an orthographic account of themorphological effects observed in tasks suchrepetition priming.

In the repetition priming task, first (prime) andsecond (target) presentations are separated by anaverage of ten and sometimes as many as 50 in­tervening items. Target latency as a function oftype of prime is examined. Changes in spelling orpronunciation tend not to significantly diminishthe effect of morphological relatedness in this taskso that prime-target pairs such as SLEPT-SLEEPor HEALTH-HEAL produce facilitation equivalentto SLEEP-SLEEP or HEALED-HEAL respectively(Fowler et a1., 1985; Feldman & Moskovljevic,1987). Similarly, formal similarity of morphologi­cally-unrelated prime and target (viz., phonologi­cally and orthographically but not morphologicallysimilar pairs such as DIET and DIE) does not re­sult in priming at these long lags (Feldman &Moskovljevic, 1987; Hanson & Wilkenfeld, 1985;Napps, 1989; Napps & Fowler, 1987). Despitean absence of effects due exclusively tospelling/pronunciation and orthographic form inthe repetition priming literature, there persists atendency to try to interpret morphological effectsas effects of orthographic structure. For example,Seidenberg (1987) following Adams (1981) sug­gested that patterns of high and low bigram fre­quency could account for morphological patterningbecause transitional probabilities of letter se­quences that straddle a (syllabic or) morphologicalboundary tend to be low (bigram troughs) relativeto probabilities of sequences internal to a unit (cf.Rapp, 1992).

The similar response patterns for regular andirregular forms described above provide evidenceagainst an orthographic account of morphologicalfacilitation in repetition priming. Othermorphological effects inconsistent with anorthographic account are based on (1)manipulations of alphabet and on the ( 2) theabsence of an effect when a target is preceded byan unrelated word with a similar orthographicand morphological structure. These findings willbe reviewed in the remainder of section one.

Many readers of Serbo-Croatian are fluent inboth the Roman and Cyrillic alphabets. For suchreaders, this situation has been exploited in thestudy of morphological processing (Feldman &

Moskovljevic, 1987; Experiment 1). The rationalewas that if the facilitation observed in repetitionpriming arises in a relatively early stage ofprocessing and represents repeated analysis of thesame orthographic pattern, then the facilitationdue to morphological relatedness should differwhen successive presentations of the target wordalternated alphabet versus when they preservedalphabet. In that study (Feldman & Moskovljevic,1987), lags ranged from 7 to 13 items andalphabet was manipulated between subjects sothat a one alphabet (preserved) and a twoalphabet (alternated) condition existed. Because itis possible that the time course of activation ofvisual form varies with lag (Monsell, 1985; Ratcliff,Hockley, & McKoon 1985), a second study wasconducted in which alphabet was manipulatedwithin subjects and a more expanded range of lags(3 to 20 intervening items) was included(Feldman, 1992). In both, words and pseudowordswere presented twice, with a lag of interveningitems. Subjects who were students at theUniversity of Belgrade were instructed to performa lexical decision to each letter string as itappeared. In the alternated alphabet condition,prime and target were transcribed in differentalphabets (e.g., HOrA-NOGOM). In the preservedalphabet condition, prime and target were in thesame alphabet (e.g., NOGA-NOGOM). Decisionlatencies to targets that were preceded by primeswhere target and prime either alternated orpreserved alphabet were compared in an attemptto find evidence for facilitation based onrepetitions of specific visual patterns.

Results of the two alphabet alternationexperiments are summarized in Table 1. Inneither experiment was the effect of alphabet(preserved/alternated) significant. Moreover, inthe latter experiments, it was the case that forwords, neither the effect of lag nor the interactionof alphabet by lag. was significant. Statedgenerally, significant target facilitation occurredwhen primes appeared in either the samealphabet or in a different alphabet from the targetand target facilitation was no greater in thealphabet preserved condition than in thealternated condition. Obviously words presentedand represented in the same alphabet are morevisually similar than are the Roman and Cyrillicalternatives of a word. Yet, in the repetitionpriming task where several items intervenedbetween first and second presentations, nosignificant increment to facilitation was observedon the alphabet preserved trials relative to thealphabet alternating trials. This finding was

Morphological Analysis in Word Recognition 183

observed at lags as short as 3 and as long as 20items.

Table 1. Mean decision latencies (ms) and errors forwords in the alphabet preserved and alphabetalternated conditions of the repetition priming task.

(Feldman and Moskovljevic. 1987; Experiment 1.)

Repetition: AlphabetLag Alternated Preserved Difference

unrelated but homographic form (e.g., BORAMAfrom BORA). An analysis of variance revealed asignificant effect of prime type in both reactiontime and errors. Results are summarized in Table2. Target decision latencies were 589 ms in theidentity condition, 617 ms in the morphologicalcondition and 656 ms in the orthographiccondition. Decision latencies in the no primecondition were 661 ms. Target error rates were3.3% in the identity, 4.4% in the morphologicaland 12.6% in the orthographic condition. Errorrates in the no prime condition were 16.7%. Theeffect of prime type was significant with bothsubjects (F1) and item (F2) as random effectvariables, with both reaction time and errors asdependent measures. Post-hoc tests indicated thatthe morphological and orthographic primeconditions were significantly different.

Table 2. Mean reaction times and errors for targets(e.g., BOROVI) following identity, morphological andorthographic primes and for first presentations of thetarget in repetition priming.

9090

506.1595.4448.2565.4

601

552

5927.3

5957.3

6.6

588

6074.5

10

1010

20

642678

FirstPresentation

words65112.7

(Feldman. 1992; Experiment 1a)

(Feldman, 1992; Experiment 1b)

words628 3 562 6610.8 8.3 2.5

562 665.9 4.9

10 567 617.9 2.9

573 557.9 2.9

A second strategy for differentiatingmorphological and orthographic effects entailedcomparing morphologically-related primes tounrelated primes that share orthographicstructure. In Serbo-Croatian, it is possible toidentify pairs of unrelated words that are formedaround homographic base morphemes. Forexample, the word "BOR" in nominative singular,meaning "pine," is masculine in gender while theword "BORA," meaning wrinkle, in nominativesingular is feminine. They have homographic basemorphemes spelled BOR but, because of genderdifferences, require different sets of inflectional af­fixes. In a repetition priming task (Feldman &Andjelkovic, 1991; Experiment 3), targets (e.g.,BOROVI from BOR) were preceded an average often items earlier in the list by an identicalrepetition (e.g., BOROVI), by a morphologically­related form (e.g., BOR) or by a morphologically-

Prime Type Example Latency Errors

Identity borovi 589 3.3Morphological bor 617 4.4Orthographic borama 656 16.7First Presentation 661 16.7

The orthographic and no prime conditions didnot differ with either the reaction time or theerror measure. That is, no facilitation withorthographically similar. but morphologicallyunrelated words was observed when an average often items intervened between first and secondpresentations in a repetition priming task. Theseresults are consistent with an earlier experimentconducted with Serbo-Croatian materials(Feldman & Moskovljevic, 1987; Experiment 2) inwhich decision latencies for target words such asSTAN meaning "APARTMENT" were reduced bythe prior presentation of a derivationally-relatedprime such as the word STANCIl: meaning"LITTLE APARTMENT." By contrast, the wordSTANICA did not reduce target latencies. Notethat this word is morphologically unrelated butorthographically similar to the target, is composedof one morpheme and means "BUS STATION." Incontrast to Feldman and Moskovljevic (1987), inthe present study, orthographic primes, asmorphological primes, were morphologicallycomplex forms consisting of a base morpheme and

184 Feldman and Andjelkovic

an inflectional affix. Nevertheless, orthographicprimes were equivalent to the no prime condition.Collectively, these studies refute the hypothesisthat orthographic similarity underliesmorphonological facilitation in repetition priming.

In summary, relative to the no prime condition,both morphological relatives and identicalrepetitions facilitated target recognition. Theorthographic prime condition was not significantlydifferent from the no prime condition. Finally, andmost important to the present discussion ofmorphological effects in repetition priming, targetlatencies and errors following morphologicalprimes and orthographic primes at long lags weresignificantly different both in the analysis bysubjects and in the analysis by items.

Orthographic and morphological primes alsodifferentially influenced target latency whenpresented immediately in succession. In atraditional immediate priming task,morphological primes facilitate and orthographicprimes may inhibit. However, the orthographiceffect is sensitive to the density of theorthographic neighborhood of the prime (Forster,Davis, Schoknecht, & Carter, 1987) as well as therelative frequency of prime and target and thepresence or absence of a mask (Segui & Grainger,1990). For example, without a mask, lowerfrequency orthographic primes tend to inhibitwhereas with a mask, it is the higher frequencyprime that shows inhibition.

A recent report with French materials isconsistent with this characterization ofmorphological as contrasted with orthographicprimes (Grainger, Cole & Segui, 1991). In thatstudy, masked primes consisted of morphological,orthographic or unrelated derivations of thetarget. Decision latencies for targets were fastestin the morphological condition (619 ms),(numerically but not statistically) slowest in theorthographic condition (653 ms) and intermediate(639 ms) in the unrelated condition. In thosematerials, however, orthographic primes (e.g.,AFFIRME-REFORME) tended to be less similarto the target than were morphological primes (e.g.,DEFORME-REFORME) and this might accountfor the marginally significant inhibition in theorthographic condition. Nevertheless, the criticalpoint is that morphological primes showedfacilitation whereas orthographic primes showedweak inhibition, at best, and conservatively, nodifference from the unrelated condition.

A study recently completed in Serbo-Croatianreplicates the difference between orthographic andmorphological primes presented in immediate

succession with an unmasked prime (Feldman &Andjelkovic, 1991). In a series of two experiments,targets consisting of morphologically-complexforms were preceded by either a morphologicalrelative, an unrelated word formed from ahomographic base morpheme or anorthographically and morphologically unrelatedword. For example, the target BOROVI waspreceded by (1) BOR which is inflectionallyrelated, (2) BORI which is not relatedmorphologically although it is orthographicallysimilar because its base morpheme ishomographic and by (3) KRV which is unrelatedalong both morphological and orthographicdimensions. In one experiment (Experiment 2),primes were of higher frequency than targets. In asecond (Experiment 3), primes were of lowerfrequency than targets. In both experiments,primes without masks were presented touniversity students in Belgrade for 250 ms andfollowed by a blank for 50 ms after which thetarget appeared for 1000 ms. Latencies greaterthan 2SD from the mean were treated as errors.Results are summarized in Tables 3 and 4.

Table 3. Mean reaction times (and percent errors) fortargets (e.g., borovi) following morphological,orthographic and unrelated primes. Primes were higherin relative frequency than their targets.

Prime Type Example Latency Errors

Morphological bor 684 24Orthographic borovi 754 45Unrelated krv 738 38

Table 4. Mean reaction times (and percent errors) fortargets (e.g., bori) following morphological,orthographic and unrelated primes. Primes were lowerin relative frequency than their targets.

Prime Type Example Latency Errors

Morphological borovi 687 15Orthographic bor 743 46Unrelated krv 746 30

Morphological and unrelated primes differedsignificantly at short lags in both experiments andthis outcome replicates the morphologicalfacilitation observed at longer lags with similarmaterials. When primes were less frequent thantargets, significant orthographic inhibition (F1and F2) was evident in the error measure but not

MorphologIcal Anall/sis In Word Recognztlon 185

in the reaction time measure; This finding isconsistent with the results of Segui and Grainger(1990) using morphologically simple materials andunmasked primes. When primes were morefrequent than targets, orthographic inhibition wasnot statistically significant although the reactiontime pattern was quite similar to the patternobtained when primes were less frequent thattheir targets.

In Italian, inhibitory effects betweenhomographic base morphemes (e.g., FINA-FINIREwhich mean "thin" in feminine singular and "tofinish," respectively) have been reported in both adouble lexical decision (viz., are both letter stringswords?) and in a lexical decision task where primeand target are presented successively (Laudanna,Badecker, & Caramazza, 1989). Results wereinterpreted as evidence of inhibitory connectionsbetween homographic base morphemes in thelexicon. By this account, a differential effect of therelative frequencies of prime and target is notanticipated although it could be accommodated.More problematic is the failure to observeinhibition among homographic forms concurrentwith facilitation among morphologically-relatedforms at longer intervals between presentationsviz., at the lags incorporated into the repetitionpriming task. If inhibition reflects a principle oflexical organization then a justification for itssensitivity to lag is needed. Regardless of whetherorthographic primes slow or impair accuracy totargets relative to unrelated primes and whetherhomographic base morphemes relative to othertypes of orthographic controls pose a specialproblem for the representation of morphology, it isuseful to focus on the reliable facilitation obtainedwhen items are morphologically-related ascompared to either unrelated or orthographicconditions. This is true both in repetition primingwith average lags of ten items and in successivepriming with or without a mask.

In summary, morphologically-related words thatundergo changes in spelling and/or pronunciationso that the base morpheme is partially obscuredproduce the same pattern of facilitation as do re­lated forms that are structurally transparent.This finding has been observed in Serbo-Croatian(Feldman & Fowler, 1987) as well as English(Fowler et al., 1985; see also Kelliher &Henderson, 1990; Nagy, Anderson, Schommer,Scott, & Stillman, 1989) and presents a challengefor an orthographic account of facilitation in therepetition priming task. In addition, for morpho­logically-related prime-target pairs, the effect ofpresenting repetitions in the same alphabet was

no different than the effect of alternating alphabet(Feldman & Moskovljevic, 1987; Feldman, 1992).This outcome suggests that the basis of facilitationmust be sufficiently abstract to tolerate changes invisual form introduced by manipulations of alpha­bet. Finally, when homographic base morphemesappeared in prime and target, facilitation was ob­served among forms that shared a base morphemebut not among unrelated forms. In fact, the con­trast between homographic and no prime condi­tions sometimes revealed marginal inhibition.Evidently, morphological effects cannot bedescribed in terms of shared orthographicstructure.

2. Distinguishing between semantic andmorphological effects

Related forms, by definition, share a basemorpheme. Because morphemes are generallydefined as units of meaning, it is plausible thatmorphological facilitation reflects the semanticsimilarity of prime and target. Linguistsdistinguish between two types of morphologicalrelatives, inflections and derivations, and theseforms differ with respect to the productivity ofrules and the predictability of their meaning froma semantic analysis of the base and itsconstituents (Aronoff, 1976). Whereas inflectionsrarely produce new shades of meaning,derivations are much less constrainedsemantically and historically often changemeaning once formed (e.g., TERRIFIC andTERROR). For example (from Henderson, 1985),the prefix UN typically modifies the base adjectiveor verb in a predictable manner (e.g., UNCLEAR,UNDRESS) but some forms are derived fromobsolete or rare bases (e.g., UNKEMPT) (Lakoff,1971; Jackendoff, 1975). Moreover, forms such asUN+base+ABLE are semantically ambiguousinsofar as the prefix can modify either thebase verb (V) or the (V+ABLE) adjective(e.g., UNDOABLE. may mean UNDO+ABLEor UN+DOABLE). Inconsistencies ofsemantic composition are obvious in semanticcomparisons of base-derivation pairs suchas DISCUSS-DISCUSSION, CONGREGATE­CONGREGATION and PROFESS-PROFESSIONand point to the unpredictability inherent in asemantic analysis of some complex forms fromtheir constituents.

Nevertheless, morphologically-related wordstend to have similar meanings. Moreover, becausesemantic facilitation has been so thoroughlystudied (see Neely, 1991, for a review), it isimportant to contrast facilitation due to

186 Fe/dman and Andje/kame

morphological relatedness and to other types ofsemantic relatedness. Semantic contributions tothe pattern of facilitation in repetition primingwere explored with derivational relatives inEnglish and in Hebrew. In a repetition primingstudy with English materials (Feldman, 1991),semantic overlap was first assessed by a separategroup of subjects who scaled the items forsemantic distance. (Due to the constraints ofEnglish, word class changes and other aspects ofsemantic predictability were not well controlled.)For each target, a morphological relative close andremote in meaning was selected. Small butstatistically equivalent effects of the priorpresentation of the same morpheme in a relatedword were observed for derivational relatives thatwere semantically close (36 ms) and semanticallyremote (30 ms) whereas identical presentationsproduced robust facilitation (93 ms). Evidently,extent of semantic overlap did not influence thepattern of facilitation in repetition priming.

In a second study (by C.A. Fowler, reported inFeldman, 1991), a target (e.g., HOT) was precededat least ten items earlier in the list by the sameitem (e.g., HOT) or by a strong antonym (e.g.,COLD). No facilitation was observed in theantonym condition relative to the identity and noprime (i.e., initial) conditions. In the sense thatthese items were highly predictable andsemantically constrained, it is difficult to arguethat semantic similarity underlies facilitation inrepetition priming.

A third study (Bentin & Feldman, 1990)compared facilitation by associative andmorphological primes at long and at short lags.Materials were Hebrew words; prime conditionsconsisted of morphological, semantic or bothmorphological and semantic relatives of the

target. For example, the word meaning"LIBRARY" was preceded by the word for"NUMBER" which is formed from the same root orbase morpheme, by the word for "LIBRARIAN"which is also formed from the same root and bythe word for "READING" which is semanticallybut not morphologically related. Magnitude ofmorphological facilitation did not change over lagswhereas that for semantic facilitation did.Moreover, when the prime immediately precededthe target, semantic and semantic-plus­morphological primes showed greater facilitationthan did morphological primes but when anaverage of ten items intervened, morphologicaland semantic-plus-morphological showedequivalent facilitation. Evidently the patterns ofassociative and morphological facilitation aredifferentially affected by lag (see Table 5).

In conclusion, similarity of meaning betweenmorphologically-related prime and target does notaffect the pattern of facilitation in repetitionpriming. Morphological relatives closely relatedand remotely related semantically both producedthe same pattern of facilitation at long lags.Moreover, closely related antonym pairs producedno effect under conditions where morphologicalrelatedness effects were observed. Patterns ofmorphological facilitation in this task are,therefore, not easily described in terms ofsemantic similarity. In addition, morphologicaleffects occurred at separations between prime andtarget that considerably exceed those at whichsemantic/associative priming has beendemonstrated (Bentin & Feldman, 1990; see alsoDannenbring & Briand, 1982; Emmorey, 1989;Henderson, Wallis & Knight, 1984; Napps, 1989).Evidently, morphological and semantic facilitationreflect different underlying mechanisms.

Table 5. Mean reaction times and errors for targets (e.g., iT"7!iO meaning "library") following morphological,semantic plus morphological. semantic and no prime in repetition priming (Bentin & Feldman. 1990).

La!:lag 0 lag 15

Prime Type Example Meaning RT Errors RT Errors

Morphological .,!'CO number 589 1.8 587 1.0

Semantic/Morphological l"!,C librarian 559 1.0 583 2.6

Semantic iTN""p reading 563 2.1 611 1.1

Filler 606 1.2 606 1.2

Morphological Analysis in Word Recognition IS7'

3. Morphological effects in sentence contexts

The study of word recognition is sometimes rep­resented as the interface between the domains ofperception and language processing. It is the case,however, that the status of linguistic codes inword recognition is problematic (Henderson,1989). In the first two sections of this chapter, weshowed that morphological effects could be exper­imentally differentiated from effects of (1) sharedorthographic and (2) shared semantic structure.In the next two sections, the relation betweenmorphology and other types of linguistic pattern­ing are examined. In section three, patterns of fa­cilitation due to morphological and syntactic simi­larity between prime and target sentences areexplored. In section four, the association betweenmorphological and phonological analysis is exam­ined in beginning readers.

Effects of morphological relatedness have beenobserved in sentence contexts as well as inisolated words. In one study (Feldman &Andjelkovic, 1990), students from the Universityof Belgrade were presented with pairs ofsentences that either shared the same syntacticstructure (subject verb (SV) or verb object (VO»and/or shared the same base morphemes. In oneexperiment, subjects were required to read thesentence aloud and onset to vocalization wasmeasured. A similar task has been reported toshow effects of syntactic structure (Bock, 1986;1990). In two related experiments, subjects wererequired to judge whether the sentence madesense and latency and errors were measured. Forexample, subjects saw target sentences such asVODICI PLIVAJU which means "The guidesswim" and has a subject (S) verb (V) structure.Across subjects, that sentence was preceded byfour types of prime sentences. These included (1)Morphologically unrelated and structurallydissimilar primes consisting of sentences such asCITA KNJIGU which means "He reads a book."This sentence has a verb (V) object (0) structure(which is grammatical in Serbo-Croatian becausepronouns need not be specified outside the verb).Also included were (2) morphologically unrelatedbut structurally similar primes consisting ofsentences such as ZENA CITA which has a SVstructure and means "The woman reads." and (3)morphologically related and structurally similarprimes consisting of sentences such as VODICPLNA which has a SV structure and means "Theguide swims." Finally, there were (4) morpho­logically related but structurally dissimilar primes

. consisting of sentences such as VODI PLIVACA

which has a VO structure and means "He guidesthe swimmer." Primes and targets wereconstructed so that the same order of the two basemorphemes (i.e., VOD- and PLIV-) wasmaintained over all prime and target sentences inwhich they appeared.

Foil sentences were morphologically related' orunrelated and had either the same or a differentconstituent structure. Morphologically relatedfoils contained the same base morphemes in ille­gal combinations such as verbal affixes on nounsand nominal affixes on verbs. Morphologically un­related foils were composed of legal morphologicalcombinations but were semantically anomalous.The primes for foil sentences were always seman­tically and syntactically acceptable.

In the oral reading task, the prime and targetmembers of a pair were presented in differentalphabets. Primes were always presented inRoman and targets in Cyrillic. In this way, thevisual similarity of successive presentations of amorpheme was reduced. Significant effects ofmorphological relatedness were observed in theoral reading task. Effects of structural similaritywere absent (see Table 6). While this outcome canbe interpreted as evidence of facilitation due torepetition of base morphemes in prime and targetsentences, an alternative account of morphologicaleffects in this task focuses on the repetition of theinitial syllable (e.g., VOD-) in all related sentencesbut not in unrelated sentences. Therefore, it isimportant to replicate the morphological effect ina task where an advantage based on repetition ofthe first syllable effect seems unlikely to occur.

In the verification task, subjects had to decidewhether each target sentence made sense.Therefore, latencies presumably reflect more thanjust processing of the first syllable. All prime sen­tences were meaningful as were half of the targetsentences. Primes and targets were printed in thesame alphabet. Otherwise, materials were identi­cal to those of the previous experiment. Resultsindicated that latencies were significantly longerfollowing morphologically unrelated primes thanfollowing morphologically related primes (seeTable 6). In addition, latencies were longer follow­ing structurally dissimilar sentences than follow­ing structurally similar sentences in both themorphologically related and the unrelated condi­tions. The interaction was significant by Fl butnot by F2. As in the previous experiment, the ef­fect of morphology could reflect priming of mor­phemic units over different sentence structures.Alternatively, it could simply reflect episodic repe­titions of the same letter sequence (e.g., VOD-) in

188 Feldman and And;elkovic

sentence initial position. The episodic accountseems unlikely in a verification task where theentire sentence must be processed before respond­ing. Moreover, it does not explain the significantdifference between morphologically related primesentences with same and different structures (i.e.,901 vs. 975). In addition, when morphemes werenot repeated (i.e., for morphologically unrelatedsentences), structurally dissimilar primes andstructurally similar primes had significantly dif­ferent effects on target latencies. This outcome in­dicates that the verification task is sensitive tosentence structure defined over different surfaceforms. In summary, when the experimental taskrequires that the entire sentence be processed, fa­cilitation can arise between sentences with similarstructures.

Table 6. Oral reading and verification times (errors inparentheses) for target sentences primed bymorphologically and structurally similar sentences(From Feldman & AndjelkoviC, 1990).

target: VODICI PLIVAJU (SV)

Sentence structureSame structure Different structure

MorphologyRelated VODIC PLIVA (SV) VODI PLIVACA (VO)Unrelated ZENA CITA (SV) CITA KNJIGU (VO)

Task

Oral reading:Related 676 690

(5) (8)Unrelated 718 730

(9) (9)

Verification:Related 901 975

(3) (16)Unrelated 973 1002

(12) (16)

An examination of the materials used in theprevious two experiments revealed that morpho­logically related and structurally related primesconsisted of words related by inflection to the tar­get sentence whereas morphologically related butstructurally dissimilar sentences generally con­sisted of words related to the target by derivation.In a final experiment in this study, effects of sen­tence structure and morphology were again inves­tigated in a sentence verification task. In contrastto the previous experiments, here all critical items

consisted of morphologically related prime-targetsentences. Moreover, in addition to sentencestructure, type of morphological relatedness (viz.,inflection/derivation) was manipulated. Theessence of stimulus construction entailed identifY­ing base morphemes that could function as part ofeither a noun or a verb. For example, subjects sawtarget sentences (constructed around the mor­phemes PLIV- which means "swim" and VOD­which means "guide") such as PLIVAJU VODICIwhich means "The guides swim" and has a VBstructure. As in the previous experiments, thatsentence was preceded by four types of prime sen­tences across different groups of subjects. Allprimes were morphologically related by either in­flection or derivation to the target. In inflectedsentences, the word class of the base morphemeswas preserved over prime and target sentences. Inderived sentences, the word class of the base mor­phemes changed in prime and target sentences. Inaddition, primes and targets varied with respectto similarity of sentence structure. Four combina­tions of sentence structure and morphology werepossible: (1) structurally dissimilar inflectionalprimes consisting of VO sentences such as PLIVAKA VODICU which means "He swims toward theguide." (2) structurally dissimilar derivationalprimes consisting ofVO sentences such as VODISPLIVACA which means "You guide the swimmer."(3) structurally similar inflectional primes consist­ing of BV sentences such as VODIC PLIVA whichmeans "The guide swims" and finally, (4) struc­turally similar derivational primes consisting ofBV sentences such as PLIVAC VODI which means"The swimmer guides" (see Table 7). Primes andtargets were constructed so that the same order ofsentence constituents (viz., B,V,O) was preservedthroughout a set. The advantage of constructingmaterials in this way is that repeated base mor­phemes (i.e., VOD- and PLIV-) do not always ap­pear in the same position in prime and target sen­tences. For example, both VODIC PLIVA andPLIVAC VODI have subject before verb but theordering of base morphemes in these sentencesdiffer. The disadvantage is that by preserving theordering of elements in a pair with similar struc­ture, the effect of changing syntactic role for a par­ticular base morpheme may be lost. Prime andtarget sentences were printed in differentalphabets.

Results indicated that the effect of morphologywas significant (by both FI and F2) for both thelatency and the error measures such that deriva­tions produced less facilitation than did inflec­tions. This finding suggests that the effect of sen-

Morphological Anall/sis in Word Recognition

Table 7. Verification times (and errors in parentheses)for target sentences primed by morphologically andstructurally similar and dissimilar sentences.

target: PLIVAJU vomCI (VS)

Sentence structureSame structure Different structure

tence structure observed in the previous verifica­tion experiment can be attributed to different ef­fects on targets of prime sentences related by in­flection and by derivation. Because inflectionallyrelated primes differed only in number from thetarget whereas derivationally related sentencestransformed the base morpheme of the noun into averb and the base morpheme of the verb into anoun, it was always the case that inflectional sen­tences were semantically more similar to the tar­get than were derivational sentences.

The effect of sentence structure was not repli­cated. The failure to obtain an effect of sentencestructure for either inflectionally or derivationallyrelated primes indicates that facilitation based onrepetition of the initial syllable is not in itself aplausible account of facilitation. The absence of asentence structure effect most likely reflects theway in which structure was manipulated in thepresent experiment. Specifically, the order of(subject, object and verb) constituents ways notalways preserved across structurally relatedprimes and their targets.

Morphological effects occur in sentence contextsas well as in isolated words. In the first two exper­iments in this series, all related targets startedwith the same initial syllable (base morpheme).Therefore, effects of morphological relatednesscould simply be anticipation effects based on repe­tition of the first syllable. This account is notplausible in the third experiment, however. Infact, in that experiment, same structure and dif­ferent structure sentences started with differentinitial syllables (morphemes) and yet there was no

effect of sentence structure of the prime (958 msvs. 949 ms). In sentences related by inflection, theabsence of an effect of structure was not antici­pated and needs to be investigated further.Specifically, in contrast to English, in Serbo­Croatian it is possible to independently manipu­late repetition of sentence constituents (subject,object and verb) and the ordering of those con­stituents.

Even in contexts and tasks that focus onsentence processing it appears that themorphemic constituents of words are analyzed.That is, morphological analysis is not restricted toisolated words in the word recognition task. Inthese experiments, it is evident that activationamong the morphological constituents of primeand target sentences is not necessarily tied totheir syntactic role in a sentence nor to theordering of morphemes. In the next section,associations between morphological processes andother analytic processes are investigated.

4. Associations between phonological andmorphological analysis

The beginning reader provides a windowthrough which to evaluate the relation betweenmorphological and phonological analysis in wordrecognition. In one study (Feldman, Andjelkovic,& Fowler, in preparation), children betweenseven and eight years of age who were nativespeakers of Serbo-Croatian were administeredboth a morphological and a phonological task. Inthe morphological task, children were auditorilypresented with a source word and a sentenceframe and their task was to complete the sentenceby adjusting the morphological affixes on thesource word to make it fit semantically andsyntactically with the sentence frame. Sentenceframes were constructed so that depending on thesource word, either an inflectional or aderivational substitution was required and frameswere paired so that for one source word both aninflectional and a derivational adjustment werenecessary. For example, some children wererequired to fit the source word KUVAR meaning"a cook" (agent in nominative singular) into thesentence frame MAMA MI POMAZE DAwhich means "Mother helps me to _." Thissentence requires the first person singular verbform KUVAM which is related by derivation to thesource word KUVAR. For other children, theinfinitive KUVATI was presented as the sourceword for the same sentence frame. Here thesource word and the response are related byinflection. Still other groups of children viewed the

949(10)

1100(21 )

PLIVA KA vomCu (Va)vODIS PLIVACA (Va)

958(6)

1084(20)

Morphology:

inflection

inflection vomC PLIVA (SV)derivation PLIVAC vom (SV)

derivation

190 Feldman and Andje/kavie

same source words (i.e., KUVAR, KUVATI) ondifferent sentence frames. For example, OVAJRESTORAN IMA DOBROG _ which means"This restaurant has a good_" requires theaccusative singular form of the agent KUVARA.This response is inflectionally related to KUVARand derivationally related to KUVATI. Allsubjects were tested on all four combinations ofsentence frame and source word and acrosssubjects the same base morpheme appeared in allconditions. This design minimized effects due tosentence frame and to the morphologicalcomplexity and the familiarity of the source wordas well as the correct response.

Thirty six sentences were constructed. Eachcontained between four and six words. The targetword was always in final position and varied withrespect to word class. Sentences were read aloudby the experimenter. The source word waspresented both before and after the sentenceframe and all source word-sentence framecombinations required at least one morphologicalsubstitution. Forty children randomly selectedfrom an urban elementary school in Belgrade,Yugoslavia, were tested individually by a nativespeaker of Serbo-Croatian and the subjects'responses were transcribed by that experimenter.Results indicated that inflectional responsestended to be correct more frequently than werederivational responses. Mean errors were 1.0 and3.5 respectively out of a maximum of 18.

Subsequent to the sentence completion task, allsubjects participated in a phoneme deletion task(Rosner & Simon, 1971). Subjects heard wordsand pronounced them aloud without thedesignated phoneme. All words becameorthographically legal but meaningless sequencesafter phoneme deletion. The position of thedeleted letter was balanced across words and,in the source word, it always constituted part ofa cluster. Responses were transcribed by the sameadult native speaker of Serbo-Croatian.Performance on the phoneme deletion taskwas correlated with performance on theinflectional and derivational conditions (summedover sentence frame) of the sentence completiontask.

Results indicated a significant correlationbetween phonemic deletion and each morphemecondition r =.37 for inflections and r =.52 forderivations, respectively. Finally, to each childwas administered verbal and nonverbalintelligence tests and these served as covariantcontrols. Verbal intelligence accounted for 33% ofthe variance and nonverbal intelligence accounted

for 25% of the variance on derivationalmorphology score. The contribution of intelligenceto the inflectional score was not significant butthis outcome may reflect the near perfectperformance and consequent lack of variability onthe inflectional task. Most important, resultsrevealed a significant relationship betweenphonological and derivational performance evenwhen effects of intelligence were partialled out.That is, with controls for verbal and nonverbalintelligence, performance on a phonological taskwas still a significant indicator of performance ona (derivational) morphology task. It accounted fora significant 14% ofthe variance.

Evidently, the ability to explicitly manipulatephonemic segments is associated with the abilityto complete sentences with a syntactically correctform. This relationship is independent of intelli­gence and can be interpreted as evidence of a gen­eral linguistic style of analysis that is not tied toparticular units. This outcome has also been ob­served for learners of English which conveys syn­tactic information through fixed word order incontrast to Serbo-Croatian where word order isrelatively free to vary (Fowler, 1988; 1990).

It is often claimed that metalinguistic skill isthe single most important factor in learning toread (e.g., Tunmer, 1988). While there is amplecompelling evidence that reading an alphabeticorthography requires phoneme awareness,evidence that awareness of linguistic units abovethe level of the phoneme makes an independentcontribution to reading skill is sparse (but seeFowler, 1988; 1990). In the present study, theawareness of morphemes in young readers isassociated with phoneme awareness and therelationship cannot be explained by generalintelligence or by vocabulary knowledge.

5. Morphological effects reflect linguisticanalysis

It is sometimes claimed that three related skillsunderlie the language user's command of mor­phology (Tyler & Nagy, 1989). Primary is an ap­preciation of the internal structure of a word suchthat the presence of a shared component amongmorphological relatives is recognized, either ex­plicitly or implicitly. Experimental evidence formorphological analysis of a word's structure comesprimarily from patterns of facilitation in a prim­ing task in which the same base morpheme is re­peated. Recognition by skilled readers is facili­tated when the same morphological componentsrecur and the basis of this facilitation can be nei­ther semantic nor orthographic in origin.

Morphological Anaillsis in Word Recognition 191

Once words can be analyzed with respect tomorphological components then it is reasonable toask whether skilled readers are sensitive to theconstraints on combinations of morphemes or tothe syntactic implications of appending particularaffixes to a base morpheme. The design of the foilsin the sentence verification and oral reading tasksforced adult subjects to attend to these dimensionsbecause some sentences were composed of illegalmorphological combinations. Although the foilswere not analyzed, morphological analysis wasevident in the facilitation to target sentences com­posed from the same morphological constituentsas their prime sentences. Here, effects of visualsimilarity were eliminated by presenting membersof a pair in contrasting alphabets. Interestingly,equivalent facilitation occurred when, acrossprime and target sentences, a base morphemechanged word class (derivational relatives) andwhen it did not (inflectional relatives) even thoughthey differed with respect to semantic similarity.It has been reported that skills of morphologicalanalysis emerge before combinatory or syntacticskills (Tyler & Nagy, 1989). In the sentencecompletion task, however, seven and eight yearolds were able to produce the appropriate in­flectional and derivation affixes for a variety ofsyntactic contexts. In order to respond accuratelyin this task, subjects had to segment the basemorpheme from its source word as well asgenerate a syntactically appropriate affix.Sometimes this entailed forming a verb from anoun or a noun from a verb and it always requiredthe addition of an inflectional affix. Evidently, themetalinguistic demands of this task allowed thechildren to utilize their knowledge of morphemesand how they combine in particular syntacticcontexts.

In conclusion, evidence for morphologicalanalysis in word recognition is not tied exclusivelyto the repetition priming task although that taskhas allowed a differentiation betweenmorphological analysis and effects of orthographicor semantic structure. Morphological effects arealso evident in a task where constituents of asentence are experimentally manipulated.Although the mechanism of syntactic effects in theverification task is not clear, it is certain thatfacilitatio'n occurs when the constituents of wordsare repeated. This finding is surprising becausethe task fosters analysis at a level more abstractthan the morpheme or the combination ofmorphemes that comprise the words of a sentence.Evidently, readers cannot refrain from engagingin analysis at the level of the morpheme.

It has recently been demonstrated that time torecognize a target word is influenced by its fre­quency relative to the other words in its ortho­graphic neighborhood (Grainger, 1990). Similarlyfor a morphologically simple word, the frequencyof words that are inflectionally and derivationallyrelated to it influences the pattern of reactiontimes in lexical decision (Katz, Rexer, & Lukatela,1991; Nagy, Anderson, Schommer, Scott, &Stallman, 1989; Taft & Forster, 1975) and it is notnecessary that the shared base morpheme of thosewords be explicitly represented in the surface form(Kelliher & Henderson, 1990). These findings sug­gest that recognition of any particular word is in­fluenced by properties of other words that are re­lated along some dimension. This similarity is of­ten captured in terms of organizational propertiesof the lexicon that are distributed rather than tiedto one lexical entry. Perhaps the illusion of ashared orthographic or semantic componentamong morphological relatives has misguided theinvestigation of morphological processing and un­dermined our understanding of the status of themorpheme as. an abstract linguistic unit.

It has been observed that children who are goodat phonological analysis also tend to be good atmorphological analysis and that their performancecannot be attributed to either verbal intelligence(and good vocabulary) or to general intelligence.This finding helps to elucidate the value ofmorphological analysis. It is well established thatgood and poor beginning readers differ in theirability to grasp the phonological structure of aword in a variety of experimental tasks.Phonological analysis helps the beginning readermap unfamiliar written words into a spoken formwhich may be familiar even when the writtenform is not. The essence of this process is anexplicit appreciation of the linguistic componentsof a written word and how they map ontophonemes. It is important because it underlies theability to read unfamiliar words and combinatorialproductivity of the writing system in general.Morphological analysis may serve a similarfunction. Insofar as words can be constructed fromand decomposed into meaningful components andthose components can be recombined into newwords, morphological analysis enhances theproductivity of the reader.

REFERENCESAronoff, M. (1976). Word formation in generative grammar.

Cambridge, MA: MIT Press.Adams, M. J. (1981). What good is orthographic redundancy? In

O. Tzeng & H. Singer (Eds.), Perception of print: Reading researchin experimental psychology. Hillsdale, NJ: Erlbaum.

192 Feldman and Andjelkovic

Bentin, S., & Feldman, L. B. (1990). The contribution ofmorphological and semantic relatedness to repetition primingat short and long lags: Evidence from Hebrew. Quarterly Journalof Experimental Psychology, 42A, 693-711.

Bock, J. K (1986). Meaning, sound, and syntax: Lexical priming insentence production. Journal of Experimental Psychology:Learning, Memory, and Cognition, 12(4), 575-586.

Bock, J. K, & Loebell, H. (1990). Framing sentences. Cognition, 35,1-39.

Dannenbring, G. L., & Briand, K. (1982). Semantic priming andthe word repetition effect in a lexical decision task. CanadianJournal ofPsychology, 36, 435-444.

Emmorey, K D. (1985). Auditory morphological priming in thelexicon. Language and Cognitive Processes, 4, 73-92.

Feldman, L. B. (1992). Bi-alphabetism and the design of a readingmechanism. In D. M. Willows, R. S. Kruk, & E. Corcos (Eds.),Visual processes in reading and reading disabilities. Hillsdale, NJ:Erlbaum.

Feldman, L. B. (1991). Morphological relationships revealedthrough the repetition priming task. In M. Noonan, P.Downing, & S. Lima (Eds.), Linguistics and literacy. (239-254).Amsterdam/Philadelphia: John Benjamins.

Feldman, L. B., & Andjelkovic, D. (1990, November). Morphemicand syntactic facilitation in sentence contexts. Paper presented atthe Psychonomic Society, New Orleans, LA.

Feldman, L. B., & Andjelkovic, D. (1991, November). Morphologicalpriming at long and short lags in visual word recognition. Paperpresented at the Psychonomic Society, San Francisco, CA.

Feldman, L. B., Andjelkovic, D. & Fowler, A. (ms in preparation).Morphological and phonological analysis In beginning readers.

Feldman, L. B., & Fowler, C. A. (1987). The inflected noun systemin Serbo-Croatian: Lexical representation of morphologicalstructure. Memory & Cognition, 15, 1-12.

Feldman, L. B., & Moskovljevic, J. (1987). Repetition priming isnot purely episodic in origin. Journal of Experimental Psychology:Learning, Memory & Cognition, 13,573-581.

Feustel, T. c., Shriffrin, R. M., & Salasoo, A. (1983). Episodic andlexical contributions to the repetition priming effect in wordidentification. Journal of Experimental Psychology: General, 112,309-346.

Forster, K. 1., & Davies, c., Schoknecht, c., & Carter, R. (1987).Masked priming with graphemically related forms: Repetitionor partial activation? Quarterly Journal of ExperimentalPsychology, 39A, 211-251.

Fowler, C. A., Napps, S. E., & Feldman, L. B. (1985). Relationsamong regular and irregular morphologically related words inthe lexicon as revealed by repetition priming. Memory &Cognition, 13, 241-255.

Fowler, A. (1988). Grammaticality judgments and reading skill ingrade 2. Annals of DysleXia, 38, 73-84.

Fowler, A. (1990). Factors contributing to performance onphonological awareness tasks. Haskins Laboratories Status Reporton Speech Research, SR-103/104, 137-152.

Grainger, J. (1990). Word frequency and lexical neighborhoodeffects in lexical decision and naming. Memory & Language, 29,28-244.

Grainger, J., Cole, P., & Segui, J. (1991). Masked morphologicalpriming in visual word recognition. Memory & Language, 30,37()"'384.

Hanson, V. L., & Feldman, L. B. (1989). Language specificity inlexical organization: Evidence from deaf signers' lexicalorganization of ASL and English. Memory & Cognition, 17, 292­301.

Hanson, V. L., & Wilkenfeld, D. (1985). Morphophonology andlexical organization in deaf readers. Language and Speech, 28,269-280.

Henderson, L. (1985). Toward a psychology of morphemes. In A.W. Ellis (Ed.), Progress in the psychology of language (pp. 15-72).London: Erlbaum.

Henderson, L. (1989). On the mental representation ofmorphology. and its diagnosis by measures of visual accessspeed. In W. Marslen-Wilson (Ed.), Lexical representation andprocess (pp. 357-391). Cambridge, MA: MIT Press.

Henderson, L., Wallis, J., & Knight, D. (1984). Morphemicstructure and lexical access. In H. Bouma & D. Bouwhuis(Eds.), Attention and performance X (pp. 211-224). Hillsdale, NJ:Erlbaum.

Jackendoff, R. (1975). Morphological and semantic regularities inthe lexicon. Language, 51, 639-671.

Katz, L., Rexer, K, & LUkatela, G. (1991). The processing ofinflected words. Psychological Research, 53, 25-31.

Kelliher, S., & Henderson, L. (1990). Morphologically basedfrequency effects in the recognition of irregularly inflectedverbs. British Journal of Psychology, 81, 527-539.

Lakoff, G. (1971). Syntactic irregularity. New York: Rinehart &Winston.

Laudanna, A., Badecker, W., & Caramazza, A. (1989). Priminghomographic sterns. Journal ofMemory & Language, 28, 531-546.

Monsell, S. (1985). Repetition and the lexicon. In A. W. Ellis (Ed.),Progress in the psychology of language (pp.147-195). London:Erlbaum.

Nagy, W., Anderson, R c., Schommer, M., Scott, J. A., &Stallman, A. C. (1989). Morphological families in the internallexicon. Reading Research Quarterly, 24, 263-282.

Napps, S. E. (1989). Morphemic relationships in the lexicon: Arethey distinct from semantic and formal relationships? Memory& Cognition, 17, 729-739.

Napps, S. E., & Fowler, C. A. (1987). Formal relationships amongwords and the organization of the mental lexicon. Journal ofPsycholinguistic Research, 16,257-272.

Rapp, B. C. (1992). The nature of sublexical orthographicorganization: The bigram trough hypothesis examined. JournalofMemory and Language, 31, 33-53.

Ratcliff, R, Hockley, W., & McKoon, G. (1985). Components ofactivation: Repetition and priming effects in lexical decisionand recognition. Journal of Experimental Psychology: General, 114,435-450.

Rosner, J., & Simon, D. P. (1971). The auditory analysis test: Aninitial report. Journal ofLearning Disabilities, 4, 384-392.

Seidenberg, M. (1987). Sublexical structures in visual wordrecognition: Access units or orthographic redundancy? In M.Coltheart (Ed.), Attention and performance XII: The psychology ofreading. (pp. 245-263). Hillsdale, NJ: Erlbaum.

Segui, J., & Grainger, J. (1990). Priming word reCOgnition withorthographic neighbors:. Effects of relative prime-targetfrequency. Journal of Experimental Psychology: Human Perceptionand Performance, 16, $-76.

Stanners, R. F., Neiser, J. J., Hernon, W. P., & Hall, R (1979).Memory representation for morphologically related words.Journal of Verbal Learning and Verbal Behavior, 18,399-412.

Tunmer, W. (1988). Metalinguistic abilities and beginningreading. Reading Research Quarterly, 23, 134-158.

Tyler, A., & Nagy, W. (1989). The acquisition of English deriva­tional morphology. Journal ofMemory & Language, 28, 649-667.

FOOTNOTES'In L. Katz & R. Frost (Eds.), Orthography, phonology, morphology,

and meaning (pp. 343-360). Amsterdam: Elsevier SciencePublishers (1992).

t Also State University of New York at Albany.tUniversity of Belgrade.