Hilskins LAbonItorYs Sutus Report on Speech Resetlrch1992, SR-109/110, 163-178

Morphological Analysis of Disrupted Morphemes:Evidence from Hebrew

Laurie Beth Feldmant and Shlomo Bentintt

In concatenated languages such as English, the morphemes of a word are linked linearly sothat words formed from the same base morpheme also resemble each other along orthographicdimensions. In Hebrew, by contrast, the morphemes of a word can be but are not generallyconcatenated. Instead, a pattern of vowels is infixed between the consonants of the rootmorpheme. Consequently, the shared portion of morphologically-related words in Hebrew isnot always an orthographic unit. In a series of three experiments using the repetition primingtask with visually-presented Hebrew materials, primes that were formed from the same basemorpheme and were morphologically-related to a target facilitated target recognition.Moreover, morphologically-related prime and target pairs that contained a disruption to theshared orthographic pattern showed the same pattern of facilitation as did nondisrupted pairs.That is, there was no effect of disrupting, over successive prime and target presentations thesequence of letters that constitutes the base morpheme or root. In addition, facilitation wassimilar across derivational, inflectional and identical primes. The conclusion of the presentstudy is that morphological effects in word recognition are distinct from effects of sharedstructure.

The internal structure of a word plays a key rolein its recognition. Whereas much work on visualword recognition has focused on phonology, morerecent efforts have focused on aspects ofmorphology. One experimental task that issensitive to the morphological components ofwords is repetition priming. Significant facilita­tion among visually-presented morphologicallyrelated words in the repetition priming variant ofthe lexical decision task is well documented(Stanners, Neiser, Hernon, & Hall 1979).Generally, responses to targets that are formedaround the same base morpheme as their(morphologically-related) primes are faster andmore accurate than to targets following unrelatedprimes. Sometimes, the facilitation with

The research reported here was conducted at HebrewUniversity in Jerusalem and was supported by funds fromNational Institute Of Child Health and Development GrantHD.01994 to Haskins Laboratories and from the IsraelFoundation Trustees to Shlomo Bentin.

163

morphological relatives as primes is equivalent tothe effect of an identical repetition of the target.Sometimes, it is numerically reduced relative toidentical repetitions but is still statisticallyreliable (Fowler, Napps, & Feldman, 1985).Effects of morphological relatedness with visuallypresented materials in the lexical decision taskhave been found across a variety of languages in­cluding Serbo-Croatian (Feldman & Fowler, 1987),English (Feldman, 1991a; Fowler et at, 1985) andHebrew (Bentin & Feldman, 1990) as well asAmerican Sign Language (Hanson & Feldman,1989; see also Emmorey, 1989). At lags largerthan zero or if more than a few seconds separatethe second presentation from the first, the patternof facilitation due to morphological relatedness isdistinct from the pattern due to semantic associa­tion (Bentin & Feldman, 1990; Dannenbring &Briand, 1982; Henderson, Wallis & Knight, 1984;Napps, 1989). At average lags of 10 items, ortho­graphic similarity of morphologically unrelatedprime and target (e.g., pairs such as DIET andDIE) produces neither facilitation nor inhibition(Bentin 1989; Feldman & Moskovljevic, 1987;

164 Feldmlln and Bentin

Hanson &Wilkenfeld, 1985; Napps & Fowler,1987). In short, the repetition priming procedureis a viable tool for studying how the morphologicalrelation among words is represented in the lexiconand how that relation distinguishes itself fromother types of similarity.

An examination of morphologically complexwords across languages reveals two basic linguis­tic principles by which such words are constructed.In one, discrete morphemic constituents arelinked linearly. There is a base morpheme towhich other elements are appended so as to form asequence. This principle defines a concatenativemorphology, of the kind characteristic of Englishand Serbo-Croatian, for example. In languageswith a concatenative morphology, suffixes andprefixes are regularly appended to the base mor­pheme in a manner that preserves its phonologicaland orthographic structure. According to the otherprinciple, morphemic units are not just appendedto a base form, but also modify its internal struc­ture. This principle defines a nonconcatenativemorphology of the kind found in Hebrew, for ex­ample (McCarthy, 1981).

In the repetition priming studies of morphologi­cal processing conducted with visually-presentedEnglish and Serbo-Croatian materials describedabove, primes and targets were typically con­structed around the same base morpheme. andonly differed with respect to affix. As a result,among morphological relatives, the base mor­pheme remained intact and unchanged.Exceptions consist of studies that explored effectsof changed spelling and/or pronunciation amongmorphologically related pairs (e.g., HEAL andHEALTH or SLEEP and SLEPT) at long lags(Fowler et al., 1985; Stanners et al., 1979; see alsoKempley & Morton, 1982), studies that examinedspelling and sound changes among morphologicalrelatives at varying short lags and SOA's (Napps& Fowler, 1987) and a study with German mate­rials that examined umlaut changes (Schriefers,Friederici, & Graetz, 1992). Even in those studies,however, the changes introduced to the base mor­pheme were relatively minor (e.g., consisting of avowel or a vowel plus consonant change) as com­pared to the portion that was preserved. Thestructure of materials in those studies reflects ageneral principle of construction for languageswith a concatenative morphology. That is, whenmorphemes are concatenated it is almost alwaysthe case that the phonological and orthographicstructure of the base morpheme will be preservedamong regular morphological relatives (but seeKelliher & Henderson, 1990). A morpheme tends

to be a sequence of consonants and vowels thatforms a syllable (or several) and concatenativeword formation processes do not disrupt the co­herence of the morpheme. The implication of thisis that in concatenated languages such as English,morphological relatives will tend to have se­quences of letters in common. As applied to theconstruction of materials in the typical repetitionpriming task where morphologically-related pairsare formed by adding a suffix, the initial portion ofprimes and targets will tend to be identical.

Nonconcatenative formation processes are lesslikely to preserve the integrity of the basemorpheme. The base morpheme in Hebrew is anabstract form which is called the "root" and iscomprised of a string of three (or four) consonants.The root is not a complete phonological unit as itincludes no vowels. Superimposed on the root isthe "word pattern" which consists primarily ofvowels. The root together with a word patternconstitute the word. Some word patterns consistexclusively of vowels and typically, the vowels areinfixed between the consonants of the root. Otherword patterns include a consonant prefIX (e.g., Mplus vowel) or a suffix (e.g., vowel plus T) as well.Both the word pattern as well as the root areproductive and convey morphological andsemantic information (Oman, 1971). For example,the root SH·M·N can take many word patternsincluding -e-e- to form the noun /SemenJ (whichmeans "oil-), and -a-e- to form the adjective/Samen/ (which means "fat-). Similarly, the root Z·M·R can take many word patterns including -a-a,-e-e-, and -i-e-. Note that roots such as Z.M·R andSH·M·N are productive in that they generateseveral words in the semantic fields related tosinging and oil respectively. Similarly, the wordpatterns are productive and tend to modify theroot in systematic ways (Berman, 1978). Forexample, the -a-a- word pattern tends to denotean agent, the -e-e- pattern an object, and the -i-e­the past tense of an active verb in the third personsingular. Thus, in Hebrew, /zamar/ meaning "asinger," /zemer/ meaning "a song" and tzimer/meaning "he sang" are all morphologically-relatedbecause they share the Z·M·R root and are all bi­morphemic because they include a word patternas well as a root.

It is useful to point out that when different rootsaccept the same word pattern, the semanticinformation carried by that word pattern is notfully consistent. Specifically, although, the wordpattern -a-a- often denotes an agent, it is alsosometimes used to denote the past tense singularform of active verbs as well as some adjective

Morphological Analysis ofDisrupted Morphemes: Evidence from Hebrew 165

fonns. Compare, for example, the contribution ofthe -a- a- pattern to the root Z-M-R (/zamar /meaning "singer") with its effect on the root L-V-N(Ilavanl meaning "white"). Similarly, the semanticcontribution of the root is not consistent in anysimple sense over all morphologically-relateditems.

The principle of building words in Hebrew, incontrast to that of languages such as English andSerbo-Croatian, dictates that the phonological andorthographic similarity of morphologically-relatedwords in Hebrew will be spread over severalsyllables. Root morphemes consist of a sequence ofconsonants and the requisite vowels for aparticular word pattern are infixed between theconsonants. Consequently, the root morphemeconstitutes neither an orthographically nor aphonologically coherent whole. Rather thanfonning continuous units, morphemes tend to bedisrupted and distributed over several syllables.

Alternative accounts of morphologicaleffects

Accounts of morphological effects in wordrecognition often minimize the role of purelylinguistic variables such as the morpheme andrely on orthographic and phonological patterningof letter units or on semantic similarity inconjunction with shared orthographic andphonological structure. For example, Seidenberg(1987) suggested that patterns of high and lowprobability of transition among sequences ofletters could account for (syllabic or) morphol­ogical patterning because transitionalprobabilities of letter sequences that straddle a(syllabic or) morphological boundary tend to below (bigram troughs) relative to probabilities ofsequences internal to a unit. In an illusoryconjunction paradigm, subjects who tended tomisidentify the color of the target letter were morelikely to assign the color of another letter fromwithin the same morphological unit than from anadjacent but different unit. Although this resultprovides support for orthographic (specifically,bigram) structure in a particular task, it does notnegate the influence of morphology in wordrecognition. Recently, in fact, morphologicaleffects have been demonstrated in a lexicaldecision task where color boundaries within aword were either consistent or inconsistent withmorphological boundaries (Rapp, 1992). Moreover,morphological boundary effects were evident bothin words with bigrams troughs at the boundaryand in words without troughs. Similar effects havealso been reported for compound words

(Prinzmetal, Hoffman, & Vest, 1991). Whether ornot orthographic factors in morphologicalprocessing prove to be relevant for languages withconcatenative morphologies such as English, it isdifficult to see how they could be adapted easily tononconcatenated languages such as Hebrewbecause morphemes are not always coherentunits. In sum, the tendency to interpretmorphological effects as orthographic patterningmakes it essential to examine orthographicinfluences on morphological processing in alanguage in which the morpheme is not always anorthographic entity.

The emphasis on orthographic patterning is alsoevident in morphological parsing models in whichthe affixes of a morphologically complex word arefirst eliminated and then the remaining portion ofthe letter string is matched to candidate entries ina lexicon (e.g., Taft & Forster, 1975). Althoughaffix parsing models may be plausible inlanguages such as English in which the repertoireof morphological affixes is relatively limited, theirpracticality is severely compromised in languageswith differing morphological structures (cf.Henderson, 1989). In Turkish, for example,sequences of morphological affixes may beappended to one root and the fonn of those affixesmay vary due to phonological factors. Moreover,some aft"txes may be applied more than once.Consequently, a process of Suff'lX stripping withsubsequent analysis of the remainder may have toundergo many iterations before the root can besuccessfully identified. It has been proposed thatfor Turkish, priority in morphological analysis of aword goes to the root and only then is its sequenceof affixes identified. That process starts at the rootand proceeds from left to right (Hankamer, 1989).In contrast, morphological parsing in Hebrewposes special problems because the root morphemeconstitutes neither a coherent phonological nororthographic unit and morphological fonnation isless systematic.

In a study of morphological analysis usingrepetition priming with Hebrew materials (Bentin& Feldman, 1990), patterns of facilitation forprime-target pairs that were related by semanticassociation and by a shared (morphological) rootwere compared. The study exploited the fact thatalthough words that are constructed around thesame root are, by definition, morphologically­related, the semantic relation amongmorphologically-related forms in Hebrew mayvary dramatically. All of the morphologicalrelations of prime and target pairs werederivational in nature. As a consequence, the

166 Feldmlln and Bentin

meaning of a derived form was not alwayspredictable in any simple way from a semanticanalysis of its component morphemes (see Aronoff,1976). Facilitation due to morphologicalrelatedness was evident at lags that averaged tenintervening items. Moreover, facilitation wasequivalent for semantically-close (e.g., kitchen­cook) and semantically-distant (e.g., slaughter­cook) prime-target relatives. That is, themagnitude of facilitation to the target meaning"cook" was equivalent following primes meaning"kitchen" and "slaughter." These findings are, infact, consistent with the claim based on Englishmaterials that at long lags semantic overlapbetween prime and target does not influence themagnitude of repetition priming (Feldman,1991a). In summary, when all related words werederivational in nature, and an average of tenitems intervened between prime and target,facilitation due to morphological relatedness inthe repetition priming task was not sensitive tothe semantic similarity of prime and target. Thisoutcome suggests that morphological analysis isnot based on the semantic overlap ofmorphological relatives.

Covariants of morphological structure inHebrew

With respect to orthographic structure of wordsin Hebrew, it is important to note that mostvowels are represented by optional diacriticsplaced beneath, above or within the precedingconsonant although some vowels are representedby letters. Because words are conventionallywritten without vowel diacritics, morphologically­complex words that share a root morpheme butdiffer with respect to word pattern will tend to beorthographically but not phonologicallyindistinguishable. For example, the words/gever/and /gavar/ are both written .,:1 (Note that incontrast to English and to phonemic notation,Hebrew is read from right to left) These words aremorphologically related and mean "man" and"overcome," respectively. Because the wordpattern is composed exclUSively of vowels andbecause the /el and /a! vowels are represented byoptional diacritics, these two words have the sameorthographic form as conventionally written. Ofcourse, although both words have phonologicalforms that are created around the G·V·R root,their phonological forms differ because of theinfixed vowels. By contrast, when vowels arewritten and particularly when one of them isrepresented by a letter, then the orthographicpattern of the root morpheme, like its phonological

pattern is no longer a coherent unit. For example,the sequence "Q,?Q is read ImiJrnar/, meaning"guard" whereas the sequence ";1~, is read/l~mer/, which means "guardian". These wordsare morphologically related as they share the rootr-m-v (SH·M·R) They differ, with respect tophonological form, as well as orthographic form,however, because in one case the letters for the /~/

vowel of the word pattern is inflXed between theconsonants of the root. In the present study, weuse patterns of facilitation for morphologicallycomplex words in the repetition priming task toask whether the morphological processing ofdisrupted roots as typically occurs in Hebrew issimilar to the processing of continuous roots astypically occurs in concatenated languages.

Linguists distinguish between two types ofmorphologically complex words. Words that sharea base morpheme but differ with respect toinflectional affixes are generally considered to beforms of the same word (e.g., CALCULATE,CALCULATED). Words that share a basemorpheme but differ with respect to derivationalaffixes are generally considered to be differentwords (e.g., CALCULATE, CALCULATOR,CALCULATION). As a secondary objective in thepresent study, we use patterns of facilitation toask whether inflectional and derivationalformations are likely to involve distinct types ofrepresentations and/or processing.

Experimental evidence for this linguisticdifference has been difficult to obtain in English.One possible reason for the failure to find evidencefor the linguistic distinction between inflectionaland derivational formations is that the similarityof orthographic form cannot be equated inEnglish. Specifically, because inflectional relativesand derivational relatives tend to differ withrespect to length of affix (or because thetransitional probability from the final letter of thebase morpheme to the initial letter of the affixdiffers for inflectional and derivational aff"lXes),these comparisons are not appropriate.

In Hebrew, by contrast, it is possible to identifypairs of words that are, respectively,inflectionally- and derivationally-related and areequated with respect to orthographic andphonological similarity to that target. Bydefinition, all such words are morphologicallyrelated to each other because they are constructedaround the same root morpheme. Words in a pairdiffer with respect to the word pattern butinflectionally-related and derivationally-relatedword patterns can be matched with respect topresence (and letter length) of prefixes and/or

Morphological Analysis of Disrupted Morphemes: Evidencefrom Hebrew 167

suffixes. In this way, the structural similarity to atarget of inflectional and derivational relativescan be matched so that types of morphologicalformations can be compared.

To summarize, the primary goal of the presentstudy was to examine the role of orthographicpatterning in morphological analysis. Accordingly,using morphologically nonconcatenated Hebrewmaterials, the orthographic integrity of the basemorpheme across morphological relatives was sys­tematically manipulated in the repetition primingtask. Sometimes prime and target presentationspreserved the same orthographic form of the rootmorpheme and sometimes they did not. Asecondary goal of the present study was tocompare facilitation by inflectional andderivational relatives. Primes and targets shareda common root and primes were eitherinflectionally- or derivationally-related oridentical to the target. Lexical decision latency tothe target was compared followingmorphologically-related and identical primes. Aseries of three experiments was conducted in anattempt to uncover the contribution to wordrecognition of orthographic similarity over andabove that of morphological relatedness.

EXPERIMENT 1Across languages, a variety of mechanisms for

forming words exist, the most common being theaddition or affixation of an element to a basemorpheme (Matthews, 1974). AffIXation includesthree processes, defined by the position relative tothe base morpheme, where addition occurs. Theseinclude prefixation, suffixation, and infixation inpositions initial, final and internal to the basemorpheme. Prefixation and suffixation entail thelinear concatenation of elements, whereasinfixation is nonconcatenative insofar as theintegrity of the base morpheme is disrupted. Asdescribed above, the characteristic morphologicalprocess of Semitic languages, such as Hebrew,relies on a skeleton of consonants into which apattern of vowels is infixed (although a prefIX orsuffix may also be appended). The morphologicalsystem of Semitic languages is distinguished forits productivity, the manner in which semanticmodification of the root occurs among complexforms that share a root, and for thenonconcatenativity of morphemes (Berman, 1978).

As noted above, the orthographic integrity of thebase morpheme is generally maintained inEnglish and in Serbo-Croatian but not alwayspreserved in written Hebrew. For processes of

infixation, morphological changes typically entailappending different word patterns to a root, wherethe word patterns specify the requisite vowels of aword. When represented by a letter, vowels in theword pattern necessarily disrupt the sequence ofconsonants that co~prise the Loot. Consider, forexample, the words "0) ~d "Q' and comparethem with the target word "01'. The target is thepresent tense of the verb "to fall", in the thirdperson singular (pronounced /n:>fe1/). The firstform is inflectionally-related to the target and ispronounced /nafal/; it is the past tense of the sameverb in the same person. The second form isderivationally-related to the target and ispronounced !n£fe1l which means "a dropout". Bydefinition, all three forms are morphologicalrelated because they share the same root' -C-'(N·F·L). Note, however, that in the target word,the root morpheme is not continuous. It isdisrupted by the vowel 0 /:>/, which is part of theword pattern. Contrast this pattern with that forthe words "T~L' and"'T;t' as compared with thetarget C"~x},: The target is pronounced/avadirnl, meaning "slaves". The first word isinflectionally-related to the target, is pronouncedleved! and is the singular form "slave." Thesecond word is derivationally-related to the target,is pronounced lavad/, which is the past tense,third person singular of the verb "to work." Notethat in this case, the orthographic root "T-~-L'

remains intact in all related forms. Theorthographic similarity (due to preservation of theorthographic pattern for the root) of themorphological relatives depicted in the latterexample is characteristic of all regularly-relatedpairs in English. In the present experiment withHebrew materials, the pattern of facilitation dueto morphological relatedness of prime-target pairswas compared when the orthographic form of theshared root was di¥:Upted over prime and targetpresentations (e.g., ,,-C-,) and when it was intact(e.g., "T-:-L').

It has already been demonstrated that inHebrew, facilitation in the repetition priming taskis sensitive to derivational relatedness of primeand target (Bentin & Feldman, 1990). Ifinflectional and derivational formations inHebrew are similarly represented in the lexiconthen it is anticipated that the magnitude offacilitation in the lexical decision repetitionpriming task will not vary with type ofmorphological relation, and a comparison ofinflectional and derivational primes is included inthe present investigation. It is anticipated that if

168 Fe1drrum and Bentin

orthographic similarity of prime to target isindependent of morphological relatedness then thepattern of facilitation for roots that are disruptedand roots that are not disrupted will not differ.

MethodsSubjects. Forty-eight first year students from

the Department of Psychology at HebrewUniversity participated in Experiment 1. All werenative speakers of Hebrew. All had vision thatwas normal or conected-to-normal and had priorexperience in reaction-time studies. None hadparticipated in other experiments in the presentstudy.

Stimulus materials. Forty-eight Hebrew wordtriplets were constructed. Each included threeforms: a target word, a word that wasinflectionally-related to it and a word that wasderivationally-related to it. All members of atriplet were constructed from the same rootmorpheme but they differed with respect to wordpattern. The orthographic and phonemic overlapof morphologically-related words to their targetswas systematically manipulated. Targetsconsisted of twenty-four verbs in present tense,third person singular and twenty-four pluralnouns. For verb targets, the inflected forms werepast tense formations (third person singular), andthe derived forms were nouns in singular case. Inthe verb set, the roots were orthographicallycontinuous in both inflected and derived forms,but the roots were disrupted in the target by theinfixation of a letter vowel. For the noun targets,the inflected forms were the same nouns insingular and the derived forms were verbs in pasttense (third person singular) In this set, the rootswere orthographically continuous in targets aswell as related forms.

Four types of words preceded each target acrossexperimental lists. Words inflectionally- andderivationally-related to the target, an identicalrepetition of the target and an (orthographically,phonologically and semantically) unrelated wordserved as primes The orthographic similarity ofthe derived and inflected primes to their targetwas matched within each triplet (All word tripletsand their English translations are listed inAppendix A). The unrelated words had the samemorphological structure (word pattern) as did therelated words (for other targets) although theynecessarily had different root morphemes.

Ninety-six pseudowords were constructed bycombining meaningless three-consonant rootmorpheme with real word patterns. Root

morphemes in nonwords were not repeated oversuccessive trials so as to enhance the orthographicsalience of the words.

Four test orders were assembled. Each list wascomprised of 96 words and 96 nonwords. All itemswere presented with their vowels. The 96 wordsconsisted of the 48 targets and their 48 primes.Twelve targets were preceded by identical repeti­tions, 12 targets were preceded by derivationally­related primes, 12 were preceded by inflectionally­related primes and 12 targets were preceded bymorphologically, orthographically and semanti­cally unrelated word primes. The lag betweenprime and target varied between 7 to 13 itemswith an average of 10. The serial position of alltarget words and pseudowords was identicalacross test orders. The primes were rotated amongthe four lists, so that within a list each type ofprime was equally represented and, across lists,each target was preceded once by each of the fourtypes ofprimes.

Procedure. Twelve subjects were randomlyassigned to each of the stimuli lists. Thus, the fourprime types were compared within subjects acrossall 48 targets and within stimuli across all 48subjects. Speed and accuracy were equallyemphasized in the instructions.

The stimuli were presented approximately 80em from the subject, at the center of a Macintoshmonochromatic screen. Each item was exposeduntil the subject responded or for 2000 ms,whichever came first. The interval between onsetof successive stimuli was 2500 ms.

The dominant hand was used for word re­sponses and the nondominant hand was used fornonword responses. Latencies were measuredfrom stimulus onset, to the nearest millisecond us­ing a special software algorithm1 and enors wereautomatically registered. Following the instruc­tions, a practice list comprised of 24 items (twoidentity, two inflectional and two derivationalprime-target pairs as well as 12 pseudowords) waspresented. After a short pause, the experimentallist followed in one block. The complete experi­mental session lasted about 20 minutes.

Results and DiscussionLexical decision reactions times more extreme

than two SO's from the mean for subjects and foritems in each condition were excluded from allanalyses. Fewer than 2% of all responses wereeliminated by these constraints. Mean lexicaldecision latencies and enors in Experiment 1 aresummarized in Table 1.

Morphologiad ATUllysis ofDisrupted Morphemes: E'Oidenc:e from Hebrew 169

Table 1. Mean lexical decision time and percent errorsfor targets following morphologically-related andunrelated primes words in Experiment 1 (SEm inparentheses).

PRIMElYPE

Unrelated Identity Inflection Derivation

RT 769 701 700 710(14) (14) (14) (13)

Errors 0.5 0.5 0.4 0.5(0.3) (0.2) (0.1) (0.2)

The statistical reliability of the repetitionpriming effect was tested in each task by ANOVAwith repeated measures across subjects (F1) andacross stimuli (F2). In the lexical decision task theeffect of prime type was significant F 1(3,141)=20.72, MSe=2279, p<.0001, and F 2(3,141)=18.67, MSe=2704, p<.OOOl. Tukey-A posthoc comparisons revealed that whereas all primetypes significantly facilitated lexical decisionrelative to the unrelated condition (p<.01), themagnitude of the effect did not differ from onetype of prime to another. In particular, it wasinteresting that facilitation with identity primeswas not significantly larger than with inflectionalor derivational primes.

The factors of Target continuity (disrupted, con­tinuous), and Prime type (unrelated, identity, in­flectional, derivational) were examined in ananalysis of variance. This analysis revealed thatthe effect of target continuity was not reliable F2(1,46)=0.43, MSe=17429, p>.50. 1 The effect ofPrime type was significant but, as suggested bythe absence of a reliable interaction betweenPrime type and continuity F2 (3,138)=0.44,MSe=2737, p>.50, facilitation from morphologi­cally-related primes to orthographically disruptedtarget did not differ from facilitation to ortho­graphically continuous targets.2

Table 2. Mean lexical decision latency in milliseconds(and SEm) for target words with disrupted andcontinuous roots following primes in the four primingconditions ofExperiment 1.

PRIMElYPE

Unrelated Identity Inflection Derivation

Disrupted 759 699 704 700(19) (14) (14) (16)

Continuous 785 704 715 717(21) (13) (17) (15)

The error rate on words was very low and didnot differ as a function of prime type Fl (3,141)=0.24, MSe=0.4, p<.80. Due to the design of theexperiment, facilitation due to repetition ofpseudowords could not be analyzed.

Experiment 1 had three important outcomes: a)The magnitude of the facilitation in lexical deci­sion was similar for prime-target pairs whosestructure preserved the orthographic continuity ofthe root and for those where the continuity of theroot was disrupted by infixing an additional letter.It was the case that all the disrupted roots wereembedded in verb targets whereas all the continu­ous roots were in nouns and that the derivation­ally-related primes (but not the inflectionally-re­lated primes) always introduced a change in wordclass between prime and target. Nevertheless,statistically nonsignificant and numerically smalldifferences between facilitation by inflectional andby derivational relatives were obtained. This out­come suggests that the morphological repetitioneffect is sensitive neither to similarity of ortho­graphic form between the prime and the targetnor to the similarity of word class. b) Significantfacilitation for inflectionally- and derivationally­related as well as for identity primes was observedand provided further evidence for morphologicalanalysis in Hebrew. However, the magnitude ofthe facilitation in lexical decision was not signifi­cantly greater for prime-target pairs related by in­flection than for pairs related by derivation. Thus,facilitation by repetition priming was not sensitiveto the type of morphological relation. c) Finallyfacilitation due to morphological relatedness i~ .Hebrew cannot be attributed to repetition of aninitial syllable. Although the initial consonant wasalways unchanged in prime and target, the follow­ing vowel did vary. Initial consonant and voweloverlap of prime and target was greater for inflec­tions than for derivations for the nondisruptedtargets whereas the vowel never overlapped forthe disrupted targets. Nevertheless, the patternwas similar for both.

In conclusion, the results of Experiment 1 repli­cate effects of morphological relatedness in therepetition priming task when the orthographicintegrity of the base morpheme is preserved overprime and target and extends the outcome tocases where the continuity of the root morphemeis disrupted. In addition, it shows that the ten­dency for enhanced semantic overlap of inflec­tionally-related prime-target pairs relative toderivationally-related pairs contributes nothing tothe pattern of facilitation. Collectively, these re­sults provide no behavioral evidence for a linguis-

170 Feld""1n and Bentin

tic distinction between morphological types.Moreover, it suggests that effects due to morpho­logical relatedness are not easily interpreted as acomposite of orthographic and semantic similarity.

EXPERIMENT 2The results of the previous experiment revealed

morphological analysis in the lexical decision task.Evidently, subjects were sensitive to repetitions ofa sequence of consonants that comprises a rootmorpheme whether or not they form an ortho­graphic unit. Importantly, inflectional relation­ships and derivational relationships produced thesame pattern of facilitation. We assume that thisoutcome can be interpreted as a failure to findevidence for a psychological distinction betweenmorphological types in Hebrew. Aspects of stimu­lus construction in Experiment 1 permit an alter­native account, however.

In Experiment 1, all nonwords were constructedfrom a meaningless string of consonants combinedwith a real word pattern and all words(necessarily) consisted of a meaningful rootcombined with an appropriate word pattern.Therefore, in order to perform the lexical decisiontask successfully, it was not logically necessary forsubjects to attend to the whole word: an analysisof the root would have been sufficient.Consequently, it is possible that the failure toobserve a difference between morphologically­related primes with inflectional and derivationalword patterns reflected the tendency of subjects toignore perceptually nonsalient vowel informationin this experimental setting. It was essential toshow that subjects were, in fact, sensitive to theword patterns that create the distinction betweeninflectional and derivational formations and thiswas the intent of the second experiment.

In Experiment 2, the informativeness of wordpattern information was enhanced by constructingpseudowords along a different principle. Here,pseudowords consisted of a real root and a realword pattern in an illegal combination. The wordsconsisted of the same items as in the previousexperiment. The differentiation between word andpseudowords therefore required the subject toprocess the word pattern as well as the root. As inthe previous experiment, word targets werepreceded by identity, unrelated, inflectionally- andderivationally-related primes.

MethodSubjects. Forty-eight first year students from

the Department of Psychology at Hebrew

University participated in Experiment 2. As in theprevious experiment, all were native speakers ofHebrew. All had vision that was normal or Cf •

rected-to-normal and all had prior experience il

reaction-time studies although none participatedin other experiments in the present study.

Stimulus materials. The words used in thepresent experiment were identical to those used inExperiment 1. There were 48 sets, each comprisedof a target, an unrelated word, a derivationally­related word, and an inflectionally-related word.Half of the targets contained orthographicall~

continuous roots and half contained roots thatwere disrupted by the inf'lXation of the vowel /~/

which is represented by the letter O. Bothinflectional and derivational primes alwaysincluded the full root morpheme in a continuousform, and primes were matched foT' orthographicsimilarity with the target.

The 96 pseudowords were conSt. ucted usingother productive roots that exist in the language.All roots were combined with legal word patternssuch that the particular combination of root andword pattern was meaningless. For example, theroot ~-~-1' (A-V-D) was combined with the wordpattern -~-a-ut in order to form the phonologicallylegal but meaningless structure n~TQ;Vwhichis pronounced /::>vdanut/. This manipulation wasintroduced so as to promote morphologicalanalysis of all letter strings.

The four test orders created for Experiment 1were modified so that a new set of nonwords wassubstituted for the old set. In all other respectsthe materials were identical to those of theprevious experiment.

Procedure. Subjects were instructed to make alexical decision judgment. The procedure as wellas the word stimuli were identical those that ofExperiment 1 except that the timing software wasmeasured from s hardware device that eliminatedthe constant that had been added to each latencyin the previous experiment.

Results & DiscussionMean lexical decision latencies were calculated

in each condition, across subjects and acrossstimuli. Errors and extreme reaction times wereeliminated according to the constraints describedfor Experiment 1. Mean reaction times and errorsfor each conditions are present:d in Table 3.

The comparison of the latencies of lexicaldecisions to target words in the differentconditions was based on ANOVA using subjects(F1) and stimuli (F2) as random factors.

Morplwlogicld Analysis ofDisrupted Morphemes: Evidence from Hebrew 171

Table 3. Mean lexical decision times in millisecondsand percentage of errors for morphologically-relatedand unrelated target words in Experiment 2 (SEm inparentheses).

PRIMElYPE

Unrelated identity Inflection Derivation

RT 628 571 576 578(12) (6) (6) (10)

Errors 1.9 1.4 1.9 1.5(0.4) (0.3) (0) (0.4)

morphological structure of the items in order toperform the lexical decision task, facilitation inmorphological repetition priming was notsensitive to a) type of morphological relation norto b) preservation (or disruption) of anorthographic pattern for the morpheme acrossprime and target words.

Table 4. Mean lexical decision latency in milliseconds(and SEm) for target words following primes withdisrupted and continuous roots in the four primingconditions ofExperiment 2.

PRIMElYPE

Unrelated Identity Inflection Derivation

Plausible accounts of facilitation .in therepetition priming task have identified response­related (episodic) as well as lexical influences (e.g.,Bentin & Feldman, 1990; Bentin & Moscovitch,1988; Bentin & Peled, 1990; Forster & Davis,1984; Monsell, 1985). One account of the presentresults places the locus of facilitation at the levelof the root morpheme that is repeated in bothinflectional and derivational pairs. Perhapsrepetition serves to facilitate the identification ofan orthographically and semantically abstractroot within the composite root plus word patternthat constitutes a word. Conjointly, facilitationmay reflect that was present in our previousexperiments. It was the case that lexical decisionresponse to a root :was also repeated. That is, rootsthat were parts of words on their firstpresentation were parts of words on their secondpresentations. It never was the case that rootsthat were parts of pseudowords on their firstpresentation were parts of words on their secondpresentation. This redundancy between roots andresponses might have facilitated the decisionprocess or the selection between the word and nota word response categories, thereby introducingan additional source of facilitation. In the thirdand final experiment, the lexical decisionassociated with a particular root was manipulatedover repetitions. The experiment was designed inorder to identify an episodic component offacilitation associated with response repetition.

This analysis showed a significant effect of typeofprime [Fl (3,141)=9.98, MSe=1826, p<.OOO1 andF2 (3,141)=20.71, MSe=1744; p<.OOOlJ. Post hocTukey-A comparisons of the means indicated thatthe inflectional, derivational, and identity primesall facilitated lexical decision relative to theunrelated condition, (p<.0l). As in Experiment 1,the magnitude of facilitation was similar for thethree related prime types. The analysis of errorscores showed no significant difference due to typeofprime Fl (3,141)=1.49, MSe=1.54, p>.14.

The responses to targets with orthographicallydisrupted roots and targets with continuous rootswere compared by a mixed model ANOVA and aresummarized in Table 4. Targets with continuousroots were marginally faster than (different)targets with disrupted roots F2 (1,46)=3.13,MSe=8787, p<.084. The effect of prime type wasreliable and, consistent with the outcome ofExperiment 1, there was no interaction betweentype of prime and target continuity F (3,138)=0.78,MSe=2459, p>.501. Because the pattern offacilitation was similar for targets withorthographically disrupted and orthographicallycontinuous roots, these data support theconclusion of Experiment 1 that preservation oforthographic pattern is not a necessary conditionfor facilitation due to morphological relatedness.Finally, neither for disrupted roots nor forcontinuous roots were inflectionally-relatedprimes and derivationally-related primessignificantly different from each other.

The outcome of the present experimentreplicated that of Experiment 1. The magnitude offacilitation in lexical decision was not significantlygreater for prime-target pairs related by inflectionthan for pairs related by derivation. Moreimportant, neither was facilitation influenced bythe orthographic integrity of the repeated rootmorpheme. Thus, even when the composition ofpseudowords forced subjects to analyze the

TARGET

Disrupted 640(22)

Continuous 602(11)

575(7)

563(9)

584(9)

568(7)

595(17)

561(10)

172 Feldman and Bentin

EXPERIMENT 3Pseudoword structure influences rejection time

in the lexical decision task. Caramazza, Laudannaand Romani (1988) reported that Italianpseudowords composed of illegal combinations ofreal morphemes were harder to reject thanpseudowords composed of one legal morpheme andone illegal (nonmorpheme) sequence. Similarresults have been reported in English (Katz,Rexer, & Lukatela, 1990). Of course, Italian is aconcatenated language like English andmorphemes consist of uninterrupted sequences ofletters whereas in Hebrew the morpheme root is amore abstract unit. Experiment 3 assesseswhether Hebrew pseudowords words formedaround a meaningful root pose special problemsrelative to pseudowords formed around ameaningless string of consonants. Hebrewpseudowords constructed by combiningmeaningful root morphemes with real wordpatterns were compared with pseudowordsconstructed of a meaningless root with a real wordpattern.

Experiment 3 also attempts to evaluateresponse repetition as a source of facilitation inthis task. In the experiments reported above aswell as in all previously reported repetitionpriming studies, the lexical status of the primeand the lexical decision to the target werematched so that if the answer to the first was"word" then the answer to the second would alsobe "word" and if the answer to first was"pseudoword" then the answer to the secondwould also be "pseudoword." In the presentexperiment, the effect of morphologically-relatedpseudoword primes on word targets wasinvestigated. That is, primes and targets werealways formed around the same root but, due toillegal combinations of root and word pattern, thelexical status of the prime was not always a realword. Failure to find facilitation when the lexicalstatus of prime and target is not matched wouldprovide evidence for a response-related componentto facilitation in the repetition priming task.

The addition of a condition in which pseudowordprimes are followed by word targets serves toeliminate another potential problem ofinterpretation. In the previous two experiments,only words were repeated so that it was possiblethat subjects used repetition of the root as acriterion for deciding the lexical status of a letterstring. That is, if a particular string of consonantshad been presented previously then respond"word." By this account, target facilitationfollowing unrelated primes would be over

estimated as these were first presentations of thatconsonant string. Accordingly, targets followingpseudoword primes formed from the same rootshould show facilitation because the root isrepeated. By contrast, if targets followingunrelated primes (with different roots) and targetsfollowing pseudoword primes (repeated roots) donot differ significantly, then it is unlikely thatsubjects are exploiting repetition of the root per seas a basis for judging the lexical status of a target.

Experiment 3 was designed to differentiate theeffect of repeating a root morpheme from the effectof repeating a lexical decision response (cf. Logan,1989). If facilitation following morphologicalrepetition reflects units for accessing the lexiconrather than lexical processes, then target wordsthat contain a root that was previously presentedshould be faster than targets whose roots werepresented for the first time. Importantly, thelexical status of the word in which the rootappeared should have no effect. That is, both wordand nonword primes that contain the rootmorpheme should facilitate targets. On the otherhand, if morphological components must activatea lexical entry in order to produce facilitation thenroots embedded in pseudowords will not facilitatewords with those same roots. Such an outcomecould also suggest that relatively late processes ofdecision and response selection contribute to thepattern of facilitation in the repetition primingtask.

MethodSubjects. Forty-eight first year students from

the Department of Psychology at HebrewUniversity participated in Experiment 3. As in theprevious experiments, all were native speakers ofHebrew. All had vision that was normal orcorrected-to-normal and all had prior experiencein reaction-time studies although noneparticipated in other experiments in the presentstudy.

Stimulus materials. The materials fromExperiment 1 were modified in the thirdexperiment so that the response for a particularroot was not necessarily constant over first andsecond presentations of that root. The materialsfor the third experiment were identical to those ofthe previous two experiments with two exceptions.First, instead of including an identical repetitionof each target word, a new prime was constructed.It consisted of an illegal combination of the targetroot and a word pattern. Accordingly, the correctlexical decision response for these primes was nota word. As a consequence of introducing a new

Morphological Analysis 0[ Disrupted Morphemes: Evidence from Hebrew 173

principle for constructing primes, two types ofpseudowords occurred within each test order. Onetype consisted of pseudowords formed by creatingan illegal combination of meaningful root and realword pattern. These were pseudoword primes forreal word targets and twelve existed in each list.The other consisted of a real word pattern on ameaningless root and these were pseudowordfillers. Both types of pseudowords were presentedto each subject so that they could be compared.

Thus, the design of the Experiment 3 wassimilar to the design of Experiment 1, except thathere the identity word primes were replaced bypseudoword primes constructed from the sameroot morpheme that appeared in the target.Within each of the four test orders, the forty-eighttargets were preceded equally often bypseudowords, by inflected primes, by derivedprimes, and by unrelated word primes. Across testorders, each target was preceded by each type ofprime.

Procedure. Subjects were instructed to make alexical decision judgment and the procedure andinstructions were identical to those of the twoprevious experiments.

Results and DiscussionMean decision latencies and error rates for

Experiment 3 are summarized in Table 5. Errorsand extreme reaction times were eliminatedaccording to the same constraints used in previousexperiments.

The ANOYA of word latencies revealed asignificant effect of type of prime [Fl (3, 141) =5.76, MSe = 2016, p<.OOl; F2 (3,141) = 11.21, MSe= 2482, p<.Oool] although the analysis of errorscores did not [Fl (3,138)=1.13, MSe=1.14, p>.33].

Table 5. Mean lexical decision times in milliseconds(and SEm) and percentage of errors formorphologically-related. unrelated target words andfor pseudowords in Experiment 3.

PRIMElYPE

Unrelated Identity Inflection Derivation

faster than targets preceded by unrelated words.In replication of previous results, the magnitudeof facilitation was similar for inflectional andderivational type primes. Reaction times totargets preceded by pseudoword primes were notsignificantly different from reaction times totargets preceded by unrelated primes, however.This outcome suggests that when the response toa root was not repeated, repetition of the root perBe was not sufficient to facilitate (or inhibit) lexicaldecision. This outcome is important because itsuggests that word target responses were notsimply facilitated because the same root wasrepeated during the experimental session.Facilitation necessitated activation of a lexicalentry.

Comparison of the meaningful root andmeaningless root pseudowords revealed that thepresence of a meaningful root delayed rejections ofpseudowords by about 200 ms. (698 ms vs. 902 ms,respectively). This difference was statisticallysignificant [Fl(1,47)=88.5, MSe=11280, p<.OOOl]and, is consistent with the results found inconcatenated languages such as Italian andEnglish.

As in the previous experiments, latencies anderrors to targets containing disrupted andcontinuous roots were compared. They aresummarized in Table 6. The ANOYA showed thatcontinuous and disrupted target types were notsignificantly different F 2 (1,46)=1.78, MSe=9960,p>.18. In replication of previous experiments, theeffect of type of prime was significant but therewas no interaction between type of prime andcontinuity F2 (1,138)=0.88, MSe=2640, p>.44.

Table 6. Mean lexical decision latency in milliseconds(and SEm) for target words with disrupted andundisrupted roots in the four priming conditions inExperiment 3.

PRIMElYPE

Unrelated Identity Inflection Derivation

TARGET

RTs (SEm) 653(13.5)

Errors 2.4(0.7)

647(10.1)

1.9(0.5)

608(7.30)

1.7(0.5)

609(7.60)

1.6(0.4)

Disrupted 634(17)

Continuous 661(19)

638(14)

650(14)

609(11)

613(10)

597(9)

631(12)

For latencies, post-hoc Tukey-A revealed thattargets preceded by inflectionally- andderivationally-related primes were significantly

The difference in the lexical decision latencybetween the two types of pseudowords suggeststhat during the process of lexical decision, roots

174 Feldman and Bentin

were examined and that readers cannot ignore themeaningfulness of the roots even when they arecomponents of pseudowords. Nevertheless, thepresence of a meaningful root in a pseudowordcould not facilitate later lexical decision to a wordformed from the same root. Because only thepseudoword-word combination was examined, thisoutcome could suggest that repetition of theepisode or particular response is a source offacilitation in the repetition priming task.Alternatively, it is plausible that morphologicalcomponents must activate a lexical entry in orderto produce facilitation at a later point. In anyevent, it appears that the locus of root facilitationcannot be prelexical.

GENERAL DISCUSSIONIn a series of three lexical decision experiments,

significant facilitation due to morphologicalrelatedness of prime and target was observed withHebrew materials. Subjects performed a lexicaldecision to both prime and target and 7 to 13items intervened between them. When relatedprimes were matched for overall orthographicsimilarity to targets, facilitation by inflectionalprimes was equivalent to facilitation byderivational primes, both of which werestatistically equivalent to facilitation by identicalrepetitions. Similar magnitudes of facilitation forthe two types of morphological primes isinteresting because forms related by derivationgenerally tend to be less similar in meaning thanforms related by inflection (Aronoff, 1976).Moreover, in our particular experiments, pairsrelated by inflection were always of the sameword-class whereas pairs related by derivationchanged word class. Evidently, the facilitationthat underlies repetition priming amongmorphologically-related forms cannot reflectpreservation of shared meaning over prime andtarget. These results are consistent with the claimthat at long lags, semantic relatedness per se isnot a primary source of facilitation in therepetition priming task (Bentin & Feldman, 1990),and support a distinction between facilitation dueto associative and morphological relatedness(Henderson, 1985).

Alternative accounts of facilitation betweenmorphologically-related prime-target pairs em­phasize the repetition of phonological and ortho­graphic patterns conveyed by a shared morpheme.As described above (see also Berman, 1971), andin contrast to concatenated morphologies such asthat of English, morphologically complex words inHebrew consist of a root morpheme of consonants

into which a word pattern is infixed.Consequently, root morphemes are abstract pat­terns that cannot be realized as unified phonologi­cal entities. In the present study, roots were re­peated over related prime and target but, becauseword patterns changed, related words were notassociated with a common phonological structure.Nevertheless, facilitation was observed. In conclu­sion, appreciation of morphological relatednessdoes not require phonological identity. As appliedto the repetition priming task, repetition of aphonological unit is not necessary in order to pro­duce morphological facilitation.

Accounts of morphological effects that empha­size orthographic structure (e.g., Seidenberg,1987; Seidenberg & McClellend,1989) may bemore appropriate for concatenated languages be­cause morphemes tend to be orthographic as wellas linguistic units. For example, in English, thebase morpheme is typically undisrupted by mor­phological manipulations.3 Nevertheless, previousstudies in English have demonstrated that formorphologically-related words, the repetition oforthographic form plays only a minimal and sta­tistically insignificant role in the morphologicalrepetition effect (e.g., Napps, 1989; Napps &Fowler, 1987). Similarly in Serbo-Croatian, facili­tation in repetition priming was numericallyequivalent when prime and target were both writ­ten in the same alphabet (e.g., NOGOM - NOGA)and when prime was in one alphabet (e.g.,nogAl and target was in the other (e.g., NOGA)(Feldman & Moskovljevi~, 1987; Feldman, inpress). In Hebrew, the root is always phonologi­cally and sometimes also orthographically dis­rupted because of its nonconcatenated structure.The major contribution of the present result is tounderscore the limitations of an orthographic ac­count of morphological analysis. This claim isbased on the following evidence.

First, the magnitude of target facilitationfollowing morphological relatives was similar tothat following identical repetitions although theorthographic similarity of the inflected andderived primes to their matched targets was, bydefinition, smaller than with identity primes.Second, in all three experiments, the comparisonbetween prime-target pairs with orthographicdisruptions to the root and pairs with continuousroots yielded no significant differences. Moreover,in Experiment 3, repetition of the root did notfacilitate Ie> decision to the target if its firstpresentatior in the context of a pseudoword,even thou the pseudoword was asorthographicaHy similar to the target as were the

Morphological Analysis ofDisrupted Morphemes: Evidence from Hebrew 175

related words. These results are consistent withthe outcome of a similar study conducted withEnglish materials (Fowler, et a1., 1985) in thatchanges in spelling (and/or pronunciation) had noeffect on the pattern of facilitation betweenmorphologically-related prime-target pairs in therepetition priming task. The implication of theabove is that facilitation in the repetition primingtask in nonconcatenated as well as concatenatedlanguages cannot be attributed to repetition of anoverall orthographic form nor to preservation,over successive presentations, of the continuity ofan orthographic pattern. In summary,morphological analysis is cannot be tied toorthographic units.

Inflections and derivations are contrasted bylinguists as representing two different types ofmorphological formations. In English, inflectionalaffixes are few and tend to be composed of three orfewer letters whereas derivational endings can becomposed of a more variable number of letters.Moreover, some derivations change the meaningand pronunciation of the base morpheme in amanner that is not characteristic of inflections(Chomsky & Halle, 1968). In Hebrew, it is possibleto find inflectional and derivational relatives of atarget that modify the structure of the root to asimilar degree although they necessarily differwith respect to their semantic similarity to thetarget. In the present repetition priming study, nodifferences between inflectional and derivationaltypes of morphological formations were observed.Consistent with the conclusion of Napps (1989)and Napps and Fowler (1987), it is evident thatfacilitation due to morphological relatedness inthe present study does not represent theconvergence of semantic, orthographic, andphonological relationships.

Locus of morphological effectsIn order to observe morphological facilitation in

lexical decision, it is not necessary thatorthographic pattern be preserved and thisfinding has been interpreted to mean thatmorphological analysis is not tied to anorthographic pattern. Similarly, facilitationpatterns are not sensitive to the semantic overlapof prime and target in either this or an earlierstudy (Feldman, 1992). Because the morphologicalcharacter of a word cannot be captured by itsorthographic and semantic properties, it seemsthat the morphological structure in general andthe Hebrew root morpheme in particular must berepresented. A morphological representation in

the lexicon has been proposed by severalinvestigators (e.g., Grainger, Cole; & Segui, 1991).

The claim that morphological effects in wordrecognition reflect lexical processes is based onseveral sources of evidence. Typically, effects ofrepeating a morpheme are numerically larger andstatistically more robust for word than forpseudoword prime-target pairs. Significantfacilitation for pseudowords in the repetitionpriming task is unreliable even when the negativelexical decision is repeated over prime and targetwith the same continuous base morpheme (e.g.,Duchek & Neely, 1989; Feldman & Moskovljevic,1987). For example, in the one repetition primingstudy where Hebrew pseudowords were repeated(Bentin & Feldman, 1990), evidence forfacilitation due to repetition with pseudowordsdepended on the choice of a baseline. Similarly, inat least one study with English materials (Fowler,et aI., 1985), evidence of facilitation withpseudowords depended on the number of itemsintervening between prime and target (see alsoScarborough, Cortese, and Scarborough, 1986).For morphologically related word pairs, bycontrast, effects tend to be larger in magnitudeand manipulations of lag are not significant(Feldman, in press). The results of Experiment 3also cast doubt on a locus for the morphologicalfacilitation that is independent of the lexicon. If itwere possible for subjects to extract a root fromboth word and pseudowords prior to accessing thelexicon, then the effect on word targets of wordand pseudoword primes should have been similar.Analogous effects for word and pseudowMdprimes were not observed, however.

A second source of evidence that (at least some)morphological effects are lexical in origin is theinteraction of morphological with frequencyeffects. Although it is not the case in repetitionpriming that (relative) frequency ofmorphologically-related prime and target had asignificant effect (Feldman, 1992), morphologicaland frequency effects often interact in otherrecognition tasks. Accordingly, more frequentwords are less sensitive to manipulations ofmorphological structure than are less frequentwords. For example, in an experimentalproduction task (Stemberger & MacWhinney,1986; 1988), the error rate on lower-frequencymorphologically-complex forms was significantlyhigher than on higher-frequency verb forms.Similarly, it has been suggested (Caramazza eta1., 1985) that both whole word and morphologicalunits may constitute viable units for accessing the

176 Feldman lind Benlin

lexicon but that the availability of the former areconstrained by the frequency of the particularsurface form.

It is important to point out that the measure ofvariance included in Table 3 provides no evidencethat performance was more variable in thepseudoword prime condition than in the unrelatedprime condition. Therefore, an account based oncompensatory processes such as facilitation due torepetition of the root being offset by a change ofresponse to that root seems implausible.

Evidence that facilitation due to morphologicalrelatedness is lexical in locus is compelling andfits well with the results of studies that useddifferent experimental paradigms. What is lessobvious is how to account for the effect ofmorphemic composition on pseudoword rejectionlatencies. Rejection latencies were prolonged forpseudowords that included a meaningful rootrelative to pseudowords that did not This outcomefor real roots in illegal combinations with wordpatterns could reflect a relatively late andstrategic re-evaluation of the decision processanalogous to the spelling check necessary forpseudohomophone rejection.

Recently, Grainger et a1. (1991) have identifiedtwo plausible lexical loci for morphological effectsin word recognition. As usually conceived,morphological effects are interpreted as sublexicalin origin so that morphological relatedness isrepresented as a system of facilitatory connectionsbetween lexical entries for morphologically-relatedwords or as a pattern of activation amongmorphological units at a level intermediatebetween word and letter level units. Whetherinterpreted as a system of connections betweenwhole word forms or as patterns of activationamong shared morphological units, the traditionallocus of morphological relatedness is sublexical(but not prelexical) in that it is intermediatebetween word and letter levels. As noted byGrainger and his colleagues (1991), according to asublexical account, one might expect to observeinhibition among morphologically-related wordsbecause of their shared orthographic structure butthis outcome has not been reported. Alternatively,morphological units may be represented at a levelabove the word so that all words formed from thesame base morpheme are linked by facilitatoryconnections to the morpheme and conversely, fromthe morpheme back to related words. By thesupralexical account, activation spreads from aspecific word to its base morpheme and then on toother words that are morphologically-related to it.An extension of the supralexical account is

consistent with the claim that facilitation in therepetition priming task with Hebrew materialsmay reflect the process of extracting the root. fromthe root plus word pattern combination thatconstitutes a word (Bentin & Feldman, 1990). Italso alleviates the problem of identifying amorpheme which, in Hebrew, is neither aphonological nor an orthographic entity.Segmenting root from word pattern in Hebrewnecessarily requires extensive lexical knowledge,therefore the process of root extraction in Hebrewmust be distinguished from prelexical processessuch as affix stripping (Taft & Forster, 1975).Almost all Hebrew pseudowords have legalorthographic (and phonological) patterns so thattheir differentiation from words must entailexamination of the root and may even include anevaluation of its semantic content. Thisidentification may require extracting the root fromthe word. It is plausible that when roots arerepeated over prime and target words inrepetition priming, it is the identification of theroot that is facilitated. Of course, even theextraction of a semantically meaningful root fromits word context is not sufficient to reliablycategorize a string as a word. The combination ofroot morpheme and word pattern must also beevaluated. It was observed in Experiment 3 thatpseudowords composed of a meaningful root inillegal combination with a word pattern were moredifficult to reject than pseudowords formed arounda meaningless root. Activation from the root couldspread down to letter level even in the absence ofword level activation and this pattern of activationthroughout the system could have the effect ofbiasing the decision process toward a wordresponse.

In summary, both lexical and postlexicalinfluences may contribute to the pattern offacilitation in the repetition priming task. Forlexical decision, response repetition about thelexical status of a particular morpheme in aparticular (word or pseudoword) contextconstitutes a postlexical contribution. Support forthe lexical aspect of morphological analysis is tiedto the pattern of facilitation in the repetitionpriming task for word targets. It could arise eithersublexically or supralexically. The noncon­catenative morphological structure of Hebrewlends itself to a supralexical representation ofmorphology. If common morphological units arecaptured at a level above the word thendiscontinuities of phonological or orthographiccomponents of a morpheme are no longerproblematic. Prolonged latencies for pseudoword

MorplwlogicaJ Analysis ofDisrupted Morphemes: Evidence from Hebrew 177

composed of illegal combinations of root and wordpattern relative to pseudowords composed fromnonroot are also anticipated. In sum, morpho­logical analysis in word recognition is not tied toorthographic form and entails lexical knowledge ateither a sublexical or a supralexicallevel.

REFERENCESAronoff, M. (1976). Word fOrmRtion in genmltiTle grllm1flllr.

Cambridge, MA: MIT Press.Bentin, 5., &: Feldman, L. B. (1990). The contribution of

morphological and semantic relatedness to repetition primingat short and long lags: Evidence from Hebrew. QuArterly founuIlofExperimentll1 Psychology, 42A, 693-711.

Bentin, 5., &: Moscovitch, M. (1988). The time course of repetitioneffects for words and unfamiliar faces. Jounllll of ExperimentlllPsycholcgy: GenerIIl, 117, 148-160.

Bentin, 5., &: Peled, 8-5., (1990) The contribution of task-relatedfactors to ERP repetition effects at short and long lags. Memorylind Cognition, 18,359-367.

Berman, R. A. (1978). Modern Hel1mv Structure. Tel Aviv:University Publishing Projects.

Caramazza, A., Laudanna, A., &: Romani, C. (1988). Lexical accessand inflectional morphology. Cognition, 28,287-332

Chomsky, N., &: Halle, M. (1968). The sound JHlHern ofEnglish. NewYork: Harper and Row.

Dannenbring, G. L., &: Briand, K. (1982). Semantic priming andthe word repetition effect in a lexical decision task. ulnlldumJournlll ofPsychology, 36, ~5-444.

Duchek, J. M. &: Neely, J. H. (1989). A dissociative word­frequency x levels-of-processing interaction in episodicrecognition and lexical decision tasks. Memory lind Cognition,17,148-162.

Emmorey, K. D. (1985). Auditory morphological priming in thelexicon. l..IInguAge lind Cognititle Processes, 4, '73-92.

Feldman, L. B. (in press). Bi-alphabetism and the design of areading mechanism. In D. M. Willows, R. S. Kruk, &: E. Corcos(Eds.), VisUill processes in reIIding lind reIIding diSllbilities.Hillsdale, NT: Erlbaum.

Feldman, L. B. (1992). Morphological relationships revealedthrough the repetition priming task. In M. Noonan, P.Downing, &: S. Urna (Eds.), Unguistics lind Utervcy. (pp. 239­254). Amsterdam/Philadelphia: John Benjamins Pub. Co.

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., &: Moskovijevi~, J. (1987). Repetition priming isnot purely episodic in origin. Journlll ofExperimentll1 Psycholcgy:Leurning, Memory & Cognition, 13,5'73-581.

Forster, K. I., &: Davis, C. (1984). Repetition priming andfrequency attenuation in lexical access. Journlll of ExperimentlllPsycholcgy; Leurning, Memory lind Cognition, 10,680-698.

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.

Grainger, J., Co~, P., &: Segui, J. (1991). Masked morphologicalpriming in visual word recognition. Memory & l..IInguAge, 30,370-384.

Hankamer, J. (1989). Morphological parsing and the lexicon. In W.Marslen-Wilson (Ed.), LexicIIl representlltion lind process. (pp. 392­4(8). Cambridge, MA: MIT Press.

Hanson, V. L., &: Feldman, L. B. (1989). Language specificity inlexical organization: Evidence from deaf signers' lexical

organization of ASL and English. Memory & Cognition, 17,292­301.

Hanson, V. L., &: Wilkenfeld, D. (1985). Morphophonology andlexical organization in deaf readers. l..IInguAge lind Speech, 28,269-280.

Henderson, L (1985). Toward a psychology of morphemes. In A.W. Ellis (Ed.), Progm;s in the psycholcgy of IllnguAge (pp. 15-72).London: ErIbaum.

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

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

Katz, L., Rexer, K., &: Lukatela, G. (1991). The processing ofinflected words. PsychologiCIIl Restrlrch, 53, 25-31.

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

Kempley, 5., &: Morton, J. (1982). The effects of priming withregularly and irregularly related words in auditory wordrecognition. British JOUnlIIl ofPsychology, 73,441-454.

Logan, G. (1988). Toward an instance theory of automatization.PsychologiCIII Review, 95,492-527.

Matthews, P. H. (1974). Morphology. Cambridge: CambridgeUniversity

McCarthy, J. J. (1981). A prosodic theory of nonconcatenativemorphology. Unguistic Inquiry, 12,3'73-346.

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

Napps, S. E. (1989). Morphemic relationships in the lexicon: Arethey distinct from semantic and formal relationships? Memory& Cognition, 17, 779-739.

Napps, S. E., &: Fowler, C. A. (1987). Formal relatiorwhips amongwords and the organization of the mental lexicon. JOUnlIIl ofPsycholinguistic Restrlrch, 16,257-272

Oman, U. (1971) Binyanim, Ubsisim, Netiyot, and Hataiyot.HRUnitln'Sita, 16,15-22. (in Hebrew)

Prinzmetal, W., Hoffman, H., &: Vest, K. (1991). Automaticprocesses in word recognition: An analysis form illusoryconjunctions. Journlll of Experimentlll Psychology: HumllnPerception lind Perfimrumce, 17, 902-923.

Rapp, B. C. (1992). The nature of sublexical orthographicorganization: The bigram trough hypothesis examined. JOUTlllllofMemory lind l..IInguAge.

Scarborough, D. L., Cortese, C. &: Scarborough, H. S. (1986).Frequency and repetition effects in lexical memory. Journlll ofE:cperimentll1 Psychology: Humtm Perception lind PerjOmumce, 4, 1­17.

Schriefers, H., Friederici, A., &: Graetz, P. (1992). Inflectional andderivational morphology in the mental lexicon: Symmetriesand asymmetries in repetition priming. Quarterly Journal ofExperimental Psychology, 44A, 3'73-390.

Seidenberg, M. (1987). Sublexical structures in visual wordrecognition: Access units or orthographic redundancy? In M.Coltheart (Ed.), Attention lind Perfomumce XlI: The Psycholcgy ofRellding. (pp. 245-263). Hillsdale, NT: Erlbaum.

Seidenberg, M., &: McClelland, J. L. (1989). A distributed,developmental model of visual word recognition and naming.PsychologiCIII Review, 96, 523-568.

Stanners, R. F., Neiser, J. J., Hernon, W. P., &: Hall, R. (1979).Memory representation for morphologically related words.Journlll ofVerbII1 Leurning lind VerbIIl Behlltlior, 18, 399-412.

178 Feldrruln and Bentin

Stemberger, J. P., &. MacWhinney, B., (1986). Frequency andlexical storage of regularly inflected words. Mnrwry lindCognition, 14, 17-26.

Sternberger, J. P., &. MacWhinney, B., (1988). Are inflected Cormsstored in the lexicon? M. Hammond &. M. Noonan (Eds.)TheoretiC'JI1 morphdogy. San Diego, CA: Academic Press.

Taft, M., &c Forster, K. I. (1975). Lexical storage and retrieval ofprefixed words. lou17lll1 of VtrbllllAlrning lind Vtrblll 8ehIwior,14,638-647.

FOOTNOTEStAlso State University of New York at Albany.

ttDep8rtment of Psychology, The Hebrew University, JenISa1em.lWe thank Len Katz for developing the software.2Because the effect of continuity was not significant over items, F

values over subjects were not included.3Exceptions include alternations of strong vowels in pairs such

as SING-SUNG and MEET-MET.