Haskins Laboratories Status Report on Speech Research1993, SR-115/116, 69-97

Beyond Orthography and Phonology: Differences betweenInflections and Derivations*

Laurie Beth Feldmant

The influence of morphological structure was investigated in two types of word recognitiontasks with Serbian materials. Morphological structure included both inflectional andderivational formations and comparisons were controlled for word class and theorthographic and phonological similarity offorms. In Experiments 1, 2, and 3, the patternof facilitation to target decision latencies was examined following morphologically-relatedprimes in a repetition priming task. Although all morphologically-related primesfacilitated targets relative to an unprimed condition, inflectionally-related primesproduced significantly greater effects than did derivationally-related primes. InExperiments 4, 5, and 6 subjects were required to segment and shift an underlined portionfrom one word onto a second word and to name the result aloud. The shifted lettersequence was sometimes morphemic (e.g., the equivalent of ER in DRUMMER) andsometimes not (e.g., the equivalent of ER in SUMMER). Morphemic letter sequences weresegmented and shifted more rapidly than their nonmorphemic controls when they wereinflectional affixes but not when they were derivational affixes. These results indicate that(a) morphological effects cannot be ascribed to orthographic and phonological structure, (b)the constituent morphemic structure of a word contributes to word recognition and (c)morphemic structure is more transparent for inflectional than for derivational formations.

Morphology underlies the productivity of theword-formation process and a word's fit into thesyntactic frame of a sentence. Linguists distin­guish between two classes of morphological forma­tions. Words that differ in their derivational af­fixes but share a base morpheme (e.g.,CALCULATION, CALCULATOR) are generallyconsidered to be different lexical items and tohave different meanings. Words that differ intheir inflectional affixes (e.g., CALCULATING,CALCULATED) but share a base morpheme aregenerally considered to be versions of the same

The research reported here was conducted at the Laboratoryfor Experimental Psychology at the University of Belgrade andwas supported by funds from National Institute of ChildHealth and Development Grant HD-01994 to HaskinsLaboratories. Portions were presented to the November 1987meeting of the Psychonomic Society. I wish to thank DarinkaAndjelkovic who conducted the experiments and DraganaBarac-Cikoja and Petar Makara who helped prepare thematerials. Valuable comments on earlier versions of thismanuscript were provided by David Balota, Gary Dell, CarolFowler, James Neely and several anonymous reviewers and Ithank them all.

69

word, with the particular version that appears ina sentence being determined by the syntax of thesentence. In general, inflectional formations aremore productive, do not change word class mem­bership relative to the base morpheme and aremore constrained by syntax (Anderson, 1982) thanare derivational formations. In addition, meaningsof inflected forms tend to be compositional of themeaning of the base and affix morphemes,whereas meanings of derived forms are less oftencompositional. The present study examines howinflectional and derivational formations areprocessed.

Four principles of lexical storage have been pro­posed for words composed of several morphemes,that is morphologically-complex words. First, aprinciple of economic storage makes it appealingto represent complex forms in terms of a basemorpheme. Accounts based on base morphemesare adequate for inflectional forms (e.g.,Caramazza, Laudanna, & Romani, 1988) but areless plausible for derivations, in part, because (a)the formation rules for derivations are complex

70 Feldman

and there is no way to ascertain whether a partic­ular form has been created and (b) the semanticcontribution of the base morpheme to the meaningof the morphologically complex derivational formis unpredictable. Second, accounts based on thestem (base morpheme plus derivational affix, ifany e.g., Burani & Laudanna, 1992) posit differentrepresentations for inflections and derivations.For example, in a lexical decision task where bothitems are formed around the same base mor­pheme, words with a derivational affix producedifferent patterns of facilitation between itemsrelative to words with only an inflectional affix(Laudanna, Badecker, & Caramazza, 1992).Although it is likely that the lexical representa­tion of inflected and derived forms differs, the re­lation between the two types of formations is un­derspecified. Third, morphologically complexwords may be represented mentally as wholeforms, without reference to their constituents(Butterworth, 1983). Fourth, Caramazza and hiscolleagues have proposed that both base mor­pheme and whole word are units for lexical access,that these alternatives are not mutually exclusive(e.g., Caramazza, Miceli, Silveri, & Laudanna,1985), and that word frequency may playa keyrole (Caramazza et al., 1988).

The repetition priming paradigm (Stanners,Neiser, Hernon, & Hall, 1979), yielded evidencethat morphological relationships constitute aprinciple of organization within the internal lexi­con. The influence of morphological relatedness isassessed by comparing lexical decision latency oraccuracy to the target preceded by a morphologicalrelative to (a) a first presentation of the targetword (i.e., no prime) and (b) an identical repetitionof the target word. Sometimes the reduction in re­action times and errors that occurs with morpho­logical relatives as primes is equivalent to the ef­fect of an identical repetition (e.g., Fowler, Napps,& Feldman, 1985). Other times, decision latenciesto targets following morphological relatives arereduced relative to first presentations but areslower than identical repetitions. The latter pat­tern is ambiguous. It has been interpreted as evi­dence of separate lexical entries (e.g., Stanners etal., 1979) and as evidence of interrelated entries(e.g., Fowler et al., 1985).

Facilitation due to morphological relatedness oc­curs in the lexical decision task across a variety oflanguages including Serbian (Feldman & Fowler,1987), Hebrew (Bentin & Feldman, 1990), as wellas English (Fowler et al., 1985; Feldman, 1992)and American Sign Language (Hanson &Feldman, 1989) and across a variety of conditions.

Facilitation in repetition priming has been ob­served when prime and target are in either thesame or different modalities (Fowler, et al., 1985;see also Kirsner, Milech, & Standen, 1983).Primes can be morphologically complex and tar­gets can be morphologically simple or primes canbe simple and targets complex (Feldman &Fowler, 1987; Schriefers, Friederici, & Graetz,1992). The latter observation is significant be­cause morphologically simple forms tend to behigher in frequency than morphologically complexforms. For complex targets, both derived and in­flected formations show effects based on morpho­logical relatedness (Fowler et al., 1985; Schrieferset al., 1992).

Several studies have tried to compare patternsof facilitation at long or at short lags for prime­target pairs related by inflection and by deriva­tion. Differences in facilitation (ms) for targets fol­lowing morphologically-related and unrelatedprimes are summarized in Table 1. As shownthere, inflectional primes typically producegreater facilitation than derivational primes butthe difference is often not statistically significant.For example, the words POTARONO andPOTETE are related by inflection and the wordsPOTATORE and POTETE are related by deriva­tion. In a lexical decision task, both pairs pro­duced faster latencies than unrelated pairs(Laudanna et al., 1992; Exp. 1). In these experi­ments, morphological relationship was defined ona single lexical item in Italian. Similarly,SPARIZIONE meaning disappearance is definedas a derivation whereas SPARIVANO meaningthey disappeared is defined as an inflection. Onlythe latter slowed recognition of (morphologically­unrelated) SPARATI which is the past participleof shot and is formed from a different but homo­graphic base morpheme (Laudanna et al., 1992;Exp. 3). In a repetition priming task with Germanmaterials (Schriefers et al., 1992), inflectionalprimes consisting of different inflected adjectiveforms produced greater facilitation than deriva­tional primes consisting of abstract nouns formedfrom adjectives. Finally, in Hebrew (Feldman &Bentin, in press), morphological relationship wasdefined over the word pair because it is not alwaysobvious which item is derived from which, but nodifferences between inflections and derivationswere observed in the repetition priming task. Inshort, the pattern of results observed with variouspriming procedures indicates that differences be­tween inflectional and derivational facilitationhave appeared, but that they are often not reliablysignificant in separate comparisons.

Beyand Orthography and Phonology: Differences between Inflections and Derivations 71

Table 1. Summary of facilitation for inflectional and derivational targets following identity, inflectional andderivational primes in immediate and long term priming tasks.

IDENTITYTYPE OF PRIME

INFLECTIONAL DERIVATIONALI

INTERVALDERrVATIONAL2

STUDY:Stanners et aI., 1979. Exp. l a

166 181160 150140 131

Stanners et aI., 1979. Exp. 2b

84 4999 39

Stanners et aI., 1979. Exp. 3c

120118

Fowler et aI., 1985d

101 7842

Feldman & Fowler, 1987. Exp. Ie54 45

Feldman & Fowler, 1987. Exp. 2f

90 74Feldman & Fowler, 1987. Exp. 3g

58 50

Feldman & Bentin, 1992. Exp. 1h68 60

Schriefers, Friederici, & Graetz, 1992, Exp. 2i

108 9990

Laudanna, Badecker, & Caramazza, 1992, Exp. Ij26

7232

47

59

5026

35

long

long

long

long

long

long

long

long

long

44short

a simple regular targets and inflected primes e.g., BURNS-BURNb simple regular targets and irregular inflected primes e.g., HUNG-HANGc simple regular targets and regular derived primes e.g., SELECTIVE-SELECTdsimple targets with sound change primes e.g., HEALTH-HEALe simple targets with regular inflected primes e.g.. DlNARA-DlNARf inflected targets with simple and inflected primes e.g., DlNAR(OM)-DlNARAg simple targets with inflected sound change targets PETKUIPETKOM-PETAKh complex targets with complex primes e.g. NAFAUNEFEL-NOFELi simple and complex targets with simple and complex primes e.g., ROTEJROTLICH-ROTj inflected targets e.g., RAPIVANOIRAPITORE-RAPIRE

The repetition priming results summarizedabove clearly demonstrate that under some condi­tions, morphological effects do arise in wordrecognition tasks. However, contrasts between theeffects of inflections and derivations have not beencompelling. Some of the experiments includedboth inflectional and derivational forms but theydid not explicitly compare these types of morpho­logical formations. When planned comparisons be­tween inflectionally- and derivationally-relatedprime-target pairs have been included, resultshave been equivocal. For example, whereasStanners et al. (1979) reported significant differ­ences in magnitude of facilitation for these two

types of formations when they were regular only,Fowler et al. (1985) found no significant differencealthough small numerical differences typicallywere evident. Although these experiments withEnglish materials included a comparison of facili­tation with inflectional and derivational primes,this comparison is not without its problems. InEnglish, inflectional formations tend to be moresimilar in form and meaning than are derivationalformations (or alternatively, forms related by in­flection share a stem as well as a base morphemewhereas forms related by derivation typicallyshare only a stem). This observation is relevantbecause at short lags, orthographic overlap is

72 Feldman

sometimes reported to influence the pattern of fa­cilitation in this and similar tasks (Emmorey,1989; Napps & Fowler, 1987; Stolz & Feldman, inpress). Moreover, the number of inflectional af­fixes for English is severely limited relative to thenumber of derivational affixes. These limitationsimpede a rigorous experimental comparison be­tween inflectional and derivational formationswith English materials.

By contrast, in Serbian it is possible to identifyinflection-derivation pairs with only minimaldifferences in form and meaning. One suchcontrast entails agents and other nouns formedfrom verbs. For example, PEVAC, meaning singer,is formed from the verbal base morpheme PEVand the derivational affix AC. The same basemorpheme appears in all present tense forms ofthe verb to sing including PEVA and PEVAM.Other sets of inflection-derivation pairs entailverb forms that share a base morpheme but differin aspect (which reflects temporal properties ofthe verb). Perfective and imperfective aspect canbe marked by the vowel ofthe suffix, by a prefix orby an infix. Although it is sometimes difficult toascertain which is the derived form, it is wellestablished that perfective and imperfective verbsin Serbian are derivationally related to each other.(Therefore, in the present study, derivation will bedefined relative to a target.) Of course, each canbe inflected to produce different verb forms. Forboth agent and aspect type of derivations, it ispossible to identify inflectional forms with thesame base morpheme so that the orthographic andphonemic overlap of primes with targets ismatched across derivational and inflectionalcomparisons.

The first three of the present experiments wererepetition priming experiments in which nativespeakers of Serbian performed a lexical decisiontask with Serbian materials. Targets were pre­ceded by other forms that were either inflection­ally or derivationally related to the target.Inflectional and derivational formations werematched for phonological and orthographic over­lap with the target. In Experiment 1, targets suchas PEVA (third person singular verb) were pre­ceded an average of ten items earlier in the list by(a) an identical repetition, PEVA (b) the inflec­tionally-related prime, PEVAM (first person sin­gular verb) or (c) the derivationally-related prime,PEVAC (nominative singular of agentive).Inflectional and derivational primes were matchedwith respect to orthographic and phonologicalsimilarity to the target but derivational forms didnot preserve the word class of the target. In

Experiment 2, prefixed or infixed imperfectiveverb targets in third person plural such asOBARE and GURNU were preceded by (a) anidentical repetition, (b) an inflectionally-relatedprime, OBARIM or GURNEM (first person singu­lar verbs), or (c) a derivationally-related prime,BARIM or GURAM (first person singular verbs)that differed in aspect. In Experiment 3, perfec­tive targets such as NOSE (third person pluralverbs) were preceded by (a) an identical repetition,(b) an inflectionally-related prime, NOSIM (firstperson singular verbs) or (c) a derivationally-re­lated prime, NOSAM (imperfective first personsingular verbs) where the last differed in aspect.Here, all primes and targets were verb forms andprime-target similarity was matched across onehalf of the inflectional and derivational primes.Using planned comparisons, target facilitation inlexical decision following inflectionally-relatedprimes and derivationally-related primes wascompared and facilitation following derivationally­related primes relative to first presentations wasassessed.

In order to ascertain that the morphological ef­fects observed in repetition priming were not spe­cific to the lexical decision task, the effect of mor­phology was also investigated in a second experi­mental task. In Experiments 4, 5, and 6, subjectswere required to segment and shift the final se­quence of letters from a visually-presented sourceword to a target word and to name the new formaloud. Morphemic segments were compared withtheir phonemically- and orthographically-matchedbut nonmorphemic controls and both inflectionaland derivational segments were examined. Thestructure of experimental materials for the pre­sent study is described in Table 2.

By linguistic accounts, the component structureof inflections is more transparent than that ofderivations. The repetition priming task hasproven itself to be sensitive to morphological rela­tions between prime and target, but attempts tocompare the patterns of facilitation in repetitionpriming for inflectionally- and derivationally-re­lated prime-targets pairs have not yielded unam­biguous results. This outcome may reflect the factthat in English, derivational affixes tend to becomposed of more letters and to be semanticallyless· compositional than are inflectional affixes.Experiment 1 was designed to compare these twotypes of morphological formations when effects ofaffix length are matched. In Serbian, it is easier tomeet these constraints than in English becauseextensive families of words are formed from thesame base morpheme.

Beyond Orthography and Phonology: Differences between Inflections and Derivations

Table 2. The morphological constituents ofmorphologically simple and complex words in Serbian.

SlEM

73

WORD

SERBIAN

PEVAMPEVAC

NOS1MNOSAM

PRESOMPRELOM

CEVICEDI

lEDEMBEDEM

ZIDARKADAR

BAlAMSAlAM

BASTICAKOSTICA

BASE

PEVPEV

NOSNOS

PRESPRELOM

CEVCED

JEDBEDEM

ZIDKADAR

BAlSAlAM

BASTKOSTICA

SUFFIX:DERIVATION

AC

AR

AM

ICA

SUFFIX:INFLECTION

AM

1MAM

OMfracture

EM

MEANING

I singsinger

I carry (perfective)I carry (imperfective)

press (instrumental)

pipes (nominative plural)he wrings

I eatembankment

brick layersequence

I do magicfair

garden (diminutive)pit

MethodsSubjects. Twenty-seven first year students from

the Department of Psychology at the University ofBelgrade participated in Experiment 1. All werenative speakers of Serbian. All had vision thatwas normal or corrected to normal and had priorexperience in reaction-time studies.

Stimulus materials. Twenty-seven Serbian wordtriples were selected. Fourteen consisted of a nountarget in nominative case with an inflectionally­related form in instrumental case and a deriva­tionally-related verb form. For example, the nom­inative target BROD, meaning boat, was pairedwith its instrumental BRODU and with BROD!,the third person singular of the verb meaning tosail which is derivationally-related to BROD, thetarget. The remaining thirteen triples consisted ofverb targets in one of three singular person formswith another inflected form of that same verb andwith the agentive derived from that verb. For ex­ample, the target PEVA, meaning he sings, waspaired with PEVAM, meaning I sing, andPEVAC, meaning singer. All words were highlyfamiliar, contained between three and seven let­ters, and were printed in Roman script. They arelisted in Appendix 1. Twenty-seven orthographi­cally and phonemically regular pseudowords were

generated by changing one or two letters (vowelwith vowel or consonant with consonant) in basesof other real words. Triplets were generated forthese pseudowords in a fashion analogous to thatfor words (i.e., affixes were real).

A member of each morphologically-related word(and pseudoword) triple appeared once as a targetand once as a prime. In the identity condition, thesame form occurred twice. In the inflectionally­related condition, the prime was another inflectedform of the target and it necessarily preservedword class. In the derivationally-related condition,the prime was a verbal form for noun targets anda noun form for verb targets. Inflectionally andderivationally related primes were each one or twoletters longer than the target and within a pair,overlap was perfectly matched phonemically aswell as orthographically. Finally, the full targetwas contained within the inflectionally andderivationally related primes. For example, bothBRaD! and BRODU are each one letter longerthan and include the target BROD. Primes andtargets were separated by an average of 10 inter­vening items with a range was of 8 to 12 items.

Procedure. Individually tested subjects per­formed a lexical decision task. On each trial, avisual fixation signal accompanied by an auditory

74 Feldman

signal appeared for 200 ms then a target letterstring printed in upper case was presented for 750ms. As each target letter string appeared on theCRT of an Apple II computer, the subject presseda telegraph key with both hands to indicatewhether or not it was a word. A press of thefarther key signaled "yes" and the closer key, "no".Reaction time was measured from the onset of theletter string. The interval between subject's re­sponse and the onset of the next experimentaltrial was 2000 ms.

Design. For Experiment 1, three test orders eachcontaining 114 items were created. Half of theitems were words and half pseudowords. Fifty­four items were primes and fifty-four items weretargets. Words and pseudowords were presentedequally often as primes and as targets. Inaddition, there were six filler items introduced tomaintain the requisite lags. Each test orderincluded nine tokens of each of the three types ofprimes (viz., identity, inflectional, derivational)and across test orders, each target was precededby all three types of prime. Subjects wererandomly assigned to one of the three test ordersand a practice list of ten items preceded eachexperimental list.

Results and Discussion

Errors and extreme response times (greaterthan 2 SD or less than -2 SD from each subject'smean) were eliminated from all reaction timeanalyses. Accordingly, about 4% of all responses

were eliminated. Table 3 summarizes the meanrecognition times over subjects for target wordsand pseudowords preceded by identity, inflectionaland derivational primes and for the firstpresentation of those same words as a prime.

Analyses of variance were performed on targetlatencies for words and pseudowords usingsubjects (FlJ and items (F2) as random variables.The analysis included the first presentation of thetarget as the no prime condition, targets precededby themselves as the identity condition, targetspreceded by an inflected form and targetspreceded by a derived form. For words, there wasa significant effect of type of prime on targetlatencies [Fl(3,78) = 15.66 MSe = 641, p < .001;F2(3,78) = 8.13, MSe = 1597, p < .001] but theeffect of prime with the error measure missedsignificance [Fl(3,78) = 2.66, MSe = 46, P < .054;F 2 (3,78) = 0.88]. The results of plannedcomparisons on decision latencies indicated thatfacilitation from derivationally-related primes wassignificantly weaker than facilitation frominflectionally-related primes [Fl(1,26) = 6.58, MSe= 383, p < .016; F2(1,26) = 3.1, MSe = 4950, p <.08] and significantly different from the no primecondition [Fl(1,26) = 6.14, MSe = 804, P < .02;F2(1,26) = 4.05, MSe = 6468, p < .05]. Targetlatencies following derivational primes tend to beslower than target latencies following inflectionalprimes. For pseudoword targets, the effect of typeof prime was significant for neither reaction timesnor errors.

Table 3. Mean lexical decision latencies and errors for targets on their first presentation, or when preceded byidentity, injlectionally- and derivationally-related primes in Experiment 1.

NONE

PEVA

RT ERRWORDS

569 (8)FACILITATION

TYPE OF PRIME

IDENTITY INFLECTIONAL DERIVATIONAL

PEVA PEVAM PEVACPEVA PEVA PEVA

RT ERR RT ERR RT ERR

524 (8) 536 (9) 550 (12)45 (0) 33 (- I) 19 (-4)

PSEUDOWORDS664 (8)

FACILITATION666-2

(8)(0)

663I

(9)(-1)

64420

(7)(I)

Beyond Orthography and Phonology: Differences between Inflections and Derivations 75

Facilitation was assessed by examining differ­ences in latencies (and errors) to targets precededby a prime and to first presentations of targets.Consequently, prime presentations necessarilyoccurred earlier in the list than did targets.Because there is evidence that latencies get fasteras subjects proceed through the list, and becausefacilitation following derivations tended to beweak relative to facilitation following inflections,it is important to determine whether or notfacilitation from derivations was correlated withserial position of the prime. In Experiment 1, thecorrelation between serial position of the primeand the difference between latencies for firstpresentations and latencies following derivationalprimes was r = -.048. Therefore, the magnitude offacilitation was not distorted by the no primebaseline. Note, however, that any potentialbaseline problem is not relevant when comparingfacilitation following inflectional and derivationalprimes because position of the target (and thetarget item) were identical.

The present experiment with Serbian materialsreplicates previous findings in the same language(Feldman & Moskovljevic, 1987; Feldman &Fowler, 1987) as well as other languages (Fowleret aI., 1985; Bentin & Feldman, 1990).Specifically, relative to a no prime condition,morphologically-related word forms facilitatedeach other at lags that average 10 interveningitems but pseudoword analogs did not. Insummary, facilitation was observed in the patternof target latencies for all types of morphologically­related primes and the amount of facilitationvaried by type of prime. It is interesting to notethat although identity and inflectional primestended to yield statistically equivalent facilitationin earlier studies (e.g., Feldman & Fowler, 1987),under some circumstances derivations have beenobserved to produce facilitation that wassignificantly reduced relative to the identitycondition (Feldman & Moskovljevic, 1987; Exp. 2;Schriefers et al., 1992). Nevertheless, no publishedexperiment with Serbian materials included, oreven permitted, a direct comparison betweeninflectional and derivational types of primes.

The present study extends previous repetitionpriming results in Serbian by contrasting twotypes of morphological formations while tightlycontrolling their similarity. With phonemic andorthographic overlap equated between inflection­ally- and derivationally-related prime forms, therewas evidence of enhanced facilitation for targetsfollowing inflectionally-related primes relative to

derivationally-related primes. This distinction canbe represented in the lexicon. Perhaps the linkagebetween whole word forms that share a base mor­pheme is stronger (or the internal coherence oftheir constituents is weaker) for inflectionally-re­lated forms than for derivationally-related forms.

Unfortunately, the composition of experimentalmaterials in the present experiment is consistentwith another account. In Experiment 1, allderivational formations differed in word classfrom their morphologically-related target whereasall inflectional formations (necessarily) preservedword class. Specifically, translations of agentivessuch as singer primed verb targets such as hesings and verb forms such as he sails primedderived noun targets such as boat. While suchchanges are, in fact, characteristic of derivationalprocesses in all languages, they make anunequivocal interpretation ofthe contrast betweeninflectional and derivational pairs more difficult.It is important to note that although, in therepetition priming task, no effects of semanticsimilarity have been reported with visuallypresented relatives and lags of 10 items (Bentin &Feldman, 1990; Napps, 1989) or with auditorilypresented materials presented successively(Emmorey, 1989; but see Radeau, 1983;Slowiaczek, 1994), it is nevertheless possible thatderivations are semantically more distinct fromtheir targets than are inflections and thatsemantic similarity can, under somecircumstances, contribute to the pattern offacilitation. Accordingly, Experiment 2 entailed acomparison of the pattern of facilitation withSerbian inflections and derivations that a)consistently preserved word class, b) weresemantically quite close in meaning and c) wereconstructed with attention to their orthographicsimilarity to the target.

EXPERIMENT 2Inflectional affixes tend to alter the meaning of

the base morpheme in predictable ways (Aronoff,1976) whereas the effect of derivational affixes isless consistent. Consequently, inflectionalformations tend to be similar in form and meaningto other forms that share a base morpheme (andstem) and differ with respect to inflectional affixwhereas derivational formations tend to differ inform and meaning from other forms that share abase morpheme and differ with respect toderivational affixes (and stem). In Serbian it ispossible to identify inflection-derivation pairs withonly minimal differences in meaning and form.

76 Feldman

One such contrast entails verbs that differ withrespect to aspect. Generally stated, aspect reflectsthe temporal properties of the verb. These includeinceptive forms of stative verbs and iterativeforms of verbs that describe discrete events.

All the experimental materials for Experiment 2were verb forms. Targets were preceded byidentity, inflectionally- or derivationally-relatedprimes. Inflected primes were other forms of thesame verbs that differed in person. Derivedprimes were forms of lexically-distinct verbscomposed from the same base morpheme thatdiffered in aspect and person from the targetword. The manipulation on derivation alternatedperfective and imperfective forms. Semantically,this distinction is relatively minor entailingcontrasts between semantic notions such ascompleted and progressive actions in HE SATDOWN and HE WAS SITTING or between eventsand states such as HE RECOGNIZES and HEKNOWS. (Note that progressivity isgrammaticalized in English whereas stativity islexicalized (Lyons, 1977)). It is important tounderscore, however, that in Serbian, unlikeEnglish, the perfective and imperfective forms ofthe verbs included in the present study areconsidered distinct lexical entries.

It is relevant to note that there is no consensusabout the morphological status of aspectualformations either across languages or acrosstheorists (compare Anderson, 1982 with Bybee,1985). In the present study, it is assumed thataspect is a derivational process. It is restricted byits meaning to a particular semantic class ofSerbian verbs (Partridge, 1964 in Bybee, 1985).Moreover, it was also always the case that twodistinct verbal entries existed in the dictionary.Note however, that these formations do notchange word class as is typical of derivationalformations. In Experiment 2, aspect was markedby the addition of either a prefix or an infix to thebase morpheme. Consequently, forms related byinflection shared both a base morpheme and astem (base morpheme plus derivational affix)whereas forms related by derivation shared a basemorpheme but differed with respect to stem. Forexample, the words OBARIM and BARIM areboth formed from the base BAR and theinflectional affix 1M. They differ with respect tothe presence of a prefix which is part of the stem.Accordingly, the stems are OBAR and BAR,respectively.

The outcome of Experiment 1 indicated thatwith controls for orthographic overlap, the lexical

representation of morphological relatedness by in­flection and derivation differed. If this outcomereflects type of morphological relation as distin­guished from effects of preserving or altering wordclass, then consistent with the results ofExperiment 1, in Experiment 2 facilitation fromprimes that are inflectionally-related to their tar­gets should be greater than from primes that arederivationally-related. Of course, the absence of adifference is ambiguous. It could indicate that theeffect observed in Experiment 1 does reflectchanges in word class between prime and target.Alternatively, it could indicate that aspect inSerbian is not a derivational relationship butrather, a less general inflectional relationship.

If, as sometimes claimed (e.g., Taft & Forster,1975; Bergman, Hudson, & Eling, 1988), prefixesbut not other affixes are stripped from the basebefore lexical access is attempted, then thepattern for prefixed primes should differ from thatof infixed primes. Alternatively, if activation inrepetition priming is based on the stem (base plusderivational affix) rather than the base alone assometimes claimed (Burani & Laudanna, 1992)then infixed forms should show a pattern similarto that of prefixed forms. Because inflectionsshared both base and stem whereas derivationsshared a base morpheme only, derivations shouldproduce weaker facilitation than inflectionswhenever the stem and base morpheme differ.

In summary, as in Experiment 1, patterns offacilitation for primes related by inflection and byderivation are examined in Experiment 2. Bothinflectional and derivational primes alwaysincluded the full base morpheme and theirinflectional affixes were matched for letter length.In contrast to Experiment 1, in which ortho­graphic and phonological overlap was perfectlymatched but word class differed betweenderivational but not inflectional primes, inExperiment 2, the presence of a prefix or an infIxrendered inflectional primes more similar to theirtargets than derivational primes (that included noaffix) but all were verb forms.

Methods

Subjects. Thirty-six first year students similarin characteristics to those of Experiment 1,participated in Experiment 2. None hadparticipated in Experiment 1.

Stimulus materials. Forty-eight Serbian wordtriples were selected. Each included three verbforms: a target verb, a prime that wasinflectionally related and a prime that was

Beyond Ortlwgraphy and Phonology: Differences between Inflections and Derivations 77

derivationally related to the target. Targetsconsisted of present-tense verb forms in the thirdperson plural. Each was composed of a basemorpheme and an aspectual affix. Inflected formswere first person singular of those same verbs.Derived forms were first person singular ofdifferent verbs formed from the same basemorpheme without an aspectual affix. (Theseforms are designated as derived because they arerelated by derivation to the target.) Inflectionaland derivational primes were always presented inthe same person and number. Items are listed inAppendix l.

Typically, the target and inflected prime wereimperfective forms and the derived prime wasperfective. They were all formed from the samebase morpheme but, because of the addition of anaffix, they differed with respect to their stems.Derivation was defined relative to the targetrather than on an isolated word. Structurally, allmembers of a triple were composed of the samebase morpheme but differed with respect to thepresence of an affix, either prefix or infix. Forexample, perfective forms of the base morphemeBAR, meaning cook, included BARIM, BARIS,BAR!.. .BARE whereas imperfective forms such asOBARIM, OBARIS, OBAR!...BARE include theprefix O. Other than the prefix or infix, theorthographic and phonemic overlap of primes andtheir morphologically-related targets was perfectlycontrolled by selecting third person plural formsending in E as targets and necessarily as identityprimes (e.g., OBARE), forms ending in 1M (e.g.,OBARIM) as inflectionally-related primes andverbs differing in aspect (e.g., BARIM) as thederivationally-related primes. Perfective forms ofthe base morpheme GUR, meaning push, includeGURAM, GURAS, GURA...GURAJU whereasimperfective forms such as GURNEM, GURNES,GURNE. ..GURNU include the infix N. For infixedrelatives, targets and identity primes ending in U(e.g., GURNU), inflectional primes ending in EM(e.g., GURNEM) and derivational primes endingin AM (e.g., GURAM) were presented whereinflectional and derivational primes were alwaysin the same person and number. In summary, theorthographic and phonological similarity of bothinflectional and derivational primes was matchedto the target so that both included the full basemorpheme although due to the prefix or infix,overall overlap for inflectional forms was slightlygreater than that for derivational forms.

Pseudoword triples were created by substitutingvowels or consonants within other base mor-

phemes in order to create meaningless bases thatwere orthographically legal. To these, real in­flected affixes were appended in order to createsets of pseudowords that differed only with re­spect to affix. The distribution of pseudoword af­fixes was matched to those for words. Pseudowordtargets were preceded by identity, inflectionally­(or derivationally-) related pseudoword primes orby a real word prime. The value in including aword prime with a pseudoword target was to ex­amine whether facilitation in repetition primingextends to strings without lexical status.

Three test orders were created. Each contained200 items and included equal numbers of wordand pseudoword targets preceded an average often items earlier in the list by a morphologically­related prime. In each test order, eight tokens foreach· of the three types of prime were presented.Across the three test orders, each word orpseudoword target was preceded by all three typesof morphologically related primes. In contrast toprevious repetition priming studies, herepseudoword targets were preceded 33% of the timeby a word prime formed from the same basemorpheme.

Procedure. The procedure was identical to thatof Experiment 1.

Results and Discussion

Mean decision latencies (for responses less than2 SD or greater than -2 SD from each subject'smean) and error rates in Experiment 2 are sum­marized in Table 3. Errors and outliers accountedfor approximately 6% of all responses. An analysisof lexical decision latencies for words revealed asignificant effect of type of prime [F 1(3,105) =45.06, MSe = 1726, p < .001; F2(3,138) = 27.73,MSe =2865, p < .0001]. Effects of affix type weresignificant [F1(l,35) = 46.51, MS e = 1321, p <.001; F2(l,46) == 5.42, MSe = 12712, P < .02]. Theinteraction of type and overlap was significant inthe subjects but not in the items analysis[F1 (3,105) =4.59, MSe =2101, P < .005). Plannedcomparisons indicated that inflectional primesproduced faster target latencies than did deriva­tional primes both for prefixed targets [F1(1,35) =4.61, MSe == 2596, p < .04; F2(l,23) == 3.88, MSe =12096, P < .053) and for infixed targets, [F1(1,35)=8.79, MSe == 1759 p < .005; F2(1,23) =6.79, MSe= 17749, P < .01). Target latencies followingderivational primes were significantly faster thanfirst presentations latencies in the prefixed[F1(l,35) =16.23, MSe =2660, P < .001; F2(l,23) =11.26, MS e = 35101, P < .001) but not in the

78 Feldman

infixed condition [Fl(1,35) =1.96, MSe =2883,p <.17; F2(l,23) = 4.74, MS e = 12384, p < .03].Correlations between serial position of the primeand the magnitude of facilitation for targets(following derivational primes relative to the noprime condition) were not significant [r = -.154]therefore it is unlikely that the magnitude of facil­itation was significantly overestimated by the noprime baseline.

The analysis of error scores revealed asignificant effect of type of prime [F 1(3,105) =17.01, MSe = 85,p <.001; F2(3, 138) = 5.84, MSe =1.8, P < .oon Effect of affix type were significantin the subjects analysis only [Fl(1,35) = 32.89,MSe = 79, p < .001; F2(1,46) = 2.99, MSe = 8.2, p <.09]. The interaction of type and overlap wassignificant in the analysis by subjects only[Fl (3,105) = 3.30, MSe = 85, p < .023; F2(3,138) =1.45, MSe = 1.8, P < .23]. Mean decision latenciesand errors for pseudowords are included in Table

4. They indicate no effect of lexical status of theprime on pseudoword target latencies.

The magnitude of facilitation was significantlygreater for prime-target pairs related by inflectionthan for pairs related by derivation. Thus,facilitation in repetition priming was once againsensitive to type of morphological relation whenword class and formal properties of the affixeswere controlled. Assuming an appropriatebaseline, derivational primes produced significantfacilitation relative to the no prime condition forprefixes although not (statistically) for infixes.More importantly, derivational primes producedsignificantly reduced facilitation relative to theinflectional primes. The present results replicatethe general pattern of morphological relatednessin the repetition priming task including thedifferent patterns of facilitation followinginflectional and derivational primes that wasobserved in Experiment 1.

Table 4. Mean lexical decision latencies (and errors) for targets on their first presentation, or when preceded byidentity, inflectionally- and derivationally-related primes in Experiment 2.

NONE

PREFIXED WORDS

OBARE

RT ERR675 (9)

FACILITATION

TYPE OF PRIME

IDENTITY INFLECTIONAL DERIVATIONAL

OBARE OBARIM BARIMOBARE OBARE OBARE

RT ERR RT ERR RT ERR57 (4) 600 (5) 626 (8)102 (5) 75 (4) 49 (I)

INFIXED WORDS

GURNU

675 (21)FACILITATION

PSEUDOWORDS678 (7)

FACILITATION

GURNU GURNEM GURAMGURNU GURNU GURNU

629 (8) 628 (7) 658 (14)46 (13) 47 (14) 17 (7)

672 (5) 668 (6) 665 (6)6 (2) 10 (I) 13 (I)

Beyond Orthography and Phonology: Differences between Inf!ectiotl$ and Derivations 79

EXPERIMENT 3

The natural confound between inflections andderivations noted above was eliminated in thethird experiment. Specifically, forms related byinflection tend to be more similar in terms oforthography and phonology than forms related byderivation. This is because derived forms share abase morpheme but differ with respect toderivational affix and therefore stem whereasinflected forms share both their base morphemeand their stem. The materials for Experiment 3consisted of another set of verbs related by aspect.In each instance, two entries were formed aroundthe same base morpheme; however, they differedwith respect to the set of inflectional affIXes eachrequired. That is, many items shared both theirbase morpheme and their stem and they differedonly with respect to their thematic vowel (Scalise,1984). If differences between facilitation byinflection and derivation are observed with thematerials of Experiment 3, they cannot beattributed to orthographic overlap or to repetitionof the base morpheme but not the stem.

MethodsSubjects. Thirty-six first year students similar

in characteristics to those of the first twoexperiments participated in Experiment 3. Nonehad participated in Experiments 1 or 2.

Stimulus materials. Twenty-six word triples inSerbian were selected. Each included three verbforms: a target verb, a prime that wasinflectionally-related, and a prime that wasderivationally-related to the target. Targetsconsisted of present tense verb forms in the firstor third person plural. Inflected forms were firstperson singular of those same verbs. Derivedforms were first person singular of different verbsformed from the same base morpheme thatdiffered in the temporal qualities of the actionthey conveyed. Inflectional and derivationalprimes were always presented in the same personand number. Items are listed in Appendix 1.

The orthographic and phonemic overlap ofprimes and their morphologically-related targetswas carefully controlled and two patterns wereincluded. Structurally, all members of a triple inthe matched pattern were verbs constructed fromthe same base morpheme and stem but theydiffered with respect to the (thematic) vowelaround which the inflectional affix was formed.For example, in one pattern, perfective forms ofthe base morpheme NOS meaning carry aregenerally formed around the vowel I such as

NOSIM, NOSIS NOSI...(but) NOSE whereasimperfective forms are generally formed around Asuch as NOSAM, NOSAS NOSA...NOSAJU.Forms ending in E served as targets andnecessarily as identity primes (e.g., NOSE), formsending in 1M or EM (e.g., NOSIM) served asinflectionally-related primes, and verbs differingin aspect (e.g., NOSAM) served as thederivationally-related primes. Thirteen such pairswere selected. Thirteen pairs followed a secondunmatched pattern in which the inflectionally­related prime overlapped by one or two lettersmore than did the derivationally-related prime.For example, forms ending in AMO (e.g.,NAZIVAMO) served as targets and as identityprimes, forms ending in AM (e.g., NAZIVAM)served as inflectional primes and forms ending inEM (e.g., NAZOVEM) served as derivationalprimes. Note that for these triples, inflectionalprimes preserved both the I and A vowels of thetarget whereas the derivational primes did not.

Pseudoword triples were created by substitutingvowels or consonants within other base mor­phemes in order to create meaningless bases thatwere orthographically legal. To these, real in­flected affixes were appended in order to createsets of pseudowords that differed only with re­spect to affix. As with words, pseudoword targetswere preceded by identity, inflected and deriva­tionally-related primes. Inflected and derivedforms consisted of a nonsense base morphemewith a legal affix. The distribution of pseudowordaffixes was matched to those for words.

In summary, as in Experiments 1 and 2, bothinflectional and derivational primes always in­cluded the full base form and their affixes werematched for letter length. In contrast toExperiment 1, in which orthographic and phono­logical overlap was perfectly matched but wordclass differed between prime and target, deriva­tional as well as inflectional primes inExperiments 2 and 3 preserved word class of thetarget. In contrast to Experiment 2, in which or­thographic and phonological overlap between in­flectional and derivations primes was not perfectlymatched, in Experiment 3 matched and mis­matched overlap was systematically manipulated.In the condition in which orthographic and phono­logical overlap were mismatched, inflectionalprimes were more similar to their targets thanwere derivational primes because the inflectionalprimes (i.e., those ending in AM) preserved thevowels of the target form whereas none of thederivational primes did. In the condition in whichorthographic and phonological overlap were

80 Feldman

matched, inflectional and derivational primeswere equally similar to their targets.

Three test orders were created. Each contained114 items and included equal numbers of wordand pseudoword targets preceded an average often items earlier in the list by a morphologically­related prime. In each test order, four or fivetokens of each of the three types of matched andunmatched primes were presented. Across thethree test orders, each word or pseudoword waspreceded by all three types of morphologicallyrelated primes.

Procedure. The procedures were identical tothose of Experiments 1 and 2.

Results and DiscussionMean decision latencies (for responses less than

2 SD or greater than -2 SD from each subject'smean) and error rates in Experiment 3 are sum­marized in Table 5. Accordingly, approximately5% of responses were eliminated. An analysis oflexical decision latencies for words revealed asignificant effect of type of prime (F1(3,105) =23.40, MSe = 2598,p < .001; F2(3,72) = 11.17, MSe= 2396, P < .0001]. Effects of orthographic andphonological overlap (match) were significantin the analysis by subjects (F1(1,35) = 21.14, MSe= 3607, p < .001] but not in the analysis byitems (F2(l,24) = 2.54; MSe = 86181; p < .12].Importantly, the interaction of type and match didnot approach significance in either analysis.

Nevertheless, comparisons between inflectionsand derivations were examined separately formatched and mismatched items. Target latenciesfollowing orthographically-matched inflectionalprimes were faster than target latencies followingderivational primes (F1(1,35) = 6.26, MSe =2197,p < .014; F2(l,12) = 4.73, MSe = 1150, P < .05].However this pattern missed significance formismatched primes [Fl(1,35) = 2.06, MSe = 2197,p < .15; F2(1,12) =1.23]. Finally, latencies for tar­gets following derivational primes were signifi­cantly faster than for first presentations whenoverlap was mismatched (Fl(l,105) =28.58 MSe =2197, p < .001; F2(1,12) = 8.81, MSe = 2814, p <.01] and in the subjects analysis, when overlapwas matched [F1(l,105) =6.97, MSe =2197, p <.01; F2(l,12) = 1.62, MSe = 4240, P < .23]. Thecorrelation between serial position of the primeand magnitude of facilitation for targets followingderivational primes relative to the no prime condi­tion was positive and nonsignificant [r = .11]. Inconjunction with previous experiments, this fmd­ing supports the appropriateness of the no primebaseline.

The analysis of error scores revealed asignificant effect of type of prime (F1(3,105) =4.60, MSe = 78,p < .005; F2(3,75) = 3.74, MSe =.61, p < .02]. For pseudowords, neither decisionlatencies nor accuracy revealed an effect of prime.Mean decision latencies and errors forpseudowords are included in Table 5.

Table 5. Mean decision and naming latencies (and errors) for targets on their first presentation, preceded by identity,injlectionally- and derivationally-related primes in Experiment 3.

TYPE OF PRIME

NONE IDENTITY INFLECTIONAL DERIVATIONAL

NOSE NOSIM NOSAMNOSE NOSE NOSE NOSE

RT ERR RT ERR RT ERR RT ERR

MATCHED WORDS630 (10) 577 (2) 573 (3) 600 (4)

FACILITATION 53 (8) 57 (7) 30 (6)

MISMATCHED WORDS676 (4) 617 2) 601 (3) 617 (2)

FACILITATION 59 (2) 75 (I) 59 (2)

NONE IDENTITY PSEUDOWORD WORD

PSEUDOWORDS674 (8) 671 (9) 665 (8) 663 (9)

FACILITATION 3 (-1 ) 9 (0) 11 (-1 )

Beyond Orthography and Phonology: Differences between Inflections and Derivations 81

The pattern of target latencies indicated thatidentical repetition and inflectional primes bothproduced significant and equivalent facilitation.Matched derivational primes produced signifi­cantly reduced facilitation relative to the inflec­tional condition and significant facilitation rela­tive to the no prime condition. The present resultsreplicate previously observed effects of morpholog­ical relatedness in the repetition priming task andextend those results by revealing a significantdistinction between the effect on targets of inflec­tional and derivational primes that share boththeir stem and their base morpheme.

Effects of orthographic and phonologic overlapbetween prime and target on the pattern offacilitation across prime types were systematicallyexamined because inflectional relatives tend to bemore similar than derivational relatives. Matchedand mismatched overlap never interacted withtype of prime although planned comparisonsindicated that the difference between targetspreceded by inflections and by derivations wasstatistically more reliable for matched pairs thanfor mismatched pairs. This pattern is notanticipated if differences in magnitude offacilitation between inflectionally- andderivationally-related primes reflects extent oforthographic overlap with the target. Moreover,because the semantic differences betweeninflectional and derivational relatives was small,it cannot readily be attributed to greater semanticoverlap for inflections relative to derivations.

The materials selected for Experiment 3 areunique in that for many items inflectional andderivational relatives were both formed aroundthe same base morpheme and differed only withrespect to the vowel from which the inflectionalaffixes were formed. Because no derivational affixwas introduced, relatives shared both their basemorpheme and their stem. Thus, the results ofExperiment 3 indicate that the difference betweeninflections and derivations in the repetitionpriming task cannot be attributed to greaterfacilitation for stems than for bases.

In the present study, morphological relativesproduced facilitation to target decision latencies inthe repetition priming task but the interpretationof these lexical decision results is not straightfor­ward. It has been suggested that the results ob­tained with this task may reflect binary decisionprocesses that are specific to this task (Balota &Chumbley, 1984) or alternatively that expectancyand post-lexical mechanisms are involved as wellas lexical activation (Neely, 1991). Obviously, it isimportant to provide converging evidence from

other experimental tasks for the contribution ofmorphology to word recognition. In the three re­maining experiments, morphological effects areinvestigated in a new experimental task.

EXPERIMENT 4The outcome of the first three experiments using

the repetition priming paradigm suggestedprocessing differences between inflectional andderivational formations. Another source ofevidence for the role of morphology in lexicalprocessing derives from the pattern of errorsobserved in the production of spontaneous speech(Cutler, 1980; Dell, 1986; Fromkin, 1973; Garrett,1980, 1982; Stemberger, 1985). One prevalenttype of error entails the reordering of morphemicelements so that the stem or affix of a wordmigrates from the intended word to another site.The pattern for stems and affixes tend to differ(Garrett, 1976). Although there are confoundedprosodic differences, this observation has beeninterpreted as evidence that the base morphemeand inflectional components of a morphologicallycomplex word are separable. Moreover, when wordfinal elements are misordered, those that aremorphemic are. more likely to shift than arephonemically equivalent but nonmorphemicsegments (Sternberger, 1984) and this differencecannot be attributed to frequency differences(Dell, 1990). Finally, inflectional affixes are morelikely to migrate than are derivational affixes(Garrett, 1982). Collectively, these observationsindicate that the constituent structure ofmorphologically complex words is available to theproduction mechanism and are consistent with theclaim that inflectional and derivational forms maybe treated differently (see also Badecker &Caramazza, 1989; Miceli & Caramazza, 1988).

In Experiments 4, 5 and 6, an experimental taskinspired by the pattern of speech errors inspontaneous speech was developed in order toprovide converging evidence for the claim that themorphological constituents of a word can beavailable to a processing mechanism. The segmentshifting task entails deliberately shifting segmentsfrom a source word to a target word and rapidlynaming the product aloud. The experimentalmanipulation exploits the fact that the morphemicstructure of many words is not wholly transparentand that the same sequence of letters (e.g., ER)can function morphemically in one context (e.g.,DRUMMER) and nonmorphemically in another(e.g., SUMMER). Letter sequences which aremorphological in the context of some source wordsand nonmorphological in the context of others

82 Feldman

were shifted onto the same target word.Pronunciation latencies for the same targetsformed from morphemic and nonmorphemicsource words are compared. The segment shiftingprocedure used in Experiment 4 is depicted inFigure L

MethodsSubjects. Twenty-six students at the University

of Belgrade participated in the experiment in par­tial fulfillment of the requirements for anIntroductory Psychology course. All had experi­ence with reaction time studies but none had par­ticipated in previous experiments in this study.The data from nine addition subjects were elimi­nated because their error rates exceeded 20%.

Stimulus materials. Sixteen pairs of Serbianwords were constructed for each of two morpholog­ical types and these constituted the source wordsfor Experiment 4. Each pair of source words in­cluded a morphologically complex word composedof a base morpheme and a morphological suffIxand a morphologically simple control word. Thecontrol word ended with the same sequence of let­ters that functioned morphemically in its pair.Morphemic and nonmorphemic endings were con­trolled for phonemic and syllabic structure ( Tyler& Nagy, 1989). The Serbian analog of inflected

•••

words such as WINNING and matched morpho­logically simple words such as INNING consti­tuted an inflectional type pair. For example, in­flectional source words consisted of masculine sin­gular instrumentals such as PRESOM, whichmeans press, and nonmorphemic controls con­sisted of morphologically simple words endingwith the same sequence of letters without a mor­phemic function such as PRELOM, which meansfracture, in nominative case. Note that the OMsequence appeared on morphemic and non­morphemic source words of equal length and thatsource and target words were semantically unre­lated.

A second morphological type consisted of homo­graphic morpheme affixes. Pairs of source wordsconsisted of morphologically complex source wordswith morphological affixes that were compatiblewith the target word (same syntactic category andgender) and morphologically complex sourcewords that were not. That is, the Serbian analogof nominal or verbal S was shifted to another wordof the same (consistent) or a different(inconsistent) word class. For example, the nomi­native plural I from CEVI, meaning pipes, or thethird person singular I from CEDI, meaning hewrings was shifted to the target word RAD in or­der to form the word RADI meaning he works.

200ms

•••

PRESOM

PRESOM

TOP

•••

"TOPOM"

Figure 1. The original segment shifting procedure.

PRELOM

PRELOM

TOP

•••

"TOPOM"

500ms

1500 msclock starts

clock stops

Beyond Orthography and Phonology: Differmces between Inflections and Derivations 83

Note that in both source words the I ismorphemic. What differs for homographic affixesis the consistency or inconsistency of the syntacticcategory of the source word and the target word.Source words for the segment shiftingexperiments are described in Table 1 and arelisted in Appendix l.

Procedure. Following the presentation of a fixa­tion point for 200 ms, a source word with a portionunderlined appeared for 500 ms. Immediately af­terwards, the target word appeared below thesource word and a clock started. Both words re­mained visible for 1500 ms. A blank field followedthe display and lasted for 2000 ms.

Subjects were instructed to segment and shiftthe designated segment from a source word onto atarget word and to name the new result aloud asrapidly as possible. For example, the OM of thesource word PRESOM was underlined andsubjects were instructed to shift that sequence ofletters to the target word TOP in nominative casein order to produce TOPOM, which means cannonin instrumental case. Onset to vocalization wasmeasured and errors were recorded. A sequence of13 practice items preceded the experimental listwhich included eight tokens each in themorphemic-nonmorphemic and morphemic­incompatible conditions.

Results and Discussion

Means for Experiment 4 are summarized inTable 6. All correct scores less extreme than 3 SDfrom the mean for each subject were included inan analysis of variance (approximately 14% of all

scores were eliminated) and revealed a significanteffect ofmorphological type (inflection/homograph)[Fl(1,25) =7.53, MSe =1278,p < .01; F2(1,30) =4.85, MSe =1222, p < .04]. The interaction ofmorphological status and morphological type wassignificant in the analysis by subjects [Fl(l,25) =11.89 MS e = 437, P < .003], but was onlymarginally significant in the analysis by items[F2(1,30) =2.96, MSe =1550, P < .10]. The effectof morphological status was not significant. Aplanned comparison between morphological andnonmorphological segments was significant for theinflectional type of affix [Fl(1,25) = 10.44, MSe =585, p < .001; F2(l,15) =3.25, MSe =1115, p <.09], but not for the homographic type. No effectswere significant for errors.

For the homographic morpheme type, where theconsistency or inconsistency of the syntacticcategory of the source word and the target wordwas varied, no significant effects of consistency (-6ms) were observed. Shifting rates for "I" segmentsderived from verbal and nominal source wordswere statistically equivalent.

The outcome of Experiment 4 was thatmorphological segments were shifted from sourcewords to targets words more rapidly than theirphonologically-matched controls, but thatsyntactically congruent and incongruentmorphological affIxes did not differ. This resultsuggests that the component structure ofmorphologically complex words is available to thelanguage processing mechanism and thatmorphemes as contrasted with phonemes are therelevant units.

Table 6. Segment shifting times and errors for morphological affixes and their nonmorphological or incompatiblecontrols in Experiment 4.

MORPHOLOGICAL TYPE

INFLECTION

HOMOGRAPH

*incompatible morpheme

SHIFTED SEGMENTNONMORPHEMIC

MORPHEMIC or INCOMPATIBLE

PRESQM PRELOM

582 60411% 12%

CEDl CEVl*

577 57113% 13%

DIFFERENCE

22I

-6o

84 Feldman

The observed effect could reflect the lexicalrepresentation of morphological structure such asthe process of segmenting, from the source, asequence of letters that constitutes a morphemiccomponent and of affixing that sequence to thetarget. That is, segmentation of morphologicalunits could underlie the effect. Similarly, it ispossible that the internal structure of wordscomposed of multiple morphemes differ in theircoherence relative to morphologically simplewords. It should be pointed out that coherence,defined in terms of sequential probabilitiesbetween letters, is a not plausible account,because the composition of morphemic andnonmorphemic sequences was well matched inthis study (see also Rapp, 1992). Nevertheless, therepresentation of morphologically complex wordsmay encompass their sub-word units, andmorphological coherence may be relevant. Insummary, morphemic affixes were more easilysegmented from a source word than werenomnorphemic controls presumably because theavailability of sublexical morphologicalcomponents determined morphological coherence.In effect, the imposed shifting of letter sequencesfrom morphologically-simple words is difficultbecause it is arbitrary, whereas the shifting ofletter sequences from morphologically-complexwords is relatively easy because it is principledand follows morphological structure.

EXPERIMENT 5The purpose of Experiment 5 was to replicate

the results of the previous experiment and toallow a new comparison between inflectional andderivational morphological types. In an attempt toincrease the magnitude of the effects observed inthe previous experiments, the segment of thesource word that subjects had to shift was notspecified when the source word appeared. Instead,it was indicated 750 ms later and wassimultaneous with the appearance of the targetword. The comparison between inflectional andderivational affixes was again examined. If theconstituent structure of inflections is moretransparent than that of derivations, then effectsshould be more systematic for inflections. Finally,if the segment shifting effect is sensitive tostrategies imposed by the subject and if subjectsanticipate segmenting morphological affixes, thenlimiting preparation time before the onset of thetarget may increase the magnitude of the effectbecause the component structure of themorphemic source word, but not its control, will beavailable before it is visually specified.

MethodsSubjects. Twenty-four students at the

University of Belgrade participated in Experiment5 in partial fulfillment of the requirements for anIntroductory Psychology course. All hadexperience with reaction time studies but nonehad participated in previous experiments in thisstudy. No subject's data were eliminated becauseof error rates in excess of 20%.

Stimulus materials. Eighteen pairs of Serbianwords were constructed for each of twomorphological types and these constituted thesource words. Each pair of source words includeda morphologicallY complex word (composed of abase morpheme and a morphological suffix) and amorphologically simple control word. The controlword ended with the same sequence of letters thatfunctioned morphemically in its pair. Inflectionaltype source words consisted of first personsingular verbs ending in EM such as KRADEM,which means I steal, and nonmorphemic controlsconsisted of morphologically simple words endingwith the same sequence of letters without amorphemic function such as BADEM, whichmeans almond, in the nominative case. Note thatthe EM sequence appeared on morphemic andnonmorphemic source words whose length differedby no more than one letter.

A second morphological type consisted of agen­tives which are derivational morphemes. Thesepairs of source words consisted of morphologicallycomplex source words and morphologically simplesource words ending in the sequence AR or AC.For example, derivaJional source words consistedof agents such as CUVAR, meaning guard, innominative case and nonmorphological controlsconsisted of morphologically simple words such asSTVAR, meaning thing, in nominative. In bothcases, the AR was shifted to the target word RADin order to form the word RADAR, meaningworker. Subjects were instructed to add theshIfted segment from the source word to the targetword and to name it aloud.

Procedure. In an attempt to increase the size ofthe effect observed in the previous experiment, thepresentation conditions of Experiment 5 weremodified. The segment of the source word thatsubjects had to shift was not indicated at the sametime that the source word appeared. That is, thesource word first appeared alone and withoutunderlining. After 750 ms, the target wordappeared below the source word, the segment ofthe source word that subjects had to shift wasunderlined, and a clock started. A blank fieldfollowed the display and lasted for 2000 ms.

Beyond Orthography and Phonology: Differences between Inflections and Derivations 85

Subjects were instructed to segment and shiftthe designated segment from a source word onto atarget word and to name the new result aloud asrapidly as possible. For example, the EM of thesource word JEDEM was underlined and subjectswere instructed to shift that sequence of letters tothe target word KUJE in order to produceKUJEM, which means I hammer. Onset tovocalization was measured and errors wererecorded. A sequence of 13 practice items precededthe experimental list which included nine tokensof morphemic and nonmorphemic source words inthe inflectional and derivational conditions.

Results and Discussion

An analysis of variance on correct latencies lessextreme than 3 SD from the mean for each subject(approximately 6% of all responses were elimi­nated) revealed significant effects of morphologicaltype (inflection/derivation) [Fl(I,23) = 13.78, MSe= 2487, p < .002] and morphological status(morpheme/nonmorpheme) [Fl(I,23) = 9.1, MSe =913, P < .007] by the subjects' analysis but onlymorphological type approached significance by theitems' analysis [F2(1,34) =2.75, MSe =5139, p <.11]. With the error measure, neither the main ef­fect nor the interaction of affix by type approachedsignificance. Means are summarized in Table 7.

Numerical differences for agentive derivationalaffixes were reduced and in the opposite directionrelative to those of inflectional affixes althoughthere was no significant interaction. Nevertheless,a planned comparison conducted on means for

each subject indicated that inflectional affixeswere shifted faster than their nonmorphologicalcontrols [Fl(I,23) =8.15, MSe =1201, P < .009]and a test conducted on means for each itemshowed the same trend [F2(1,17) =2.68, MSe =2393, P < .12]. No effects were significant forderivational affixes, however.

The outcome of Experiment 5 was that inflec­tional but not derivational segments were shiftedfrom source words to targets words more rapidlythan from their phonologically-matched controls.This result suggests that the component structureof morphologically complex words is sometimesavailable to the language processing mechanismand again, that base morphemes as contrastedwith phonemes are the relevant units.Numerically, the effect was comparable to that ofExperiment 4 suggesting that restricted prepara­tion time did not alter the processes involved inthis task. Times were longer overall but errorrates decreased. Importantly, the relation betweenspeed and accuracy across experiment did not dif­fer by experimental condition.

The linguistic productivity and lexical structureof inflectional as contrasted with derivationalformations noted above leads one to expectinflectional affixes to show enhanced effectsrelative to derivational affixes and the effect ofmorphological status was significant only forinflections in Experiment 5. The sixth experimentin this series also compares inflections andderivations in a more complex version of thesegment shifting task.

Table 7. Segment shifting times and errors for morphological affixes and their nonmorphological controls inExperiment 5.

MORPHOLOGICAL TYPE

INFLECTION

DERIVATION

SHIFfED SEGMENTMORPHEMIC NONMORPHEMIC

JEDEM BEDEM

781 8095% 4%

ZIDAR KADAR

753 7614% 5%

DIFFERENCE

28- I

81

86 Feldmnn

EXPERIMENT 6An attempt at replication of differences in in­

flectional and derivational processing with differ­ent materials necessitated a modification of thesegment shifting procedure described above. Inthis experiment, as in the previous segment shift­ing experiments, subjects had to shift the affixfrom the source word to the target word. In con­trast to the procedure of the previous experiments,in Experiment 6, subjects had to delete the origi­nal affix on the target word before substitutingthe shifted segment. As in Experiments 4 and 5,subjects had to name the resulting word aloud.The addition of this step whereby subjects had todelete the original affix (or its orthographicallyand phonemically matched control) rendered thetask more difficult but it permitted the compari­son of morphological constructions for inflectionaland derivational formations to be expanded.

MethodsSubjects. Twenty-six students from the same

population as those in previous experimentsparticipated in Experiment 6. None hadparticipated in previous experiments in this study.

Stimulus materials. Materials consisted ofthirty-six Serbian word pairs including equalnumbers of inflectional and derivational morpho­logical types and their nonmorphemic controls. Asin the previous experiment, the inflectional typeconsisted of first person singular verbs such asPROGONIM, meaning I capture, and their non­morphemic controls such as SINONIM, meaningsynonym. They were shifted to inflected targetssuch as DELE, meaning they share. In the presentexperiment, in order to respond DELIM, meaningI share, subjects had to delete the original affix(viz., E) and substitute the 1M affix. The deriva­tional type contrast consisted of singular diminu-

tives ending in ICA such as BASTICA, meaninglittle garden, and their controls such as KOSTICA,meaning seed. They were shifted to target such asBUBA, meaning bug and subjects respondedBUBICA, meaning little bug.

Procedure. The procedure of Experiment 6 waslike that of Experiment 5 (but not 4), in that thesegment of the source word that subjects had tosl).ift was not specified at the same time that thesource word appeared. Instead, it was indicatedafter 750 ms when the target word appeared.However, in both the inflectional and derivationalconditions of Experiment 6, subjects had to dropthe original (morphemic) affix on the target and tosubstitute the affix from the source word. That is,the final vowel on words such as BACE and BUBAwas deleted before adding 1M or ICA respectively.Finally, filler trials in which no portion wasunderlined were also included. In those cases,subjects were required to repeat the target wordin its original form.

Nine tokens in the morphemic andnonmorphemic conditions were included for boththe inflectional and derivational conditions.

Results and DiscussionAn analysis of variance on correct latencies less

extreme than 3 SD from the mean (so that approx­imately 6% of all responses were eliminated) re­vealed a significant effect of morphological type(inflection/derivation) [Fl(l,25) = 60.03, MSe =3295, p < .001; F2(l,34) = 17.12, MSe = 8177 p <.001] and a marginally significant interaction ofmorphological status and morphological type[Fl(l,25) = 10.02, MSe = 2960,p < .005; F2(l,34) =2.77, MSe = 7182, P < .10]. The effect of morpho­logical status just missed significance with sub­jects as a random variable [Fl(l,25) = 4.12, MSe =3250, p < .053]. Means are summarized in Table 8.

Table 8. Segment shifting times and errors for morphological affixes and their nonmorphological controls inExperiment 6.

MORPHOLOGICAL TYPE

SHIFTED SEGMENTMORPHEMIC NONMORPHEMIC DIFFERENCE

INFLECTION

DERIVATION

BAJAM SAlAM

829 886 579% 6% -3

BASTICA KOSTICA

776 765 -93% 3% 0

Belfand Orthography and Phonology: Differences between Inflections and Derivations 87

The effect of morphological status of affix (56ms) was significant for inflectional type pairs[Fl(l,25) =14.01, MSe =2960, P < .001)] by sub­jects and was marginally significant by items[F2(l,34) =3.24, MSe =7182, p < .08)]. For deriva­tional pairs, the effect was in the opposite direc­tion (-11 ms) and was not significant [Fl(1,25) =.54]. The significant effect for inflections and thenonsignificant change in direction for derivationsproduced the marginally significant interaction ofaffix by morphological type. For errors, the effectof morphological type was significant by subjects[Fl(1,25) =9.82, MSe =575, P < .005; F2(1,34) =7.73, MSe = 398, p < .09] but the main effect ofmorphological status and the interaction of con­trast by type did not approach significance.Because latency and error patterns for the targetsfollowing inflected primes in Experiment 6 sug­gested a speed accuracy tradeoff, correlations be­tween measures were computed. Neither the cor­relations for morphemic and nonmorphemic condi­tions separately nor the pooled correlation ap­proached significance. Evidently, latencies did notdecrease as errors increased.

The results of Experiment 6 are consistent withthe segment shifting results of the previous exper­iments whereby morphological segments areshifted faster than their nonmorphemic controls.The pattern of errors goes in the opposite directionbut it was not statistically significant nor was itproduced by a speed-accuracy tradeoff. Althoughthe results with items as a random factor areweak, the pattern was replicated with (a) the in­flectional affixes for instrumental nouns inExperiment 4 (b) first person singular verbs inExperiments 5 and 6. The set of experimentalmaterials for Experiment 6 required a modifica­tion to the experimental procedure whereby theoriginal affix on the target word had to be deletedbefore the shifted segment could be appended andit allowed a valuable replication of the previousresults. Specifically, the effect of segment shiftingwas significant for inflectional pairs but not forderivational pairs. These results are consistentwith the linguistic distinction between morpholog­ical types noted above and with the pattern ofproduction error whereby inflections enter intospeech errors more frequently than do derivations(Garrett, 1980). This finding suggests that themorphological structure of inflectional and deriva­tional formations does differ.

GENERAL DISCUSSIONIn the repetition priming paradigm, the pattern

of facilitation among lexical decision latencies for

target words that were preceded by morphologicalrelatives provided evidence that skilled readers ofSerbian are sensitive to the constituent structureof morphologically complex words. It was notnecessary for identical forms to be repeated inorder to reduce target decision latencies.Repetition of the same base morpheme in adifferent but related morphologically-complexword also produced facilitation. Evidence ofmorphological relatedness in repetition priming isconsistent with the results of similar studiesconducted across a variety of languages andmorphological contexts, and generally, it isinterpreted as evidence that morphology isrepresented in the lexicon.

Similarly, the failure to find facilitation inlexical decision among target pseudowords thatwere preceded by other pseudowords constructedfrom the same meaningless base morpheme andreal morphemic suffixes, or by words constructedfrom the same meaningful base morpheme in anillegal combination with a real affIX, is consistentwith the outcome of other studies that have usedthis experimental task. Although small facilitationeffects for pseudoword targets sometimes havebeen reported in lexical decision with repetitionpriming (e.g., Bentin & Feldman, 1990), it is neverthe case that pseudoword effects are numericallylarger than word effects and most typically theyare smaller. It has been suggested that forpseudowords, under some encoding conditions, theadvantage of repeating the same or a very similarorthographic and phonemic pattern is offset by theinappropriateness of responding "word" to afamiliar pseudoword pattern (Balota & Chumbley,1984; Duchek & Neely, 1989; Feustel, Shriffrin &Salasoo, 1983). That is, familiarity offsets anyadvantage associated with repeating a "no"response.

The present experiments conducted withSerbian materials permitted a rigorous compari­son of two types of morphological formations.When inflectionally- and derivationally-relatedprime-target pairs were compared, significantlygreater facilitation was observed for inflectionalrelatives than for derivational relatives. Thisfinding was observed in Experiment 1, in whichderivational formations differed in word classfrom inflectional formations but were equallysimilar with respect to phonological and ortho­graphic overlap, in Experiment 2, in which all for­matives were verbs and targets following deriva­tionally-related primes differed with respect to theaddition of either a prefix or a suffix, and inExperiment 3, in which one half of the primes

88 Feldman

were perfectly matched for overlap as well asword class with their targets and one half sharedone letter more in the inflectional condition thanin the derivational condition.

When derivationally-related prime-target pairswere compared with first presentation, significantfacilitation was observed for agentives inExperiment 1, for prefixed targets in Experiment2, and for mismatched pairs in Experiment 3.Facilitation following derivational primes was notsignificant in the analysis by subjects for infixeditems (Experiment 2). The planned comparison byitems for matched items (Experiment 3) was notsignificant although the latency differences wereconsistent with a pattern of facilitation. Thereliability of facilitation from derivationally­related primes may be low and the no primebaseline may overestimate the magnitude offacilitation (but see discussion of results forExperiment 1-3). However, the same pattern wasobserved in three experiments. Moreover, whenthe planned comparisons for no prime andderivational conditions in the three experiments(five conditions) were combined into one statisticaltest (Winer, 1971 p.49), results indicated thatfacilitation was significant [X2 (10) = 48.21 P <.001for subjects and X2 (10) = 38.97]p <.001 for items.In summary, although there is a tendency towardsfacilitation of targets following derivationalprimes, because targets always occurred slightlylater in the session than their primes, thefacilitation effects with derivational relativesshould be interpreted with caution.

It is unlikely that the effect of morphologicalrelatedness can be described in terms of thepattern of activation between the orthographicand perhaps phonological units that constitute aword. This claim is based on a) the pattern offacilitation for morphologically-related prime­target pairs in which the base morpheme does notalways retain the same form, b) the absence offacilitation among morphologically-unrelatedprime-target pairs that are similar in form, c)statistically significant differences in patterns offacilitation to targets primed by inflectional andderivational relatives that are matched or nearlymatched to the target for similarity of form, and d)the effect of morphological status on segmentshifting for inflectional affixes but not forderivational affixes. The separate bases for theseclaims will now be summarized.

In previous repetition priming studies, changesin spelling or pronunciation between morphologi­cally-related prime and target did not diminishthe magnitude of facilitation to target decision la-

tencies relative to morphologically-related prime­target pairs with no change. For example, Serbianforms that undergo palatalization (e.g., NOZI),forms with letter deletion (e.g., PETKU) and regu­lar forms (e.g., NOGOM) all produced equivalenttarget (e.g., NOGA, PETAK) facilitation (Feldman& Fowler, 1987). Similar results in repetitionpriming have been reported in English for moder­ately irregular forms such as SLEPT-SLEEP(Fowler et al., 1985; Napps, 1989; Stolz &Feldman, in press; cf. Stanners et al., 1979). Inaddition, recognition latency to inflected verbforms was sensitive to frequency of citation forms(and cumulative frequency for all regular forms)both when they differed in spelling (Kelliher &Henderson, 1990) and when spelling was pre­served (Katz, Rexer & Lukatela, 1991; Nagy,Anderson, Schommer, Scott & Stallman, 1989).That is, a contribution of both citation frequencyand cumulative frequency of morphologically-re­lated forms to recognition latency was observedeven when the orthographic and phonological formof the base morpheme was not preserved.Equivalent patterns of influence for morphologi­cally-related words with differing orthographicand phonological form and for words with similarform are problematic for any model that assumesthat the base morpheme alone or a pattern of acti­vation over its letter or phoneme units underliesmorphological effects. In summary, patterns of fa­cilitation in repetition priming suggest that theunderlying morphemic representation is abstractenough to tolerate at least moderate orthographicand phonological variation.

Whereas formal similarity of morphologicallyunrelated words can produce inhibition in somepresentation formats when items are presentedclose in succession (Grainger, Cole, & Segui, 1990;Laudanna et al., 1992; Segui & Grainger, 1990;Stolz & Feldman, in press ), at long lags it is thecase that the formal similarity of morphologicallyunrelated primes and targets (e.g., pairs such asDIET and DIE) does not result in priming eitherin English (Hanson & Wilkenfeld, 1985; Napps &Fowler, 1987; Stolz & Feldman, in press) or inSerbian (Feldman & Andjelkovic, 1992; Feldman& Moskovljevic, 1987; ). For example, for prime­target pairs formed from unrelated homographicbase morphemes (e.g., BOR) such as BOHAMA(dative plural of BORA meaning wrinkle) andBOROVI (nominative plural of BOR meaningpine), no effect of formal similarity was observedfor inflectionally complex words at lags of tenitems in repetition priming (Feldman &Andjelkovic, 1992). Stated generally, whereas or-

Beyond Orthography and Phonology: Differences between Inflections and Derivations 89

thographically similar form plays a role at shortlags, at long lags, as in the repetition primingtask, there was neither a facilitative nor inhibitoryeffect on the target of orthographic andphonological similarity between morphologically­unrelated prime and target.

The primary finding of Experiments 1, 2 and 3was that inflectional primes produced signifi­cantly greater facilitation than did derivationalprimes and that derivational primes produced fa­cilitation relative to the no prime condition. Thisoutcome was observed under experimental condi­tions that a) perfectly matched formal overlap ofprime and target but left word class free to vary(Exp. 1), b) perfectly controlled word class by us­ing only verb forms and manipulated position inwhich affix was added and c) perfectly controlledword class by using only verb forms but matchedletter overlap of inflectional and derivational rel­atives on only one half of the items (Exp. 3).

It is evident that the lexical representation ofinflectional and derivational formations must dif­fer. Several accounts have been proposed. It hasbeen suggested that, in the lexicon, the linkagebetween whole word forms that share a base mor­pheme is stronger for inflectionally-related formsthan for derivationally-related forms or that theconnections between components must be strongerfor inflectional than for derivational formations.This is consistent with the linguistic claim thatthe component structure of inflectionally-relatedforms is more transparent than that of deriva­tionally-related forms. Alternatively, as notedabove, inflectional formations tend to share a basemorpheme and stem and differ with respect to in­flectional affix whereas derivational formationstend to share a base morpheme and differ with re­spect to derivational affixes and stem.Accordingly, if both stems and bases are taken asunits to be activated in repetition priming, thenthe difference between inflections and derivationscould reflect redundant activation for inflectionsrelative to derivations. Results of Experiment 3indicate that this account is incomplete, at best.Inflectional relatives like NOSIM and derivationalrelatives like NOSAM share both base morphemeand stem but they did not produce equivalent tar­get facilitation.

Results of the segment shifting task also providecompelling evidence that the morphological statusof a word's constituents influences performance inrecognition tasks. Although differences betweeninflectional and derivational affixes were weakerin this task, in that analyses by items tended tohave significance levels around p < .10, a similar

outcome was observed in three experiments withdifferent manipulations on inflectional affixes.Importantly, the results were consistent withthose obtained in repetition priming. In the seg­ment shifting task, morphological effects weremore reliable for inflectional affixes relative totheir controls than for derivational affixes andtheir controls. Phonological and orthographicproperties of a source word were matched in theexperimental design and sequences that createdmorphological segments were manipulated moreefficiently than nonmorphological controls over avariety of inflectional environments. The morpho­logical origins of a segment were evident despitethe fact that all of the responses articulated bysubjects were frequent words and that responseswere the same in the morphological and nonmor­phological conditions.

In that the internal structure of words formedwith inflectional affixation may be more transpar­ent than that of words with derivational affixa­tion, and effects were more reliable for inflectionsthan for derivations, an account of the segmentshifting experiments based on ease of segmenta­tion is plausible. Similarly, the emphasis could beon coherence at the boundary between sublexicalword units. In languages such as Serbian (andEnglish) in which morphemes are concatenated toform complex forms, there is a temptation to de­scribe morphemes in terms of boundanes betweenorthographic or phonological sequences of units. Asshould be obvious from the present discussion,however, these lexical representations must besufficiently abstract to accommodate changes inform as well as the word context of which the se­quence is a part. Admittedly, it is difficult to dis­tinguish between segmentation of and coherencebetween morphemic components when morphol­ogy is concatenated.

An investigation of morphological effects in anonconcatenative language such as Hebrew is lessamenable to an account based on sequence,however. In Hebrew, the root or base morpheme ofa word is represented by a discontinuous patternof (usually) three consonants. Vowels carrying aninflectional function are infixed between theseunits. Consequently the base morpheme is notrealized as an uninterrupted unit within the word.It is necessarily abstract with respect tophonological (and sometimes orthographic)patterning and yet, effects of morphologicalrelatedness have been demonstrated both inrepetition priming (Bentin & Feldman, 1990;Feldman & Bentin, 1994) and in segment shifting(Feldman, Frost & Pnini, in press). In Hebrew,

90 Feldman

ease of segmentation rather than coherence mayprovide the more accurate account.

Alternatively, the locus of the segment shiftingeffect could reflect strategies that vary on a trialby trial basis and reflect compatibility betweensource and target word. Accordingly, word classcompatibility effects for affixes from source andtarget words would be anticipated. The CEVI­CEDI comparison in Experiment 4 indicated thatwhen affixes were shifted from a source word ofthe same or a different word class as the target,latencies were equivalent. Similarly, the results ofsegment shifting experiments in Hebrew(Feldman, Frost & Pnini, in press) indicate thatmorphological effects can be observed when theaffix is shifted to a meaningless target string.Segment shifting effects are not expected on pseu­doword targets if the effect reflects compatibility.Evidently, the locus of the segment shifting effectis not yet well understood but, at this point, a lexi­cal locus tied to morphemic as distinguished fromorthographic components seems likely.

In summary, two very different experimentalparadigms provide strong support for the psycho­logical processing of the morpheme and for a dis­tinction between the processing of inflectional andderivational formations. Similarity of form definedby orthographic and phonological overlap of mor­phologically-related primes and targets is not anecessary condition to produce facilitation.Similarity of form in the absence of morphologicalrelatedness is not a sufficient condition to producetarget facilitation or inhibition at long lags.Patterns of activation over orthographic or phono­logical units cannot describe morphological effects.Importantly, they cannot account for the differ­ences between inflections and derivations whensemantic similarity is controlled. Evidently pro­cessing of a word is sensitive to that word's con­stituent morphemic structure.

REFERENCESAnderson, S. R. (1982). Where's morphology? Linguistic Inquiry,

13,571-612.Aronoff, M. (1976). Word formation in generative Grammar.

Cambridge, MA: MIT Press.Balota, D. A., & Chumbley, J. 1. (1984). Are lexical decision a good

measure of lexical access? The role of word frequency in theneglected decision stage. Journal of Experimental Psychology:Human Perception and Performance, 10, 340-357.

Badecker, W., & Caramazza, A. (1989). A lexical distinction be..tween inflection and derivation. Linguistic Inquiry, 20,108..116.

Bentin, S., & Feldman, 1. 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.

Bergman, B., Hudson, P., & Eling, P., (1988). How simple complexwords can be: Morphological processing and wordrepresentations. Quarterly Journal of Experimental Psychology,4OA,41-72.

Burani, C, & Laudanna, A., (1992). Units of representation forderived words in the lexicon. In R Frost and L. Katz (Eds.)Orthography, phonology, morphology and meaning (pp. 361-376).Amsterdam: Elsevier Science Publishers RV.

Butterworth, R (1983). Lexical representation. In B. Butterworth(Ed.), Language production (Vol. 1). London and San Diego:Academic Press.

Bybee, J. H. (1985). Morphology. Amsterdam: John Benjamins.Caramazza, A, Laudanna, A., & Romani, C (1988). Lexical access

and inflectional morphology. Cognition, 28, 287-332.Caramazza, A., Miceli, G., Silveri, M. C, & Laudanna, A. (1985).

Reading mechanisms and the organization of the lexicon:Evidence from acquired dyslexia. Cognitive Neuropsychology, 2,81..114.

Cutler, A. (1980). Errors of stress and intonation. In V. A. Fromkin(Ed.), Errors in linguistic performance. New York: AcademicPress.

Dell, G. S. (1986). A spreading-activation theory of retrieval insentence production. Psychological Review, 93, 285-321.

Dell, G. S. (1990). Effects of frequency and vocabulary type onphonological speech errors. Language and Cognitive Processes, 5,313-349.

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

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

Feldman, 1. B. (1992). Morphological relationships revealedthrough the repetition priming task. In M. Noonan, P.Downing, & S. Lima (Eds.), Literacy & Linguistics.Amsterdam/Philadelphia: John Benjamins Pub. Co.

FEldman, L. B., & Andjelkovit, D. (1992). Morphological analysisin word recognition. In R. Frost & L. Katz (Eds.), Phonology,orthography, morphology and meaning (pp. 343..360). Amsterdam:Elsevier Science Publishers B.V.

Feldman, L. B., & Bentin, S. (1994). Morphological analysis ofdisrupted morphemes: Evidence from Hebrew. QuarterlyJournal of Experimental Psychology, 47A, 407-435.

Feldman, L. B., Frost, R. & Pnini, D. (in press). Morphemes neednot be contiguous units. Journal of Experimental Psychology:Learning, Memory and Cognition.

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. & Fowler, C. A. (1987b) Morphemic segments shiftfaster than their nonmorphemic controls. Paper presented tothe Psychonomic Society, Seattle, WA.

Feldman, L. B., & Moskovljevit, 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.

Fowler, CA., 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.

Fromkin, V. A. (1973). Speech errors as linguistic evidence. Mouton:The Hague.

Beyond Orthography and Phonology: Differences between Inflections and Derivations 91

Garrett, M. F. (1976). Syntactic processes in language production.In R. J. Wales & E. C. T. Walker (Eds.), New approaches tolanguage mechanisms (pp. 231-256). Amsterdam: North Holland.

Garrett, M. F. (1980). A perspective on research in languageproduction. In E. C. Mehler, T. Walker, & M. F. Garrett (Eds.),Perspectives on mental representation (pp. 179-220). Hillsdale, NJ:Lawrence Erlbaum Associates.

Garrett, M. F. (1982). Production of speech. Observations fromnormal and pathological language use. In A. Ellis (Ed.),Normality and pathology in cognitive functions (pp. 19-76).London: Academic Press.

Grainger, J., Cole P., & Segui, J. (1991). Masked morphologicalpriming in visual word recognition. Journal of Memory &Language, 30, 370-384.

Hanson, V. L., & Feldman, L. B. (1989). Language specificity inlexical organization: Evidence from deaf signers' lexicalorganization of ASL and English. Memory & Cognition, 17, 292­30l.

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.

Hudson, P. T. W. (1990). What's in a word? Levels ofrepresentation and word recognition. In D. A. Balota, G. B.Flores d'Arcais & K. Rayner (Eds.), Comprehension processes inreading. Hillsdale: Erlbaum.

Katz, L., Rexner, 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.

Kirsner, K., Milech, D., & Standen, P. (1983). Common andmodality-specific processes in the mental lexical. Memory &Cognition, 11,621-630.

Laudanna, A, Badecker, W., & Caramazza, A (1992). Processinginflectional and derivational morphology. Journal of Memory &Language, 31, 333-348.

Lyons, J. (1977). Semantics (Vol. 2). Cambridge: CambridgeUniversity Press.

Miceli, G. , & Caramazza , A. (1988). Dissociation of inflectionaland derivational morphology. Brain and Language, 35, 24-65.

Nagy, W., Anderson, R. c., Schommer, M., Scott, J. A., &Stallman, AC. (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 Reseorch, 16, 257-272.

Neely, J. (1991). Semantic priming effects in visual wordrecognition: A selective review of current findings and theories.In D. Besner & G. W. Humphreys (Eds.), Basic processes inreading: Visual word recognition. Hillsdale, NJ: LawrenceErlbaum Associates.

Partridge, M. (1964). Serbo-Croatian practical grammar andreader. Belgrade: University of Belgrade Press.

Radeau, M. (1983). Semantic priming between spoken words inadults and children. Canadian Journal ofPsychology, 37, 547-556.

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

Scalise, S. (1984). Generative morphology, Dordrecht: ForisPublications.

Schriefers, H., Friederid , A., & Graetz, P., (1992). Inflectional andderivational morphology in the mental lexicon: Symmetriesand asymmetries in repetition priming. Quarterly Journal ofExperimental Psychology, 44, 373-390.

Segui, J., & Grainger, J. ( 1990). Priming word recognition withorthographic neighbors: Effects of relative prime-targetfrequency. Journal of Experimental Psychology: HumanPerception and Performance, 16365-76.

Siowiaczek, L. M. (1994). Semantic priming in a single-wordshadowing task. American Journal ofPsychology, 107,245-260.

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

Sternberger, ]. P. (1984). Structural errors in normal andagrammatic speech. Cognitive Neuropsychology, 1, 281-313.

Sternberger, ]. P. (1985). An interactive activation model oflanguage production. Progress in the psychology of language (pp.143-186). London: Lawrence Erlbaum Associates.

Stolz, J. A., & Feldman, L. B. (in press). The role of orthographicand semantic transparency of the base morpheme inmorphological processing. In L. B. Feldman (Ed.), Morphologicalaspects of language processing. Hillsdale: Lawrence ErlbaumAssociates.

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

Winer, B. J. (1971). Statistical principles in experimental design. NewYork: McGraw-Hill.

FOOTNOTES"Appears in Journal of Memory and Language, 33.442-470 (1994).t Also State University of New York at Albany.

92 Feldman

APPENDIX 1Experiment 1 materials

TARGET IDENTITY INFLECTION DERIVATION

blud blud bludom bludimbol bol bolom bolanbranim branim branis branikbrod brod brodu brodibroj broj broju brojibura bura burom burancud cud cudom eudimcvet cvet cvetom cvetandeo deo delom delimdocek docek docekom docekangovor govor govorom govorimgrad grad gradom gradimhlad hlad hladom hladanigras igras igram igraclorn lorn lomom lomimnosam nosam nosas nosasobaram obaram obaras obaracpeva peva pevam pevacplivam plivam plivas plivacracuna racuna racunam racunarrad rad radom radimslikas slikas slikam slikarspava spava spavas spavactragas tragas tragam tragacvaja vaja vajam vajarvodis vodis vodim vodiczvoni zvonl zvonim zvonik

Beyond Orthography and Phonology: Differences between In(lections and Derivations 93

Experiment 2 materials

TARGET IDENTITY INFLECTION DERIVATION

PREFIXED:obare obare obarim barimocede ocede ocedim cedimociste oCiste ocistim Cistimobodre obodre obodrim bodrimoderu oderu oderem deremoljuste oljuste oljustim ljustimiseku iseku isecem secemiskoce iskoce iskocim skocimisele isele iselim selimulepe ulepe ulepim lepimurade urade uradim radimubodu ubodu ubodem bodemzgrabe zgrabe zgrabim grabimzbroje zbroje zbrojim brojimzbace zbace zbacim bacimzdrobe zdrobe zdrobim drobimzgnjece zgnjece zgnjeCim gnjecimzbrisu zbrisu zbrisem brisemslete slete sletim letimsmute smute smutim mutimsprze sprze sprzim przimsmrve smrve smrvim mrvimstresu stresu stresem tresemslome slome slomim lomim

INFIXED:birkaju birkaju birkam biramcarnu carou carnem caramdirnu dirnu dirnem diramdzarnu dzarnu dzarnem dzaramgurnu gurnu gurnem guramjavnu javnu javnem javimJurnu jurnu jurnem jurimkidnu kidnu kidnem kidamkucnu kucnu kucnem kucammere mere merkam merimmrdnu mrdnu mrdnem mrdamnjuse njuse njuskam njusimpadnu padnu padnem padampirkaju pirkaju pirka pirisednu sednu sednem sedamsevnu sevnu sevnem sevamsuste suste suskam sustimsvirkaju svirkaju svirkam sviramturnu turnu turnem turamvirkaju virkaju virkam viramvirnu virnu virnem virimvrdnu vrdnu vrdnem vrdamzivnu zivnu zivnem zivimzovnu zovnu zovnem zovem

94 Feldman

Expariment 3 materials

TARGET IDENTITY INFLECTION DERIVATION

MATCHED:bace bace bacim bacamcite cite ci6m cicemdobiju dobiju dobijam dobijemklize klize klizim klizamlupe lupe lupim lupamnose nose nosim nosamodbiju odbiju odbijam odbijemopuste opuste opustim opustamsede sede sedim sedamspero spero speram speremvode vode vodim vodamvoze voze vozim vozamcuce tute tu6m tutnem*

MISMATCHED:dirnemo dirnemo dirnem diramduvnemo duvnemo duvnem duvamnapijamo napijamo napijam napijemnatapamo natapamo natapam natopimnazivamo nazivamo nazivam nazovemnaturamo naturamo naturam naturnemobaramo obaramo obaram oborimobijamo obijamo obijam obijemodvajamo odvajamo odvajam odvojimpotapamo potapamo potapam potopimpovijamo povijamo povijam povijemskidamo skidamo skidam skinemzovnemo zovnemo zovnem zovem

*not included in the analysis

Beyond Orthography and Phonology: Differences between Inflections and Derivations

Experiment 4 materials

95

MORPHOLOGICALLYCOMPLEX

WORD

INFLECTIONSslavompresomprozomakeijommaljembojempaleemhieemprolecemtucemadresomprincemarkadometikomkafanomantenom

HOMOGRAPHIC MORPHEME

MORPHOLOGICALLYSIMPLE

WORD

slalomprelomprolomaksiommelemboemparfemharemproblemtotemagronomprijemastronomekonomkareinomanatom

TARGET

radtopgrofdankrajmuzbrojnozcajkonjkumkoslayzidzetsat

cedidavideligadjakidamrZI

camikostabujacrpikrstiprskapadamastaprastagleda

eevi jaddani cuddedi gostgara rakkipa Ianmravi gledcari reckosa nosbuta zeeervi cadjkosti zubprsta lukpoda klubmosta domplasta sirgliba sin

96 Feldman

Experiment 5 materials

i%

MORPHOLOGICALLY MORPHOLOGICALLY TARGETCOMPLEX SIMPLE

WORD WORD

DERIVATIONS

berat kolat letbirat vrat pIestuvar stvar redkuvar ajvar mlinlimar ormar lugmerat :tarat kovorat trat vozpekar bakar kalemribar dabar stanrudar sudar domslikar plakar grobsumar samar brodsvirat otirat glastrubat korbat krojvidar radar drugvinar bunar zubzidar kadar lekzlatar litar put

INFLECTIONS

bajam sajam kidabiram jaram pitaderem bagrem berediktiram dijagram diagramjedem bedem klujekradem badem ka:temajam zajam diraorem harem dignepajam pojam bujaperem melem pasepijem prijem dodepletem sistem baktepostim kostim daviprogonim sinonim delitonem fonem pisuudaram epigram kupavajam najam padazidam islam peva

Bet(ond Orthography and Phonology: Dif(erences between Inflections and Derivations

Experiment 6 materials

97

MORPHOLOGICALLYCOMPLEX

WORD

DERIVATIONS

vranicaspravicapelenicabasticasavanicasobicamasnicakadicanajavicakravicabananicabanjicapesmicarolnicazabavicatasnicasarmicasaunica

INFLECTIONS

bajamdererndiktirarnjedernkrademmajamorempajarnperempijempletempostirnprogonirntonernudaramvajamzajamzidarn

MORPHOLOGICALLYSIMPLE

WORD

stanicazdravicavodenicakosticatavanicaubicamasnicaladicakijavicakrivicapijanicabrnjicapesnicabolnicabradavicakosnicasamicasapunica

sajambagremdiagrambedernbademzajarnharemjaramrnelempnJemsistemkostimsinonimfonemepigramnajampojamislam

TARGET

kantabundakrpabubatablakucapritagumakastabaraguzvakiflakorpabasnapalmapostaranategla

kidajuberngadjajuklujukazudirajudignukupajupasudodjupodjudavedelepisubirajupadajububajupevaju