Event-related Brain Potentials Elicited by MorphologicalHomographic Orthographic and Semantic Priming

Alberto Domrsquonguez Manuel de Vega and Horacio Barber

Abstract

amp The morphological structure of words in terms of theirstem morphemes and affixes could influence word access andrepresentation in lexical memory Three experiments werecarried out to explore the attributes of event-related potentialsevoked by different types of priming Morphological primingwith pairs of words related by their stem (hijohija [sondaughter]) produced a sustained attenuation (and even atendency to positivity) of the N400 shown by unrelated wordsacross the three experiments Homographic priming (Experi-ment 1) using pairs of words with a superficially similar stem

but without morphological or semantic relation ( focofoca[ floodlightseal]) produced an initial attenuation similar to themorphological pairs but which rapidly tended to form adelayed N400 due to the impossibility of integration Howeverorthographic priming (rasarana [ flatfrog]) in Experiment 2does not produce attenuation of the N400 but an effect similarto that of unrelated pairs Experiment 3 shows that synonymsadvance more slowly than morphological pairs to meaningcoherence but finally produce a more positive peak around600 msec amp

INTRODUCTION

Words are complex structures generally composed ofmore than one morpheme For example in the wordwalker the suffix -er adds to the stem walk a variationthat transforms the meaning of the stem This morpho-logical compositionality could be a structural property ofwords taken into account by the human linguistic pro-cessor If words are organized in memory according totheir morphological properties then stem and affixesare the best candidates as access units But if wholewords are represented in the mental lexicon then theseunits could be letters These two possibilities differ intheir demands of cognitive resources and processesWhole word representation is very expensive in termsof memory resources because this requires a lexicalentry for each derived or inflected word On the otherhand morphological representation drastically reducesthe memory demands although this requires a previoussegmentation of the input words into their constituentmorphemes which may slow down word recognition Inaddition to the problems of access and lexical represen-tation research is concerned with the functional archi-tecture of the system whether modular or interactive aswe will refer to later

This article focuses on the size of the access unitswhether letters or morphemes that guide the ascentfrom input words to meaning This topic has beenapproached using different methodologies for example

manipulating the frequency of the root building mor-phologically structured pseudowords (nonwords madeup of existing morphemes) exploring the recognition ofpast tense irregular verbal forms or evaluating produc-tion and comprehension in patients with cerebral dam-age in language areas Perhaps the most powerfulmanipulation however at least for our purposes is thecontrast between morphological priming (ie a primeword such as nino [boy] is briefly presented and thenfollowed by a target word such as nina [girl] bothsharing the stem but dif fer ing in the gendersuffixes) and some other types of priming In particularLaudanna Badecker and Caramazza (1989 1992) re-ported in Italian slower lexical decision times for a pairof stem homographs such as port-areport-e (to carrydoor) than for pairs of words with similar orthographicoverlap and different stem descriptions such as coll-ocolp-o (neckblow) The special relationship betweenthese words is based on the letter overlap in the stemof both prime and target There is in fact no semantic ormorphological relationship between these but superfi-cial processing of the prime could lead to the targetbeing taken as a morphological relative of the prime andtherefore to a garden path in the meaning access Similarresults were obtained in Spanish by Allen and Badecker(1999 2002) The inhibition obtained for mor-otildeamor-os(the prime is the third person singular imperfect form ofthe verb mor-ir lsquolsquoto diersquorsquo the target means Arabic) washigher than the inhibition for moralmor-os (moralityArabic) These results point firstly to an orthographic ac-tivation of the stem mor- and secondly to a competitionUniversity of La Laguna Tenerife Spain

copy 2004 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 164 pp 598ndash608

between the two morphemes and meanings (death andrace) activated by the orthographic features of the stemmor- The homographic stem inhibition is not explicablein terms of letter overlap but only in terms of morpho-logical parsing The study of Laudanna et al (1989) inItalian and that of Allen and Badecker (1999) in Spanishhave assumed that the AAM model provided the bestframe to explain their results (Caramazza Miceli Silveriamp Laudana 1985 Caramazza Laudana amp Romani 1988)In this model the entries are stored in an Input Ortho-graphic Lexicon as a structured sequence whose firstelement is the orthographic form of the stem and all thegrammatical features associated with this form Thus theentries for morotildea and moros initially have the sameorthographic form (they are stem homographs) andconsequently a competition or inhibition between thesewould take place at the moment of lexical selection

In Spanish Domotildenguez Cuetos and Segu otilde (2002)compared stem homographic priming such as in foc-ofoc-a (lightbulbseal) with morphological primingsuch as hij-ohij-a (sondaughter) obtaining significantfacilitation for morphological pairs and a tendency toinhibition for stem homographic pairs Using similarstimuli Garcotildea-Albea Sanchez-Casas and Igoa (1998)also found a tendency to inhibition for stem homo-graphs However in none of these studies was theinhibitory tendency for stem homographs statisticallysignificant in comparison to unrelated pairs

Although the above behavioral measures do not re-veal a significant stem homographic effect electrophys-iological measures such as event-related potentials(ERPs) have shown a clearer dissociation between stemhomographs and morphological pairs Thus BarberDomotildenguez and de Vega (2002) obtained for bothconditions a similar attenuation of the N400 componentin the 350ndash500 msec window and a broader negativityfor stem homographs in the 500ndash600 msec window Thislate negativity was compatible with the Laudana et al(1989) proposal that the early inhibitory effect of stemhomographs delays the stage of meaning integration

Experiment 1 addresses the first aim of the presentstudy To replicate the dissociation between stem homo-graphs and morphological pairs and to extend the ERPanalyses up to 1000 msec Extending the temporal win-dow beyond 600 msec is potentially interesting because itallows us to see whether the stem homographic curveinvolves a sustained negativity or conversely a later peakfollowed by a positive recuperation (delayed N400) Thelatter would support Barber et alrsquos (2002) suggestionthat stem homographs cause an initial interferencefollowed by a delay in the meaning integration stage Ifthis is not the case the sustained negativity would notfind easy agreement in the literature

A second objective was to rule out a possible con-founding between stem homographic effects and ortho-graphic priming With this aim a new orthographicpriming condition was introduced in Experiment 2

involving pairs of words such as rasarana (flatfrog)These orthographic pairs are semantically unrelated butthey overlap in all letters except one In this way theformal similarity between primes and targets parallelsthat of morphological pairs or stem homographs If thepreviously obtained results for stem homographs de-note initial morphological stem segmentation then noN400 attenuation should be expected for orthographicpairs since they do not imply stem overlap

The final objective was to solve another confoundingof the previous experiments Morphological priming isalso semantic priming In Experiment 3 pairs of syno-nym words such as ciriovela (candlecandle) wereintroduced These synonym pairs maintain the samesemantic relationship as morphological pairs but avoidany orthographic or morphological relation If the mor-phological effect denotes only semantic integration wewill find a similar N400 attenuation both in morpho-logical pairs and in synonym pairs In contrast if mor-phological effects are due to genuine morphologicalpriming then the N400 attenuation will be clearer formorphological pairs than for synonyms Thus for themorphological pair hijohija processing of the primestem would facilitate the correct lexical entry to thetarget whereas no morphological facilitation wouldoccur for ciriovela despite their semantic relation

To sum up the aims of the present research will befirstly to know whether or not the unit of lexical accessis the morpheme secondly to ensure that the stemhomographic effect is morphological and not just anorthographic effect and thirdly to separate morpho-logical from merely semantic effects

EXPERIMENT 1 STEM HOMOGRAPHS

The goal of this experiment was to replicate the exper-iment of Barber et al (2002) taking a longer window ofERP for analysis In this way we may observe whetherstem homographs produce a delayed N400 This wouldsupport the hypothesis of a later difficulty in meaningintegration following the initial failure in morphologicalparsing A morphological primingmdashhijohija (sondaughter)mdashwas compared with a stem homographicprimingmdashfocafoco (sealfloodlight)mdashand an unrelatedcontrol conditionmdashpavometa(turkeygoal) It was ex-pected that stem homograph pairs would produce adelayed N400 whereas morphological pairs would resultin a more permanent attenuation of N400 It was alsoexpected that stem homographs would not produce asignificant inhibition in reaction times in comparisonwith the unrelated condition

Results

Behavioral data

Mean reaction times for correct responses in each con-dition are presented in Table 1 Statistical analyses

Domotildenguez de Vega and Barber 599

showed that the 49-msec difference between morpho-logical priming and unrelated priming [F(110) = 1449p lt 01] and the 49-msec difference between morpho-logical priming and stem homographic priming[F(110) = 1108 p lt 01] are statistically significantNo difference was evident between responses to stemhomographs and unrelated pairs

ERP data

Figure 1A shows the grand ERP average for the threepriming conditions at Fz Cz and Pz sites In thiswindow we see a center-parietal negativity for unrelatedpairs peaking around 380 msec which can be identifiedwith the N400 component found in other studies Themorphological priming and the stem homograph con-dition produced a clear attenuation of this componentbeginning around 250 msec after stimulus onset Bothconditions differ later however Whereas stem homo-

graphs immediately decline starting at 350 msec andreach a negative peak around 450 msec morphologicalpairs produce only a slow and sustained trend tonegativity

250ndash350 msec

At this initial window morphological priming did notdiffer from stem homographic priming and both weredifferent from unrelated pairs [morphological primingF(110) = 2464 p = 0001 marginally for stem homo-graphic priming F(110) = 446 p = 0062] This initialoverlap between the two types of priming supports anearly orthographic processing of the stem of the word

350ndash450 msec

In this second window morphological priming main-tained a sustained positivity evidenced in a significantdifference with the nonrelated pairs [F(110) = 2678p lt 001] An interaction between priming and elec-trode [F(220) = 674 p lt 05 F(220) = 778 p lt 005normalized] situated this morphological effect at theposterior sites although the separate analysis for eachelectrode was significant for the three localizationsAlso morphological priming differs from stem homo-graphs in disagreement with the previous window

Table 1 Mean Reaction Times (msec) and Standard Devia-tions in Experiment 1 for Each Experimental Condition

Type of Priming Mean (SD)

Morphological Stem Homographic Unrelated

686 (175) 735 (155) 735 (161)

Figure 1 Grand averagewaveforms in the middle linefor the three experiments (A)introducing stem homographsin Experiment 1 (B) withorthographically related wordsin Experiment 2 and (C) withsynonyms in Experiment 3

600 Journal of Cognitive Neuroscience Volume 16 Number 4

[F(110) = 2898 p lt 001] Stem homographic primingon the unrelated items did not produce a significanteffect but interacted with the electrode factor[F(220) = 1185 p lt 001 F(220) = 1737 p lt 001normalized] the difference only being significant at thePz electrode [F(110) = 927 p lt 05]

450ndash650 msec

Figure 1A clearly shows that at this late window stemhomographic pairs markedly differ from morphologicalpairs The former shows an N400-like component sim-ilar to that of the unrelated pairs but approximately100 msec later Statistical analyses showed that the meanamplitude was more negative for stem homographs thanfor morphological pairs [F(110) = 155 p lt 005] Alsostem homographs were more negative than unrelatedpriming but in this case the difference was only mar-ginally significant [F(110) = 422 p = 067] No statis-tical differences were found between the morphologicaland the unrelated priming conditions

A peak latency analysis was carried out to determinemore exactly the difference between the N400 producedby unrelated pairs and what we called the N400 retardedcomponent produced by the stem homographs condi-tion (the mean of peak latencies can be seen in Table 2)This analysis was run on the more negative latency in thecurves for each subject between 250 and 650 msec andit resulted in a significant difference between the mo-ments of the two peaks [F(110)= 1484 p lt 005]

Discussion

The results obtained by Barber et al (2002) were repli-cated in this experiment Morphological priming pro-duced an attenuation of the N400 as a result of the easyintegration between the prime and the target that notonly share a part of their meaning but also their stemmorpheme Homographic pairs show a wave that over-laps that of the morphological pairs until 350 msecsuggesting that participants initially try to integrate wordsas if they were morphological pairs However the later

rapid decay to negativity peaking around 500 msec forhomographs (a retarded N400 component) greater atthe parietal site should be interpreted as a word inhibi-tion caused by the meaning disparity found at later stagesof word processing By contrast the morphological con-dition produces a more distributed and sustained nega-tivity in the three central electrodes The results seemto support a model of lexical access partially based onthe morphological organization of words but beforeaccepting this interpretation we have to rule out alterna-tive explanation of the data Thus we are assuming thatthe retarded N400 for stem homographs indicates initialmorphological processing but this could also be ex-plained as simply the result of the orthographic relationbetween the prime and the target The next experimenttries to test this possibility

EXPERIMENT 2 ORTHOGRAPHIC PRIMING

Stem homographs such as focafoco orthographicallyoverlap in their stem morphemes (eg foc-) But in factfoc- in foca is a different morpheme from foc- in focoThis might be the cause of the inhibition found withreaction time measures in Italian (Laudanna et al 19891992) and in Spanish (Allen amp Badecker 1999) and alsothe cause of the N400 retardation in Experiment 1However we need to discard the alternative explanationthat the inhibition is merely caused by the orthographicoverlap Many reaction time studies have demonstratedthat pairs of words sharing all except one letter (neigh-bors) but without any morphological relation producedinhibition These results were obtained with a similarexperimental procedure to our experiments large primeduration (200 msec) and primendashtarget frequency rela-tion (low-frequency primes and high-frequency targets)Furthermore some ERP studies compared unrelatedpriming with formal priming (Doyle Rugg amp Wells1996) and rhyme priming (Radeau Bensson Fonteneauand Luis-Castro 1998) obtaining an N400 amplitudemodulation for formal priming pairs in comparison withunrelated pairs In a wider experimental context wordswith many lexical neighbors produced larger N400 thanwords with fewer lexical neighbors (Holcomb Graingeramp OrsquoRourke 2002) denoting that this component issensitive to orthographic variables of the lexicon

Experiment 2 compared morphological priming (hijohija [sondaughter]) with orthographic priming (rasarana [flatfrog]) If the orthographic condition causesthe same delayed N400 we obtained in Experiment 1 forstem homographs then we might consider that ortho-graphic overlapping rather than morphological processing is the appropriate explanation for all primingeffects observed in Experiment 1 If the orthographiccondition however produces no delayed N400 then wemay accept that morphological processing is a betterexplanation for both morphological priming and stemhomographic priming

Table 2 Peak Latencies (msec) from the Onset of the Targetand Standard Deviations for Unrelated Priming and StemHomographic Priming at the Three Analyzed Sites Fz Cz andPz (Experiment 1)

Type of Priming

Unrelated Stem Homographic PrimingElectrode

Sites Mean SD Mean SD

Fz 372 45 463 80

Cz 385 44 486 101

Pz 438 91 521 100

Domotildenguez de Vega and Barber 601

[F(110) = 2898 p lt 001] Stem homographic primingon the unrelated items did not produce a significanteffect but interacted with the electrode factor[F(220) = 1185 p lt 001 F(220) = 1737 p lt 001normalized] the difference only being significant at thePz electrode [F(110) = 927 p lt 05]

450ndash650 msec

Figure 1A clearly shows that at this late window stemhomographic pairs markedly differ from morphologicalpairs The former shows an N400-like component sim-ilar to that of the unrelated pairs but approximately100 msec later Statistical analyses showed that the meanamplitude was more negative for stem homographs thanfor morphological pairs [F(110) = 155 p lt 005] Alsostem homographs were more negative than unrelatedpriming but in this case the difference was only mar-ginally significant [F(110) = 422 p = 067] No statis-tical differences were found between the morphologicaland the unrelated priming conditions

A peak latency analysis was carried out to determinemore exactly the difference between the N400 producedby unrelated pairs and what we called the N400 retardedcomponent produced by the stem homographs condi-tion (the mean of peak latencies can be seen in Table 2)This analysis was run on the more negative latency in thecurves for each subject between 250 and 650 msec andit resulted in a significant difference between the mo-ments of the two peaks [F(110)= 1484 p lt 005]

Discussion

The results obtained by Barber et al (2002) were repli-cated in this experiment Morphological priming pro-duced an attenuation of the N400 as a result of the easyintegration between the prime and the target that notonly share a part of their meaning but also their stemmorpheme Homographic pairs show a wave that over-laps that of the morphological pairs until 350 msecsuggesting that participants initially try to integrate wordsas if they were morphological pairs However the later

rapid decay to negativity peaking around 500 msec forhomographs (a retarded N400 component) greater atthe parietal site should be interpreted as a word inhibi-tion caused by the meaning disparity found at later stagesof word processing By contrast the morphological con-dition produces a more distributed and sustained nega-tivity in the three central electrodes The results seemto support a model of lexical access partially based onthe morphological organization of words but beforeaccepting this interpretation we have to rule out alterna-tive explanation of the data Thus we are assuming thatthe retarded N400 for stem homographs indicates initialmorphological processing but this could also be ex-plained as simply the result of the orthographic relationbetween the prime and the target The next experimenttries to test this possibility

EXPERIMENT 2 ORTHOGRAPHIC PRIMING

Stem homographs such as focafoco orthographicallyoverlap in their stem morphemes (eg foc-) But in factfoc- in foca is a different morpheme from foc- in focoThis might be the cause of the inhibition found withreaction time measures in Italian (Laudanna et al 19891992) and in Spanish (Allen amp Badecker 1999) and alsothe cause of the N400 retardation in Experiment 1However we need to discard the alternative explanationthat the inhibition is merely caused by the orthographicoverlap Many reaction time studies have demonstratedthat pairs of words sharing all except one letter (neigh-bors) but without any morphological relation producedinhibition These results were obtained with a similarexperimental procedure to our experiments large primeduration (200 msec) and primendashtarget frequency rela-tion (low-frequency primes and high-frequency targets)Furthermore some ERP studies compared unrelatedpriming with formal priming (Doyle Rugg amp Wells1996) and rhyme priming (Radeau Bensson Fonteneauand Luis-Castro 1998) obtaining an N400 amplitudemodulation for formal priming pairs in comparison withunrelated pairs In a wider experimental context wordswith many lexical neighbors produced larger N400 thanwords with fewer lexical neighbors (Holcomb Graingeramp OrsquoRourke 2002) denoting that this component issensitive to orthographic variables of the lexicon

Experiment 2 compared morphological priming (hijohija [sondaughter]) with orthographic priming (rasarana [flatfrog]) If the orthographic condition causesthe same delayed N400 we obtained in Experiment 1 forstem homographs then we might consider that ortho-graphic overlapping rather than morphological processing is the appropriate explanation for all primingeffects observed in Experiment 1 If the orthographiccondition however produces no delayed N400 then wemay accept that morphological processing is a betterexplanation for both morphological priming and stemhomographic priming

Table 2 Peak Latencies (msec) from the Onset of the Targetand Standard Deviations for Unrelated Priming and StemHomographic Priming at the Three Analyzed Sites Fz Cz andPz (Experiment 1)

Type of Priming

Unrelated Stem Homographic PrimingElectrode

Sites Mean SD Mean SD

Fz 372 45 463 80

Cz 385 44 486 101

Pz 438 91 521 100

Domotildenguez de Vega and Barber 601

Note that in this experiment the orthographic pairsmaintain in both words the same final vowel which inSpanish is the gender suffix -a for feminine words and -ofor masculine words No other possibility exists to obtainpairs of words sharing all letters except one and withoutmorphological or stem overlapping A potential problemis that unlike in stem homographs and morphologicalpairs in orthographic pairs a gender andor rhymepriming could take place for instance rasarana (flatfrog) To control this difference between orthographicand morphological priming an additional baseline con-dition (rhyming) was introduced in this experimentpairs of words overlapping only in their final letter suchas cubagoma (barrelrubber)

Results

Behavioral Data

Mean reaction times for correct responses in each condi-tion are presented in Table 3 The morphological condi-tion produced a facilitation of 70 msec on the unrelatedpairs [F(19) = 4115 p lt 001] and 57 msec on the rhymepairs [F(19) = 7228 p lt 001] Also morphological pairswere processed 105 msec faster than orthographicallyrelated words [F(19) = 7640 p lt 001] In contrastorthographically related pairs produced 35 msec of retar-dation on the unrelated words [F(19) = 1020 p = 01]and 48 msec on the rhyming words [F(19) = 2048p = 001] No significant difference was produced be-tween unrelated and rhyming words

ERP Data

Figure 1B clearly shows an overlap of the orthographicpriming and the two control conditions in this windowThe orthographic condition looks in fact undistinguish-able from the unrelated and the rhyming conditionsThe three waves show a negative deflection peakingaround 400 msec after target presentation Howevermorphological priming has a wave with the same shapeobserved in Experiment 1 namely a positivity (above5 mv) starting at 250 msec and remaining throughout thethree windows

250ndash450 msec

The positive wave of the morphological condition dif-fers significantly from the negativity of unrelated words

[F(19) = 3649 p lt 001] from the control rhymingwords [F(19) = 6308 p lt 001] and from the ortho-graphic priming [F(19) = 6950 p lt 001] Morpholog-ical priming in the unrelated pairs interacted with theelectrode factor [F(218) = 477 p lt 05] but this effectwas lost when normalized [F(218) = 286 p = 08]Also morphological priming in the rhyming words in-teracted with electrodes [F(218) = 471 p lt 05] theeffect being marginal when normalized [F(218) = 318p = 066] Although orthographic priming is not signif-icantly different from unrelated or rhyming pairs therewas a difference at certain electrode sites [F(218) = 453p = 05] Orthographic priming was more negative thanrhyming priming at the Pz site however when thisdifference was analyzed electrode by electrode nosignificant effect was reached

350ndash450 msec

In this window morphological priming maintained thepositivity in the unrelated words [F(19) = 5671p lt 001] rhyming words [F(19) = 7308 p lt 001]and orthographically related words [F(19) = 6353p lt 001] No other difference was significant

450ndash650 msec

Morphological pairs show a significant difference withunrelated pairs [F(19) = 606 p lt 05] larger at thefrontal site [F(218) = 511 p lt 05 normalized F = 592p lt 05] and also with rhyming pairs [F(19) = 900p lt 05] especially at the Fz site [F(218) = 627 p lt 01normalized F = 640 p lt 01] This result contrasts withthe central or parietal localization of the morphologicaleffect at the previous epochs

Orthographic pairs after their maximum negative peakat 400 msec still show a very slow recovery of positivity Infact the wave for orthographic pairs is significantly morenegative than for morphological pairs [F(19) = 3293p lt 001] and rhyming pairs [F(19) = 621 p lt 05]There were no orthographic priming effects when mea-sured on the totally unrelated pairs

Discussion

The unrelated priming condition produced a standardN400 whereas the morphological priming conditionshowed a clear positive wave in the same temporalregion This positivity found in both Experiments 1and 2 suggests that the semantic integration betweenprime and target is facilitated when they share the samemorpheme The only variation with respect to Experi-ment 1 was that the effect of morphology was alsovisible at the late temporal window and was larger atthe anterior site Fz (see Figure 1B)

Orthographic pairs on the other hand tend to bemore negative than rhyming pairs at the early and late

Table 3 Mean Reaction Times (msec) and StandardDeviations in Experiment 2 for Each Experimental Condition

Type of Priming Mean (SD)

Morphological Orthographic Rhyming Unrelated

668 (74) 773 (75) 725 (83) 738 (87)

602 Journal of Cognitive Neuroscience Volume 16 Number 4

windows very different from the positive tendency ofmorphological pairs The late effect is quite similar tothe stem homographic effect found in Experiment 1 forthe same window of 450ndash650 msec However unlikeorthographic pairs stem homograph pairs also pro-duced a modulation of N400 at the 250ndash450 msecwindow overlapping the morphological wave duringthe first 100 msec Putting together the results of Experi-ments 1 and 2 a clear dissociation emerges betweenstem homographs and orthographic pairs at the earlywindow despite the fact that in both conditions theprime and the target share the same number of lettersThus stem homographs were initially processed asmorphologically related words (early positivity) startinga sort of morphological lsquolsquogarden pathrsquorsquo (Figure 1A) Inother words for stem homographs a morphologicalstage of stem segmentation takes place which is similarto that for morphological pairs and that occurs as a firstcomponent between 250 and 350 msec

The orthographic pairs however were initially pro-cessed as unrelated pairs until 450 msec (see Figure 1B)and from this point they produced a late negativity atthe window of 450ndash650 msec possibly as a consequenceof orthographic overlapping rather than morphologicalprocessing

Concerning reaction times facilitation was found formorphological pairs in Experiment 1 and Experiment 2an inhibition effect was observed for orthographic pairsin Experiment 2 replicating some other studies whichfound inhibition for neighbors

In conclusion the morphological condition producesfacilitation on RTs and attenuation of N400 stem homo-graphs produce a delayed N400 and the orthographiccondition produces inhibition on RTs and a late nega-tivity in ERP These differences support an early mor-phological treatment of words when the prime and thetarget have an orthographically similar stem such asoccurs for morphological pairs and stem homographsThis morphological processing cannot be merely re-duced to orthographic overlapping

EXPERIMENT 3 SYNONYM PRIMING

The previous experiments demonstrate that stem ho-mographic inhibition is a morphological phenomenonthat cannot be confused with orthographic formal prim-ing Experiment 3 aims to rule out another confoundingissue To what extent could morphological priming bereduced to a semantic priming effect Notice that unlikeorthographic and stem homographic words morpho-logical pairs are semantically related (primes and targetsconstitute the feminine and the masculine version of thesame root morpheme) To address this question asynonyms condition was introduced including ortho-graphically and morphologically unrelated words thatmaintain however a semantic relation and a change ingender suffix similar to those of morphological pairs for

instance ciriovela (candle masculinecandle feminine)or varapalo (stick masculinestick feminine) Usingthese pairs it is possible to know whether the sourceof the observed morphological priming is simply thesemantic relation between the prime and the target Ifthis is the case the ERP waves for the morphological andthe synonym condition would be similar otherwise themorphological wave would differ from that of the syn-onym pairs

Results

Behavioral Data

Mean reaction times for correct responses in eachcondition are presented in Table 4 In comparison withunrelated pairs both morphological pairs [F(110) =899 p lt 001] and synonyms produce facilitation[F(110) = 2653 p lt 001] Synonyms produced aneven larger facilitation of 56 msec than morphologicalpairs (45 msec) although no significant difference wasfound between them

ERP data

Again unrelated words produced the typical N400whereas morphological pairs and synonyms showed aclear attenuation of N400 (see Figure 1C)

250ndash350 msec

Morphological priming was more positive than unrelat-ed pairs [F(110) = 1903 p = 001] The contrastbetween synonyms and unrelated words was significant[F(110) = 582 p lt 05] and also the differencebetween synonyms and morphologically related words[F(110) = 1051 p lt 01]

350ndash450 msec

At this window the morphological priming persists in apositive tendency significantly different from the nega-tive curve for unrelated words [F(110) = 4504p lt 001] and more important at the central andposterior sites [F(220) = 1032 p = 00l normalizedF(220) = 937 p = 001] The attenuation of the N400for synonyms obtained at the previous window does notnow reach statistical significance as perhaps a largervariability of the data was obscuring this effect

Table 4 Mean Reaction Times (msec) and StandardDeviations in Experiment 3 for Each Experimental Condition

Type of Priming Mean (SD)

Morphological Synonym Unrelated

687 (146) 676 (136) 732 (113)

Domotildenguez de Vega and Barber 603

450ndash650 msec

The morphological positivity stabilizes whereas thesemantic priming wave becomes even more positivepeaking around 600 msec The only significant differ-ence was between synonyms and unrelated pairs[F(110) = 501 p lt 05] especially important at Czand Pz [F(220) = 556 p lt 05 normalizedF(220) = 578 p lt 05] No statistical difference wasfound between semantic and morphological primingbut there is a small region in the middle of the epochin which differences are quite visible Therefore a sepa-rate analysis was carried out for this region in the win-dow 500 to 600 msec The results showed that synonymsmaintain a significant difference with unrelated words atthe posterior sites [F(220) = 433 p lt 05] but inaddition synonyms were also significantly more positivethan morphological words [F(110) = 649 p lt 05]

Discussion

Semantic integration in the early epoch seems easier formorphologically related words than for synonyms Inboth cases the tendency to positivity appears butmorphologically related words produce an even morepronounced slope that stabilizes approximately at 10 mv400 msec after the target onset In contrast synonymssurpass the positivity of morphologically related wordsat the third window about 600 msec These effects areprobably caused by different processes In the earlyepoch morphological pairs are more effectively inte-grated than synonyms because the synonym target maynot take advantage of the previous stem treatment ofthe synonym prime The morphological treatment isonly successful for morphological pairs which are morerapidly integrated

GENERAL DISCUSSION

This ERP study made a systematic contrast betweenmorphological priming and three other sources of pos-sible confounding stem homographic orthographicand synonym priming The electrophysiological wavefor morphological priming was clearly dissociated fromthe waves observed for the other priming conditionssuggesting that during lexical access readers apply agenuine rule-based process of stem segmentation In allthree experiments the morphologically related wordsshowed a clear modulation of N400 in the 250ndash450 msecepoch maintaining the positivity later (see Figure 1) Theinitial overlap between morphological and stem homo-graphic waves was interpreted as an attempt to apply arule-based procedure to segment the word stems and tointegrate them This action is effective for morphologicalpriming whereas the delayed N400 indicates an integra-tion failure for stem homographs An explanation for this

is that stem homographs like morphologically relatedwords have the same orthographic root in spite of notbeing morphological relatives

The results support for Spanish word processing atime-course starting with morphological parsing of wordstem and affixes followed by lexical contact and finallyproceeding to wordndashword meaning integration (Mar-slen-Wilson amp Tyler 1998 see Domotildenguez et al 2002for similar conclusions with RT measures) Our ap-proach is compatible with the AAM model (Caramazzaet al 1985 1988) presented in the introduction to thisarticle and basically coinciding with a model defendedby Allen and Badecker (2002) They propose a two-stagemodel to explain the inhibitory effect they obtained onreaction times with stem homographs The form of thestems is represented at the lexeme level whereas anabstract representation of the stem including grammat-ical class and meaning information is located at thelemma level Stem homographs use a common repre-sentation at the lexeme level and different representa-tions at the lemma level where competition for lexicalselection takes place This model could be applied toexplain our ERP results Stem homographic pairs initiallyaccede to a common lexeme unit and this explains whyat the 250ndash350 msec window the stem homographicpriming and the morphological priming waves overlapHowever after 350 msec the lemma contact stagebegins Stem homographs produce a delayed N400because they activate two different lemmas resultingin inhibition However morphological pairs maintain alate positivity because they activate the same lemmaproducing facilitation

As to orthographic pairs these do not share a lexemerepresentation and therefore do not show an earlyfacilitation (positivity at the 250ndash350 msec window)When the lemma level is reached orthographic pairsare processed as ordinary unrelated words that is theyactivate two different lemmas showing the standardN400 effect (neither activation nor inhibition)

Finally when synonym words are presented the lexicalselectionshouldbe facilitated because they share thesamelemma Therefore as for morphological pairs a largepositivity could be predicted However the time courseof this positivity was rather different between morpholog-ical and synonym pairs (see Figure 1C) Morphologicalpriminginvolvesanearly positivitymaintainedthroughoutboth temporal windows because the facilitation is pro-duced both at the lexeme and at the lemma stage Incontrast synonyms produce a slow developing positivitywhich reaches maximum at the late window because onlyat the lemma stage does facilitation take place

This is not the only theoretical framework to interpretthese experimental results It might also be possible toimplement our proposal in terms of the InteractiveActivation Models on the condition of including mor-phological units midway between letters and meanings(Taft 1994 Taft amp Zhu 1995) In fact Allen and

604 Journal of Cognitive Neuroscience Volume 16 Number 4

windows very different from the positive tendency ofmorphological pairs The late effect is quite similar tothe stem homographic effect found in Experiment 1 forthe same window of 450ndash650 msec However unlikeorthographic pairs stem homograph pairs also pro-duced a modulation of N400 at the 250ndash450 msecwindow overlapping the morphological wave duringthe first 100 msec Putting together the results of Experi-ments 1 and 2 a clear dissociation emerges betweenstem homographs and orthographic pairs at the earlywindow despite the fact that in both conditions theprime and the target share the same number of lettersThus stem homographs were initially processed asmorphologically related words (early positivity) startinga sort of morphological lsquolsquogarden pathrsquorsquo (Figure 1A) Inother words for stem homographs a morphologicalstage of stem segmentation takes place which is similarto that for morphological pairs and that occurs as a firstcomponent between 250 and 350 msec

The orthographic pairs however were initially pro-cessed as unrelated pairs until 450 msec (see Figure 1B)and from this point they produced a late negativity atthe window of 450ndash650 msec possibly as a consequenceof orthographic overlapping rather than morphologicalprocessing

Concerning reaction times facilitation was found formorphological pairs in Experiment 1 and Experiment 2an inhibition effect was observed for orthographic pairsin Experiment 2 replicating some other studies whichfound inhibition for neighbors

In conclusion the morphological condition producesfacilitation on RTs and attenuation of N400 stem homo-graphs produce a delayed N400 and the orthographiccondition produces inhibition on RTs and a late nega-tivity in ERP These differences support an early mor-phological treatment of words when the prime and thetarget have an orthographically similar stem such asoccurs for morphological pairs and stem homographsThis morphological processing cannot be merely re-duced to orthographic overlapping

EXPERIMENT 3 SYNONYM PRIMING

The previous experiments demonstrate that stem ho-mographic inhibition is a morphological phenomenonthat cannot be confused with orthographic formal prim-ing Experiment 3 aims to rule out another confoundingissue To what extent could morphological priming bereduced to a semantic priming effect Notice that unlikeorthographic and stem homographic words morpho-logical pairs are semantically related (primes and targetsconstitute the feminine and the masculine version of thesame root morpheme) To address this question asynonyms condition was introduced including ortho-graphically and morphologically unrelated words thatmaintain however a semantic relation and a change ingender suffix similar to those of morphological pairs for

instance ciriovela (candle masculinecandle feminine)or varapalo (stick masculinestick feminine) Usingthese pairs it is possible to know whether the sourceof the observed morphological priming is simply thesemantic relation between the prime and the target Ifthis is the case the ERP waves for the morphological andthe synonym condition would be similar otherwise themorphological wave would differ from that of the syn-onym pairs

Results

Behavioral Data

Mean reaction times for correct responses in eachcondition are presented in Table 4 In comparison withunrelated pairs both morphological pairs [F(110) =899 p lt 001] and synonyms produce facilitation[F(110) = 2653 p lt 001] Synonyms produced aneven larger facilitation of 56 msec than morphologicalpairs (45 msec) although no significant difference wasfound between them

ERP data

Again unrelated words produced the typical N400whereas morphological pairs and synonyms showed aclear attenuation of N400 (see Figure 1C)

250ndash350 msec

Morphological priming was more positive than unrelat-ed pairs [F(110) = 1903 p = 001] The contrastbetween synonyms and unrelated words was significant[F(110) = 582 p lt 05] and also the differencebetween synonyms and morphologically related words[F(110) = 1051 p lt 01]

350ndash450 msec

At this window the morphological priming persists in apositive tendency significantly different from the nega-tive curve for unrelated words [F(110) = 4504p lt 001] and more important at the central andposterior sites [F(220) = 1032 p = 00l normalizedF(220) = 937 p = 001] The attenuation of the N400for synonyms obtained at the previous window does notnow reach statistical significance as perhaps a largervariability of the data was obscuring this effect

Table 4 Mean Reaction Times (msec) and StandardDeviations in Experiment 3 for Each Experimental Condition

Type of Priming Mean (SD)

Morphological Synonym Unrelated

687 (146) 676 (136) 732 (113)

Domotildenguez de Vega and Barber 603

450ndash650 msec

The morphological positivity stabilizes whereas thesemantic priming wave becomes even more positivepeaking around 600 msec The only significant differ-ence was between synonyms and unrelated pairs[F(110) = 501 p lt 05] especially important at Czand Pz [F(220) = 556 p lt 05 normalizedF(220) = 578 p lt 05] No statistical difference wasfound between semantic and morphological primingbut there is a small region in the middle of the epochin which differences are quite visible Therefore a sepa-rate analysis was carried out for this region in the win-dow 500 to 600 msec The results showed that synonymsmaintain a significant difference with unrelated words atthe posterior sites [F(220) = 433 p lt 05] but inaddition synonyms were also significantly more positivethan morphological words [F(110) = 649 p lt 05]

Discussion

Semantic integration in the early epoch seems easier formorphologically related words than for synonyms Inboth cases the tendency to positivity appears butmorphologically related words produce an even morepronounced slope that stabilizes approximately at 10 mv400 msec after the target onset In contrast synonymssurpass the positivity of morphologically related wordsat the third window about 600 msec These effects areprobably caused by different processes In the earlyepoch morphological pairs are more effectively inte-grated than synonyms because the synonym target maynot take advantage of the previous stem treatment ofthe synonym prime The morphological treatment isonly successful for morphological pairs which are morerapidly integrated

GENERAL DISCUSSION

This ERP study made a systematic contrast betweenmorphological priming and three other sources of pos-sible confounding stem homographic orthographicand synonym priming The electrophysiological wavefor morphological priming was clearly dissociated fromthe waves observed for the other priming conditionssuggesting that during lexical access readers apply agenuine rule-based process of stem segmentation In allthree experiments the morphologically related wordsshowed a clear modulation of N400 in the 250ndash450 msecepoch maintaining the positivity later (see Figure 1) Theinitial overlap between morphological and stem homo-graphic waves was interpreted as an attempt to apply arule-based procedure to segment the word stems and tointegrate them This action is effective for morphologicalpriming whereas the delayed N400 indicates an integra-tion failure for stem homographs An explanation for this

is that stem homographs like morphologically relatedwords have the same orthographic root in spite of notbeing morphological relatives

The results support for Spanish word processing atime-course starting with morphological parsing of wordstem and affixes followed by lexical contact and finallyproceeding to wordndashword meaning integration (Mar-slen-Wilson amp Tyler 1998 see Domotildenguez et al 2002for similar conclusions with RT measures) Our ap-proach is compatible with the AAM model (Caramazzaet al 1985 1988) presented in the introduction to thisarticle and basically coinciding with a model defendedby Allen and Badecker (2002) They propose a two-stagemodel to explain the inhibitory effect they obtained onreaction times with stem homographs The form of thestems is represented at the lexeme level whereas anabstract representation of the stem including grammat-ical class and meaning information is located at thelemma level Stem homographs use a common repre-sentation at the lexeme level and different representa-tions at the lemma level where competition for lexicalselection takes place This model could be applied toexplain our ERP results Stem homographic pairs initiallyaccede to a common lexeme unit and this explains whyat the 250ndash350 msec window the stem homographicpriming and the morphological priming waves overlapHowever after 350 msec the lemma contact stagebegins Stem homographs produce a delayed N400because they activate two different lemmas resultingin inhibition However morphological pairs maintain alate positivity because they activate the same lemmaproducing facilitation

As to orthographic pairs these do not share a lexemerepresentation and therefore do not show an earlyfacilitation (positivity at the 250ndash350 msec window)When the lemma level is reached orthographic pairsare processed as ordinary unrelated words that is theyactivate two different lemmas showing the standardN400 effect (neither activation nor inhibition)

Finally when synonym words are presented the lexicalselectionshouldbe facilitated because they share thesamelemma Therefore as for morphological pairs a largepositivity could be predicted However the time courseof this positivity was rather different between morpholog-ical and synonym pairs (see Figure 1C) Morphologicalpriminginvolvesanearly positivitymaintainedthroughoutboth temporal windows because the facilitation is pro-duced both at the lexeme and at the lemma stage Incontrast synonyms produce a slow developing positivitywhich reaches maximum at the late window because onlyat the lemma stage does facilitation take place

This is not the only theoretical framework to interpretthese experimental results It might also be possible toimplement our proposal in terms of the InteractiveActivation Models on the condition of including mor-phological units midway between letters and meanings(Taft 1994 Taft amp Zhu 1995) In fact Allen and

604 Journal of Cognitive Neuroscience Volume 16 Number 4

450ndash650 msec

The morphological positivity stabilizes whereas thesemantic priming wave becomes even more positivepeaking around 600 msec The only significant differ-ence was between synonyms and unrelated pairs[F(110) = 501 p lt 05] especially important at Czand Pz [F(220) = 556 p lt 05 normalizedF(220) = 578 p lt 05] No statistical difference wasfound between semantic and morphological primingbut there is a small region in the middle of the epochin which differences are quite visible Therefore a sepa-rate analysis was carried out for this region in the win-dow 500 to 600 msec The results showed that synonymsmaintain a significant difference with unrelated words atthe posterior sites [F(220) = 433 p lt 05] but inaddition synonyms were also significantly more positivethan morphological words [F(110) = 649 p lt 05]

Discussion

Semantic integration in the early epoch seems easier formorphologically related words than for synonyms Inboth cases the tendency to positivity appears butmorphologically related words produce an even morepronounced slope that stabilizes approximately at 10 mv400 msec after the target onset In contrast synonymssurpass the positivity of morphologically related wordsat the third window about 600 msec These effects areprobably caused by different processes In the earlyepoch morphological pairs are more effectively inte-grated than synonyms because the synonym target maynot take advantage of the previous stem treatment ofthe synonym prime The morphological treatment isonly successful for morphological pairs which are morerapidly integrated

GENERAL DISCUSSION

This ERP study made a systematic contrast betweenmorphological priming and three other sources of pos-sible confounding stem homographic orthographicand synonym priming The electrophysiological wavefor morphological priming was clearly dissociated fromthe waves observed for the other priming conditionssuggesting that during lexical access readers apply agenuine rule-based process of stem segmentation In allthree experiments the morphologically related wordsshowed a clear modulation of N400 in the 250ndash450 msecepoch maintaining the positivity later (see Figure 1) Theinitial overlap between morphological and stem homo-graphic waves was interpreted as an attempt to apply arule-based procedure to segment the word stems and tointegrate them This action is effective for morphologicalpriming whereas the delayed N400 indicates an integra-tion failure for stem homographs An explanation for this

is that stem homographs like morphologically relatedwords have the same orthographic root in spite of notbeing morphological relatives

The results support for Spanish word processing atime-course starting with morphological parsing of wordstem and affixes followed by lexical contact and finallyproceeding to wordndashword meaning integration (Mar-slen-Wilson amp Tyler 1998 see Domotildenguez et al 2002for similar conclusions with RT measures) Our ap-proach is compatible with the AAM model (Caramazzaet al 1985 1988) presented in the introduction to thisarticle and basically coinciding with a model defendedby Allen and Badecker (2002) They propose a two-stagemodel to explain the inhibitory effect they obtained onreaction times with stem homographs The form of thestems is represented at the lexeme level whereas anabstract representation of the stem including grammat-ical class and meaning information is located at thelemma level Stem homographs use a common repre-sentation at the lexeme level and different representa-tions at the lemma level where competition for lexicalselection takes place This model could be applied toexplain our ERP results Stem homographic pairs initiallyaccede to a common lexeme unit and this explains whyat the 250ndash350 msec window the stem homographicpriming and the morphological priming waves overlapHowever after 350 msec the lemma contact stagebegins Stem homographs produce a delayed N400because they activate two different lemmas resultingin inhibition However morphological pairs maintain alate positivity because they activate the same lemmaproducing facilitation

As to orthographic pairs these do not share a lexemerepresentation and therefore do not show an earlyfacilitation (positivity at the 250ndash350 msec window)When the lemma level is reached orthographic pairsare processed as ordinary unrelated words that is theyactivate two different lemmas showing the standardN400 effect (neither activation nor inhibition)

Finally when synonym words are presented the lexicalselectionshouldbe facilitated because they share thesamelemma Therefore as for morphological pairs a largepositivity could be predicted However the time courseof this positivity was rather different between morpholog-ical and synonym pairs (see Figure 1C) Morphologicalpriminginvolvesanearly positivitymaintainedthroughoutboth temporal windows because the facilitation is pro-duced both at the lexeme and at the lemma stage Incontrast synonyms produce a slow developing positivitywhich reaches maximum at the late window because onlyat the lemma stage does facilitation take place

This is not the only theoretical framework to interpretthese experimental results It might also be possible toimplement our proposal in terms of the InteractiveActivation Models on the condition of including mor-phological units midway between letters and meanings(Taft 1994 Taft amp Zhu 1995) In fact Allen and

604 Journal of Cognitive Neuroscience Volume 16 Number 4

Badeckerrsquos (2002) model and their explanations interms of activation and inhibition suggests interactiveand partially distributed processing

In contrast with these two-step models totally dis-tributed connectionist models defend the position thatthe morphological properties of words emerge unique-ly from the correlation between orthographic andsemantic features of morphological relatives and there-fore they reject a genuine morphological level ofprocessing (Seidenberg amp Gonnerman 2000 Joanisseamp Seidenberg 1999) We think that distributed modelsmight not account easily for our results because theywould predict a similar priming effect for stem homo-graphic priming as for neighbor orthographic priming

In short the use of ERPs to study the access andrepresentation of morphologically related words allowedsupport for a model of word recognition that starts bysegmenting the stem and then proceeds to select anappropriate semantic representation at a higher levelThese two levels could be related to the two-step modelproposed by Allen and Badecker (1999 2002) A secondconsequence derived from the results is that N400modulation could be reflecting two different subcompo-nents an early component (250ndash350 msec) at the level ofmorphological decomposition (the lexeme level) and alater component related to lexical and meaning selection(the lemma level)

METHODS

The three experiments conducted share the same expe-rimental procedures EEG-recording protocol signaltreatment and statistical analysis as well as some ofthe materials We will describe them in the followingsections

Materials

The morphological priming condition and two othercontrol conditions were common to all three experi-ments whereas the stem homographic orthographicand synonym conditions were introduced in Experi-ments 1 2 and 3 respectively All conditions consistedof 42 pairs of primendashtarget stimuli Morphological pairsshared the same stem and varied only their gendersuffix the final letter -a for feminine and -o for mascu-line for example hijahijo (daughterson) Unrelatedwords involved words with one standard suffix ofgender (-o or -a) and differing in all their letters andmeanings for instance pavometa (turkeygoal) Anoth-er list of nonexperimental fillers whose final letters werenot the suffixes -a or -o was also composed azarpais(chancendashcountry) All experiments also included nonex-perimental wordndashpseudoword pairs Eighty-four ofthese shared all letters except one (lagolaga [lakelaga]) and the other eighty-four were totally unrelated(besofoja [kissfoja]) The distribution of the ortho-

graphic similarity in the wordndashpseudoword pairs andwordndashword pairs aims to reduce the possibility thatparticipants may guess whether the target is a word ornot merely by observing the orthographic overlapbetween prime and target

In addition to the above conditions Experiment 1introduced stem homographic pairs which shared allthe letters of their stems but differed morphologicallyand therefore semantically for example focafoco (sealfloodlight) The specific stimuli of Experiment 2 were theorthographic pairs and the unrelated rhyming pairs Inthe orthographic condition primes and targets shared allletters except one that was not the final letter forexample rasarana (flatfrog) In this way prime andtarget were neighbor words but not stem homographsIn the rhyming condition unrelated words differed in allletters except the final letter for instance cubagoma(barrelrubber) This condition may be a better baselinethan totally unrelated pairs because neighbor pairsshared the last letter Finally the specific condition ofExperiment 3 was synonym pairs Primes and targetswere synonym words differing in gender for exampleciriovela (candle [masculine]candle [feminine])

The experimental stimuli were Spanish nouns oradjectives of four five or six letters long The lexicalfrequency of targets obtained from Alameda and Cuetos(1995) was controlled across conditions and theirmeans are shown in Table 5 The frequency of the primewas maintained lower than the frequency of the targetThis condition has been demonstrated effectively toproduce inhibition in the recognition of targets ortho-graphically related with their primes at least whenexposure time for primes is long enough to be con-sciously recognized This is the case for our primes thatwere exposed for 250 msec Each subject received210 word pairs (168 experimental) and 168 wordndashnonword pairs The characteristics of the stimuli maybe seen in Table 5 and the total list of experimental sti-muli is offered in the Appendix

Procedure

Each trial was organized as follows (1) a pure tone of200 msec indicated the beginning of the trial (2) a blankinterval of 300 msec 3) a fixation point in the center ofthe screen for 200 msec (4) a blank interval of 500 msec(5) the prime was presented in lowercase letters for200 msec (6) an interstimulus blank interval for100 msec (7) the target word or nonword in uppercaseletters remained 1 sec or until the participant re-sponded (8) the word PARPADEA (BLINK) appearedfor 2 sec (9) and finally another blank interval remainedfor 1500 msec until the next trial Participants were toldthat a word would appear in the center of the screenfollowed by a second uppercase letter string the targetstimulus They were asked to pay attention to bothwords and were required to indicate whether the target

Domotildenguez de Vega and Barber 605

was or was not a word by pressing one of two buttons onthe response keyboard (lexical decision task) Theirright index finger was used for the button lsquolsquoYESrsquorsquo andtheir left index finger for the button lsquolsquoNOrsquorsquo Participantswere also asked to maintain fixation on the center of thescreen and to avoid eye movements during the intervalfrom the fixation point until the appearance of the wordPARPADEA (BLINK)

EEG Recording

The EEG was recorded from 15 thin electrodes mountedon an electrode elastic cap according to the 10 stan-dard system of the American ElectroencephalographicSociety and amplified using a Medicid-4 system Therecording sites were placed on F7 Fz F8 FT7 FT8 T3C5 Cz C6 T4 CP5 CP6 Pz O1 and O2 All EEGchannels were referred to linked mastoids Interelec-trode impedance was kept below 5 Klaquo The bio-signalswere recorded in the 005ndash50 Hz band and were digi-tized continuously with a sampling rate of 200 Hz TwoEOG electrodes one suborbital left eye and another onthe external canthi of the right eye monitored forvertical and horizontal eye movements and blinks

Analyses

Baseline correction was performed using the averageEEG activity in the 100 msec preceding the onset of thetarget word as a reference signal value After the baselinecorrection trials with ocular artifacts exceeding 80 mnwere rejected Separate ERPs were averaged for eachexperimental condition for each of participants and foreach of the electrode sites

Three epochs were selected for analyses A visualinspection of waves evidenced that 250 msec was ap-proximately the point at which all the related experi-mental conditions began to differ from the unrelatedcondition in the three experiments From this pointthere is a positive peak that could be considered a firstcomponent followed by a negative tendency to conforma second component which peaks around 400 msec The

first epoch to analyze was extended from 250 to350 msec The second which includes the N400 beganat 350 msec and ended at 450 msec Finally the thirdepoch was established from 450 to 650 msec in order toclarify how some effects initiated in the previous windowevolved or the emergence of new later effects

Multiple analysis of variance (MANOVA) was appliedon the average voltage window for each experimentalcondition Analyses were carried out only on the Fz Czand Pz of midline sites because the effects were morepowerful on the central axis Because no laterality effectswere evident in the total wave distribution the left andright electrode sites were discarded from the analysesAll interactions described in the manuscript betweenelectrode sites and conditions were normalized with theprocedure established by McCarthy and Wood (1985)

MANOVAs on reaction times excluding errors werecarried out by participants excluding latencies exceed-ing or not reaching 25 standard deviations from themean reaction time of each category Because the dataof interest were the differences between pairs of con-ditions MANOVAs were applied to calculate significantdifferences between the morphological and the unrela-ted condition between the morphological and the stemhomographs condition and finally between the stemhomographs and the unrelated condition These analy-ses constitute a primary source of evidence to appreciateto what extent behavioral measures and ERPs are eithercomplementary or conflicting

Participants

Thirty-two psychology undergraduates at the Universityof La Laguna Spain participated in the experiments forcourse credit 11 in Experiment 1 (8 women and 3 menage range 18ndash26 years mean 207) 10 in Experiment 2(9 women and 1 man age range 19ndash33 mean 229)and 11 in Experiment 3 (9 women and 2 men agerange 20ndash28 mean 213) All were Spanish native speak-ers with normal or corrected-to-normal vision andclassified as right-handed according to the EdinburghHandedness Inventory (Oldfield 1971)

Table 5 Lexical Frequency Longitude and Target Gender in All Experimental Conditions

Prime Frequency Target Frequency No of 456 Letter Words -o-a Target Words

Stem homographs 18 117 21174 2121

Orthographic 13 114 23154 1923

Semantic 55 110 22164 1626

Morphological 52 115 22164 1626

Unrelated rhyming 28 113 22164 2121

Unrelated 25 112 22164 2022

606 Journal of Cognitive Neuroscience Volume 16 Number 4

MORPHOLOGICAL STEM HOMOGRAPHS NEIGHBORS SYNONYMS UNRELATED RHYME

acida ACIDO acero ACERA bruma BROMA forro FUNDA ficha JUNIO noria RUBIA

fino FINA acta ACTO bala BATA ganso OCA limbo CARTA ronco TEXTO

neta NETO arca ARCO topo TOMO asno BURRA cerdo FALSA plomo RASGO

bello BELLA bando BANDA pauta PAUSA rabo COLA rollo DUCHA prado DISCO

guapo GUAPA barbo BARBA turba TUMBA gluteo NALGA pavo META caro SACO

maja MAJO bazo BAZA rifa RIMA laxo FLOJA cava YESO zaga LUPA

mona MONO cano CANA lacio LABIO cirio VELA hucha TIBIO sidra TALLA

hado HADA cepo CEPA bulo BUZO jeta MORRO dina POMO cana FOFA

rica RICO cero CERA rueca RUEDA clavo PUNTA ruda POZO gula PATA

soso SOSA coco COCA pana PAJA jaca POTRO pasmo TROPA tieso GLOBO

moza MOZO copo COPA cota COLA edredon MANTA lija DEDO fuga VIVA

preso PRESA costo COSTA cuna CURA vara PALO treta REINO bizca CARGA

moro MORA coto COTA rasa RANA lazada NUDO jira TAJO galo CEBO

sola SOLO faja FAJO credo CRUDO esposa MARIDO peto LONA cazo HIPO

vana VANO filo FILA seto SENO obesa GORDO vaca FIJO nena GOTA

gata GATO foca FOCO trepa TRIPA sortija ANILLO halo SOGA apto PINO

india INDIO gesta GESTO cieno CIELO fosa HOYO poema CAMPO silla MEDIA

corta CORTO grada GRADO daga DAMA solida DURO video LUCHA barro TECHO

lindo LINDA grifa GRIFO musgo MUSLO loma CERRO liga CELO cuba GOMA

amiga AMIGO labia LABIO calco CALVO espiritu ALMA terca QUESO horma VENTA

malo MALA lapo LAPA rosco RISCO zoco PLAZA cisma MANSO larva MONDA

mulo MULA loma LOMO manco MANTO sebo GRASA socio FURIA grato CASCO

mango MANGA maya MAYO foro FOTO oda VERSO lila MURO beca ROSA

llana LLANO micra MICRO lima LIRA mago BRUJA salmo GUASA cauto RUEDO

baja BAJO pala PALO trapo TRAGO modo FORMA surco FIRMA horn PACTO

roja ROJO papo PAPA luto LUJO brisa VIENTO mosto PLENA denso HUECO

seca SECO plato PLATA placa PLAYA pompa LUJO teja PURO jefa MODA

viudo VIUDA plazo PLAZA tela TEMA manto CAPA piso ROPA mito HUMO

burdo BURDA porra PORRO tala TASA rudo BRUTA rezo LOSA veto MIMO

docta DOCTO bombo BOMBA pica PIPA plagio COPIA capo MISA remo OLMO

listo LISTA punta PUNTO lazo LADO sepulcro TUMBA pena TIPO rara COSA

bobo BOBA rata RATO toro TONO caldo SOPA suma PELO pupa ZONA

hija HIJO rejo REJA torta TARTA folio HOJA yugo MASA rizo LORO

honda HONDO rumba RUMBO timo TIRO cima PICO caza HILO gafa BODA

alta ALTO seta SETO aspa ARPA rostro CARA mirra TALLO farra NECIA

obvia OBVIO grupa GRUPO hola HORA trazo RAYA eter OBRA nata MESA

regia REGIO suela SUELO maga MAPA curada SANO nodo LUNA pura CAMA

liso LISA tuba TUBO panza DANZA chucho PERRA tufo GIRA sapo PITOv

aliada ALIADO correa CORREO tarado TAPADO sarao FIESTA eslora NEGADO modosa RETINA

diaria DIARIO garito GARITA calima CALIFA paga SUELDO vivero PALETA rumano COCIDO

fulana FULANA navajo NAVAJA espeso ESPOSO alimento COMIDA dibujo HERIDA divino CASADO

limpia LIMPIO puerto PUERTA cuerno CUERPO heno HIERBA rutina PASADO locura MIRADA

APPENDIX

Type of Priming

Domotildenguez de Vega and Barber 607

Acknowledgments

This research was supported by the grants PB98-0431 andBSO2002-04301-C02-02 by the Ministerio de Ciencia y Tech-nologotildea (Spain) We thank Margaret Gillon Dowens for herrevision of earlier versions of this manuscript

Reprint requests should be sent to Alberto DomotildenguezDepartamento de Psicologotildea Cognitiva Universidad de LaLaguna Campus de Guajara 38205 La Laguna Canary IslandsSpain or via e-mail adominulles

REFERENCES

Alameda J R amp Cuetos F (1995) Diccionario defrecuencias de las unidades linguotildesticas del castellanoOviedo Servicio de Publicaciones de la Universidadde Oviedo

Allen M amp Badecker W (1999) Stem homograph inhibitionand stem allomorphy Representing and processing inflectedforms in a multilevel lexical system Journal of Memory andLanguage 41 105ndash123

Allen M amp Badecker W (2002) Stem homographs andlemma level representations Brain and Language 8179ndash88

Barber H Domotildenguez A amp De Vega M (2002) Human brainpotentials indicate morphological decomposition in visualword recognition Neuroscience Letters 318 149ndash152

Caramazza A Laudanna A amp Romani C (1988) Lexicalaccess and inflectional morphology Cognition 28 297ndash332

Caramazza A Miceli G Silveri M C amp Laudanna A (1985)Reading mechanisms and the organisation of the lexiconEvidence from acquired dyslexia CognitiveNeuropsychology 2 81ndash114

Domotildenguez A Cuetos F amp Segu otilde J (2002) The time courseof inflexional morphological priming Linguotildestics 40235ndash259

Doyle M C Rugg M D amp Wells T (1996) A comparison onthe electrophysiological effects of formal and repetitionpriming Psychophysiology 33 132ndash147

Garcotildea-Albea J E Sanchez-Casas R amp Igoa J M (1998) The

contribution of word form and meaning to languageprocessing in Spanish Some evidence from monolingualand bilingual studies In D Hilbert (Ed) Sentenceprocessing A cross-linguistic approach New YorkAcademic Press

Holcomb Ph J Grainger J amp OrsquoRourke T (2002) Anelectrophysiological study of the effects of orthographicneighborhood size on printed word perception Journalof Cognitive Neuroscience 14 938ndash950

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist modelProceedings of the National Academy of Sciences USA

Laudanna A Badecker W amp Caramazza (1989) Priminghomographic stems Journal of Memory and Language28 53lndash546

Laudanna A Badecker W amp Caramazza (1992) Processinginflectional and derivational morphology Journal ofMemory and Language 31 333ndash348

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Sciences 2 428ndash435

McCarthy G amp Wood Ch C (1985) Scalp distributions ofevent-related potentials An ambiguity associated withanalysis of variance models Electroencephalography andClinical Neuropsychology 62 203ndash208

Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia97 97ndash113

Radeau M Bensson M Fonteneau E amp Luis-Castro S(1998) Semantic repetition and rime priming betweenspoken words Behavioral and electrophysiologicalevidence Biological Psychology 48 183ndash204

Seidenberg M S amp Gonnerman L M (2000) Explainingderivational morphology as the convergence of codesTrends in Cognitive Sciences 4 353ndash361

Taft M (1994) Interactive activation as a framework forunderstanding morphological processing Language andCognitive Processes 9 271ndash294

Taft M amp Zhu X (1995) The representation of boundmorphemes in the lexicon A Chinese studyIn L B Feldman (Ed) Morphological aspects oflanguage processing (pp 293ndash316) Hove UK Erlbaum

608 Journal of Cognitive Neuroscience Volume 16 Number 4

Acknowledgments

This research was supported by the grants PB98-0431 andBSO2002-04301-C02-02 by the Ministerio de Ciencia y Tech-nologotildea (Spain) We thank Margaret Gillon Dowens for herrevision of earlier versions of this manuscript

Reprint requests should be sent to Alberto DomotildenguezDepartamento de Psicologotildea Cognitiva Universidad de LaLaguna Campus de Guajara 38205 La Laguna Canary IslandsSpain or via e-mail adominulles

REFERENCES

Alameda J R amp Cuetos F (1995) Diccionario defrecuencias de las unidades linguotildesticas del castellanoOviedo Servicio de Publicaciones de la Universidadde Oviedo

Allen M amp Badecker W (1999) Stem homograph inhibitionand stem allomorphy Representing and processing inflectedforms in a multilevel lexical system Journal of Memory andLanguage 41 105ndash123

Allen M amp Badecker W (2002) Stem homographs andlemma level representations Brain and Language 8179ndash88

Barber H Domotildenguez A amp De Vega M (2002) Human brainpotentials indicate morphological decomposition in visualword recognition Neuroscience Letters 318 149ndash152

Caramazza A Laudanna A amp Romani C (1988) Lexicalaccess and inflectional morphology Cognition 28 297ndash332

Caramazza A Miceli G Silveri M C amp Laudanna A (1985)Reading mechanisms and the organisation of the lexiconEvidence from acquired dyslexia CognitiveNeuropsychology 2 81ndash114

Domotildenguez A Cuetos F amp Segu otilde J (2002) The time courseof inflexional morphological priming Linguotildestics 40235ndash259

Doyle M C Rugg M D amp Wells T (1996) A comparison onthe electrophysiological effects of formal and repetitionpriming Psychophysiology 33 132ndash147

Garcotildea-Albea J E Sanchez-Casas R amp Igoa J M (1998) The

contribution of word form and meaning to languageprocessing in Spanish Some evidence from monolingualand bilingual studies In D Hilbert (Ed) Sentenceprocessing A cross-linguistic approach New YorkAcademic Press

Holcomb Ph J Grainger J amp OrsquoRourke T (2002) Anelectrophysiological study of the effects of orthographicneighborhood size on printed word perception Journalof Cognitive Neuroscience 14 938ndash950

Joanisse M F amp Seidenberg M S (1999) Impairments inverb morphology after brain injury A connectionist modelProceedings of the National Academy of Sciences USA

Laudanna A Badecker W amp Caramazza (1989) Priminghomographic stems Journal of Memory and Language28 53lndash546

Laudanna A Badecker W amp Caramazza (1992) Processinginflectional and derivational morphology Journal ofMemory and Language 31 333ndash348

Marslen-Wilson W D amp Tyler L K (1998) Rulesrepresentations and the English past tense Trends inCognitive Sciences 2 428ndash435

McCarthy G amp Wood Ch C (1985) Scalp distributions ofevent-related potentials An ambiguity associated withanalysis of variance models Electroencephalography andClinical Neuropsychology 62 203ndash208

Oldfield R C (1971) The assessment and analysis ofhandedness The Edinburgh Inventory Neuropsychologia97 97ndash113

Radeau M Bensson M Fonteneau E amp Luis-Castro S(1998) Semantic repetition and rime priming betweenspoken words Behavioral and electrophysiologicalevidence Biological Psychology 48 183ndash204

Seidenberg M S amp Gonnerman L M (2000) Explainingderivational morphology as the convergence of codesTrends in Cognitive Sciences 4 353ndash361

Taft M (1994) Interactive activation as a framework forunderstanding morphological processing Language andCognitive Processes 9 271ndash294

Taft M amp Zhu X (1995) The representation of boundmorphemes in the lexicon A Chinese studyIn L B Feldman (Ed) Morphological aspects oflanguage processing (pp 293ndash316) Hove UK Erlbaum

608 Journal of Cognitive Neuroscience Volume 16 Number 4