Early, equivalent ERP masked priming effects for regular and irregular morphology

Brain & Language 123 (2012) 81–93

Contents lists available at SciVerse ScienceDirect

Brain & Language

journal homepage: www.elsevier .com/locate /b&l

Early, equivalent ERP masked priming effects for regular and irregular morphology

Joanna Morris a,⇑, Linnaea Stockall b

a School of Cognitive Science, Hampshire College, 893 West St., Amherst, MA 01002, United Statesb Department of Linguistics, Queen Mary, University of London, Mile End Road, London E1 4NS, UK

a r t i c l e i n f o

Article history:Accepted 2 July 2012Available online 20 August 2012

Keywords:Irregular allomorphyMorphologyMorpho-orthographic decompositionVisual word formElectroencephalographyMasked priming

0093-934X/$ - see front matter � 2012 Elsevier Inc. Ahttp://dx.doi.org/10.1016/j.bandl.2012.07.001

⇑ Corresponding author.E-mail addresses: [email protected] (J. M

(L. Stockall).

a b s t r a c t

Converging evidence from behavioral masked priming (Rastle & Davis, 2008), EEG masked priming (Mor-ris, Frank, Grainger, & Holcomb, 2007) and single word MEG (Zweig & Pylkkänen, 2008) experiments hasprovided robust support for a model of lexical processing which includes an early, automatic, visual wordform based stage of morphological parsing that applies to all derivationally affixed words. The mecha-nisms by which regularly (walked, birds) and irregularly (gave, geese) inflected forms are processed areless well established. We combine the masked priming paradigm with EEG recording to directly comparethe ERPs evoked by regularly and irregularly inflected forms. We find equivalent N250 priming effects forboth types of morphological complexity, which argues for rapid, form based morphological parsing of allmorphologically complex word forms.

� 2012 Elsevier Inc. All rights reserved.

1. Introduction

The question of how words like teacher and walked are parsedand recognized has been debated for more than 30 years (Rastle& Davis, 2008; Stockall & Marantz, 2006; Taft & Forster, 1975).On the one hand, these words can easily be parsed into a stemand affix, each of which makes a transparent, predictable contribu-tion to the meaning and grammatical function of the whole word,and each of which is robustly attested in many other forms in thelanguage. These facts motivate models such as that of Taft and For-ster (1975); Taft (2004) or Stockall and Marantz (2006), that in-clude an early, form based, affix-stripping mechanism thatprecedes and feeds into subsequent activation of stored lexicalstems. On the other hand, words like teacher are familiar and fre-quently heard, read and produced, and many words that can beformally parsed into a stem and affix are not so obviously transpar-ent semantically (a folder is not usually understood as ‘a personwho folds’). These facts have motivated a range of models withoutearly, form based decomposition, in which complex familiar wordsare stored whole in the lexicon (Albright & Hayes, 2003; Butter-worth, 1983) or recognized via associative networks with no mor-phological units at all (Rumelhart & McClelland, 1986, chap. 18).

A second, related question is how words like sang or geese, thatdo not follow the regular morphological patterns of the language,are parsed and stored. Models that do not include decompositionfor regular allomorphy also, obviously, do not assume decomposi-tion for irregulars. The same form and meaning similarity based

ll rights reserved.

orris), [email protected]

associations that link walk with its relatives walked, walking, andwalkable link sing with its relatives sang, singing, singer in thesemodels (Devlin, Jamison, Matthews, & Gonnerman, 2004; Rumel-hart & McClelland, 1986, chap. 18; Seidenberg & McClelland, 1989).

By contrast, models that assume rule based decomposition forregulars are split between those, such as the extremely influentialWords & Rules model (Pinker & Prince (1988, chap. II) inter alia)that adopt the whole word storage + similarity associations modelfor irregulars, and those that extend the rule based approach togenerate irregulars as well (Chomsky & Halle, 1968; Halle & Mar-antz, 1994; Stockall & Marantz, 2006; Yang, 2002).

These latter models posit ‘full, across the board, decomposition’(Stockall & Marantz, 2006), with no categorical differences be-tween regular and irregular allomorphy, or between concatenative(affixal) and non-concatenative morphology. The rules in (1 and 2)show how Distributed Morphology (Halle & Marantz, 1994) wouldaccount for ‘sold’, ‘sang’ and ‘slept’. In (1), we have the fragment ofgrammar that generates the three possible allomorphs of the Eng-lish past tense morpheme. The abstract past tense morpheme Tpast

can be realized in any of three ways: ;,/t/, or /d/. The first two op-tions are restricted to a small set of roots, which are listed with theirregular allomorph, while the/d/ option is freely available and ap-plies to any root that does not appear on one of the memorizedlists.

Tpast ! ;fHIT;SING;SIT;BID;...g

! =t=fLEAVE;BEND;BUY ;TEACH;...g

! =d=

ð1Þ

The rules in (1) are sufficient to account for all regular past tenseforms and for all zero-change forms (hit, cut), which simply take

http://dx.doi.org/10.1016/j.bandl.2012.07.001

mailto:[email protected]

mailto:[email protected]

http://dx.doi.org/10.1016/j.bandl.2012.07.001

http://www.sciencedirect.com/science/journal/0093934X

http://www.elsevier.com/locate/b&l

82 J. Morris, L. Stockall / Brain & Language 123 (2012) 81–93

the null allomorph of the past tense, rather than the default/d/.Forms like sang or sold require additional stem vowel adjustmentrules such as the rule in (2-a), which maps the high front vowel/I/to the low front vowel/æ/ in stems like sing and swim, when thosestems occur together with the past tense morpheme.

a: =I=! =æ= TpastfSING;SWIM;RING;...g

b: =i=! =e= TpastfKEEP;SWEEP;LEAP;SLEEP;DREAM;...g

c: . . .

ð2Þ

Thus the irregular past tense form sold is hypothesized to be gener-ated from the verb root

ffiffiffiffiffiffiffisellp

+ the regular/d/ allomorph of the pasttense + a morpho-phonological adjustment rule that maps/e/ ? /o/.The irregular past tense sang, is generated from

ffiffiffiffiffiffiffiffiffising

p+ the null

allomorph of the past tense + the rule /I/ ? /æ/, and irregular sleptfrom

ffiffiffiffiffiffiffiffiffiffiffisleep

p+ the/t/ allomorph of the past tense + the rule /i/ ? /

e/.The models proposed by Yang (2002) and Stockall and Marantz

(2006) use slightly different formalisms that account for the varia-tion in the realization of the past tense morpheme and the stemallomorphy across the �170 irregular roots in slightly differentways, but the critical core claim shared by all these theories is thatprocessing the letter string ‘taught’ involves activating the rootffiffiffiffiffiffiffiffiffiffiffiffi

teachp

, and that this activation is the result of successfully recog-nizing the surface [Ot] sound (or aught letter string) as the output ofa rule that operates over underlying/itS/ sequences.

The prediction of these models, then, is that the early word formrecognition processes must be sensitive not just to the patternsassociated with regular allomorphy, but also to those associatedwith irregular allomorphy. The work of Albright (2002) and Alb-right and Hayes (2003) on ‘islands of reliability’ in the irregular all-omorphy of English and Italian suggests that this claim is not asunlikely as it might otherwise seem. They show that speakers aresensitive to even subtle stochastic subregularities in the morpho-logical patterning of their language. For example, although thestem/past tense alternation found in bleed � bled, lead � led,feed � fed, read � read, and breed � bred is certainly irregular (onlya small set of stems participate in the alternation), it is actuallyhighly reliable (by Albright and Hayes’ counts, 6/7 stems endingin eed have past tense allomorphs that rhyme with bled). This highdegree of consistency in even irregular allomorphy means the pat-tern recognition system responsible for initial form based decom-position (discussed in detail below) could plausibly detectpossible irregular morphemes as well as regular. Given that ablautand similar morphologically conditioned stem alternations arewidely attested across the world’s languages, it would not be sur-prising for an early, word form based, morphological parsingmechanism to be capable of detecting this kind of pattern.

The experiment reported here addresses both the general issueof whether regular morphological inflection is rapidly detected bythe same early, morpho-orthographic parsing mechanisms thathave been argued to operate over derivational affixes, and themore specific question of whether irregular allomorphy is alsoparsed by these same mechanisms, by combining the behavioralmasked priming + lexical decision paradigm with EEG recordings.

1.1. Behavioral masked morphological priming

In the masked priming paradigm, a prime is visually presentedfor only 30–50 ms and the prime is masked by the prior presenta-tion of a masking stimulus, typically a series of hash marks(#####) or random consonant strings (SDFGHJK). The prime iseither immediately followed by another mask or by the targetwhich serves as a backward mask. The short prime duration, aswell as the presence of the forward (and backward) mask, preventsthe subject from consciously perceiving the prime. Moreover, the

close temporal proximity of the prime and target allows little timefor the prime to be processed in isolation from the target, and thusany effects of the prime on responses to the target are presumed toreflect early automatic lexical processes rather than strategic and/or episodic memory effects (Forster & Davis, 1984). When primesare fully visible to the subject, it is unclear whether the priming ef-fects observed are the result of automatic lexical processes, (e.g.activation of a lexical entry) or the result of an episodic memoryof the prime influencing the decision process. Such priming studiesare also vulnerable to the use of predictive strategies by subjects ifthe relationship between prime-target pairs becomes obvious(Feustel, Shiffrin, & Salasoo, 1983; Jacoby, 1983). These concernscan be mitigated with the use of a masked prime (Forster & Davis,1984). The classic finding using this paradigm is faster responsesfor target words preceded by identical or similar prime wordswhen compared to primes that are unrelated to the target.

Feldman, O’Connor, and Del Prado Martı́n (2009) and Rastle andDavis (2008) both provide an overview of the recent literatureusing this paradigm to investigate morphological complexity. Thecore finding is that when a masked prime word is related to thetarget by a possible morpho-orthographic relationship, lexicaldecision times to the target are significantly faster than when thesame target is preceded by an unrelated prime. This result is nicelycaptured in the title of Rastle, Davis, and New (2004)’s paper: ‘‘Thebroth in my brother’s brothel: Morpho-orthographic segmentationin visual word recognition’’. This title derives from the key compar-ison between a prime target pair like brother � broth, in which theprime can plausibly be decomposed into a stem ‘broth’ and a famil-iar suffix ‘-er’, and a prime target pair like brothel � broth, in whichthe prime can not be decomposed, because ‘-el’ is not an attestedsuffix in English. Rastle et al. find that pseudo-complex primessuch as brother and adder significantly facilitate lexical decisionreaction times to their pseudo-stem targets (broth, add), whileprimes which contain the target as an orthographic substring,but do not contain an existing affix (brothel, addict) are not associ-ated with any such facilitation.

Starting with the work of Rastle, Davis, Marslen-Wilson, and Ty-ler (2000), at least 20 experiments using the masked priming para-digm to compare real and pseudo/opaque derivationalmorphology have been published. The consensus of this literatureusing masked priming and the behavioral lexical decision task toinvestigate morphological complexity is strong support for a modelin which there is an early, automatic, obligatory, visual word formbased stage of morpho-orthographic segmentation in which all pos-sible affixes are stripped from their stems (Rastle & Davis, 2008).

There is debate within this literature about whether this earlieststage of morphological processing is purely word form based (Rastle& Davis, 2008) or may involve lexical-semantic factors (Feldmanet al., 2009). Feldman et al. (2009)’s review of the literature finds thatacross the 18 experiments they survey, genuinely morphologicallyrelated primes are, on average, associated with greater facilitationeffects than pseudo-related primes. They argue that this means thatthe semantic overlap between morphological relatives must beplaying a role at the earliest stages of processing, and thus that theearly segmentation is not purely form based. However, an explana-tion for differences in prime magnitude can also be found in stem:w-hole word transition probabilities (Hay, 2001), which are a purelyword form based measure. Stem:whole word transition probabili-ties are calculated as the ratio of the stem frequency to the wholeword frequency. For transparently complex words like taxable orteacher, the probability of the whole word given the stem (tax orteach) is fairly low: tax and teach occur as whole words on theirown, and in combination with many other derivational and inflec-tional morphemes. Conversely, for pseudo-complex words likebrother, the probability of the whole word given the stem is veryhigh: broth is a low frequency word itself, and this stem occurs in

J. Morris, L. Stockall / Brain & Language 123 (2012) 81–93 83

no common derived or inflected words. Thus the transitional proba-bility from the (putative) stem to the whole word is an excellent in-dex of the probability that the whole word is morphologicallycomplex. The higher the stem/whole word transition probability,the lower the likelihood that the whole word is inflected or derivedfrom the stem, and vice versa. Lewis, Solomyak, and Marantz (2011);Lewis and Marantz (submitted for publication) (discussed in Sec-tion 1.3) show that these transition probabilities are, indeed, anexcellent predictor of early neural sensitivity to morphological com-plexity, and that no appeal to semantic factors is required to accountfor differences in early processing responses between transparentand opaque morphologically complex word forms.

1.2. Masked morphological priming from irregularly inflected primes

Meunier and Marslen-Wilson (2000) use the masked primingparadigm to examine the representation of regular and irregularverb forms in French. They find that regular and irregular verbforms prime their infinitive forms equally, suggesting that theyare both represented in a single system, but this experiment didnot include an orthographic overlap condition, thus limiting theconclusions that can be drawn. Other research investigating irreg-ular allomorph priming with the masked priming paradigm (Kielar,Joanisse, & Hare, 2008; Pastizzo & Feldman, 2002) provides addi-tional evidence suggesting that at least some irregulars do primetheir stem targets, but the two studies come to conflicting conclu-sions about exactly which irregulars these are. Pastizzo and Feld-man find priming for high overlap irregulars like gave � give, butnot for low overlap pairs like taught � teach, while Kielar et al. findessentially the opposite pattern.

In an effort to resolve these conflicts, Crepaldi, Rastle, Coltheart,and Nickels (2010) compared genuinely related irregular prime-tar-get pairs like sold� sell and mice�mouse with pseudo-irregular pairslike bold� bell and spice � spouse. Using the masked priming design,Crepaldi et al. found that irregularly inflected primes significantlyfacilitated reaction times to their targets, but that there was no facil-itation for prime-target pairs where the prime and target were simi-lar to one another in the same way as an existing irregular prime andtarget, but had no plausible morphological relationship to one an-other (e.g. bold� bell). Crepaldi et al. did not directly compare regu-lars and irregulars. Their results, and the results of the other studiesreporting significant masked priming effects for irregular allomorphprimes, suggest that irregular allomorphs are very rapidly analyzed,leading to rapid activation of their stem correlates, and thus facilita-tion in the masked priming paradigm. However, the lack of a pseudo-irregular priming effect in the Crepaldi et al. experiments raisesdoubts about whether irregulars are analyzed by the same mor-pho-orthographic mechanisms as regular allomorphs with overt,segmentable affixes. Crepaldi et al. conclude that irregulars are notsegmented morpho-orthographically, and thus that regulars shouldbe associated with greater priming effects than irregulars.

By contrast, theories such as Albright and Hayes (2003), Halleand Marantz (1994) and Stockall and Marantz (2006), make noprincipled distinction between regular and irregular allomorphy,even at the orthographic form level of representation. As discussedabove (Section 1), irregular verbs cluster into what Albright andHayes (2003) term ‘islands of reliability’. Given that many of thepast tense forms that exhibit these sub-regularities are themselveshighly frequent, and thus regularly encountered, these patterns arepredicted to be available to an early visual linguistic pattern recog-nition system.

1.3. Neural correlate of early, morpho-orthographic segmentation

Given the robust body of behavioral evidence for early, visualword form based, morphological segmentation for affixed forms,

the question naturally arrises as to whether the visual word formarea/response identified by Cohen et al. (2000) as being the firstprocessing response specifically sensitive to linguistic word formfeatures in a number of languages (Lehtonen et al., 2007; McCand-liss, Cohen, & Dehaene, 2003; Nakamura, Dehaene, Jobert, Le Bihan,& Kouider, 2005; Tarkiainen, Cornelissen, & Salmelin, 2002) alsoshows sensitivity to morpho-orthographic features.

Zweig and Pylkkänen (2008) addressed this question using a sin-gle word reading paradigm (no primes or masks) and a visual lexicaldecision task. Zweig and Pylkkänen (2008) experiment 1 comparedthree types of word stimuli: (a) transparently complex suffixedforms like teacher (b) unsegmentable monomorphs such as stretchand (c) forms with an -er ending, but that could not be plausibly seg-mented such as winter (there is no stem ‘wint’). Experiment 2 madethe same comparisons with prefixes: complex: refill vs. unsegment-able: throng vs. no-stem: resume. Zweig and Pylkkänen (2008) usedthe Tailerach coordinates of the Visual Word Form Area as reportedin Cohen et al. (2000) to model the MEG activation. In both experi-ments, the forms that could be segmented into two discrete piecesdiffered from those that could not: they evoked greater amplitudesfrom the visual word form area response component peakingapproximately 170 ms after the onset of the visually presentedword. There were no differences between forms like winter or re-sume which contain a substring that could be an affix, but no plausi-ble stem, and unsegmentable monomorphs like stretch or throng.Zweig and Pylkkänen (2008) argue that this increased activationfor morphologically complex words indexes precisely the earlystage of visual word form based morphological segmentation arguedfor from the behavioral masked morphological priming literature.

Solomyak and Marantz (2010); Lewis et al. (2011) and Lewisand Marantz (submitted for publication), in a series of experimentsusing distributed source modeling of the MEG signal evoked bysingle words, have confirmed this finding that activity originatingin the fusiform gyrus is differentially sensitive to morphologicalcomplexity. In these studies, Marantz and his colleagues find thatactivity originating in the left fusiform gyrus and peaking between140 and 190 ms after the onset of a visually presented word (theM170 component) increases as a function of the decomposabilityof that word, but is not modulated by semantic factors.

Lewis et al. (2011) and Lewis and Marantz (submitted for pub-lication) show that the magnitude of the M170 response is, in fact,highly correlated with the stem:whole word transition probabilityof a potentially morphologically complex word. Words with lowtransition probabilities (and thus high probability of being com-plex) are associated with greater M170 amplitudes than wordswith higher TPs (lower probability of being complex). This purelyword form based measure predicts precisely the difference be-tween teacher and brother reported by Feldman et al. (2009), withno need to appeal to semantic factors.

The behavioral masked priming literature and MEG single wordreading literature, then, provide convergent, consistent evidencesupporting a model of visual lexical processing in which morpho-logical constituents are very rapidly and automatically detectedon the basis of their orthographic form.

Interestingly however, research directly combining the maskedmorphological priming paradigm with measurements of theevoked neural correlates from EEG recordings, has so far produceda less consistent picture.

Research combining masked repetition priming with EEG hasresulted in a clear pattern of evoked ERP components that can berelated to a well motivated functional architecture for word recog-nition (Grainger & Holcomb, 2009; Holcomb & Grainger, 2006,2007). A series of ERP components, whose amplitudes are modu-lated by priming, appear to reflect processing that proceeds fromvisual features to orthographic representations and finally tomeaning.


The earliest of these components, the N/P150, is also the mostfocal, producing priming effects that are visible only at themost anterior and posterior sites primarily in the right hemisphere;positive-going at occipital sites (especially over the right hemi-sphere) and negative-going at more anterior sites. This effect hasbeen observed across several studies using words (Holcomb &Grainger, 2006), single letters (Petit, Midgley, Holcomb, & Grainger,2006) and pictures of objects (Eddy, Schmid, & Holcomb, 2006).Evidence that the N/P150 is modulated by such a diverse range ofstimuli, including non-linguistic stimuli, suggests that it reflectsan early, low level process, possibly one that is involved in mappingvisual features onto higher level representations (Holcomb &Grainger, 2006). This component is, then, unlikely to be the ERP ana-log of the MEG M170/Visual Word Form component, and thus is notexpected to show the same sensitivity to morpho-orthographicpatterns.

The second component found to be sensitive to linguisticprocesses operating in the masked priming paradigm is the N250(Holcomb & Grainger, 2006). The N250 is a negative going wavewith an onset around 175 ms, a duration of approximately150 ms and a peak at around 250 ms. It has a broad scalp distribu-tion with the largest effects over the more frontal sites. It is larger,and peaks earlier, to targets following primes with which they shareno letters than to targets that are partial repetitions of their primes,while targets that are partial repetitions of their primes producelarger N250s than full repetitions. (Holcomb & Grainger, 2006,2007) suggest that the N250 may reflect the processing of sub-lettervisual feature representations and sub-word orthographic repre-sentations (i.e. letters and letter clusters). In particular, theypropose that the amplitude of the N250 may reflect the degree ofmismatch between letter and letter-cluster representations thatare activated by the prime stimulus, and those representationsreceiving activation from the target. This component is the likeliestto reflect the early VWFA processing indexed by the MEG M170.

Finally the third component implicated in masked repetitionpriming is the N400 component (Brown & Hagoort, 1993; Kutas& Hillyard, 1984), starting around 350 ms and ending around550 ms with a central-posterior scalp distribution that was mostnegative to targets that were completely unrelated to their primes,least negative to targets that were complete repetitions of theirprimes and intermediate to targets that partially overlapped theirprimes. Holcomb and Grainger (2006) suggest that in its earlyphase it reflects the mapping of lexical form onto meaning, with la-ter N400 effects reflecting integration across semantic representa-tions (see Lau, Phillips, & Poeppel (2008) for an extremely thoroughreview of the literature on the timing, anatomical generators andfunctional interpretations of the N400 response).

The masked priming paradigm combined with EEG recordinghas now been used in four studies investigating morphologicalpriming, with results that are roughly consistent with this threecomponent/three stage model. Lavric, Clapp, and Rastle (2007)(LCR); Morris et al. (2007) (MTGH) and Morris, Grainger, and Hol-comb (2008) (MGH) all compared genuine, transparent deriva-tional morphological complexity (Morph: teacher �teach) withpseudo complexity (Psuedo: corner � corn) and orthographic over-lap (Ortho: brothel � broth) in English. The two Morris et al. studiesused the same materials, but while the 2007 study used the morecommon lexical decision task, the 2008 study used a go/no-gosemantic categorization task. Royle, Drury, Bourguignon, andSteinhauer (2010)(RDBS) compared French inflectional complexity(Morph: cassait � casse ‘broke’ � ‘break’) with orthographic over-lap (Ortho: cassis � casse ‘blackcurrant’ � ‘break’) and semanticsimilarity (Sem: brise � casse‘break’ � ‘break’) and used the lexicaldecision task.

These four studies used slightly different variations on themasked priming design (duration of masked prime, presence or

absence of backwards mask, SOA between prime and target, num-ber of items, etc.), and focused on slightly different analysis timewindows and groupings of electrodes, which lead to inevitable dif-ferences in the results, summarized in (Table 1).

This pattern of results is roughly what we expect given thebehavioral masked priming, MEG single word and ERP repetitionpriming studies: at the earliest latencies (100–200 ms) we see sen-sitivity to visual feature overlap between prime and target, but notto semantics or to whether the overlap is morphological or not; be-tween 200 and 300 ms, we see differentiation between morpho-orthographic overlap and pure form overlap (in some studies);and between 300 and 500 ms there is further dissociation betweengenuine and pseudo morphological overlap.

But given the variability in each of these three time windowsacross very similar studies, it’s clear that more research is requiredto establish the reliability, and therefore interpretability, of thispattern.

Our research, then, was motivated by three goals: first, to di-rectly compare regular and irregular morphological inflection pro-cessing in English using the masked priming + lexical decisionparadigm, which is so well established as a tool for investigatingthe early stages of morphological processing; second, to use EEGto track the time course of this processing and determine when dif-ferences between inflection types emerge; and third, to help re-solve the inconsistencies in the ERP masked morphologicalpriming literature by adding data from English inflectionalallomorphy.

If regular inflectional morphology is detected on the basis of thesame morpho-orthograpic segmentation processes as affixal deri-vational morphology, we should see robust facilitation effects forregular past tense priming in the N250 time window. If irregularinflectional allomorphy is also detected and parsed on the basisof morpho-orthographic features, then we should see similarly ro-bust priming effects for irregular past tense to stem priming in thesame early time window. And if the N250 response is genuinelysensitive to these processes of morpho-orthographic parsing, weshould see significant differences between morphological andorthographic priming conditions (Lavric et al., 2007; Morris et al.,2008).

2. Materials and methods

2.1. Participants

Twenty participants from the Tufts University community (15female, aged 18–25, mean = 21.3) were paid for their participationand gave informed consent to participate. The data from three sub-jects, one male and two females, were excluded from analysis, onebecause of failure to complete the study, one for reporting a diag-nosis of dyslexia, and one for excessive eye movement artifacts.Participants were right-handed native English speakers with nor-mal or corrected-to-normal vision, with no reported linguistic orneurological impairment.

2.2. Stimuli

The stimuli were 120 regular and 120 irregular past tense andstem verb forms, chosen from the CELEX English database. These120 irregular verbs almost exhaust the set of English irregular verbroots (there are about 170 such roots in total, but 26 of these arezero-change (hit, cut), and a further 18 are sufficiently uncommonas to have lexical decision accuracy scores below 75% in the EnglishLexicon Project Database (shrive-shrove, rend-rent, clothe-clad, etc)).Thus unlike the studies by Crepaldi et al. (2010), Kielar et al. (2008)and Pastizzo and Feldman (2002), our experiment offers a compre-

Table 1Significant effects of masked priming manipulations on ERPs as reported in previous literature.

MTGH2007 LCR2007 MGH2008 RDBS2010

Conditions M,P,O M,P,O M,P,O M,S,O100–200

(ms)No analysis for this epoch increased positivities,

all conditionsIncreased positivities at frontal sites andnegativities at occipital sites, all conditionsa

No significant effects

200–300(ms)

Reduced negativities for M and P across allelectrodes, significant effects for O only at frontalelectrodes

Reduced negativitiesfor M and P

Reduced negativities for M across all electrodes,for P at midline electrodes, no significant effectsfor O

Reduced negativities forM and O, not for S

300–500(ms)

Decreased negativities for M, not P or O Reduced negativitiesfor M and P, but notfor O

Reduced negativities for all three conditions (nointeractions)

Reduced negativities forM, but not O or Sconditions

M = Morphologically Related, P = Pseudo-morphologically Related, O = Orthographically Related, S = Semantically Related1 analyses do not include midline electrodes for anytime window.

a Only analyzed anterior and posterior sites in the 100–200 ms time window.


hensive investigation of English past tense irregular allomorphyprocessing.

Each stem form appeared in four conditions: (a) primed by itself(ID), (b) primed by its past tense form (PT), (c) primed by an Ortho-graphic Control (OC) that differed in only one letter, and (d) primedby an unrelated item with which it shared no letters (UN).

We matched regular and irregular verb stems on length and logsurface frequency (CELEX database, Baayen, Piepenbrock, & VanRijn (1993)) and past tense primes on log surface frequency as seenin Table 2. An ANOVA on log surface frequency, with verb type(regular, irregular) and Verb Tense (past, stem) as factors revealedno main effects of either verb type (F(1,238) = 2.005, p = .158) orverb tense (F(1,238) = 1.086, p = .298) and no significant interac-tion (F(1,238) = 2.20, p = .139). A similar ANOVA on length revealeda main effect of both Prime Type (F(1,238) = 545.03, p < .001) andverb tense (F(1,238) = 52.512, p < .001) as well a significant inter-action (F(1,238) = 88.408, p < .001), due to the fact that regular pasttense primes were longer than all other conditions.

Across the two verb types we also matched the orthographiccontrol and unrelated prime conditions on length (OC and UN:F(1,238) = 3.497, p = .063) and log surface frequency (OC:F(1,238) = .427, p = .514; UN: F(1,238) = .528, p = .468).

From these stimuli 4 lists were constructed, each containing 30items per condition (240/list). No item appeared in more than onecondition in any given list. In a testing session each participant wasgiven 2 of the four lists, separated by a short break, and the order ofpresentation of the two lists was counterbalanced across partici-pants. Thus, each participant saw each target verb in two of thefour possible conditions. In total, each participant saw 480 exper-imental items, 60 in each of the 4 conditions. In addition to theexperimental items, each list contained 480 pseudoword filleritems created by changing one or two letters of an existing word(nonwords were phonotactically well formed). As with the exper-imental items, 60 nonword targets were preceded by identityprimes, 60 by unrelated primes that shared no letters with thetarget, 60 by primes that differed in only one letter, 30 by primes

Table 2Mean (SE) length and log lexical frequencies for the three prime types for the two verbconditions.

PRIMETYPE Length Log frequency

Regular Irregular Regular Irregular

Mean (SE) Mean (SE) Mean (SE) Mean (SE)

Verb stema 4.27 (0.06) 4.44 (0.07) 2.12 (0.04) 2.06 (0.08)Past tense 6.08 (0.07) 4.53 (0.08) 2.20 (0.03) 2.04 (0.07)Ortho control 4.27 (0.06) 4.44 (0.07) 2.34 (0.08) 2.26 (0.08)Unrelated 4.27 (0.06) 4.44 (0.07) 2.29 (0.07) 2.36 (0.07)

a Verb stem = targets and identity primes.

consisting of the target followed by the suffix ‘-ed’ (analogous toregular past tense primes), and 30 by primes that differed fromtheir targets due to a vowel change (analogous to irregular pasttense primes).

2.3. Procedure

Participants were seated in a comfortable chair is a darkenedroom at a distance of 140 cm from the computer monitor. Eachtesting session began with a short practice block, followed by theexperimental block. Participants were told that they would see alist of words and nonwords on the computer monitor and were in-structed to respond as quickly and as accurately as possible indi-cating whether the stimulus was a word (dominant hand) or not(non-dominant hand) by pressing one of two response keys. Visualstimuli were presented on a 19-inch monitor, with a diagonalviewable screen size of 18 inches, and a width of approximately14.5 in., set to a refresh rate of 100 Hz (which allows 10 ms resolu-tion of stimulus control). Stimuli were displayed at high contrast aswhite letters (Verdana font) on a black background. Each letter was40 pixels tall by 20 pixels wide. The screen resolution was 800 by600 pixels, and the visual angle subtended by stimuli ranged from1.1� to 3.25�. Primes were presented in lower case letters for 50 ms,preceded by a 500 ms forward mask and a 20 ms backward maskcomposed of hash marks (#######). The target was then pre-sented in upper case letters for 300 ms followed by a 1200 ms ITI.

2.3.1. Recording procedureThe electroencephalogram (EEG) was recorded from 29 active

tin electrodes held in place on the scalp by an elastic cap (Elec-trode-Cap International). In addition to the 29 scalp sites, addi-tional electrodes were attached to below the left eye (to monitorfor vertical eye movement/blinks), to the right of the right eye(to monitor for horizontal eye movements), over the left mastoidbone (reference) and over the right mastoid bone (recorded ac-tively to monitor for differential mastoid activity). All EEG elec-trode impedances were maintained below 5 kX (impedance foreye electrodes was less than 10 kX). The EEG was amplified byan SA Bioamplifier with a bandpass of 0.01 and 40 Hz and theEEG was continuously sampled at a rate of 200 Hz throughoutthe experiment.

2.3.2. Data analysisERPs time locked to the onset of target words in each category

were formed off-line from trials free of excessive artefact or re-sponse error. Trials characterized by EOG artefact in excess of70 mV were rejected, resulting in 8.25% of trials being discarded.This percentage did not vary significantly across experimental con-ditions (p > 0.9). In addition, any trials with incorrect behavioral re-sponses were also excluded from the averages. All trials were


baselined to the average of activity in the 100 ms pre-target periodand were low-pass filtered at 15 Hz. We calculated the mean volt-age in the 100–200, 200–300 ms and 300–500 ms time windowsrelative to the 100 ms pre-target baseline. The calculations in the100–200 ms epoch were based on a different set of electrodes thanfor the other epochs as explained below. These time epochs werechosen because they correspond to peaks in the waveforms elicitedby the stimuli that were identified by visual inspection and also tothe latency ranges that have been found for the N/P150, the N250and the N400 (Holcomb and Grainger, 2009).

We calculated difference scores by subtracting the latency ormean amplitude for each of the three types of related trials (iden-tity, past tense, and orthographic control) from those for the unre-lated trials.

The univariate approach to repeated measures ANOVA requiresthe assumption of sphericity. One form of sphericity is compoundsymmetry, which requires that all variances of the repeated mea-surements are equal, and that all correlations between the pairsof repeated measurements are equal (O’Brien & Kaiser, 1985).When this assumption is violated, as is frequently the case withERP data sets, then various adjustments are introduced to compen-sate for the violations (e.g. Greenhouse & Geisser, 1959; Huynh &Feldt, 1970; (techniques still widely used)). However, the contrastsinvolved in testing repeated measures effects do not need to beindependent of each other if we use multivariate criteria to simul-taneously test the statistical significance of the two or more re-peated measures contrasts, hence the MANOVA approach torepeated measures ANOVA has gained popularity in recent years(Groh-Bordin, Zimmer, & Ecker, 2006; Maurer, Blau, Yoncheva, &McCandliss, 2010; Pizzagalli, Regard, & Lehmann, 1999). In ouranalyses, we follow the recommendations of O’Brien and Kaiser(1985) and report multivariate test statistics.

All MANOVAs were run using the general linear model approachin SPSS, and subjects were a random factor in all analyses. Fixedfactors were Verb Type, Prime Type, Anteriority and Laterality,and all factors were crossed in the model. The Verb Type factorcontrasted mean ERP amplitude difference scores for the regularand irregular verbs, while the Prime Type factor contrasted meanERP amplitude difference scores in the identity, past tense andorthographic control conditions. To analyze the scalp distributionof the ERP effects, we included two factors, Anteriority and Lateral-ity. The Anteriority factor represented the anterior–posterior dis-tribution of effects and included three levels contrastingelectrode locations from the back to the front of the head, whilethe Laterality factor represented the left–right distribution and in-cluded three levels contrasting electrode locations at left hemi-sphere, midline and right hemisphere locations.

We report any significant main effects of the experimental fac-tors verb type and prime type and any interaction of the topo-graphical factors anteriority and laterality with the experimentalfactors. We analyzed the mean amplitude difference scores byselecting 9 representative sites distributed across the scalp (F3,Fz, F4, C3, Cz, C4, P3, Pz, P4) [see Fig. 2].

The N/P150 was analyzed with a specific set of six occipital andfrontal electrodes (O1,Oz, O2, FP1, FPz, FP2), selected because pre-vious masked priming ERP studies have indicated that N/P150priming effects are largely restricted to frontal pole and occipitalsites (Chauncey, Holcomb, & Grainger, 2008; Morris et al., 2008).For these data, we again used a repeated-measures MANOVA withfour within-subjects factors (verb type, prime type, anteriority andlaterality), but for this data set, the Anteriority factor comprisedtwo, as opposed to three levels.

Voltage amplitudes vary considerably over the topography ofthe head, but in the absence of interactions with the experimen-tally manipulated variables, these differences mean little, so we re-port only results concerning the main effects of the experimentally

manipulated factors, and the interaction of these factors with thetopographic factors anteriority and laterality.

We analyzed reaction times and accuracy rates with a 2 � 3 re-peated measures ANOVA with verb type (regular, irregular), andprime type (identity, past tense, orthographic control) as factors.As with the electrophysiological data, difference scores were com-puted by subtracting the reaction times or error rates for each ofthe three types of related trials (identity, past tense and ortho-graphic control) from those for the unrelated trials. Any responsesthat were below 200 ms or above 1500 ms were excluded from theanalysis.

The a level was set at 0.05. Planned comparisons for all analyseswere computed to investigate any effects involving more than twoconditions. Because we were interested primarily in the differencesbetween prime types, to investigate the main effect of Prime Typeor any interactions involving this factor, we compared identity topast tense primes, identity to orthographic control primes and pasttense to orthographic control primes, producing three plannedcomparisons. To correct for multiple comparison, we performedBonferroni corrections by multiplying each p-value by the numberof comparisons made. For all comparisons, we report the mean dif-ference, the standard error, and the corrected p-values. Correctedp-values that meet the a-threshold are marked with ⁄.

3. Results

3.1. Behavioral results

Analyses of the reaction time data (see Table 3) yielded a signif-icant effect of Prime Type (F(2,18) = 89.4, p < .001⁄) as well as aPrime Type by Verb Type interaction (F(2,18) = 15.9, p < .001⁄).There was no main effect of verb type. Inspection of the mean dif-ference scores for the main effect of prime type showed that iden-tity primes elicited significantly greater priming effects than pasttense primes (MID � PT = 13.6(2.1), t = 6.38, p < .001⁄⁄). In turn,the priming effect for past tense primes was greater than for ortho-graphic control primes (MPT � OC = 35.0(4.4), t = 7.95, p < .001⁄⁄).

The verb type by prime type interaction arose because for reg-ular verbs, past tense primes produced greater priming effects thandid orthographic control primes (MPT � OC = 49.8(6.3), t = 7.89,p < .001⁄⁄) but did not differ from identity primes (MPT � ID = �1.0(3.2), t = 0.31, p = 1.0). In contrast, for irregular verbs, past tenseprimes were less effective than identity primes (MPT � ID = �26.2(4.1), t = 4.66, p < .001⁄⁄), but produced greater effects than Ortho-graphic Control primes (MPT � OC = 20.1 (4.3), t = 9.11, p < .001⁄⁄).

Analysis of the accuracy scores (Table 4) revealed a significantmain effect of prime type (F(2,18) = 8.0, p < .003⁄) but no effect ofverb type and no prime type by verb type interaction (all ps > .1).Planned comparisons indicated that responses to targets precededby their stems and to those preceded by past tense forms did notdiffer (MID � PT = 1.04 (0.6), t = 1.71, p = .3), but orthographic con-trols primes were significantly less effective than both identityprimes (MID � OC = 4.1 (1.0), t = 3.99, p = .002⁄) and past tenseprimes (MPT � OC = 3.0 (0.8), t = 3.76, p = .004⁄).

3.2. ERP results

Fig. 1 plots the mean sensor activation for regular and irregularverb targets preceded by identity, unrelated, and past tense primesover all sensors. In the epoch from 0–800 ms after the onset of thetarget we observed an initial small negative-going potential (N1)peaking between 40 and 70 ms immediately followed by a largerpositivity peaking between 140 and 180 ms. These early potentialsare likely to reflect sensory processing related to processing themask, prime and target stimuli. Following these early potentials,

Table 3Mean reaction times (ms) and difference scores. Difference scores were calculated by subtracting the mean reaction time for each of the three types of related trials (past tense,identity and orthographic control) from those for the unrelated trials.

PRIMETYPE Regular Irregular

Mean (SE) Diff. (SE) Mean (SE) Diff. (SE)

Unrelated 533.20 (12.23) 539.04 (13.75)Past tense 501.11 (12.84) 32.09 (3.74) 525.75 (13.93) 13.29 (3.43)Identity 500.11 (13.11) 33.09 (4.23) 499.59 (12.75) 39.46 (3.96)Orthographic control 550.96 (15.47) �17.76 (5.38) 545.90 (14.50) �6.86 (4.72)

Table 4Mean accuracy rates (% correct) and difference scores. Difference scores werecalculated as for Table 3.

PRIMETYPE Regular Irregular


Unrelated 5.50 (1.90) 6.33 (1.35)Past tense 3.92 (1.19) 1.58 (0.98) 5.17 (1.28) 1.17 (0.71)Identity 3.58 (0.93) 1.92 (1.20) 3.42 (1.04) 2.92 (0.97)Orthographic

control8.08 (1.66) �2.58 (1.12) 7.08 (1.31) �0.75 (0.67)


we observed a series of two negative deflections, the first peakingaround 250 ms post-target (N250) and the second at around400 ms post target (N400). Both were maximal at centro-parietalsites.

The two negativities (N250 and N400) were followed by a largepositive deflection peaking around 500 ms (which we call the latepositive component, or LPC). A number of studies suggest that thislate positive component (LPC) of the event related potential can beviewed as an index of stimulus evaluation and comparison pro-cesses (Kutas, McCarthy, & Donchin, 1977; McCarthy, 1981). In a‘same/different’ decision paradigm as well as a semantic categori-zation paradigm, Chabot, York, and Waugh (1984) found high cor-relations between mean response latency and the latency valuewhere the LPC reached its maximum value at the electrode posi-tions monitored (Pz, Cz, C30) (midway between C3 and P3) andC40 (midway between C4 and P4). Thus we decided to conduct posthoc tests to investigate this component.

Fig. 2 plots the difference waves, computed by subtracting themean voltage for each of the three types of related trials (pasttense, identity and orthographic control) from those for the unre-lated trials, at nine electrode sites. The difference waves isolatethe components of interest more clearly by removing variationsin voltage that are common to all conditions leaving only the dif-ferences related to the experimental manipulation. These differ-ence waves were the dependent measure in the statisticalanalyses reported below.

Fig. 3 plots the subtraction of the related (identity, past tense,orthographic control) conditions from the unrelated conditions at150 ms, 250 ms and 400 ms post-target for regular (Panel A) andirregular (Panel B) verbs.

3.2.1. 100–200 ms epochTo analyse the N/P150 component we used a specific set of six

occipital, central and frontal electrodes (O1, Oz, O2, FP1, FPz, FP2),selected because previous masked priming ERP studies have indi-cated that N/P150 priming effects are largely restricted to thesesites (Chauncey et al., 2008; Morris et al., 2008). We analyzed thedata with a repeated measures MANOVA with verb type, primetype, anteriority and laterality as factors, as described in the dataanalysis section. We did not find any significant differences in thistime window related to either of our experimental factors of verbtype or prime type (all ps > 0.5).

3.2.2. 200–300 ms epochAnalyses of the mean amplitude data in this time window

yielded a significant effect of prime type (F(2,18) = 18.4,p < .001⁄) as well as a significant prime type by laterality interac-tion (F(4,16) = 4.1, p = 0.17). There was no main effect of verb type,nor did verb type participate in any significant interactions (allps > 0.1).

Inspection of the mean amplitude difference for the main effectof prime type (see Table 5) showed that while targets in both thepast tense and identity conditions elicited N250 priming effects,those in the orthographic control condition showed no significantpriming. Planned comparisons confirmed these findings; the meanamplitude difference for targets in the identity and past tense con-ditions did not differ (MPT � ID = 0.3 (0.3), t = 1.16, p = 0.78), butthe mean amplitude difference for targets in the orthographic con-trol condition was significantly more positive (a smaller N250priming effect) than for those in the identity (MOC � ID = 1.1(0.2), t = 4.77, p < .001⁄) and past tense conditions (M OC � PT = 0.8(0.2), t = 4.78, p = .001⁄). Visual inspection of the voltage maps re-vealed that the Prime Type by Laterality interaction was probablydue to the fact that the effect appeared greatest at midline sites.

3.2.3. 300–500 ms epochAnalyses of the mean amplitude data in this time window

yielded a significant effect of prime type (F(2,18) = 16.6,p < .001⁄) as well as a significant prime type by anteriority interac-tion (F(4,16) = 3.9, p = .021⁄). There was no main effect of VerbType, nor did Verb Type participate in any significant interactions(all ps > 0.1).

Inspection of the mean amplitude differences (Table 6) for themain effect of prime type revealed that prime-target pairs in boththe past tense and identity conditions elicited large N400 primingeffects, i.e. more negative responses to unrelated than to relatedpairs. In contrast, there was no significant priming effect forprime-target pairs in the orthographic control condition.

Planned comparisons confirmed these findings, showing thatmean amplitude differences for items in the identity and past tenseconditions did not differ (MID � PT = 0.04 (0.3), t = 0.16, p = 1.0),but that the difference for items in the orthographic control condi-tion was significantly more positive than for those in the identity(MOC � ID = 1.2 (0.3), t = 3.97, p = .002⁄) and past tense conditions(MOC � PT = 1.25 (0.2), t = 5.84, p < .001⁄).

Visual inspection of the voltage maps, and pairwise compari-sons (Table 7), revealed that the prime type by anteriority interac-tion was due to the centro-parietal distribution of the effect,typical of the N400; at centro-parietal sites the N400 priming effectfor prime-target pairs in the identity and past tense conditions wasgreater than for pairs in the orthographic control condition. How-ever at frontal sites, these differences between conditions were notas great.

3.2.4. Late Positivity (LPC)In the response time and accuracy measures, regular past

tense and identity primes evoked equivalent priming effects but

Fig. 1. Grand average difference waveforms at 15 scalp electrode sites for regular and irregular verbs preceded by unrelated (black), past tense (red) and identity (blue)primes. Target onset is marked by the vertical calibrating bar, and each tick mark on the x-axis represents 100 ms. Negative voltages are plotted upward.


Fig. 2. Grand average difference waveforms at 9 scalp electrode sites for regular (Panel A) and irregular (Panel B) verbs. The difference waveforms were computed bysubtracting the amplitude for each of the three types of related trials (past tense (dashed line), identity (dotted/dashed line) and orthographic control (solid line)) from thosefor the unrelated trials, thus larger amplitude difference waves reflect a larger difference between Related and unrelated trials. Negative voltages are plotted upward. Whereamplitudes for both the Related and unrelated trials have the same sign (±) and priming is associated with a decrease in amplitude, the resulting difference waves share thissign. Target onset is marked by the vertical calibrating bar, and each tick mark on the x-axis represents 100 ms. To the right of each panel is shown an enlarged view ofelectrode site CZ to illustrate in greater detail the N250, N400 and LPC. At the center of the figure we show the electrode montage. The nine sites used for the MANOVA arehighlighted.


irregular past tense primes, although they produced greater effectsthan orthographic control primes, were less effective primes thanidentity primes.

In contrast, the electrophysiological data showed that for bothregular and irregular verbs, past tense and identity primes wereequally effective in reducing the amplitude of the N250 andN400 components. Visual inspection of our data suggested thatthe peak latency of the late positive component (LPC) might corre-late with our behavioral data. Thus we decided to conduct a posthoc analysis of peak latency difference scores in the 300–700 msinterval post stimulus onset at electrode site Pz where the ampli-tude of the LPC was maximal, to see test this hypothesis. As with

the amplitude data, difference scores, i.e. priming effects, werecomputed by subtracting the peak latency for each of the threetypes of related trials (identity, past tense and orthographic con-trol) from that for the unrelated trials.

We conducted a 2 � 3 repeated measures MANOVA with verbtype (regular, irregular), and prime type (identity, past tense,orthographic control) as factors. We found a significant effect ofprime type (F(2,18) = 48.6, p < .001⁄) as well as a significant primetype by verb type interaction (F(2,18) = 4.13, p = .03).

Planned comparisons conducted to examine the main effect ofprime type showed that orthographic control primes elicited smal-ler priming effects (longer latencies) than both past tense

Fig. 3. Voltage maps reflecting the subtraction of the related (identity, past tense, orthographic control) conditions from the unrelated conditions at 150 ms, 250 ms and400 ms post-target for regular (Panel A) and irregular (Panel B) verbs.

Table 7Mean amplitude differences for all pairwise comparisons of Past Tense (PT), Identity(ID) & Orthographic Control (OC) primes in the N400 time window. A negative scorereflects a decrease in negativity for the related compared to unrelated trials.

Electrodes Comparison Mean diff. (SE) t-statistic p-value

Frontal PT � ID �0.080 (0.32) 0.25 1.00PT � OC �0.903⁄ (0.23) 3.98 <0.01ID � OC �0.820 (0.34) 2.42 0.08

Central PT � ID �0.100 (0.27) 0.36 1.00PT � OC �1.306⁄ (0.24) 5.53 <0.01ID � OC �1.211⁄ (0.32) 3.81 <0.01

Parietal PT � ID 0.050 (0.25) 0.18 1.00PT � OC �1.533⁄ (0.22) 7.04 <0.01ID � OC �1.577⁄ (0.30) 5.24 <0.01

Table 5Mean amplitude differences (lV) for Past tense, identity and orthographic controlprimes in the N250 time window (200–300 ms). Amplitude differences werecalculated by subtracting the mean voltage for each of the three types of relatedtrials (past tense, identity and orthographic control) from those for the unrelatedtrials. A negative score reflects a decrease in negativity for the related compared tounrelated trials.

Verb type Prime type Mean (SE)

Regular Past tense �0.45 (0.30)Identity �0.90 (0.43)Orthographic control 0.28 (0.35)

Irregular Past tense �0.74 (0.22)Identity �0.90 (0.28)Orthographic control 0.14 (0.26)

Table 6Mean amplitude differences (lV) for past tense, identity & orthographic controlprimes in the N400 time window (300–500 ms). Amplitude differences werecalculated as for Table 5. A negative score reflects a decrease in negativity for therelated compared to unrelated trials.

Verb type Prime type Mean (SE)

Regular Past tense �0.35 (0.27)Identity �0.52 (0.34)Orthographic control 0.86 (0.25)

Irregular Past tense �0.73 (0.30)Identity �0.47 (0.31)Orthographic control 0.55 (0.24)

Table 8Mean latencies and differences in mean latency (ms) for the LPC measured atelectrode site Pz. A positive score reflects a faster peak response for the relatedcompared to unrelated trials.

PRIMETYPE Regulars Irregulars


Unrelated 487.25 (11.14) 486.50 (14.53)Past tense 425.50 (11.92) 61.75 (12.35) 452.75 (11.88) 33.75 (15.74)Identity 426.75 (13.53) 60.50 (13.35) 420.00 (12.16) 66.50 (12.09)Orthographic

control486.75 (18.80) 0.50 (14.66) 497.75 (12.40) �11.25 (14.34)


(MPT � OC = 53.1 (8.9), t = 5.95, p < .001⁄) and identity primes(MID � OC = 68.9 (6.8), t = 10.13, p < .001⁄), but that past tenseand identity primed conditions did not differ from each other in la-tency (MID � PT = 15.75 (7.3), t = 2.16, p = .13) (see Table 8).

The verb type by prime type interaction arose because, for reg-ular verbs, past tense primes produced greater effects (shorterlatencies) than did orthographic control primes (MPT � OC = 61.25(13.5), t = 4.54, p = .001⁄) but did not differ from identity primes(MPT � ID = 1.25 (8.8), t = 0.14, p = 1.0). In contrast, for irregularverbs, past tense primes were less effective than identity primes(MPT � ID = �32.75 (9.8), t = 3.35, p = .01⁄), but produced greatereffects (shorter latencies) than orthographic control primes(MPT � OC = 45.0 (16.6), t = 2.71, p = .04⁄). This pattern was similarto that found in the reaction time data.

4. Discussion

Our behavioral data were similar to those of previous overtpriming studies that have found full priming for regular verbsand reduced priming for irregulars (Marslen-Wilson, Hare, & Older,1993; Stanners, Neiser, Hernon, & Hall, 1979; Sonnenstuhl, Eisen-beiss, & Clahsen, 1999; Stockall & Marantz, 2006), and fit the pat-tern predicted by Crepaldi et al. (2010). Both regular and irregularpast tense primes significantly facilitated lexical decision times totheir stem targets as compared both to unrelated and to ortho-graphically similar primes, but the magnitude of the effect wasgreater by 18 ms for the regular than the irregular past tenseprimes. Moreover, the priming effect for the regular verbs wasequivalent to the priming effect for identity primes, while irregularpast tense forms evoked less facilitation than identity primes did.

The ERP data revealed a different pattern. In the earliest mea-sure, the N/P150, we found evidence of a main effect of primingfor all related vs. unrelated conditions. There was a reduction in

Table 9Summary of differences between related and unrelated trials. ID and OC means areaveraged across regular and irregular verb target conditions.

CONDITION N250 (lV) N400 (lV) LPC (ms) RT (ms)

ID �0.90 �0.49 +63.5 +36.3Reg �0.45 �0.35 +61.8 +32.1Irreg �0.74 �0.73 +33.8 +13.3OC +0.21 +0.71 �5.4 �12.31


amplitude for targets following primes with which they sharemany more physical features than unrelated primes, consistentwith previous interpretations of this response component (Hol-comb & Grainger, 2006).

In both the N250, and the N400 time windows, our main epochsof interest, we found significant effects of the priming manipula-tion for both regular and irregular past tense primes, and no differ-ences in the magnitude of the responses to the two types of target(but significant differences between morphological priming andorthographic overlap). By contrast, the latency of the late positivityshows greater facilitation effects for regular and identity primeconditions than for the irregular, consistent with the RT data. Thisdifference between the early and late response components (Ta-ble 9) suggests that the effects of masked priming that we observedreflect at least two distinct underlying processes, only one ofwhich—indexed by the LPC and by reaction times—is sensitive tothe verb type distinction.

Because the N250 and the N400 reflect early stages of lexicalprocessing, namely the mapping of orthographic representationsonto whole-word orthographic representations and the subse-quent mapping of lexical form onto meaning, they allow us a snap-shot of the very earliest stages of complex word recognition, wherethe effects of the priming manipulation for regular and irregularpairs on the N250 and N400 appear equivalent.

What we add to the results of previous masked priming inves-tigations of regular and irregular allomorphy is the detailed tempo-ral resolution provided by ERPs, which enables us to determinewhether differences in the behavioral responses to irregular andregular masked priming conditions are a result of early, automaticor late, decision related processes. We find that at the earlieststages where we can expect to observe effects of linguistic process-ing, both regular and irregular allomorphy are associated with thesame pattern of evoked responses, consistent with theories of mor-phology that make no categorical distinction between the two.

These results also help to resolve some of the uncertainty aboutinterpreting previous results from ERP studies investigatingmasked morphological priming. The clear dissociation betweenmorphological and orthographic overlap priming in the N250 re-sponse is consistent with the findings of Lavric et al. (2007) andMorris et al. (2008), and, more generally, with the single wordreading MEG studies (Zweig & Pylkkänen, 2008; Solomyak & Mar-antz, 2010; Lewis et al., 2011) that find early, visual word formbased morphological complexity effects.1 The fact that we get thissame dissociation between morphological and orthographic formrelatedness for regular inflectional priming instils confidence thatthis body of results really reflects general morpho-orthographicanalysis mechanisms (rather than processes specific to derivationalmorphology). And finally, the fact that we also get equivalent resultsfor irregular inflectional priming suggests that these analysis mech-

1 The fact that the single word reading MEG studies which use cortical source-space models to determine regions of interest find effects 50–100 ms earlier than theEEG, masked priming studies reporting averaged sensor data is not cause for alarm. Infact Mohanan, Fiorentino, and Poeppel (2008) investigate masked repetition primingusing MEG. Employing averaged sensor data analysis techniques, they find significantpriming effects peaking around 225 ms post stimulus onset.

anisms are capable of handling a wider range of morphological expo-nence than previously thought.

However, it is worth noting that although irregular past tensepriming is not associated with significantly different effects thanregular Past Tense priming or identity priming in these earlyepochs, numerically the impact of the related prime is differentfor these three conditions, and different in the two early time win-dows. The magnitude of the N250 priming effect is greatest for theidentity priming condition, and weakest for the regular verb prim-ing condition, with the irregular verb priming effect lying almostexactly between the two. For the N400 response, however, irregu-lar past tense priming is associated with the largest effect, regularpast tense priming with the smallest effect, and identity priming inthe middle. Given that these are mere numerical trends, and notpredicted by either the Crepaldi et al. (2010) model or by theStockall and Marantz (2006) model, we resist the temptation tospeculate about their possible interpretation. Further research,perhaps using techniques like the correlational analyses of Solom-yak and Marantz (2010) is clearly required.

What we can conclude from our results is that a mere 50 ms ofexposure to ‘sold’ is sufficient to affect the early, visual word formbased processing of ‘sell’ just as strongly as 50 ms of exposure to‘walked’ facilitates processing of ‘walk’ (or even as strongly as50 ms of exposure to a word affects reactivation of that same wordin the identity priming condition), despite the lack of an ortho-graphically transparent morphological relationship between thetwo forms in the sold � sell cases. For this to be possible, the mor-phological relationship between ‘sold’ and ‘sell’ must be accessibleto early stages of form based, pre-semantic processing — with only50 ms of prime exposure and 20 ms SOA between prime and target,there is no time for ‘sold’ to activate ‘sell’ on the basis of the kindsof lexical semantic similarity associations posited in the Word andRules (Pinker & Prince, 1988, chap. II) model.

This result has clear consequences for our models of the mentallexicon. The combined results of Crepaldi et al. (2010), Kielar et al.(2008), Pastizzo and Feldman (2002), and Meunier and Marslen-Wilson (2000) and the data reported above (together with the sub-stantial masked priming and MEG single word reading evidence forregular decomposition) clearly favor a model in which all poten-tially morphologically complex letter strings are very rapidlyparsed into their constituent morphemes, even when that processof parsing does not involve simply stripping off linearly adjacentaffixes.

Given Crepaldi et al. (2010)’s failure to find pseudo-irregularpriming effects (‘bold’ did not prime ‘bell’), but our finding of ro-bust irregular allomorph priming effects in the N250 time windowassociated with visual word form processing (Holcomb & Grainger,2006), it is clear that further research is required to pin down theexact mechanisms involved in word form based parsing of irregu-lar allomorphy. Kielar et al. (2008) found that irregulars ending in‘-t’ or ‘-d’ evoked greater reaction time priming responses thanirregulars without any overt exponent of the regular past tensemorpheme—exactly what we would expect given morpho-ortho-graphic analysis mechanisms sensitive to sub-regularities (Albright& Hayes, 2003).

If we have now established that such a highly irregular set offorms as the English irregular past tense verbs are rapidly and reli-ably parsed as morphologically complex, then this early formbased parsing mechanism is hypothesized to be very general in-deed. Other ‘irregular’ morphological phenomena, including Semi-tic templatic morphology (McCarthy, 1981), process morphologysuch as reduplication (Marantz, 1982), productive ablaut (Ander-son, 1985), and polysynthesis (Baker, 1996) are all predicted tobe processed by similar mechanisms (modulo variations inorthographic systems). To the extent that the processing of thesephenomena have been investigated using paradigms that allow


us to distinguish early from late stages of processing, the evidenceis consistent with this prediction (Boudelaa & Marslen-Wilson,2001; Boudelaa & Marslen-Wilson, 2004; Badecker & Allen, 2002;Frost, Deutsch, & Forster, 2000; Kazanina, Dukova-zheleva, Geber,Kharlamov, & Tonciulescu, 2008), but further confirmatory re-search is of course required. The research reported here, whichcombines the brief stimulus presentation of masked priming withthe detailed temporal resolution of EEG provides a clear startingpoint for this research.

References

Albright, A. (2002). The identification of bases in morphological paradigms. PhD thesis.Los Angeles: University of California.

Albright, A., & Hayes, B. (2003). Rules vs. analogy in English past tenses: Acomputational/experimental study. Cognition, 90, 119–161.

Anderson, S. R. (1985). Inflectional morphology. In T. Shopen (Ed.). Languagetypology and syntactic description (Vol. 48, pp. 74–85). Cambridge UniversityPress.

Baayen, R. H., Piepenbrock, R., & Van Rijn, H. (1993). The CELEX lexical database.Philadelphia, PA: CD-ROM.

Badecker, W., & Allen, M. (2002). Morphological parsing and the perception oflexical identity: A masked priming study of stem homographs. Journal ofMemory and Language, 47, 125–144.

Baker, M. C. (1996). The polysynthesis parameter. New York: Oxford University Press.Boudelaa, S., & Marslen-Wilson, W. D. (2001). Morphological units in the Arabic

mental lexicon. Cognition, 81, 65–92.Boudelaa, S., & Marslen-Wilson, W. D. (2004). Abstract morphemes and lexical

representation: The CV-Skeleton in Arabic. Cognition, 92, 271–303.Brown, C., & Hagoort, P. (1993). The processing nature of the N400: Evidence from

masked priming. Journal of Cognitive Neuroscience, 5, 34–44.Butterworth, B. (1983). Lexical representation. In B. Butterworth (Ed.). Language

production development writing and other language processes (Vol. 2,pp. 257–294). Academic Press.

Chabot York & Waugh (1984). The late positive component of the evokedwaveform: Relationship to word recognition processes. International Journal ofNeuroscience, 22(3–4), 299–314.

Chauncey, K., Holcomb, P. J., & Grainger, J. (2008). Effects of stimulus font and sizeon masked repetition priming: An event-related potentials (ERP) investigation.Language and Cognitive Processes, 23, 183–200.

Chomsky, N., & Halle, M. (1968). The sound pattern of English (Vol. 26). Harper &Row.

Cohen, L., Dehaene, S., Naccache, L., Lehéricy, S., Dehaene-Lambertz, G., Hénaff, M.A., et al. (2000). The visual word form area. Brain, 2000, 291–307.

Crepaldi, D., Rastle, K., Coltheart, M., & Nickels, L. (2010). Fell primes fall but doesbell prime ball? Masked priming with irregularly-inflected primes. Journal ofMemory and Language, 63, 83–99.

Devlin, J. T., Jamison, H. L., Matthews, P. M., & Gonnerman, L. M. (2004). Morphologyand the internal structure of words. Proceedings of the National Academy ofSciences of the United States of America, 101, 14984–14988.

Eddy, M., Schmid, A., & Holcomb, P. J. (2006). Masked repetition priming and event-related brain potentials: A new approach for tracking the time-course of objectperception. Psychophysiology, 43, 564–568.

Feldman, L. B., O’Connor, P. A., & Del Prado Martı́n, F. M. (2009). Early morphologicalprocessing is morphosemantic and not simply morpho-orthographic: Aviolation of form-then-meaning accounts of word recognition. PsychonomicBulletin and Review, 16, 684–691.

Feustel, T. C., Shiffrin, R. M., & Salasoo, A. (1983). Episodic and lexical contributionsto the repetition effect in word identification. Journal of ExperimentalPsychology: General, 112, 309–346.

Forster, K. I., & Davis, C. (1984). Repetition priming and frequency attenuation inlexical access. Journal of Experimental Psychology: Learning, Memory andCognition, 10, 680–698.

Frost, R., Deutsch, A., & Forster, K. I. (2000). Decomposing morphologically complexwords in a nonlinear morphology. Journal of Experimental Psychology: Learning,Memory and Cognition, 26, 751–765.

Grainger, J., & Holcomb, P. J. (2009). Watching the word go by: On the time-courseof component processes in visual word recognition. Language and LinguisticsCompass, 3, 128–156.

Greenhouse, S., & Geisser, S. (1959). On methods in the analysis of profile data.Psychometrika, 24, 95–112.

Groh-Bordin, C., Zimmer, H. D., & Ecker, U. K. H. (2006). Has the butcher on the busdyed his hair? When color changes modulate ERP correlates of familiarity andrecollection. NeuroImage, 32, 1879–1890.

Halle, M., & Marantz, A. (1994). Some key features of distributed morphology. MITWorking Papers in Linguistics, 21, 275–288.

Hay, J. (2001). Lexical frequency in morphology: Is everything relative? Linguistics,39, 1041–1070.

Holcomb, P. J., & Grainger, J. (2006). On the time course of visual word recognition:An event-related potential investigation using masked repetition priming.Journal of Cognitive Neuroscience, 18, 1631–1643.

Holcomb, P. J., & Grainger, J. (2007). Exploring the temporal dynamics of visual wordrecognition in the masked repetition priming paradigm using event-relatedpotentials. Brain Research, 1180, 39–58.

Huynh, H., & Feldt, L. (1970). Conditions under which mean square ratios inrepeated measurements designs have exact F-distributions. Journal of theAmerican Statistical Association, 65, 1582–1589.

Jacoby, L. L. (1983). Perceptual enhancement: Persistent effects of an experience.Journal of Experimental Psychology: Learning, Memory and Cognition, 9, 21–38.

Kazanina, N., Dukova-zheleva, G., Geber, D., Kharlamov, V., & Tonciulescu, K. (2008).Decomposition into multiple morphemes during lexical access: A maskedpriming study of Russian nouns. Language and Cognitive Processes, 23, 800–823.

Kielar, A., Joanisse, M. F., & Hare, M. L. (2008). Priming English past tense verbs:Rules or statistics? Journal of Memory and Language, 58, 327–346.

Kutas, M., & Hillyard, S. A. (1984). Brain potentials during reading reflect wordexpectancy and semantic association. Nature, 307, 161–163.

Kutas, M., McCarthy, G., & Donchin, E. (1977). Augmenting mental chronometry:The P300 as a measure of stimulus evaluation time. Science, 197, 792–795.

Lau, E., Phillips, C., & Poeppel, D. (2008). A cortical network for semantics:(De)constructing the N400. Nature Reviews Neuroscience, 9, 920–933.

Lavric, A., Clapp, A., & Rastle, K. (2007). ERP evidence of morphological analysis fromorthography: A masked priming study. Journal of Cognitive Neuroscience, 19,866–877.

Lehtonen, M., Cunillera, T., Rodrı́guez-Fornells, A., Hultén, A., Tuomainen, J., & Laine,M. (2007). Recognition of morphologically complex words in Finnish: Evidencefrom event-related potentials. Brain Research, 1148, 123–137.

Lewis, G., & Marantz, A. (submitted for publication). Tracking the -ic in frantic:Evidence for full decomposition of unique-stem words. Language and CognitiveProcesses.

Lewis, G., Solomyak, O., & Marantz, A. (2011). The neural basis of obligatorydecomposition of suffixed words. Brain & Language, 118, 118–127.

Marantz, A. (1982). Re: Reduplication. Linguistic Inquiry, 13, 435–482.Marslen-Wilson, W., Hare, M., & Older, L. (1993). Inflectional morphology and

phonological regularity in the English mental lexicon. In Proceedings of the 15thannual conference of the cognitive science society (pp. 693–698).

Maurer, U., Blau, V. C., Yoncheva, Y. N., & McCandliss, B. D. (2010). Development ofvisual expertise for reading: Rapid emergence of visual familiarity for anartificial script. Developmental Neuropsychology, 35, 404–422.

McCandliss, B. D., Cohen, L., & Dehaene, S. (2003). The visual word form area:Expertise for reading in the fusiform gyrus. Trends in Cognitive Science, 7,293–299.

McCarthy, J. (1981). A prosodic theory of nonconcatenative morphology. LinguisticInquiry, 12, 373–418.

Meunier, F., & Marslen-Wilson, W.D. (2000). Regularity and irregularity in Frenchinflectional morphology. In Proceedings of the twenty-second annual conference ofthe cognitive science society: August 13–15, 2000. Institute for Research inCognitive Science, University of Pennsylvania, Philadelphia, PA, LawrenceErlbaum Associates (p. 346).

Mohanan, P. J., Fiorentino, R., & Poeppel, D. (2008). Masked repetition priming usingmagnetoencephalography. Brain and Language, 106, 65–71.

Morris, J., Frank, T., Grainger, J., & Holcomb, P. J. (2007). Semantic transparency andmasked morphological priming: An ERP investigation. Psychophysiology, 44,506–521.

Morris, J., Grainger, J., & Holcomb, P. J. (2008). An electrophysiological investigationof early effects of masked morphological priming. Language and CognitiveProcesses, 23, 1021–1056.

Nakamura, K., Dehaene, S., Jobert, A., Le Bihan, D., & Kouider, S. (2005). Subliminalconvergence of Kanji and Kana words: Further evidence for functionalparcellation of the posterior temporal cortex in visual word perception.Journal of Cognitive Neuroscience, 17, 954–968.

O’Brien, R., & Kaiser, M. (1985). MANOVA method for analyzing repeated measuresdesigns: An extensive primer. Psychological Bulletin, 97, 316–333.

Pastizzo, M., & Feldman, L. (2002). Discrepancies between orthographic andunrelated baselines in masked priming undermine a decompositional accountof morphological facilitation. Journal of Experimental Psychology: Learning,Memory and Cognition, 28, 244–249.

Petit, J. P., Midgley, K. J., Holcomb, P. J., & Grainger, J. (2006). On the time course ofletter perception: A masked priming ERP investigation. Psychonomic BulletinReview, 13, 674–681.

Pinker, S., & Prince, A. (1988). On language and connectionism: Analysis of a paralleldistributed processing model of language acquisition. In S. Pinker & J. Mehler(Eds.), Connections and symbols (pp. 73–193). Cambridge, MA: MIT Press.

Pizzagalli, D., Regard, M., & Lehmann, D. (1999). Rapid emotional face processing inthe human right and left brain hemispheres: An ERP study. NeuroReport 10.

Rastle, K., Davis, M., Marslen-Wilson, W., & Tyler, L. (2000). Morphological andsemantic effects in visual word recognition: A time-course study. Language andCognitive Processes, 15, 507–537.

Rastle, K., & Davis, M. H. (2008). Morphological decomposition based on the analysisof orthography. Language and Cognitive Processes, 23, 942–971.

Rastle, K., Davis, M. H., & New, B. (2004). The broth in my brothers brothel: Morpho-orthographic segmentation in visual word recognition. Psychonomic Bulletin andReview, 11, 1090–1098.

Royle, P., Drury, J., Bourguignon, N., & Steinhauer, K. (2010). In Morphology and wordrecognition: An ERP approach. proceedings of the 2010 annual conference of theCanadian linguistic association (pp. 1–13).

Rumelhart, D. E., & McClelland, J. L. (1986). On learning the past tenses of Englishverbs. In J. L. McClelland & D. E. Rumelhart (Eds.), Parallel distributed


processing explorations in the microstructure of cognition. Parallel distributedprocessing: Explorations in the microstructure of cognition (Vol. 2, pp. 216–271).MIT Press.

Seidenberg, M. S., & McClelland, J. L. (1989). A distributed, developmental model ofword recognition and naming. Psychological Review, 96, 523–568.

Solomyak, O., & Marantz, A. (2010). MEG Evidence for early morphologicaldecomposition in visual word recognition: A single-trial correlational MEGstudy. Journal of Cognitive Neuroscience, 22, 2042–2057.

Sonnenstuhl, I., Eisenbeiss, S., & Clahsen, H. (1999). Morphological priming in theGerman mental lexicon. Cognition, 72, 203–236.

Stanners, R., Neiser, J. J., Hernon, W., & Hall, R. (1979). Memory representation formorphologically related words. Journal of Verbal Learning and Verbal Behavior,18, 399–412.

Stockall, L., & Marantz, A. (2006). A single route, full decomposition model ofmorphological complexity: MEG evidence. The Mental Lexicon, 1, 85–123.

Taft, M. (2004). Morphological decomposition and the reverse base frequency effect.The Quarterly Journal of Experimental Psychology Section A, 57, 745–765.

Taft, M., & Forster, K. I. (1975). Lexical storage and retrieval of prefixed words.Journal of Verbal Learning and Verbal Behavior, 14, 638–647.

Tarkiainen, A., Cornelissen, P. L., & Salmelin, R. (2002). Dynamics of visual featureanalysis and object-level processing in face versus letter-string perception.Brain: A Journal of Neurology, 125, 1125–1136.

Yang, C. (2002). Knowledge and learning in natural language. Oxford, New York:Oxford University Press.

Zweig, E., & Pylkkänen, L. (2008). A visual M170 effect of morphological complexity.Language and Cognitive Processes, 24, 412–439.

Early, equivalent ERP masked priming effects for regular and irregular morphology

Documents

Transcript of Early, equivalent ERP masked priming effects for regular and irregular morphology