author's personal copy - binghamton...

This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

http://www.elsevier.com/copyright

Author's personal copy

Research Report

Morphological agreement at a distance: Dissociation betweenearly and late components of the event-related brain potential

Polly L. O'Rourkea, Cyma Van Pettenb,⁎aCenter for Advanced Study of Language, University of Maryland, College Park, MD, USAbBinghamton University, State University of New York, Binghamton, NY, USA

A R T I C L E I N F O A B S T R A C T

Article history:Accepted 30 March 2011Available online 3 April 2011

Syntactic relationships among non-adjacent words are a core aspect of sentence structure.Research on complex sentences with displaced elements has concluded that resolving long-distance dependencies can tax working memory. Here we examine a simpler relationship—morphological agreement between the elements of a noun phrase—across a gradient ofdistance. Participants read sentences with violations of gender agreement among Spanishnouns, determiners and adjectives. For those explicitly assigned the task of detecting errors,accuracy was uniformly high across the four levels of distance between (dis)agreeing words.A second group performed a comprehension task as ERPs were recorded. Gender agreementerrors elicited a left anterior negativity (LAN) regardless of the distance between (dis)agreeing words, indicating that the errors were detected. In contrast, a temporally latercomponent of the ERP (P600) showed decreasing amplitudes as the number of wordsbetween (dis)agreeing elements increased. Smaller P600 responses were also associatedwith slower responses to the comprehension questions. Given other work suggesting thatthe P600 indexes attempted repair of a problematic sentence structure, the results suggestthat the participants became increasingly unwilling to re-visit their initial parse of asentence as the required effort increased, despite having noted an error. The results arediscussed within the context of studies showing that readers often compute inadequatestructural representations of sentences. We suggest that P600 amplitude may reflect thecosts versus benefits of sentence re-analysis, determined by a combination of sentencestructure, task requirements, and the degree to which sentencemeaning hinges on a correctstructural analysis.

© 2011 Elsevier B.V. All rights reserved.

Keywords:Event-related potentialGender agreementNumber agreementMorphologySentenceProficiency

1. Introduction

Most of the world's languages use morphological agreement toflag relationships among words in sentences. For instance,although the morphology of English is simplified relative to

many languages, it retains a fewovert agreement requirements,such as person and number agreement between nouns andverbs (e.g., the student says versus the students say versus I say).Other languages possess richer morphological systems, suchthat German includes six forms of the indefinite article ein (“a”),

B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9

⁎ Corresponding author at: Department of Psychology, Science IV, Binghamton University, 6600 Vestal Parkway East, Binghamton,NY 13902, USA.

E-mail address: [email protected] (C. Van Petten).

0006-8993/$ – see front matter © 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.brainres.2011.03.071

ava i l ab l e a t www.sc i enced i r ec t . com

www.e l sev i e r . com/ loca te /b ra i n res


with the correct variant depending on the case, number andgender of the subsequentnoun.Morphological agreementhelpsto identify constituent boundaries and the relationshipsbetween constituents (see Bates et al., 1999 andMacWhinneyand Bates, 1989 for discussion of the variable importanceof morphological agreement versus word order across lan-guages). Overt morphological agreement is especially useful inidentifying structural relationships among words that areseparated by intervening words (long-distance dependencies),which in turn influence the final interpretation of a sentence.Consider, for example, the ambiguity of the English sentenceThe footprint of the suspect found last night was analyzed—was it thefootprint or the suspect thatwas located last night? The parallelsentence in Spanish contains no such ambiguity becausenouns and their modifiers are marked for gender, and thefeminine past participle encontrada (found), matches only one ofthe possible antecedents, huella (footprint): La huellaFEM delsospechosoMASC encontradaFEM la noche anterior fue analizadaFEM.

Although the potential utility of morphological agreementfor sentence parsing is most evident when there are inter-vening words, successful resolution of long-distance depen-dencies has long been identified as troublesome for readersand listeners due to the burden placed on working memory.Indeed, it has been argued that readers often fail to use genderagreement information during initial sentence parsing, evenwhen it may prevent ambiguity (Brysbaert and Mitchell, 2000).The current experiment examines the processing of genderagreement in Spanish across a gradient of distance betweenthe words that should agree. Native speakers of Spanish readcorrect sentences and those with violations of gender agree-ment as their electrical brain activity was recorded, affordinga non-intrusive measure of how and when agreement isprocessed. Below, we review grammatical gender in Spanish,the difficulty of processing long-distance dependencies, anddifferent components of the event-related brain potentialassociated with earlier and later syntactic processing.

1.1. Grammatical gender

Grammatical gender is pervasive in Indo-European and otherlanguage families and, in most instances, is a purely morpholog-ical feature. In Spanish, a small set of animate nouns havemasculine and feminine forms corresponding to their naturalgender (e.g., chico/chica, boy/girl). The much larger majority ofnouns have no correspondence between their grammaticalgender and any real-world notions of masculinity or femininity(even when one might think that an inanimate noun is closelyassociated with men or women, e.g., ovarioMASC /ovary andbarbaFEM/beard). Gender is, for the most part, merely a wayin which nouns are categorized in a large number of languages,including German, Dutch, French, Italian, Greek, Arabic,Norwegian, Icelandic, etc.

In some languages, such as Dutch and French, the gender ofan unknown word is difficult to predict from its pronunciationor spelling (Blom et al., 2008; Franck et al., 2008). In contrast,the two-gender system of Spanish is fairly transparent andregular in the mapping between gender and the phonology/orthography of nouns. It is estimated that for 68% of nouns,endings of “-o” and “-a” indicate masculine and femininegender, respectively (Moreno-Sandoval and Goñi-Mendoyo,

2002). The remainder end with “-e” or with a consonant andare variably masculine or feminine. A very small numberof nouns—less than 2%—are inconsistently marked (e.g.,manoFEM; Teschner and Russell, 1984). Like nouns, a majorityof adjectives are consistently inflected with “-o” and “-a” formasculine and feminine gender (estimated as 62% by Moreno-Sandoval and Goñi-Mendoyo, 2002).

1.2. Agreement: near and far

Whether or not noun gender is transparently marked, the overtmanifestation of grammatical gender is through agreement. InSpanish, all words within a noun phrase (determiner and anyadjectives)mustagree ingenderandnumberwith theheadnoun,as must any adjectives or participles outside of the noun phrasethat modify that head noun. The words involved in the genderagreement structure may thus be adjacent or nonadjacent. Inexample 1, the masculine noun piano selects an immediatelyadjacentmasculinearticle (el), adjacent (antiquo) andnonadjacentadjectives ( feo), andalso amoredistantmasculine adjective (caro)after the embedded clause.

1. El piano antiguo y feo que compramos ayer fue caro.(The ugly antique piano that we bought yesterday was expensive.)

The current experiment examines whether evaluation ofagreement becomes less likely or more difficult with increasingdistance, by comparing correct and incorrect agreement acrossa gradient of distance, via sets of sentences like those below.

2A. Acabo de llegar y creo que el (la) piano está aquí.(I've just arrived and think that the piano is here.)

2B. Llegamoshace poco y vimos el piano roto (rota) en la sala.(Wearrived not long ago and saw the broken piano in the room.)

2C. Me han dicho que el piano está roto (rota) y ya nofunciona.(They've told me that the piano is broken and no longerworks.)

2D. El piano que compramos ayer está roto (rota) y nofunciona.(The piano that we bought yesterday is broken and doesn'twork.)

Examples 2A and 2B contain two agreement structureswith immediately adjacent words, a determiner–noun se-quence and a noun–adjective sequence, respectively. Al-though these are equivalent in lacking any interveningwords, we hypothesize that the article–noun sequence has aspecial status for Spanish speakers due to the statisticalproperties of the language. Determiners are a small set ofwords, the large majority of which occur in masculine andfeminine forms (e.g., el/la, un/una, ese/esa, este/esta, otro/otra,aquel/aquella, exceptions are the personal possessives mi, tu,su, which are gender-invariant). Gender-matching betweendeterminers and nouns is strongly deterministic, such that thegender of a determiner predicts the gender of the subsequentnoun with greater than 99% validity. The exceptional case isthat feminine nouns beginning with a stressed “a” are notpreceded by la, so that elMASC aguaFEM is correct. Becausedeterminers are an obligatory part of a Spanish noun phrase

63B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


and nearly always occur in immediately pronominal position,they provide very strong cues that a masculine or femininenoun will occur next. When participants were cued tocomplete sentence preambles with a word of the correctgender, Franck et al. (2008) found that inconsistent article–noun sequences (of the el agua variety) were more disruptivethan the (also rare) cases of nouns with misleading gendermarkings like the feminine mano. When participants wereasked to name pictures with a determiner–noun phrase,O'Rourke (2008) found that Spanish speakers were remarkablyresistant to interference from distractor words of the oppositegender. Both sets of results suggest a particularly tightrelationship between determiner and noun gender in Spanish.In contrast, noun gender has less predictive validity for theform of a subsequent adjective, because there need be noadjective at all, and only some two-thirds of adjectives aretransparently marked for gender. We thus predictedthat violations of gender agreement in determiner–nounsequences (as in 2A) would be more salient to Spanish readersthan violations of even immediately adjacent noun–adjectivesequences (as in 2B). Sentence examples 2C and 2D includeagreement between a noun and adjectival past participle, butwith a single word (2C) or a relative clause (2D) intervening, sothat the distance between agreeing elements becomes pro-gressively longer across the sentence types.

1.3. Distance as a processing factor: theoretical andempirical work

Distance between structurally-related elements is a criticalfactor in most, but not all, models of sentence parsing (seee.g., Frazier and Clifton, 1998; Gibson, 1998 versus Fodor andInoue, 1994). Empirical studies have often focused on depen-dencies between fronted wh-phrases and their predicates,which can be separated by variable amounts of interveningmaterial. Sentences with longer dependencies (as in 3B below)are rated asmore complex, elicit longer reading times, and aregenerally dispreferred as compared to those with shorterdependencies like 3A (Crain and Fodor, 1985; Frazier, 1987;Frazier and Flores d'Arcais, 1989; Stowe, 1986; example fromPhillips et al., 2005).

3A. The detective hoped that the lieutenant knew whichaccomplice the shrewd witness would recognize in thelineup.

3B. The lieutenant knew which accomplice the detectivehoped that the shrewd witness would recognize in thelineup.

Gibson (1998) suggests two sources of processing difficultyassociated with increasing dependency length: amaintenancecost of keeping a predicted syntactic category in workingmemory until it occurs, and an integration cost of reviving priorlexical itemswhen their related constituents occur. In accordwith this general framework, several studies have suggestedthat readers with lowworkingmemory capacity have greaterdifficulty with long-distance dependencies than those withhigher working memory spans (Friederici et al., 1998; Kingand Just, 1991; MacDonald et al., 1992). In Gibson's model andothers, the relevant metric of distance is not mere number of

intervening words, but is instead specified in structuralterms, such as the number of syntactic heads with unfilleddependencies, accounting for the much-documented factthat object-relative sentences are more difficult to processthan subject-relative sentences of similar length (e.g., Ford,1983; Foss and Cairns, 1970; Hickok et al., 1993; Holmes, 1973;King and Just, 1991).

In contrast to the intensive research effort devoted tofiller-gap relations in wh-constructions, a small number ofstudies have examined the behavioral impact of distance onmorphological agreement. Our literature search revealed foursuch studies, two using production measures and two usingcomprehension measures, all four devoted to number agree-ment between nouns and verbs.

In production, a prominent paradigm for studying agree-ment is the attractor effect—the likelihood that speakers willproduce aword that agreeswith a nearbyword rather than thecorrect, but more distant constituent. Bock and Cutting (1992)argued that distance should be defined structurally ratherthan linearly, given that agreement errors were more likelywhen a verb's subject and themisleading attractor occurred inthe same clause than when they occurred in different clauses(e.g., a preamble like “The report of the destructive fires…”wasmore likely to elicit the incorrect continuation “are” than “Thereport that they controlled the fires…”). However, the samestudy showed that linear distance was also relevant, asincreasing the distance between subject and verb within aclause led to an increase in attraction errors. Haskell andMacDonald (2005) confirmed the importance of linear distanceby analyzing the verbs produced after noun phrases with botha singular and a plural element. Their participants were morelikely to select a verb that agreed with the second (morerecent) noun, continuing “the owner and the players” with aplural verb, but “the players and the owner” with a singular.

Comprehensionstudiesmanipulatingdistance innoun–verbagreement have compared correct agreement to violations ofagreement, as in the current experiment. In an eye-trackingexperiment with Hebrew sentences, Deutsch (1998) foundlonger gaze durations on verbs with incorrect number whenthe separation between noun and verbwas short, but not whenthe separation was increased by the addition of an adjectivalphrase1. Strikingly, the short-distance violations elicited regres-sive eye movements back to prior portions of the sentence, butthe long-distance violations did not. Deutsch suggested that, inhis long-distance conditions, the subject nounhadalready beensemantically interpreted by the time the verb occurred, so thatreaders did not feel compelled to evaluate the distant (mis)agreement. An alternative account is that readers had simplyforgotten the noun number and thus did not detect amismatchwith the verb number. With Dutch sentences, Kaan (2002)similarly manipulated the distance between nouns and verbs

1 An English gloss of Deutsch's (1998) long and short sentencetypes in their correct versions (noun and verb match in number) is:

Short: The policeman reported that a carSING had beenstolenSING at five in the morning.Long: The policeman reported that a carSING of the nicest andmost recent model had been stolenSING at five in the morning.

64 B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


via the insertion of a modifying phrase2, and found thataccuracy in judging sentence acceptability was significantlylower in the long-distance condition, although this effect wasmodest (some 2% accuracy reduction).

Thehandful of studies investigating thebehavioral impact ofsentential distanceonmorphological agreement all suggest thatdistance is relevant, but are amenable to different processingaccounts. The two production studies included competition forthe control of agreement, with the result that near nounsdominated far nouns for the selection of verb number, at least ifthey occurred in the same clause (Bock and Cutting, 1992;Haskell and MacDonald, 2005). This result cannot tell uswhether, in the absence of competition, it is more difficult toestablish an agreement relation between widely-separatedwords. The reading studies of Deutsch (1998) and Kaan (2002)suggest that long-distance agreement violations are lessdisruptive than short-distance violations, but are consistentwith at least two possibilities: 1) the memory record ofmorphological features fades over time, such that agreementviolationswith distant words are not always noticed and 2) thatalthough distant constituents are potentially accessible inmemory, readers sometimes fail to attempt integration overlonger distances. Below, we review evidence that an earlycomponent of the ERP is associated with detection of morpho-syntactic violations, while a later component is associated withattempted repair or revision of a problematic structure.

1.4. Event-related potentials and agreement

Non-invasive recordings of brain electrical activity—ERPs—have been used extensively in research on sentence proces-sing, including the processing of agreement. A major strengthof ERPs for psycholinguistic research is the multidimensionalnature of the signal: spatially and temporally distinct aspectsof the voltage waveforms (components) can be influenced bydifferent manipulations, and multiple components can beinfluenced by a single manipulation (see Kutas et al., 2006 forreview).

1.4.1. P600A reliable sign of difficulty in syntactic processing is theP600—a late positive wave that often lacks a clear peak butis evident in an interval some 500–800 ms after the onset ofproblematic word, and is typically largest at centroparietalscalp sites. Within sentences, larger P600s for errorsthan for correct words have been observed for a variety ofagreement errors, violations of local phrase structure, andfor errors involving higher-level syntactic structure (Hagoortet al., 1993; McKinnon and Osterhout, 1996; Osterhout andHolcomb, 1992; Osterhout and Mobley, 1995; see Kutas et al.,2006 for review). In contrast, errors of agreement in word pairsdonot elicit P600s (Barber andCarreiras, 2005;Münte et al., 1993;

Münte and Heinze, 1994). Within sentences, P600 effects do notrequire outright syntactic violations: correct but syntacticallycomplex sentences elicit larger P600s than simpler sentences,low-frequency verb argument structures elicit larger P600s thanpreferred argument structures, and garden path sentences elicitsubstantial P600s as compared to straightforward sentences(Kaan et al., 2000; Osterhout et al., 1994; Phillips et al., 2005; vanBerkum et al., 1999). Together, these last two observationssupport the widely-accepted view that the P600 componentdoes not reflect the detection of a syntactic anomaly, but ratherthe subsequent attempt to revise or repair one's initial parseof a sentence (Friederici et al., 1996; Hahne and Friederici, 1999;Kuperberg, 2007; van Herten et al., 2006).

An ERP phenomenon that closely resembles the P600grammaticality effect is observed in studies of episodicmemory, in which successful retrieval of studied materialelicits larger positive potentials than unsuccessful retrieval(Rubin et al., 1999; Van Petten and Senkfor, 1996). Thissimilarity is consistent with the idea that an attempt to reviseone's parse of a sentence necessitates retrieval or reactivationof the preceding words in the sentence. The enhanced P600 forsyntactic violations as compared to control items is also likelyto receive a substantial contribution from a more general ERPcomponent, the P3, which is elicited by stimuli that areunusual and task-relevant, or more broadly, those that forceupdating of working memory (Donchin and Coles, 1988). Theamplitude of P600 violation effects is sensitive to some of thesame factors that influence P3 amplitude in nonlinguisticparadigms, namely theprobability of a violation (larger for low-probability, Coulson et al., 1998; Gunter et al., 1997; Hahne andFriederici, 1999) and the relevance of the violation to theparticipants' assigned task (larger during sentence acceptabil-ity/grammaticality tasks than in reading for comprehension,Gunter et al., 1997; Osterhout and Mobley, 1995; see alsoFriederici et al., 2001; Osterhout and Hagoort, 1999 fordiscussion of the separability of the P3 and P600). In thecurrent experiment, we assign a post-sentence comprehen-sion task for which the syntactic status of a sentence (correctversuswrong) is irrelevant, with 50% correct and 50% incorrectsentences overall.

1.4.2. Left anterior negativityAn ERP component occurring with a shorter latency than theP600 is a stronger candidate for the neural correlate of theinitial detection of a problematic sentence structure. Morenegative potentials in response to syntactic errors have beenvariably observed since Kutas and Hillyard's (1983) manipu-lation of noun–verb number agreement (“All turtles have fourleg…”, “Some shells is…”) and of verb tense (“This allowsthem to stayed…”). These authors reported that the syntacticviolation effect was both smaller in amplitude and moreanterior in scalp distribution than the large centroparietalN400 elicited by semantic violations (“Others eat leaves,radios…”) in the same texts, but like the N400, was evidentwithin 300 ms of word onset. In subsequent studies, thisfrontocentral negative component has frequently been left-lateralized, and thus dubbed the left anterior negativity or LAN.LANs have been observed in response to many varieties ofmorphosyntactic error across languages, including viola-tions of noun–verb number agreement, verb tense, and noun

2 An English gloss of Kaan's (2002) long and short sentence typesin their correct versions (noun and verb match in number) is:

Short: Although the emperorSING the dissident will banSING

according to the rumor, there is a lot of opposition.Long: Although the emperorSING according to the rumor thedissident will banSING, there is a lot of opposition.

65B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


case (see Kutas et al., 2006; Friederici and Weissenborn, 2007for reviews)3. In contrast to the P600, LAN effects havebeen described as relatively impervious to the probability ofviolations in the stimulus set, and to the assigned task(Coulson et al., 1998; Hahne and Friederici, 1999, 2002).

In addition to the typical finding of a P600, violations ofgender agreement between nouns and their associated de-terminers, adjectives, or verbs have variably elicited clear LANeffects or negative potentials that more closely resemble anN400, apotential that ismore standardlyelicitedby semanticallyunpredictable words rather than by syntactic difficulties. Afactor that is likely tobe relevant for thisdiscrepancy is theuseofanimate nounswith natural gender (andmatching grammaticalgender) versus inanimate nouns whose gender is purelygrammatical. Studies using animate nouns have yielded N400-like effects (Demestre et al., 1999; Schmitt et al., 2002); thoseusing inanimate nouns have yielded LAN-type effects (Barberand Carreiras, 2005; Gunter et al., 2000). Two studiesmanipulat-ing the animate/inanimate status of the noun governing genderagreement confirm this dissociation (Deutsch and Bentin, 2001;Hammer et al., 2008; see Wicha et al., 2003 for related evidencethat the N400 is sensitive to a mismatch between grammaticalgender and semantic expectations). However, a third studycomparing animate nouns with both biological and syntacticgender to inanimate nouns with only syntactic gender obtainedLAN-type effects for violations of noun–adjective gender agree-ment in both cases (Barber et al., 2004). The current experimentincludes only inanimate nouns for which gender is a purelygrammatical feature, so that only LAN and/or P600 effects areexpected.

1.4.3. Are the LAN and P600 differentially sensitive toagreement distance?The behavioral studies reviewed above suggest that readers areless likely to establish agreement betweendistant thanadjacentwords, but these studies were not able to differentiate twopossible sources of a distance effect: impaired access to therelevant features versus a disinclination to engage in integra-tion/repair over long distances. The ERP review above suggeststhat the LAN and P600 responses tomorphosyntactic errors canserve as independent measures of detection versus repair, sothat it is of some interest to knowwhether these two responsesshow parallel sensitivity to integration distance. Reductions inboth the LAN and P600 responses to agreement violation acrossdistancewould favor the reducedaccessibility account,whereasa stable LANaccompaniedby a reducedP600would suggest thatagreement violations can be detected but dismissed.

Studies in German and Dutch that might be relevant to theaccessibility-versus-integration question have been inconclu-

sive. In comparisons of correct and incorrect agreement insimple and complex sentences, smaller P600 responses incomplex sentences have been reported in three experiments(Gunter et al., 1997; Kaan and Swaab, 2003; Kolk et al., 2003), butthe opposite result was reported in a fourth study (Münte et al.,1997). Three of these experiments did not observe a robust LANcomponent so that no comparison between LAN and P600effects could be made; one reported that P600 but not LAN wassensitive to syntactic complexity (Gunter et al., 1997). A handfulof studies have manipulated linear distance between (dis)agreeing elements in a manner more closely related to thecurrent experiment. Hammer et al. (2008) found smaller P600responses to errors of gender agreement between nouns andpronouns in their long-distance condition as compared to theirshort distance condition, but as no LAN was present in eithercondition, no comparison between components was possible.Kaan (2002) observed a P600 and atypical LAN (central andbilateral rather than left anterior) in response to errors of noun–verb number agreement, and found both components insensi-tive to linear distance. These studies are taken up again in theDiscussion (Section 3.3), together with two studies in Spanishwith both LAN and P600 effects (Barber and Carreiras, 2005;Dowens et al., 2010).

As detailed above, the question of whether LAN and P600showparallel or divergent results has beenhampered by the lowreliability of the LANresponse tomorphosyntactic violation.Oneclear source of this low reliability is simply that LAN effects aremuch smaller in amplitude than P600 effects, and thusintrinsically more difficult to record. Some studies of Spanishgender agreement have observed significant LAN effects (Barberet al., 2004; Barber and Carreiras, 2005; Dowens et al., 2010), butotherhave reportednosignificantnegative components (Martín-Loeches et al., 2006; Wicha et al., 2004). A less well-documentedpossibility is that LAN effects aremore prominent for some sortsof violations than others, as found in at least one experiment(Coulson et al., 1998). The current experiment includes agradient of distance rather than only two levels to increase thelikelihood of detecting a change in LAN amplitude acrossdistance.We additionally compare gender violations to a secondtype of violation, number disagreement. Both sorts of agreementviolation are compared to semantic violations, so that the LANeffect canbeclearly identifiedvia itsdistinctive scalpdistributionrelative to the expected centroparietal topography of the N400.

In the current experiments, native speakers of Spanish readsentences like those in example 2 (Section 1.2),which vary in thedistance between words subject to gender agreement. The fourpairs of gender agreement conditionswere correct and incorrectdeterminer–noun agreement (LA CASA versus EL CASA, example2A), correct and incorrect noun–adjective agreement (LA CASAROJA versus LA CASA ROJO, 2B), correct and incorrect noun–copula–adjective phrases (EL PIANO ESTÁ ROTO versus EL PIANOESTÁ ROTA, 2C) and correct and incorrect noun–X–adjectivephrases in which an average of four words separated theagreeing/disagreeing elements (2D). We will refer to the fourpairs of conditions as the distance manipulation. The sentencescontained no potential ambiguity for gender agreement; even inthe longest distance condition, no gendered nouns occurredbetween the (dis)agreeing elements. Critical words occurred inmatched sentence-medial positions, as shown in Table 1. Thegender sentences were intermixedwith semantically congruent

3 Like the P600, larger LANs are also observed in contrasts betweensyntactically complex and simpler sentences, particularly incomparisons of sentenceswith andwithout filler-gap dependencies(Felser et al., 2003;KluenderandKutas, 1993; Phillips et al., 2005).The“complexity LAN” is not a transitory response to a single word, butinstead persists across the duration of a long-distance dependency,as between a wh-word and its corresponding gap. This response isthus presumed to be a fairly direct index of working memorymaintenance; the relationship between this response and the LANtriggered by immediate agreement errors is still a matter requiringtheoretical refinement (Kutas et al., 2006).

66 B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


and incongruent sentences, and sentences with correct andincorrect number agreement, as shown in Tables 2 and 3.

In themain experiment, ERPswere recorded as participantswere asked to read for comprehension, enforced by yes/nocomprehension questions that unpredictably followed 100 ofthe 500 sentences (see Table 4 for examples). They wereinstructed to read sentences silently for comprehension, thatquestions would follow some sentences, and that somesentences would include errors. In a separate experimentwith behavioral measures only, we asked a different group ofparticipants to detect grammatical errors and to signal theiroccurrence with a button press.We expected that, when error-

detection was the assigned task, all of the gender agreementerrors would be readily detectable. Like the ERP subjects, theerror-detection subjects also answered post-sentence com-prehension questions.

2. Results

2.1. Error detection experiment

The group of 12 participants assigned the task of detectinggrammatical errorswere fairly accurate in their other assignedtask of answering post-sentence comprehension questions,achieving amean of 84.5% correct (chance=50%, range of 76 to92% across participants). Their comprehension accuracy wassimilar but slightly worse than that of the ERP participants(who had amean of 88.1%, described in Section 2.5), t(34)=3.63,p=.059.

An average of 86.4% (se 3.9) of the agreement errors werenoted (hits), and 8.7% of the correct sentences were errone-ously labeled as containing grammatical errors (false alarms).For each condition, accuracy was computed as hit rateminus

Table 2 – Sample sentences for number agreement.

UnaSG chaquetaSG (chaquetasPL) monaSG es lo que necesitas parala entrevista.(A pretty jacket is what you need for the interview.)UnSG barcoSG (barcosPL) cargadoSG de médicos llegará prontopara ayudar a la gente.(A ship loaded with doctors will arrive soon to help the people.)Hay medicinasPL genéricasPL (genéricaSG) en la farmacia cercade donde vivo.(There are generic medicines in the pharmacy close to where I live.)Hay minutosPL silenciososPL (silenciosoSG) en que muchopuede ocurrir.(There are silent moments in which a lot can happen.)

Note: Subscripts “SG” and “PL” denote singular and plural(subscripts not presented to participants). Critical words areunderlined; variant words for violation conditions are shown inparentheses.

Table 3 – Sample semantically congruent and incongruentsentences.

La señora en el vestido de gala ha perdido un guante (tanque)de satén.(The lady in the ball gown has lost a satin glove (tank).)El niño se enfermó con la gripe y tenía una tos (flor) muypersistente.(The boy got sick with the flu and had a very persistent cough (flower).)Será difícil nadar contra la corriente (merced) pero tenemosque hacerlo.(It will be difficult to swim against the current (mercy) but we have to do it.)Cuando lleguemos a Pisa quiero ver la torre (miel) inclinada.(When we arrive in Pisa I want to see the leaning tower (honey).)

Note: Critical words are underlined; variant words for violationconditions are shown in parentheses.

Table 1 – Sample sentences for gender agreement.

Determiner–nounAcabo de llegar y creo que elM (laF) pianoM está aqui.(I've just arrived and I think that the piano is here.)Me han dicho que no entienden laF (elM) palabraF alemanaF queusé en el diálogo.(They have told me that they don't understand the German word I usedin the dialogue.)Es importante que tengamos laF (elM) llantaF en buena condición.(It is important that we have the tire in good condition.)

Noun–adjectiveSupongo que es normal que elM bañoM privadoM (privadaF)tenga jabón y champú.(I suppose that it is normal that the prívate bathroom has soapand shampoo.)Creo que voy a intentar tener laF ropaF limpiaF (limpioM)para mañana.(I believe that I will try to have the clothes clean for tomorrow.)No deben pensar que elM recursoM escasoM (escasaF) va a durarpara siempre.(You shouldn't think that the scarce resource will last forever.)

Noun–copula–adjectivePienso que prefiero que laF camisaF sea moradaF (moradoM)porque me gusta el color.(I think that I prefer the shirt to be purple because I like the color.)No es posible que laF obraF sea terminadaF (terminadoM) porqueel artista está de viaje.(It is not possible that the artwork is finished because the artist ison a trip.)Es importante que elM quesoM esté ralladoM (ralladaF) para queno haya pedazos grandes.(It is important for the cheese to be grated so that there are nobig pieces.)

Noun–X–adjectiveElM documentoM que es tan importante está firmadoM (firmadaF)y listo para el juicio.(The document that is so important is signed and ready for the judge.)LaF repuestaF que nos dieron es falsaF (falsoM) y no lapodemos aceptar.(The answer that they gave us is false and we can't accept it.)LaF piscinaF en donde nos bañaremos es hondaF (hondoM)y podemos zambullirnos.(The pool where we swim is deep and we can dive.)

Note: Subscripts “M” and “F” denote masculine and femininesyntactic gender (subscripts not presented to participants). Criticalwords are underlined; variant words for violation conditions areshown in parentheses.

67B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


false alarm rate (chance performance in this measure is 0%).Table 5 shows that accuracy was roughly 10% lower in thenumber conditions than the gender conditions (t(11)=3.40,p< .01, collapsed across distance for the gender conditions).The reduced accuracy for the number conditions relative tothe gender conditions was due to both a lower hit rate and ahigher false alarm rate, although neither of the individualcomparisons were statistically robust (hit rate, t(11)=2.14,p= .056; false alarm rate, t(11)=1.68, p= .11).

Table 5 shows that there was little variation in accuracyacross the four pairs of gender conditions. A repeated-measure ANOVA with distance between (dis)agreeing words

as a factor showed a nonsignificant main effect of distance(F(3,33)=1.63), as did separate analyses of hit and false alarmrates (both Fs<1). However, the overall interaction term(with 3,33 df) in these ANOVAs is neither an appropriate nora sensitive statistical test of the question examined here,which concerns the shape of the accuracy function acrossdistance. We thus used a polynomial contrast (trendanalysis) to decompose the interaction term into its linear,quadratic and cubic components. The polynomial contrastconsiders the levels of a variable as ordered, rather than assimply different (see e.g., Keppel and Wickens, 2004). Alinear decline (or increase) in accuracy would thus beapparent as a significant linear component, a U-shaped func-tion as a significant quadratic component, and a W-shapedfunction as a significant cubic component (neither of the lattertwo effects would be particularly interpretable here). There wasno hint of a linear gradient of accuracy in analyses of hit minusfalse alarm rate, hit rate, or false alarm rate (Flinear (1,11)=1.09,0.14, 0.17, respectively), nor were any of the quadratic or cubiccomponents significant.

Overall, the accuracy results suggest that—when the experi-mental task directed attention to grammatical structure—number agreement was less salient than gender agreement.Critically however, the participantswere equally able to judgethe correctness of gender agreement in the determiner–noun, noun–adjective, noun–copula–adjective, and noun–X–adjective conditions.

2.2. Semantic congruity

For the main experiment in which ERPs were recorded as 24participants read for comprehension, Fig. 1 shows thatsemantically incongruent sentence continuations elicitedmore negative ERPs than congruent continuations, with thedifference peaking at about 400 ms after onset. Figs. 1 and 2show that the semantic congruity effect was largest atcentroparietal scalp sites, as is typical for the N400 congruityeffect for visual words (Kutas et al., 2006; Van Petten andLuka, 2006). We defined a spatiotemporal region-of-interest(ROI) that captures the latency range and scalp regionwhere the congruity effect was largest: 300–600 ms (relativeto the 100 ms prestimulus baseline) at central, parietaland temporal sites Cpz, Pz, Cp3, Cp4, P3, P4, Tp7, Tp8, T5,and T6. Mean amplitudes in this region were analyzed via ananalysis of variance (ANOVA) taking congruity (congruentversus incongruent) and scalp site as repeated measures,resulting in a main effect of congruity (F(1,23)=97.8, p<.0001).

Table 5 – Percent accuracy of detecting agreement errors in error-detection experiment (mean and se).

Hit rate False alarm rate Hits minus false alarms

Gender, determiner–noun 88.4 (4.4) 5.4 (1.5) 83.3 (4.5)Gender, noun–adjective 86.3 (3.8) 7.7 (2.1) 78.5 (4.2)Gender, noun–copula–adjective 87.5 (3.4) 7.1 (1.3) 80.4 (4.4)Gender, noun–X–adjective 87.7 (4.3) 7.1 (2.0) 80.6 (4.4)Gender, all 87.6 (3.8) 6.8 (1.4) 80.7 (4.2)Number 81.1 (4.9) 10.0 (2.3) 71.1 (4.5)

Table 4 – Sample comprehension questions.

Sentence: Dicen que recuerdan la compañía que usamos pero yolo olvidé.(They say that they remember the company that we used but I forgotabout it.)Question: ¿Se han olvidado de la compañía?(Have they forgotten the company?)Sentence: Ahora es evidente que el requisito fue impuesto porel presidente de la universidad.(Now it is evident that the requirement was imposed by the president ofthe university.)Question: ¿Fue el presidente el que impuso ese requisito?(Was the president the one who imposed the requirement?)Sentence: El escenario que se usaría ya estaba construido y listopara la ópera.(The stage that was going to be used was already built and ready forthe opera.)Question: ¿Estaba listo el escenario?(Was the stage ready?)Sentence: El helado que les gusta tanto es italiano pero lo hacenpor todas partes.(The ice-cream that they like so much is Italian but they make iteverywhere.)Question: ¿Hay que comprar el helado en Italia?(Is it necessary to buy the ice-cream in Italy?)Sentence: Me parece que no usan la lámpara eléctrica porqueno funciona bien.(It seems to me that they don't use the electric lamp because it doesn'twork well.)Question: ¿Funciona bien la lámpara?(Does the lamp work well?)Sentence: Hoy va a empezar el feria y tenemos que hacerpreparativos.(Today the fair starts and we have to make preparations.)Question: ¿Empieza la feria hoy?(Does the fair start today?)

68 B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


Comparison of the four left sites to the four right sites did notshow a significant lateral asymmetry (F(1,23)=1.04).

2.3. Gender agreement

Fig. 3 compares the ERPs elicited by critical words that agreewith their antecedents to those that violate gender agree-ment, collapsed across distance between the agreeing/dis-agreeing words. Gender violations elicited more negative ERPsthan correct words at frontal and frontotemporal sitesbeginning at about 250 ms after word onset, followed bymore positive ERPs at posterior scalp sites beginning at about500 ms after stimulus onset. Fig. 2 shows that the frontaldifference had a pronounced leftward asymmetry and thusmatches the description of a LAN (left anterior negativity).

The later, more posterior effect—P600—lacked any obviousasymmetry.

2.3.1. LANAn initial analysis examined the asymmetry of the frontaleffect by comparing the impact of gender agreementviolations in the left frontal and anterior temporal region tohomologous sites over the right hemisphere (F3, F7, Ft7, C3,T3 versus F4, F8, Ft8, C4, T4), collapsed across the distancemanipulation. An ANOVA on mean amplitudes in the 300–500 ms latency range (factors of agreement/violation, left/right, and 5 levels of scalp site) yielded a main effect ofagreement/violation (F(1,23)=4.57, p< .05), accompanied byan interaction between agreement and laterality (F(1,23)=4.78, p< .05). Follow-up analyses showed that the frontal

Fig. 1 – Grand average ERPs from 24 subjects, contrasting semantically congruent and incongruent words in sentence-medialpositions. “Le” is an electrode site below the right eye, referenced to an average of the two mastoids; “He” is a right-to-leftmontage of two electrodes just lateral to the outer canthi of the eyes.

69B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


agreement effect was significant at the left (F(1,23)=7.37,p< .02) but not right (F(1,23)=1.23) scalp sites4. Subsequentanalyses of the frontal agreement effect (LAN) were thusperformed on the left frontotemporal ROI.

The impact of distance between agreeing elements—fromthe immediately adjacent determiner–noun condition to thelongest noun–X–adjective condition—was evaluated in anANOVA with factors of agreement (correct versus violation)and distance (4 levels), using a polynomial contrast todetermine if any observed effect of distance showed a lineargradient across the four levels. This analysis confirmed theinitial result of a main effect of agreement (F(1,23)=7.68,p< .02), but yielded no hint of an interaction betweenagreement and distance (distance by agreement, F(3,69)<1;distancelinear by agreement, distancequadratic by agreement,distancecubic by agreement, all F(1,23)<1), as shown in Fig. 4.

2.3.2. P600As above, we collapsed across distance between agreeing/disagreeing elements for the initial examination of the lateposterior effect of gender agreement. The P600 was defined bya spatiotemporal region-of-interest (ROI) that is typical of thiscomponent (Coulson et al., 1998; Hagoort and Brown, 1999;Osterhout and Mobley, 1995) and where it was largest in thecurrent data: 500–800 ms at centroparietal and posteriortemporal sites Cpz, Pz, Cp3, Cp4, P3, P4, Tp7, Tp8, T5, and T6.Mean amplitudes in this region were analyzed via an ANOVAwith agreement and scalp site as repeatedmeasures, resultingin a main effect of agreement (F(1,23)=17.9, p<.0005). Com-parison of the four left sites to the four right sites yielded anonsignificant interaction between agreement and laterality(F<1), so that subsequent analyses of the P600 effect used thefull bilateral parietotemporal ROI.

The impact of distance between agreeing elements wasevaluated as above for the LAN, in an ANOVA with factors ofagreement (correct versus violation) and distance (4 levels),using a polynomial contrast to determine if any observed effectof distance showed a linear gradient across the four levels. In

contrast to the LAN, the P600 effect declined in amplitude withincreasingdistancebetween thegender-markedwords, as showin Figs. 4 and 5 (distance by agreement, F(3,69)=2.60, ε=1.0,p= .059; distancelinear by agreement, F(1,23)=5.19, p< .05;distancequadratic by agreement and distancecubic by agreement,both F<1.3).

Follow-up analyses examined the impact of genderagreement violation for the four pairs of sentence types.Themain effect of agreement violationwas significant for thethree shorter distances between (dis)agreeing words, but notfor the longest-distance noun–X–adjective condition inwhich a short clause occurred between the two gender-marked words (determiner–noun: F(1,23)=18.3, p< .0005;noun–adjective, F(1,23)=7.63, p<.02; noun–copula–adjective,F(1,23)=11.8, p<.005; noun–X–adjective, F(1,23)<1).

The analyses above yielded contrasting results for thesensitivity of the LAN and P600 to the distance manipulation.The dissociation between components was pursued in ananalysis on the ERP differences between correct and incorrectagreement at each level of distance, in an ANOVAwith factors ofcomponent (LAN versus P600) and distance (4 levels). Thisanalysis confirmedthedifferential sensitivityof the twoviolationeffects to distance (component by distance, F(3,69)=4.17, ε=0.93,p=.02; component by distancelinear, F(1,23)=14.6, p<.001; compo-nent by distancequadratic, component by distancecubic, both F<1).

2.4. Number agreement

Fig. 6 shows that although violations of number agreement(e.g., UNA CASAS) elicited a larger anterior negativity and alarger late positivity than correct agreement, these effectswere very small. The LAN and P600 effects of agreementviolation were analyzed as above, but neither were statisti-cally significant (LAN, F(1,23)=1.59; P600, F(1,23)=2.06).

It is of some interest to directly compare the two varietiesof morphosyntactic agreement, so that the number condi-tions were contrasted with the most comparable genderconditions—those in which the agreeing/disagreeing wordswere immediately adjacent to one another (determiner–nounand noun–adjective, combined). For the LAN, an ANOVA withfactors of agreement (correct versus violation) and type-of-agreement (gender versus number) yielded a main effect ofagreement (F(1,23)=9.26, p<.01), but the overall similarity ofthe two violation effects led to a nonsignificant interactionbetween agreement and type-of-agreement (F<1). For theP600, smaller responses to violations of number than genderagreement yielded a marginally significant interaction be-tween agreement and type-of-agreement (F(1,23) = 3.98,p= .058), in addition to the expected main effect of agreement(F(1,23)=16.5, p<.0005).

2.5. ERPs and comprehension performance

Mean accuracy in answering the 100 post-sentence compre-hension questions was 88.1%, with a range of 79 to 98% acrossthe participants (chance performance=50%). Reaction times(RTs) for the questions averaged 2333 ms, with a range of 1898to 2670 ms across participants. Accuracy and RT wereuncorrelated across participants (r=− .09); RT showed greatervariability than accuracy (coefficients of variation were 0.113

4 Supplementary analyses were performed to confirm the sugges-tion of Fig. 2, that the enhanced negative potentials elicited by bothgender agreement errors and semantic incongruities (relative to thecorrect control items) had distinct scalp topographies. Meanamplitudes measured within a common latency window of 300–500 ms were subjected to ANOVAs with sentence type (semanticversus gender), violation (correct versus erroneous) and scalplocation as repeated measures. To avoid spurious interactionsbetween sentence type and scalp location due to the largeramplitude of the N400 as compared to the LAN, amplitudes werenormalized by dividing each value by the square root of the sum ofsquared values in each condition, for each participant (McCarthyandWood, 1985). For the sevenmidlinescalp sites, themoreanteriormaximum of the LAN as compared to the N400 yielded a significantinteraction between sentence type, violation, and the anterior-to-posterior location factor after applying the Huhyn–Feldt correctionfor nonsphericity of variance (F(6,138)=6.15, ε=.47, p<.001). Analysisof the five pairs of scalp sites close to the midline (Fp1 and Fp2through P3 and P4, as shown in Fig. 1) yielded the same three-wayinteraction (F(4,92)=4.74, ε=.46, p<.02), although analysis of the farlateral sites (F7/8 through T5/6) did not. Overall, the results showthat thenegativity elicited bygender agreement violations (i.e., LAN)has amore anterior scalp distribution than that elicited by semanticincongruities (i.e., N400).

70 B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


versus .067, respectively). We take the overall level ofperformance to indicate that all participants were engagedin the task of reading for comprehension.

Given several reports that ERP responses to syntactic errorscan vary with language proficiency in both native and secondlanguages (Kotz, 2009; Pakulak and Neville, 2010; Steinhaueret al., 2009), we explored whether accuracy or speed ofresponding were correlated with the amplitude of ERPdifferences between acceptable and problematic criticalwords. Latencies and scalp regions for the N400, P600 andLAN were defined as above, and Pearson correlations betweenERP effect amplitudes (e.g., the amplitude of the differencebetween an agreement violation and correct agreement) and

the behavioral measures of accuracy and RT were conducted.Accuracy was not significantly correlated with any of the ERPmeasures. Larger P600 responses to violations of genderagreement were associated with faster responses to thepost-sentence comprehension questions (r=− .45, p<.05), asshown in Fig. 75 (only correct answers were included in the RTmeasure). RT was not significantly correlated with the

5 The correlation between the P600 effect and reaction time wasevident when all the gender trials were included. Only non-significant negative correlations were obtained in lower-poweranalyses of each distance condition separately.

Fig. 2 – Left: Spline-interpolated topographic maps showing the scalp distribution of differences between incorrect and correctsentence-medial words. Right: Grand average ERPs elicited by correct and incorrect sentence-medial words, showing the LANeffect for gender violations at a left frontal scalp site (top), the P600 effect for gender violations at the midline parietal site(middle), and the N400 effect for semantic incongruities at the midline parietal site (bottom). Correct and incorrect genderagreement is collapsed across the distance between (dis)agreeing words. Note that the light/dark scale is inverted for thepositive component (P600) relative to the two negative components (LAN and N400).

71B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


amplitude of the LAN response to agreement errors or theN400response to semantic incongruities.

3. Discussion

The central experimental result was that distance betweenwords that should agree modulated the amplitude of the P600response to violations of gender agreement, but that the earlierLAN response was constant across distance. Secondary resultswere that violations of gender agreement elicited more reliableERPeffects thanviolationsofnumberagreement, and that largerP600s were observed in readers who responded more quicklyto subsequent comprehension questions. We briefly discussthe secondary findings before turning to the central result.

3.1. Gender versus number agreement

An unanticipated result was that violations of numberagreement led to very weak LAN and P600 responses, despite

the fact that the disagreeing words were immediately adjacentto one another. In contrast to the results of the distancemanipulation on gender agreement, the small amplitude ofboth components suggests that the mismatches between nounnumber and determiner or adjective number were likely to beentirely overlooked. The results of the error-detection experi-ment indicated that, even when readers were explicitlyinstructed to search for agreementerrors, thenumber violationswere less frequently noticed than the violations of genderagreement. The possibility that some varieties of morphosyn-tactic violation are easier to detect than others has not seen agreat deal of attention in ERP studies; some studies with bothgender and number violations have collapsed them in analyses(Hagoort, 2003; Martín-Loeches et al., 2006; Osterhout andMobley, 1995). With a sentence acceptability task that is likelyto focus attention on morphosyntactic errors, Barber andCarreiras (2005) observed slightly larger P600s for gender thannumber violations, but a later experiment with the samematerials and an older group of participants yielded equivalentP600s (Dowens et al., 2010), as did an experiment in Hindi withan acceptability task (Nevins et al., 2007). In a comprehension

Fig. 3 – Grand average ERPs from 24 subjects, contrasting correct and incorrect gender agreement in sentence-medial words,collapsed across the distance between (dis)agreeing words.

72 B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


task more like that used here, Coulson et al. (1998) did observelarge differences between two varieties of morphosyntacticerror, which they attributed to differential saliency. The currentresults are consistent with the conclusion that gender agree-ment is more salient to native Spanish speakers than numberagreement. It should be noted, however, that the experimentwas not designed to directly compare the saliency of numberand gender agreement, as the number violations occurred inearlier portions of their sentences than the gender violations(e.g., “Una casas…” versus “Acabo de llegar y creo que el piano…”).Barber and Carreiras (2005) observed larger P600 effects forviolations that occurred later in sentences as compared to thosethat occurred earlier. Differential saliency of the two varieties ofviolation may thus have reflected increased attention to laterportions of the sentence rather than intrinsic differences.

3.2. P600 amplitude and response speed

Post-sentence comprehension questions were included toensure that participants were actively engaged in reading.

Comprehension accuracy was acceptably high for all partici-pants, but correct response times showed a wide range ofvariability, which proved to be inversely correlated with theP600 difference between correct and incorrect gender agree-ment. The questions were presented as a single line of text,rather than oneword at a time, so that response time includedboth the time required to read the question and the time tomake a decision based on the remembered content of thesentence. Because the evaluation of individual differenceswasnot a central goal of the experiment, we do not know whetherresponse speed reflected the transient state of a participant onthe day of the experiment, or a more durable trait, such astheir reading speed or general verbal ability. The observedresult of larger P600 differences between conditions inparticipants with faster responses to comprehension ques-tions is similar to an early report (Mecklinger et al., 1995) withvery differentmaterials (object-relative versus subject-relativesentences as compared to the morphosyntactic violationsassessed here). Other work indicates that P600 amplitudemaybe a reliable index of language proficiency in both second-language learners and in native speakers (Kotz, 2009; Pakulakand Neville, 2010; Steinhauer et al., 2009).

3.3. Agreement distance

The sentences used here were not especially complex,consisting of only two clauses and average length of lessthan 12 words. Moreover, the critical words occurred (on

Fig. 4 – Amplitudes of the differences between incorrect andcorrect gender agreement, across the four sentence typeswith variable distance between the (dis)agreeing words. Top:LAN amplitudes calculated across five left frontal andanterior temporal sites F3, F7, Ft7, C3 and T3. Bottom: P600amplitudes calculated across centroparietal and posteriortemporal sites Cpz, Pz, Cp3, Cp4, P3, P4, Tp7, Tp8, T5, and T6.Error bars depict the standard error of the mean, computedacross subjects as the random factor.

Fig. 5 – Grand average ERPs from 24 subjects showing theP600 effect for correct versus incorrect gender agreement,across the four sentence types with variable distancebetween the (dis)agreeing words. Shown is a compositeelectrode site that is an average of the centroparietal andposterior temporal sites (Cpz, Pz, Cp3, Cp4, P3, P4, Tp7, Tp8,T5, T6) where the P600 was measured.

73B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


average) as only the seventhword of their sentences. There islittle reason to imagine that parsing or comprehension of theexperimental sentences over-taxed working memory. Simi-larly, gender agreement between nouns and their deter-miners and adjectives was transparent. All nouns andadjectives were regularly marked with masculine “-o” orfeminine “-a” endings, and only a single gendered nounoccurred in each sentence so that there was no possibility forconfusion about which noun should be matched to an articleor adjective. In the behavioral experiment, native speakers ofSpanish were able to detect the gender agreement errors witha reasonably high (although not perfect) degree of accuracy.Critically, error-detection performance did not vary with thedistance between (dis)agreeing words. Given the relativelytrivial level of difficulty for establishing agreement relation-ships among the words, it is not surprising that each ofthe four varieties of gender disagreement elicited a larger leftanterior negativity than its correct control. The data did notsuggest a decline in LAN amplitude across increasingdistance between (dis)agreeing words. This finding indicatesthat—at some level, conscious or unconscious—the viola-tions of gender agreement were detected regardless ofdistance.

In Section 1, we noted that “distance” as a processingvariable can be defined in syntactic structural terms, by lineardistance (number of intervening words), and by co-occurrenceprobabilities within a language. The metric used here was ablend of co-occurrence probability (yielding very high predict-ability for the gender of a noun following a gendered article inSpanish) and linear distance (intervening words for the noun–adjective conditions). In future work, it may be of some interest

to contrast the three possible metrics of distance in a morecomplex design.

In contrast to the invariant LAN effect, the temporally laterP600 elicited by agreement violations declined across distance.The largest effect was observed for determiner–noun combina-tions, a package subject to particularly strong constraints inSpanish given that a gendered article nearly always predicts theimmediate occurrence of a gendered noun. Smaller butsignificant P600 effects were observed for (grammaticallyoptional) adjectives that followed their nouns immediately orafter one intervening word, but no effect was observed when anoun and adjective were separated by a short relative clause.The large majority of ERP studies implementing agreementviolations have used immediately adjacent words and reportedsignificant P600s, but four previous studies have contrastedconditions with two levels of distance between agreeingelements. One observed a result like the current one: Hammeret al. (2008) found a P600 for gender disagreement between aGerman noun and pronoun at short separations, but no effectwitha longseparation. Incontrast, Kaan (2002) foundequivalentP600s across distance for Dutch noun–verb number agreement.Two studies in Spanish compared agreement between articlesand nouns at the beginnings of sentences (“El/La/Los piano…”)to agreement between nouns and adjectives separated by oneword that also occurred in later sentence positions (“El faro esalto/alta/altos…”). In the first of these, Barber and Carreiras(2005) found a larger P600 for the longer separation later in thesentence; in the second studywith the samematerials, Dowenset al. (2010) found equivalent P600s in a slightly older group ofnative speakers. We suspect that the critical difference amongstudies is the task assigned to the participants. Like the currentparticipants, Hammer et al.'s subjects answered comprehen-sion questions presented unpredictably after a subset ofsentences, whereas participants in the other three studiesjudged the grammaticality of each sentence. A grammaticalitytaskmandates a complete evaluation of sentence structure and

Fig. 7 – Scatterplot of the relationship between response timefor answering post-sentence comprehension questionscorrectly, and P600 difference between incorrect and correctgender agreement. Each dot represents one experimentalparticipant.

Fig. 6 – Grand average ERPs from 24 subjects, contrastingcorrect and incorrect number agreement. Shown arecomposite electrodes where the LAN and P600 effects werelargest for the gender violations: left frontal and anteriortemporal (F3, F7, Ft7, C3, T3) at top, and centroparietal andposterior temporal (Cpz, Pz, Cp3, Cp4, P3, P4, Tp7, Tp8, T5, T6)at bottom.

74 B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


thus assesses readers' ability to determine whether a potentialviolation is, in fact, erroneous. Questions about the semanticcontent of a sentence—if their answers do not hinge onagreement (as here)—do not enforce syntactic evaluation. Wethus take the amplitude of P600 violation responses incomprehension tasks to reflect readers' propensity to assesssentence structure when allowed to select their own strategy.Our suggestion that the nature of the assigned task is critical forthe presence/absence of agreement distance effects is consis-tent with other reports that grammaticality tasks lead to largerP600s than free-reading or comprehension tasks (Gunter et al.,1997; Osterhout and Mobley, 1995).

The current dissociation between LAN and P600 effectscontingent on sentence type has rarely been observed, butlends further support to the proposal that the LAN reflects afairly reflexive response to a readily-detectable error whereasthe P600 reflects a subsequent voluntary attempt to repair aproblematic structure (Friederici et al., 1996; Hahne andFriederici, 1999). The decline in P600 amplitude acrossdistance suggests that our participants became increasinglyunwilling to re-visit their initial parse of a sentence as thenumber of intervening words (words presumably requiringretrieval or re-activation) increased. For the materials usedhere, this “lazy approach” to parsing was a perfectly viablestrategy for deriving sentence meaning, because sentencescontained only one noun and thus no competition for thegovernance of gender agreement. It would be of some interestto implement a distance gradient with sentences that includesecondary nouns that could serve as misleading candidatesfor gender agreement. In her study of Dutch numberagreement, Kaan (2002) contrasted sentences with singlenouns (subjects) to sentences with an additional object nounand found larger P600s for the sentences with a potentiallymisleading candidate for number agreement, but no interac-tion between distance and number of nouns. That resultwas obtained with a grammaticality task; whether—and howoften—readers fully resolve agreement structures when notcompelled to do so remains an important question.

Behavioral research suggests that readers sometimesperform inadequate structural analyses even when themeaning of a sentence does hinge on a correct parse.Christianson et al. (2001) presented comprehension ques-tions after garden-path sentences like While Anna dressed thebaby played in the crib. By the end of this sentence, it is clearthat Anna dressed herselfwhile the baby amusedhimself, butthe first few words of the sentence suggest a more commonsubject–verb–object structure in which Anna dressed thebaby. However, Christianson et al.'s subjects showed highrates of erroneously responding “yes” to questions like “DidAnna dress the baby?”. A subsequent study in which readersparaphrased sentences without exposure to potentiallymisleading questions showed a mixture of correct interpre-tations, interpretations based solely on the initial incorrectparse, and interpretations reflecting some blend of the two(i.e., the baby played after being dressed by Anna; Patsonet al., 2009). Ferreira et al. (2002) have suggested thatcomprehenders frequently compute fairly sloppy represen-tations of input sentences based on syntax alone, but thatwhen combined with the discourse or communicativecontext typically present in real-world language use, these

representations are “good enough”. These authors suggestthat a “good enough” approach to parsing may also explainwhy comprehenders can understand sentences with gram-matical errors due to speaker disfluency.

From a study closely relevant to the current experiment,Brysbaert and Mitchell (2000) argued that readers of Dutch canfail to use gender agreement to guide sentence parsing. Theypresented sentences with relative clauses that, in English,would be ambiguous: The son of the actress who was on thebalcony was shot. In Dutch, the gender of the relative pronoun(common or neuter) can disambiguate the sentence, so that(4A) indicates that the son was on the balcony, but (4B)indicates that the actress was on the balcony.

4A. HetNEU zoontjeNEU van deCOM actriceCOM datNEU op hetbalkon zat . . .(The son of the actress who was on the balcony . . .)

4B. HetNEU zoontjeNEU van deCOM actriceCOM dieCOM op hetbalkon zat . . .(The son of the actress who was on the balcony . . .)

In parallel with the English garden-path studies above,Brysbaert and Mitchell's subjects sometimes failed to use theavailable gender information to override the initially preferredparse in which the relative clause modifies the first noun, sothat they incorrectly reported that the son was on the balconysome 33% of the time after sentences like (4B).

Brysbaert and Mitchell (2000) further suggested that therewas a conflict between their results and ERP studies suggest-ing that grammatical gender is consulted during parsing (vanBerkum et al., 1999, 2000). We suggest that there is littleconflict between the behavioral and ERP results if oneremembers that the behavioral results show a probability offailure to use agreement information that, while surprisinglyhigh, is less than 100%. We assume that the amplitude of theP600 derived from an average of multiple trials similarlyreflects the proportion of individual trials in which re-analysisis attempted—maximally large when every trial in a givencondition triggers re-analysis, smallest when no initial parseis revisited, and of intermediate amplitude when there is amixture of trials within a condition. The gradient of P600amplitude observed here serves to document one factor thatcontributes to a reader's decision to attempt re-analysis: theeffort required to do so. As we suggest above, an importantquestion for future research is whether P600 amplitude willalso reflect a tradeoff between the effort required for re-analysis and the potential payoff. The costs versus benefits ofre-analysis could, for instance, be manipulated by crossing adistancemanipulation like that here with a taskmanipulationthat makes full resolution of agreement relations more or lessnecessary.

4. Experimental procedures

4.1. Participants

All participants were free of neurological/psychiatric condi-tions and medications according to self report, and had

75B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


normal or corrected-to-normal vision. All were paid for theirparticipation and offered informed consent. All were nativespeakers of Spanish for whom English was a second language.

Twelve healthy young adults (5 men) with a mean age of30.6 years (range 22–36 years) participated in the error-detectionexperiment. Nine were born in Mexico, 1 in Spain, 1 in Boliviaand 1 in the United States (this last individual spent 21 of his36 years in Mexico, including most of his childhood). Formaleducation ranged from four years of college to several years ofdoctoral study, with a mean of 19.3 years6. Three otherindividuals began, but did not complete the study: one due toequipment failure and two who were unable to return for thesecond session.

Twenty-four healthy young adults (12 men) with a meanage of 26.7 years (range 18 to 37) participated in the ERPexperiment. Formal education ranged from one year ofcollege to several years of doctoral study, with a mean of17.4 years. Six were left-handed; of the right-handers, threereported a left-handed immediate family member. Datafrom three other individuals were not analyzed due toexcessive blink artifacts in the electroencephalogram. Allbut two were born in Spanish speaking countries: 6 inColumbia, 6 in Mexico, 4 in Puerto Rico, 2 in Spain, 2 in Cuba,1 in Peru and 1 in El Salvador. The two participants born inthe United States lived in Mexico for most of their childhood,moving to the U.S. at ages 9 and 16. Subjects reported initialexposure to English at a mean age of 11 years. Seven beganto learn English in school before age 11, but none of theseexperienced an immersion setting until after age 11. Allparticipants reported current daily use of Spanish in alanguage history questionnaire.

4.2. Stimuli

In both the ERP and error-detection experiments, participantsread 500 sentences across two experimental sessions: 250correct sentences and 250 with errors of gender agreement,errors of number agreement, or semantically incongruouswords, as described in more detail below.

The gender agreement conditions were comprised of 40sentences in each of 8 conditions: correct and incorrectdeterminer–noun agreement (LA CASA versus EL CASA),correct and incorrect noun–adjective agreement (LA CASAROJA versus LA CASA ROJO), correct and incorrect noun–copula–adjective phrases (EL PIANO ESTÁ ROTO versus ELPIANO ESTÁ ROTA) and correct and incorrect noun–X–adjective phrases in which an average of four words sep-arated the agreeing/disagreeing elements. We will refer tothe four pairs of conditions as the distance manipulation,although it should be noted that the agreeing/disagreeingelements are immediately adjacent in both the determiner–noun and noun–adjective conditions.

All nouns and adjectives were singular and bore overt,transparent gender marking (-a or -o). Half of the nouns weremasculine and half feminine, so that the correct and incorrectcritical words were also evenly divided between syntacticgenders. All of the critical nouns were inanimate, such thattheir grammatical gender was unassociated with biologicalgender.

Sentences for the gender agreement conditions were con-structed in octets like those in examples 2A–2D in Section 1,using the same noun for each (and the same adjective in the sixversions with an adjective). ERPS were timelocked to the onsetof the critical (underlined) words. Eight stimulus lists wereformed so that a given participant read only one of the eightrelated variants, but across subjects, each sentence appearedequally often in its correct and incorrect form. Additionalsentence examples are provided in Table 1.

Critical words occurred in roughly the seventh positionwithin their sentences, and never occurred in initial or finalsentence positions (mean position 7.1 for the determiner–noun sentences, 7.6 for noun–adjective sentences, 7.7 fornoun–copula–adjective sentences, and 7.2 for noun–X–adjec-tive sentences). Even in the longest-distance noun–X–adjec-tive sentences, no noun phrases occurred between theagreeing/disagreeing elements, so that there were no poten-tially misleading candidates for agreement with the criticaladjective.

4.2.1. Number agreementTwo additional sets of sentences were included for compar-isonwith the gender agreement conditions, as seen inTable 2.Each stimulus list contained 60 sentences with correct and 60with incorrect number agreement, for which the agreeing/disagreeing elementswere always immediately adjacent. Thenumber agreement structures were evenly divided betweendeterminer–noun (UNACASA) and noun–adjective (MINUTOSSILENCIOS) sequences. Correct and incorrect words wereequally divided between singular and plural, andwere alwayscorrectly marked for gender. Critical words in the numberconditionswere roughlymatched to the gender conditions formean frequency of usage (27.5 versus 34.3 per million, Davisand Perea, 2005) and length. In contrast to the genderconditions, number violations occurred in early sentencepositions, at the second or third word, to introduce variabilityin potential error position across the stimuli.

4.2.2. Semantic conditionsThe second contrast for the gender conditions was betweensemantically correct and incongruous sentence-medialwords, which were expected to differ in N400 amplituderather than in LAN or P600 amplitude. Thirty correct and 30incongruous sentences occurred in each stimulus list; incon-gruous versions were formed by swapping words betweensentences so that the same words appeared in both versionsacross subjects, although each subject received only oneversion. Critical words in the semantic conditions werematched to the gender conditions for length and frequencyof usage, but occurred in slightly later sentence positions(mean position 9.2 words from sentence beginning, but neverin sentence-final position). Sample sentences are shown inTable 3.

6 Years of education were calculated as 12 for a high schooldegree (or foreign equivalent), plus years of college (up to amaximum of 4 for a Bachelor's degree), plus years of postgraduatestudy (up to a maximum of 4).

76 B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


4.2.3. Comprehension questionsYes/no comprehension questions followed 100 of the 500sentences, half correctly answered by “yes” and half “no”, asillustrated in Table 4. Questions were equally likely to occurafter a sentence in any condition, with the exception thatsemantically incongruous sentences were not queried.

4.3. Procedure

The ERP experiment was conducted in two sessions lastingapproximately 2 h each. After a language history question-naire and electrode application, subjects were seated in acomfortable chair with keypads on each armrest. They wereinstructed to read sentences silently for comprehension, thatquestions would follow some sentences, and that somesentences would include errors. A practice set of 15 sentencesincluding correct sentences, agreement errors and semanticincongruities preceded the experimental stimuli. Participantsin the error-detection experiment followed the same pro-cedures (without electrode application), but were asked topress a key whenever they detected an agreement error in asentence; examples of both number and agreement errorswere included in their instructions.

Stimulus sentences were arranged in random order, suchthat the occurrence of the different sentence types and ofcomprehension questions was unpredictable. Sentenceswere presented word by word within a continuously-presentrectangular frame (to aid eye fixation) in the center of acomputer monitor screen. Word duration was 200 ms with a300 ms interstimulus interval. After the end of a sentence,either a comprehension question or the next sentencefollowed after 5500 ms. Questions were presented in theirentirety above the fixation frame for 2500 ms, followed by3500 ms before the onset of the next sentence. Key presseswith the right and left index fingers were used for yes and noresponses to the questions, with the mapping betweenresponse and right or left hand counterbalanced betweenthe two sessions. Short rest breaks occurred after every 50sentences.

4.4. Electrophysiological methods

The electroencephalogram (EEG) was recorded by 25 tinelectrodes mounted in an elastic cap (Electrocap Interna-tional, Eaton OH). Electrode sites were midline Fpz, Fz, Fcz,Cz, Pz and Oz, and lateral pairs Fp1/2, F3/4, F7/8, Ft7/8, C3/4,T3/4, Cp3/4, Tp7/8, P3/4, and T5/6. Vertical eye movementsand blinks were monitored with an electrode placed belowthe right eye referred to the left mastoid. Horizontal eyemovements were monitored with a bipolar montage usingtwo electrodes lateral to the right and left external canthi.Electrode impedances were below 5 kohm. The EEG wasdigitized at 250 Hz with bandpass filters of 0.01 to 100 Hzand a gain of 50,000 (amplifiers from Grass Instruments,Warwick RI). The left mastoid was used a reference duringrecording; ERPs were then re-referenced offline to anaverage of the left and right mastoids. Trials containingblinks, eye movement or amplifier saturation artifacts wererejected prior to averaging the trials into ERPs for eachcondition.

R E F E R E N C E S

Barber, H., Carreiras, M., 2005. Grammatical gender and numberagreement in Spanish: an ERP comparison. J. Cogn. Neurosci.17, 137–153.

Barber, H., Salillas, E., Carreiras, M., 2004. Gender or gendersagreement? In: Carreiras, M., Clifton, C. (Eds.), On-line Study ofSentence Comprehension; Eye-tracking, ERPs and Beyond.Psychology Press, Brighton, UK.

Bates, E., Devescovi, A., D'Amico, S., 1999. Processing complexsentences: a cross-linguistic study. Lang. Cogn. Proc. 14, 69–123.

Blom, E., Polisenská, D., Unsworth, S., 2008. The acquisition ofgrammatical gender in Dutch. Second Lang. Res. 24, 259–265.

Bock, K., Cutting, J.C., 1992. Regulating mental energy: performanceunits in language production. J. Mem. Lang. 31, 99–127.

Brysbaert, M., Mitchell, D.C., 2000. The failure to use genderinformation in parsing: a comment on van Berkum, Brown, andHagoort., 1999. J. Psycholinguist. Res. 29, 453–466.

Christianson, K., Hollingworth, A., Halliwell, J., Ferreira, F., 2001.Thematic roles assigned along the garden path linger. Cogn.Psychol. 42, 368–407.

Coulson, S., King, J.W., Kutas, M., 1998. Expect the unexpected:event-related brain response to morphosyntactic violations.Lang. Cogn. Proc. 13, 21–58.

Crain, S., Fodor, J.D., 1985. How can grammars help parsers? In:Dowty, D.R., Karttunen, L., Zwicky, A. (Eds.), Natural LanguageParsing: Psychological, Computational, and TheoreticalPerspectives. Cambridge University Press, pp. 94–128.

Davis, C.J., Perea, M., 2005. BuscaPalabras: a program for derivingorthographic and phonological neighborhood statistics andother psycholinguistic indices in Spanish. Behav. Res. Methods37, 665–671.

Demestre, J., Meltzer, S., García-Albea, J.E., Vigil, A., 1999.Identifying the null subject: evidence from event-related brainpotentials. J. Psycholinguist. Res. 28, 293–312.

Deutsch, A., 1998. Subject–predicate agreement in Hebrew:interrelations with semantic processes. Lang. Cogn. Proc. 13,575–597.

Deutsch, A., Bentin, S., 2001. Syntactic and semantic factors inprocessing gender agreement in Hebrew: evidence from ERPsand eye movements. J. Mem. Lang. 45, 200–224.

Donchin, E., Coles, M.G.H., 1988. Is the P300 component amanifestation of context updating? Behav. Brain Sci. 11,357–374.

Dowens, M.G., Vergara, M., Barber, H.A., 2010. Morphosyntacticprocessing in late second-language learners. J. Cogn. Neurosci.22, 1870–1887.

Felser, C., Clahsen, H., Münte, T.F., 2003. Storage and integration inthe processing of filler-gap dependencies: an ERP study oftopicalization and wh-movement in German. Brain Lang. 87,345–354.

Ferreira, F., Bailey, K.G.D., Ferraro, V., 2002. Good-enoughrepresentations in language comprehension. Cur. Dir. Psych.Sci. 11, 11–15.

Fodor, J.D., Inoue, A., 1994. The diagnosis and cure of garden paths.J. Psycholinguist. Res. 23, 407–434.

Ford, M., 1983. A method for obtaining measures of local parsingcomplexity throughout sentences. J. Verb. Learn. Verb. Behav.22, 203–218.

Foss, D.J., Cairns, H.S., 1970. Some effects of memory limitationupon sentence comprehension and recall. J. Verb. Learn. Verb.Behav. 9, 541–547.

Franck, J., Vigliocco, G., Antón-Méndez, I., Collina, S., Frauenfelder, U.H., 2008. The interplay of syntax and form in sentence production:a cross-linguistic study of form effects on agreement. Lang. Cogn.Proc. 23, 329–374.

Frazier, L., 1987. Syntactic processing: evidence from Dutch. Nat.Lang. Ling. Theory 5.4, 519–559.

77B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


Frazier, L., Clifton, C., Jr. 1998. Sentence reanalysis and visibility. In:Fodor, J.D., Ferreira, F. (Eds.), Reanalysis in Sentence Processing.Kluwer, Dordrecht, pp. 143–177.

Frazier, L., Flores d'Arcais, G.B., 1989. Filler-driven parsing: a studyof gap filling in Dutch. J. Mem. Lang. 28, 331–344.

Friederici, A.D., Weissenborn, J., 2007. Mapping sentence form ontomeaning: the syntax–semantic interface. Brain Res. 46, 50–58.

Friederici, A.D., Hahne, A., Mecklinger, A., 1996. Temporal structureof syntactic parsing: early and late event-related brain potentialeffects. J. Exp. Psychol. Learn. Mem. Cogn. 22, 1219–1248.

Friederici, A.D., Steinhauer, K., Mecklinger, A., Meyer, M., 1998.Working memory constraints on syntactic ambiguity resolutionas revealed by electrical brain responses. Bio. Psychol. 47,193–221.

Friederici, A.D., Mecklinger, A., Spencer, K.M., Steinhauer, K.,Donchin, E., 2001. Syntactic parsing preferences and their on-linerevisions: a spatio-temporal analysis of event-related brainpotentials. Cogn. Brain Res. 11, 305–323.

Gibson, E., 1998. Linguistic complexity: locality of syntacticdependencies. Cogn. 68, 1–76.

Gunter, T.C., Stowe, L.A., Mulder, G., 1997. When syntax meetssemantics. Psychophys. 34, 660–676.

Gunter, T.C., Friederici, A.D., Schriefers, H., 2000. Syntactic genderand semantic expectancy: ERPs reveal early autonomy and lateinteraction. J. Cogn. Neurosci. 12, 556–568.

Hagoort, P., 2003. Interplay between syntax and semantics duringsentence comprehension: ERP effects of combining syntacticand semantic violations. J. Cogn. Neurosci. 15, 883–899.

Hagoort, P., Brown, C.M., 1999. Gender electrified: ERP evidence onthe syntactic nature of gender processing. J. Psycholinguist.Res. 6, 715–728.

Hagoort, P., Brown, C., Groothusen, J.T., 1993. The syntacticpositive shift (SPS) as an ERP measure of syntactic processing.Lang. Cogn. Proc. 8, 439–483.

Hahne, A., Friederici, A.D., 1999. Electrophysiological evidence fortwo steps in syntactic analysis: early automatic and latecontrolled processes. J. Cogn. Neurosci. 11, 194–205.

Hahne, A., Friederici, A.D., 2002. Differential task effects onsemantic and syntactic processes as revealed by ERPs. Cogn.Brain Res. 13, 339–356.

Hammer, A., Jansma, B.M., Lamers, M., Münte, T.F., 2008. Interplayof meaning, syntax and working memory during pronounresolution investigated by ERPs. Brain Res. 16, 177–191.

Haskell, T.R., MacDonald, M.C., 2005. Constituent structure andlinear order in language production: evidence from subject–verbagreement. J. Exp. Psychol. Learn. Mem. Cogn. 31, 891–904.

Hickok, G., Zurif, E., Canseco-Gonzalez, E., 1993. Structural descriptionof agrammatic comprehension. Brain Lang. 45, 371–395.

Holmes, V., 1973. Order of main and subordinate clauses insentence perception. J. Verb. Learn. Verb. Behav. 12, 285–293.

Kaan, E., 2002. Investigating the effects of distance and numberinterference in processing subject–verb dependencies: an ERPstudy. J. Psycholinguist. Res. 31, 165–193.

Kaan, E., Swaab, T.Y.Y., 2003. Repair, revision, and complexity insyntactic analysis: an electrophysiological differentiation. J.Cogn. Neurosci. 15, 98–110.

Kaan, E., Harris, T., Gibson, E., Holcomb, P.J., 2000. The P600 as anindex of syntactic integration difficulty. Lang. Cogn. Proc. 15,159–201.

Keppel, G., Wickens, T.D., 2004. Design and Analysis: A Researcher'sHandbook, 4th edition. Prentice Hall, Upper Saddle River, NJ.

King, J., Just, A., 1991. Individual differences in syntactic processing:the role of working memory. J. Mem. Lang. 30, 580–602.

Kluender, R., Kutas, M., 1993. Bridging the gap: evidence from ERPson the processing of unbounded dependencies. J. Cogn.Neurosci. 5, 196–214.

Kolk, H.H.J., Chwilla, D.J., van Herten, M., Oor, P.J.W., 2003.Structure and limited capacity in verbal working memory: astudy with event-related potentials. Brain Lang. 85, 1–36.

Kotz, S.A., 2009. A critical review of ERP and fMRI evidence on L2syntactic processing. Brain Lang. 109, 68–74.

Kuperberg, G., 2007. Neuralmechanisms of language comprehension:challenges to syntax. Brain Res. 1146, 23–49.

Kutas, M., Hillyard, S.A., 1983. Event-related brain potentials togrammatical errors and semantic anomalies. Mem. Cogn. 11,539–550.

Kutas, M., Van Petten, C., Kluender, R., 2006. Psycholinguisticselectrified II: 1995–2005, In: Traxler, M., Gernsbacher, M.A.(Eds.), Handbook of Psycholinguistics, Second Edition. Elsevier,New York, pp. 659–724.

MacDonald, M.C., Just, M.A., Carpenter, P.A., 1992. Workingmemory constraints on the processing of syntactic ambiguity.Cogn. Psychol. 24, 56–98.

MacWhinney, B., Bates, E., 1989. The Crosslinguistic Study ofSentence Processing. Cambridge University Press, New York.

Martín-Loeches,M., Nigbur, R., Casado, P., Hohlfeld, A., Sommer,W.,2006. Semantics prevalence over syntax during sentenceprocessing: a brain potential study of noun–adjective agreementin Spanish. Brain Res. 1093, 178–189.

McCarthy, G., Wood, C.C., 1985. Scalp distributions of event-relatedpotentials: an ambiguity associated with analysis of variancemodels. Electroenceph. Clin. Neurophys. Evoked Potentials 62,203–208.

McKinnon, R., Osterhout, L., 1996. Constraints on movementphenomena in sentence processing: evidence fromevent-related brain potentials. Lang. Cogn. Proc. 11, 495–523.

Mecklinger, A., Schriefers, H., Steinhauer, K., Friederici, A.D., 1995.Processing relative clauses varying on syntactic and semanticdimensions: an analysis with event-related potentials. Mem.Cognit. 23, 477–494.

Moreno-Sandoval, A., Goñi-Mendoyo, J., 2002. Spanish inflectionalmorphology in DATR*. J. Logic Lang. Info. 11, 79–150.

Münte, T.E., Heinze, H.J., 1994. ERP negativities to syntactic violationsin written text. In: Heinze, H.J., Münte, T.F., Mangun, G.R. (Eds.),Cognitive Electrophysiology. Birkhäuser, Boston, pp. 211–238.

Münte, T.E., Heinze, H.J., Mangun, G.R., 1993. Dissociation of brainactivity related to syntactic and semantic aspects of language.J. Cogn. Neurosci. 5, 335–344.

Münte, T.F., Szentkuti, A., Wieringa, B.M., Matzke, M., Johannes, S.,1997. Human brain potentials to reading syntactic errors insentences of different complexity. Neurosci. Lett. 235, 105–108.

Nevins, A., Dillon, B., Malhotra, S., Phillips, C., 2007. The role offeature-number and feature-type in processing Hindi verbagreement violations. Brain Res. 1164, 81–94.

O'Rourke, P.L., 2008. The nature of syntactic gender processing inSpanish: an ERP study. Doctoral dissertation. Retrieved fromProQuest Dissertations and Theses. (Accession Order No.3336981).

Osterhout, L., Hagoort, P., 1999. A superficial resemblance does notnecessarilymean you are part of the family: counterargumentsto Coulson, King and Kutas (1998) in the P600/SPS-P300 debate.Lang. Cogn. Proc. 14, 1–14.

Osterhout, L., Holcomb, P.J., 1992. Event-related potentials elicitedby syntactic anomaly. J. Mem. Lang. 31, 785–806.

Osterhout, L., Mobley, L.A., 1995. Event-related potentials elicitedby failure to agree. J. Mem. Lang. 34, 739–773.

Osterhout, L., Holcomb, P.J., Swinney, D.A., 1994. Brain potentialselicited by gardenpath sentences: evidence of the applicationof verb information during parsing. J. Exp. Psychol. Learn. Mem.Cogn. 20, 786–803.

Pakulak, E., Neville, H., 2010. Proficiency differences insyntactic processing of monolingual native speakersindexed by event-related potentials. J. Cogn. Neurosci. 22,2728–2744.

Patson, N.D., Darowski, E.S., Moon, N., Ferreira, F., 2009. Lingeringmisinterpretations in garden-path sentences: evidence fromaparaphrasing task. J. Exp. Psychol. Learn. Mem. Cogn. 35,280–285.

78 B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9


Phillips, C., Kazanina, N., Abada, S.H., 2005. ERP effects of theprocessing of syntactic long-distance dependencies. Cogn.Brain Res. 22, 407–428.

Rubin, S.R., VanPetten, C., Glisky, E.L., Newberg,W.M., 1999.Memoryconjunction errors in younger and older adults: event-relatedpotential and neuropsychological evidence. Cogn. Neuropsych.16, 459–488.

Schmitt, B.M., Lamers, M., Münte, T.F., 2002. Electrophysiologicalestimates of biological and syntactic gender violation duringpronoun processing. Cogn. Brain Res. 14, 333–346.

Steinhauer, K., White, E.J., Drury, J.E., 2009. Temporal dynamics ofsecond language acquisition: evidence from ERPs. SecondLang. Res. 25, 13–41.

Stowe, L., 1986. Evidence for online gap creation. Lang. Cogn. Proc.1, 227–245.

Teschner, R.V., Russell, W.M., 1984. The gender patterns ofSpanish nouns: an inverse dictionary-based analysis. HispanicLing. 1, 111–132.

Van Berkum, J.J.A., Brown, C.M., Hagoort, P.M., 1999. When doesgender constrain parsing? Evidence from ERPs. J. Psycholing.Res. 29, 467–481.

Van Berkum, J.J.A., Hagoort, P., Brown, C.M., 2000. The use ofreferential context and grammatical gender in parsing: a reply toBrysbaert and Mitchell (2000). J. Psycholing. Res. 29, 467–481.

VanHerten,M., Chwilla, D.J., Kolk, H.H., 2006.When heuristics clashwith parsing routines: ERP evidence for conflict monitoring insentence perception. J. Cogn. Neurosci. 18, 1181–1197.

Van Petten, C., Luka, B.L., 2006. Neural localization of semanticcontext effects in electromagnetic and hemodynamic studies.Brain Lang. 97, 279–293.

Van Petten, C., Senkfor, A.J., 1996. Memory for words and novelvisual patterns: repetition, recognition, and encoding effects inthe event-related brain potential. Psychophys. 33, 491–506.

Wicha, N.Y.Y., Bates, E.A., Moreno, E.M., Kutas, M., 2003.Potato not pope: human brain potentials to genderexpectation and agreement in Spanish spoken sentences.Neurosci. Lett. 346, 165–168.

Wicha, N.Y.Y., Moreno, E.M., Kutas, M., 2004. Anticipatingwords and their gender: an event-related brain potentialstudy of semantic integration, gender expectancy, andgender agreement in Spanish sentence reading. J. Cogn.Neurosci. 16, 1272–1288.

79B R A I N R E S E A R C H 1 3 9 2 ( 2 0 1 1 ) 6 2 – 7 9

author's personal copy - binghamton...

Documents