Prospective and retrospective semantic processing: Prediction, time, and relationship strength in...

Brain & Language 135 (2014) 115–129

Contents lists available at ScienceDirect

Brain & Language

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

Prospective and retrospective semantic processing: Prediction, time,and relationship strength in event-related potentials

http://dx.doi.org/10.1016/j.bandl.2014.06.0010093-934X/� 2014 Elsevier Inc. All rights reserved.

⇑ Corresponding author. Address: 30 Campus Road, Annandale-on-Hudson,NY 12504, USA.

E-mail address: [email protected] (B.J. Luka).

Barbara J. Luka a,⇑, Cyma Van Petten b

a Bard College, Annandale-on-Hudson, NY, USAb Binghamton University, 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 2 June 2014

Keywords:Semantic associationWord pairsN400PredictionIntegrationSerialParallel

Semantic context effects have variously been attributed to prospective processing – predictions aboutupcoming words – or to retrospective appreciation of relationships after reading both context and target.In two experiments, we altered the core variable distinguishing prospective from retrospective process-ing, namely time. Word pairs varying in strength of relationship were presented sequentially, to allowtime for anticipation of the second word, or simultaneously. For both sorts of presentation, the amplitudeof the N400 component of the event-related potential was graded from Unrelated to Moderate/Weak toStrong associates. Strong associates showed a temporal advantage over weaker associates – an earliercontext effect – only during sequential presentation. Spatial distributions of the N400 context effects alsodiffered for simultaneous versus sequential presentation.

� 2014 Elsevier Inc. All rights reserved.

1. Introduction

Semantic context effects are evident across a broad swath ofdependent measures in cognitive psychology. Whether the contextis a single word or a sentence fragment, words preceded by relatedcontexts can be accurately identified with briefer exposure dura-tions or higher levels of noise than words preceded by unrelatedcontexts (Miller & Isard, 1963; Tulving & Gold, 1963), and receivefaster responses in a variety of tasks including deciding whethera letter string is actually a word (lexical decision, Meyer &Schvaneveldt, 1971), naming aloud (Jacobson, 1973), and semanticjudgments such as whether the item refers to an animate orconcrete entity (McRae, de Sa, & Seidenberg, 1997). Across severaldecades, many functional mechanisms have been proposed toaccount for these effects, and these accounts can be categorizedin a variety of ways. Here, we focus on one sort of division, namelythe role of time.

1.1. Prospective versus retrospective accounts of semantic contexteffects

Prospective explanations of semantic context effects stipulatethat, after presentation of a related context, some aspect of thetarget word is pre-activated, so that the target has already been

partially processed in advance of its physical occurrence. Prospec-tive theories vary in their description of what, exactly, is pre-activated, from discrete individual words in models with localistrepresentations of items in a mental lexicon, to semantic featuresof words in models with more distributed representations. Localistversions include both passive spreading activation along linksbetween related words (Collins & Loftus, 1975), and more activeanticipation of related words (Becker, 1980; Neely, 1977). Distrib-uted versions include partial pre-activation of the target word’smeaning due to semantic features that are shared with the context(Masson, 1995; Plaut, 1995; Sharkey, 1989), or activation of event-based schemas that include instruments, objects and actors thatmight participate in the same activity, such that ‘‘broom’’ canprime ‘‘floor’’ (Hare, Jones, Thomson, Kelly, & McRae, 2009; Moss,Ostrin, Tyler, & Marslen-Wilson, 1995). These accounts vary dra-matically in their assumptions about how words and semanticknowledge are represented in long-term memory, and in theirapplicability to single-word versus sentence contexts (see VanPetten & Kutas, 1991 for discussion). Prospective explanationsare, however, united by the idea that the critical activity leadingto facilitated processing of a target word occurs in the intervalbetween the presentation of the context and the target.

In contrast to such prospective accounts, retrospective accountsof semantic context effects stress the idea that readers and listen-ers spontaneously try to find relationships among sequentialwords, and that performance of many tasks is improved (or at leastnot hindered) when such relationships can be found. For instance,one influential retrospective account is the compound cue model,

http://crossmark.crossref.org/dialog/?doi=10.1016/j.bandl.2014.06.001&domain=pdf

http://dx.doi.org/10.1016/j.bandl.2014.06.001

mailto:[email protected]


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

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

116 B.J. Luka, C. Van Petten / Brain & Language 135 (2014) 115–129

which states that sequentially presented words are combined inworking memory, and the combination matched to the contentsof long-term memory. Compound cues formed from related wordsare better matches – better retrieval cues – than unrelated pairs orthe target alone (Ratcliff & McKoon, 1988, 1995). Other retrospec-tive accounts are more closely linked to particular tasks. For thelexical decision task that has dominated behavioral experiments,it has long been noted that finding a relationship between the tar-get item and the preceding context serves as a clear signal that thetarget cannot be a nonword and should receive a ‘‘yes’’ decision(de Groot, 1983; Neely, Keefe, & Ross, 1989). The defining featureof retrospective accounts is that semantic context effects can arisefrom the comparison and combination of word meanings afterboth the context and target have been presented.

The earliest reports of semantic context effects on accuracy andreaction time were interpreted as evidence for active prediction ofupcoming words (e.g., Miller & Isard, 1963; Tulving & Gold, 1963),but prospective accounts became generally less popular in the late1970s and early 1980s (e.g., Forster, 1981; see Van Petten & Luka,2012 for review). Over the last decade, researchers using behav-ioral methods have once again begun to favor the idea that readersand listeners actively anticipate the semantic features of upcomingwords, at least, and perhaps individual words as well (Kamide,Altmann, & Haywood, 2003; Pickering & Garrod, 2007; Roland,Yun, Koenig, & Mauner, 2012).

The N400 component of the event-related potential is also verysensitive to semantic context, such that words preceded by con-gruent sentence fragments or by related single words elicit smallerN400s that those preceded by unrelated contexts (see Kutas, VanPetten, & Kluender, 2006 for review). N400 context effects havebeen observed with a variety of assigned tasks including lexicaldecision, monitoring for a target semantic category (e.g., occasionalanimal names in a stream of words), detecting repeated words orsentences, preparing to answer comprehension questions thatoccur after sentences, preparing to judge whether a probe letteroccurred in the previous word, or similarly, preparing to judgewhether a probe word occurred in the previous sentence. Critically,N400 context effects can also be observed in situations that do notrequire any specific choices or decisions imposed by the laboratorytask, with instructions only to read for comprehension. AlthoughN400 context effects are not impervious to requirements to per-form additional tasks that might detract from semantic processing(see Van Petten, 2014 for recent review), it is clear that the sensi-tivity of this component to semantic context is not tied to a partic-ular task-specific strategy. Nonetheless, there have been debatesabout the contribution of prospective versus retrospective mecha-nisms to N400 context effects or, more specifically, about theconfirmation of predictions about upcoming words versus the easeof semantic integration across multiple words after they haveoccurred (Chwilla, Brown, & Hagoort, 1995; Holcomb, 1993; Lau,Phillips, & Poeppel, 2008; Osterhout & Holcomb, 1995).

As in the behavioral literature, recent N400 studies haveemphasized the anticipation of upcoming words and/or theirmeanings (see Van Petten & Luka, 2012 for review). Many N400sentence-processing results have been interpreted as evidencefor anticipation of semantic features, if not words per se(Federmeier, 2007; Szewczyk & Schriefers, 2013; Thornhill & VanPetten, 2012; Wlotko, Federmeier, & Kutas, 2012; see alsoFedermeier, Kutas, & Schul, 2010 for similar interpretation of aword-pair study). Other results are more plausibly interpreted asevidence for prediction of full words. For instance, DeLong et al.have examined ERPs to the articles ‘‘a’’ and ‘‘an’’ in English(DeLong, Groppe, Urbach, & Kutas, 2012; DeLong, Urbach, &Kutas, 2005). These articles have identical (minimal) meaning, sothat they should never create difficulty for semantic integration.However, given a sentence about someone flying something, if

‘‘kite’’ was the most favored sentence completion, then the word‘‘a’’ in penultimate position elicited a smaller N400 than the word‘‘an’’ (as in ‘‘an airplane’’, an acceptable but less preferred ending).These results suggest that the subjects were actively predicting aspecific word, over and above some class of fly-able nouns. Simi-larly, Laszlo and Federmeier (2009) observed smaller N400s forsemantically incongruent sentence completions when these wereorthographic neighbors of a congruent word (as compared toincongruent words that were not neighbors), which may also sug-gest that semantically-based predictions extend to visual wordforms.

1.1.1. Revisiting the utility of retrospective processingDespite the clear evidence for prospective contributions to the

N400, it seems premature to dismiss the importance of retrospec-tive evaluation of semantic relationships. Language is used notonly to refer to information known to both the speaker/writerand the listener/reader, but also to communicate novel conceptsthat cannot be fully appreciated or predicted in advance. Much ofeveryday language comprehension is likely to involve a complexinterplay between the retrieval of existing concepts from memoryand the construction of new meaning from these building blocks(Coulson, 2006). Some ERP studies indicate that the overall contextof a sentence or an extended passage of discourse can overriderelationships that are likely to be pre-stored. Word-pair relation-ships that ordinarily reduce N400 amplitude can be made ineffec-tive if the pair relationships conflict with the newly-constructedmeaning of an entire sentence. For instance, although ‘‘olive-OIL’’elicited a smaller N400 than ‘‘olive-SHOES’’ for isolated pairs, thispattern of results was reversed given a sentence frame like‘‘Although they were uncomfortable to walk in, she loved her olive. . .’’ (Coulson, Federmeier, Van Petten, & Kutas, 2005). A similarreversal was observed for full sentences depending on discoursecontext. Although the final word of ‘‘The peanut was in LOVE’’ elic-ited a larger N400 than ‘‘The peanut was SALTED’’ in a list ofunconnected sentences, a story about the adventures of a peanutcharacter produced the opposite pattern of results (Nieuwland &van Berkum, 2006). In these two studies, it is difficult to determinewhen, exactly, the novel relationships were constructed given thestrong constraints imposed by the preceding context. A differentexperiment provides stronger evidence for the retrospective appre-ciation of meaning. Chwilla and colleagues found that, given anappropriate lead-in – such as a protagonist’s strong desire to gocanoeing despite the absence of paddles – readers were able tomake sense of a novel scenario like paddling a canoe with a Frisbee.As compared to equally unpredictable but uninterpretable combi-nations (like paddling with a pullover), the newly-constructedinterpretations yielded a smaller N400 (Chwilla, Kolk, & Vissers,2007). Because of the very weak predictability of words like ‘‘Fris-bee’’, this result shows a clear influence of retrospective (althoughrelatively rapid) appreciation of semantic relations. At the sametime, other studies suggest a cost for new meaning constructionas compared to relationships that are more likely to be pre-stored.As compared to more familiar or more conventional combinations,novel metaphors and novel combinations of literal word meaningshave variably shown larger N400s (Arzouan, Goldstein, & Faust,2007; Coulson & Van Petten, 2002, 2007) or enhancements of alater frontal positivity (Davenport & Coulson, 2011, 2013). Overall,the existing literature suggests contributions of both anticipatoryprocesses and of successful integration performed after all theinput has occurred.

1.2. The current study: prediction, time, and association strength

The current study uses much simpler materials than the sen-tence experiments that have suggested roles for both predictive

B.J. Luka, C. Van Petten / Brain & Language 135 (2014) 115–129 117

and retrospective semantic processing: word pairs that are likely tohave pre-existing relationships in long-term memory. Theextended duration of sentences – especially when combined withthe sequential word-by-word presentation method that is neces-sary to timelock ERPs to specific events and avoid eye movementartifacts – makes sentence processing particularly amenable topredictions about what is coming next. Word pairs more readilyallow manipulation of the key element that distinguishes prospec-tive from retrospective accounts of semantic combination, namelytime. In Experiment 1 of the current study, words are separated bya moderate delay of 700 ms, a stimulus-onset-asynchrony (SOA)that is fairly typical of ERP sentence experiments. The 700 msSOA should allow plenty of time to read the first word of a pairand, perhaps, to begin thinking about what might occur next. InExperiment 2, the two members of a pair are presented simulta-neously. Simultaneous presentation removes both any encourage-ment to engage in predictive processing, and the necessaryresource of spare time in which to do so. The semantic contexteffects in Exp. 1 may receive some contribution from predictiveprocessing, but this is likely to be much reduced in Exp. 2.

We are aware of only two previous ERP experiments usingsimultaneous presentation of word pairs. In a brief report, VanPetten and Kutas (1988) showed robust N400 differences betweensemantically related and unrelated pairs that was later in onset(�400 ms) than what is typically observed for sequentially pre-sented pairs (200–300 ms), but did not compare simultaneous tosequential presentation. Anderson and Holcomb (1995) comparedERPs to the same pairs presented simultaneously and with an SOAof 800 ms. Both presentation methods led to smaller N400s forrelated than unrelated pairs, but these semantic context effects dif-fered in both scalp distribution and onset latency. The contexteffect during sequential presentation was maximal at frontal scalpsites, whereas simultaneous presentation led to a parietally-maximal effect. Anderson and Holcomb additionally observed thatthe semantically related and unrelated conditions differed in alatency window of 200–300 ms after presentation of the secondword; with simultaneous presentation, the context effect was firstsignificant in a latency window of 300–400 ms after pair presenta-tion. However, it is unclear whether the latency advantage forsequential presentation should be attributed to the benefit of cor-rectly anticipating a related word, or simply to the greater ease ofreading one word instead of two.1

Both word-pair studies with simultaneous presentation exam-ined only a binary contrast between related and unrelated wordpairs. In contrast, graded manipulations of semantic context haveplayed a central role in the ERP sentence processing studies thathave pointed to a strong role for predictive processing. For sen-tence contexts, the standard measure of contextual strength iscloze probability, the percentage of subjects who offer a particularword to complete a sentence frame when asked to generate ‘‘thebest completion’’ or ‘‘the first word that comes to mind’’. Clozeprobabilities are derived from a normative group of participants,separate from those that participate in the subsequent ERPexperiment. N400 amplitudes are inversely correlated with clozeprobability, such that more predictable words elicit smallerN400s (DeLong et al., 2005, 2012; Kutas & Hillyard, 1984; Kutaset al.,1984; Thornhill & Van Petten, 2012; Wlotko & Federmeier,2013). For word pairs, strength of relationship is derived from aparallel generative procedure: the percentage of subjects who pro-duce Word B in response to cue Word A, association strength. Thecurrent experiments implement a gradient of relationship strength

1 Both studies split the pairs across the visual midline – one in the left visual fieldand one in the right – such that the onset delays of the semantic context effect in thesimultaneous condition may have additionally included hemispheric transfer times.Current Experiment 2 uses vertically-arranged pairs to avoid this factor.

by using a published word association norm to select strongly,moderately and weakly-associated cues for the same criticalwords.

With a delay between the presentation of the first and secondwords of a pair, several published ERP experiments show gradedERP amplitudes: most negative for unrelated words, intermediatefor weak relationships, and least negative for strong relationships.Most of these have included an overt relationship judgment(Frishkoff, 2007; Kandhadai & Federmeier, 2010a, 2010b; Kutas& Hillyard, 1989; Roehm, Bornkessel-Schlesewsky, Rösler, &Schlesewsky, 2007; Rösler, Streb, & Haan, 2001), such that it is pos-sible that the results reflected some combination of a smaller N400and a larger decision-related P300 for stronger relationships (seeHillyard, Squires, Bauer, & Lindsay, 1971; Paul & Sutton, 1972;Squires, Squires, & Hillyard, 1975 for the impact of decisionconfidence on P300 amplitude). Recently, we isolated the strengtheffect to the N400 by instead asking participants to make a lexicaldecision about the second item of a pair, or to perform a delayedletter-matching task which postpones all decisions until well afterthe ERP elicited by the second word of a pair (Luka & Van Petten, inpress). In those experiments, we examined only N400 amplitudes,like most prior reports.

Here, we examine whether the greater predictability of strongassociates might also influence the latency of the N400. A simpleproposal is that after viewing the first word of a pair, subjects oftenpredict/guess/generate a second word, if there is a sufficiently longdelay before some second word is actually presented. This men-tally generated word is more likely to correspond to the actual sec-ond word for strongly- than weakly-associated pairs. (Recall thatthe definition of association strength is the probability with whicha normative group of subjects generate Word 2 in response toWord 1.) In question is whether such ‘‘lucky guesses’’ also speedprocessing of the second word via top-down support for the phys-ically-presented item. In our recent work, we found that althoughstrength-of-association influenced N400 amplitude, it did notinfluence reaction times in the lexical decision task (Luka & VanPetten, in press). Together with other reports of null strength-effects in lexical decision RTs (Anaki & Henik, 2003; Fischler,1977; Fischler & Goodman, 1978; Hodgson, 1991; Koriat, 1981;Kroll & Potter, 1984; Nation & Snowling, 1999; Sánchez-Casas,Ferré, Demestre, García-Chico, & García-Albea, 2012), this patternof results suggests that ERPs might offer a better view of processingspeed. In current Experiment 1 (sequential presentation), weexamine the timing of N400 effects from one of our recent exper-iments (Exp. 2 of Luka & Van Petten, in press) and compare these tonew Exp. 2 with the same stimuli presented simultaneously. Ifsuccessful prediction speeds processing, we expect associationstrength to influence ERP latency during sequential but not simul-taneous presentation of word pairs. More specifically, we examineonset latency of the difference between associated and unrelatedword pairs because the onset of an averaged ERP is determinedby those trials within a condition that have the earliest latencies(i.e., the hypothetical ‘‘lucky guesses’’ for the strongly-associatedpairs).

2. Methods

2.1. Participants

Participants in both experiments had normal or corrected-to-normal vision and reported no history of psychiatric disorder,neurological disorder, learning disability, or current use of medica-tions thought to affect the central nervous system. Thirty volun-teers (14 men, 16 women) were paid for their participation inExperiment 1. Their mean age was 23.4 years (range 19–33), with

Table 2Critical noun characteristics (mean and se).

Weak Medium Strong

Association strength 5.9 (0.3) 11.7 (0.5) 23.6 (1.0)Backward association strength 1.7 (0.3) 1.8 (0.3) 3.4 (0.5)KF word frequency 151 (14) – –HAL word frequency 512 (57) – –Word length in letters 4.9 (0.1) – –Orthographic neighbors 5.7 (0.3) – –

Note: The four conditions used the same critical nouns, so that word frequency,length and number of orthographic neighbors are the same across the weakly,moderately and strongly associated pairs, as well as in the unassociated pairsformed by recombining words from the related pairs. Association strengths fromthe Edinburgh Associative Thesaurus (Kiss et al., 1973) as the percent of subjectswho respond with the critical word when given the context word as a cue.Backward association strength is the percent of subjects who responded with thecontext word when given the critical word as a cue. Word frequency computed asthe sum of all regularly inflected forms in the Kucera and Francis (1967) corpus (KFword frequency) and the newer HAL corpus (HAL word frequency, available from theEnglish Lexicon Project, Balota et al., 2007), both as counts per million. Ortho-graphic neighbors are the number of other words than can be formed by changingone letter, these counts also from the English Lexicon Project.


16.2 years of formal education (2 years of college to 4 years ofgraduate study). Twenty-four were right-handed and 6 left-handed; 10 of the right-handers reported an immediate familymember (parent or sibling) who was left-handed. Data from 4additional participants were not analyzed: 3 had high numbersof trials contaminated by non-EEG artifacts (more than 50% in atleast one condition), and 1 showed very low accuracy in the let-ter-probe task described below (59.7% versus a mean of 95.6% forthe retained subjects).

Twenty-four volunteers (11 men, 13 women) were paid fortheir participation in Experiment 2. Their mean age was 22.6 years(range 19–32), with 15.2 years of formal education (1 year of col-lege to 2 years of graduate study). All were right-handed; 12reported an immediate family member (parent or sibling) whowas left-handed. Data from 3 additional participants were not ana-lyzed because they had high numbers of trials contaminated bynon-EEG artifacts (more than 50% in at least one condition).

2.2. Stimuli

Two hundred and forty nouns were paired with three cue wordseach, to form triplets of semantically related pairs, as in ORE-METAL, WELD-METAL, and SCRAP-METAL (see Table 1 for moreexamples). The nouns were offered as responses to the cue wordsin a free association procedure with 100 subjects (the EdinburghAssociative Thesaurus, EAT; Kiss, Armstrong, Milroy, & Piper,1973). For the Strong condition, the cue elicited the critical nounfrom an average of 23.6% of the subjects in the EAT. This associativestrength is very close to the average for the most popular responseto the more than 8000 cue words in the EAT (24.7%). The Moderateand Weak conditions included pairs whose associative strengthswere roughly half (11.7%) and a quarter (5.9%) of the Strongcondition; the pairwise differences in association strength werestatistically robust (Strong vs. Moderate, Moderate vs. Weak, botht(239) > 14, p < .00001). The critical nouns’ frequency of usage andnumber of orthographic neighbors are listed in Table 2 becauseboth factors are known to influence N400 amplitude (Laszlo &Federmeier, 2011; Van Petten, 1995).

Each subject received all 240 critical nouns, one quarter with aStrong cue, one quarter with a Moderate cue, one quarter with aWeak cue, and one quarter with an Unrelated cue. Related contextswere rotated across subjects so that each critical noun appearedequally often in a Strong, Moderate or Weak pair although an indi-vidual subject viewed each critical noun only once. The Unrelatedpairs were formed by recombining cue words and critical nouns.The same cue words thus appeared in both related and unrelatedpairs across subjects, but each individual subject viewed a cue onlyonce. To equate the proportion of semantically related and unre-lated pairs, 120 semantically unrelated word pairs (unanalyzed fill-ers) were added to each stimulus list, so that subjects viewed 180related and 180 unrelated word pairs in total. The four pair types

Table 1Stimulus examples.

Unassociated cue Weak cue Medium cue Strong cue Noun

Yogurt Violent Deed Delayed ActionGround Consent Mature Child AdultSmooth Gate Barn Knock DoorViolent Dollar Pocket Invest MoneyElbow Pour Coconut Yogurt MilkMature Matching Shades Rainbow ColorsKnock Forest Burning Flame FireMatching Smooth Peeling Flesh SkinPocket Till Ground Fertile SoilFlame Space Decorate Elbow Room

(Strong, Moderate, Weak, Unrelated) and the unrelated filler pairswere randomly intermixed.

In Experiment 2, the two words of a pair were presented simul-taneously, so that the ERPs indexed the processing of both wordsas well as their relationship. The second words of each pair arethe same across conditions, but the Strong, Moderate and Weakcue words are necessarily different. Table 3 thus shows the basiccharacteristics of the cue words. The three varieties of cue wordwere very similar in word length and number of orthographicneighbors. Table 3 shows two measures of frequency of usage,one derived from an older word count (Kucera & Francis, 1967)used by most previous ERP experiments with English stimuli, andone based on the newer HAL corpus (Balota et al., 2007). Theseshowed differences among the means, but small differences giventheir variances. The differences were also in opposite directions forthe two frequency counts, so that we consider these cues ade-quately matched for word frequency.

2.3. Procedure

Subjects were seated in a comfortable chair facing a CRT moni-tor. During a block of trials, the screen continuously displayed acentral frame in which all text stimuli appeared. The sequence ofevents for each trial in Experiment 1 was: (a) cue word (200 msduration), (b) eye fixation frame only (500 ms), (c) critical noun(200 ms), (d) fixation frame only (1500 ms), and (e) a single letterof the alphabet with a question mark (200 ms). Subjects used theindex fingers of each hand to indicate whether the probe letteroccurred in either of the preceding words or in neither word. Forboth related and unrelated pairs, half of the correct answers were

Table 3Cue word characteristics (mean and se).

Unassociated Weak Medium Strong

KF word frequency 64 (4) 53 (8) 69 (7) 71 (6)HAL word frequency 200 (15) 231 (24) 203 (25) 167 (27)Word length 6.0 (0.1) 6.1 (0.1) 5.9 (0.1) 6.1 (0.1)Orthographic neighbors 3.9 (0.2) 3.9 (0.3) 3.8 (0.3) 3.8 (0.3)

Note: Weak, medium and strong cue words also served as unassociated cues(rotated across subjects), so that characteristics of the unassociated cues are themeans of the weakly, moderately and strongly associated cues. Word frequencycomputed as the sum of all regularly inflected forms in the Kucera and Francis(1967) corpus (KF word frequency) and the newer HAL corpus (HAL word frequency,available from the English Lexicon Project, Balota et al., 2007), both as counts permillion. Orthographic neighbors are the number of other words that can be formedby changing one letter; these counts also from the English Lexicon Project.

2 Subjecti = (n � subaveragemean) � ((n � 1) � subaveragei), where subjecti is therecovered latency for that subject, n is the number of subjects, subaveragemean isthe mean latency across all n subaverages, and subaveragei is the latency from thesubaverage that excludes the i-th participant.


‘‘present’’ and half were ‘‘absent’’. ‘‘Present’’ letters were equallylikely to occur in the first or second word of a pair. The mappingbetween right and left index fingers and present/absent was coun-terbalanced across subjects. The letter probe task was assigned toencourage attentive reading of the word pairs, but postponestask-related decisions beyond the time window in which the ERPsare expected to be influenced by the semantic relationshipbetween words of a pair. The delayed letter-probe task thus elim-inates decision-related P300s from the ERPs of interest and allowsisolation of the N400 semantic context effects (Kutas & Hillyard,1989).

In Experiment 2, the two members of each word pair were pre-sented together, with the cue words centered above the criticalnouns. Words were presented inside the continuously-present fix-ation frame in the same font used in Experiment 1. The display oftwo words occupied approximately 1.4 degrees of vertical angle,with about 0.6 degrees separating the lower edge of a top wordfrom the upper edge of a bottom word. A five-letter word sub-tended about 2 degrees horizontally. Both words of a pair werethus likely to fall within foveal vision. However, pilot work indi-cated that it was difficult to read two words with the 200 ms dis-play duration used in Experiment 1, so that this was increased to400 ms (see Anderson & Holcomb, 1995 for the same durationdecision for simultaneous pairs). The probe letter followed after1500 ms.

In both experiments, the 360 word pairs were divided into 8blocks of 45, with brief rest periods between blocks. A short prac-tice block with unrelated pairs and related pairs varying in associ-ation strength preceded the regular trials, to familiarize subjectswith the stimulus timing, key-press apparatus, and letter-probetask.

2.4. Electrophysiological methods

The electroencephalogram was recorded from 29 scalp sites: 7spanning the midline of the scalp from prefrontal to occipital(midline, Fpz, Fz, Fcz, Cz, Cpz, Pz, Oz), 7 lateral pairs closer to themidline (near-lateral, Fp1, Fp2, F3, F4, Fc3, Fc4, C3, C4, Cp3, Cp4,P3, P4, O1, O2), and 4 lateral pairs farther from the midline(far-lateral, Ft7, Ft8, T3, T4, Tp7, Tp8, T5, T6). Scalp sites werereferenced to the left mastoid during recording, as was a right mas-toid site and an electrode below the right eye used to detect blinksand vertical eye movements. An additional pair of electrodes nearthe external canthi of the two eyes (right referenced to left) wasused to detect horizontal eye movements. Amplifier bandpasswas 0.01–100 Hz; sampling rate was 250 Hz, and gain was 50,000.

Trials contaminated by blink, eye movement, or amplifier satu-ration artifacts were rejected prior to averaging the trials into ERPsfor each condition, in a epoch beginning 200 ms before the onset ofthe second words of each pair and continuing to 1000 ms afteronset. After artifact rejection, ERPs for each subject were com-prised of a mean of 53.8–55.7 trials each for the conditions ofStrong, Moderate, Weak and Unrelated in both experiments (min-imum of 32 trials). Prior to measurement, ERPs were re-referencedto an average of the right and left mastoids.

2.5. ERP measurements

2.5.1. Onset latencies of the context effectsOnset latency of the difference between a semantically related

and unrelated condition is relevant for two distinct issues in thecurrent study. The first is to provide an estimate of the minimumdelay in semantic processing when the visual display includestwo words as compared to one. The sole prior study to comparesimultaneous to sequential presentation of word pairs observed adelay in the onset of the N400 context effect for simultaneous

presentation (Anderson & Holcomb, 1995). We expect some delayhere as well, which may arise from any stage of word processingfrom basic visual processes forward. The second is an assessmentof the hypothesis that extra time between the two members of apair may afford a ‘‘head start’’ for appreciation of the second word’smeaning, if the relationship between the two words is sufficientlystrong to allow successful prediction of the word itself, or some ofits semantic features. If so, the onset of the N400 context effect forstrongly-associated pairs should be earlier than for weaker pairswith sequential presentation, but this temporal advantage shouldbe reduced or eliminated with simultaneous presentation.

The onset latency of an ERP difference between conditions isnotoriously difficult to identify, largely because the earliest portionof an effect is small in amplitude and thus requires analysis of thatportion of the waveform with low signal-to-noise ratio (SNR). Overthe last twenty years, a variety of methods for identifying onsetlatency have been developed, each with strengths and weaknesses.Here, we apply two very different methods with the expectationthat their outcomes should converge if the influences of SOAand/or association strength are robust.

2.5.1.1. Jacknife. All ERP methods incorporate averaging to reducenoise. In the most standard procedure, the EEG during each trialwithin a condition is averaged for each subject to form an ERP,some amplitude measure is computed for each subject’s averagedERP, and the mean and variance of those measures across subjectsis submitted to some statistical test. The so-called grand average –the average across individuals typically displayed in figures – rep-resents the mean, whereas the variance is computed across theindividual subjects’ ERPs. The jackknife approach takes the noisereduction offered by averaging one step further, by using grandaverages to compute the variance as well as the mean (Miller,Patterson, & Ulrich, 1998). This is achieved by creating multiplegrand averages from the sample of participants, each leaving outthe data from one person. A sample of 30 subjects would lead to30 such subaverages, the first leaving out Subject #1, the secondleaving out Subject #2, and the last leaving out Subject #30. Eachindividual subject thus contributes to the variance around themean by his absence, rather than by his presence as in the standardprocedure.

The jackknife is a general procedure that can be applied to anysort of data. In ERP research, it was developed to estimate onsetlatency of the Lateralized Readiness Potential (Miller et al., 1998;Ulrich & Miller, 2001) and has only recently been applied to onsetlatencies of other experimental effects (Eimer & Grubert, 2014;Grubert & Eimer, 2013; Sprondel, Kipp, & Mecklinger, 2013). Todate, these have not included N400 effects. To estimate an onsetlatency, the amplitude of a difference waveform (one conditionminus another) is measured at each time point in each of the suba-verages, and the time at which amplitude exceeds some criterion isscored as the onset for that subaverage. These numbers can then besubjected to any statistical procedure together with a correction inthe degrees of freedom (Ulrich & Miller, 2001). Alternatively, thecontribution of each participant can be recovered via a simple for-mula2 and these numbers can be fed to any statistical test withoutcorrecting the degrees of freedom (Smulders, 2010); this was themethod used here.

It is important to note that the jackknife procedure per se andthe amplitude criterion used to define the onset of an effect areorthogonal, and that the latter is selected by the user. Amplitudecriteria have been defined in various ways: A) as absolute


microvolts (e.g., the point at which the difference waveformreaches 1 lV, see Eimer & Grubert, 2014 for an example of thisapproach), B) as a percentage of maximum amplitude of the differ-ence wave (e.g., 50%, Grubert & Eimer, 2013), or C) as a multiple ofthe estimated noise level of the data (Osman, Bashore, Coles,Donchin, & Meyer, 1993). In a comparison of these criteria withsimulated data, Miller et al. (1998) note that both absolute andpercent criteria can lead to distorted latency estimates if the con-ditions to be compared differ in maximum amplitude or in theslope of the amplitude rise across time. For the comparisons acrosslevels of association strength here, prior results indicate that themaximum amplitude of the N400 context effect will be larger forstrongly- than weakly-associated word pairs (Luka & Van Petten,in press). We thus use a criterion based on (C), which we alsobelieve is less arbitrary than selecting some absolute amplitudeor percentage.

Specifically, the Strong, Moderate, and Weak context effectswere isolated as differences from the Unrelated condition in thegrand average with data from all participants and all scalp sites.Noise in each difference wave was estimated as the standard devi-ation of amplitude during the prestimulus epoch of 200 ms – avalue that, in theory, would be zero with an infinite signal-to-noiseratio (a perfectly flat prestimulus baseline). Onset times for thejackknife were then defined as the post-stimulus time at which adifference wave exceeded two standard deviations of the presti-mulus amplitudes measured at each data point (every 4 ms).

2.5.1.2. Sequential significance tests. Instead of setting an amplitudecriterion for the onset of an effect, an alternative procedure tests thesignificance of differences between conditions by comparing themat each sequential data point (e.g., every 4 ms for a digital samplingrate of 250 Hz). This procedure raises the problem of how to correctfor false positives arising from multiple comparisons. One commonsolution is to impose a ‘‘temporal cluster’’ criterion, for instance bydefining onset time as the first test with probability less than somealpha level (e.g., p < .05), if that is followed by some number of teststhat also have probability less than that alpha (see Anderson &Holcomb, 1995; Schmitt, Münte, & Kutas, 2000; Thorpe, Fize, &Marlot, 1996; Timmer, Vahid-Gharavi, & Schiller, 2012 for applica-tions of this approach). However, the exact degree of protectionagainst false positives offered by a temporal-cluster criterion is dif-ficult to estimate without knowing, in advance, the correlationbetween sequential data points for the signal (the true differencebetween conditions) and for the noise (EEG that is unrelated toeither condition; see Guthrie & Buchwald, 1991 for discussion oftemporal autocorrelation).

Permutation tests are also designed to control Type I (false posi-tive) error rates, by determining the exact distribution of a test sta-tistic across all of the comparisons conducted. This is achieved bycomputing the test statistic under all possible rearrangements ofthe condition labels for data points. For instance, for a paired t-testwith a sample of three subjects, the t-statistic would be computedin the standard way (all condition labels in their true order), butalso after reversing the condition labels for one of the three subjects(three possibilities), reversing the labels for two subjects (threepossible combinations) and, finally, reversing the labels for all threesubjects. For a family of comparisons – such as comparing condi-tions at multiple time points – this procedure would be repeatedto generate the distribution of t cumulated across all of the desiredtime points. The standard t-statistic is then deemed significant if itis larger than 95% of those in the permutation distribution if thedesired family-wise alpha level is .05, or greater than 99% if thedesired alpha level is .01, etc. In practice, full permutation testsare rarely conducted because the number of computations quicklybecomes impractical with realistic size samples. E.g., for a pairedcomparison with 30 subjects, creating the full permutation

distribution would require 230 or 1,073,741,824 t-tests for eachtime point. More typically, researchers use a Monte Carlo procedureof sampling some smaller number of randomly selected rearrange-ments of the data in order to estimate the full permutation distribu-tion (Dwass, 1957; Edgington, 1969). Monte Carlo samples of10,000 permutations are generally considered to provide a closeapproximation of the full distribution (Blair & Karniski, 1993;Jöckel, 1986; see Backer, Hill, Shahin, & Miller, 2010; Dimigen,Sommer, Hohlfield, Jacobs, & Kliegl, 2011; Doi & Shinohara, 2012;Kuefner, Jacques, Prieto, & Rossion, 2010; Laszlo & Federmeier,2014 for applications of permutation tests to determining latencyranges of ERP and EEG effects).

Here, we assessed the onset latency of the Strong, Moderate andWeak context effects by comparing each of the related conditionsto the Unrelated control condition at each time point (every4 ms) with 30,000 permutations of each, and an alpha of p < .01,one-tailed (Camargo, Azuaje, Wang, & Zheng, 2008).

Overall, the jackknife procedure focuses on one aspect of theproblem of measuring onset latencies, namely low signal-to-noiseat the low-amplitude onset of an ERP effect. In contrast, permuta-tion tests offer no special advantage for noise reduction, but insteadfocus on avoiding false positives from large numbers of statisticaltests. For both the jackknife and the permutation approaches todetermining onset latency, we also applied a temporal-cluster cri-terion to protect against spurious identification of noise as genuineonset, namely that differences between conditions persist for aminimum of 40 ms. Both procedures were applied to data averagedacross all scalp sites to provide the highest possible signal-to-noiseratio.

2.5.2. AmplitudesThe impact of association strength on ERP amplitudes was

assessed by comparing mean amplitudes for the Strong, Moderateand Weak conditions in a 300 ms time window, with respect to the200 ms prestimulus baseline. For each experiment, the beginningof the time window was determined by the onset latencies identi-fied in the jackknife and permutation procedures. As elaborated inthe Results, different latency windows were used for the twoexperiments because, as expected, simultaneous presentation ledto an overall delay in the onset of the N400 difference betweenrelated and unrelated conditions.

2.6. Statistical tests

Statistical tests on onset latencies were ANOVAs with associa-tion strength as a repeated measure. Statistical tests on amplitudeswere ANOVAs with association strength and scalp site as repeatedmeasures. The Huhyn–Feldt correction for nonsphericity ofvariance was applied to all F-ratios with more than one degree offreedom in the numerator; reported are the original df, correctedprobability level, and the epsilon (e) correction factor. Partial eta-squared (gp

2) and Cohen’s d are reported as measures of effect sizefor ANOVAs and t-tests, respectively. Confidence intervals for effectsizes were computed in SPSS as described by Smithson (2001).

2.7. Results, sequential presentation (Experiment 1)

Mean accuracy for the letter probe task was 96.4% (se 0.4%),indicating that the participants read the words attentively.

Fig. 1 shows ERPs elicited by the critical nouns when theyoccurred in semantically unrelated versus associated pairs (col-lapsed across the three levels of strength). After a series of positiveand negative peaks typical of visual ERPs, the waveforms show anegative peak around 400 ms that is smaller for the associatedwords. The semantic context effect is evident at all scalp sitesand lasts for several hundred milliseconds.

Fig. 1. Grand average ERPs from 30 participants, elicited by the second words of the pairs in Experiment 1 at all scalp sites. ‘‘All associated’’ collapses across strong, moderate,and weak associates. Negative plotted up here and in subsequent figures. Data low-pass filtered at 5 Hz for display. See Supplementary Material for waveforms filtered at30 Hz, low-pass.


The left side of Fig. 2 displays the ERPs elicited by strongly,moderately and weakly-associated words; strongly-associatedwords elicited smaller N400s than the weakly- or moderately-associated words (which appear to be much the same). The rightside of Fig. 2 compares the semantic context effects for the Strong,Moderate and Weak pairs, by subtracting the ERPs elicited by eachof the associated conditions from those elicited by their sharedUnrelated control condition. The difference waves are consistentwith the initial waveforms in showing a larger context effect inthe strongly-associated condition. Fig. 2 additionally suggests thatthe semantic context effect began earlier for the Strong pairs thanfor the Moderate and Weak pairs.

2.7.1. Strong, moderate and weak association: onset latenciesFor the jackknife procedure, the Strong, Moderate and Weak

context effects were isolated as difference waves from the sharedcontrol condition of Unrelated words, collapsed across the 29 scalpsites. The noise level of each difference wave was estimated as fluc-tuations in amplitude during the 200 ms prestimulus baseline. Thestandard deviations of prestimulus amplitudes were very similarfor the Strong, Moderate, and Weak difference waves: 0.44, 0.42,and 0.39 lV respectively. After applying the 2-SD criterion for

the onset of each effect, the jackknife procedure yielded estimatedonset times of 206 ms for the Strong association effect, and 302 msfor both the Moderate and Weak association effects (F(2,58) = 18.1,p < .001, e = 0.72, gp

2 = .38, 95% CI [.18, .52]). Onset latency in theStrong condition was earlier than both the Moderate (pairedt(29) = 5.3, p < .0001, d = 0.97, 95% CI [0.53,1.40]) and Weak condi-tions (t(29) = 4.13, p < .0005, d = 0.75, 95% CI [0.34,1.15]). For thepermutation procedure, amplitudes in the Strong, Moderate andWeak conditions were measured every 4 ms (with respect to theprestimulus baseline) in the ERPs for each subject and comparedto the Unrelated condition via paired t-tests. This procedureyielded onset latency estimates similar to the jackknife: 212 msfor the Strong condition, 328 ms for the Moderate condition and332 ms for the Weak condition. Fig. 3 displays the full timecourseof the significant results from the permutation tests.

2.7.2. Strong, moderate and weak association: amplitudesMean amplitudes from the 300–600 ms latency range were ana-

lyzed in order to compare the magnitudes of the three contexteffects after all three had begun (i.e., an amplitude measure thatis distinct from the onset latency measures). An ANOVA withStrength (3 levels) and scalp Site (29 levels) yielded a main effect

Fig. 2. Grand average ERPs from 30 participants, elicited by the second words ofassociated pairs in Experiment 1 at midline scalp sites from prefrontal (top) tooccipital (bottom). Left: Raw ERPs. Right: Difference waves formed by subtractingERPs elicited by associated words from those elicited by unrelated words.

Fig. 3. Results of the permutation t-tests comparing the weakly-, moderately- andstrongly-associated conditions to the Unrelated control condition at each 4 mssample point from stimulus onset to 1000 ms later. Solid lines indicate contiguoussignificant tests (sequences of at least 40 ms). Circles indicate a single significanttest.


of Strength (F(2,58) = 6.17, p < .005, e = 1.00, gp2 = .18, 90% CI

[.04, .30]). Given the broad spatial distribution of the effect,the interaction between Strength and Site was nonsignificant(F(56,1624) = 0.83). The main effect of association strength was dri-ven by a smaller N400 in the Strong condition; the Moderate andWeak conditions did not differ from each other (F(1,29) = 0.63).

It is important to note that, although strong associations weremore effective than weaker associations in reducing N400amplitude, all three types of related pairs elicited smaller N400sthan the Unrelated condition (Strong: F(1,29) = 18.5, p < .0005,gp

2 = .39, 90% CI [.15, .55]; Moderate: F(1,29) = 6.94, p < .02,gp

2 = .19, 90% CI [.20, .38]; Weak: F(1,29) = 11.1, p < .002, gp2 = .28,

90% CI [.07, .46]).

2.8. Results, simultaneous presentation (Experiment 2)

Mean accuracy for the letter probe task was 93.5% (se 0.5%),indicating that the participants read the words attentively.

Fig. 4 shows ERPs elicited by the semantically unrelated versusassociated pairs (collapsed across the three levels of strength).After a prominent negative peak at about 400 ms after stimulusonset, the waveforms at many scalp sites show a second nega-tive-going peak at roughly 750 ms. It is tempting to speculate thatthis is a ‘‘second N400’’ elicited after reading the second word ofthe pair. But it is equally clear that the impact of a semantic rela-tionship between the two words began well before this late second

negative peak, so that participants must have apprehended themeaning of both words fairly early during most of the trials.

The left side of Fig. 5 displays the ERPs elicited by strongly, mod-erately and weakly-associated pairs. Strongly-associated pairs elic-ited smaller N400s at posterior scalp sites, although the threeconditions appear fairly similar at anterior sites. The right side ofFig. 5 compares the semantic context effects for the Strong, Moder-ate and Weak pairs, by subtracting the ERPs elicited by each of theassociated conditions from those elicited by the Unrelated controlcondition. These difference waves are consistent with the initialwaveforms in showing a larger context effect in the strongly-associated condition, at some but not all scalp sites. In contrast toExperiment 1, Fig. 4 suggests that the semantic context effect beganat much the same time for the Strong, Moderate and Weak pairs.

2.8.1. Strong, moderate and weak association: onset latenciesFor the jackknife procedure, the Strong, Moderate and Weak

context effects were isolated as difference waves from the sharedcontrol condition of Unrelated words, collapsed across the 29 scalpsites. The noise level of each difference wave was estimated as fluc-tuations in amplitude during the 200 ms prestimulus baseline. Thestandard deviations of prestimulus amplitudes were very similarfor the Strong, Moderate, and Weak difference waves: 0.56, 0.55,and 0.51 lV respectively. After applying the 2 SD criterion for theonset of each effect, the jackknife procedure yielded estimatedonset times of 345 ms for the Strong association effect, 342 msfor the Moderate, and 346 ms for the Weak association effects.These were not significantly different (F(2,46) = 0.10). The permu-tation procedure yielded onset latency estimates that were verysimilar to the jackknife: 356 ms for the Strong condition, 344 msfor the Moderate condition and 356 ms for the Weak condition.

2.8.2. Strong, moderate and weak association: amplitudesMean amplitudes from the 350–650 ms latency range were ana-

lyzed in order to compare the magnitudes of the three contexteffects, after all three had begun. An ANOVA with Strength (3 lev-els) and scalp Site (29 levels) did not yield a main effect of Strength(F(2,46) = 0.37) but an interaction between Strength and Site(F(56,1288) = 2.27, p < .05, e = 0.11, gp

2 = .09, 90% CI [.01, .13]).Followup analyses showed that the interaction between

Strength and scalp site was driven by the Strong pairs. Comparisonof the Moderate and Weak pairs yielded neither a main effect ofStrength nor a Strength by Site interaction. In contrast, comparisonof Strong to Moderate pairs yielded a Strength by Site interaction(F(28,644) = 2.59, p < .05, e = 0.18, gp

2 = .10, 90% CI [.01, .16]), as

Fig. 4. Grand average ERPs from 24 participants, elicited by word pairs in Experiment 2 at all scalp sites. ‘‘All associated’’ collapses across strong, moderate, and weakassociates.


did comparison of the Strong to Weak pairs (F(28,644) = 3.27,p < .05, e = 0.10, gp

2 = .13, 90% CI [.01, .23]). We examined whichscalp sites had smaller N400s for the Strong pairs as compared tothe Moderate and Weak pairs (collapsed) via permutation testson amplitudes in the 350–650 ms latency range (30 K permuta-tions, alpha = .01, one-tailed). Individual sites showing significanteffects of association strength fell in a contiguous region of poster-ior midline and right scalp (Cpz, Cp4, Pz, P4, Oz, O2, T6) plus a pos-terior temporal site on the left (T5).

It is important to note that, although strong associations weremore effective than weaker associations in reducing N400amplitude, all three types of related pairs elicited smaller N400sthan the Unrelated condition (Strong: F(1,23) = 33.3, p < .0001,gp

2 = .59, 90% CI [.30, .66]; Moderate: F(1,23) = 19.3, p < .0005,gp

2 = .46, 90% CI [.16, .56]; Weak: F(1,29) = 13.8, p < .001, gp2 = .38,

90% CI [.10, .50]).

2.9. Cross-experiment comparisons: Impact of sequential versussimultaneous presentation

2.9.1. Onset latencies of the context effectsThe jackknife and permutation procedures indicated that onset

latencies of the context effects (differences between associated and

unrelated pairs) were roughly 200 ms for Strong pairs presentedsequentially, 300 ms for Moderate and Weak pairs presentedsequentially, and 350 ms for pairs of all strengths presented simul-taneously. The analyses above showed that strongly-related pairshad earlier onset latencies than weaker pairs only during sequen-tial presentation. We also performed a cross-experiment analysisto examine the overall delay in differentiating related from unre-lated pairs due to simultaneous presentation of the two words inone display. The estimates of onset latency derived from thejackknife procedure were submitted to an ANOVA with abetween-subject factor of SOA (sequential vs simultaneous) and awithin-subject factor of association strength (Strong, Moderate,Weak). This yielded a main effect of SOA (F(1,52) = 12.2, p < .001,gp

2 = .19, 90% CI [.05, .34]), a main effect of Strength(F(2,104) = 5.69, p < .005, e = 0.87, gp

2 = .10, 90% CI [.02, .20]), andan SOA by Strength interaction F(2,104) = 5.81, p < .005, e = 0.87,gp

2 = .10, 90% CI [.02, .20]). Given the single-experiment analysesabove, all three of these outcomes are much as expected. Followupanalyses showed that the substantially later differentiationbetween Strong and Unrelated pairs during simultaneous presen-tation (as opposed to sequential) presentation was robust(t(52) = 3.87, p < .0005, d = 1.06, 95% CI [0.48,1.63]). The smaller�50 ms delay in the onset of the Moderate and Weak context

Fig. 5. Grand average ERPs from 24 participants, elicited by associated pairs inExperiment 2 at midline scalp sites from prefrontal (top) to occipital (bottom). Left:raw ERPs. Right: difference waves formed by subtracting ERPs elicited by associatedpairs from those elicited by unrelated pairs.

3 The centroparietal maximum of the context effects in Exp. 2 (simultaneouspresentation) is considered more typical of N400 effects than the fairly flat scalpdistribution observed in Exp. 1, which instead showed a very gradual taper fromanterior to posterior (see reviews by Friederici, 2011; Kutas & Federmeier, 2011; Lauet al., 2008; Van Petten & Luka, 2006). However, descriptions of the ‘‘standard N400scalp distribution’’ are largely grounded in sentence-processing experiments; lesseffort has been devoted to descriptions of typical topographies for word pairs. In theonly previous comparison between sequential and simultaneous word pairs,Anderson and Holcomb (1995) observed a shift from a frontally-maximal effect forsequential presentation to a parietally-maximal context effect for simultaneouspresentation, similar to the results here. The functional interpretation of thesimultaneous/sequential topographic difference is not obvious, in particular whetherit should be linked to engagement in a more prospective versus more retrospectivemode of semantic processing. We do note two other experiments that invitedparticipants to engage in semantic prediction, with the result of anterior N400 effects.In both, subjects were cued on a trial-by-trial basis about the likely semantic categoryof an upcoming word, such that an ‘‘x’’ predicted an animal name with 80% validity,and a ‘‘+’’ predicted a tool name with 80% validity (Cristescu, Devlin, & Nobre, 2006;Kanske, Plitschka, & Kotz, 2011). Words from the invalidly cued category (a tool whenan animal was likely or vice versa) elicited more negative ERPs in the N400 latencyrange than words from the validly-cued category. The validity effects in theseexperiments are similar to the context effect induced by sequential word presentationhere – visible along the full anterior-to-posterior axis, but largest frontally (partic-ularly in Kanske et al., 2011). An understanding of the factors that lead to fairlyanterior versus centroparietal N400 effects, and the functional significance of thesedifferences awaits further research.


effects due to simultaneous presentation was also reliable (maineffect of SOA: F(1,52) = 4.48, p < .05, gp

2 = .08, 90% CI [.01, .21]; maineffect of Strength and Strength � SOA, Fs < 0.1).

2.9.2. Amplitudes and scalp distributions of the context effectsThe single-experiment analyses above indicated that associa-

tion strength influenced N400 amplitude during both sequentialand simultaneous presentation, although the strength effect duringsequential presentation was spatially widespread (no interactionwith scalp site) whereas the analogous effect during simultaneouspresentation was restricted to posterior scalp sites. A combinedanalysis defined the context effects as in the single-experimentanalyses: amplitudes of the differences between the relatedconditions and the Unrelated condition during the 300–600 msinterval for sequential presentation and the 350–650 ms intervalfor simultaneous presentation, with association strength and scalpsite as repeated measures and SOA as a between-subject factor.The results showed that association strength was equally effectiveat modulating N400 amplitude in the two experiments (maineffect of Strength, F(2,104) = 4.70, p < .02, e = 1.0, gp

2 = .08, 90% CI[.01, .17]; main effect of SOA and Strength � SOA interaction, ns).However, interactions between SOA and scalp location indicatethat sequential versus simultaneous presentation influenced theoverall topography of the semantic context effects (SOA � Site,F(28,1456) = 2.45, p < .05, e = 0.19, gp

2 = .05, 90% CI [.01, .07];Strength � SOA � Site, F(56,2912) = 1.96, p = .059, e = 0.13,gp

2 = .04, 90% CI [.00, .05]). Fig. 6 shows that both the overall differ-ence between related and unrelated pairs, as well as the difference

between strongly- and weakly-associated pairs was more posteriorfor simultaneous than sequential presentation.3

3. Discussion

3.1. Simultaneous versus sequential word presentation: reading twowords at once?

Simultaneous presentation of the two members of a word pairin Exp. 2 was designed to reduce the time available for anticipationor prediction or pre-activation of the second word or aspects of itsmeaning. One can wonder, however, about the degree of timecompression created by simultaneous presentation. Can simulta-neously presented pairs actually be processed simultaneously?The current results, as well as Anderson and Holcomb’s prior work,indicate that – unsurprisingly – two words cannot be processed asquickly as one. Anderson and Holcomb (1995) observed significantN400 differences between related and unrelated pairs in a200–300 ms latency window with sequential presentation, butnot until a 300–400 ms window with simultaneous presentation.We observed a roughly 50 ms delay in the onset of the semanticcontext effect for weakly-related pairs, and a delay approaching150 ms for strongly-related pairs. However, both studies also indi-cate substantial time compression for reading two-word displaysas compared to two single-word displays in that the latencies fortwo-word displays were much less than double those of the singleword displays. This underadditivity indicates a high degree oftemporal overlap in the processing of two words when they arepresented simultaneously.

The two-word displays used here were, of course, very differentfrom typical reading in which words are displayed in a horizontalseries. As a first approximation, natural reading along a horizontalline of text is an inherently serial activity in which words are fix-ated one at a time. However, extensive research on eye movementsduring reading also shows substantial deviation from purely serialprocessing. A fairly large proportion of words in text are not fix-ated, and word-skipping is more prevalent for words that are highin frequency of usage and/or predictable from the prior context,indicating that readers extract some information about wordN + 1 in a line of text while their eyes are focused on word N. Otherwork indicates that readers slow down when they are unable tobenefit from a parafoveal preview of the word just to the right ofcurrent fixation (see Brysbaert, Drieghe, & Vitu, 2005; Schotter,

Fig. 6. Top: spline-interpolated topographic maps of the difference between semantically-associated and unrelated conditions during sequential (mean amplitude 300–600 ms) and simultaneous (mean amplitude 350–650 ms) presentation of the word pairs. Bottom: amplitudes of the semantic context effects for strongly-, moderately- andweakly-associated pairs (Unrelated minus associated), measured at scalp sites from anterior to posterior. Fp is the mean of prefrontal sites Fpz, Fp1, Fp2; F is the mean offrontal sites Fz, F3, F4; Fc is the mean of frontocentral sites Fcz, Fc3, Fc4; C is the mean of central sites Cz, C3, C4; Cp is the mean of centroparietal sites Cpz, Cp3, Cp4; P is themean of parietal sites Pz, P3, P4; O is the mean of occipital sites Oz, O1, O2. Error bars are standard error of the mean.


Angele, & Rayner, 2012 for reviews). Some division of cognitiveresources between a currently fixated and parafoveal word is clearfrom the finding that the magnitude of the preview effect dependson the word frequency of the currently fixated word (Henderson &Ferreira, 1990; Schroyens, Vitu, Brysbaert, & d’Ydewalle, 1999). Thedegree to which semantic – as opposed to orthographic – informa-tion is extracted from parafoveal words has been a controversialtopic, with some evidence favoring semantic processing whenthe words are short (i.e., do not extend very far into the region ofthe retina with reduced acuity) and semantic constraints are strong(Barber, van der Meij, & Kutas, 2013; Schotter et al., 2012). Othershave argued that strong semantic constraints lead to an expecta-tion for the orthographic form of an upcoming word, such that par-afoveal preview effects on eye movements reflect the perception oforthographic information to the right of fixation (Kretzschmar,Bornkessel-Schlesewsky, & Schlesewsky, 2009).

Regardless of its orthographic versus semantic nature, the par-afoveal preview effect indicates that even fairly natural readingincludes some degree of parallel processing, or temporal overlapin the processing of adjacent words. Here, the simultaneous worddisplays were configured to encourage such processing, in thatthe vertical arrangement of the two-word displays allowed bothwords to fall largely or completely within the fovea. A different lineof evidence indicates that temporal proximity also creates tempo-ral overlap in the processing of two words that is lost with greaterseparation. Lexical decisions to Word 1 of a pair are speeded when

a related word is presented simultaneously, or up to some 300 mslater, but not when presentation of the second word is delayed fora longer time (Kiger & Glass, 1983; Logan & Schulkind, 2000).

In the current study, we cannot rule out the possibility that par-ticipants shifted their attention from top to bottom in the simulta-neous displays such that processing the cue (top) word had sometemporal priority over the target (bottom) word. Processing thecue before the target is, of course, the precondition for any sortof prospective processing of a predictable target. However, the�50 ms disparity in onset times for the Weak/Moderate contexteffects for simultaneous versus sequential presentation suggestsan upper limit for the amount of processing that could have beendevoted solely to the cue word before initiating processing of thetarget word in simultaneous displays. This small span of time sug-gests that any residue of serial processing in the two-word displayswas modest and that, instead, there was substantial temporal over-lap in the processing of the two simultaneously-presented words.That temporal overlap means, in turn, that simultaneous presenta-tion led to a very substantial reduction of the necessary resourcefor any sort of pre-activation of the second word of a pair – namelyan interval in which the cue has been understood but the target hasnot yet been viewed. In contrast, the 700 ms gap between wordsduring sequential presentation should have allowed some ‘‘slacktime’’ between the two words, during which readers could pre-activate a full word or some semantic features that may or maynot actually occur in the target.


3.2. Association strength and ERP latencies

The manipulation of association strength from stronger toweaker to null was designed to influence the likelihood that anyanticipatory or predictive activity prior to the presentation of Word2 would prove helpful for processing it. During both sequential andsimultaneous presentation of the pairs, all of the associated wordselicited smaller N400s than semantically unrelated words, but thisdifference was larger in the strongly-associated condition, as inother studies (Frishkoff, 2007; Kandhadai & Federmeier, 2010a,2010b; Kutas & Hillyard, 1989; Luka & Van Petten, in press;Roehm et al., 2007; Rösler et al., 2001). Of greater interest hereis that the semantic context effect for the strongly-associated pairsbut also began earlier than those for moderately and weakly-associated pairs, but only with sequential presentation.

The onset latency difference observed here is a somewhat unu-sual result for ERP studies of semantic processing. In contrast toERP components such as the P300 or Lateralized Readiness Poten-tial that show fairly dramatic latency differences across stimuli andtasks (Kutas, McCarthy, & Donchin, 1977; Osman et al., 1993),N400 effects are often described as changes in the amplitude of apotential with stable latencies across conditions (Federmeier &Laszlo, 2009; Kutas & Federmeier, 2011). There are, however, someprior observations of earlier N400 effects for stronger contexts ascompared to weaker. In an experiment with related and unrelatedword pairs, Federmeier et al. (2010) found the peak latency of therelated/unrelated difference to be about 40 ms earlier for antonympairs (association strength 41%) as compared to pairs consisting ofcategory names and exemplars (association strength 8%). Anadvantage for antonyms in both onset and peak latency wasobserved in a prior experiment using similar materials (Kutas &Iragui, 1998). At least one other experiment with word pairs vary-ing in association strength appears to show an earlier divergence ofstrongly-associated targets from unrelated, although latency wasnot analyzed (Kandhadai & Federmeier, 2010a, Fig. 4). Latency dif-ferences among N400 context effects are likely to be underrepre-sented in the literature because they can only be characterized inexperimental designs with at least two levels of relationshipstrength plus an unrelated condition, a task that avoids overlapwith decision-related P300 components, and are likely to be clear-est when lexical characteristics (length, frequency, orthographicneighborhood) of the eliciting words are very closely matchedamong conditions – the conditions implemented here.

The earlier onset of the context effect after stronger associates isvery consistent with what one might expect of predictive process-ing – that the strong associates led to correct anticipation of thecritical nouns during a higher proportion of the trials than theweaker associates. The earlier onset of the semantic context effectshows the benefit of correct prediction, namely a more rapid abilityto distinguish the predicted word from some unrelated item. Whenthe word pairs were presented simultaneously in Exp. 2, strongly-associated pairs continued to elicit smaller N400s than moreweakly related pairs, but onset latencies of the effects were equiv-alent. This pattern of results is what one would expect from a ret-rospective appreciation of the relationship between the words (orfrom co-processing the two words), in the absence of a temporal‘‘head start’’ from successful prediction of the upcoming word.

The current set of materials was not designed to reveal whetherthe anticipatory processing in Exp. 1 consisted of predicting a spe-cific word form, or more broadly expecting semantic features thatcould be instantiated by multiple words. This latter question mightbe addressed by comparing equally low-strength associates thatare semantically similar versus dissimilar to a strong (i.e., predict-able) associate. It is also important to note that the use of wordpairs with established semantic relationships places limits on theapplicability of the results to the understanding of novel meaning,

for which retrospective appreciation of conceptual relationships islikely to be most prevalent. The current contrast between sequen-tial and simultaneous pairs is largely capable of separating antici-patory processing from non-prospective varieties of semanticcontext effects, but is less well-suited to distinguishing truly pos-thoc semantic synthesis (a third stage after two terms have beenfully processed as independent concepts) from parallel processingof two items during which a pre-existing relationship is detected.

The comparison of sequential to simultaneous presentationhere has the advantage of directly manipulating the dimensionthat distinguishes prospective from non-prospective processing,namely time. A different experimental approach is to use a con-stant long SOA, but indirectly encourage or discourage participantsfrom predictive processing by altering the ratio of related to unre-lated word pairs. The logic here is that a high relatedness-proportion(RP) can lead to high rates of success in anticipating the secondword of a pair (if all of the pairs are strongly associated), while alow RP will stimulate participants to abandon such a strategybecause their predictions will usually be wrong. Several ERP stud-ies have examined the impact of relatedness proportion on N400differences between related and unrelated word pairs, with thegeneral result of larger effects in the high-RP than low-RP condi-tion (Brown, Hagoort, & Chwilla, 2000; Holcomb, 1988; Küper &Heil, 2010; Silva-Pereyra et al., 1999). More recently, Lau,Holcomb, and Kuperberg (2013) have repeated this manipulationwith a careful analysis of onset latencies. When 50% of the wordpairs were strongly associated, ERPs to related and unrelatedwords diverged at around 200 ms after the onset of the secondword, but the context effect did not begin until about 350 ms whenthese strong pairs comprised only 10% of the total stimulus list. Theonset latency difference observed by Lau et al. is very compatiblewith that observed for the strong versus weak associates in Exp.1 (roughly 200 versus 300 ms), yielding converging evidence thatsuccessful prediction influences the speed of word processing. Amajor difference in the outcome of the RP paradigm and thesequential/simultaneous comparison here is that the amplitudesof the context effects were not diminished in the simultaneouscondition as they are in low-RP conditions. It seems likely thatsimultaneous presentation, in and of itself, provides strong encour-agement to attempt semantic integration, leading to smaller N400swhen successful.

Published discussions about the roles of predictive processingversus integration in driving N400 amplitude have sometimes pre-sented these as dichotomous alternatives. The current resultsinstead argue that both prospective and retrospective semanticprocessing are reflected in N400 amplitudes (although successfulprediction yields earlier responses), so that the correct answer is‘‘both’’.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.bandl.2014.06.001.

References

Anaki, D., & Henik, A. (2003). Is there a ‘‘strength effect’’ in automatic semanticpriming? Memory and Cognition, 31, 262–272. http://dx.doi.org/10.3758/BF03194385.

Anderson, J. E., & Holcomb, P. J. (1995). Auditory and visual semantic priming usingdifferent stimulus onset asynchronies: An event-related brain potential study.Psychophysiology, 32, 177–190. http://dx.doi.org/10.1111/j.1469-8986.1995.tb03310.x.

Arzouan, Y., Goldstein, A., & Faust, M. (2007). Brainwaves are stethoscopes: ERPcorrelates of novel metaphor comprehension. Brain Research, 1160, 69–81.http://dx.doi.org/10.1016/j.brainres.2007.05.034.



http://dx.doi.org/10.3758/BF03194385

http://dx.doi.org/10.3758/BF03194385

http://dx.doi.org/10.1111/j.1469-8986.1995.tb03310.x


http://dx.doi.org/10.1016/j.brainres.2007.05.034


Backer, K. C., Hill, K. T., Shahin, A. J., & Miller, L. M. (2010). Neural time course ofecho suppression in humans. Journal of Neuroscience, 30, 1905–1913. http://dx.doi.org/10.1523/JNEUROSCI.4391-09.2010.

Balota, D. A., Yap, M. J., Cortese, M. J., Hutchison, K. A., Kessler, B., Loftis, B., et al.(2007). The English Lexicon project. Behavior Research Methods, 39, 445–459.http://dx.doi.org/10.3758/BF03193014. http://elexicon.wust.edu.

Barber, H. A., van der Meij, M., & Kutas, M. (2013). An electrophysiological analysisof contextual and temporal constraints on parafoveal word processing.Psychophysiology, 50, 48–59. http://dx.doi.org/10.1111/j.1469-8986.2012.01489.x.

Becker, C. A. (1980). Semantic context effects in visual word recognition: Ananalysis of semantic strategies. Memory and Cognition, 8, 493–512. http://dx.doi.org/10.3758/BF03213769.

Blair, R. C., & Karniski, W. (1993). An alternative method for significance testing ofwaveform difference potentials. Psychophysiology, 30, 518–524. http://dx.doi.org/10.1111/j.1469-8986.1993.tb02075.x.

Brown, C. M., Hagoort, P., & Chwilla, D. J. (2000). An event-related brain potentialanalysis of visual word priming effects. Brain and Language, 72, 158–190. http://dx.doi.org/10.1006/brln.1999.2284.

Brysbaert, M., Drieghe, D., & Vitu, F. (2005). Word skipping: Implications fortheories of eye movement control in reading. In G. Underwood (Ed.), Cognitiveprocesses in eye guidance (pp. 53–77). Oxford: Oxford University Press. http://dx.doi.org/10.1093/acprof:oso/9780198566816.003.0003.

Camargo, A., Azuaje, F., Wang, H., & Zheng, H. (2008). Permutation-based statisticaltests for multiple hypotheses. Source Code for Biology and Medicine, 3, 15. http://dx.doi.org/10.1186/1751-0473-3-15.

Chwilla, D. J., Brown, C. M., & Hagoort, P. (1995). The N400 as a function of the levelof processing. Psychophysiology, 32, 274–285. http://dx.doi.org/10.1111/j.1469-8986.1995.tb02956.x.

Chwilla, D. J., Kolk, H. H. J., & Vissers, C. T. W. M. (2007). Immediate integration ofnovel meanings: N400 support for an embodied view of languagecomprehension. Brain Research, 1183, 109–123. http://dx.doi.org/10.1016/j.brainres.2007.09.014.

Collins, A. M., & Loftus, E. F. (1975). A spreading activation theory of semanticprocessing. Psychological Review, 82, 407–428. http://dx.doi.org/10.1037/0033-295X.82.6.407.

Coulson, S. (2006). Constructing meaning. Metaphor and Symbol, 21, 245–266.http://dx.doi.org/10.1207/s15327868ms2104_3.

Coulson, S., Federmeier, K., Van Petten, C., & Kutas, M. (2005). Right hemispheresensitivity to word and sentence level context: Evidence from event-relatedbrain potentials. Journal of Experimental Psychology: Learning, Memory, andCognition, 31, 129–147. http://dx.doi.org/10.1037/0278-7393.31.1.129.

Coulson, S., & Van Petten, C. (2002). Conceptual integration and metaphor: Anevent-related potential study. Memory and Cognition, 30, 958–968. http://dx.doi.org/10.3758/BF03195780.

Coulson, S., & Van Petten, C. (2007). A special role for the right hemisphere inmetaphor comprehension? ERP evidence from hemifield presentation.Brain Research, 1146, 128–145. http://dx.doi.org/10.1016/j.brainres.2007.03.008.

Cristescu, T., Devlin, J. T., & Nobre, A. C. (2006). Orienting attention to semanticcategories. NeuroImage, 33(4), 1178–1187.

Davenport, T., & Coulson, S. (2011). Predictability and novelty in literal languagecomprehension: An ERP study. Brain Research, 1418, 70–82. http://dx.doi.org/10.1016/j.brainres.2011.07.039.

Davenport, T., & Coulson, S. (2013). Hemispheric asymmetry in interpreting novelliteral language: An event-related potential study. Neuropsychologia, 51,907–921. http://dx.doi.org/10.1016/j.neuropsychologia.2013.01.018.

De Groot, A. M. B. (1983). The range of automatic spreading activation in wordpriming. Journal of Verbal Learning and Verbal Behavior, 22, 417–436. http://dx.doi.org/10.1016/S0022-5371(83)90273-6.

DeLong, K. A., Groppe, D. M., Urbach, T. P., & Kutas, M. (2012). Thinking ahead ornot? Natural aging and anticipation during reading. Brain and Language, 121,226–239. http://dx.doi.org/10.1016/j.bandl.2012.02.006.

DeLong, K. A., Urbach, T. P., & Kutas, M. (2005). Probabilistic word preactivationduring language comprehension inferred from electrical brain activity. NatureNeuroscience, 8, 1117–1121. http://dx.doi.org/10.1038/nn1504.

Dimigen, O., Sommer, W., Hohlfield, A., Jacobs, A. M., & Kliegl, R. (2011).Coregregistration of eye movements and EEG in natural reading: Analysesand review. Journal of Experimental Psychology: General, 140, 552–572. http://dx.doi.org/10.1037/a0023885.

Doi, H., & Shinohara, K. (2012). Event-related potentials elicited in mothers by theirown and unfamiliar infants’ faces with crying and smiling expression.Neuropsychologia, 50, 1297–1307. http://dx.doi.org/10.1016/j.neuropsychologia.2012.02.013.

Dwass, M. (1957). Modified randomization tests for nonparametric hypotheses.Annals of Mathematical Statistics, 28, 181–187. http://dx.doi.org/10.1214/aoms/1177707045.

Edgington, E. S. (1969). Approximate randomization tests. Journal of Psychology, 72,143–149. http://dx.doi.org/10.1080/00223980.1969.10543491.

Eimer, M., & Grubert, A. (2014). Spatial attention can be allocated rapidly and inparallel to new visual objects. Current Biology, 24, 193–198. http://dx.doi.org/10.1016/j.cub.2013.12.001.

Federmeier, K. D. (2007). Thinking ahead: The role and roots of prediction inlanguage comprehension. Psychophysiology, 44, 491–505. http://dx.doi.org/10.1111/j.1469-8986.2007.00531.x.

Federmeier, K. D., Kutas, M., & Schul, R. (2010). Age-related and individualdifferences in the use of prediction during language comprehension. Brain andLanguage, 115, 149–161. http://dx.doi.org/10.1016/j.bandl.2010.07.006.

Federmeier, K. D, & Laszlo, S. (2009). Time for meaning: Electrophysiology providesinsights into the dynamics of representation and processing in semanticmemory. In B. H. Ross (Ed.), The psychology of learning and motivation (pp. 1–44).San Diego, CA: Elsevier. http://dx.doi.org/10.1016/s0079-7421(09)51001-8.

Fischler, I. (1977). Semantic facilitation without association in a lexical decisiontask. Memory and Cognition, 5, 335–339. http://dx.doi.org/10.3758/BF03197580.

Fischler, I., & Goodman, G. O. (1978). Latency of associative activation in memory.Journal of Experimental Psychology: Human Perception and Performance, 4,455–470. http://dx.doi.org/10.1037/0096-1523.4.3.455.

Forster, K. I. (1981). Priming and the effects of sentence and lexical contexts onnaming time: Evidence for autonomous lexical processing. Quarterly Journal ofExperimental Psychology, 33A, 465–495.

Friederici, A. D. (2011). The brain basis of language processing: From structure tofunction. Physiological Reviews, 91, 1357–1392. http://dx.doi.org/10.1152/physrev.00006.2011.

Frishkoff, G. A. (2007). Hemispheric differences in strong versus weak semanticpriming: Evidence from event-related brain potentials. Brain and Language, 100,23–43. http://dx.doi.org/10.1016/j.bandl.2006.06.117.

Guthrie, D., & Buchwald, J. S. (1991). Significance testing of difference potentials.Psychophysiology, 28, 240–244. http://dx.doi.org/10.1111/j.1469-8986.1991.tb00417.x.

Grubert, A., & Eimer, M. (2013). Qualitative differences in the guidance of attentionduring single-color and multiple-color visual search: Behavioral andelectrophysiological evidence. Journal of Experimental Psychology: HumanPerception and Performance, 39, 1433–1442. http://dx.doi.org/10.1037/a0031046.

Hare, M., Jones, M., Thomson, C., Kelly, S., & McRae, K. (2009). Activating eventknowledge. Cognition, 111, 151–167. http://dx.doi.org/10.1016/j.cognition.2009.01.009.

Henderson, J. M., & Ferreira, F. (1990). Effects of foveal processing difficulty on theperceptual span in reading: Implications for attention and eye movementcontrol. Journal of Experimental Psychology: Learning, Memory and Cognition, 16,417–429. http://dx.doi.org/10.1037/0278-7393.16.3.417.

Hillyard, S. A., Squires, K. C., Bauer, J. W., & Lindsay, P. H. (1971). Evoked potentialcorrelates of auditory signal detection. Science, 172, 1357–1360. http://dx.doi.org/10.1126/science.172.3990.1357.

Hodgson, J. M. (1991). Informational constraints on prelexical priming. Languageand Cognitive Processes, 6, 169–205. http://dx.doi.org/10.1080/01690969108406942.

Holcomb, P. J. (1988). Automatic and attentional processing: An event-related brainpotential analysis of semantic priming. Brain and Language, 35, 66–85. http://dx.doi.org/10.1016/0093-934X(88)90101-0.

Holcomb, P. J. (1993). Semantic priming and stimulus degradation: Implications forthe role of the N400 in language processing. Psychophysiology, 30, 47–61. http://dx.doi.org/10.1111/j.1469-8986.1993.tb03204.x.

Jacobson, J. Z. (1973). Effects of association upon masking and reading latency.Canadian Journal of Psychology, 27, 58–69. http://dx.doi.org/10.1037/h0082455.

Jöckel, K.-H. (1986). Finite sample properties and asymptotic efficiency of MonteCarlo tests. Annals of Statistics, 14, 336–347. http://dx.doi.org/10.1214/aos/1176349860.

Kamide, Y., Altmann, G. T. M., & Haywood, S. L. (2003). The timecourse of predictionin incremental sentence processing: Evidence from anticipatory eyemovements. Journal of Memory and Language, 49, 133–156. http://dx.doi.org/10.1016/S0749-596X(03)00023-8.

Kandhadai, P., & Federmeier, K. D. (2010a). Hemispheric differences in therecruitment of semantic processing mechanisms. Neuropsychologia, 48,3772–3781. http://dx.doi.org/10.1016/j.neuropsychologia.2010.07.018.

Kandhadai, P., & Federmeier, K. D. (2010b). Automatic and controlled aspects oflexical associative processing in the two cerebral hemispheres.Psychophysiology, 47, 774–785. http://dx.doi.org/10.1111/j.1469-8986.2009.00969.x.

Kanske, P., Plitschka, J., & Kotz, S. A. (2011). Attentional orienting towards emotion:P2 and N400 ERP effects. Neuropsychologia, 49, 3121–3129. http://dx.doi.org/10.1016/j.neuropsychologia.2011.07.022.

Kiger, J. I., & Glass, A. L. (1983). The facilitation of lexical decisions by a primeoccurring after the target. Memory and Cognition, 11, 356–365. http://dx.doi.org/10.3758/bf03202450.

Kiss, G. R, Armstrong, C., Milroy, R., & Piper, J. (1973). An associative thesaurus ofEnglish and its computer analysis. In A. J. Aitken, R. W. Bailey, & N. Hamilton-Smith (Eds.), The computer and literary studies (pp. 271–286). Edinburgh,Scotland: Edinburgh University Press. <http://www.eat.rl.ac.uk/>.

Koriat, A. (1981). Semantic facilitation in lexical decision as a function of prime-target association. Memory and Cognition, 9, 587–598. http://dx.doi.org/10.3758/BF03202353.

Kretzschmar, F., Bornkessel-Schlesewsky, I., & Schlesewsky, M. (2009). Parafovealversus foveal N400s dissociate spreading activation from contextual fit.NeuroReport, 20, 1613–1618. http://dx.doi.org/10.1097/WNR.0b013e328332c4f4.

Kroll, J. F., & Potter, M. C. (1984). Recognizing words, pictures, and concepts: Acomparison of lexical, object, and reality decisions. Journal of Verbal Learningand Verbal Behavior, 23, 39–66. http://dx.doi.org/10.1016/S0022-5371(84)90499-7.

http://dx.doi.org/10.1523/JNEUROSCI.4391-09.2010

http://dx.doi.org/10.1523/JNEUROSCI.4391-09.2010

http://dx.doi.org/10.3758/BF03193014.http://elexicon.wust.edu

http://dx.doi.org/10.1111/j.1469-8986.2012.01489.x

http://dx.doi.org/10.1111/j.1469-8986.2012.01489.x

http://dx.doi.org/10.3758/BF03213769

http://dx.doi.org/10.3758/BF03213769



http://dx.doi.org/10.1006/brln.1999.2284

http://dx.doi.org/10.1006/brln.1999.2284

http://dx.doi.org/10.1093/acprof:oso/9780198566816.003.0003

http://dx.doi.org/10.1093/acprof:oso/9780198566816.003.0003

http://dx.doi.org/10.1186/1751-0473-3-15

http://dx.doi.org/10.1186/1751-0473-3-15





http://dx.doi.org/10.1037/0033-295X.82.6.407

http://dx.doi.org/10.1037/0033-295X.82.6.407

http://dx.doi.org/10.1207/s15327868ms2104_3

http://dx.doi.org/10.1037/0278-7393.31.1.129

http://dx.doi.org/10.3758/BF03195780

http://dx.doi.org/10.3758/BF03195780



http://refhub.elsevier.com/S0093-934X(14)00089-3/h9005




http://dx.doi.org/10.1016/j.neuropsychologia.2013.01.018

http://dx.doi.org/10.1016/S0022-5371(83)90273-6

http://dx.doi.org/10.1016/S0022-5371(83)90273-6


http://dx.doi.org/10.1038/nn1504

http://dx.doi.org/10.1037/a0023885

http://dx.doi.org/10.1037/a0023885



http://dx.doi.org/10.1214/aoms/1177707045

http://dx.doi.org/10.1214/aoms/1177707045

http://dx.doi.org/10.1080/00223980.1969.10543491

http://dx.doi.org/10.1016/j.cub.2013.12.001

http://dx.doi.org/10.1016/j.cub.2013.12.001

http://dx.doi.org/10.1111/j.1469-8986.2007.00531.x

http://dx.doi.org/10.1111/j.1469-8986.2007.00531.x


http://dx.doi.org/10.1016/s0079-7421(09)51001-8

http://dx.doi.org/10.3758/BF03197580

http://dx.doi.org/10.1037/0096-1523.4.3.455




http://dx.doi.org/10.1152/physrev.00006.2011

http://dx.doi.org/10.1152/physrev.00006.2011




http://dx.doi.org/10.1037/a0031046

http://dx.doi.org/10.1037/a0031046

http://dx.doi.org/10.1016/j.cognition.2009.01.009


http://dx.doi.org/10.1037/0278-7393.16.3.417

http://dx.doi.org/10.1126/science.172.3990.1357


http://dx.doi.org/10.1080/01690969108406942

http://dx.doi.org/10.1080/01690969108406942

http://dx.doi.org/10.1016/0093-934X(88)90101-0

http://dx.doi.org/10.1016/0093-934X(88)90101-0



http://dx.doi.org/10.1037/h0082455

http://dx.doi.org/10.1214/aos/1176349860

http://dx.doi.org/10.1214/aos/1176349860

http://dx.doi.org/10.1016/S0749-596X(03)00023-8

http://dx.doi.org/10.1016/S0749-596X(03)00023-8


http://dx.doi.org/10.1111/j.1469-8986.2009.00969.x

http://dx.doi.org/10.1111/j.1469-8986.2009.00969.x



http://dx.doi.org/10.3758/bf03202450

http://dx.doi.org/10.3758/bf03202450

http://www.eat.rl.ac.uk/

http://dx.doi.org/10.3758/BF03202353

http://dx.doi.org/10.3758/BF03202353

http://dx.doi.org/10.1097/WNR.0b013e328332c4f4

http://dx.doi.org/10.1097/WNR.0b013e328332c4f4

http://dx.doi.org/10.1016/S0022-5371(84)90499-7

http://dx.doi.org/10.1016/S0022-5371(84)90499-7


Kucera, H., & Francis, W. (1967). Computational analysis of present day AmericanEnglish. Providence, RI: Brown University Press.

Kuefner, D., Jacques, C., Prieto, E. A., & Rossion, B. (2010). Electrophysiologicalcorrelates of the composite face illusion: Disentangling perceptual anddecisional components of holistic face processing in the human brain. Brainand Cognition, 74, 225–238. http://dx.doi.org/10.1016/j.bandc.2010.08.001.

Küper, K., & Heil, M. (2010). Letter search and relatedness proportion: Furtherelectrophysiological evidence for the automaticity of semantic activation.Neuroscience Letters, 482, 26–30. http://dx.doi.org/10.1016/j.neulet.2010.06.084.

Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: Finding meaning inthe N400 component of the event-related brain potential (ERP). Annual Reviewof Psychology, 62, 621–647.

Kutas, M., & Hillyard, S. A. (1984). Brain potentials during reading reflect wordexpectancy and semantic association. Nature, 307, 161–163. http://dx.doi.org/10.1038/307161a0.

Kutas, M., & Hillyard, S. A. (1989). An electrophysiological probe of incidentalsemantic association. Journal of Cognitive Neuroscience, 1, 38–49. http://dx.doi.org/10.1162/jocn.1989.1.1.38.

Kutas, M., & Iragui, V. (1998). The N400 in a semantic categorization task across 6decades. Electroencephalography and Clinical Neurophysiology, 108, 456–471.http://dx.doi.org/10.1016/S0168-5597(98)00023-9.

Kutas, M., Lindamood, T. E., & Hillyard, S. A. (1984). Word expectancy and event-related brain potentials during sentence processing. In S. Kornblum & J. Requin(Eds.), Preparatory states and processes (pp. 217–237). Hillsdale, NJ: Erlbaum.

Kutas, M., McCarthy, G., & Donchin, E. (1977). Augmenting mental chronometry:The P300 as a measure of stimulus evaluation time. Science, 197, 792–795.http://dx.doi.org/10.1126/science.887923.

Kutas, M., Van Petten, C., & Kluender, R. (2006). Psycholinguistics electrified II:1995–2005. In M. Traxler & M. A. Gernsbacher (Eds.), Handbook ofpsycholinguistics (2nd ed., pp. 659–724). New York: Elsevier. http://dx.doi.org/10.1016/B978-012369374-7/50018-3.

Laszlo, S., & Federmeier, K. D. (2009). A beautiful day in the neighborhood: Anevent-related potential study of lexical relationships and prediction in context.Journal of Memory and Language, 61, 326–338. http://dx.doi.org/10.1016/j.jml.2009.06.004.

Laszlo, S., & Federmeier, K. D. (2011). The N400 as a snapshot of interactiveprocessing: Evidence from regression analyses of orthographic neighbor andlexical associate effects. Psychophysiology, 48, 176–186. http://dx.doi.org/10.1111/j.1469-8986.2010.01058.x.

Laszlo, S., & Federmeier, K. D. (2014). Never seem to find the time: Evaluating thephysiological time course of visual word recognition with regression analysis ofsingle item ERPs. Language and Cognitive Processes, 29, 642–661. http://dx.doi.org/10.1080/01690965.2013.866259.

Lau, E. F., Holcomb, P. J., & Kuperberg, G. R. (2013). Dissociating N400 effects ofprediction from association in single-word contexts. Journal of CognitiveNeuroscience, 25, 484–502. http://dx.doi.org/10.1162/jocn_a_00328.

Lau, E., Phillips, C., & Poeppel, D. (2008). A cortical network for semantics:(de)Constructing the N400. Nature Reviews Neuroscience, 9, 920–933. http://dx.doi.org/10.1038/nrn2532.

Logan, G. D., & Schulkind, M. D. (2000). Parallel memory retrieval in dual-tasksituations: I. Semantic memory. Journal of Experimental Psychology: HumanPerception and Performance, 26, 1072–1090. http://dx.doi.org/10.1037/0096-1523.26.3.1072.

Luka, B. J., & Van Petten, C. (in press). Gradients versus dichotomies: How strengthof semantic context influences event-related potentials and lexical decisiontimes. Cognitive, Affective, and Behavioral Neuroscience. <http://dx.doi.org/10.3758/s13415-013-0223-1>.

Masson, M. E. J. (1995). A distributed memory model of semantic priming. Journal ofExperimental Psychology: Learning, Memory, & Cognition, 21, 3–23. http://dx.doi.org/10.1037/0278-7393.21.1.3.

McRae, K., de Sa, V., & Seidenberg, M. S. (1997). On the nature and scope of featuralrepresentations of word meaning. Journal of Experimental Psychology: General,126, 99–130. http://dx.doi.org/10.1037/0096-3445.126.2.99.

Meyer, D. E., & Schvaneveldt, R. W. (1971). Facilitation in recognizing pairsof words: Evidence of a dependence between retrieval operations. Journalof Experimental Psychology, 90, 227–234. http://dx.doi.org/10.1037/h0031564.

Miller, G. A., & Isard, S. (1963). Some perceptual consequences of linguistic rules.Journal of Verbal Learning and Verbal Behavior, 2, 217–228. http://dx.doi.org/10.1016/S0022-5371(63)80087-0.

Miller, J., Patterson, T., & Ulrich, R. (1998). Jacknife-based method for measuring LRPonset latency differences. Psychophysiology, 35, 99–115. http://dx.doi.org/10.1111/1469-8986.3510099.

Moss, H. E., Ostrin, R. K., Tyler, L. K., & Marslen-Wilson, W. D. (1995). Accessingdifferent types of lexical semantic information: Evidence from priming. Journalof Experimental Psychology: Learning, Memory, and Cognition, 21, 864–883.http://dx.doi.org/10.1037/0278-7393.21.4.863.

Nation, K., & Snowling, M. J. (1999). Developmental differences in sensitivity tosemantic relations among good and poor comprehenders: Evidence fromsemantic priming. Cognition, 70, B1–B13. http://dx.doi.org/10.1016/S0010-0277(99)00004-9.

Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles ofinhibitionless spreading activation and limited-capacity attention. Journal ofExperimental Psychology: General, 106, 226–254. http://dx.doi.org/10.1037/0096-3445.106.3.226.

Neely, J. H., Keefe, D. E., & Ross, K. L. (1989). Semantic priming in the lexical decisiontask: Roles of prospective prime-generated expectancies and retrospectivesemantic matching. Journal of Experimental Psychology: Learning, Memory, andCognition, 15, 1003–1019. http://dx.doi.org/10.1037/0278-7393.15.6.1003.

Nieuwland, M. S., & van Berkum, J. J. A. (2006). When peanuts fall in love: N400evidence for the power of discourse. Journal of Cognitive Neuroscience, 18,1098–1111. http://dx.doi.org/10.1162/jocn.2006.18.7.1098.

Osman, A., Bashore, T. R., Coles, M. G. H., Donchin, E., & Meyer, D. E. (1993). On thetransmission of partial information: Inferences from movement-related brainpotentials. Journal of Experimental Psychology: Human Perception andPerformance, 18, 217–232. http://dx.doi.org/10.1037/0096-1523.18.1.217.

Osterhout, L., & Holcomb, P. J. (1995). Event-related potentials and languagecomprehension. In M. D. Rugg & M. G. H. Coles (Eds.), Electrophysiology of mind:Event-related brain potentials and cognition. Oxford: Oxford University Press.

Paul, D. D., & Sutton, S. (1972). Evoked potential correlates of response criterion inauditory signal detection. Science, 177, 362–364. http://dx.doi.org/10.1126/science.177.4046.362.

Pickering, M., & Garrod, S. (2007). Do people use language production to makepredictions during comprehension? Trends in Cognitive Science, 11, 105–110.http://dx.doi.org/10.1016/j.tics.2006.12.002.

Plaut, D. C. (1995). Semantic and associative priming in a distributed attractornetwork. In J. F. Lehman & J. D. Moore (Eds.), Proceedings of the 17th annualconference of the cognitive science society (pp. 37–42). Hillsdale, NJ: Erlbaum.

Ratcliff, R., & McKoon, G. (1988). A retrieval theory of priming in memory.Psychological Review, 95, 385–408. http://dx.doi.org/10.1037/0033-295X.95.3.385.

Ratcliff, R., & McKoon, G. (1995). Sequential effects in lexical decision: Tests ofcompound-cue retrieval theory. Journal of Experimental Psychology: Learning,Memory, and Cognition, 21, 1380–1388. http://dx.doi.org/10.1037/0278-7393.21.5.1380.

Roehm, D., Bornkessel-Schlesewsky, I., Rösler, F., & Schlesewsky, M. (2007). Topredict or not to predict: Influences of task and strategy on the processing ofsemantic relations. Journal of Cognitive Neuroscience, 19, 1259–1274. http://dx.doi.org/10.1162/jocn.2007.19.8.1259.

Roland, D., Yun, H., Koenig, J.-P., & Mauner, G. (2012). Semantic similarity,predictability, and models of sentence processing. Cognition, 122, 257–279.http://dx.doi.org/10.1016/j.cognition.2011.11.011.

Rösler, F., Streb, J., & Haan, H. (2001). Event-related potentials evoked by verbs andnouns in a primed lexical decision task. Psychophysiology, 38, 694–703. http://dx.doi.org/10.1111/1469-8986.3840694.

Sánchez-Casas, R., Ferré, P., Demestre, J., García-Chico, T., & García-Albea, J. E.(2012). Masked and unmasked priming effects as a function of semanticrelatedness and association strength. Spanish Journal of Psychology, 15, 891–900.http://dx.doi.org/10.5209/rev_SJOP.2012.v15.n3.39382.

Schmitt, B. M., Münte, T. F., & Kutas, M. (2000). Electrophysiological estimates of thetime course of semantic and phonological encoding during implicit picturenaming. Psychophysiology, 37, 473–487. http://dx.doi.org/10.1111/1469-8986.3740473.

Schotter, E. R., Angele, B., & Rayner, K. (2012). Parafoveal processing in reading.Attention, Perception, & Psychophysics, 74, 5–35. http://dx.doi.org/10.3758/s13414-011-0219-2.

Schroyens, W., Vitu, F., Brysbaert, M., & d’Ydewalle, G. (1999). Eye movementcontrol during reading: Foveal load and parafoveal processing. Quarterly Journalof Experimental Psychology, 52, 1021–1046. http://dx.doi.org/10.1080/027249899390909.

Sharkey, N. E. (1989). The lexical distance model and word priming. In Proceedingsof the eleventh cognitive science conference (pp. 860–867). Hillsdale, NJ:Lawrence Erlbaum Associates.

Silva-Pereyra, J., Harmony, T., Villanueva, G., Fernández, T., Rodríguez, M., Galán, L.,et al. (1999). N400 and lexical decisions: Automatic or controlled processing?Clinical Neurophysiology, 110, 813–824. http://dx.doi.org/10.1016/S1388-2457(99)00009-7.

Smithson, M. (2001). Correct confidence intervals for various regression effect sizesand parameters: The importance of noncentral distributions in computingintervals. Education and Psychological Measurement, 61, 605. http://dx.doi.org/10.1177/00131640121971392.

Smulders, F. T. Y. (2010). Simplifying jackknifing of ERPs and getting more out of it:Retrieving estimates of participants’ latencies. Psychophysiology, 47, 387–392.http://dx.doi.org/10.1111/j.1469-8986.2009.00934.x.

Sprondel, V., Kipp, K. H., & Mecklinger, A. (2013). Timing matters: Age-relatedchanges in episodic retrieval control as revealed by event-related potentials.Brain Research, 1537, 143–155. http://dx.doi.org/10.1016/j.brainres.2013.09.005.

Squires, K. C., Squires, N. K., & Hillyard, S. A. (1975). Decision-related corticalpotentials during an auditory signal detection task with cued observationintervals. Journal of Experimental Psychology: Human Perception and Performance,1, 268–279. http://dx.doi.org/10.1037/0096-1523.1.3.268.

Szewczyk, J. M., & Schriefers, H. (2013). Prediction in language comprehension: AnERP study on sentence comprehension in Polish. Journal of Memory andLanguage, 68, 297–314. http://dx.doi.org/10.1016/j.jml.2012.12.002.

Thornhill, D. E., & Van Petten, C. (2012). Lexical versus conceptual anticipationduring sentence processing: Frontal positivity and N400 ERP components.International Journal of Psychophysiology, 83, 382–392. http://dx.doi.org/10.1016/j.ijpsycho.2011.12.007.

Thorpe, S., Fize, D., & Marlot, C. (1996). Speed of processing in the human visualsystem. Nature, 381, 520–522. http://dx.doi.org/10.1038/381520a0.



http://dx.doi.org/10.1016/j.bandc.2010.08.001

http://dx.doi.org/10.1016/j.neulet.2010.06.084

http://dx.doi.org/10.1016/j.neulet.2010.06.084




http://dx.doi.org/10.1038/307161a0

http://dx.doi.org/10.1038/307161a0

http://dx.doi.org/10.1162/jocn.1989.1.1.38


http://dx.doi.org/10.1016/S0168-5597(98)00023-9




http://dx.doi.org/10.1126/science.887923

http://dx.doi.org/10.1016/B978-012369374-7/50018-3

http://dx.doi.org/10.1016/B978-012369374-7/50018-3

http://dx.doi.org/10.1016/j.jml.2009.06.004


http://dx.doi.org/10.1111/j.1469-8986.2010.01058.x

http://dx.doi.org/10.1111/j.1469-8986.2010.01058.x

http://dx.doi.org/10.1080/01690965.2013.866259

http://dx.doi.org/10.1080/01690965.2013.866259

http://dx.doi.org/10.1162/jocn_a_00328

http://dx.doi.org/10.1038/nrn2532

http://dx.doi.org/10.1038/nrn2532

http://dx.doi.org/10.1037/0096-1523.26.3.1072

http://dx.doi.org/10.1037/0096-1523.26.3.1072

http://dx.doi.org/10.3758/s13415-013-0223-1

http://dx.doi.org/10.3758/s13415-013-0223-1

http://dx.doi.org/10.1037/0278-7393.21.1.3

http://dx.doi.org/10.1037/0278-7393.21.1.3

http://dx.doi.org/10.1037/0096-3445.126.2.99

http://dx.doi.org/10.1037/h0031564

http://dx.doi.org/10.1037/h0031564

http://dx.doi.org/10.1016/S0022-5371(63)80087-0

http://dx.doi.org/10.1016/S0022-5371(63)80087-0

http://dx.doi.org/10.1111/1469-8986.3510099

http://dx.doi.org/10.1111/1469-8986.3510099

http://dx.doi.org/10.1037/0278-7393.21.4.863

http://dx.doi.org/10.1016/S0010-0277(99)00004-9

http://dx.doi.org/10.1016/S0010-0277(99)00004-9

http://dx.doi.org/10.1037/0096-3445.106.3.226

http://dx.doi.org/10.1037/0096-3445.106.3.226

http://dx.doi.org/10.1037/0278-7393.15.6.1003


http://dx.doi.org/10.1037/0096-1523.18.1.217






http://dx.doi.org/10.1016/j.tics.2006.12.002




http://dx.doi.org/10.1037/0033-295X.95.3.385

http://dx.doi.org/10.1037/0033-295X.95.3.385

http://dx.doi.org/10.1037/0278-7393.21.5.1380

http://dx.doi.org/10.1037/0278-7393.21.5.1380




http://dx.doi.org/10.1111/1469-8986.3840694

http://dx.doi.org/10.1111/1469-8986.3840694

http://dx.doi.org/10.5209/rev_SJOP.2012.v15.n3.39382

http://dx.doi.org/10.1111/1469-8986.3740473

http://dx.doi.org/10.1111/1469-8986.3740473

http://dx.doi.org/10.3758/s13414-011-0219-2

http://dx.doi.org/10.3758/s13414-011-0219-2

http://dx.doi.org/10.1080/027249899390909

http://dx.doi.org/10.1080/027249899390909




http://dx.doi.org/10.1016/S1388-2457(99)00009-7

http://dx.doi.org/10.1016/S1388-2457(99)00009-7

http://dx.doi.org/10.1177/00131640121971392

http://dx.doi.org/10.1177/00131640121971392

http://dx.doi.org/10.1111/j.1469-8986.2009.00934.x



http://dx.doi.org/10.1037/0096-1523.1.3.268


http://dx.doi.org/10.1016/j.ijpsycho.2011.12.007


http://dx.doi.org/10.1038/381520a0


Timmer, K., Vahid-Gharavi, N., & Schiller, N. O. (2012). Reading aloud in Persian: ERPevidence for an early locus of the masked onset priming effect. Brain andLanguage, 122, 34–41. http://dx.doi.org/10.1016/j.bandl.2012.04.013.

Tulving, E., & Gold, C. (1963). Stimulus information and contextual information asdeterminants of tachistoscopic recognition of words. Journal of ExperimentalPsychology, 66, 319–327. http://dx.doi.org/10.1037/h0048802.

Ulrich, R., & Miller, J. (2001). Using the jackknife-based scoring method formeasuring LRP onset effects in factorial designs. Psychophysiology, 38, 816–827.http://dx.doi.org/10.1111/1469-8986.3510099.

Van Petten, C. (1995). Words and sentences: Event-related brain potentialmeasures. Psychophysiology, 32, 511–525. http://dx.doi.org/10.1111/j.1469-8986.1995.tb01228.x.

Van Petten, C. (2014). Selective attention, processing load, and semantics: Insightsfrom human electrophysiology. In G. R. R Mangun (Ed.), Cognitiveelectrophysiology of attention (pp. 236–253). New York: Elsevier. http://dx.doi.org/10.1016/b978-0-12-398451-7.00019-1.

Van Petten, C., & Kutas, M. (1988). The use of event-related potentials in the study ofbrain asymmetries. International Journal of Neuroscience, 39, 91–99. http://dx.doi.org/10.3109/00207458808985695.

Van Petten, C., & Kutas, M. (1991). Electrophysiological evidence for the flexibility oflexical processing. In G. Simpson (Ed.), Word and sentence (pp. 129–174).Amsterdam: North Holland Press.

Van Petten, C., & Luka, B. J. (2006). Neural localization of semantic context effects inelectromagnetic and hemodynamic studies. Brain and Language, 97, 279–293.http://dx.doi.org/10.1016/j.bandl.2005.11.003.

Van Petten, C., & Luka, B. J. (2012). Prediction during language comprehension:Benefits, costs, and ERP components. International Journal of Psychophysiology,83, 176–190. http://dx.doi.org/10.1016/j.ijpsycho.2011.09.015.

Wlotko, E. W., & Federmeier, K. D. (2013). Two sides of meaning: The scalp-recordedN400 reflects distinct contributions from the cerebral hemispheres. Frontiers inPsychology, 4, 181. http://dx.doi.org/10.3389/fpsyg.2013.00181.

Wlotko, E. W., Federmeier, K. D., & Kutas, M. (2012). To predict or not to predict:Age-related differences in the use of sentential context. Psychology and Aging,27, 975–988. http://dx.doi.org/10.1037/a0029206.


http://dx.doi.org/10.1037/h0048802

http://dx.doi.org/10.1111/1469-8986.3510099



http://dx.doi.org/10.1016/b978-0-12-398451-7.00019-1

http://dx.doi.org/10.1016/b978-0-12-398451-7.00019-1

http://dx.doi.org/10.3109/00207458808985695

http://dx.doi.org/10.3109/00207458808985695






http://dx.doi.org/10.3389/fpsyg.2013.00181

http://dx.doi.org/10.1037/a0029206

Date post:	01-Feb-2017
Category:	Documents
Upload:	cyma
View:	216 times
Download:	1 times

Prospective and retrospective semantic processing: Prediction, time, and relationship strength in...

Documents