NEUROREPORT

COGNITIVE NEUROSCIENCE AND NEUROPSYCHOLOGY

Event-related brain potential evidence for early e¡ects of neighborhood density in word recognition Vanessa Talera and Natalie A. Phillipsb a

Concordia University and bLady Davis Institute for Medical Research, Montreal, Quebec, Canada

Correspondence to Natalie A. Phillips, Department of Psychology, Concordia University, 7141Sherbrooke Street West, Montre¤al, Que¤bec, Canada H4B1R6 Tel: + 1514 848 2424 ext. 2218; fax: + 1514 848 4537; e-mail: [email protected] Received 5 September 2007; accepted 7 September 2007

Orthographic neighborhood density (ND) has been shown to play an important role in lexical access. In this study, event-related potentials were recorded as participants read sentences terminating in a high or low ND word, which was either congruent or incongruent with the sentence context. High ND words elicited a signi¢cantly more positive-going waveform at 150^300 ms poststimulus onset than low ND words, whereas items seen in

incongruent sentence contexts resulted in a larger N400 amplitude than those seen in congruent sentence contexts. We suggest that high ND items result in greater initial global activation in the lexical system, re£ected in the early positivity, and that this activation proceeds automatically and independently of context. c 2007 Wolters Kluwer Health | NeuroReport 18:1957^1961
Lippincott Williams & Wilkins.

Keywords: event-related potentials, lexical access, neighborhood density

Introduction A central question arising in the lexical processing literature is how lexical activation unfolds over time. One important variable is a word’s orthographic neighborhood density (ND), defined as the number of similar words, or neighbors, existing in the lexicon. Neighbors may be defined by the Nmetric [1], whereby a neighbor is a word that differs from the target word by one letter (e.g. badge has two neighbors, barge and budge). Previous research indicates that high ND (HND) words engender shorter response times (RTs) than low ND (LND) words in lexical decision and word-naming tasks [2–8], but that HND has an inhibitory effect in word identification [9–11]. HND pseudowords, however, exhibit longer RTs than LND pseudowords in lexical decision tasks [1,2]. Grainger and Jacobs [7] account for these seemingly contradictory findings within their multiple readout model of word recognition, which holds that lexical decisions may be made either through the activation of individual word representations in memory, or via global lexical activation, reflecting overall activity in the lexicon. Owing to activation of neighbors, an HND letter string elicits greater global lexical activity than a LND letter string, leading to faster ‘yes’ responses, but inhibiting ‘no’ responses; this accounts for the opposing effects of ND on words and pseudowords. Holcomb et al. [12] assessed the multiple readout model using event-related brain potentials (ERPs). Although the behavioral results were in line with those from previous studies, HND words and pseudowords generated larger N400s than did LND items. These results were interpreted as consistent with Grainger and Jacobs’ [7] hypothesis that a

large orthographic neighborhood causes greater global activation within the lexical system, reflected in a larger N400. Holcomb et al. [12] also found a greater negativity to HND items between 150 and 300 ms in experiment 2. Although this finding was not explored further, we believe that it is of interest, especially given that HND exerts an inhibitory effect in word recognition [9–11], which is generally taken to occur prior to 400 ms. A number of studies suggest that lexical effects may occur as early as 200 ms poststimulus onset (e.g. [13–15]). Syllable frequency effects have been reported at 150 ms [16], likely reflecting initial activation of multiple candidates when the target has a high frequency initial syllable. This study aimed to further investigate the effects of ND on the brain’s electrophysiological response. Previous research [12] has used lexical decision and semantic categorization tasks. Given the conflicting findings with respect to the effects of ND in different tasks, however, it is important to determine whether similar effects are seen in other tasks also. Therefore, we recorded ERPs as participants performed a sentence reading task and responded to periodic comprehension questions. As ND likely plays a role in early word recognition, we predicted that ND effects would be seen very early in the ERP waveform. Holcomb et al. [12] interpreted the larger N400 to HND items as indexing greater lexical activation. Following Barber et al. [16], we suggest that this effect may be due to more effortful lateral inhibition (for evidence that the N400 may index aspects of inhibition, see [17]). Thus, we presented HND and LND words in both congruent and incongruent

c Wolters Kluwer Health | Lippincott Williams & Wilkins 0959- 4965
Vol 18 No 18 3 December 2007 19 5 7 Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

NEUROREPORT sentence contexts, reasoning that inhibition should be less effortful when lexical selection is supported by a congruent context, and that, if the larger N400 to HND items [12] reflects inhibitory demands, then it should be attenuated in a congruent context. If, in contrast, it results from greater global lexical activity in the system, then sentence congruency should not interact with ND, as context should not influence the increased global activity.

Materials and methods Participants Twenty-two healthy English speakers between 18 and 35 years of age (eight men and 14 women) were recruited from the Concordia University community and through word of mouth (average age¼23.773.9; average education¼ 15.172.7). All had normal or corrected-to-normal vision, and no history of neurological or psychiatric disease. All but one was right-handed, according to the participants’ report. They were remunerated for their participation. The study received ethical approval from the Research Ethics Committee at the Jewish General Hospital, Montreal, Canada and participants signed an informed consent form prior to the experiment. Stimuli Stimulus materials comprised five or six-letter nouns, 90 HND (average ND¼3.1) and 90 LND (ND¼0), appearing in either a congruent or an incongruent sentence context (n¼45/condition). Stimuli were matched for length and for frequency using data from the English lexicon project of Washington University in St Louis (ELP, [18]) and Kuc˘era and Francis [19]. Mean bigram frequency, sum bigram frequency and bigram frequency by position were controlled using the ELP [18]. Concreteness ratings were collected from 32 English-speaking young adults, and were matched across the conditions (mean concreteness rating¼4.67 on a scale of 1–5, where 5 represents the highest concreteness rating). The ELP also provides mean naming and lexical decision RT and accuracy for each item. The stimuli show the traditional HND advantage in both RT and accuracy measures (lexical decision: HND- mean RT¼654 ms, mean accuracy¼93%; LND- mean RT¼712 ms, mean accuracy¼85%; naming: HND- mean RT¼637 ms, mean accuracy¼98%; LND- mean RT¼677 ms, mean accuracy¼94%; Po0.001 for all measures). Procedure Participants were tested individually in a dimly-lit room in a single session of around 2 h. (Participants also completed an unrelated experiment examining lexical ambiguity, which comprised two 12-min blocks, one performed before and one after this experiment. Participants silently read the words and answered comprehension questions. Critical stimuli differed for the two experiments. The 2-h session included both the experiments.) Sentences appeared in a pseudorandom order, with at most three congruent or incongruent sentences or three HND or LND stimuli in a row. Sentences were presented in 10 blocks of approximately 3 min each, and appeared one word at a time, in yellow 36-point Arial font on a black background. The maximal visual angle for a critical stimulus was 6.141. Each word appeared for 600 ms; interstimulus interval was 0 ms. The critical stimulus was always the sentence-final word; it

TALER AND PHILLIPS

included a sentence-final period and appeared for 600 ms. The interstimulus interval between the critical stimulus and the onset of the following sentence was 1400 ms. To ensure attentiveness, participants responded to periodic comprehension questions, presented in white 24-point Arial font on a black background, and remaining on the screen until a response was given. Comprehension questions were asked on 16.7% of trials (40 questions), randomly distributed throughout the session. Participants were verbally instructed to read sentences appearing on the screen word by word. They were told that some sentences would not make sense, and that they would sometimes be asked questions about the last sentence they had read, and were asked to respond verbally to the questions. They were then verbally provided with a sample sentence and question that were not stimuli in the study. Electroencephalogram recording and data analysis Electroencephalogram (EEG) was sampled continuously from 32 Ag/AgCl electrodes fitted into a commercially available nylon EEG cap (EZ Cap, Herrsching-Breitbunn, Germany). Impedances were kept below 5 kO. Electrodes were positioned following the 10–20 system convention. The EEG was recorded from five midline sites (Fz, FCz, Cz, CPz, Pz) and 20 lateral sites (FP1, FP2, F3, F4, F7, F8, FT7, FT8, FC3, FC4, P3, P4, P7, P8, CP3, CP4, TP7, TP8, O1, O2), relative to a left ear reference, which was recomputed to a linked ear reference off-line. Electrooculogram activity (EOG) was recorded from electrodes placed at the outer canthi of both the eyes (horizontal EOG) and above and below the left eye (vertical EOG). A cephalic (forehead) location served as the ground. Critical EEG epochs were time-locked to the onset of each sentence-final word, amplified using Neuroscan Synamps (Compumedics, Inc., El Paso, Texas, USA) in a DC-30 Hz bandwidth, sampled at 100 Hz for 1100 ms (100 ms prestimulus) and processed offline using Neuroscan Edit 4.3 software (Compumedics, Inc., El Paso, Texas, USA). The file was rereferenced off-line and a DC drift correction was applied. We then specified and corrected it for vertical EOG artifact; this procedure included data from an EEG recording containing no EOG artifact, allowing retention of legitimate EEG signals at the EEG electrodes. The file was then epoched into discrete EEG trials on the basis of stimulus onset. A two-stage artifact rejection procedure was applied, rejecting trials contaminated by horizontal EOG activity (750 mV) or where the EEG epoch was contaminated by clipping, movement artifact, etc. (7100 mV). Average ERP waveforms for each participant in each condition were calculated from the remaining EEG trials (average number of trials per cell¼42.273.5). All participants with less than 30 trials in any condition were excluded from analyses (n¼4), leaving behind a total of 18 participants.

Results Repeated measures analyses of variance (ANOVAs) were conducted using SPSS (Version 10.0, SPSS Inc., Chicago, Illinois, USA) statistical software, and the Greenhouse and Geisser [20] nonsphericity correction was used when appropriate. Interactions significant at Po0.05 were further decomposed using simple effects post-hoc analyses. Following convention, we report unadjusted degrees of freedom,

19 5 8 Vol 18 No 18 3 December 2007 Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

NEIGHBORHOOD DENSITYAND WORD RECOGNITION

as well as the Greenhouse and Geisser epsilon value (e) and adjusted P value. The ERP grand average waveforms for HND and LND items in congruent and incongruent contexts, averaged across participants, are shown in Fig. 1. At around 200 ms, HND items can be seen to elicit a more positive peak than LND items in posterior sites. Data analyses proceeded as follows: Separate analyses were conducted for midline sites (Fz, FCz, Cz, CPz, Pz; yielding a factor of Site) and for lateral sites as a function of anteriority and laterality (left anterior: F7, F3, FT7, FC3; right anterior: F8, F4, FT8, FC4; left posterior: TP7, CP3, P7, P3; right posterior: TP8, CP4, P8, P4). ANOVAs were conducted to examine the effects of length, ND and congruency from 150 to 300 ms and from 300 to 600 ms. Mean amplitude was calculated in 50-ms windows (i.e. 150–200 ms, 200–250 ms, etc.), yielding a factor of time. Thus, four ANOVAs were conducted, two examining response in midline sites (one in each time window), and two examining response in lateral sites in each time window. Early effects (150–300 ms) HND stimuli elicited a more positive-going waveform at electrode Pz [ND  site, F(4,68)¼6.32, Po0.011] and in posterior lateral sites [ND  anteriority, F(1,17)¼18.24, Po0.001]. In lateral sites, HND stimuli additionally elicited a more positive-going waveform in right hemisphere sites [ND  hemisphere, F(1,17)¼4.55, Po0.048]. The early positivity to HND items across both hemispheres was strongest between 200 and 250 ms [ND  time, F(2,34)¼3.28, Po0.051]. Later effects (300–600 ms) No significant effects of congruency or ND were seen in midline sites in this time window. A trend towards a more negative-going response to HND items in anterior midline sites and more positive-going response to HND items in posterior midline sites was seen [ND  site, F(4,68)¼3.29, P¼0.59], although LSD post-hoc tests found no significant effects. In the lateral sites analysis, incongruent stimuli elicited a more negative-going waveform in anterior lateral sites between 400 and 500 ms [congruency  anteriority  time, F(5,85)¼4.66, Po0.005]. Additionally, a more positive-going waveform was seen to HND than to LND items in the incongruent condition in the right hemisphere [congruency  ND  hemisphere, F(1,17)¼7.92, Po0.012]. Finally, HND stimuli elicited a more positivegoing response from 300 to 350 ms in posterior sites in the congruent condition, and from 400 to 500 ms in posterior sites in the incongruent condition [congruency  ND  anteriority  time, F(5,85)¼3.02, Po0.05].

Discussion ND exerted an effect on the ERP waveform from 150 to 300 ms, with HND items eliciting a more positive deflection in posterior lateral sites than LND items. Congruency played a role from 300 to 600 ms, with incongruent stimuli eliciting a more negative peak in anterior lateral sites than congruent stimuli. Additionally, HND lexical items elicited a more positive-going waveform in posterior and right hemisphere sites from 300 to 600 ms poststimulus onset, reflecting a sustained positivity to HND items that is stronger in the incongruent condition.

NEUROREPORT The early effects of ND suggests that it plays an important role in the initial access and/or activation of lexical candidates. Unlike previous research [12], the early ND effect, however, was characterized by an enhanced positivity for HND items and a different topographical distribution. We suggest that this effect may be a modulation of either the visual P2 or the recognition potential (RP) [13,21]. The RP is a peak at 200–250 ms elicited by recognizable images, including words, and modulated by semantic factors such as concreteness [22] and word class [23]. Its polarity varies depending on task demands, and it has been suggested to originate in the visual word form area (for a review, see [24]). Orthographic ND is a good candidate to modulate the RP, as it likely plays an important role in the recognition of the orthographic word forms. If the effect seen here is an RP, this suggests that readers experience greater difficulty recognizing LND than HND words, consistent with data from lexical decision tasks (e.g. [2]). The RP, however, is typically elicited using procedures such as rapid stream stimulation, which we did not use; furthermore, it is usually left-lateralized, whereas the early potential seen here was bilaterally distributed. Further research is thus necessary to determine whether the effect seen here is related to the RP. Additionally, HND items elicited more positivity at lateralized sites in the N400 time window, which was stronger and long-lasting in the incongruent condition. This likely reflects sustained positivity to HND items that is long-lasting when lexical activation is not constrained by context. The congruency effect was seen only in the N400 window. One question that arises is why the results reported here differ from those reported by Holcomb et al. [12], who found a larger N400 and no early positivity to HND items. Although in this study HND items did elicit a more negative-going response in anterior midline sites, this effect did not reach significance. One possibility is that the different stimulus presentation and task demands in the two studies, particularly the availability of sentence context in this study, may have resulted in reduced inhibitory requirements and hence a reduction in N400 activity. A second possible explanation rests upon the question of neighbor frequency. HND stimuli of Holcomb et al. had an average of 5.3 neighbors of higher frequency, whereas the HND stimuli used in this study had an average of 1.37 [t(247)¼12.948, Po0.001]. The existence of many higher frequency neighbors likely increases inhibitory requirements, resulting in a larger N400 response.

Conclusion This study examined the effects of ND and congruency on the brain’s electrophysiological response. HND items engendered a more positive-going response from 150 to 300 ms poststimulus onset, whereas congruency exerted an effect on the N400. The lack of interaction between ND and congruency prior to 300 ms indicates that ND influences initial lexical activation early and is independent of context effects, although activation appears to be long-lasting when an item is seen in an incongruent context. This finding has important implications for models of word recognition, suggesting that the role of sublexical factors such as ND in initial word recognition and the concomitant influences on

Vol 18 No 18 3 December 2007 19 59 Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

NEUROREPORT

TALER AND PHILLIPS

Congruent HND, for example, 'The scientist labelled the flask' Congruent LND, for example, 'Johnny has an infected tonsil' Incongruent HND, for example, 'There are mice in the hammer' Incongruent LND, for example, 'They sheltered under the muscle'

Fz

F8

F7

FCz

FT8

FT7

Cz

TP8

TP7

CPz

T6

T5

–6

Pz Stimulus onset

–4 –2 0 2 4 6 0

200

400

600

800

Fig. 1 Grand average waveforms for 18 participants, showing HND and LND items in congruent and incongruent conditions. Time is plotted on the x-axis and amplitude in microvolts on the y-axis; following convention, negative amplitudes are plotted upwards. HND, high neighborhood density; LND, low neighborhood density.

overall lexical activation proceed automatically and regardless of context.

Lezley Ingenito and Tobias Leim for their assistance in participant recruitment and testing.

Acknowledgements

References

This study was supported by a postdoctoral fellowship awarded to V.T. by the Alzheimer’s Society of Canada/ Fonds de recherche en sante´ du Que´bec. The authors thank

1. Coltheart M, Davelaar E, Jonasson JT, Besner D. Access to the internal lexicon. In: Dornic S, editor. Attention and performance IV. Hillsdale, New Jersey: Erlbaum; 1977. pp. 535–555.

19 6 0 Vol 18 No 18 3 December 2007 Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

NEIGHBORHOOD DENSITYAND WORD RECOGNITION

2. Andrews S. Frequency and neighborhood effects on lexical access: activation or search? J Exp Psychol Learn Mem Cogn 1989; 15: 802–814. 3. Andrews S. Frequency and neighborhood effects on lexical access: lexical similarity or orthographic redundancy? J Exp Psychol Learn Mem Cogn 1992; 18:234–254. 4. Andrews S. The effect of orthographic similarity on lexical retrieval: resolving neighborhood conflicts. Psychon Bull Rev 1997; 4:439–461. 5. Carreiras M, Perea M, Grainger J. Effects of the orthographic neighborhood in visual word recognition: cross-task comparisons. J Exp Psychol Learn Mem Cogn 1997; 23:857–871. 6. Forster KI, Shen D. No enemies in the neighborhood: absence of inhibitory neighborhood effects in lexical decision and semantic categorization. J Exp Psychol Learn Mem Cogn 1996; 22:696–713. 7. Grainger J, Jacobs AM. Orthographic processing in visual word recognition: a multiple read-out model. Psychol Rev 1996; 103:518–565. 8. Sears CR, Hino Y, Lupker SJ. Neighborhood frequency and neighborhood size effects in visual word recognition. J Exp Psychol Hum Percept Perform 1995; 21:876–900. 9. Grainger J, Segui J. Neighborhood frequency effects in visual word recognition: a comparison of lexical decision and masked identification latencies. Percept Psychophys 1990; 47:191–198. 10. Perea M, Carreiras M, Grainger J. Blocking by word frequency and neighborhood density in visual word recognition: a task-specific response criteria account. Mem Cognit 2004; 32:1090–1102. 11. Pollatsek A, Perea M, Binder KS. The effects of ‘neighborhood size’ in reading and lexical decision. J Exp Psychol Hum Percept Perform 1999; 25:1142–1158. 12. Holcomb PJ, Grainger J, O’Rourke T. An electrophysiological study of the effects of orthographic neighborhood size on printed word perception. J Cogn Neurosci 2002; 14:938–950.

NEUROREPORT 13. Martin-Loeches M, Hinojosa JA, Gomez-Jarabo G, Rubia FJ. The recognition potential: an ERP index of lexical access. Brain Lang 1999; 70:364–384. 14. Pulvermuller F, Lutzenberger W, Preissl H. Nouns and verbs in the intact brain: evidence from event-related potentials and high-frequency cortical responses. Cereb Cortex 1999; 9:497–506. 15. Sereno SC, Rayner K, Posner MI. Establishing a time-line of word recognition: evidence from eye movements and event-related potentials. NeuroReport 1998; 9:2195–2200. 16. Barber HA, Vergara M, Carreiras M. Syllable-frequency effects in visual word recognition: evidence from ERPs. NeuroReport 2004; 15:545–548. 17. Debruille JB. Knowledge inhibition and N400: a study with words that look like common words. Brain Lang 1998; 62:202–220. 18. Balota DA, Cortese MJ, Hutchison KA, Neely JH, Nelson D, Simpson GB, et al. The English Lexicon Project: a web-based repository of descriptive and behavioral measures for 40 481 English words and nonwords (Electronic database). St Louis, Missouri: Washington University; 2002. 19. Kuc˘era H, Francis WN. Computational analysis of present day American English. Providence: Brown University Press; 1967. 20. Greenhouse SW, Geisser S. On methods in the analysis of profile data. Psychometrika 1959; 24:95–112. 21. Ruddell AP. The recognition potential: a visual response evoked by recognizable images. Abstracts-Neurosci 1992; 16:106. 22. Martin-Loeches M, Hinojosa JA, Fernandez-Frias C, Rubia FJ. Functional differences in the semantic processing of concrete and abstract words. Neuropsychologia 2001; 39:1086–1096. 23. Hinojosa JA, Martin-Loeches M, Casado P, Munoz F, Carretie L, Fernandez-Frias C et al. Semantic processing of open- and closed-class words: an event-related potential study. Cogn Brain Res 2001; 11:397–407. 24. Martin-Loeches M. The gate for reading: Reflections on the rec0ognition potential. Brain Res 2007; 53:89–97.

Vol 18 No 18 3 December 2007 19 61 Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.