ERPs and text verification processes 1

Electrophysiological Evidence for the Time-Course of Verifying Text Ideas

Todd R. Ferretti Wilfrid Laurier University

Murray Singer University of Manitoba

Courtney Patterson Wilfrid Laurier University

Running head: Text verification processes

Please address correspondence to: Todd Ferretti Center for Cognitive Neuroscience Department of Psychology Wilfrid Laurier University Waterloo, Ontario N2L 3C5

ERPs and text verification processes 2 Abstract We examined how verb factivity influences the ability of readers to detect and resolve the mismatch of receiving false referents in relation to true referents in discourse contexts. Factive verbs (e.g., know), but not nonfactive verbs (believe), entail the truth of their complements. Recent research by Singer (2006) suggests that there are pragmatic costs associated with knowing something that is clearly false and only believing something that is clearly true. However, because Singer measured reading times for full sentences, it could not be determined whether these costs were initiated upon the appearance of the critical target word (i.e., the word that validated or invalidated previous text ideas) or at a later point in the sentences. In the present research we recorded event-related brain potentials while people read the same passages for comprehension and analyzed potentials evoked to the critical target words. Our results demonstrate that the brain distinguishes between true and false target words by at least 200 ms after their onset, and that the pragmatic costs identified by Singer lead to interactions between verb factivity and truth in both early (P2) and later occurring brain components (late phase of N400 and late frontal positivity). In general, the results suggest readers had greater difficulty integrating false nouns than true nouns following factive than nonfactive verbs, and that detection of this mismatch also occurred earlier following factive verbs. Our results provide insight into the time-course of the processes that underlie the verification of text ideas, and extend neurocognitive research on anaphoric resolution.

ERPs and text verification processes 3 Text and discourse comprehension is central to a vast range of human enterprises. For text information to effectively guide people's behavior, it is necessary that the reader scrutinize the accuracy of text ideas. Consistent with the proposal, there is extensive evidence that readers are sensitive to the congruence between the current portion of a message and its antecedents. Readers can detect discrepancies on the basis of relatively apparent mismatches at the surface level of discourse (e.g., Klin, 1995; Long & Chong, 2000). However, they also notice inference-based inconsistencies stemming from numerous dimensions of the text situation model (Zwaan, Magliano, & Graesser, 1995); including spatial (O'Brien & Albrecht, 1992), temporal (Rinck, Hahnel, & Becker, 2001), logical (Lea & Mulligan, 2002), and causal (Suh & Trabasso (1993) facets of text. Monitoring these inferential relations depends, in part, on the reader's access to the world knowledge relevant to the message (Singer & Halldorson, 1996; Singer, Halldorson, Lear, & Andrusiak, 1992). Continual scrutiny of discourse congruence is likely a manifestation of the more general cognitive principle that stimulus identification is regularly followed by a processing stage of stimulus evaluation or comparison (Tulving, 1983). Numerous theoretical frameworks, such as construction-integration (Kintsch, 1988, 1998) and activation monitoring (Gallo & Roediger, 2002), posit such a stage (see also Garrod & Terras, 2000; Kolodner, 1983; Singer & Halldorson, 1996; Traxler, Sanford, Aked, & Moxey, 1997; Wyers & Radvansky, 1999). Processes of Verifying Text Ideas Readers' evaluation of text ideas was recently addressed within the theoretical framework of memory-based text processing (Singer, 2006). From this perspective, the current text chunk comprises memory cues for the passive retrieval of antecedent ideas as well as relevant knowledge in long-term memory (Greene, Gerrig, McKoon, & Ratcliff, 1994; O'Brien, Lorch, &

ERPs and text verification processes 4 Myers, 1998). The accessed or “resonating” (Ratcliff, 1978) ideas are available to participate in processes central to comprehension, such as anaphoric resolution (O'Brien & Albrecht, 1991) and inferential construction (Klin, 1995). These memory-based processes are critical to readers’ ability to detect text inconsistencies, as discussed at the outset. Singer (2006) presented evidence that memory-based retrieval also supports ensuing verification processes. The passive nature of memory-based retrieval is supported by the irrelevance of some resonating ideas to the gist of the text (McKoon, Gerrig, & Greene, 1996). In this framework, Singer (2006) inspected the sensitivity to text consistency among adults who read simply in anticipation of answering a comprehension question. In a typical text, people first read that Ken and his brother gobbled some apples, and later encountered target statements such as The coach determined that it was oranges that Ken ate. A full understanding of the narrative entails noticing the discrepancy between the two sentences. Reading time for targets such as the coach sentence varied systematically with the consistency between the target and the prior text. Had the reader not scrutinized text accuracy, reading time would have been invariant across the experimental conditions. More specifically, Singer's (2006) target sentences varied in their affirmative versus negative expression and in their truth and with reference to their antecedents. The reading time patterns suggested that processes of finding the antecedent and comparing it with the current text idea bear considerable similarity to processes of intentionally verifying sentences such as Two is not an odd number (e.g., Carpenter & Just, 1975; Wason & Jones, 1963). However, the find-andcompare processes were modulated by discourse pragmatic influences regulated by verb factivity (Lyons, 1977). In this regard, factives such as The coach DETERMINED that it was oranges that Ken ate entail the truth of their complements (viz. it was oranges) whereas nonfactives such as

ERPs and text verification processes 5 The coach FIGURED that it was oranges do not. Singer presented evidence that cognitive costs associated with reading that a character "determines" something false or only "figures" something true are superimposed on the times needed to find and compare text antecedents. He concluded that readers continually verify the current text with reference to preceding text ideas. The Present Study This study was designed to scrutinize the verification of text ideas using event-related brain potential methodology (ERP). According to Singer’s (2006) memory-based analysis, ERPs should reveal, at a critical word such as oranges, the signatures of the conceptual matching and of interactions of pragmatics and truth that he documented. In contrast, if Singer’s reading-time data reflected controlled processes that accompany the wrap-up of sentence comprehension (Haberlandt & Graesser, 1985), then these ERP patterns will not emerge. The whole-sentence reading times of Singer could not distinguish these alternatives. Throughout, the focus is on materials such as those shown in Table 1. Sentence 2 specified one of the two alternatives: On this day, it was very hot and Ken and his brother gobbled some oranges/apples. At sentence 5, a narrative character's view of the crucial idea was described in the factive or nonfactive form, The coach determined/figured that it was oranges that Ken ate. The reading time measure of Singer (2006) revealed a stable Truth by Factivity interaction: False reading time was reliably greater than true reading time with factive verbs, but not with nonfactive verbs. This difference was attributed to a nonfactive reading-time cost associated with reading that the coach is not fully certain (figured) of something clearly true according to prior text. The resulting inflation of the true nonfactive reading times was proposed to abolish the true-false reading time difference in that condition.1 Table 1

ERPs and text verification processes 6 Sample Materials of Experiment 1 _____________________________________________________________________________ Passage Ken enjoyed riding his bike to football practice in the afternoon with his brother. (Sentence 1) On this day, it was very hot and Ken and his brother gobbled some oranges/apples. (Sentence 2) Since it was about a five mile ride from their house to the practice field, they figured they were getting a better workout than most of the other guys on the team. (Sentence 3) By the time they got to practice, Ken was feeling sick to his stomach. (Sentence 4) The coach determined/figured that it was oranges that Ken ate. (Sentence 5) Everyone knew that they were sour at this time of year. (Sentence 6)

Question Did Ken ride his bike to football practice? (Yes) _____________________________________________________________________________ As Table 1 illustrates, the critical word that permits the evaluation of text accuracy (e.g., oranges) sometimes appeared in mid-sentence. Because Singer (2006) measured reading times for full sentences, it could not be determined whether the processes posited to establish text consistency were initiated (a) upon the appearance of the critical word (orange) or (b) only at a later time. ERP research indicates that discourse incongruence is detected at the earliest opportunity (e.g., van Berkum, Hagoort, & Brown, 1999). We applied this powerful on-line measure to achieve a higher resolution of these phenomena than was permitted by full-sentence reading time.

ERPs and text verification processes 7 Our main ERP components of interest were the N400 and late positive components. The N400 indexes semantic expectancies at both lexical and message levels within word, sentence, and discourse contexts (Brown, Hagoort, & Kutas, 2000). Words incongruent with semantic expectancies produce more negativity than congruent words between 300-500 ms after their onset, and this difference typically is maximal at central and posterior head locations. The N400 is also sensitive to integration costs associated with coreferential processing between nouns and their antecedents. For example, the N400 is reduced as a function of word repetition (Anderson & Holcomb, 2005; Burkhardt, 2006; Ledoux, Gordon, Camblin, & Swaab, 2007; Van Petten, Kutas, Kluender, Mitchiner, & McIsaac, 1991), degree of semantic mismatch between coreferring expressions (Anderson & Holcomb, 2005), the saliency and linguistic prominence of antecedents in the situation model (Burkhardt, 2006; Burkhardt & Roehm, 2007; Ledoux et al., 2007), and as a function of the type of referential connection (Burkhardt & Roehm, 2007). Recent research by Burkhardt (2006) has demonstrated differences in the early versus late phases of the N400 component that distinguish between the establishment of dependency relations as opposed to novel or independent relations. In this case, an extended negativity was found in the late phase for novel or independent relations as opposed to dependent relations (i.e., repetition and bridging relations). The N400 is sometimes followed by positivity in amplitudes at posterior head locations and/or frontal locations. Late positivity that appears approximately 600-1000 ms post stimulus onset has been suggested to reflect increased memory demands during syntactic and semantic reanalysis, difficulty in conceptual integration, and elaboration of text with information from long-term memory (Federmeier, Wlotko, De Ochoa, & Kutas, 2007; Kaan & Swaab, 2003; Van Petten et al., 1991). Importantly, late positivity has been shown to index integration costs

ERPs and text verification processes 8 associated with anaphoric complexity (Burkhardt, 2005, 2006; Ledoux et al., 1997). For example, research has demonstrated increased positivity for anaphoric connections derived from the introduction of discourse-new proper names in contrast to referentially dependent pronominal elements (Burkhardt, 2005), and for discourse-new referents and referents involving the drawing of bridging relations in contrast to repeated referents (Burkhardt, 2006). According to Burkhardt, these results suggest that the N400 and late positive components are revealing of functionally dissociable integration processes during the resolution of anaphors in discourse. Based on previous research, the N400 and late positive components should be informative about how the pragmatic constraints afforded by factive and nonfactive verbs influence the ability of readers to detect and resolve the mismatch of receiving a false referent in relation to a true referent. Specifically, if Singer’s (2006) analysis is correct, then amplitudes in these components should reflect anaphoric integration costs that are greater for false than true nouns, but this difference should be largest following factive verbs (i.e., truth and factivity should interact). Alternatively, if these pragmatic costs are not captured by these two components, then we would expect a main effect of truth in all analyses. Method Participants Sixty undergraduate students from Wilfrid Laurier University participated for course credit. All participants were native English speaking, had normal or corrected-to-normal visual acuity, and all were right-handed. Materials

ERPs and text verification processes 9 The experimental materials were 32 narrative passages derived from the materials of O’Brien, Plewes, and Albrecht (1990), including the 28 that had been adapted by Singer (2006; see Table 1). Sentence 2 always introduced one of the two alternative nouns that defined the true or false conditions. Sentence 5 always involved a character saying or thinking something about the critical target noun, and included either a factive or nonfactive verb. These passages were presented across four lists, such that each list contained eight passages representing each of the conditions obtained by crossing Truth (true, false) and verb Factivity (factive, nonfactive). The lists also included the 21 filler passages of Singer (2006). The fillers were similar to the experimental passages in length and narrative style but did not include a truth or factivity manipulation. A simple comprehension question was written for each experimental and filler passage. Procedure Participants sat in front of a computer monitor located in an electrically-shielded chamber. They were instructed to read each passage for comprehension and to answer a post-passage comprehension question by pressing buttons labeled “Yes” and “No”. Sentences 1-3 of every passage appeared in full one at time in the center of the computer screen and participants were required to press a button to signal understanding them. Sentences 4 to 6 appeared in a fixed serial visual presentation (SVP) format: Each word was displayed for 300 ms plus a 200 ms interstimulus interval, with an additional 2000 ms of blank interval following each sentence.2 Recording and Analysis. The electroencephalogram (EEG) was recorded from 64 electrodes distributed evenly over the scalp. Eye movements and blinks were monitored via electrodes placed on the outer canthus and infraorbital ridge of each eye. Electrode impedances

ERPs and text verification processes 10 were kept below 5ΩK. EEG was processed with a bandpass of 0.05 - 100 Hz, and was digitized at 250 Hz. Results The data were re-referenced off-line to the average of the left and right mastoids. High frequency noise was removed by applying a low-pass filter set at 30 Hz. ERPs were computed in epochs that extended from 100 ms before the target words to 1000 ms after their onset. Trials contaminated by EEG artifacts and trials with incorrect answers were removed from the analysis before averaging. Table 2 shows the mean amplitudes at four temporal regions of interest: 200-300 ms (P2), 300-500 ms (N400), 600-1000 ms, 800-1000 ms (frontal electrodes only). These regions were chosen based on visual inspection of the waveforms and on previous ERP research. Analyses of variance (ANOVA) were applied to these values. In all analyses, the factors of interest were Truth (true vs. false), verb Factivity (factive vs. nonfactive) and Electrode site, all of which were within-participants variables. List functioned as a between-participants factor to stabilize variance caused by rotating the experimental passages across lists. All p-values are reported after epsilon correction (Huynh-Felt) for repeated measures with greater than one degree of freedom. All F-values are significant at p < .05 unless reported otherwise.

ERPs and text verification processes 11

F7

FP1

FPZ

FP2

F3

FZ

F4

P2 T7

F8

F7

FP1

FPZ

F3

FZ

Extended N400 C3

CZ

C4

P3

PZ

P4

P7

O1

F4

F8

Late Frontal Positivity T8

T7

P8

P7

C3

CZ

C4

P3

PZ

P4

T8

P8

O1

O2

OZ

FP2

OZ

O2

1000 ms

False Noun True Noun Factive

+

False Noun True Noun Nonfactive

Figure 1. Mean amplitudes for true and false target nouns following factive (left panel) and nonfactive verbs (right panel) for selected electrode locations from all head locations.

ERPs and text verification processes 12 F7

FZ

F8

0.0

0.0

0.0

2.5

2.5

2.5

µV

µV

µV

5.0

5.0 7.5

7.5

Late Frontal Positivity

10.0 -100.0

150.0

400.0

650.0

5.0 7.5

P2 10.0 -100.0

900.0

150.0

400.0

650.0

10.0 -100.0

900.0

T7

2.5 µV

2.5 µV

5.0

5.0

5.0

7.5

7.5

7.5

400.0

650.0

10.0 -100.0

900.0

150.0

400.0

ms

650.0

900.0

10.0 -100.0

P7

2.5

2.5 µV

5.0

5.0

5.0

7.5

7.5

7.5

400.0

650.0

900.0

900.0

0.0

µV

150.0

650.0

P8

0.0

2.5

400.0 ms

PZ

µV

10.0 -100.0

150.0

ms

0.0

900.0

0.0

Extended N400

2.5

µV

650.0

T8

0.0

150.0

400.0 ms

CZ

0.0

10.0 -100.0

150.0

ms

ms

10.0 -100.0

ms

150.0

400.0

650.0 ms

False Factive True Factive

900.0

10.0 -100.0

150.0

400.0

650.0

900.0

ms

False Nonfactive True Nonfactive

Figure 2. Mean amplitudes for all conditions at Frontal (F7, FZ, F8), Central (T7, CZ, T8), and Parietal electrodes (P7, PZ, P8).

ERPs and text verification processes 13 Table 2 Mean amplitudes (µV) for each time region of interest. ___________________________________________________________ Time Region (ms) Condition 200-300 300-500 600-1000 800-1000 ___________________________________________________________ Factive True 4.42 3.71 4.45 4.39 False 3.00 .96 3.03 4.10 Difference 1.42* 2.75*** 1.42* .29 Nonfactive True 3.52 3.5 4.35 3.76 False 3.33 1.37 4.37 5.29 Difference .19 2.13*** .02 1.53* ___________________________________________________________ * p < .05, *** p < .001

ERP Time Region 200-300 ms (P2). Visual inspection of Figures 1 and 2 show that waveforms for target nouns varied as a function of truth and verb factivity. For factive verbs only, differences between true and false target nouns began approximately 200 ms following their onset. This difference was broadly distributed, and maximal at frontal electrodes. Consistent with these observations, there was a Truth X Factivity interaction, F(1, 56) = 5.28. Planned comparisons revealed that mean amplitudes for true nouns were more positive than false nouns following factive verbs, F(1, 56) = 14.01, but not nonfactive verbs, F < 1.3 The large truth effect for factive verbs contributed to a significant main effect of truth, F(1, 56) = 4.90. The main effect of factivity was nonsignificant, F < 1.

ERPs and text verification processes 14 300-500 ms (N400). The Truth X Factivity interaction was nonsignificant, F < 1.10. The truth main effect was significant, F(1, 56) = 42.00: Amplitudes for false nouns were more negative than for true nouns following factive verbs, F(1, 56) = 39.06, and nonfactive verbs, F(1, 56) = 23.40. Truth also interacted with electrode site, F(61, 3416) = 3.72. As Figures 1 and 2 show, this interaction reflects a typical N400 distribution -- broadly distributed, slightly larger over the right than left hemisphere, and maximal at central and parietal locations. The main effect of factivity was nonsignificant, F < 1. 600-1000 ms. There was a significant Truth X Factivity interaction, F(1, 56) = 4.14. This interaction occurred because amplitudes for false nouns continued to be more negative than for true nouns following factive verbs, F(1, 56) = 8.10, but not nonfactive verbs, F < 1. There was a marginal main effect of truth, F(1, 56) = 3.84, p < .06, and truth interacted with electrode site, F(61, 3416) = 2.76. The distribution of the truth effect in the factive condition resembled the same effect in the earlier N400 region (i.e., broadly distributed and largest at central and parietal locations). This suggests that the continued truth effect for the factive condition primarily reflects an extended late phase in the N400 component.4 Finally, amplitudes to target nouns in nonfactive passages were more positive than target nouns in factive passages, F(1, 56) = 3.47, p < .07. 800-1000 ms. Visual inspection of the figures also indicates that between 800-1000 ms mean amplitudes for false nouns at frontal head locations were more positive than for true nouns. ANOVA was applied to frontal electrodes to mainly capture the frontal positivity and not the extended late phase of the N400 effect observed at central and posterior locations. Truth and factivity interacted in this analysis, F(1, 56) = 4.65. Amplitudes for false nouns were more positive than for true nouns in the nonfactive condition, F(1, 56) = 6.59, but not for the factive

ERPs and text verification processes 15 condition, F < 1. Neither the truth, F(1, 56) = 2.43, p > .12, nor the factivity, F < 1, main effect was significant. Discussion The results provide insight into the time-course of text verification processes in the brain. Within 200 ms of the onset of the critical nouns, brain potentials differentiated those that were congruent with previous text ideas from those that were incongruent. In addition, verb factivity and truth exerted an interactive effect on the evaluation of text ideas, in both early and late temporal regions. Furthermore, the very rapid emergence of these patterns supports Singer’s (2006) hypotheses about a passive, memory-based contribution to the ongoing verification of text ideas (Till, Mross, & Kintsch, 1988). Singer’s whole-sentence reading time data, in contrast, could not unequivocally distinguish between the involvement of relatively automatic memory processes as opposed to controlled processes conceivably associated with a stage of sentence wrap-up. The analysis for the N400 region demonstrated a larger truth effect in the factive than nonfactive condition, although the difference was not large enough to produce a significant interaction. However, our analysis for the subsequent 600-1000 ms region demonstrated the predicted interaction more clearly in the form of differences in the late phase of the N400 component -- for factive verbs only, false nouns continued to be significantly more negative than for true nouns. According to Burkhardt (2006), the N400 region is the point at which coreferential processing for dependency relations occurs. When readers encounter repeated words, the dependency relation is resolved quickly relative to novel referents. Burkhardt also interpreted an extended N400 negativity for novel referents as evidence of continued integration costs associated with the unavailability of antecedents. If the pragmatic implications associated

ERPs and text verification processes 16 with verb factivity had no influence in this region, then in contrast to the present findings, we should have found the new (false) nouns to elicit an extended N400 negativity following both factive and nonfactive verbs. Our results extend previous findings by showing pragmatic costs associated with factive verbs (i.e., determining something that is clearly false) enhances the late phase of the N400. They also show that the pragmatic costs associated with nonfactive verbs (only "figuring" something clearly true) reduce the relative difference in continued integration difficulty between true and false nouns. The results also demonstrated late positivity at frontal head locations for false nouns following nonfactive verbs. Previous research (Burkhardt, 2006) has demonstrated increased late posterior positivity for anaphoric connections derived from the introduction of new entities into the discourse in contrast to repeated words. This positivity has been suggested to index integration costs associated with establishment of a new referent into the situation model (Burkhardt, 2006).5 In the present research, the presence of a late positivity in the nonfactive condition and its absence in the factive condition is informative about how pragmatic costs associated with verb factivity influence the establishment of referents into a reader’s situation model. Specifically, the absence of this effect following factive verbs, coupled with the extended N400 response, suggests that people fail to establish the false nouns as new or independent referents in their situation model. Alternatively, the absence of the extended N400 following nonfactive verbs, coupled with the appearance of the late positivity, suggests successful establishment of the novel referents has occurred. Our results also demonstrated that truth and factivity interact in the P2 region. Little is known about the relationship between language processing and the P2 component and thus we can only speculate on the implications of our present finding. The P2 is known to be sensitive to

ERPs and text verification processes 17 visual feature detection and extraction (Luck & Hillyard, 1994). Recent language research suggests that the P2 component is sometimes sensitive to semantic expectancy (Federmeier & Kutas, 2002; Federmeier et al., 2005), sometimes in combination with morphosyntactic information associated with verbs (Ferretti, Kutas, & McRae, 2007). Our finding of enhanced P2 amplitudes for true factive sentences may suggest that the lack of a pragmatic cost leads to more efficient visual feature extraction for expected nouns. In summary, Singer’s (2006) self-paced reading results for the full target sentences are most comparable to our P2 and late phase N400 data as these ERP components show evidence of greater processing difficulty for false than true nouns, but only when they followed factive verbs. Furthermore, the early and late phase of the N400 also helped distinguish pragmatic effects from effects of lexical repetition, and the Late Positivity results provided crucial information about whether referential integration for the false nouns was ultimately successful. In conclusion, our results extend the findings of Singer (2006) and recent ERP research on anaphoric resolution by providing electrophysiological evidence for text verification processes during reading, by demonstrating the time-course of these verification processes to the word/concept that validates or invalidates antecedent concepts mentioned in discourse, and by showing how a subtle aspect of verb meaning can determine whether referential integration is ultimately successful.

ERPs and text verification processes 18 References Albrecht, J. E., & Myers, J. L. (1995). Role of context in accessing distant information during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 1459-1468. Anderson, J. E., & Holcomb, P. J. (2005). An electrophysiological investigation of the effects of coreference on word repetition and synonymy. Brain and Language, 94, 200216. Brown, C. M., Hagoort, P., & Kutas, M. (2000). Postlexical Integration Processes in Language Comprehension: Evidence from Brain-Imaging Research. In M. S. Gazzaniga (Ed.), The New Cognitive Neurosciences (pp. 881-898). Cambridge: MIT Press. Burkhardt, P. (2005). The syntax-discourse interface: Representing and interpreting dependency. Amsterdam, Philadelphia: John Benjamins. Burkhardt, P. (2006). Inferential bridging relations reveal distinct neural mechanisms: Evidence from event-related brain potentials. Brain and Language, 98, 159-168. Burkhardt, P., & Roehm, D. (2007). Differential effects of saliency: An event-related brain potential study. Neuroscience Letters, 413, 115-120. Carpenter, P. A., & Just, M. A. (1975). Sentence comprehension: A psycholinguistic processing model of verification. Psychological Review, 82, 45-73. Coulson, S., Federmeier, K. D., Van Petten, C., & Kutas, M. (2005). Right Hemisphere Sensitivity to Word- and Sentence-Level Context: Evidence From Event-Related Brain Potentials. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31, 129-147.

ERPs and text verification processes 19 Gallo, D. A., & Roediger, H. L. III. (2002). Variability among word lists in eliciting memory illusions: evidence for associative activation and monitoring. Journal of Memory and Language, 47, 469-497. Haberlandt, K. F., & Graesser, A. C. (1985). Component processes in text comprehension and some of their interactions. Journal of Experimental Psychology: General, 114, 357-374. Federmeier, K. D., & Kutas, M. (2002). Picture the difference: electrophysiological investigations of picture processing in the two cerebral hemispheres. Neuropsychologia, 40, 730-747. Federmeier, K. D., Mai, H., & Kutas, M. (2005). Both sides get the point: Hemispheric sensitivities to sentential constraint. Memory and Cognition, 33, 871-886. Federmeier, K. D., Wlotko, E. W., De Ochoa-Dewald, E., & Kutas, M. (2007). Multiple effects of sentential constraint on word processing. Brain Research, 1146, 75-84. Ferretti, T. R., Kutas, M., & McRae (2007). Verb Aspect and the Activation of Event Knowledge. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33, 182-196. Friederici, A. D., Hahne, A., & Saddy, D. (2002). Distinct neurophysiological patterns reflecting aspects of syntactic complexity and syntactic repair. Journal of Psycholinguistic Research, 31(1), 45-63. Garrod, S., & Terras, M. (2000). The contribution of lexical and situational knowledge to resolving discourse roles: Bonding and resolution. Journal of Memory and Language, 42, 526-544. Greene, S. B., Gerrig, R. J., McKoon, G., & Ratcliff, R. (1994). Unheralded pronouns and management by common ground. Journal of Memory and Language, 35, 511-526.

ERPs and text verification processes 20 Kintsch, W. (1988). The role of knowledge in discourse comprehension: A constructionintegration model. Psychological Review, 95, 163-182. Kintsch, W. (1998). Comprehension. New York: Cambridge University Press. Klin, C. M. (1995). Causal inferences in reading: From immediate activation to long term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 14831494. Kolodner, J. L. (1983). Maintaining organization in a dynamic long-term memory. Cognitive Science, 7, 243-280. Lea, R. B., & Mulligan, E. J. (2002). The effect of negation on deductive inferences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28, 303-317. Ledoux, K., Gordon, P. C., Chamblin, C. C., & Swaab, T. Y. (2007). Coreference and lexical repetition: Mechanisms of discourse integration. Memory and Cognition, 35, 801-815. Long, D. L., & Chong, J. L. (2001). Comprehension skill and global coherence: A paradoxical picture of poor comprehenders' abilities. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 1424-1429. Luck, S. J., & Hillyard, S. A. (1994). Spatial filtering during visual search: Evidence from human electrophysiology. Journal of Experimental Psychology: Human Perception and Performance, 20, 1000-1014. Lyons, J. (1977). Semantics, Vol. 2. Cambridge: Cambridge University Press. McKoon, G., Gerrig, R. J., & Greene, S. B. (1996). Pronoun resolution without pronouns: Some consequences of memory based text processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 919-932.

ERPs and text verification processes 21 O'Brien, E. J., & Albrecht, J. E. (1991). The role of context in accessing antecedents in text. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 94102. O'Brien, E. J., Lorch, R. F., & Myers, J. L. (1998). Memory-based text processing [Special issue]. Discourse Processes, 26(2-3). Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85, 59-108. Rinck, M., Hahnel, A., & Becker, B. (2001). Using temporal information to construct, update, and retrieve situation models of narratives. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 67-80. Singer, M. (2006). Verification of text ideas during reading. Journal of Memory and Language, 54, 574-591. Singer, M., & Halldorson, M. (1996). Constructing and validating motive bridging inferences. Cognitive Psychology, 30, 1-38. Singer, M., Halldorson, M., Lear, J. C., & Andrusiak, P. (1992). Validation of causal bridging inferences. Journal of Memory and Language, 31, 507-524. Suh, S., & Trabasso, T. (1993). Inferences during reading: Converging evidence from discourse analysis, talk-aloud protocols and recognition priming. Journal of Memory and Language, 32, 279-300. Till, R. E., Mross, E. F., & Kintsch, W. (1988). Time course of priming for associate and inference words in a discourse context. Memory & Cognition, 16, 283-298. Traxler, M. J., Sanford, A. J., Aked, J. P., & Moxey, L. M. (1997). Processing causal and diagnostic statements in discourse. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 88-101.

ERPs and text verification processes 22 Tulving, E. (1983). Elements of episodic memory. New York: Oxford University Press. van Berkum, J. J. A., Hagoort, P., & Brown, C. M. (1999). Semantic integration in sentences and discourse: evidence from the N400. Journal of Cognitive Neuroscience, 11, 657-671. van Petten, C., Kutas, M., Kluender, R., Mitchiner, M., & McIsaac, H. (1995). Fractioning the word repetition effect with event-related potentials. Journal of Cognitive Neuroscience, 3, 131-150. Wason, P. C., & Jones, S. (1963). Negatives: Denotation and connotation. British Journal of Psychology, 54, 299-307. Wyers, R. S., & Radvansky, G. A. (1999). The comprehension and validation of social information. Psychological Review, 106, 89-118. Zwaan, R.A., Magliano, J.P., & Graesser, A.C. (1995) Dimensions of situation-model construction in narrative comprehension. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 386-397.

ERPs and text verification processes 23 Author Notes This research was supported by a CFI grant to the first author, and by separate NSERC discovery grants awarded to the first and second authors. We would like to thank Kara Federmeier, Marta Kutas, Gabriel Radavansky, Ed O’Brien, Gerry Altmann, and an anomalous reviewer for helpful comments on aspects of this article.

ERPs and text verification processes 24 Footnotes 1. There was also a cost, in the factive condition, of reading that a character "determined" something false. This cost enhanced the true-false reading time difference in the factive condition. 2. Using full-sentence display for sentences 1 to 3 rather than the SVP format typical of ERP methodology was intended to reduce eye-strain through the long passages and thus also reduce eye movement artifacts in the ERP data. Furthermore, using SVP for sentence 4 rehabituated the participants to that format before they reached the critical target word in sentence 5. 3. Following Federmeier et al. (2005), we also conducted an analysis on only frontal electrode locations, where P2 amplitudes have been shown to be most sensitive to higher order attention and perceptual related processing. This analysis also helps index P2 effects more independently from the leading edge of the N400 region where differences in amplitudes are maximal at central and parietal locations. Thus, the present analysis included all 23 electrodes anterior to central electrodes. The results of this analysis demonstrated a stronger interaction between truth and factivity, F(1, 56) = 7.61. Mean amplitudes for true target nouns were more positive than false nouns following factive verbs (true: 6.96 µV; false: 4.88 µV), F(1, 56) = 19.86, but not following nonfactive verbs (true: 5.71 µV; false: 5.45 µV), F < 1. The larger truth effect for factive verbs also led to a more robust main effect of truth, F(1, 56) = 6.98. The main effect of factivity was again nonsignificant, F < 1. 4. Note that extended late phases in the N400 component have been shown to last up to at least 900 ms (e.g., Coulson, Federmeier, Van Petten, & Kutas, 2005).

ERPs and text verification processes 25 5. Burkhardt (2005, 2006) reported a posterior positivity as opposed to our frontal positivity. It is beyond the scope of this article to speculate why late positivity is sometimes frontal and sometimes posterior. However, for our present purposes it is important to note that both frontal (Kaan & Schwab, 2003; Friederici, Hahne, & Saddy, 2002) and posterior (Burkhardt, 2006) positivity has been linked to integration costs associated with discourse and sentence complexity.