Brain potentials elicited by prose-embedded linguistic anomalies

LEE OSTERHOUT 0 1 MARK D. ALLEN 0 1 JUDITH MCLAUGHLIN 0 1 KAYO INOUE 0 1 0 This work was supported by Grant 2 R01DC01947 from the National Institute on Deafness and Other Communication Disorders, National Institutes of Health , awarded to L.O. We thank Rick McKinnon for help hout, Department of Psychology , Box 351525, University of Washing- ton , Seattle, WA 98195 ( 1 University of Washington , Seattle, Washington Linguistic theories distinguish between syntax (sentence form) and semantics (sentence meaning). Correspondingly, recent studies have shown that syntactic and semantic anomalies elicit distinct changes in the event-related brain potential (ERP). However, these results have been obtained with highly artificial methodologies and have not yet been generalized to more natural reading conditions. Here, we recorded ERPs while subjects read a naturalistic prose passage. The subjects either read for comprehension with no other task being assigned or read for comprehension and made acceptability judgments after each sentence. Consistent with prior work and regardlessof the subjects' assigned task, syntactic anomalies elicited a large positive wave (P600), whereas semantic anomalies elicited a large increase in N400 amplitude. These results demonstrate that the qualitatively distinct ERP responses elicited by syntactic and semantic anomalies are not artifacts of unnatural aspects of previously used stimuli, thereby providing additional evidence that separable syntactic and semantic processes exist. - One fundamental claim of current linguistic theories is that syntax (sentence form) and semantics (sentence meaning) are separable and independent (see Chomsky, 1986). Whether this claim is an accurate description of the processes underlying language comprehension has been a matter of debate. A common assumption within psycholinguistics is that separable, largely independent processes construct distinct syntactic and semantic representations of a sentence (Ferreira & Clifton, 1986; Frazier & Rayner, 1982). An alternative view is that sentence meaning can be derived directly, without an intervening syntactic level of representation (Johnson-Laird, 1983; Marslen-Wilson & Tyler, 1987; McClelland, St. John, & Taraban, 1989). One means for contrasting these views involves recording event-related brain potentials (ERPs) elicited during language comprehension (see Hagoort, Brown, & Osterhout, 1999; Osterhout, McLaughlin,& Bersick, 1997). The advantage of this approach derives from the fact that ERPs provide an on-line, millisecond-by-millisecond record of the brains electrical activity during comprehension. Furthermore, ERPs are multidimensional, varying in polarity, timing, morphology, and scalp distribution. If one assumes that cognitivelydistinct processes are mediated by neurally distinct brain systems, evidence that events occurring at the syntactic and the semantic levels elicit distinct brain responses (i.e., responses that differ in timing, distribution,or polarity) could be construed as supporting the claim that separable syntactic and semantic processes exist. Particularly relevant to this issue is recent evidence that (at least under certain experimental conditions) syntactic and semantic anomalies elicit distinct changes in the ERP (for a review, see Hagoort et al., 1999, or Osterhout et al., 1997). Semantically inappropriate words (e.g., The cat will bake the food) elicit a centroparietal negative wave that peaks at about 400 msec (the N400 effect ; Kutas & Hillyard, 1980, 1984). Syntactically anomalous words (e.g., The cat will eating the food) elicit a centroparietal positive wave that begins about 500 msec after word onset (the P600 effect) and persists for at least several hundred milliseconds (Hagoort, Brown, & Groothusen, 1993; Osterhout, 1997; Osterhout & Holcomb, 1992, 1993; Osterhout, Holcomb, & Swinney, 1994; Osterhout & Mobley, 1995; Osterhout & Nicol, 1999). In some reports, syntactic anomalies have also elicited an anterior negative wave peaking between 125 and 500 msec (Friederici, Hahne, & Mecklinger, 1996; Neville, Nicol, Barss, Forster, & Garrett, 1991; Osterhout & Holcomb, 1992). These findings have been shown to generalize well across types of linguistic anomalies (see Osterhout et al., 1997), types of languages (including English, Dutch, Italian, and French; Angrilli et al., 2002; Hagoort et al., 1993; Osterhout, McLaughlin,Allen, Inoue, & Loveless, 2002), modality (visual and auditory; Hagoort & Brown, 2000; Holcomb & Neville, 1991; Osterhout & Holcomb, 1993), and rates of word presentation (Allen, Badecker, & Osterhout, in press; Hagoort & Brown, 2000; Kutas, 1993; McKinnon & Osterhout, 1996). Such evidence seems to indicate that at least some aspects of syntactic and semantic processing are neurobiologically distinct. However, this evidence has been obtained with highly artificial methods. For example, in the vast majority of these studies, subjects were instructed to read or listen to long lists of unrelated sentences. As far as we can ascertain, only two published studies have contrasted the ERP responses to syntactic and semantic anomalies embedded in naturalistic prose. Kutas and Hillyard (1983) had subjects read an English prose passage containing semantic anomalies and morphosyntactic anomalies involving errors in number agreement or verb tense (e.g., As a turtle grows its shell grow too). The semantic anomalies elicited an increase in N400 amplitude. Few reliable differences were found between the syntactically well-formed and the ill-formed conditions, although the syntactically anomalous words elicited a small increase in negativity between 300 and 400 msec at some anterior electrode sites. In a similar study (Mnte, Heinze, Matzke, Wieringa, & Johannes, 1998), subjects read a German prose passage containing semantically anomalous, morphosyntactically anomalous (involving case errors), and misspelled words. Only semantically anomalous words elicited a robust N400 effect. However, all anomaly types elicited positive waves onsetting at about 500 msec and persisting for at least 500 msec. The results of these two studies are inconsistent with each other and with the results of studies in which lists of unrelated sentences were used. Procedural differences between Kutas and Hillyards (1983) and Mnte et al.s (1998) experiments (and between these two studies and sentence list studies) might partly explain the discrepancies. For example, Kutas and Hillyard (1983) presented English stimuli, whereas Mnte et al. (1998) presented German stimuli. Kutas and Hillyard (1983) used wordonset intervals ranging from 640 to 760 msec, whereas Mnte et al. (1998) used word onset intervals of 1 sec. Furthermore, Kutas and Hillyard (1983) plotted only the initial 600 msec of activity subsequent to critical word onset; this epoch might have been too short in duration to observe late positive waves, even if they were present in the data.1 In any event, i (...truncated)