Processing relative clauses varying on syntactic and semantic dimensions: An analysis with event-related potentials

K. STEINHAUERand A. D. FRIEDERICI 0 1 2 3 4 0 Max-Planck-Institute for Cognitive Neuroscience , Leipzig, Germany 1 H. SCHRIEFERS The Nijmegen Institute for Cognition and Information , Nijmegen, The Netherlands 2 A. MECKLINGER Max-Planck-Institutefor Cognitive Neuroscience , Leipzig, Germany 3 This work was supported by grants from the Alfried Krupp von Bohlen und Halbach Foundation and the Deutsche Forschungs gemeinschaft (Fr 519/12-2). We wish to thank Erdmut Pfeifer for his support in software production for the present experiment. We are grateful to Lee Osterhout and an anonymous reviewer for helpful com ments on an earlier draft ofthis article and to Douglas Saddy for fruit ful discussions. Requests for reprints can be sent to A. Mecklinger, Max-Planck-Institute for Cognitive Neuroscience , Inselstrasse 22, 04103 Leipzig, Germany 4 MECKLINGER, SCHRIEFERS, STEINHAUER, AND FRIEDERICI Event-related potentials were used to study how parsing of German relative clauses is influenced by semantic information, Subjects read well-formed sentences containing either a subject or an object relative clause and answered questions concerning the thematic roles expressed in those sentences. Half of the sentences contained past participles that on grounds of semantic plausibility biased either a subject or an object relative reading; the other half contained past participles that provided no semantic information favoring either reading. The past participle elicited an N400 component, larger in amplitude for neutral than for semantically biased verbs, but this occurred only in the case of subject relative clauses. More specific effects were obtained only for a subgroup of subjects, when these were grouped into fast and slow comprehenders on the basis of their questionanswering reaction times. Fast comprehenders showed larger N400 amplitudes for neutral than for semantically biased past participles in general and larger N400s for the latter when there was a bias for an object relative reading as opposed to a subject relative reading. Syntactic ambiguity resolution, indicated by an auxiliary in sentence final position, was associated in this subgroup with a positive component (P345), larger in amplitude for auxiliaries indicating an object relative reading than for those indicating a subject relative reading. The latter component was independent of semantically biasing information given by a preceding past participle. Implications of these findings for models of language comprehension are considered. - and integrated to form a final representation of a sen tence (e.g., Altmann, 1989; Clifton, Speer, & Abney, 1991; Friederici, 1985; Marslen-Wilson & Tyler, 1980; Mitchell, Corley, & Garnham, 1992). The best known model postulating priority of syntac tic processes during language comprehension is the "garden-path" model proposed by Frazier and col leagues (e.g., Frazier, 1987a, 1987b; Frazier & Rayner, 1982). According to the garden-path model, semantic and pragmatic information cannot influence the initial syntactic analysis of the sentence. Rather, this first syn tactic analysis is determined exclusively by a set of syn tactic principles (cf. Clifton et al., 1991; Frazier, 1978, 1987b). Only in a second stage is this initial "preferred" syntactic representation evaluated with respect to other information sources, such as semantic and pragmatic in formation. If later information is inconsistent with the initially favored syntactic structure, a syntactic reanaly sis will be initiated (see, e.g., Clifton & Ferreira, 1987; Rayner & Frazier, 1987). In contrast to the garden-path model, "lexical-entry-driven" models claim that informa tion in the lexical entry of a verb-that is, its argument structure-is consulted to guide the initial parse (Shapiro, Nagel, & Levine, 1993; Trueswell, Tanenhaus, & Kello, 1993). "Interactive" models suggest that all information sources-that is, semantic/pragmatic and syntactic-i-con tinuously interact so that the subject can arrive at the most plausible analysis of a sentence (Marslen-Wilson & Tyler, 1980; Taraban & McClelland, 1988). In the study reported here, we examined the nature and timing of mental processes underlying language com prehension by recording event-related brain potentials (ERPs) elicited by the presentation of specific parts of a sentence. The ERP is a transient series of recordable voltage oscillations in the brain that occur in response to discrete events. ERPs can provide a record of a subject's response to every word within a visually presented sen tence without the intrusive requirement of a manual or verbal response. Moreover, unlike reaction time (RT) measures, ERPs are multidimensional: single ERP com ponents can be distinguished by their latency, amplitude, polarity, and scalp topography. Given that functionally distinct processes are modulated by neuronally distinct brain systems, we would expect their ERP correlates to be different, too. In other words, if, in the course of forming a representation of a linguistic input, syntactic and semantic processes yield different representational levels, the ERP components associated with these pro cesses should be distinct with respect to latency, scalp topography, and/or polarity. Recent studies in which ERP measures of language processing have been used suggest that the N400, a neg ative component peaking around 400 msec after stimu lus onset that is largest over posterior scalp regions, re flects the ease with which a given lexical element can be integrated in the preceding context. It appears that the amplitude of the N400 is a function of the semantic fit between the target word eliciting the N400 and a prior context (Fischler, Bloom, Childers, Roucos, & Perry, 1983; Fischler & Raney, 1991; Kutas & Hillyard, 1984; Mecklinger, Kramer, & Strayer, 1992). Kutas and Hill yard (1984) found that the amplitude of the N400 elicited by a sentence final word was inversely related to the subjective predictability (i.e., the cloze probability) of this word. It was suggested that the N400 reflects the amount of semantic or lexical priming or constraints from the preceding context for a given target (Fischler et al., 1983; Holcomb & Neville, 1990; Kutas & Van Petten, 1988; Van Petten & Kutas, 1991). While the role of the N400 as a measure of semantic/ lexical integration processes is well established, there is an increasing number of studies identifying ERPs that are associated with syntactic processes during language comprehension. The results ofthese studies are twofold. First, they have shown that a large variety of syntactic anomalies elicit a parietally focused positive component in the ERP (Hagoort & Brown, 1994; Hagoort, Brown, & Groothusen, 1993; Neville, Nicol, Barss, Forster, & Gar rett, 1991; Osterhout, Holcomb, & Swinney, 1994). Sec ond, some of these anomalies additionally produce a negativity which is dominant over the left hemisphere (Friederici, Pfeifer, & Hahne, 1993; Miinte (...truncated)