The neurophysiological basis of word order variations in German

Abstract

This paper aims to dissociate grammatical and general cognitive (e.g., working-memory based) accounts of the processing costs elicited by word order variations in German. To this end, we present a study using event-related brain potentials (ERPs), in which dislocated arguments were instantiated either by non-pronominal or by pronominal noun phrases. This manipulation allows for a dissociation of the two competing accounts, since only a dislocation of non-pronominal arguments gives rise to a non-canonical structure. The results show that (a) in sentences with non-pronominal arguments, the determiner of a non-canonical noun phrase elicited a broadly distributed negativity, and (b) in sentences with pronominal arguments, no differences were observed for object- vs. subject-initial word orders. These findings show that the human parser is sensitive to fine-grained grammatical regularities. We therefore argue that the negativity is a reflection of a local syntactic mismatch, rather than of an increase in working-memory load.

Introduction

A number of studies have shown that event-related brain potentials (ERPs) are a sensitive and powerful tool for the description of on-line processes in human parsing. Whereas earlier studies using this method were primarily concerned with the question of what happens when the human parser is confronted with a local ungrammaticality (e.g., Coulson, King, & Kutas, 1998; Friederici, Hahne, & Mecklinger, 1996; Münte, Heinze, & Mangun, 1993; Neville, Nicol, Barss, Forster, & Garrett, 1991; Rösler, Friederici, Pütz, & Hahne, 1993), recent research, especially in languages with a relatively free word order like German, has focused on the relation between syntactic structure and working memory in sentences with dislocated elements (e.g. Fiebach, Schlesewsky, & Friederici, 2001; Friederici, 2002; Friederici, Schlesewsky, & Fiebach, in press; Friederici, Steinhauer, Mecklinger, & Meyer, 1998; King & Kutas, 1995; Rösler, Pechmann, Streb, Röder, & Hennighausen, 1998; Vos, Gunter, Schriefers, & Friederici, 2001).

In both types of studies, left-anterior negative ERP components were observed, although the interpretations of these effects have tended to be quite divergent, partly on account of the different initial research questions involved. The primary goal of this paper is to examine whether the two classes of negativities reflect similar underlying processing mechanisms. To this end, we will firstly examine some of the crucial previous findings in more detail.

In unambiguous sentences such as (1) (Fiebach et al., 2001; for details of the ERP study see Fiebach, Schlesewsky, & Friederici, 2002) found a sustained left-anterior negativity starting at the prepositional phrase following the initial wh-pronoun when this pronoun indicated an object-first question (1b in comparison to 1a).

• Peter wusste nicht, wer am Sonntag nach dem Unfall den Arzt abgeholt hat. Peter did not know who_NOM on Sunday after the accident the_ACC doctor picked up has

• ‘Peter didn’t know who picked up the doctor on Sunday after the accident.’

• Peter wusste nicht, wen am Sonntag nach dem Unfall der Arzt abgeholt hat. Peter did not know who_ACC on Sunday after the accident the_NOM doctor picked up has

• ‘Peter didn’t know who the doctor picked up on Sunday after the accident.’

The authors argue that this type of negativity, which has also been observed for the processing of similar constructions in English (Phillips, Kazanina, Wong, & Ellis, 2001), reflects the additional memory resources required in the more complex object-first question. In such sentences, the initial dislocated object pronoun must be held in working memory until its trace position can be structurally postulated, thereby allowing the element to be integrated into the current partial structure. On the second argument der/den Arzt, a late positivity shows that the structural integration of the object requires additional processing cost (cf. Kaan, Harris, Gibson, & Holcomb, 2000).

Besides word order variations induced by wh-movement, languages such as German allow a further type of non-canonical argument order.¹ The dislocation operation in question, which is referred to as “scrambling,” leads to sentences such as (2).

• Dann hat dem Sohn der Vater den Schnuller gegeben.

• Then has the_DAT son the_NOM father the_ACC pacifier given

• ‘Then the father gave the pacifier to the son.’

In contrast to wh-movement, the dislocated argument does not move to a sentence initial position but rather to a position preceding the subject clause-medially. Whereas there are a number of theoretical arguments for distinguishing between both types of dislocation (cf. Friederici et al., in press), experimental results also appear to support such a distinction (e.g., Bader & Meng, 1999).

In a recent ERP study, Rösler et al. (1998) investigated the processing of case marked scrambled noun phrases in German. The authors set out to study case marked double object sentences such as (2). The ERPs revealed a number of different effects with respect to polarity, topography and timing. Relevant for the issue under discussion is the observation of a negativity with a left-anterior focus on the determiner of a scrambled (and therefore, non-canonical) argument (e.g., dem in 2). Rösler and colleagues argue that this effect is related to working memory load insofar as an argument that cannot be immediately assigned to its canonical position must be held in working memory. Because the negativity is restricted to the determiner and not visible on the following noun, this effect does not reflect storage as such but rather “a manifestation of some preparatory processing step which enables storage of the forthcoming noun” (Rösler et al., 1998, p. 171). Furthermore, the authors consider it unlikely that the negativity reflects a “purely” syntactic problem because the structural differences in scrambled sentences also lead to differences in information structure, which is closely related to semantic/pragmatic functions of language.

Relating the negativity reported by Rösler and colleagues, which these authors refer to as a left anterior negativity (LAN), to a working memory effect gives rise to several problems. Firstly, there are a number of studies showing that LAN effects reflect first pass parsing processes and are induced by syntactic violations, such as morphosyntactic or subcategorization violations (Coulson et al., 1998; Friederici, 1995, Friederici, 1999; Friederici & Frisch, 2000; Gunter, Friederici, & Hahne, 1999; Gunter, Friederici, & Schriefers, 2000; Münte, Szentkuti, Wieringa, Matzke, & Johannes, 1997; Neville et al., 1991; Rösler et al., 1993).² Secondly, it is difficult to envisage why the storage of an object wh-element (see above) should give rise to sustained memory effects while the storage of a scrambled object is local. Furthermore, recent experimental evidence has shown that the negativity observed by Rösler et al. also surfaces when the determiner and noun of a scrambled NP are presented together (Bornkessel, Schlesewsky, & Friederici, 2002b), thus indicating that a characterization of this negativity in terms of “preparatory” processing costs pertaining to the forthcoming object noun is questionable.

In view of the above discussion, it appears that the negativity reported by Rösler et al. (1998) may alternatively be characterized as a linguistic rather than a working-memory-related phenomenon. From this perspective, the component would result from a local syntactic mismatch between the features of the predicted subject position and the object encountered in this position (cf. Friederici et al., in press), thereby reflecting the violation of grammatical canonicity principles. In order to test this hypothesis, the present study was concerned with a more detailed investigation of the nature of scrambling in German.