Information structure licenses syntax: An ERP correlate of focus processing

Abstract:

The question of how contextual factors may influence syntactic processes in sentence comprehension is central to an understanding of language architecture.  While this issue has been addressed by several studies examining contextual influences on ambiguity resolution, this paper presents ERP evidence for the relevance of context to early syntactic integration processes in unambiguous sentences.

Unambiguously marked clause medial word order variations in German ('scrambling') give rise to processing difficulties, which, in ERP measures, are reflected in a centrally-distributed negativity between 300 and 450 ms post-onset of the non-canonically positioned NP (e.g., den Sänger in (1); cf. Rösler et al., 1998).

This effect cannot be reduced to the relative infrequency of an accusative NP in this position and therefore most likely reflects the local violation of grammatical canonicity principles (Schlesewsky et al., 2001).

The present study examined whether the 'scrambling negativity' may be influenced by the information structural status of the scrambled element.  Participants read question-answer pairs, in which the question was either neutral ("What happened?") or established the first NP of the ensuing answer as a topic (given) or a focus (new) constituent.  Answer sentences were presented in a phrase-by-phrase manner and were either subject-initial (canonical) or object-initial (scrambled).

The ERP recordings revealed the following.  In a neutral context, scrambled NPs gave rise to a centrally distributed negativity between 300 and 500 ms post-onset of the phrase (thus replicating previous results).  Whereas the ERP patterns for scrambled NPs that were topics did not differ from those for scrambled NPs in a neutral context, scrambled NPs that were focused gave rise to a broadly distributed positivity between 300 and 500 ms post onset of the phrase in comparison to the (neutral) canonical control condition.  A comparable positivity was also observable for focused initial subjects and focused second NPs (irrespective of their case marking).

Our results essentially show two things.  Firstly, the general appearance of a distinct ERP pattern for focused arguments indicates that constituents which are expressly focused by the context are processed differently from other constituents, most likely on account of the explicit expectations that the context generates with regard to these constituents.  Secondly, since for scrambled focus constituents, the 'focus positivity' superceded the negativity generally observed for scrambled arguments, we may conclude that information structural considerations derived from the context may influence even early phases of structural integration by licensing non-canonical word orders.