Do all lexical categories carry equal weight in generating predictions about upcoming sentence material?

On-line sentence comprehension is incremental and continuous (rather than, for example, individual items being stored in memory and analysed in chunks corresponding to syntactic units; Delong et al. 2005). The original findings by Kutas and Hillyard (1980, 1984) established that ERPs (specifically, the N400 response) are sensitive to context: for instance, the final item in He spread the warm bread with socks elicits a greater negativity than a context-congruous item like butter. This has been interpreted as evidence that sentence processing depends largely on prediction for its efficiency. To date, most work on prediction has focused on noun prediction. Yet is it correct to assume that the prediction process is the same for all word classes? Verbs play a key part in determining the syntactic structure of a sentence. Does the process of prediction differ between verbs and nouns? If so, in what ways? Specifically, are there differential time-courses in the two types of prediction? This issue lies at the heart of a current debate in language processing studies: is syntactic information accessed and evaluated first, or does input from semantics take precedence in sentence comprehension?
This study (a collaboration with Colin Phillips) aims to investigate this question by contrasting N400 responses to sentence-final items in German subordinate clauses with transitive verbs. Depending on the choice of complementiser, such clauses can have canonical SVO or non-canonical SOV word order. The research question outlined above will be investigated by contrasting fully grammatical sentences which vary systematically a. in word order and b. in the Cloze probability of the final item. The N400 response, which has been reliably shown (see Kutas and Hillyard (1980, 1984), Delong (2005) and many others) to index Cloze probability, will serve as the dependent variable in this investigation.