Discourse-linking and long-distance syntactic dependency formation in real-time

A self-paced reading experiment examined the processing of long-distance wh-dependencies. Some wh-dependencies were D(iscourse)-linked (i.e. involved a lexically-specified wh-phrase like which horse) and others involved a non D-linked wh-phrase like who. Both kinds of wh-phrase show evidence of being represented at available intermediate positions within the long-distance dependencies. It is argued that this is inconsistent with Pesetsky (1987) whose hypothesis claimed that D-linked dependencies are formed via binding, and as such are predicted to be able to bypass intermediate reactivations. At points in the time-course where the potential for the violation of dependency-formation constraints becomes apparent, a D-linked antecedent seems to benefit the parse. This motivated the idea that the reduction in constraint-sensitivity which D-linking seems to elicit could be the result of time-locked “stabiliser” effects during moments of complex processing. Finally, the data may indicate a behavioural difference between intermediate and verbal antecedent reactivation during processing.