Direct association and sentence processing - a reply

Papers by Gorrell and by Gibson and Hickok in this issue question Pickering and Barry's (1991) arguments against empty categories in sentence processing. This reply disputes Gorrell's claims that Pickering and Barry's (PB's) interpretation of the data is inadequate, and, in agreement with Gibson and Hickok, reinforces the arguments that the gap location is irrelevant to the formation of an unbounded dependency. However, it accepts that it is possible to provide ''predictive'' accounts where a gap can be postulated before the gap location is reached. But any such proposal is likely to be more complicated and less natural than PB's, since there is no processing reason to assume the reality of gaps. In order to demonstrate the naturalness of the gap-free account, this reply next shows how PB's proposals can be modelled in an approach to incremental processing based on a version of flexible categorial grammar, and how the differential complexity of various recursive constructions can be accounted for. Finally, the paper considers the implications of PB's proposals for various processing strategies, and suggests that PB's conclusions are reinforced by Bock, Loebell and Morey's (1992) evidence from language production. It concludes that processing evidence gives us no reason to postulate empty categories.