Memory mechanisms for Wh-Dependency formation and their implications for Islandhood

The goal of this chapter is to provide a functionally specified model of how long-distance dependencies are understood in real time. In particular its emphasis will be how long-distance dependencies interact with working memory (WM) and how that interaction might inform our understanding of the form or origin of grammatical constraints. By long-distance dependencies, I refer primarily to unbounded displacement dependencies, such as those exhibited in wh-questions, relative clauses, topicalizations, comparative clauses, clefts, etc. The grammar appears to place no limit on the hierarchical distance that such dependencies can span. The head of the dependency – the displaced constituent itself – can occur many clauses away from the tail of the dependency: a gap or pronoun. The example in (1) illustrates this basic observation with relativization: the bold-face constituent, cookies, is only one clause distant from its gap in (1a). In (1b) and (1c), however, it is two and three clauses away, respectively. a. The cookies [S that Phil bakes ____] contain pistachios. b. The cookies [S that Billy bothers Phil to bake ____] contain pistachios. c. The cookies [S that everyone knows that Billy bothers Phil to bake ____] contain pistachios. There has long been a mutual influence between the study of long-distance dependencies and the study of WM constraints in language processing (Miller and Chomsky 1963). Between its head and tail, a large number of processing events could occur which are essentially irrelevant to the long-distance dependency itself. At the syntactic level, these include events like establishing other thematic dependencies, interpreting modifier relations, checking case and agreement, or resolving anaphora. The processing system must thus have a means of retaining information about the head of the long-distance dependencies until its “tail” can be constructed, at which point it must be able to effectively recover the head. And it must do so without suspending the intervening processing events that span the dependency. These requirements describe exactly the sort of cognitive juggling act that working memory systems are proposed to accomplish (e.g., Baddeley 1986, Miyake and Shah 1999)