Computational models of acquisition for Islands

Why look at language acquisition? Though it is not always directly stated, the debate at the center of this volume is in many ways driven by language acquisition considerations. Long-distance dependencies are themselves relatively complex, as they involve context-sensitive grammatical operations (e.g., wh-movement or slash-passing). The existence of context-sensitive operations alone increases the complexity of the hypothesis space of possible grammars that must be considered by children during the acquisition process. If island effects are indeed the result of grammatical constraints, then the hypothesis space increases yet again, as the grammar must also contain complex constraints on context-sensitive operations. A common hypothesis in the generative syntax literature is that this level of complexity (constraints on context-sensitive grammatical operations) cannot be learned directly from the input that children receive (i.e., this is a poverty of the stimulus problem). As such, many generative syntacticians have postulated the existence of innate domain-specific knowledge about the form that such constraints must take. In other words, the grammatical approach to island effects is often correlated with a nativist, or Universal Grammar (UG) based, view of language acquisition. In this way, a reductionist approach to island effects could be seen as a type of simplifying approach to the grammar, as it could eliminate the need for one set of innate constraints on the shape of human grammars. Because of this, it seems to us that discussions of “parsimony” and “simplification” in the reductionist literature either directly or indirectly concern the presumed problem that occurs during language acquisition. Given the amount of research that has been conducted on the debate between grammatical and reductionist approaches to island effects, it seems important at this stage to determine exactly what type of innate knowledge (if any) would be necessary for learning the grammatical constraints that give rise to island effects, given the input that children receive during language acquisition. Such an investigation will help determine exactly what is at stake in this debate. If grammatical island constraints cannot be learned from the input available to children without innate domain-specific knowledge (UG), then this debate has direct implications for the language acquisition process. However, if grammatical island constraints can be learned from the input available to children without UG-like knowledge, then this debate is simply one empirical question among the hundreds that must be answered in order to have a complete theory of language.