This dissertation examines the acquisition of a phenomenon that lies at the interface of logic and grammar. Polarity-sensitive items (PSIs), including negative polarity items (NPIs) such as English any, are often characterized by their restricted distribution, and analyzed in terms of their licensing condition (compare John doesn’t have any books, where any is licensed by negation, with the unlicensed *John has any books). Any moreover oscillates between NPI uses and so-called ‘free choice’ uses (John may choose any book). While a small handful of previous acquisition studies on English any have targeted children’s knowledge of the licensing condition, no previous study has systematically investigated children’s knowledge of the complex underlying semantics of PSIs like any, let alone the question of how children are to reconcile the dual nature of any. The series of studies in this dissertation presents novel evidence from experiments and corpora demonstrating that children have incredibly sophisticated semantic knowledge of any, which includes the ability to generate subdomain alternatives, to (pre-)exhaustify these alternatives, to perform domain widening, and to compute so-called free choice inferences. Yet samples of parental spontaneous production reveal very little evidence that could inform the learner as to how to carry out the semantic operations required for adult-like interpretation of any. I propose that the solution to this learning problem lies in innately constraining the hypothesis space of PSI types. Such a restricted hypothesis space is available to us in the form of a generative typology put forth in Chierchia (2013), an analysis that derives the possible classes of PSIs on the basis of free variation along two dimensions: the kind of alternatives that the target PSI activates, and the mode of exhaustification that factors the alternatives into meaning. On the assumption that these two dimensions are innately specified, only a finite set of PSI types can be generated; I discuss how the learner might use any’s unique distributional properties in the input to map the string any to the target PSI within the typology of restricted options.
Stored only when both papers are in the corpus — 81 citations exist upstream.