The many meanings of an NPI: the view from artificial language learning

Cross-linguistically, Negative Polarity Items (NPIs) are often ambiguous, allowing additional, non--polarity sensitive uses. We document several common patterns of ambiguity and show that in each case, there is a systematic logical relation between the two forms. For example, in English and other languages, two interpretations commonly available to NPIs are each other's logical duals. We explore the hypothesis that these systematic ambiguities arise due to a bias in the learning process.Specifically, when a given word is restricted to a specific logical environment (such as the scope of negation), multiple equally `correct' analyses are available to a language learner. Learners might find some of these analyses more attractive than others, explaining why an ambiguity arises when the distributional constraint is relaxed.We employ an artificial grammar learning paradigm to test which interpretations of NPIs are preferred by learners when they encounter NPIs in a positive environment. Our results show strong evidence of a `wide-scope dual' interpretation, although specific properties of the training paradigm may influence these preferences.