Title: “Inductive Bias in Language Acquisition: UG vs. Deep Learning"
Abstract: Generative approaches to language acquisition emphasize the need for language-specific inductive bias, Universal Grammar (UG), to guide learners in the face of limited data. In contrast, computational models of language learning, particularly those rooted in contemporary neural network models, have achieved high levels of performance on practical NLP tasks, largely without the imposition of any such bias. While UG-based approaches have led to important insights into the stages and processes underlying language acquisition, they have not yielded a concrete, mechanistic model of the process by which language is learned. At the same time, practical computational models have not been widely tested with respect to their ability to extract linguistically significant generalizations from training data. As a result the ability of such systems to face the challenges identified in the generative tradition remains unproven. In this talk, I will review several experiments that explore the ability of network models to take on such challenges. Looking at question formation and subject-verb agreement, we find that there is considerable variety in the degree to which network architectures are capable of learning significant grammatical generalizations through gradient descent learning, suggesting that the architectures themselves may be able to impose some of the necessary bias that is often assumed to motivate the need for UG. Inadequacies remain in the generalizations acquired, however, which points to the need for hybrid models that integrate language specific information into network models.