Friday, March 04 @ 3 pm, Harper 148

Looking Beyond: What Indirect Evidence Can Tell Us About Universal Grammar

Abstract:

One of the most controversial claims in linguistics is that children learning their native language face an induction problem: the data in their input are insufficient to identify the correct language knowledge as rapidly as children do. If this is true, then children must bring some helpful learning biases to the language acquisition problem – and the nature of these biases is often debated. For example: Are they innate or derived? Are they domain-specific or domain-general? Are they about what to learn or how to learn? Induction problems are often used to motivate innate, domain-specific knowledge about language (sometimes called Universal Grammar), but there are clearly other kinds of learning biases that might be used. In this talk, I look at the case study of English anaphoric one, an induction problem that has received considerable recent attention, particularly in the computational modeling literature. I will consider whether indirect evidence leveraged by a probabilistic learner from a broader data set could be effective, and what this tells us about the nature of the necessary learning biases. By doing so, I will be able to offer a concrete proposal about the content of Universal Grammar (for this linguistic phenomenon) as well as shed light on the acquisition trajectory for anaphoric one.