April 26, 2005
2004-2005/077
Yale linguist at UIUC for talks May 5 & 6
Charles Yang, assistant professor of linguistics at Yale University, will present two lectures at the University of Illinois, Thursday, May 5 and Friday, May 6. Both talks are free and open to the public.
Yang will discuss “On Storage and Computation” on May 5 at 2 p.m. in Room G-13, Foreign Languages Building (FLB), 707 S. Mathews, U.
On May 6 Professor Yang will address “Word Segmentation: Quick but not Dirty,” in Room G-13, FLB, at noon.
Both lectures are sponsored by the UIUC Department of Linguistics.
In his first talk, on Thursday, Wang will look at a current trend in linguistics and psychology that returns to the construction-specific approach to language that had been abandoned since the late 1960s. According to this perspective, says, Wang, both grammar and learning are “conceived as piecemeal accumulations and generalizations of specific instances of linguistic data, resulting in the abandonment of the ‘core’ components of the linguistic system—such as the syntactic parameters—that have overarching effects over language.”
There are strong theoretical arguments against this approach, but the empirical issues raised by the construction-based approach are nevertheless interesting, according to Wang. It is his belief that the responsible move is to articulate a theory that retains the core but at the same time “accommodates the more idiosyncratic cases—or ‘nuts’, to use Culicover's term—that lie somewhat beyond the core system.”
Wang argues that the nuts challenge is reducible to the traditional problem of storage and computation in the grammatical system. “From a learning perspective,” he notes, “the child must extract regularities about her language despite a certain number of exceptions.”
Drawing on a variety of psycholinguistic evidence, Wang sketches a mathematical model of how the balance between storage and computation is maintained. Empirical consequences in morphological learning, processing, and change will be discussed.
In Friday’s lecture Wang turns his attention to statistical learning. According to the speaker, despite the enthusiasm for statistical learning (Saffran, et. al., 1996) in language acquisition research, there has been no empirical test for the effectiveness of statistical learning in a realistic setting of language acquisition—or whether statistical learning is used at all.
Wang will survey a number of segmentation strategies including statistical learning and put them to test in a series of computational models using child-directed input from the CHILDES database. Wang shows that statistical learning does not scale up to the segmentation of actual languages (e.g. English). In addition, he asserts, if statistical learning is relevant to language learning, “it must be constrained by what appears to be innate and domain-specific constraints on phonological structures.”
He will also discuss the proper treatment of other segmentation strategies along with the relevance of corpus studies to language acquisition research.
Wang is the author of “Knowledge and Learning in Natural Language” (2002, Oxford), which deals with grammar and word learning, and language change.
|
