LingLang Lunch (4/19/2017): Kristine Yu (University of Massachusetts Amherst)

Kristine Yu’s research focuses on tone and intonation at all levels of language from the speech signal to grammar and human language processing, as well as ways to model them mathematically. For more information, her website is here.

Parsing with prosody: towards a computational model of prosodically-informed syntactic parsing in Samoan

It has long been clear that syntax determines certain aspects of prosody, and that prosody should therefore be part of the grammar influencing the action of the syntactic parser. Moreover, a body of theoretical work exists proposing general theories of how syntax determines certain aspects of prosody (but not others). However, what has remained unclear is how to bring prosody into the grammar to inform the parser. This is because of the many interacting, conditioning factors on prosody that may obscure the informativity of prosodic information for syntactic analysis. Our strategy for moving forward is to define and compare computational models of the interface which capture fundamental properties that distinguish proposed theories from one another—starting with case studies where syntax is clearly the primary determining factor for prosody. This talk is about a first case study on Samoan, which provides clear cases of prosodic events that are under the control of the syntax. We briefly review empirical evidence from original fieldwork about the syntax-prosody interface in Samoan. Then we introduce work towards a computational model of parsing in Samoan inspired by current ideas about the syntax-prosody interface in natural language: (1) a finite state optimality-theoretic transduction that generates licit parses of a sentence into phonological constituents, and (2) a syntactic parser that compares possible syntactic parses with the prosodic parses and computes optimal parses based on syntax-prosody interface constraints relating syntactic and phonological constituents.