Abstract

Learning the surface structure of wh-questions in English and French with a non-parametric Bayesian model

Authors
  • An Nguyen (Johns Hopkins University)
  • Colin Wilson

Abstract

The overt structure of wh-questions varies across and within languages. How does a child learn the number of wh-question types that are present in her language and the surface properties of each type? We propose a non-parametric Bayesian model of this aspect of language acquisition, focusing on discrete morphosyntactic properties of questions such as displacement and continuous prosodic properties such as wh-word duration, and apply it to data based on child-directed speech in English and French. The model successfully infers that English has fewer wh-question types than French, identifies the properties of the main question types in each language, and achieves reasonable classification accuracy on naturalistic test utterances. Non-parametric Bayesian inference is a promising method for addressing cross-linguistic and language-internal syntactic variation.

Keywords: wh-question, non-parametric, bayesian model

How to Cite:

Nguyen, A. & Wilson, C., (2021) “Learning the surface structure of wh-questions in English and French with a non-parametric Bayesian model”, Society for Computation in Linguistics 4(1), 403-405. doi: https://doi.org/10.7275/kp97-zx82

Downloads:
Download PDF

139 Views

39 Downloads

Published on
01 Jan 2021