A Statistical Approach to Parsing Ill-Formed Input

概要

論文の詳細を見る
This paper describes an algorithm used for parsing spontaneous speech and attempting to deal in an efficient way with three types of ill-formedness: false starts,filled pauses,and substitution.The appraroch proposed to recover from these frequenrly occuring errors is based on a stochastic language model called Bayesian Language Lnference(BLI).aFTER After a brief review of previous work on parsing ill-formed input,a general pverview of the BLI algorithm will be given,emphasizing characteristics that make it an appopiate tool capble of partially dealing with disfluencies and substitutions.This paper will focus on how that partial information can be reorganized in order to accrately parse sentences displaying these kinds of ill-formedness,leading to the descripton of a statisical processing system capable of analysing ill-formed input with mathematically sound consideration of full syntactic context.
一般社団法人情報処理学会の論文
1994-09-20