Pierre Hudry, Yasuharu Den
A Statistical Approach to
Parsing Ill-Formed Input
Abstract:In this report we argue for the use of statistical models in the handling of ill-formed
input, coupled to a global consideration of syntactic structure. We describe an algorithm based on an already existing language model, Bayesian Language Inference.
The algorithm was implemented and tested on artificially altered tagged data taken
from the ATR Dialogue Database. Partial conclusions are drawn from these results
focusing on remaining problems and potential future developments.