M. TOMOKIYO
Ambiguity labelling
for disambiguation experiments
Abstract:This report describes two month research stay, from 9 October 1995 to 8 December
1995, at GETA in Grenoble in the framework of the MIDDIM project. The project
was agreed between CNRS-IMAG-GETA and ATR-ITL, and cooperative research
officially started on June 1993. This project aims at studying ambiguities and their
disambiguation methods on multimodal system for spontaneously-spoken dialogues
as well as for written texts, for the Japanese, English and French languages.
The current work has extended previous one in two directions. First, we investigated
ambiguities in automatically-processed data by a parser and compared them with
ambiguities in non-processed data. Second, we conducted disambiguation experiments for Japanese dialogues on a man-machine interactive system called LIDIA-II.
The experiments satisfied us that user's intervention would be helpful for disambiguation in Japanese dialogue parsing as well as in English and French.
Keywords: ambiguity classification, ambiguity database, ambiguity kernel, ambiguity scope,
ambiguity labelling, ambiguity representation, disambiguation