Detection of Agreement and Disagreement: An investigation of linguistic coordination and conversational features

Abstract

The focus of this thesis is detection of agreement and disagreement in multiparty conversations using existing transcripts from the ICSI corpus. We use an unsupervised lexicon-based method to create our baseline and then follow a supervised approach to study the effect on performance of different feature sets. The feature sets we are interested in are the following: a. lexicosyntactic, b. Dialog act tag-based features, c. Linguistic style coordination features and d. Conversational features (baseline, individual and non-individual ones). The results enabled us to study the presence of coordination in agreeing and disagreeing statements and showed that the performance can improve when adding further features on top of the lexicosyntactic ones. Additionally, we saw that non-individual conversational features can improve the performance of agreement detection while individual conversational ones appear to have a similar effect on disagreement detection.