N-gramを用いた対話文の言い直し表現の検出法

概要

論文の詳細を見る
This paper proposes a method to detect the strings of self-repairs occuring in spontaneous speech, which are assumed to be represented with strings of syllables obtained correctly by accoustic processing, using n-gram model and Markov models of syllables. This method os decomposed of the following two steps: The first step is to detect all of the pairs (celled as n-gram pair) with the same substrings appeared in dialogue sentence using n-gram extraction methods. The second step is to determine whether n-gram pair is self-repair or not. using Markov models of syllables. This method is experimentally evaluated to be effective using ATR database of dialogue.
福井大学工学部の論文