[1N2-J-9] Natural language processing, information retrieval: dialogue

[1N2-J-9-05] Empirical study of conditional interactions based on longitudinal survey data

~ From the phonetic approach of natural spoken sentences ~

〇Hiromitsu Ota¹ (1. The Open University of Japan)

Keywords:conversation analysis, Conditional interaction analysis, Bayesian statistics, Ethnomethodology, pragmatics

In this paper, we consider the first and second oral tests (interviews) of OPI as the pre- and posterior events and the utterances of the second and third times and up to the fifth time as a pre- and posterior event, and use Bayesian approach to omit utterances In recent years, just as Bayesian statistics became mainstream from frequency statistics, there are many cases where mutual acts can be explained by conditional interactions in fact. This is because posterior events will occur based on prior events.In the longitudinal survey of the Japanese learners' conversation database provided by the National Institute of Japanese Language Studies, the same person was orally examined and pursued up to five times The unique point of this discussion is that although the utterance chain is completed at once, it is inherently different, but despite being completed in advance (first time) and after (second time), " Yes Yes Yes "or" Oh yeah yeah "It is a point that it turns out to be a casual expression, and it is a change of polite way of thinking to be seen with a strong consent such as" It repeats the same thing after the second time rude like I do not remember the last time, etc. Etc. Here, non-verbal fillers and the like are practical and applicable to interactive systems such as artificial intelligence (AI) and robot.