End-to-end dialogue structure parsing on multi-floor dialogue based on multi-task learning

Abstract

A multi-floor dialogue consists of multiple sets of dialogue participants, each conversing within their own floor. In the multi-floor dialogue, at least one multi-communicating member who is a participant of multiple floors and coordinates each to achieve a shared dialogue goal. The structure of such dialogues can be complex, involving intentional structure and relations that are within or across floors. In this study, We proposed a neural dialogue structure parser with an attention mechanism that applies multi-task learning to automatically identify the dialogue structure of multi-floor dialogues in a collaborative robot navigation domain. Furthermore, we propose to use dialogue response prediction as an auxiliary objective of the multi-floor dialogue structure parser to enhance the consistency of the multi-floor dialogue structure parsing. Our experimental results show that our proposed model improved the dialogue structure parsing performance more than conventional models in multi-floor dialogue.