Conversation analysis structured dialogue for multi-domain dialogue management

Duran, Nathan; Battle, Steve

Abstract

Dialogue state tracking is a vital component of task-oriented dialogue systems. Often, dialogue states are constrained by their target domains entity's, slots and values. Adding new domains and knowledge may require laborious hand-crafting or retraining using new corpora and training data. This makes the development of multi-domain dialogue systems a considerable challenge. To address this problem, we propose a method of structuring dialogue that is independent of domain, and closely related to constructs defined by the sociological research of Conversation Analysis, a study of human interaction in conversations. First, we summarise the applicable theories of Adjacency Pairs and Dialogue Acts and their relevance to dialogue systems. We then introduce a schema for structuring dialogue and an accompanying corpus, that utilise the Conversation Analysis inspired constructs, and discuss their potential advantages in moving towards domain agnostic dialogue management.

• Presentation Conference Type: Conference Paper (unpublished)
• Conference Name: The International Workshop on Dialogue, Explanation and Argumentation in Human-Agent Interaction (DEXAHAI)
• Start Date: Dec 15, 2018
• End Date: Dec 15, 2018
• Deposit Date: May 14, 2021
• Publicly Available Date: May 14, 2021
• Keywords: Conversation Analysis; Adjacency Pairs; Dialogue Act; Dialogue Management; State Tracking
• Public URL: https://uwe-repository.worktribe.com/output/5347221
• Publisher URL: https://sites.google.com/view/dexahai-18/home