Establishing the informational requirements for modelling open domain dialogue and prototyping a retrieval open domain dialogue system

Meier, Trent; Pimenidis, Elias

Abstract

Open domain dialogue systems aim to coherently respond to users over long conversations through multiple conversational turns. Modelling open domain dialogue is challenging as both the syntactic and semantic features of language play a role in response formation. As similarity to human dialogue has been considered the goal of open domain dialogue systems, this paper takes the view that human linguistic reasoning research can be informative to the requirement engineering process of modelling open domain dialogue. Through a review of linguistic reasoning research and modern approaches in open domain dialogue systems, the authors present informational hypotheses impacting the modelling of open domain dialogue systems. Furthermore, this paper discusses the design and testing of an open domain dialogue system presenting response BLEU-1 scores of 35.41% based on the DailyDialogue Dataset.

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 13th International Conference on Computational Collective Intelligence
• Start Date: Sep 29, 2021
• End Date: Oct 1, 2021
• Acceptance Date: Aug 15, 2021
• Online Publication Date: Sep 30, 2021
• Publication Date: Sep 30, 2021
• Deposit Date: Sep 30, 2021
• Publicly Available Date: Oct 1, 2022
• Publisher: Springer (part of Springer Nature)
• Volume: 12876
• Pages: 655-667
• Series Title: Lecture Notes in Artificial Intelligence
• Edition: 1
• Book Title: Computational Collective Intelligence
• Chapter Number: 49
• ISBN: 9783030880804
• DOI: https://doi.org/10.1007/978-3-030-88081-1_49
• Keywords: Natural language processing (NLP), Open domain dialogue modelling, DailyDialogue dataset
• Public URL: https://uwe-repository.worktribe.com/output/7864605
• Publisher URL: https://link.springer.com/book/10.1007/978-3-030-88081-1