The Role of Personal Perspectives in Open-Domain Dialogue: Towards Enhanced Data Modelling and Long-term Memory
Year: 2023
Type: conference-paper
Abstract: As Conversational AI becomes integral to various digitally enhanced services, there is a growing need for fluent and personalized dialogue agents. Yet, recent state-of-the-art neural dialogue systems face challenges in transparency and memory precision, often leading to in- coherent and factually incorrect interactions. In this position paper, we argue that open-domain dialogue plays a crucial role in building effec- tive dialogue agents by fostering shared memories and capturing personal perspectives and social information. Specifically, we see the need for (i) datasets that capture long-term interaction; and (ii) a representation scheme that enables tracking of factoids and perspectives over time. We present an analysis of several existing open-domain dialogue datasets and highlight the richness and complexity of conversational data. We further present a data modeling scheme that represents and tracks factoids and perspectives over time and makes use of shared, long-term memories crucial for facilitating fluid and meaningful interaction. Our scheme is designed to improve long-term memory, increase transparency, and can serve as inspiration for designing neuro-symbolic dialogue agents capable of long-term social interaction.
Source: VU Research Portal
Authors: Krause, Lea, Baez Santamaria, Selene, Donatelli, Lucia, Vossen, Piek
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6987099221
