Publication
NAACL 2022
Conference paper

Learning as Conversation: Dialogue Systems Reinforced for Information Acquisition

Abstract

Obtaining information through reading passages could be inefficient and even challenging for a large portion of population. This paper proposes a novel task of facilitating the information acquisition process by replacing or augmenting passage reading with conversing with an intelligent dialogue agent. For this task, we propose an information acquisition-oriented dialogue system. By applying reinforced self-play, the dialogue system could be transferred to various domains without annotated dialogue dataset, and carry out conversations that are both informative and attentive to users. Experiments on three large public data corpora show that our system delivers knowledge-intensive and attentive conversations to help end-users obtain information with ease. Our code and datasets will be publicly available upon publication.

Date

10 Jul 2022

Publication

NAACL 2022

Authors

Topics

Share