Improving Factual Consistency Between a Response and Persona Facts

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 549–562 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 549 Improving Factual Consistency Between a Response and Persona Facts Mohsen Mesgar Edwin Simpson⇤ Iryna Gurevych Ubiquitous Knowledge Processing Lab (UKP Lab) Department of Computer Science, Technical University of Darmstadt https://www.ukp.tu-darmstadt.de Abstract Neural models for response generation pro- duce responses that are semantically plausible but not necessarily factually consistent with facts describing the speaker’s persona. These models are trained with fully supervised learn- ing where the objective function barely cap- tures factual consistency. We propose to fine- tune these models by reinforcement learning and an efficient reward function that explicitly captures the consistency between a response and persona facts as well as semantic plausi- bility1. Our automatic and human evaluations on the PersonaChat corpus confirm that our ap- proach increases the rate of responses that are factually consistent with persona facts over its supervised counterpart while retaining the lan- guage quality of responses. 1 Introduction Response generation models should ideally gener- ate an appropriate response to a given context con- sisting of utterances previously exchanged between dialogue partners and facts describing the speak- ers’ persona. These models have applications in developing dialogue systems as user interfaces for digital assistants (Bobrow et al., 1977) and also in asynchronous interactions in social media in which speakers define themselves by their profiles. In this work, we focus on the aspects of persona that can be captured by a set of factual statements, a.k.a., profiles. Table 1 illustrates the persona of the speaker who should respond to the given message. The first response is topically coherent with the message and also linguistically fluent (or in general, semantically plausible) but factually inconsistent, unlike the second response, with the second fact ⇤* Now at the Intelligent Systems Lab, Dept. of Computer Science, University of Bristol. 1https://github.com/UKPLab/ EACL21-personalized-conversational-system Persona fact 1: i hate my job fact 2: i ’ m 40 years old fact 3: i work as a car salesman fact 4: my wife spends all my money fact 5: i am planning on getting a divorce Dialogue History message: hi , want to be friends? Generated Responses: inconsistent: i ’ d love to be friends . i ’ m 50 years old consistent: sure , i am 40 , i can tell you about myself Table 1: A speaker’s persona, dialogue history, one in- consistent, and one consistent possible response. in the speaker’s persona. We aim to improve the response quality in terms of its factual consistency with facts about the given speaker’s persona while retaining its semantic plausibility. Recent approaches to this problem (Zhang et al., 2018; Dinan et al., 2019; Wolf et al., 2019) gen- erate a response conditioned on persona facts and dialogue history and then use human-generated re- sponses as demonstrations to train their models by fully supervised learning (SL). While this strategy has led to markedly improved performance, there is still a misalignment …

Content truncated to 500 words