Toward Stance-based Personas for Opinionated Dialogues

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 2625–2635 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 2625 Toward Stance-based Personas for Opinionated Dialogues Thomas Scialom⋆‡, Serra Sinem Tekiro˘glu⋄, Jacopo Staiano⋆, Marco Guerini⋄ ‡ Sorbonne Universit´e, CNRS, LIP6, F-75005 Paris, France ⋄ Fondazione Bruno Kessler, Via Sommarive 18, Povo, Trento, Italy ⋆ reciTAL, Paris, France {thomas,jacopo}@recital.ai {tekiroglu,guerini}@fbk.eu Abstract In the context of chit-chat dialogues it has been shown that endowing systems with a persona profile is important to produce more coherent and meaningful conversations. Still, the rep- resentation of such personas has thus far been limited to a fact-based representation (e.g. “I have two cats.”). We argue that these represen- tations remain superficial w.r.t. the complexity of human personality. In this work, we pro- pose to make a step forward and investigate stance-based persona, trying to grasp more profound characteristics, such as opinions, val- ues, and beliefs to drive language generation. To this end, we introduce a novel dataset al- lowing to explore different stance-based per- sona representations and their impact on claim generation, showing that they are able to grasp abstract and profound aspects of the author per- sona. 1 Introduction While chit-chat neural models have obtained im- pressive improvements in recent years, they are known to suffer from key limitations: they tend to lack specificity and to lose coherence as the con- versation unfolds, becoming less captivating. One explanation is that they do not have a consistent personality; for this reason, some approaches pro- posed to explicitly encode the persona via a small set of claims describing the characteristics of the agent, such as “My dad has a car dealership”, “I have two cats” (Zhang et al., 2018a). Such repre- sentations provide a fact-based background con- text useful to drive and ground the relevance of the conversational acts for the dialogue at hand, but with little generalization capability. Pushing this approach a step beyond, we thus investigate the construction of stance-based personas, in or- der to grasp profound and intimate characteristics – such as opinions, values, and beliefs. This could allow agents to sustain personal points of view both within the same conversation and across different discussions. In this paper, we make a first attempt at rep- resenting persona with different approaches and levels of abstraction. We build a new conversa- tional dataset from a social platform dedicated to argumentative interaction,1 and report experiments for stance-based personas with varying degrees of abstraction (e.g. implicit and explicit stance representation). Our experiments show that stance- based personas enable the agents to intervene, con- sistently with their representation, across topics unseen at training time. 2 Related Work Dialogue datasets and approaches Open- domain dialogue or chit-chat scenarios were considered as intractable problems until recently. The research community has made significant progress thanks to two factors: (i) large datasets and (ii) end-to-end neural approaches based on pre-trained language models. In particular, the idea of using large pre-trained language models finetuned on dialogue tasks has proved very effective (Zhang …

Content truncated to 500 words