Modelling Multi-issue Bargaining Dialogues: Data Collection, Annotation Design and Corpus

Modelling Multi-Issue Bargaining Dialogues: Data Collection, Annotation Design and Corpus Volha Petukhova1, Christopher Stevens2, Harmen de Weerd2, Niels Taatgen2, Fokie Cnossen2, Andrei Malchanau1 1Saarland University, Spoken Language Systems, Saarbr¨ucken, Germany 2University of Groningen, Department of Artificial Intelligence, The Netherlands 1{v.petukhova,andrei.malchanau}@lsv.uni-saarland.de; 2{c.a.stevens,h.a.de.weerd,n.a.taatgen,f.cnossen}@rug.nl Abstract The paper describes experimental dialogue data collection activities, as well semantically annotated corpus creation undertaken within EU-funded METALOGUE project. The project aims to develop a dialogue system with flexible dialogue management to enable systems adaptive, reactive, interactive and proactive dialogue behaviour in setting goals, choosing appropriate strategies and monitoring numerous parallel interpretation and management processes. To achieve these goals negotiation (or more precisely multi-issue bargaining) scenario has been considered as the specific setting and application domain. The dialogue corpus forms the basis for the design of task and interaction models of participants negotiation behaviour, and subsequently for dialogue system development which would be capable to replace one of the negotiators. The METALOGUE corpus will be released to the community for research purposes. Keywords: negotiation corpus collection, dialogue act annotation, ISO 24617-2 dialogue act annotation scheme extension, dia- logue modelling 1. Introduction Recently, we have been witnessing the steady increasing demand for human-computer systems and interfaces of var- ious complexity. The current research efforts in human- computer system design diverge more and more from tra- ditional paradigms to modelling of two-party task-oriented systems like information-seeking dialogues. The research community is currently targeting more flexible, adaptable, open-domain dialogue systems driven by modelling natu- ral human multimodal behaviour. Advances are also be- ing made in modelling and managing multi-party interac- tions, e.g. for meetings or multi-player games. Existing ap- proaches developed for two-party dialogue have undergone certain changes. For instance, it has been acknowledged that assumptions that conversational agents act fully ratio- nally and cooperatively do not hold in many conversational settings, see e.g. (Traum et al., 2008) and (Asher and Quin- ley, 2011). This is particularly true in competitive games, debates, and negotiations where participants do not have fully aligned preferences and do not adopt shared intentions or goals. In this paper we focus on modelling negotiations, more precisely multi-issue bargaining dialogues. Much good work has been done on simple, well-structured negotiations - interactions among a few parties with fixed interests and alternatives, see (Georgila and Traum, 2011); (Efstathiou and Lemon, 2015) and (DeVault et al., 2015). In many real-life negotiations, parties negotiate over not one but multiple issues. Moreover, negotiators bargain- ing over one or multiple issues today may, and in real life most certainly will, come back to the negotiation table. So, there may be delays in making complete agreements, and previously reached agreements can be cancelled. In this paper we discuss multi-issue repetitive bargaining inter- actions collection and analysis as important steps towards computational modelling of such conversations. The paper is structured as follows. Section 2 discusses the application task domain, specifying participants roles and goals, and possible interactive phenomena to be encountered. Section 3 presents the designed scenario, interfaces and data col- lection procedure. In Section …

Content truncated to 500 words