Addressee and Response Selection for Multi-Party Conversation

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2133–2143, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Addressee and Response Selection for Multi-Party Conversation Hiroki Ouchi Nara Institute of Science and Technology ouchi.hiroki.nt6@is.naist.jp Yuta Tsuboi IBM Research - Tokyo yutat@jp.ibm.com Abstract To create conversational systems working in actual situations, it is crucial to assume that they interact with multiple agents. In this work, we tackle addressee and response se- lection for multi-party conversation, in which systems are expected to select whom they ad- dress as well as what they say. The key chal- lenge of this task is to jointly model who is talking about what in a previous context. For the joint modeling, we propose two model- ing frameworks: 1) static modeling and 2) dy- namic modeling. To show benchmark results of our frameworks, we created a multi-party conversation corpus. Our experiments on the dataset show that the recurrent neural network based models of our frameworks robustly pre- dict addressees and responses in conversations with a large number of agents. 1 Introduction Short text conversation (STC) has been gaining pop- ularity: given an input message, predict an appropri- ate response in a single-round, two-party conversa- tion (Wang et al., 2013; Shang et al., 2015). Model- ing STC is simpler than modeling a complete con- versation, but instantly helps applications such as chat-bots and automatic short-message replies (Ji et al., 2014). Beyond two-party conversations, there is also a need for modeling multi-party conversation, a form of conversation with several interlocutors convers- ing with each other (Traum, 2003; Dignum and Vreeswijk, 2003; Uthus and Aha, 2013). For exam- ple, in the Ubuntu Internet Relay Chat (IRC), sev- Figure 1: Addressee and response selection for multi-party conversation. A SYSTEM is required to select an appropriate addressee from the interlocutors in the conversational context and an appropriate response from the fixed set of candidates. eral users cooperate to find a solution for a techni- cal issue contributed by another user. Each agent might have one part of the solution, and these pieces have to be combined through conversation in order to come up with the whole solution. A unique issue of such multi-party conversations is addressing, a behavior whereby interlocutors in- dicate to whom they are speaking (Jovanovi´c and Akker, 2004; Akker and Traum, 2009). In face- to-face communication, the basic clue for speci- fying addressees is turning one’s face toward the addressee. In contrast, in voice-only or text- based communication, the explicit declaration of ad- dressee’s names is more common. In this work, we tackle addressee and response selection for multi-party conversation: given a con- text, predict an addressee and response. As Fig- ure 1 shows, a system is required to select an ad- dressee from the agents appearing in the previous context and a response from a fixed set of candidate responses (Section 3). 2133 The key challenge for predicting appropriate ad- dressees and responses is to jointly capture who is talking about what …

Content truncated to 500 words