Neural Belief Tracker: Data-Driven Dialogue State Tracking

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1777–1788 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1163 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1777–1788 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1163 Neural Belief Tracker: Data-Driven Dialogue State Tracking Nikola Mrkˇsi´c1, Diarmuid ´O S´eaghdha2 Tsung-Hsien Wen1, Blaise Thomson2, Steve Young1 1 University of Cambridge 2 Apple Inc. {nm480, thw28, sjy}@cam.ac.uk {doseaghdha, blaisethom}@apple.com Abstract One of the core components of modern spo- ken dialogue systems is the belief tracker, which estimates the user’s goal at every step of the dialogue. However, most current approaches have difficulty scaling to larger, more complex dialogue domains. This is due to their dependency on either: a) Spo- ken Language Understanding models that require large amounts of annotated training data; or b) hand-crafted lexicons for captur- ing some of the linguistic variation in users’ language. We propose a novel Neural Be- lief Tracking (NBT) framework which over- comes these problems by building on re- cent advances in representation learning. NBT models reason over pre-trained word vectors, learning to compose them into dis- tributed representations of user utterances and dialogue context. Our evaluation on two datasets shows that this approach sur- passes past limitations, matching the per- formance of state-of-the-art models which rely on hand-crafted semantic lexicons and outperforming them when such lexicons are not provided. 1 Introduction Spoken dialogue systems (SDS) allow users to in- teract with computer applications through conversa- tion. Task-based systems help users achieve goals such as finding restaurants or booking flights. The dialogue state tracking (DST) component of an SDS serves to interpret user input and update the belief state, which is the system’s internal repre- sentation of the state of the conversation (Young et al., 2010). This is a probability distribution over dialogue states used by the downstream dialogue manager to decide which action the system should User: I’m looking for a cheaper restaurant inform(price=cheap) System: Sure. What kind - and where? User: Thai food, somewhere downtown inform(price=cheap, food=Thai, area=centre) System: The House serves cheap Thai food User: Where is it? inform(price=cheap, food=Thai, area=centre); request(address) System: The House is at 106 Regent Street Figure 1: Annotated dialogue states in a sample di- alogue. Underlined words show rephrasings which are typically handled using semantic dictionaries. perform next (Su et al., 2016a,b); the system action is then verbalised by the natural language generator (Wen et al., 2015a,b; Duˇsek and Jurˇc´ıˇcek, 2015). The Dialogue State Tracking Challenge (DSTC) series of shared tasks has provided a common evalu- ation framework accompanied by labelled datasets (Williams et al., 2016). In this framework, the di- alogue system is supported by a domain ontology which describes the range of user intents the sys- tem can process. The ontology defines a collection of slots and the values that each slot can take. The system must track the search constraints expressed by users (goals or informable …

Content truncated to 500 words