Incremental Dialogue Processing in a Micro-Domain

Proceedings of the 12th Conference of the European Chapter of the ACL, pages 745–753, Athens, Greece, 30 March – 3 April 2009. c⃝2009 Association for Computational Linguistics Incremental Dialogue Processing in a Micro-Domain Gabriel Skantze1 Dept. of Speech, Music and Hearing KTH, Stockholm, Sweden gabriel@speech.kth.se David Schlangen Department of Linguistics University of Potsdam, Germany das@ling.uni-potsdam.de Abstract This paper describes a fully incremental dia- logue system that can engage in dialogues in a simple domain, number dictation. Be- cause it uses incremental speech recognition and prosodic analysis, the system can give rapid feedback as the user is speaking, with a very short latency of around 200ms. Be- cause it uses incremental speech synthesis and self-monitoring, the system can react to feedback from the user as the system is speaking. A comparative evaluation shows that naïve users preferred this system over a non-incremental version, and that it was perceived as more human-like. 1 1 Introduction A traditional simplifying assumption for spoken dialogue systems is that the dialogue proceeds with strict turn-taking between user and system. The minimal unit of processing in such systems is the utterance, which is processed in whole by each module of the system before it is handed on to the next. When the system is speaking an ut- terance, it assumes that the user will wait for it to end before responding. (Some systems accept barge-ins, but then treat the interrupted utterance as basically unsaid.) Obviously, this is not how natural human- human dialogue proceeds. Humans understand and produce language incrementally – they use multiple knowledge sources to determine when it is appropriate to speak, they give and receive backchannels in the middle of utterances, they start to speak before knowing exactly what to say, and they incrementally monitor the listener’s reactions to what they say (Clark, 1996). 1 The work reported in this paper was done while the first author was at the University of Potsdam. This paper presents a dialogue system, called NUMBERS, in which all components operate in- crementally. We had two aims: First, to explore technical questions such as how the components of a modularized dialogue system should be ar- ranged and made to interoperate to support in- cremental processing, and which requirements incremental processing puts on dialogue system components (e.g., speech recognition, prosodic analysis, parsing, discourse modelling, action selection and speech synthesis). Second, to in- vestigate whether incremental processing can help us to better model certain aspects of human behaviour in dialogue systems – especially turn- taking and feedback – and whether this improves the user’s experience of using such a system. 2 Incremental dialogue processing All dialogue systems are ‘incremental’, in some sense – they proceed in steps through the ex- change of ‘utterances’. However, incremental processing typically means more than this; a common requirement is that processing starts before the input is complete and that the first output increments are produced as soon as possi- ble (e.g., Kilger & Finkler, 1995). Incremental modules hence are those where “Each processing component will …

Content truncated to 500 words