Human pause and resume behaviours for unobtrusive humanlike in-car spoken dialogue systems

Proceedings of the of the EACL 2014 Workshop on Dialogue in Motion (DM), pages 73–77, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Human pause and resume behaviours for unobtrusive humanlike in-car spoken dialogue systems Jens Edlund KTH Speech, Music and Hearing Stockholm Sweden edlund@speech.kth.se Fredrik Edelstam KTH Speech, Music and Hearing Stockholm Sweden freede41@kth.se Joakim Gustafson KTH Speech, Music and Hearing Stockholm Sweden jocke@speech.kth.se Abstract This paper presents a first, largely qualitative analysis of a set of human-human dialogues recorded specifically to provide insights in how humans handle pauses and resumptions in situations where the speakers cannot see each other, but have to rely on the acoustic signal alone. The work presented is part of a larger effort to find unobtrusive human dialogue behaviours that can be mimicked and implemented in-car spoken dialogue systems within in the EU project Get Home Safe, a collaboration between KTH, DFKI, Nuance, IBM and Daimler aiming to find ways of driver interaction that minimizes safety issues,. The analysis reveals several human temporal, semantic/pragmatic, and structural behaviours that are good candidates for inclusion in spoken dialogue systems. 1 Introduction In-car spoken dialogue systems face specific challenges that are of little or no relevance for systems designed for other environments. The two most striking of these are (1) the very strong focus on safety in the driving situation and (2) the fact that the person who speaks to the system – its user, in other words the driver in the majority of cases – does so in an environment that may change quite drastically from the beginning of an interaction to its completion. The most straightforward source for this change is the fact that the car (and the user) moves through the environment while the dialogue progresses. The dynamic and mobile nature of the surrounding traffic adds to the complexity. Generally speaking, safety is the key concern when designing spoken dialogue systems for in-car use. While poor performance in spoken dialogue systems can clearly be a nuisance to a driver, the promise of using properly designed spoken dialogue instead of other interfaces is increased safety. This promise is based in the nature of speech: it does not require the driver to divert the use hands and eyes from the driving, and it is a mode of communication that most are quite used to and comfortable with, so should not induce great amounts of cognitive load. We present a corpus consisting of a set of human-human dialogues recorded specifically to provide insights in how humans handle interruptions - how they pause and resume speaking - in situations where the speakers cannot see each other, but have to rely on the acoustic signal alone, and a preliminary analysis of these which reveals several candidates for inclusion in in-car spoken dialogue systems. Finally, we discuss how these can be implemented and how a selection of them are included in the Get Home Safe experiment implementation. 2 Background and related work In a government-commissioned survey from 2011, …

Content truncated to 500 words