The Benefits of (Automated) Dialogue

Introduction

The potential for commercial application of automated interfaces that use text or speech-based dialogue to exchange information and guide users is increasing as people become more familiar with computers and the internet (McTear, 2004; Bickmore and Giorgino, 2006). From a technical perspective the field is fairly mature, with a body of research on the implementation of automated dialogue (Androutsopoulos and Aretoulaki, 2003), and recent advances on the underlying dimensions influencing its efficacy (e.g. De Boni, Richardson and Hurling, 2007). Research on dialogue content has explored the role of empathy (Liu and Picard, 2005), the use of ‘small talk’ (Bickmore and Picard 2004), trust (Bickmore and Cassell 2001), emotions, personalisation and narration (Stock, 1996), information exchange and ‘like-mindedness’ (Svennevig, 1999). Turunen et al. (2004) examined the role of user experience, tailoring system output to differentiate between novices and experts, whilst Bernsen and Dybkjær (1996) compared co-operation between humans with that between humans and computers, highlighting that, in the latter, a clear communication of what the system can do should enhance the interaction.

De Boni et al. (2007) have shown that dialogue systems using relationship maintenance (e.g. elements of continuity between sessions) and appropriately positioned humour have a more positive impact on users’ perception of the dialogue. There has however been much less (if any) evaluation of the benefits of dialogue versus simpler forms of information exchange, such as a pick list of alternatives. Here we report a study directly comparing user perception of two systems, both designed to help identify solutions to overcome exercise barriers, but one guiding the user via an automated text based dialogue whilst the other employed a simple list of alternatives. We expected that user perception of dialogue versus simple pick-list systems might be dependent on the domain (e.g. railway times versus beliefs about exercising), the context (e.g. relaxed private versus rushed public situations) and the individual (e.g. degree of perceived control over own behaviour). In this study we constrained the domain (to finding solutions for exercise barriers) and the context (both were internet based systems used in the privacy of the user’s own home) and searched the literature for evidence on key factors that might influence user perception. First, we briefly reviewed the efficacy of Automated Behaviour Change Programs to determine the likely benefit of including dialogue-like elements, before considering the role of Individual Differences (such as Locus of Control, Self Efficacy and Personality). Finally we reviewed Usability Guidelines for computer based programs so that we could develop systems that were easy to use, and to minimise the influence of this factor when comparing the two types of system. We briefly summarise the literature in each of these areas and its bearing on our experiment design.