Exploring the Effect of Cultural Background and Interaction Language on Preferences for Repair Strategies in Spoken Dialogue Systems

Abstract

Spoken dialogue systems (SDSs) are increasingly integrated into everyday life, appearing in forms such as embodied robots, in-car systems, and, most commonly, digital voice assistants (DVAs) like Amazon Alexa, Apple’s Siri, and Google Assistant. Despite their widespread use, conversational breakdowns remain inevitable and problematic. Repair strategies, defined as system responses designed to recover from such breakdowns, play a crucial role in shaping user experience. While prior research has focused on technical solutions and the development of effective repair strategies, limited attention has been paid to how user characteristics influence preferences and perspectives on communication repair strategies in voice-based interactions.

This thesis addresses this gap by examining how user characteristics shape preferences and perspectives on repair strategies in goal-oriented tasks. Specifically, it considers (a) cultural background (high- vs. low-context cultures), (b) interaction language (native vs. non-native), and (c) individual-level factors. To achieve this, the thesis begins with a systematic scoping review, followed by two user-centered empirical studies. The overarching goal is to provide a deeper understanding of how user characteristics influence repair strategy preferences and perspectives, thereby informing the design of adaptive SDSs that deliver improved user experiences.

The scoping review synthesized the literature on repair strategies in SDSs and produced two comprehensive frameworks: one categorizing system-initiated repair strategies and another addressing user-initiated repair strategies. Building on these foundations, the first empirical study examined how cultural background influences preferences by comparing users from the United Kingdom and Saudi Arabia. The second empirical study examined the influence of interaction language (native vs. non-native) alongside individual-level factors, including prior experience and computer self-efficacy, on repair strategy preferences. Complementing these quantitative studies, a qualitative investigation using semi-structured interviews explored user perspectives on communication breakdowns and repair strategies when interacting with SDSs.

Overall, the findings reveal cultural differences, particularly in preferences for elaborative and explanation-based repair strategies. However, the effects of interaction language and individual-level factors often outweighed cultural influences. The qualitative insights provided deeper explanations of these preferences, highlighting how contextual factors and task demands shaped user perspectives on breakdowns and repair. Collectively, this thesis extends the theoretical understanding of repair strategies in SDSs and offers practical recommendations for developing adaptive, user-centered systems that are sensitive to cultural, linguistic, and individual diversity.