What Do We See in Them? Identifying Dimensions of Partner Models for Speech Interfaces Using a Psycholexical Approach

Abstract

Perceptions of system competence and communicative ability, termed partner models, play a signifcant role in speech interface interaction. Yet we do not know what the core dimensions of this concept are. Taking a psycholexical approach, our paper is the frst to identify the key dimensions that defne partner models in speech agent interaction. Through a repertory grid study (N=21), a review of key subjective questionnaires, an expert review of resulting word pairs and an online study of 356 users of speech interfaces, we identify three key dimensions that make up a users' partner model: 1) perceptions towards partner competence and dependability; 2) assessment of human-likeness; and 3) a system's perceived cognitive fexibility. We discuss the implications for partner modelling as a concept, emphasising the importance of salience and the dynamic nature of these perceptions. CCS CONCEPTS• Human-centered computing → Natural language interfaces; HCI theory, concepts and models; Interaction design theory, concepts and paradigms; • Applied computing → Psychology.

Creation through Conversation - A Provocation

“…While years of work has investigated the effects of different 'personalities' that CUIs have on us, more recent work looks at how approaches from user modelling can be used to understand their human interlocutors and tailor responses accordingly. Doyle et al [6] investigated lexical alignment and partner modelling in the context of CUIs. Peña et al [12] build on this idea of partner models, demonstrating how allocentric (attention focused on the conversational partner) language production increases when a voice user interface is perceived as a dialogue partner rather than a tool.…”

Audience Design and Egocentrism in Reference Production During Human-Computer Dialogue

“…Research on the mechanisms that govern language production in HCD is limited, with recent work calling for further research on this topic (Clark et al, 2019;Peña et al, In Press;Cowan et al, 2023). Most existing literature tends to echo an audience design account, suggesting that language production in HCD is adaptive, being informed by preconceptions of a computer partner!s abilities and perceived knowledge (that is, a user!s partner model; Doyle et al, 2021) (Amalberti et al, 1993;Brennan, 1998;Cowan et al, 2019a;Le Bigot et al, 2007;Meddeb & Frenz-Belkin, 2010a). In comparison to HHD, people tend to use fewer fillers and coherence markers when speaking to a computer (Amalberti et al, 1993), reduce their use of pronominal anaphora, use more basic lexical choices, and make shorter utterances (Kennedy et al, 1988).…”

“…Previous work has noted that, although anchored to estimates of what they feel others know, people tend to assume more knowledge of computers than humans in object naming tasks (Cowan et al, 2017). These knowledge expectations are critical to people!s partner models of computers as dialogue partners (Doyle et al, 2021). Although this may lead to what seems like egocentric based behaviours, the mechanism by which this occurs may in fact be one of audience design based on assuming the knowledge state of the computer is more complete (i.e., they can see both occluded and non-occluded objects) than when interacting with the human partner.…”

“…As mentioned in previous work, these perceptions (i.e., user!s partner models) may be an important component to guiding audience design. Indeed recent definitions of this concept suggest that experiences people have with speech interfaces (Doyle et al, 2021) may be important to consider. Within our study, past experience data was collected through asking participants how often they used speech agents, by asking them to select whether they use them daily, a few times per week, a few times per month, rarely or never.…”

From C-3PO to HAL: Opening The Discourse About The Dark Side of Multi-Modal Social Agents

“…Communicative bots such as voice assistants, social (ro-)bots, and taskbots are increasingly becoming part of everyday life [17,34], with work showing how we integrate these agents in our daily routines [11] while creating emotional bonds with them [24,25]. As these artificial agents grow in numbers and level of sophistication, so does the number of questions relating to human-agent communication.…”