Persistence in Human-Voice Assistant Interaction

Yannick Nicolai Frommherz

AutorIn

Yannick Nicolai Frommherz

Titel

Persistence in Human-Voice Assistant Interaction

Zitierfähige Url:

https://nbn-resolving.org/urn:nbn:de:bsz:14-qucosa2-1025115

Erstveröffentlichung

2026

Datum der Einreichung

02.07.2025

Datum der Verteidigung

15.12.2025

Abstract (EN)

Voice assistants help humans with tasks like scheduling meetings by providing an interface in the most natural modality, speech. Being a fundamental mechanism in human-human interaction, persistence – the tendency for interlocutors to converge linguistically when interacting – has been shown to be equally pervasive in interaction with machines, and specifically, voice assistants, though mainly by experiments relying on artificial simulations, limiting the ecological validity of such findings. As complementary corpus analyses hardly exist, this study contributes with a much-needed examination of more authentic data, drawing on three German-language corpora, two of which involve the commercial voice assistant Alexa, with the third featuring a well-simulated one. Proposing a viable mixed-methods approach, on the one hand, this study seeks to provide an exploratory quantitative assessment of the role that persistence plays in human-voice assistant interaction. For that, select alternation sets consisting of semantically equivalent variants are modelled in the variationist tradition using logistic regression, supported by descriptive statistics. Irrespective of the quantitative showing, on the other hand, this study seeks to provide a structured qualitative account specifically of cases of allo-persistence, that is, instances where human speakers re-use the voice assistant’s language. Strikingly, while not being uniform across alternations, results from the quantitative analysis suggest that overall persistence only partially plays a role in human-voice assistant interaction. While human speakers tend to re-use their own language, they seem not inclined to re-use the voice assistant’s linguistic material. These unexpected results may be attributed to experiments failing to account for the ‘messy’ complexity of talking to an actual voice assistant as well as to the latter seemingly being programmed to vary linguistically, essentially undercutting the ‘enterprise’ of persistence which between humans is a mutual one. Notwithstanding the smaller role it seems to play, allo-persistence can fruitfully be analysed qualitatively. Two sequential patterns proved relevant. Based on the voice assistant’s apparent non-involvement in persistence, human speakers can either adopt language for some previously unexpressed ‘thing’ or, crucially, shift from their own previous wording to one that the voice assistant (disruptively) introduced in the meantime. Reviewing examples for both adoption and shift persistence reveals that when human speakers do re-use the voice assistant’s language, this can be seen as reflecting a unique interplay of fine-grained factors, among them parsing success, dialogue coherence and the voice assistant using consistent terminology. Depending on whether these variables generate a smooth interaction context, cases of persistence therein may be accounted for by human speakers pre-consciously transferring their habitual communicative mechanisms to human-voice assistant interaction (mostly adoption persistence), or by them strategically navigating limitations of the latter (shift persistence). By addressing persistence in human-voice assistant interaction from two complementary angles, resting upon naturalistic data, this study prompts that, on balance, persistence may have been overestimated by previous experimental studies, while also carving out some contextual conditions under which cases of persistence do emerge in the given interaction format, ultimately underscoring the value of methodological triangulation in two distinct ways.

Forschungsdatenverweis