Embedded Files
Building trust through linguistic alignment
Building trust through linguistic alignment
A study on how conversational agents can foster collaboration by adapting their language.
A study on how conversational agents can foster collaboration by adapting their language.
The challenge
The challenge
What communication style makes an agent or partner seem trustworthy?
Understanding this informs human and human-AI interactions that foster trust and cooperation.
My role & methods
My role & methods
At the Institute of Communication and Cognition Sciences, I designed and programmed an online coordination game (LIONESS Lab), recruited participants (Prolific), ran data collection, and analyzed the results.
Participants described objects to two partners: one partner reused the participant’s own wording (“aligned”), while the other deliberately changed it (“unaligned”).
Key insight: Linguistic alignment directly builds trust.
Key insight: Linguistic alignment directly builds trust.
Participants trusted and preferred the aligned partner both epistemically (“Who to believe?”) and practically (“Who to work with again?”).
This finding suggests that humans and conversational AI can establish rapport and credibility by subtly mirroring others’ linguistic choices.
Impact & recognition
Impact & recognition
Sam Glucksberg Student Prize, XPrag X, Paris (2023)
Presented, among other venues, at The pragmatics of trusting (artificial) others, IPrA (2023)
Published in Cognitive Science (2024)
Experiment & results
Experiment & results
One partner reuses the participant’s own words; the other uses different ones.
Stimulus image adapted from Brown-Schmidt (2009) and Arnold et al. (2007).One partner reuses the participant’s own words; the other uses different ones.
Participants saw the aligned partner as a more reliable advisor.
Participants saw the aligned partner as a more reliable advisor.
Participants preferred to interact again with the aligned partner.
Participants preferred to interact again with the aligned partner.
Technical skills
Technical skills
Applied NLP (JavaScript)
Applied NLP (JavaScript)
I programmed the experimental manipulation, varying the agent's language based on the condition, and developed scripts to validate and clean user-generated text inputs in real-time.
Statistical analysis (R)
Statistical analysis (R)
I analyzed the binary choice data with a Generalized Linear Model, and the ordinal trust ratings with a Cumulative Link Model, to test the effect of coordination while controlling for demographics.
Page updated
Google Sites
Report abuse