Emotion-Aware AI Chatbots as Emotional Support: Reducing Communication Anxiety and Boosting Willingness to Communicate among EAP Learners

Résumé fourni par la source

Foreign language anxiety (FLA) persistently constrains willingness to communicate (WTC) among English for Academic Purposes (EAP) learners, and existing AI language tools address this affective dimension inadequately. This exploratory study investigates the extent to which an emotion-aware AI chatbot, deployed as a real-time affective scaffold, is associated with reduced FLA and enhanced WTC among EAP learners at the College of Management Sciences, Egypt. A quasi-experimental mixed-methods design engaged 59 EAP students (aged 18–20) in two intact classes (experimental: n = 27; control: n = 32) across a 14-week semester. Pre- and post-test administrations of the Foreign Language Classroom Anxiety Scale (FLCAS) and WTC scale were triangulated with semi-structured interviews and structured classroom observations, and a delayed post-test followed six weeks after the immediate post-test. Treatment effects were tested with between-groups analyses of gain scores, equivalent to the group × time interaction of a 2 × 2 mixed-design ANOVA; the chatbot delivered adaptive scaffolding in response to behavioural cues treated as assumed proxies for anxiety. Baseline equivalence was confirmed (FLCAS: p = .745; WTC: p = .850). Group × time interactions significantly favoured the experimental group on the FLCAS total, F(1, 57) = 41.31, p < .001, between-group d = 1.68, and the WTC total, F(1, 57) = 62.06, p < .001, between-group d = 2.06, with significant interactions across all subdimensions (all p ? .002); the control group showed no significant within-group change, and experimental-group scores remained stable at the six-week follow-up. Qualitative analysis identified three operative mechanisms: pre-task anxiety buffering, perceived evaluative neutrality facilitating communicative risk-taking, and metacognitive self-regulation through post-session reflection. The study concludes that emotion-aware AI chatbots show preliminary promise as affective scaffolds complementing human EAP instruction. Because intact classes were taught by different instructors and the intervention added time-on-task, the observed associations cannot be interpreted causally; the reported effects are best treated as upper-bound estimates awaiting replication in larger, randomized, practice-equivalent designs. Received: 18 April 2026 / Accepted: 06 August 2026 / Published: September 2026

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.