The Effect of Using a Language Learning Application with Generative Ai Avatars (Praktika) On EFL Learners' Speaking Skills

This study examines the impact of Praktika, an AI-driven immersive language learning application featuring generative avatars, on the speaking proficiency of intermediate-level EFL learners in Iran. A total of ۳۰ male participants (aged ۱۷-۲۴), verified as B۲ level according to the CEFR, were randomly assigned to either an experimental group receiving AI-based instruction via Praktika or a control group taught through conventional classroom methods. Both groups engaged with identical thematic modules-Hobbies, Career, and Social Media-over six ۴۵-minute sessions spanning three weeks. Speaking proficiency was evaluated across four dimensions: fluency, accuracy, vocabulary range, and pronunciation. Data were triangulated through three sources: examiner-rated IELTS-style interviews, learner self-assessment questionnaires, and session-level interaction logs from the Praktika platform. Examiner scoring involved blind double-coding of ۳۰% of audio-recorded responses to ensure inter-rater reliability (Cohen's kappa = ۰.۸۲). Quantitative analysis was conducted using IBM SPSS (Version ۲۶). Paired sample t-tests revealed statistically significant improvements (p < .۰۵) in all four components for the experimental group, with effect sizes ranging from moderate to large (Cohen's d = ۰.۵۸-۰.۸۴). In contrast, the control group showed smaller and less consistent gains. Notably, triangulated data confirmed alignment between learners' self-perceived improvement and observed performance, particularly in vocabulary development and pronunciation accuracy. Analysis of interaction logs further revealed increased lexical diversity and reduced hesitation markers over time in the AI-assisted sessions. These findings underscore the pedagogical value of AI-based applications like Praktika in facilitating meaningful oral practice and targeted feedback, especially for learners with limited access to native interlocutors. By incorporating multidimensional assessment and comparative instruction, the study contributes robust evidence to the evolving field of computer-assisted language learning and offers practical insights for educators, curriculum designers, and policy-makers seeking to integrate AI into language education frameworks.