Interaction Features in Predicting Common European Framework of Reference Level

The study examined how interaction features could predict test takers’ English proficiency levels in high-stakes contexts. Using candidates’ interactive task responses in the Cambridge English Language Assessment (CELA), it explored salient interaction features that could distinguish across Common European Framework of References (CEFR) speaking levels (B1- C2).  It further ascertained the degree of accuracy in the proficiency prediction through a canonical discriminant analysis (CDA) in which weighted contributions of the 11 interaction features were created to best predict CEFR level of each test-taker. Fifty-eight video files (i.e., 116 interactive speech samples) were coded for the categories of four interactive features: (1) co-operation, (2) coherence, (3) turn-taking, and (4) strategy use. The results suggest that the selected interaction features distinguish test-takers’ CEFR levels with over 50-60 % accuracy especially regarding interactive turn and initiation. Findings offer direct implications to ESL classrooms and provide evidence to enhance our understanding of the complex nature of interactional competence in the context of the high-stakes speaking assessment.