Situated L2 pronunciation instruction during small-group robot-assisted language learning activities

Chatbots and other conversational agents based on speech recognition and processing technologies have been gaining ground in the field of language education. Although previous research has shown that automatic recognition of second language (L2) speech is difficult, little attention has been paid to how L2 teachers and learners interact with such technology when used as an interactional participant in classroom settings. Addressing this gap, this article provides a qualitative analysis of interactional practices of unplanned and situated pronunciation instruction as a teacher and 10- to 13-year-old young learners of L2 English complete robot-assisted language learning (RALL) activities in a primary school English-as-a-foreign-language (EFL) context in Finland. Drawing on 14 hours of video recordings, we use multimodal conversation analysis (CA) to analyse extended repair sequences that involve interactional problems related to word recognition by a social robot. Through a sequential analysis of selected data extracts, we show how the teacher and learners correct these problems by establishing a corrective focus for providing instruction on and modifying learners’ word-level pronunciation, such as the quality of individual sounds or word stress. From the teacher’s perspective, this consists of drawing learners’ attention to pronunciation details by highlighting sounds in learners’ talk and the robot’s talk, using embodied conduct, and modelling a target-like word pronunciation. Our findings shed light on the interactional organisation of RALL activities and some of the real-life consequences of limitations in speech recognition technologies for L2 teaching and learning interactions with conversational agents. The work conducted by the teacher to convert interactional troubles into meaningful learning opportunities suggests that human agency is needed to optimally guide and mediate language learning interactions with conversational agents based on artificial intelligence (AI) and automatic speech recognition (ASR), as these agents are less capable of showing the kind of interactional and instructional adaptation that is part of human–human interaction.