Abstract: The potential affordances of Artificial Intelligence (AI), and Speech and Language Technologies (SaLT) in particular, for language learning and teaching have long been recognised, with the first Speech and Language Technology in Education (SLATE; formerly Integrating Speech Technology in (Language) Learning (InSTIL) conference taking place in 1998. Having signposted this work, I introduce three problems in language education relevant to Education Policy in the UK to which I believe speech and language technologies may provide solutions: (1) the limited-exposure foreign languages schools curriculum (Marsden & Hawkes, 2023), (2) ensuring all adults achieve functional literacy (Dugdale & Clark, 2008; Hill et al., 2023; HMIP & Ofsted, 2022), and (3) enabling children with English as an Additional Language (EAL) to access learning (Hessel & Strand, 2023). In each case, having outlined the problem and its significance, I explore some ways in which speech and language technology might be harnessed to address the problem, and signpost some (second) language acquisition research that might inform the design of such solutions. I conclude the talk with a discussion of the potential for speech and language technologies and AI more broadly, including learning analytics, to support the development of apps based on current best understanding of (second) language acquisition and run large-scale learning experiments to further our understanding of the learning process and refine app designs. As a first step towards developing such apps, I highlight the importance of working across disciplines to develop theories of learning with “engineering power” specific enough to translate into learning designs and aligned with practising teachers local “craft knowledge” (Burkhardt and Schoenfeld, 2003). 

Abstract: Social robots offer innovative avenues for enhancing second language learning and teaching methodologies. These humanoid companions can serve as interactive language tutors, providing personalized feedback, engaging conversational practice, and immersive language experiences. However, state-of-the-art social robots remain relatively constrained in their capability to engage human users from diverse cultural backgrounds. For example, most social robots feature a 'universal' set of facial expressions, which are not universally recognised. In addition, it remains unclear which interactive pattern between the robot and the students – for example, which communication style the robot should equip and whether the robot should play as tutors or peers – can efficiently facilitate language learning. In this talk, I would like to propose the possible approach for developing culturally aware social robots and highlight their prospective role in assisting second language learning and intercultural communication training.

Abstract: This talk introduces the “ALTA Speaking” Spoken Language Processing Technology Project within the Automated Language Teaching and Assessment (ALTA) Institute, at the University of Cambridge. ALTA is a virtual institute that brings together researchers from Cambridge University Press & Assessment and the departments of Computer Science and Technology and Engineering. Employing advanced Artificial Intelligence (AI) and Machine Learning (ML) techniques, ALTA's primary objective is to develop innovative technologies to enhance the experience of language learning and to develop cutting-edge approaches to automated assessment that will benefit learners and teachers worldwide. Based at the Department of Engineering, the ALTA Speaking project adopts spoken language technologies to revolutionise tools for language teaching and assessment. The presentation will delve into how core speech technologies, such as automatic speech recognition (ASR) and speaker diarisation, and natural language processing (NLP), are integrated into the development of cutting-edge educational tools. We will explore the application of these AI-driven technologies in facilitating comprehensive automatic speech assessment, including holistic assessment and multi-view assessment that focuses on various aspects of language learning, such as grammar, pronunciation, fluency, and coherence. A critical element of the ALTA Speaking project is to provide constructive feedback to learners, essential for effective language acquisition and improvement. While the broader ALTA Institute, particularly colleagues in the Computer Science department, explore interactive content creation for language teaching and assessment, the ALTA Speaking project focuses on enhancing and applying spoken language technologies to improve educational tools and methodologies. Alongside these efforts, we are actively engaging in supporting initiatives for content creation. This collaboration underscores our commitment to pushing boundaries of educational innovation, ensuing our innovations remain impactful and accessible to learners worldwide.

Abstract: Due to the limited availability of learner language data, the speech and language technologies used in Computer Assisted Language Learning (CALL) applications are often trained using native language data instead. This can lead to reduced performance of such applications when used in real-life settings with learners. This talk provides an overview of how adapting disfluency detection models to better accommodate the linguistic features of learner speech leads to significant increase in system performance. Findings from this work motivate a call for the continued development of high-quality, linguistically-annotated learner language data.

Abstract: Second language (L2) learning is a complex process that is difficult to model. This work aims to develop a computational model of the teacher–learner interaction as used for L2 learning. The teacher model simulates a native English speaker, which uses repetition as a teaching strategy, while the learner model simulates a native Chinese speaker at an early stage of L2 English learning. Joint simulation may allow valuable insights into the entire learning process. In this study, speakers from the speechocean762 corpus were enlisted, using a word list that includes phonemes known to pose difficulties for Chinese speakers. The similarity between the output of the learning process and real learner data is evaluated using MCD, PPG, and wav2vec 2.0 distortion measures. The results indicate that the similarity between the process output and real learners with low proficiency is higher compared to that with real learners with high proficiency.