Engagement and motivation are crucial for second-language acquisition, yet maintaining learner interest in educational conversations remains a challenge. While prior research has explored what makes educational texts interesting, still little is known about the linguistic features that drive engagement in conversations. To address this gap, we introduce IntrEx, the first large dataset annotated for interestingness and expected interestingness in teacher-student interactions. Built upon the Teacher-Student Chatroom Corpus (TSCC), IntrEx extends prior work by incorporating sequence-level annotations, allowing for the study of engagement beyond isolated turns to capture how interest evolves over extended dialogues. We employ a rigorous annotation process with over 100 second-language learners, using a comparison-based rating approach inspired by reinforcement learning from human feedback (RLHF) to improve agreement. We investigate whether large language models (LLMs) can predict human interestingness judgments. We find that LLMs (7B/8B parameters) fine-tuned on interestingness ratings outperform larger proprietary models like GPT-4o, demonstrating the potential for specialised datasets to model engagement in educational settings. Finally, we analyze how linguistic and cognitive factors, such as concreteness, comprehensibility (readability), and uptake, influence engagement in educational dialogues.

Language acquisition research has benefitted from the use of annotated corpora of child-directed speech to examine key questions about how children learn and process language in real-world contexts. However, a lack of sense-annotated corpora has limited investigations of child word sense disambiguation in naturalistic contexts. In this work, we sense-tagged 53 corpora of American and English speech directed to 958 target children up to 59 months of age, comprising a large-scale sample of 15,581 utterances for 12 ambiguous words. Importantly, we carefully selected target senses that we know - from previous investigations - young children understand. As such work was part of a project focused on investigating the role of verbs in child word sense disambiguation, we additionally coded for verb instances which took a target ambiguous word as verb object. We present experimental work where we leveraged our sense-tagged corpus ChiSense-12 to examine the role of verb-event structure in child word sense disambiguation, and we outline our plan to use Transformer-based computational architectures to test hypotheses on the role of different learning mechanisms underlying children word sense disambiguation performance.