[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-585384-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-585384-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","cantotalk-probing-teacher-expertise-from-fine-tuned-talk-move-representations","CantoTalk - Probing Teacher Expertise From Fine-Tuned Talk Move Representations","","Classroom discourse shapes student learning, but analyzing teacher talk at scale is difficult in non-Western, low-resource language settings. The paper presents CantoTalk, a dataset of 7,518 Cantonese teacher utterances from Hong Kong mathematics classrooms labeled with ten talk-move categories. It evaluates whether large language models can classify talk moves and capture differences in teacher expertise. Fine-tuning five open-weight models achieves strong results, and probing embeddings shows expertise separability with high balanced accuracy.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/cantotalk-probing-teacher-expertise-from-fine-tuned-talk-move-representations/585384/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/cantotalk-probing-teacher-expertise-from-fine-tuned-talk-move-representations/585384.png","ImageObject",300,407,{"name":42,"@type":43},"Aria Callaghan","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-30","2026-10-15",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",7,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What is CantoTalk and what does it contain?","Question",{"text":63,"@type":64},"CantoTalk is a dataset of 7,518 Cantonese teacher utterances from Hong Kong mathematics classrooms annotated with ten talk-move categories.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How do the authors evaluate whether LLMs can handle the task?",{"text":68,"@type":64},"They fine-tune multiple open-weight LLMs and assess talk-move classification performance, then probe utterance-level embeddings to test whether teacher expertise is encoded.",{"name":70,"@type":61,"acceptedAnswer":71},"What do the representation analyses reveal about teacher expertise?",{"text":72,"@type":64},"The results show that teacher expertise is linearly separable from fine-tuned LLM embeddings with performance well above chance, and clustering reveals coherent discourse styles differing in authority, scaffolding, and dialogic engagement.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},585384,1792917451,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,110,114,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":107,"show_sort_weight":108,"slug":109},"Healthcare",40,"healthcare",{"id":111,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":112,"slug":113},8,30,"research-report",{"id":115,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},9,"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":111,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":144,"read_time":145,"preview_page_count":4},962084926284,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","CantoTalk: Probing Teacher Expertise From Fine-Tuned  \nTalk Move Representations  \nMayank Sharma∗  \nStanford University  \n[masharma@stanford.edu](masharma@stanford.edu)  \nXinman Liu∗  \nStanford University  \n[xinman@stanford.edu](xinman@stanford.edu)  \nGordon Wing-Leung Yeung  \nStanford University  \n[wlyeung@stanford.edu](wlyeung@stanford.edu)  \nABSTRACT  \nClassroom discourse profoundly shapes student learning, yet analyzing teacher talk at scale remains challenging in nonWestern and low-resource language contexts. This paper introduces CantoTalk, a dataset of 7,518 Cantonese teacher utterances from Hong Kong mathematics classrooms annotated with ten talk-move categories. We investigate whether LLMs can reliably classify these moves and encode systematic differences in teacher expertise. Fine-tuning five openweight LLMs yields strong performance, with the best model (Qwen3-8B) achieving micro-F1 of 0.81 and macro-F1 of 0.77. Probing utterance-level embeddings reveals that teacher expertise is linearly separable with 0.79 balanced accuracy, well above chance even after controlling for surface features. Clustering analyses uncover three coherent discourse styles differing in pedagogical authority, scaffolding, and dialogic engagement, with qualitative analysis showing systematic differences in how experienced and novice teachers execute similar talk moves. These findings demonstrate that finetuned LLM representations capture teacher expertise, offering a new lens for analyzing classroom discourse and informing teacher feedback tools.  \nKeywords  \nTalk Move Classification, LLM Fine-tuning, Representation Learning, Teacher Expertise Detection, Low-Resource NLP, Educational Discourse Analysis  \n1. INTRODUCTION  \nClassroom discourse profoundly shapes student learning and is influenced by pedagogical traditions and cultural norms [1, 2] . Yet in typical classrooms, students contribute far  \n∗ Equal contribution.  \nMayank Sharma, Xinman Liu, and Gordon Yeung. CantoTalk: Probing Teacher Expertise From Fine-Tuned Talk Move Representations. In Anthony Botelho, Maria Mercedes T. Rodrigo, Adish Singla, Hiroaki Ogata, Hyojeong So, and Young Hoan Cho (eds.) Proceedings of the 19th International Conference on Educational Data Mining, Seoul, Republic of Korea, June, 2026, pp. 44–56. International Educational Data Mining Society (2026) .  \n© 2026 Copyright is held by the author(s) . This work is distributed under the Creative Commons Attribution NonCommercial NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license. [https://doi.org/10.5281/zenodo.21039963](https://doi.org/10.5281/zenodo.21039963)  \nless than teachers: they speak for only 27 seconds per hour of teacher talk and ask just 2 academic questions per day, compared to teachers’ 142 questions [3] . Ensuring equitable classroom talk therefore depends on teachers’ ability to facilitate discourse that invites reasoning, dialogue, and participation from all learners [4] . To support these goals, researchers have developed structured“talk-move”frameworks to help teachers better engage students in class and achieve their instructional objectives [5, 6] . Traditional analysis of these talk moves, however, has relied on labor-intensive manual coding methods that cannot scale to support widespread teacher development or classroom research [4] .  \nRecent advances in large language models (LLMs) offer the potential to automate the identification of talk moves and provide immediate actionable insights to teachers [7] . However, classroom discourse remains largely an out-of-distribution domain for foundational models, which exhibit biases and misalignment to learning outcomes [8] . While there is emerging work on enhancing talk-move analysis and aligning models with pedagogical frameworks [9, 10], most studies focus on Western, English-speaking corpora and remain limited to classification tasks and frequency counts. Such analyses overlook cross-cultural adaptations of talk moves [11], the challenge","cbCaim1272RUSUrS","https://ap.wps.com/l/cbCaim1272RUSUrS","pdf",1430690,13,"English","# Abstract\n# 1. Introduction\n# 2. Related Work\n## 2.1 Talk Moves in Mathematics Classrooms","[{\"question\":\"What is CantoTalk and what does it contain?\",\"answer\":\"CantoTalk is a dataset of 7,518 Cantonese teacher utterances from Hong Kong mathematics classrooms annotated with ten talk-move categories.\"},{\"question\":\"How do the authors evaluate whether LLMs can handle the task?\",\"answer\":\"They fine-tune multiple open-weight LLMs and assess talk-move classification performance, then probe utterance-level embeddings to test whether teacher expertise is encoded.\"},{\"question\":\"What do the representation analyses reveal about teacher expertise?\",\"answer\":\"The results show that teacher expertise is linearly separable from fine-tuned LLM embeddings with performance well above chance, and clustering reveals coherent discourse styles differing in authority, scaffolding, and dialogic engagement.\"}]","CantoTalk - Probing Teacher Expertise From Fine-Tuned Talk Move Representations | PDF",1792054572,33]