Elamur Research⌕
☰
home / papers / 2026 / Cross-Lingual and Cross-Cultural Transfer of Talk Move Classification to German Science Classrooms

Cross-Lingual and Cross-Cultural Transfer of Talk Move Classification to German Science Classrooms

Summary

Automatic talk move classification labels teacher and student utterances by their discourse function. It can support research on teaching quality and feedback to teachers, but existing training data are almost entirely English and drawn from mathematics or literacy lessons. This paper asks whether the approach carries over to German biology, chemistry and physics lessons, a shift in language, subject, national classroom culture and annotation team all at once.

The authors transcribed German science lessons from schools in Bavaria, Hesse and Lower Saxony. They coded them using the scheme of Suresh et al. (2021), which was designed for US math classes. They then compared three setups:

  • training only on the small German set;
  • training a multilingual model on the large English TalkMoves data;
  • fine-tuning first on English and then on German.

All three setups work, but the German-only model clearly beats the English-trained multilingual one. The two-stage approach is only marginally better than German-only. The main lesson is that a small in-language annotated corpus is worth more than a large foreign-language one.

Key points

  • The English source data come from the TalkMoves dataset (Suresh et al., 2021): 501 annotated math lessons with 235,926 sentences. For training, there are 151,264 sentences spoken by teachers and 53,079 spoken by students, while testing uses 23,303 teacher sentences and 7,262 student sentences.
  • The label set has seven teacher categories, including "no talk move", "keeping everyone together", "revoicing", "press for accuracy" and "press for reasoning". It has five student categories, including "making a claim" and "providing evidence".
  • Three annotators coded the German data. Their agreement, measured by Fleiss-κ, lay between 0.73 and 0.92 depending on subject and speaker role.
  • Only base-size models were tested. The authors reason that schools would likely not permit sending transcripts to external servers, so smaller locally run models reflect realistic use.
  • Data protection rules mean the German transcripts cannot be shared, not even in anonymized form.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3nynifnswvtk

@inproceedings{wartena2026cross,
  title = {{Cross-Lingual and Cross-Cultural Transfer of Talk Move Classification to German Science Classrooms}},
  author = {Wartena, Christian and Schumburg, Christian and Nehring, Andreas and Ebert, Marcel and Korneck, Friederike and Schmitt, David and Irmer, Marie and Neuhaus, Birgit},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/3nynifnswvtk},
  url = {https://doi.org/10.63317/3nynifnswvtk}
}

References

20 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Branden Chan, Stefan Schweter, and Timo Möller. 2020. German’s next language model. find our card
  2. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. find
  3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. BERT: pre-training of deep bidirectional transformers for language understanding. CoRR, abs/1810.04805. ↗
  4. Tobias Dorfner, Christian Förtsch, and Birgit J. Neuhaus. 2017. Die methodische und inhaltliche Ausrichtung quantitativer Videostu-dien zur Unterrichtsqualität im mathematisch-naturwissenschaftlichen Unterricht. Zeitschrift für Didaktik der Naturwissenschaften, 23(1):261– 285. find
  5. Achyutarama R. Ganti, Steven R. Wilson, and Geoffrey Louie Wing-Yue. 2025. Cross domain classification of education talk turns. In Proceedings of the 31st International Conference on Computational Linguistics, pages 6897–6917, Abu Dhabi, UAE. Association for Computational Linguistics. find
  6. Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2021a. Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding sharing. ↗ our card
  7. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2021b. Deberta: Decoding-enhanced bert with disentangled attention. In International Conference on Learning Representations. ↗ our card
  8. James Hiebert, James W. Stigler, and Alfred B. Manaster. 1999. Mathematical features of lessons in the timss video study. ZDM, 31(6):196–201. ↗
  9. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. find
  10. Jennifer Jacobs, Karla Scornavacco, Charis Harty, Abhijit Suresh, Vivian Lai, and Tamara Sumner. 2022. Promoting rich discussions in mathematics classrooms: Using personalized, automated feedback to support reflection and instructional change. Teaching and Teacher Education, 112:103631. ↗
  11. Jennifer Jacobs, Abhijit Suresh, Brandon M Booth, Tamara Sumner, Jeffrey Bush, Chelsea Brown, and Sidney K D’Mello. 2025. Automating feedback from recorded instructional observations: using ai to detect and support dialogic teaching. In Research Handbook on Classroom Observation, pages 341–365. Edward Elgar Publishing. ↗
  12. Eckhard Klieme and Jonathan Schweig. 2020. Opportunities to learn. Technical report, OECD, Paris. ↗
  13. Ashlee Kupor, Candice Morgan, and Dorottya Demszky. 2023. Measuring five accountable talk moves to improve instruction at scale. ↗
  14. Baptiste Moreau-Pernet, Yu Tian, Sandra Sawaya, Peter Foltz, Jie Cao, Brent Milne, and Thomas Christie. 2024. Classifying tutor discursive moves at scale in mathematics classrooms with large language models. In Proceedings of the Eleventh ACM Conference on Learning @ Scale, L@S ’24, page 361–365, New York, NY, USA. Association for Computing Machinery. find
  15. Catherine O’Connor, Sarah Michaels, and Suzanne Chapin. 2015. "Scaling Down" to explore the role of talk in learning: From district intervention to controlled classroom study. In Lauren Resnick, Christa Asterhan, and Sherice Clarke, editors, Socializing Intelligence through Talk and Dialogue, pages 111–126. American Educational Research Association. find
  16. Abhijit Suresh, Jennifer Jacobs, Charis Harty, Margaret Perkoff, James H. Martin, and Tamara Sumner. 2022a. The TalkMoves dataset: K-12 mathematics lesson transcripts annotated for teacher and student discursive moves. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 4654–4662, Marseille, France. European Language Resources Association. find
  17. Abhijit Suresh, Jennifer Jacobs, Vivian Lai, Chenhao Tan, Wayne Ward, James H. Martin, and Tamara Sumner. 2021. Using transformers to provide teachers with personalized feedback on their classroom discourse: The talkmoves application. ↗
  18. Abhijit Suresh, Jennifer Jacobs, Margaret Perkoff, James H. Martin, and Tamara Sumner. 2022b. Fine-tuning transformers with additional context to classify discursive moves in mathematics classrooms. In Proceedings of the 17th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2022), pages 71–81, Seattle, Washington. Association for Computational Linguistics. find
  19. Deliang Wang. 2025. Private communication. E-Mail. find
  20. Deliang Wang, Yaqian Zheng, Jinjiang Li, and Gaowei Chen. 2025. Parameter-efficiently fine-tuning large language models for classroom dialogue analysis. IEEE Transactions on Learning Technologies, pages 1–15. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wartena-2026-cross-lingual-cross-cultural-transfer/