How Much Context Is Enough? Dependency-Aware Context Budgeting for LLM-Based Classroom Discourse Analysis
DOI:
https://doi.org/10.66421/int.j.datasci.intell.syst62Keywords:
classroom discourse, large language models, context length, long-context modeling, TalkMoves, NCTE, educational data miningAbstract
The rapid growth of large-language-model context windows creates an attractive but incomplete assumption for classroom analytics: more context should produce better discourse judgments. Classroom interaction contains at least two temporal regimes. Local discourse moves are often resolved by neighboring turns, whereas cumulative reasoning, revoicing, and teacher uptake can depend on events separated by many minutes. This paper introduces an evidence-calibrated framework for studying this distinction through a Context Utility Curve, Gain Retention, Marginal Context Utility, and a saturation horizon. We reanalyze two complementary controlled context studies grounded in real classroom corpora: a fine-grained TalkMoves context-window experiment and the long-context NSPA experiment on the NCTE transcript corpus. On TalkMoves, F1 rises from 71.93% without context to 78.92% at a ±7-turn window, but 95% of the maximum observed gain is already reached at ±3 turns (78.66%), leaving only 0.26 percentage points between that compact setting and the best reported setting. Under an equal 10-turn budget, bidirectional ±5 context exceeds a previous-only 10-turn window by 1.28 F1 points. In the lesson-scale LLM experiment, full long-context processing reaches 79.3% Macro-F1 versus 75.1% under 4k truncation, a 4.2-point absolute gain. These results support Dependency-Aware Context Budgeting: use compact bidirectional windows for locally identifiable moves and escalate to lesson-scale context for high-inference constructs. The central implication is that effective context is task-dependent; maximizing the available window is not equivalent to maximizing pedagogical measurement quality.
Index Terms—
References
[1] D. Demszky and H. Hill, “The NCTE Transcripts: A Dataset of Elementary Math Classroom Transcripts,” in Proc. 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA), 2023, pp. 528–538, doi: 10.18653/v1/2023.bea-1.44.
[2] A. Suresh, J. Jacobs, C. Harty, M. Perkoff, J. H. Martin, and T. Sumner, “The TalkMoves Dataset: K-12 Mathematics Lesson Transcripts Annotated for Teacher and Student Discursive Moves,” in Proc. 13th Language Resources and Evaluation Conf. (LREC), 2022, pp. 4654–4662.
[3] A. Suresh, J. Jacobs, M. Perkoff, J. H. Martin, and T. Sumner, “Fine-tuning Transformers with Additional Context to Classify Discursive Moves in Mathematics Classrooms,” in Proc. 17th Workshop on Innovative Use of NLP for Building Educational Applications (BEA), 2022, pp. 71–81, doi: 10.18653/v1/2022.bea-1.11.
[4] A. Suresh, J. Jacobs, V. Lai, C. Tan, W. Ward, J. H. Martin, and T. Sumner, “Using Transformers to Provide Teachers with Personalized Feedback on their Classroom Discourse: The TalkMoves Application,” AAAI 2021 Spring Symposium on Artificial Intelligence for K-12 Education, 2021, arXiv:2105.07949.
[5] S. Alic, D. Demszky, Z. Mancenido, J. Liu, H. Hill, and D. Jurafsky, “Computationally Identifying Funneling and Focusing Questions in Classroom Discourse,” in Proc. BEA, 2022, pp. 224–233, doi: 10.18653/v1/2022.bea-1.27.
[6] D. Demszky, J. Liu, H. C. Hill, D. Jurafsky, and C. Piech, “Can Automated Feedback Improve Teachers’ Uptake of Student Ideas? Evidence From a Randomized Controlled Trial in a Large-Scale Online Course,” Educational Evaluation and Policy Analysis, vol. 46, no. 3, pp. 483–505, 2024, doi: 10.3102/01623737231169270.
[7] H. C. Hill, M. L. Blunk, C. Y. Charalambous, J. M. Lewis, G. C. Phelps, L. Sleep, and D. L. Ball, “Mathematical Knowledge for Teaching and the Mathematical Quality of Instruction: An Exploratory Study,” Cognition and Instruction, vol. 26, no. 4, pp. 430–511, 2008, doi: 10.1080/07370000802177235.
[8] Q. Fang and W. Liu, “Neuro-symbolic pedagogical alignment (NSPA) for long-horizon classroom discourse analysis: Mitigating dialect bias via counterfactual preference optimization,” Computers and Education: Artificial Intelligence, vol. 11, Art. no. 100664, 2026, doi: 10.1016/j.caeai.2026.100664.
[9] S. Michaels, C. O’Connor, and L. B. Resnick, “Deliberative Discourse Idealized and Realized: Accountable Talk in the Classroom and in Civic Life,” Studies in Philosophy and Education, vol. 27, no. 4, pp. 283–297, 2008, doi: 10.1007/s11217-007-9071-1.
[10] L. B. Resnick, C. S. C. Asterhan, and S. N. Clarke, Accountable Talk: Instructional Dialogue that Builds the Mind. UNESCO International Bureau of Education, 2018.
[11] R. Alexander, “Developing dialogic teaching: genesis, process, trial,” Research Papers in Education, vol. 33, no. 5, pp. 561–598, 2018, doi: 10.1080/02671522.2018.1481140.
[12] M. T. H. Chi and R. Wylie, “The ICAP Framework: Linking Cognitive Engagement to Active Learning Outcomes,” Educational Psychologist, vol. 49, no. 4, pp. 219–243, 2014, doi: 10.1080/00461520.2014.965823.
[13] C. Howe, S. Hennessy, N. Mercer, M. Vrikki, and L. Wheatley, “Teacher–Student Dialogue During Classroom Teaching: Does It Really Impact on Student Outcomes?” Journal of the Learning Sciences, vol. 28, nos. 4–5, pp. 462–512, 2019, doi: 10.1080/10508406.2019.1573730.
[14] M. Nystrand, Opening Dialogue: Understanding the Dynamics of Language and Learning in the English Classroom. New York, NY, USA: Teachers College Press, 1997.
[15] E. Kasneci et al., “ChatGPT for good? On opportunities and challenges of large language models for education,” Learning and Individual Differences, vol. 103, Art. no. 102274, 2023, doi: 10.1016/j.lindif.2023.102274.
[16] D. Demszky et al., “Using large language models in psychology,” Nature Reviews Psychology, vol. 2, pp. 688–701, 2023, doi: 10.1038/s44159-023-00241-5.
[17] Yue, X., & Zhu, R. (2024). An explainable hybrid machine learning framework for cash-flow-conditioned multi-horizon liquidity risk early warning: An SME-oriented benchmark study. World Journal of Management Science, 2(1), 63-72.
[18] Jiao, Y., & Liang, Y. (2024). An integrated process mining and explainable machine learning framework for bottleneck and internal control risk detection in the multi entity financial close. Social Science and Management, 1(3), 96-107.
[19] Bian, M., Wang, Y., & Lin, Y. (2024). Explainable supplier risk prediction for critical telecommunications infrastructure procurement: An empirical XGBoost–SHAP framework. Social Science and Management, 1(1), 64-77.
[20] Liang, Y., & Jiao, Y. (2024). Real-time financial fraud detection using a feature-light temporal dynamic graph neural network with contrastive learning. Journal of Trends in Finance and Economics, 1(1), 76–84.
[21] Shang, W., & Chen, T. (2024). PrefixKD-Edge: Real-time locomotion-intent prediction with prefix self-distillation on lightweight temporal networks. Innovation and Technology Studies, 1(1), 30–38.
[22] Rhee, M., Zou, J., & Qin, Y. (2024). CLARITY: A closed loop framework for duplicate aware failure prediction and budgeted automated remediation in cloud infrastructure. AI and Data Science Journal, 1(1), 76-85.
[23] Zhang, C., & Wang, Y. (2024). A data-driven multi-objective production scheduling framework for battery pack manufacturing under material arrival uncertainty. AI and Data Science Journal, 1(1), 86–93.
[24] Lin, Y., Wang, Y., & Bian, M. (2024). XOR EW: An explainable and cost aware artificial intelligence framework for operational risk early warning in financial institutions. AI and Data Science Journal, 1(1), 94-102.
[25] Zhao, W., & Wang, B. (2022). CAPA: A prediction‑driven autoscaling framework for SLO‑aware cloud‑native machine learning inference. Journal of Computer Science and Electrical Engineering, 4(1), 8-16.
[26] Fang, Q., & Liu, W. (2025). HARLA-ED: Resolving Information Asymmetry and Enhancing Algorithmic Symmetry in Intelligent Educational Assessment via Hybrid Reinforcement Learning. Symmetry, 18(1), 58.
[27] Xing, S., & Wang, Y. (2025). Cross-modal attention networks for multi-modal anomaly detection in system software. IEEE Open Journal of the Computer Society.
[28] Wang, Y., & Zhang, C. (2023). AERIS: Adaptive executor scoped resource management for privacy preserving data processing in Apache Spark. Eurasia Journal of Science and Technology, 5(2), 25-33.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Charlotte Miller, Ethan Brooks, Sophie Turner

This work is licensed under a Creative Commons Attribution 4.0 International License.
All articles are published under the Creative Commons Attribution 4.0 (CC BY 4.0) License. Users may share, copy, adapt, and distribute the work for any purpose, provided proper credit is given to the original author(s) and source. Authors retain copyright.
