stanford-nlpxed/classroom_management_data

Dataset Card for Classroom Management

Dataset Summary

Three annotated datasets of teachers' classroom management language. The annotations identify utterances as 1) classroom management, 2) behavior management, and 3) specific behavior management talk moves such as command, reprimands, threats, and exclusionary consequences. This repository also contains two notebooks for training RoBERTa models using this data, train_binary.ipynb and train_multiclass.ipynb.

Supported Tasks

Classification of teachers' classroom management talk moves in utterances from elementary math classroom transcripts

Languages

English

Dataset Description

File 1: classroom_management.csv

• —Number of instances: 10354
• —Features/Columns: 2

Data Fields

• —text — 7 utterances, composed of a target utterance marked by [TARGET][/TARGET] and the 3 utterances preceding and following. Utterances include speaker markings, but the target utterance is always the teacher.
• —labels — 1 if the utterance is an instance of classroom management language, 0 otherwise

File 2: behavior_management.csv

• —Number of instances: 8057
• —Features/Columns: 2

Data Fields

• —text — 7 utterances, composed of a target utterance marked by [TARGET][/TARGET] and the 3 utterances preceding and following. Utterances include speaker markings, but the target utterance is always the teacher.
• —labels — 1 if the utterance is an instance of behavior management language, 0 otherwise

File 3: talkmoves.csv

• —Number of instances: 5720
• —Features/Columns: 7

Data Fields

• —text — A single teacher utterance, pre-determined as being an instance of behavior management language.
• —vs — 1 if the utterance is an instance of verbal sanctioning, 0 otherwise
• —vs_submove — 1 if the utterance is an instance of a simple desist, 2 if the utterance is an instance of a command, 3 if the utterance is an instance of a reprimand, 4 if the utterance is an instance of a threat
• —ms — 1 if the utterance is an instance of material sanctioning, 0 otherwise
• —ms_excl — 1 if the utterance is an instance of the kind of material sanction that involves exclusionary discipline, 0 otherwise
• —ms_excl_iso — 1 if the utterance is an instance of exclusionary discipline that involves isolating the student outside the classroom, 0 otherwise
• —ms_excl_ch — 1 if the utterance is an instance of exclusionary discipline that involves calling home, 0 otherwise

Data Collection

Transcript data from the NCTE dataset [1] transcribed from the NCTE main study [2].

• —[1] Demszky, D., & Hill, H. (2023). The NCTE Transcripts: A Dataset of Elementary Math Classroom Transcripts. In 18th Workshop on Innovative Use of NLP for Building Educational Applications.
• —[2] Kane, Thomas, Hill, Heather, and Staiger, Douglas. National Center for Teacher Effectiveness Main Study. Inter-university Consortium for Political and Social Research [distributor], 2022-06-16. https://doi.org/10.3886/ICPSR36095.v4

Annotation

Annotators are 6 experienced elementary math teachers trained on this task. Fleiss' Kappa scores range from 0.719 to 0.964.

Ethical Considerations

Not intended for evaluation of teaching quality. What is appropriate in a given classroom is highly contextual and relational in a way that these annotations to not capture.

Citation

If you use this dataset, please cite:

Tan, Mei, and Dorottya Demszky. (2025). Do As I Say: What Teachers’ Language Reveals About Classroom Management Practices. (EdWorkingPaper: 23-844). Retrieved from Annenberg Institute at Brown University: https://doi.org/10.26300/9yj6-jn52