discourse-sensitivity-attraction-effect

Material, code, and results for the paper:

Sanghee J. Kim* and Forrest Davis*. Discourse sensitivity in attraction effects: the interplay between language model size and training data. To appear in the Proceedings of the Society for Computation in Linguistics (SCiL) 2025.

---

Material

Test data from:
Sanghee J. Kim & Ming Xiang. Incremental discourse-update constrains number agreement attraction effect. Cognitive Science, 48(9): e13497.

Experimental materials used in human reading experiments (Experiments 1–3):

• attraction_exp1.csv
• attraction_exp2.csv
• attraction_exp3.csv

Code

Code for running tests and compiling results:

• AgrAttr.ipynb: Main code for running most models
• godel.py: Runs GODEL
• CompileResults.ipynb: Compiles output

Results

Output files generated by CompileResults.ipynb:

• combined_exp1_results.tsv
• combined_exp2_results.tsv
• combined_exp3_results.tsv
• by_clause_difference_table.tsv
• clause_comparison_table.tsv