cgpotts / pipps

Repository for work studying the English Preposing in PP (PiPP) construction

Characterizing English Preposing in PP constructions

Christopher Potts

Paper

Potts, Christopher. 2023. Characterizing English Preposing in PP constructions. To be submitted to English language and linguistic theory: A tribute to Geoff Pullum.

Contents

This code is intended to be run with Python 3.9.

• books.ipynb: High-level stats for BookCorpusOpen, and sampling examples to annotate.
• c4.ipynb: High-level stats for C4, and sampling examples to annotate.
• annotated/: Hand-annotated samples.
• frequency_estimates.ipynb: PiPP frequency estimates for BookCorpusOpen and C4 based on the annotated samples.
• materials.txt: Sentences used in the LLM experiments.
• materials_check.ipynb: Notebook to facilitate checking examples in materials.txt in different variants.
• experiments_wh_effects.ipynb: Notebook for running the filler-gap experiments and summarizing the results.
• experiments_prep_effects.ipynb: Notebook for running the prepostional head experiments.
• experiments_llm_transformations.ipynb: Notebook for the exploratory pilot experiments seeing whether models can map PPs to PiPPs via few-shot in-context learning.
• materials-stress-test.csv: Additional hard examples for the transformations experiment.
• results/: All experimental results and visualizations.
• pipp.mplstyle: Matplotlib style file for the project.