Codes for the survey paper: A Brief Survey and Comparative Study of Recent Development of Pronoun Coreference Resolution This reporsitory contains code and models adapted from SpanBERT, Winogrande, and GPT2. Additional, we include our analysis and replication code for the regular PCR (CoNLL-2012 solved by SpanBERT) and Hard PCR (Winograd Schema Challenge solved by Winogrande and GPT2) problems. The analysis includes finegrained pronoun setting, cross-domain setting, model comparison, dataset split and etc.

Please refer to the links of the repositories to learn about the setup of each model.

After the setup from the original repository, use GPU=0 python evaluate.py <experiment> for the evaluation. independent.py and experiments.conf have been modified to include our experments.

We have replicate the WSC experiment for GPT-2 in ./hard_PCR (WSC)/gpt2/src/gpt2_classification.ipynb. The result is 69.23% accuracy in prediction.

You can run ./hard_PCR (WSC)/winogrande/scripts/run_experiment.py with the command provided by the original repository. You can run ./hard_PCR (WSC)/winogrande/wsc_prune_exp.ipynb to run the pruning experiement and ./hard_PCR (WSC)/winogrande/data/finetuning_similarity_measurement.ipynb for splitting the WSC alike datasets by the relevancy to the original 273 questions in WSC.

Please check our arxiv draft A Brief Survey and Comparative Study of Recent Development of Pronoun Coreference Resolution for more information.

If you have any other questions about this repo, you are welcome to open an issue or send me an email, I will respond to that as soon as possible.