adzhua / Coref-Constituent-Graph

Pytorch implementation of our AAAI 2022 paper "Incorporating Constituent Syntax for Coreference Resolution"

Incorporating Constituent Syntax in Coreference Resolution

This repo provides the source code of our paper: Incorporating Constituent Syntax for Coreference Resolution accepted at AAAI 2022.

Setup

• pip -r install requirements.txt
• python setup.py install. This will build kernel for extracting top spans implemented using the C++ interface of PyTorch.

English

Preprocess

• ./setup_training.sh <ontonotes/path/ontonotes-release-5.0> conll_data. Please change python to python2 in line 196 of conll_data/conll-2012/v3/scripts/skeleton2conll.sh to make sure the script run using python2. This assumes that you have access to OntoNotes 5.0. The preprocessed data will be included under conll_data.

Training

• python train.py <experiment>
• Results are stored in the log_root directory.
• For getting the result of using SpanBERT-Base and SpanBERT-Large model, use python train_spanbert_base_mention_cons_dual_plus_multi_order2 and python train.py train_spanbert_large_mention_cons_dual_plus_multi_order2
• Finetuning a SpanBERT large model on OntoNotes requires access to a 32GB GPU, while the base model can be trained in a 16GB GPU.

Chinese

Preprocess

• ./setup_training.sh <ontonotes/path/ontonotes-release-5.0> conll_data. This assumes that you have access to OntoNotes 5.0. The preprocessed data will be included under conll_data.

Training

• python train.py <experiment>
• Results are stored in the log_root directory.
• For getting the result of using BERT-WWM-Base and RoBERTa-WWM-ext model, use python train_bert_base_mention_cons_dual_plus_multi_order2 and python train.py train_roberta_large_mention_cons_dual_plus_multi_order2