Multi-task Learning Based Neural Bridging Reference Resolution

Introduction

This repository contains code introduced in the following paper:

Multi-task Learning Based Neural Bridging Reference Resolution
Juntao Yu and Massimo Poesio In Proceedings of he 28th International Conference on Computational Linguistics (COLING), 2020

Setup Environments

• The code is written in Python 2, the compatibility to Python 3 is not guaranteed.
• Before starting, you need to install all the required packages listed in the requirment.txt using pip install -r requirements.txt.
• After that modify and run extract_bert_features/extract_bert_features.sh to compute the BERT embeddings for your training or testing.
• You also need to download context-independent word embeddings such as fasttext or GloVe embeddings that required by the system.

To use a pre-trained model

• Pre-trained models can be download from this link. We provide pre-trained models for ARRAU RST reported in our paper, if you need other models please contact me.

• Choose the model you want to use and copy them to the logs/ folder.

• Modifiy the test_path accordingly in the experiments.conf:

  • the test_path is the path to .jsonlines file, each line of the .jsonlines file is a batch of sentences and must in the following format:

  {
  "clusters": [[[0,0],[5,5]],[[2,3],[7,8]], #Coreference
  "bridging_pairs"[[[14,15],[2,3]],....] #Bridging 
  "doc_key": "nw",
  "sentences": [["John", "has", "a", "car", "."], ["He", "washed", "the", "car", "yesteday","."],["How","is","the", "left", "wheel","?"]],
  "speakers": [["sp1", "sp1", "sp1", "sp1", "sp1"], ["sp1", "sp1", "sp1", "sp1", "sp1","sp1"],["sp2","sp2","sp2","sp2","sp2","sp2","sp2"]] #Optional
  }
  

  • For coreference the mentions only contain two properties [start_index, end_index] the indices are counted in document level and both inclusive.
  • For bridging pairs, each pair contains two mentions the first one is the anaphora and the second one is the antecedent.

• Then use python evaluate.py config_name to start your evaluation

To train your own model

• You will need additionally to create the character vocabulary by using python get_char_vocab.py train.jsonlines dev.jsonlines
• Then you can start training by using python train.py config_name