• This repository contains "Korean Coreference Resolution model".
• This model is based on Kenton Lee's English Coreference Resolution model. And we applied it to Korean Coreference Resolution with referencing Shin et al.

• Install python3 requirements: pip3 install -r requirements.txt
• Build custom kernels by running setup_all.sh.
  • There are 3 platform-dependent ways to build custom TensorFlow kernels. Please comment/uncomment the appropriate lines in the script.
• Download word2vec.txt and save at here.

• ./train_coref.sh preprocesses data before train.
• Experiment configurations are found in experiments.conf
• Choose an experiment. Change paths of data, word embedding and other parameters which you would like.
• Training: python3 train.py <experiment>
• Results are stored in the logs directory and can be viewed via TensorBoard.
• Evaluation: python3 evaluate.py <experiment>

• logs directory have a pretrained model, MTA02-test.
  • MTA02-test is a pretrained model of crowdsourcing data set.
• If you want to use pretrained ELMo embedding, download it in the input directory.

• The training terminates automatically at 30k steps. The model generally converges at about 25k steps.
• If there are some errors when evaluating the development set, v4_gold_conll file may have errors. So, you should change the train, dev. set path of verify_conll.py and run it. Then, you may find some errors and fix them.
  • Most of these kind of errors are caused by ETRI morphological analysis.

• Kenton Lee et al., Higher-order Coreference Resolution with Coarse-to-fine Inference in NAACL 2018
• Shin et al., Korean Co-reference Resolution End-to-End Learning using Bi-LSTM with Mention Features in HCLT 2018

• CC BY-NC-SA Attribution-NonCommercial-ShareAlike
• If you want to commercialize this resource, please contact to us

Machine Reading Lab @ KAIST

This work was supported by Institute for Information & communications Technology Promotion(IITP) grant funded by the Korea government(MSIT) (2013-0-00109, WiseKB: Big data based self-evolving knowledge base and reasoning platform)