Skip to content
Python Shell C++
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Rewarding Coreference Resolvers for Being Consistent with World Knowledge

To appear in EMNLP 2019.


For convinience, create a symlink: cd e2e-coref && ln -s ../wiki ./wiki

For pre-training the coreference resolution system, OntoNotes 5.0 is required. [Download] [Create splits]

Data for training the reward models and fine-tuning the coreference resolver (place in <PROJECT_HOME>/data):

  • 2M triples for RE-Text [Train] [Dev]
  • 12M triples for RE-KG [Train] [Dev]
  • 60k triples for RE-Joint [Train] [Dev]
  • 10k wikipedia summaries for fine-tuning [Download]

Note: If you want to make these files from scratch, follow the instructions in the triples folder.

Pre-trained models

  • Best performing reward model (RE-Distill) [Download]
  • Best performing coreference resolver (Coref-Distill) [Download]


Unzip Coref-Distill into e2e-coref/logs folder and run GPU=x python final


Reward models

  • Download pytorch big-graph embeddings (~40G, place in <PROJECT_HOME>/embeddings) [Download]
  • Run wiki/ to create an index of the embeddings (you need to do this only once)
  • Run reward module training with cd wiki/reward && python <dataset-name>

Coreference resolver

  • Follow e2e-coref/ to setup environment, create ELMO embeddings, etc.
  • Run coreference pre-training with cd e2e-coref && GPU=x python <experiment>
  • Start the sling server with python wiki/reward/
  • Change SLING_IP in wiki/reward/ to the IP of the sling server
  • Run coreference fine-tuning with cd e2e-coref && GPU=x python <experiment> (see e2e-coref/experiments.conf for the different configurations)


  • wiki/reward/ can be used to distill the various reward models
  • e2e-coref/ can be used to save the weights of the fine-tuned coreference models so that they can be combined by setting the distill flag in the configuration file


  author    = {Rahul Aralikatte and
               Heather Lent and
               Ana Valeria Gonz{\'{a}}lez{-}Gardu{\~{n}}o and
               Daniel Hershcovich and
               Chen Qiu and
               Anders Sandholm and
               Michael Ringaard and
               Anders S{\o}gaard},
  title     = {Rewarding Coreference Resolvers for Being Consistent with World Knowledge},
  journal   = {CoRR},
  volume    = {abs/1909.02392},
  year      = {2019},
  url       = {},
  archivePrefix = {arXiv},
  eprint    = {1909.02392},
  timestamp = {Mon, 16 Sep 2019 17:27:14 +0200},
  biburl    = {},
  bibsource = {dblp computer science bibliography,}
You can’t perform that action at this time.