Skip to content

Code for paper "Document-Level Argument Extraction by Conditional Generation". NAACL 21'

License

Notifications You must be signed in to change notification settings

raspberryice/gen-arg

Repository files navigation

Argument Extraction by Generation

Code for paper "Document-Level Argument Extraction by Conditional Generation". NAACL 21'

Dependencies

  • pytorch=1.6
  • transformers=3.1.0
  • pytorch-lightning=1.0.6
  • spacy=2.3.2

Model Checkpoints

Checkpoints trained from this repo are shared for the WikiEvents dataset and the ACE dataset are available at: [s3://gen-arg-data/checkpoints/].

You can download all the contents from the S3 bucket using AWS cli: aws s3 cp s3://gen-arg-data/checkpoints/ ./ --recursive

Model Predictions

The model predictions on WikiEvents is provided in outputs/wikievents-pointer-pred. Running this file through the scorer.py function should give you the exact same numbers as Table 5.

Datasets

You can download the data through the AWS cli or AWS console. Alternatively, you can download individual files by

  • wget https://gen-arg-data.s3.us-east-2.amazonaws.com/wikievents/data/<split>.jsonl for split={train, dev,test}.
  • wget https://gen-arg-data.s3.us-east-2.amazonaws.com/wikievents/data/coref/<split>.jsonlines for split={train, dev, test}.

Additional processed test files for RAMS can be downloaded by

  • wget https://gen-arg-data.s3.us-east-2.amazonaws.com/RAMS/test_head_coref.jsonlines
  • wget https://gen-arg-data.s3.us-east-2.amazonaws.com/RAMS/test_head.jsonlines

About

Code for paper "Document-Level Argument Extraction by Conditional Generation". NAACL 21'

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published