uwnlp / Easysrl

EasySRL builds logical forms for natural language sentences, by jointly modelling CCG and semantic role labelling. It uses an efficient A* parsing algorithm, meaning it can be used to process large corpora. A demo is online here.

If you use the parser for research, please cite the following paper:

@inproceedings{lewis:2015,
  title={Joint A* CCG Parsing and Semantic Role Labelling},
  author={Lewis, Mike and He, Luheng and Zettlemoyer, Luke},
  booktitle={Empirical Methods in Natural Language Processing},
  year={2015}
}

A pretrained model is available here.

If parsing questions, use this model instead.

Basic usage:

    java -jar easysrl.jar --model modelFolder

For CCG syntactic output, use:

    java -jar easysrl.jar --model modelFolder --outputFormat ccgbank

For semantic role labelling output, use:

    java -jar easysrl.jar --model modelFolder --outputFormat srl

To get n-best parses:

    java -jar easysrl.jar --model modelFolder --nbest 10 --supertaggerbeam 0.001

Please contact Mike Lewis with any questions or feature requests (email address in the paper).