epar

epar is a Java implementation by Kilian Evang of Zhang and Clark (2011)'s shift-reduce CCG parser.

It differs in some details, such as using a hash kernel (Bohnet 2010) for weight lookup, but gives very similar evaluation results.

External dependencies

Before you run the experiment described below, you will need to get some external dependencies. Create a directory called ext and make sure it contains the following repositories as subdirectories (you may also symlink them):

• supertagging, some software for producing the POS-tagged and supertagged inputs
• candc, the C&C tools, for evaluation scripts
• zpar, Yue Zhangs ZPar, for evaluation scripts

Follow the instructions in the supertagging README file for producing the needed POS-tagged and supertagged data.

The C&C tools need to be compiled, this can be done by running

make all bin/generate

from the candc directory.

To follow the steps below, you will also need:

• Apache Ant
• Java 7 or higher
• Produce

Make sure ant, java and produce are on your PATH.

Compiling

To compile epar, run:

ant

Training the parser model

To train the parser on the CCGbank sections 02-21 for 10 iterations, run:

produce output/wsj/train.model10

Parsing the development test corpus

To parse the development test corpus with the above model, run:

produce output/wsj/dev.trees10

Evaluation

For dependency evaluation:

produce output/wsj/dev.depeval10

For PARSEVAL evaluation:

produce output/wsj/dev.eval10

Bug reports

Bug reports are very welcome, preferably as GitHub issues.

Literature

Bernd Bohnet (2010): Very High Accuracy and Fast Dependency Parsing Is Not a Contradiction. In Proceedings of the 23rd International Conference on Computational Linguistics, pages 89–97. Association for Computational Linguistics.

Yue Zhang and Stephen Clark (2011): Shift-reduce CCG Parsing. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies – Volume 1, pages 683–692. Association for Computational Linguistics.