Skip to content

Tools for all things related to Combinatory Categorial Grammar

License

Notifications You must be signed in to change notification settings

stanojevic/ccgtools

Repository files navigation

CCGtools

This is a collection of Python and Cython tools for processing CCG. It contains a pretrained parser, efficient loading of trees, A* search, latex style files, visualization, extraction of semantic dependencies, evaluation scripts etc. It is not very polished but you may find it useful. Probably the most interesting part to play with is a colab notebook for interactive parsing and visualiztion (see below).

Installation

To install the full parser run

pip install "ccgtools[parser]@git+https://github.com/stanojevic/ccgtools"

To install just the tools for building and evaluating parsers run

pip install "ccgtools@git+https://github.com/stanojevic/ccgtools"

All pretrained models have preffix "pretrained:" and will be automatically downloaded when needed. The list of available models is available here.

Command Line Usage

After the pip line above is exectued there are several commands that will be available on the PATH: ccg-eval, ccg-parser, ccg-supertagger and ccg-train. If you run them with --help flag they will provide the usage directions.

Notebook

There are two notebooks included in the repository:

  • demo.ipynb Open In Colab that shows how to use the provided parser, different ways of visualizing derivations, extraction of semantic dependencies, and extraction of logical representation. It can take a couple of minutes for the Colab to install all the necessary dependencies, but after that is done everything should be fast.
  • processing_effort.ipynb Open In Colab computes processing effort predictions for different CCG derivation trees. This follows descriptions from the papers that relate CCG parsing effort to brain activity (paper 1 and paper 2).

Author

Miloš Stanojević [email protected]

About

Tools for all things related to Combinatory Categorial Grammar

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published