Skip to content

Latest commit

 

History

History
33 lines (28 loc) · 2.33 KB

README.md

File metadata and controls

33 lines (28 loc) · 2.33 KB
Raw

Automatic Sense Disambiguation of Potentially Idiomatic Expressions

This is the source code for a system to automatically disambiguate potentially idiomatic expressions (PIEs, for short) in text. It implements four methods of doing so: a baseline most-frequent-sense method, a baseline canonical form-based method (Fazly et al., 2009), a lexical cohesion graph-based method (Sporleder & Li, 2009), and a variation on that method using literal representations of idioms' figurative senses. It evaluates those methods on a combination of four corpora, the VNC-Tokens corpus, the IDIX corpus, the PIE Corpus, and the SemEval-2013 Task 5b dataset. For a detailed description of the systems, see our LAW-MWE-CxG paper.

Requirements

To run this code, you'll need the following Python setup:

  • Python 2.7.6
  • beautifulsoup4 4.5.1
  • numpy 1.14.0
  • scipy 0.19.1
  • spacy 2.0.6 + en_core_web_sm 2.0.0

Different versions might work just as well, but cannot be guaranteed.

You'll also need:

Getting Started

  • Clone the repository
  • Create subdirectories called working and ext
  • Add these symlinks (or edit config.py):
    • create a symlink ext/BNC to the Texts directory of your copy of the BNC
    • create a symlink ext/glove to the directory containing the GloVe embeddings
    • create symlinks ext/VNC, ext/IDIX, ext/PIE_Corpus, and ext/SemEval to the main directory of the respective corpora
  • Try and run the system with python psd.py -c 0 -m cg -gs 0s. This should run a basic lexical cohesion graph method and evaluate on the development set of the combined corpora.
  • Get an overview of all options by simply running python psd.py --help

Contact

For any questions about (running) the system, feel free to contact me.