Grex: automatic grammar extraction from treebank

Grex is a tool for automatic grammar rule extraction from treebanks created by Santiago Herrera, Caio Corro and Sylvain Kahane. A full description of the method can be found in the paper: https://arxiv.org/abs/2403.17534

If you use this software, please cite the following work:

@inproceedings{herrera2024grex,
    title = "Sparse Logistic Regression with High-order Features for Automatic Grammar Rule Extraction from Treebanks",
    author = "Herrera, Santiago and Corro, Caio and Kahane, Sylvain",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    url = "https://arxiv.org/abs/2403.17534",
}

Usage

You first need to install the Skglm library: https://contrib.scikit-learn.org/skglm/auto_examples/index.html

At the moment, there are only two scripts:

• autogramm_agreement.py: search rules of morphological agreements (e.g. when is there a number agreement between a word and its head?)
• autogramm_activation.py: search rules that activate a feature (e.g. when is the subject after the verb?)

Examples are given in run.sh. The extracted rules are exported in json format.

Common arguments

• --treebank: directory where treebanks are stored
• --treebank-filter: list of treebanks to use (partial match will be used)
• --feature-filter: features to remove (must be lowercased + partial match will be used)
• --dep-filter: filter the dataset. For example, --dep-filter=head_upos=VERB,mod_upos=NOUN will check only dependencies between a VERB and a NOUN
• --json: output file
• --error: error file

autogramm_agreement.py

Search rules that forces two features to agree.

• --feature1: first feature
• --feature2: second feature

autogramm_activation.py

• --feature-name: name of the feature
• --feature-value: value of the feature we are interested in