hopsparser 0.6.0

A graph dependency parser to tinker with

HOPS, an honest parser of sentences

It ain't much but it's honest work.

This is a graph-based dependency parser inspired by Dozat and Manning (2017)'s biaffine graph parser. Contrary to Dozat, the parser performs its own tagging and can use several lexers such as FastText, BERT and others. It has been originally designed within the FlauBERT initiative.

The parser comes with pretrained models ready for parsing French, but it might be trained for other languages without difficulties.

Installation

The parser is meant to work with python >= 3.8. Install with pip, which should take care of all the dependencies and install the console entry points

pip install hopsparser

Inference

This assumes you have an already trained model in the directory MODEL. You can parse a file INPUT_FILE using the following command

hopsparser parse MODEL INPUT_FILE OUTPUT_FILE

This results in a parsed file at OUTPUT_FILE. Both INPUT_FILE and OUTPUT_FILEcan be set to-` to use the standard i/o streams, which can be convenient if you want to use the parser in a pipe.

The INPUT_FILE argument is supposed to be the path to a file in the CONLL-U format, possibly with missing columns. For instance:

1	Flaubert
2	a
3	écrit
4	Madame
5	Bovary
6	.

That is, we require word indexation and word forms only. Empty words are currently not supported. Multi-word tokens are not taken into account by the parsing models but are preserved in the outputs.

HOPS modifies the columns UPOS, HEAD and DEPREL, all the other columns and tree metadata are preserved.

Alternatively, you may add the --raw flag to the command above, in which case the parser expects a pre-tokenized raw text file with one sentence per line and individual tokens separated by blanks.

Depending on the model, the parser will be more or less fast and more or less accurate. We can however expect the parser to process several hundred sentences per second with a decent GPU. The GPU actually used for performing computations can be specified using the --device command line option.

Running in server mode

See the server mode documentation.

Pretrained models

We provide some pretrained models, see the list in models.md.

Training

Instead of using a pretrained model, one can train their own model. Training a model with BERT definitely requires a GPU. Unless you have a GPU with a large amount of onboard memory, using small batch sizes (2, 4, 8, 16, 32, 64) for training is probably a good idea. Otherwise, you are likely to run out of memory.

Training can be performed with the following steps:

1. Create a directory OUT for storing your new model
2. Copy a config YAML file from the examples directory
3. Edit it according to your needs
4. Run the command:

hopsparser train CONFIG.yaml TRAIN.conllu OUTPUT_DIR --dev-file DEV.conllu --test-file TEST.conllu 

After some time (minutes, hours…) you are done and the model is ready to run (go back to the parsing section). There are other options, see hopsparser train --help.

Partial annotations

The CoNLL-U files used for training may include missing annotations in either of the UPOS, HEAD and DEPREL columns, denoted by an underscore. In that case, the missing annotation will simply be ignored for training. For nodes where the HEAD information is missing, DEPREL will also be ignored.

Citation

If you use this parser for your scientific publication, or if you find the resources in this repository useful, please cite the following paper

@inproceedings{grobol:hal-03223424,
    title = {{Analyse en dépendances du français avec des plongements contextualisés}},
    author = {Grobol, Loïc and Crabbé, Benoît},
    url = {https://hal.archives-ouvertes.fr/hal-03223424},
    booktitle = {{Actes de la 28ème Conférence sur le Traitement Automatique des Langues Naturelles}},
    eventtitle = {{TALN-RÉCITAL 2021}},
    venue = {Lille, France},
    pdf = {https://hal.archives-ouvertes.fr/hal-03223424/file/HOPS_final.pdf},
    hal_id = {hal-03223424},
    hal_version = {v1},
}

Development

If you want a development install (so you can modify the code locally and directly run it), you can install it in editable mode with the tests extras after cloning the repository

git clone https://github.com/hopsparser/npdependency
cd npdependency
pip install -e .[tests]

In that case, you can run the smoke tests with tox to ensure that everything works on your end.

Note that using the editable mode requires pip >= 21.3.1.

Licence

This software is released under the MIT Licence, with some files released under compatible free licences, see LICENCE.md for the details.