Dataset origin: https://zenodo.org/records/6413804

French verbal frequencies in the Open Subtitle corpus

This repository provides frequencies for French verbal paradigm cells in the French Open Subtitles Corpus.

To obtain these frequencies we:

Trained a tagger based on Flaubert-base-cased (see FlauBERT) on the French_GSD part of the Universal Dependency treebank. In order to improve performances on rarer cells, the corpus was augmented with synthetic sentences which we generated using manually written templates and French verbal orthographic data from Demonext.
Cleaned and tagged the Open Subtitle Corpus
Counted the occurences of each cells for words tagged as VERB or AUX
Kept only the predicted feature combinations which were valid French verbal paradigm cells.
Converted to the paradigm cell scheme used in Vlexique

The frequencies are extracted from a 401360174 token corpus. We found a total of 74148211, from which 415176 occurences were excluded because of inexistant paradigm cells (prediction errors). 73733035 remaining occurences are documented.

published under the license: GPL-3.0
version: 1.0.0
Contributors:
- Sacha Beniamine (author)
- Maximin Coavoux (author)
- Olivier Bonami (author)

Please cite as: Sacha Beniamine, Maximin Coavoux & Olivier Bonami (2022), French verbal frequencies in the Open Subtitle corpus. DOI: 10.5281/zenodo.6412876

Files

frequencies (frequencies_subtitles.csv)

French paradigm cell frequencies

cell (string): paradigm cell, in the vlexique notation.
freq (integer): total raw frequency for this cell.