Pretrained models#
Sorted by language and corpus.
Usage note: The -camembert and -flaubert models use the eponymous embeddings and as such put a relatively heavy load on hardware. We recommend using them on GPUs with at least 10 GiB memory. Otherwise, running them on CPUs is still possible, albeit slow.
French#
FTB-UD#
GSD-UD#
Sequoia-UD#
French-spoken-UD#
Old French#
SRCMF-UD#
Due to changes in the parser in the meantime, the performances of these models differ from those presented in Grobol et al. (2022).
Model name |
UPOS (dev) |
LAS (dev) |
UPOS (test) |
LAS (test) |
Download |
|---|---|---|---|---|---|
UD_Old_French-SRCMF-2.9-bertrade_base |
97.29 |
88.35 |
97.33 |
88.97 |
|
UD_Old_French-SRCMF-2.9-camembert_base+mlm-fro |
97.61 |
90.37 |
97.66 |
91.19 |
|
UD_Old_French-SRCMF-2.9-flaubert_base_cased+mlm-fro |
97.65 |
90.91 |
97.69 |
91.00 |
If you use these models, please cite
@inproceedings{grobol2022BERTradeUsingContextual,
title = {{{BERTrade}}: {{Using Contextual Embeddings}} to {{Parse Old French}}},
booktitle = {Proceedings of the {{Thirteenth Language Resources}} and {{Evaluation Conference}}},
author = {Grobol, Loïc and Regnault, Mathilde and Ortiz Suárez, Pedro Javier and Sagot, Benoît and Romary, Laurent and Crabbé, Benoit},
date = {2022-06},
pages = {1104--1113},
publisher = {{European Language Resource Association}},
url = {https://aclanthology.org/2022.lrec-1.119},
eventtitle = {{{LREC}} 2022},
langid = {english},
venue = {Marseille, France}
}