emBERT
2020
-
Models for emBERT, the emtsv wrapper module for Huggingface's transformers package.
The following models are available from the emBERT-models repository:
- BIOE1 NER model, trained on the Szeged NER corpus. F1 97%.
- BIOE1 minimal NP chunking model, trained on Szeged TreeBank 2.0. F1 95.6% (SotA).
- BIOE1 maximal NP chunking model, trained on Szeged TreeBank 2.0. F1 95% (SotA).
Note that emBERT automatically downloads these models when needed.
If you use these models (or emBERT) in your work, please cite the following paper (see link for the bib; Hungarian):
Nemeskey Dávid Márk 2020. Egy emBERT próbáló feladat. In Proceedings of the 16th Conference on Hungarian Computational Linguistics (MSZNY 2020). pp. 409-418.
