BERTikal (aka legalnlp-bert)

BERTikal [1] is a cased BERT-base model for the Brazilian legal language and was trained from the BERTimbau's [2] checkpoint using Brazilian legal texts. More details on the datasets and training procedures can be found in [1].

Please check Legal-NLP out for more resources on (PT-BR) legal natural language processing (https://github.com/felipemaiapolo/legalnlp).

Please cite as Polo, Felipe Maia, et al. "LegalNLP-Natural Language Processing methods for the Brazilian Legal Language." Anais do XVIII Encontro Nacional de Inteligência Artificial e Computacional. SBC, 2021.

  @inproceedings{polo2021legalnlp,
    title={LegalNLP-Natural Language Processing methods for the Brazilian Legal Language},
    author={Polo, Felipe Maia and Mendon{\c{c}}a, Gabriel Caiaffa Floriano and Parreira, Kau{\^e} Capellato J and Gianvechio, Lucka and Cordeiro, Peterson and Ferreira, Jonathan Batista and de Lima, Leticia Maria Paz and do Amaral Maia, Ant{\^o}nio Carlos and Vicente, Renato},
    booktitle={Anais do XVIII Encontro Nacional de Intelig{\^e}ncia Artificial e Computacional},
    pages={763--774},
    year={2021},
    organization={SBC}
}

Usage

Ex. Loading model for general use

from transformers import AutoTokenizer  # Or BertTokenizer
from transformers import AutoModelForPreTraining  # Or BertForPreTraining for loading pretraining heads
from transformers import AutoModel  # or BertModel, for BERT without pretraining heads

model = AutoModelForPreTraining.from_pretrained('felipemaiapolo/legalnlp-bert')
tokenizer = AutoTokenizer.from_pretrained('felipemaiapolo/legalnlp-bert', do_lower_case=False)

Ex. BERT embeddings

from transformers import pipeline

pipe = pipeline("feature-extraction", model='felipemaiapolo/legalnlp-bert')
encoded_sentence = pipe('Juíz negou o recurso.')

Ex. Masked language modeling prediction

from transformers import pipeline

pipe = pipeline('fill-mask', model='felipemaiapolo/legalnlp-bert')

pipe('Juíz negou o [MASK].')
#  [{'score': 0.6387444734573364,
#  'token': 7608,
#  'token_str': 'julgamento',
#  'sequence': 'juiz negou o julgamento.'},
# {'score': 0.09632532298564911,
#  'token': 7509,
#  'token_str': 'voto',
#  'sequence': 'juiz negou o voto.'},
# {'score': 0.06424401700496674,
#  'token': 17225,
#  'token_str': 'julgado',
#  'sequence': 'juiz negou o julgado.'},
# {'score': 0.05929475650191307,
#  'token': 8190,
#  'token_str': 'recurso',
#  'sequence': 'juiz negou o recurso.'},
# {'score': 0.011442390270531178,
#  'token': 6330,
#  'token_str': 'registro',
#  'sequence': 'juiz negou o registro.'}]

References

[1] Polo, Felipe Maia, et al. "LegalNLP-Natural Language Processing methods for the Brazilian Legal Language." Anais do XVIII Encontro Nacional de Inteligência Artificial e Computacional. SBC, 2021.

[2] Souza, F., Nogueira, R., and Lotufo, R. (2020). BERTimbau: pretrained BERT models for Brazilian Portuguese. In 9th Brazilian Conference on Intelligent Systems, BRACIS, Rio Grande do Sul, Brazil, October 20-23