RoBERTaLexPT: A Legal RoBERTa Model pretrained with deduplication for Portuguese
Abstract
This work investigates the application of Natural Language Processing (NLP) in the legal context for the Portuguese language, emphasizing the importance of adapting pre-trained models, such as RoBERTa, from specialized corpora in the legal domain. We compiled and pre-processed a Portuguese Legal corpus, LegalPT corpus, addressing challenges of high document duplication in legal corpora, and measuring the impact of hyperparameters and embedding initialization. Experiments revealed that pre-training on legal and general data resulted in more effective models for legal tasks, with RoBERTaLexPT outperforming larger models trained on generic corpora, and other legal models from related works. We also aggregated a legal benchmark, PortuLex benchmark. This study contributes to improving NLP solutions in the Brazilian legal context, providing enhanced models, a specialized corpus, and a benchmark dataset. For reproducibility, we will make related code, data, and models available.
Copyright (c) 2024 Eduardo Garcia, Nádia Félix Felipe da Silva, Juliana Gomes, Hidelberg Albuquerque, Ellen Souza, Felipe Siqueira, Eliomar Lima, André Carvalho
This work is licensed under a Creative Commons Attribution 4.0 International License.
Authors who publish with this journal agree to the following terms:
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).
