Avaliação In-Domain e Cross-Domain em Restauração de Pontuação utilizando Processamento de Linguagem Natural

Punctuation plays a fundamental role in conveying the correct meaningin written texts. As a result, punctuation errors can occur, significantlyimpairing the way a message is interpreted, whether informal or informal contexts. In this sense, the use of machine learning,combined with recent techniques in natural language processing, hasbeen widely used in the task of punctuation restoration, in languagessuch as English. However, despite the wide application of this taskin other languages, its use in Portuguese is still quite limited. In thiswork, we propose to adapt a punctuation restoration model for itsapplication in formal texts in the Portuguese language, in additionto evaluating the model’s behavior in informal texts. The PortugueseLegal Sentences v3 dataset was used to train the model, which wasalso used for the in-domain evaluation. Regarding the cross-domainevaluation, the IWSLT (International Workshop on Spoken LanguageTranslation) database was used, consisting of transcripts of lecturesknown as TED Talks. The results indicate that the model with thelargest amount of training data and that mapped all question marksto full stops performed satisfactorily in the formal context, suggestingthat the methodology adopted was adequate for the proposedtask. Furthermore, it was found that the scarcity of question marksnegatively impacts the model’s performance and that, in the informalcontext, the results were unsatisfactory in the evaluation metrics, suggestingthat formal and informal sentences have their own structures,which the model was unable to generalize adequately in the informalcontext.