📄 Proceedings of the 2nd Edition of the Universal Dependencies Brazilian Festival

Summary

This volume contains the proceedings of the second UDFest-BR. It is a workshop on applying the Universal Dependencies (UD) model to the Portuguese language. It took place on September 25, 2023, co-located with the 14th STIL symposium. The workshop was intended as a venue for exchanging annotation practices and jointly seeking solutions to common problems. The organizers note that treebanks and lexical resources for Portuguese already exist, but research on this language within UD is still scarce. The aim of the event is therefore to make such work more visible and to encourage new joint projects. The proceedings include ten papers. Among them are theoretical studies on the subtleties of the UD tagset and analyses of constructions on which parsers make errors. There are reports on annotating new genres: medical package inserts, financial tweets, and signs in the Warao language. The rest describe software tools. The preface reports no quantitative results. The volume's value lies in its overview of current problems in adapting UD to a language that still has few annotated resources. This experience may also be useful to other national UD communities.

Key points

• UDFest-BR 2023 was organized by the POeTiSA project. It is part of the NLP2 initiative of the C4AI center at the University of São Paulo, funded by IBM and FAPESP. The organizing committee included T. A. S. Pardo, M. S. Duran and L. Lopes (ICMC USP, São Carlos).
• The proceedings contain 10 papers in English and Portuguese. Topics include annotation of fixed MWEs in a Portuguese treebank, enhanced dependencies for Brazilian Portuguese, and the dar + para + V[infinitivo] construction in UD.
• The tools include the UD annotation verifier Verifica-UD and a method for lexical noun phrase chunking based on UD.
• For context, the preface mentions that UD already has about 200 annotated corpora in more than 100 languages.
• The program committee included, among others, Joakim Nivre, Alexandre Rademaker and Cláudia Freitas.