Letícia Guedes Guimarães
Author directory2026
Clause Linkage in Brazilian Portuguese Narratives: Evidence from Enhanced Dependency Annotation
Letícia Guedes Guimarães | Adriana Pagano | André V. Lopes Coneglian | Magali Duran
Proceedings of the 17th Brazilian Symposium in Information and Human Language Technology
Letícia Guedes Guimarães | Adriana Pagano | André V. Lopes Coneglian | Magali Duran
Proceedings of the 17th Brazilian Symposium in Information and Human Language Technology
This paper explores the potential of Enhanced and Extended Enhanced Universal Dependencies for narrative analysis in Brazilian Portuguese, focusing on how enhanced dependency relations contribute to clause linkage. The study examines narrative sentences annotated in CoNLL-U format and compares basic and enhanced dependency layers. The results show that enhanced dependencies are particularly useful for recovering implicit subject relations in Portuguese and for enriching clause relations through information from prepositions and conjunctions. These enhancements enable a more explicit representation of participant continuity and event relations in narratives, supporting the transition from syntactic dependency analysis to richer semantic annotation.
Developing Multilingual Multiword Expressions (MWEs) Resources: An Annotated Corpus of Fables following the PARSEME guidelines
Letícia Guedes Guimarães | Adriana Pagano | André V. Lopes Coneglian | Aline Villavicencio
Proceedings of the 17th Brazilian Symposium in Information and Human Language Technology
Letícia Guedes Guimarães | Adriana Pagano | André V. Lopes Coneglian | Aline Villavicencio
Proceedings of the 17th Brazilian Symposium in Information and Human Language Technology
This paper introduces a Brazilian Portuguese corpus of fables developed as part of a broader multilingual initiative. It comprises the narratives and their concluding moral statements (epimythia), annotated for morphosyntax and multiword expressions (MWEs) according to the Universal Dependencies (UD) and PARSEME guidelines, respectively. The analysis reveals that 53.59% of the sentences contain at least one MWE, with inherently reflexive verbs (IRVs) constituting the most frequent category. MWEs are also distributed differently: narratives contain a higher proportion of adverbial MWEs, whereas adpositional MWEs are frequent in epimythia. Future work will enrich the corpus through semantic annotation based on Uniform Meaning Representation (UMR).