This work evaluates AMR graph construction rules, originally developed for literary texts, in Portuguese news data to support building a symbolic AMR parser for this genre. Using the AMRNews-PT gold-standard corpus, we conduct an ablation analysis to assess rule contributions to AMR representation quality. The results reveal a hierarchical organization in which a small set of core structural rules drives most parsing performance, while additional rules provide finer semantic refinements. This rule hierarchy serves as the basis for building the parser and enabling the annotation of a larger news corpus.

In this article, we present the current state of DANTE (Dependency-ANalysed corpora of TwEets), a treebank focusing on tweets (currently X posts) written in Brazilian Portuguese, featuring four subcorpora varying in size, domain, and level of annotation under the Universal Dependencies framework. This article aims to inform the Natural Language Processing and Linguistic communities about currently available corpora, as well as what to expect from the treebank in the near future, while also bringing together the tools and language resources developed within the DANTE project.