AMR-based Semantic Parsing for the Portuguese Language

Meaning Representation (AMR) is a semantic formalism designed to capture the meaning of a sentence, representing it as a single rooted directed acyclic graph with labeled nodes (concepts) and edges (relations) among them. This representation has received growing attention from the Natural Language Processing community as many authors have proposed several models to produce an AMR graph from a sentence, aiming to improve natural language understanding. However, most of these models have focused on the English language due to the lack of large annotated corpora for other languages, producing a gap between English and other languages. To overcome this issue, in this paper, we carried out a fine-grained analysis of several parsers, adapted three different models to Portuguese, and proposed some improvements. Furthermore, we extended a previous rule-based AMR parser designed for Portuguese. We evaluated these models on a manually annotated corpus in Portuguese. Then, we performed a detailed error analysis to identify the major challenges in Portuguese AMR parsing that we hope will inform future research in this area.