A Rule-Based AMR Parser for Portuguese

Semantic parsers help to better understand a language and may produce better computer systems. They map natural language statements into meaning representations. Abstract Meaning Representation (AMR) is a new semantic representation designed to capture the meaning of a sentence, representing it as a single rooted acyclic directed graph with labeled nodes (concepts) and edged (relations) among them. Although it is receiving growing attention in the Natural Language Processing community, most of the works have focused on the English language due to the lack of large annotated corpora for other languages. Thus, the task of developing parsers becomes difficult, producing a gap between English and other languages. In this paper, we introduce an approach for a rule-based parser with generic rules in order to overcome this gap. We evaluate the parser on a manually annotated corpus in Portuguese, achieving promising results and outperforming one of the current parser development strategies in the area.