Learning to Map Dependency Parses to Abstract Meaning Representations

Abstract

Abstract Meaning Representation (AMR) is a semantic representation language used to capture the meaning of English sen- tences. In this work, we propose an AMR parser based on dependency parse rewrite rules. This approach transfers dependency parses into AMRs by integrating the syn- tactic dependencies, semantic arguments, named entity and co-reference informa- tion. A dependency parse to AMR graph aligner is also introduced as a preliminary step for designing the parser.