Semantic Analysis and Evaluation of Translation Based on Abstract Meaning Representation.

Abstract Meaning Representation (AMR) offers a novel scheme and perspective on the relation between linguistic form and semantic of various sentences. In order to explore the semantic equivalence among homologous translations, the paper analyzes the scenarios of same semantic structures covered by AMR and proposes a framework of variation in homologous translations. In addition, the framework is mapped into the annotation of AMR and used for common semantic characteristics mining in homologous translations. Accordingly AMR semantic structure matching (Smatch) score is applied to machine translation quality evaluation task. Experiments on small scale dataset preliminarily prove the effectiveness of AMR in translation quality evaluation.