AMR2FRED, A Tool for Translating Abstract Meaning Representation to Motif-Based Linguistic Knowledge Graphs

摘要

In this paper we present AMR2FRED, a software application to translate Abstract Meaning Representation (AMR) to RDF using the knowledge patterns applied by the FRED machine reading method. AMR and FRED representations are both graph-based, and event-centric (neo-Davidsonian), but they differ in several logical, conceptual, and design assumptions. The former has become a de facto standard for the Natural Language Processing community, whereas FRED adds semantics to the extracted information using several ontologies and best practices from the Semantic Web. With the increasing availability of manually AMR-annotated datasets, this tool provides straightforward means to adapt annotated datasets for AMR according to the design patterns used by FRED, and to evaluate machine reading tools with gold-standard data. AMR2FRED takes as input an AMR representation of a text, and prints a FRED-like RDF output. The system is open source and can be freely downloaded from https://github.com/infovillasimius/amr2Fred.