Linking knowledge graphs across languages with semantic similarity and machine translation

Abstract

Knowledge graphs and ontologies underpin many natural language processing applications, and to apply these to new languages, these knowledge graphs must be translated. Up until now, this has been achieved either by direct label translation or by cross-lingual alignment, which matches the concepts in the graph to another graph in the target languages. We show that these two approaches can, in fact, be combined and that the combination of machine translation and crosslingual alignment can obtain improved results for translating a biomedical ontology from English to German.