Retrieving Lexical Semantics from Multilingual Corpora

This paper presents a technique to build a lexical resource used for annotation of parallel corpora where the tags can be seen as multilingual ‘synsets’. The approach can be extended to add relationships between these synsets that are akin to WordNet relationships of synonymy and hypernymy. The paper also discusses how the success of this approach can be measured. The reported results are for English, German, French, and Greek using the Europarl parallel corpus.