Variation and Semantic Relation Interpretation: Linguistic and Processing Issues

International audienceStudies in linguistics define lexico-syntactic patterns to characterize the linguistic utterances that can be interpreted with semantic relations. Because patterns are assumed to reflect linguistic regularities that have a stable interpretation, several software implement such patterns to extract semantic relations from text. Nevertheless, a thorough analysis of pattern occurrences in various corpora proved that variation may affect their interpretation. In this paper, we report the linguistic variations that impact relation interpretation in language, and may lead to errors in relation extraction systems. We analyze several features of state-of-the-art pattern-based relation extraction tools, mostly how patterns are represented and matched with text, and discuss their role in the tool ability to manage variation