A Hybrid Approach for the Identification of Multiword Expressions
Marion Weller, Fabienne Fritzinger · 2010
According to Bannard (2007), “A multiword expression is usually taken to be any word combination that has some feature (syntactic, semantic or purely statistic) that cannot be predicted on the basis of its component words and/or the combinatorial process of language.”. We present a procedure for the identification of German multiword expressions (MWES) by making use of the morpho-syntactic restrictions of MWEs and of their semantic opacity which we approximate by their translational behaviour. As morpho-syntactic and translational features are independent, we combine their benefits assuming that they complement each other (hybrid approach). Based on the different features, we compute scores indicating a degree of idiomaticity. The procedure can be divided into two parts: candidate extraction from preprocessed text and feature evaluation, cf. Figure 1. We concentrate on preposition-noun-verb (PNV) triples (e.g. ’unter (den) Teppich kehren’ – lit. ’to sweep under the rug’, idiom. ’to hide sth.’), as this is a high-frequent pattern covering many idiomatic MWEs.