Extracting Multiword Translations from Aligned Comparable Documents
Reinhard Rapp, Serge Sharoff · 2014
Most previous attempts to identify translations of multiword expressions using comparable corpora relied on dictionaries of single words.The translation of a multiword was then constructed from the translations of its components.In contrast, in this work we try to determine the translation of a multiword unit by analyzing its contextual behaviour in aligned comparable documents, thereby not presupposing any given dictionary.Whereas with this method translation results for single words are rather good, the results for multiword units are considerably worse.This is an indication that the type of multiword expressions considered here is too infrequent to provide a sufficient amount of contextual information.Thus indirectly it is confirmed that it should make sense to look at the contextual behaviour of the components of a multiword expression individually, and to combine the results.