Paper: Wordsyoudontknow: Evaluation of lexicon-based decompounding with unknown handling

ACL ID W14-5707
Title Wordsyoudontknow: Evaluation of lexicon-based decompounding with unknown handling
Venue Computational Approaches to Compound Analysis
Session
Year 2014
Authors

In this paper we present a cross-linguistic evaluation of a lexicon-based decomposition method for decompounding, augmented with a ?guesser? for unknown components. Using a gold standard test set, for which the correct decompositions are known, we optimize the method?s parameters and show correlations between each parameter and the resulting scores. The results show that even with optimal parameter settings, the performance on compounds with unknown elements is low in terms of matching the expected lemma components, but much higher in terms of correct string segmentation.