Using compound lists for German decompounding in a back-off scenario
Pedro Bispo Santos · 2014
Lexical resources like GermaNet offer compound lists of reasonable size. These lists can be used as a prior step to existing decompounding algorithms, wherein decompounding algorithms would function as a back-off mechanism. We investigate whether the use of compound lists can enhance dictionary and corpus-based decompounding algorithms. We analyze the effect of using an initial decompounding step based on a compound list derived from GermaNet with a gold standard in German. The obtained results show that applying information from GermaNet can significantly improve all tested decompounding approaches across all metrics. Precision and recall increases statistically significant by .004-.018 and .011- .022 respectively.