Exploiting Fine-grained Syntactic Transfer Features to Predict the Compositionality of German Particle Verbs
Stefan Bott, Sabine Schulte im Walde · 2015
This article presents a distributional approach to predict the compositionality of German particle verbs by modelling changes in syntactic argument structure. We justify the experiments on theoretical grounds and employ GermaNet, Topic Models and Singular Value Decomposition for generalization, to compensate for data sparseness. Evaluating against three human-rated gold standards, our fine-grained syntactic approach is able to predict the level of compositionality of the particle verbs but is nevertheless inferior to a coarse-grained bag-of-words approach. 1