Toward Better Understanding of Hebrew NP Chunks (original) (raw)

In we presented two techniques (SVM Model Tampering and Anchored Learning) for investigating the SVM learning process and resulting models. These techniques were applied to the task of SVM based Hebrew NP Chunking. The results were better understanding of SVM based chunking, of the role lexical features play in the learned models, of the Hebrew NP chunking task definition, and identification of several deficencies in the NP Chunking corpus. This paper focuses on the Hebrew specific issues raised by that work, which are presented with more detail.