Decomposition Kernels for Natural Language Processing
2006
We propose a simple solution to the sequence labeling problem based on an extension of weighted decomposition kernels. We additionally introduce a multiinstance kernel approach for representing lexical word sense information. These new ideas have been preliminarily tested on named entity recognition and PP attachment disambiguation. We finally suggest how these techniques could be potentially merged using a declarative formalism that may provide a basis for the integration of multiple sources of information when using kernel-based learning in NLP.