A Local Alignment Kernel in the Context of NLP

This paper discusses local alignment kernels in the context of the relation extraction task. We define a local alignment kernel based on the Smith-Waterman measure as a sequence similarity metric and proceed with a range of possibilities for computing a similarity between elements of sequences. We propose to use distributional similarity measures on elements and by doing so we are able to incorporate extra information from the unlabeled data into a learning task. Our experiments suggest that a LA kernel provides promising results on some biomedical corpora largely outperforming a baseline.