Nschneid writeup of Bunescu 2005

This is Nschneid's review of Bunescu_2005_a_shortest_path_dependency_kernel_for_relation_extraction

Relation extraction with SVMs. They extract features for each token and consider paths between tokens in a dependency graph, derived either from a dependency parse (local dependencies only) or a CCG parse (local+nonlocal dependencies). They define a kernel which computes similarity between two dependency paths, counting how many features are shared for each token along the path. The kernel is used to determine which of the training relation instances is closed to the test instance.

The shortest-path dependency kernels outperform the dependency kernel from (Culotta and Sorensen, 2004) in both scenarios, with a more significant difference for SP-CFG. An error analysis revealed that Collins’ parser was better at capturing local dependencies, hence the increased accuracy of SP-CFG. Another advantage of our shortest-path dependency kernels is that their training and testing are very fast.

• Did they simply use a dependency tree in the (more successful) CFG case? Is the shortest-path issue then irrelevant, since there is exactly one path between a pair of nodes in a tree?