Automatic discovery of feature sets for dependency parsing

This paper describes a search procedure to discover optimal feature sets for dependency parsers. The search applies to the shift–reduce algorithm and the feature sets are extracted from the parser configuration. The initial feature is limited to the first word in the input queue. Then, the procedure uses a set of rules founded on the assumption that topological neighbors of significant features in the dependency graph may also have a significant contribution. The search can be fully automated and the level of greediness adjusted with the number of features examined at each iteration of the discovery procedure. Using our automated feature discovery on two corpora, the Swedish corpus in CoNLL-X and the English corpus in CoNLL 2008, and a single parser system, we could reach results comparable or better than the best scores reported in these evaluations. The CoNLL 2008 test set contains, in addition to a Wall Street Journal (WSJ) section, an out-of-domain sample from the Brown corpus. With sets of 15 features, we obtained a labeled attachment score of 84.21 for Swedish, 88.11 on the WSJ test set, and 81.33 on the Brown test set.