Improving Syntactic Coordination Resolution using Language Modeling
Philip V. Ogren · 2010
Determining the correct structure of coordinating conjunctions and the syntactic constituents that they coordinate is a difficult task. This subtask of syntactic parsing is explored here for biomedical scientific literature. In particular, the intuition that sentences containing coordinating conjunctions can often be rephrased as two or more smaller sentences derived from the coordination structure is exploited. Generating candidate sentences corresponding to different possible coordination structures and comparing them with a language model is employed to help determine which coordination structure is best. This strategy is used to augment a simple baseline system for coordination resolution which outperforms both the baseline system and a constituent parser on the same task. 1