Evaluation of Two-level Dependency Representations of Argument Structure in Long-Distance Dependencies
Paola Merlo · 2015
Full recovery of argument structure information for question answering or information extraction requires that parsers can analyse long-distance dependencies. Previous work on statistical dependency parsing has used post-processing or additional training data to tackle this complex problem. We evaluate an alternative approach to recovering long-distance dependencies. This approach uses a two-level parsing model to recover both grammatical dependencies, such as subject and object, and full argument structure. We show that this two-level approach is competitive, while also providing useful semantic role information.