|
I was wondering about this area of coreference resolution. What are the challenges in this area, especially from a machine learning perspective rather than an NLP perspective? |
|
The currently best coreference resolution systems (see Haghighi and Klein, Coreference resolution in a modular, entity-centered model and Raghunathan et al, A multi-pass sieve for coreference resolution) have almost no machine learning, as apparently semantic constraints are far more important than the sort of features that can be easily incorporated in a discriminative model. From reading these papers I'd say that the main challenge in coreference resolution from a machine learning perspective is how to learn these relationships and constraints from minimally labeled data, followed by how to improve upon these techniques to get semantic informations. |
