Cristina Nicolae, Gabriel Nicolae
This paper presents a syntactic path-based learning algorithm (Candel from Candidate-Elimination) for the coreference resolution of pronouns that have their antecedents in the same sentence. Syntactic paths are treated as hypotheses to be learned. The hypotheses make up a version space that is delimited by a specific set and a general set, which grow closer to each other as the algorithm runs, in order to be consistent with the training examples encountered. Experiments on the MUC-6 and MUC-7 datasets reveal that this resolution method is a viable alternative to acquiring large amounts of data from the web.
Subjects: 13.1 Discourse; 12. Machine Learning and Discovery
Submitted: Feb 25, 2008