Exploring Spanish Corpora for Portuguese Coreference Resolution

The task of coreference resolution has attracted a great deal of attention in the literature due to its importance in deep language understanding, and its potential as a subtask in a variety of complex natural language processing problems. We experiment with different methods for generating training data and architectures for extracting meaningful mention representations. Coreference resolution in lesser-resourced languages is challenging, and transfer learning is a promising technique to overcome the comparatively smaller available corpora. We explore direct transfer learning from Spanish to Portuguese. We present state-of-the-art systems on both Spanish and Portuguese, and report promising results in a cross-language setting.