A Discriminative Hierarchical Model for Fast Coreference at Large Scale

Methods that measure compatibility between mention pairs are currently the dominant approach to coreference. However, they suffer from a number of drawbacks including difficulties scaling to large numbers of mentions and limited representational power. As the severity of these drawbacks continue to progress with the growing demand for more data, the need to replace the pairwise approaches with a more expressive, highly scalable alternative is becoming increasingly urgent. In this paper we propose a novel discriminative hierarchical model that recursively structures entities into trees. These trees succinctly summarize the mentions providing a highly-compact information-rich structure for reasoning about entities and coreference uncertainty at small, large, and massive scales. The unique recursive structure of our entities allows our model to adapt to entities of various sizes, express features over entity hierarchies, and scale to massive data, making our approach a desirable new standard to replace the antiquated pairwise model