Title:Coreference Resolution without Span Representations

Abstract:Since the introduction of deep pretrained language models, most task-specific NLP models were reduced to simple lightweight layers. An exception to this trend is the challenging task of coreference resolution, where a sophisticated end-to-end model is appended to a pretrained transformer encoder. While highly effective, the model has a very large memory footprint -- primarily due to dynamically-constructed span and span-pair representations -- which hinders the processing of complete documents and the ability to train on multiple instances in a single batch. We introduce a lightweight coreference model that removes the dependency on span representations, handcrafted features, and heuristics. Our model performs competitively with the current end-to-end model, while being simpler and more efficient.