Coreference resolution is a natural language processing task that involves identifying and linking all expressions in a text that refer to the same entity. For example, consider the following sentence: “John went to the store. He bought some groceries.” The pronoun “he” in the second sentence refers back to “John” in the first sentence. Coreference resolution algorithms are used to automatically identify such relationships in text.

To perform coreference resolution, a model needs to be able to identify all the mentions of entities in the text and determine which of these mentions refer to the same entity. Currently the most popular method is using a span ranking head on top of a pretrained model.

Span ranking head uses a scoring mechanism to eliminate unnecessary spans from all possible spans in a given text passage. There are pruning mechanisms on following papers to speed up this elimination process. The goal of span ranking head is to predict each tokens that forms an entity. Then there is a clustering mechanism to identify mentions that belong to the same entity.

paper : https://arxiv.org/pdf/1707.07045.pdf