Coreference Resolution

Coreference resolution determines when different words or phrases in a text refer to the same entity. In the sentence 'Maria said she would call', coreference resolution links 'she' back to 'Maria'. Expressions that share a referent are called coreferent, and the set of all coreferent mentions forms a coreference chain or cluster.

This task is critical for coherent text understanding. Without resolving pronouns and nominal references, a system reading 'The president signed the bill. He argued it would help the economy' cannot correctly attribute the action to the president. Question answering, information extraction, and summarization all depend on correctly tracking entities across sentences.

Coreference resolution is one of the harder NLP problems because it requires world knowledge, pragmatic inference, and long-range context. Neural models — particularly those built on large pretrained encoders like BERT — have significantly improved performance, but challenging cases involving bridging anaphora, implicit references, and ambiguous pronouns remain active research areas.