Coreference

Coreference is the linguistic relation in which multiple expressions in a discourse refer to the same real-world entity. Expressions can be proper names, common nouns, or pronouns, and a coreference chain groups all mentions that refer to the same entity within a text. Coreference is essential for coherent interpretation, discourse structure, and information integration.

Anaphora is a common type of coreference where a later expression refers back to an earlier antecedent, as in "Alice arrived late because she was stuck in traffic," where "she" refers to "Alice." Cataphora is the opposite, where an expression refers forward to a later antecedent, such as "When he spoke, John smiled," where "he" points to "John."

Coreference resolution (also called anaphora resolution) is the task of identifying antecedents for all mentions and forming coherent clusters of co-referring expressions. Approaches range from rule-based and statistical methods to modern neural models that learn to form clusters of spans without explicit mention-pair rules. End-to-end neural models compute representations for candidate mentions and decide which mentions belong together in the same entity.

In practice, coreference resolution supports many natural language processing applications, including machine translation, document summarization, question answering, and information extraction. Evaluation typically uses benchmark datasets and metrics such as MUC, B3, CEAF, and related CO NLL shared-task scores. Key challenges include resolving long-distance dependencies, pronoun ambiguity, zero pronouns in pro-drop languages, and the need for world knowledge and discourse context to disambiguate references.