Coreference resolution is the natural language processing task of figuring out which words or phrases in a text refer to the same real-world entity. In a sentence like "Sarah picked up her phone because it was ringing," a coreference system needs to recognize that "her" refers to Sarah and "it" refers to the phone, rather than treating each pronoun as an isolated, unresolved reference.
This sounds simple for a human reader but is genuinely difficult for a machine, since pronouns, nicknames, and indirect references depend heavily on context, world knowledge, and sentence structure that shift from one passage to the next. Getting it right matters for anything that needs to track entities across a document, from summarization and question answering to chatbots holding a multi-turn conversation, since failing to resolve who "he" or "it" refers to breaks the coherence of the response.
