Coreference Resolution: Linking Pronouns to Their Antecedents

Every time a language model reads "Maria met John... she smiled at him," it has to solve a small but critical puzzle: which earlier noun does each pronoun actually point to? This simulator renders that mention-pair scoring process in 3D — each mention in a short passage becomes a floating node on a timeline, and resolving a pronoun draws live-scored arcs back to every legal candidate antecedent. Two sliders let you trade semantic-embedding similarity against recency decay exactly the way a real coreference model's antecedent scoring function does, and a toggle switches gender/number agreement between a hard linguistic filter and a soft, purely-learned penalty. Three preset passages isolate different failure modes — clean gender cues, an all-neuter "it" chain that only semantics and recency can untangle, and a plural "they" pair that agreement alone can't disambiguate — with a live accuracy readout showing how each weighting choice changes how many pronouns the model actually gets right.