The paper introduces the first end-to-end neural coreference resolution model, which significantly outperforms previous methods without using syntactic parsers or hand-engineered mention detectors. The key idea is to consider all spans in a document as potential mentions and learn distributions over possible antecedents for each. The model computes span embeddings that combine context-dependent boundary representations with a head-finding attention mechanism. It is trained to maximize the marginal likelihood of gold antecedent spans from coreference clusters and is factored to enable aggressive pruning of potential mentions. Experiments demonstrate state-of-the-art performance, with a gain of 1.5 F1 on the OntoNotes benchmark and 3.1 F1 using a 5-model ensemble. The model is also relatively interpretable, providing insights into the strengths and weaknesses of its coreference decisions.The paper introduces the first end-to-end neural coreference resolution model, which significantly outperforms previous methods without using syntactic parsers or hand-engineered mention detectors. The key idea is to consider all spans in a document as potential mentions and learn distributions over possible antecedents for each. The model computes span embeddings that combine context-dependent boundary representations with a head-finding attention mechanism. It is trained to maximize the marginal likelihood of gold antecedent spans from coreference clusters and is factored to enable aggressive pruning of potential mentions. Experiments demonstrate state-of-the-art performance, with a gain of 1.5 F1 on the OntoNotes benchmark and 3.1 F1 using a 5-model ensemble. The model is also relatively interpretable, providing insights into the strengths and weaknesses of its coreference decisions.