End-to-end Neural Coreference Resolution

We introduce the first end-to-end coreference resolution model and show that it significantly outperforms all previous work without using a syntactic parser or hand-engineered mention detector. The key idea is to directly 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 by 3.1 F1 using a 5-model ensemble, despite the fact that this is the first approach to be successfully trained with no external resources.

PDF Abstract EMNLP 2017 PDF EMNLP 2017 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Coreference Resolution CoNLL 2012 e2e-coref + ELMo Avg F1 70.4 # 15
Coreference Resolution CoNLL 2012 e2e-coref (single) Avg F1 67.2 # 17
Coreference Resolution CoNLL 2012 e2e-coref (ensemble) Avg F1 68.8 # 16
Coreference Resolution OntoNotes e2e-coref F1 67.2 # 22

Methods