Learning Structured Perceptrons for Coreference Resolution with Latent Antecedents and Non-local Features

• DOI: 10.3115/v1/p14-1005
• Publication Date: 2015-06-17T03:05:16Z
• Authors: Anders Björkelund; Jonas Kuhn
• Keywords: Coreference; Perceptron; Baseline (sea)

We investigate different ways of learning structured perceptron models for coreference resolution when using non-local features and beam search. Our experimental results indicate that standard techniques such as early updates or Learning Search Optimization (LaSO) perform worse than a greedy baseline only uses local features. By modifying LaSO to delay until the end each instance we obtain significant improvements over baseline. model obtains best date on recent shared task data Arabic, Chinese, English.