Coreference Resolution in Ambiguous Pronouns Using BERT and SVM

Coreference resolution is the method of matching the expressions in a text to the corresponding entity which it is referring. This is a major task in many Natural Language Processing problems and a prolonged challenge in the field of NLP. One of the major challenges in the existing coreference resolution systems is the absence of a better solution in the task of ambiguous pronoun resolution. So in this work, a coreference resolution system for ambiguous pronoun resolution using BERT based approach is proposed. First, the dataset is trained on a BERT model for obtaining the contextual embeddings in the text and then applying it to the SVM classifier for classification and thus obtain the coreference resolution for the target pronoun. The dataset used in this work is Gendered Ambiguous Pronouns (GAP) dataset, released by Google AI Language.