Coreference Resolution with Mention Representation Using a Convolutional Neural Network

초록

Coreference resolution is to identify the reference relationship of different expressions in a document. In the previous studies, machine-learning models using sieve rules as input features have showed good performances. However, it is very time-consuming and labor-intensive that design effective sieve rules. In this paper, we propose a coreference resolution model that reduces the feature engineering cost by using a convolutional neural network. The proposed model showed better performances (= F1-measure of 0.6857 at CoNLL) than the previous model using complex sieve rules at all representative evaluation measures.

키워드