Using Coreference Information to Improve Neural Multi-Document Summarization
Abstract
While coreference information has been shown to be a useful feature in summarization, it has received little attention with the advent of neural methods for text summarization. Recent work, however, suggests that neural approaches can also profit from explicitly modeling coreference.
While prior work by Sharma et. al. (2019) has focused exclusively on single document summarization, in this talk, I will present ongoing work on leveraging coreference information to improve content selection in neural networks for multi-document summarization. The talk will cover preliminary results on a multi-document news summarization task and highlight the challenges associated with integrating coreference information in neural MDS.
