Abstract
This paper describes an experiment aimed at improving the quality of coreference resolution for Russian by combining one of the most recent developments in the field, employment of neural networks, with benefits of using semantic information. The task of coreference resolution has been the target of intensive research, and the interest at using neural networks, successfully tested in other tasks of natural language processing, has been gradually growing. The role that semantic information plays for the task of coreference resolution has been recognized by researchers, but the impact of semantic features on the performance of neural networks has not been yet described in detail. Here we describe the process of integrating features derived from open-source semantic information into the coreference resolution model based on a neural network, and evaluate its performance in comparison with the base model. The obtained results demonstrate quality on par with state-of-the-art systems, which serves to re-establish the importance of semantic features in coreference resolution, as well as the applicability of neural networks for the task.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call