Movie Scene Argument Extraction with Trigger Action Information

Abstract

Movie scene argument is an essential part of the movie scene. Movie scene argument extraction can help to understand the movie plot. In this paper we propose a movie scene argument extraction model, which utilizes the trigger action paraphrase as extra information to help improve the argument extraction. Specifically, we obtain the paraphrase of trigger from the dictionary and employ attention mechanism to encode them into an argument oriented embedding vector. Then we use the argument oriented embedding vector and the instance embedding for argument extraction. Experimental results on a movie scene event extraction dataset and a widely used open domain event extraction dataset prove effectiveness of our model.