RST Discourse Parsing with Tree-Structured Neural Networks

Discourse structure has a central role in several NLP tasks, such as document translation, text summarization and dialogue generation. Also, text-level discourse parsing is notoriously difficult for the long distance of discourse and deep structures of discourse trees. In this paper, we build a tree-structured neural network for RST discourse parsing. We also introduce two tracking LSTMs to store long-distance information of a document to strengthen the representations for sentences and the entire document. Experimental results show that our proposed method obtains comparable performance regarding standard discourse parsing evaluations when compared with state-of-the-art systems.