Abstract

Shallow discourse parsing is the task of parsing a piece of text into a set of discourse relations. Although the characteristics of a discourse relation varies between theories, they are regarded as the building blocks of the discourse consisting of two discourse units and an optional connective. The relations without any overt connective are known as implicit discourse relations and are shown to be the bottleneck in discourse parsing. In my talk, I will present my ongoing work on shallow discourse parsing where I focus on implicit relations and aim to: (i) improve the performance of the existing systems by transferring knowledge from similar sentence level tasks, e.g. natural language inference, connective prediction (ii) develop a multilingual model which does not require any annotations in the low resource (Non-English) languages.