Generating diverse sentential arguments on controversial topics with a memory-augmented generation model

Abstract

An automatic system that provides diverse perspectives on a controversial topic is one of the most important researches, which natural language processing can address. Many kinds of researches based on information retrieval and extraction techniques are in progress. However, they may be hard to provide stable results with a query that is absent from the repository. To address these problems, we propose a sentential argument generation model from diverse perspectives, augmented with the common characteristics extracted from the arguments in a training corpus. Our proposed model generates sentential arguments with higher diversity than other baseline models. We hope that our proposed model can help to provide users with a variety of perspectives that could be considered on a controversial topic, even when the topic has not been actively discussed or related information is not accessible.