In this paper, authors explore the problem of developing an argumentative dialogue agent that can be able to discuss with human users on controversial topics. The authors describe two systems that use retrieval-based and generative models to make argumentative responses to the users. The experiments show promising results although they have been trained on a small dataset.

Research in argument mining has mainly focused on the problem of identifying claims, premises (Boltuzic and Snajder, 2014, 2015; Levy et al., 2014), assessing arguments, classifying stances, detecting political beliefs (Hasan and Ng, 2013; Iyyer et al., 2014; Bamman and Smith, 2015) or finding the connection between claims (Stab and Gurevych, 2014). Very few research has addressed the problem of generating arguments directly in a conversational form.

To study and analyze debates, it is important to understand how to formulate claims, how arguments develop and relate to each other, what factors influence the next argument. In this work, the authors explore the question of whether the authors can teach computers to make or generate arguments and follow the ideas/stances/sides of actors in a debate. To start inspecting this challenging problem, the authors develop two debate dialogue systems, a retrieval based and a generative model. The aim of the system is to mimic a debater, make arguments and give relevant responses to users on given topics.

Such argumentative dialogue systems could be useful in a lot of future applications, such as in information campaigns, where the users can get objective answers for controversial topics to make evidence-based decisions; in an interactive argumentative dialogue system, where the users can practice making arguments, learning to persuade people.

Authors: Dieu Thu Le (University of Stuttgart), Cam-Tu Nguyen (Nanjing University), Kim Anh Nguyen (FPT University)

See more HERE.

Related posts: