Conversational AI: Dialogue-based Adaptive Argumentative Writing Support

This is a project in the area of Natural language processing (NLP). This is a subfield of Computer Science, Artificial Intelligence (AI) and Computational Linguistics concerned with the interactions between computers and human language. Within the scope of this research project, we aim to explore the potential of adaptive argumentation skill learning with designing, implementing and evaluating dialogue-based adaptive writing support systems to help students to improve their ability to argue in a structured, logical and reflective way. For this purpose, we plan to research and develop a novel machine learning-based method to automatically assess the quality of students’ argumentative texts. Moreover, we aim to investigate the ability of chatbots, using sophisticated statistical and neural network language models, as a feedback provider to guide students towards building persuasive arguments. To gain the important training date for the supervised machine learning, we first need to develop a extensive labeling scheme. Therefore, we will develop a novel rich semantic representation for fine-grained argumentation annotations capturing logical, rhetoric and dialectical quality dimensions. Based on this innovative annotation scheme, we will build a large, reusable argumentation corpus. To our knowledge, it is the first of its kind. Therefore, it can become a valuable resource for all researchers in our field. The argumentation corpus constitutes the data set for our machine learning algorithm, i.e. the algorithm for extracting argumentative discourse structures and assessing the quality of the underlying argumentation. Furthermore, we will explore and build novel dialogue-based argumentation learning frameworks. Hence, we will develop an argumentative dialogue agent, also called a chatbot, that debates with students with the objective of playfully increasing their arguing skills. This dialogue agent utilises our comprehensive argumentation corpus, as well as large language models and BERT-like architectures. Next, we will extend the dialogue system in the direction of a feedback provider that evaluates the quality of students’ argumentative texts and gives interactive feedback on their lines of argumentation. The main outcomes of this project will be (i) a novel comprehensive approach to automatically assess the quality of argumentative texts for providing adaptive static feedback; (ii) a proof-of-concept for an argumentative dialogue agent for debating with students on controversial topics with the objective of playfully improving their arguing skills; and (iii) a first prototype of a dialogue-based approach for providing students with adaptive interactive feedback on the quality of their arguments and reasoning chains. The expected results of our project are interesting for both scientific communities and practitioners. We explore an emerging application field of Natural Language Processing (NLP) and Artificial Intelligence (AI), in particular the potential of educational agents to teach students how to write persuasive texts by engaging in a conversation with the user and providing adaptive feedback about the quality of their argumentation. Moreover, the results of this research project will help companies to develop argumentation learning tools to support the continuous education of their employees, aiming to improve their arguing skills to better match the requirements of today’s job profiles. Nowadays, with the tasks demanding more and more creativity and interdisciplinary cooperation, the ability to argue convincingly is indispensable.