Classifying Argumentative Relations Using Logical Mechanisms and Argumentation Schemes

While argument mining has achieved significant success in classifying argumentative relations between statements (support, attack, and neutral), we have a limited computational understanding of logical mechanisms that constitute those relations. Most recent studies rely on black-box models, which are not as linguistically insightful desired. On the other hand, earlier use rather simple lexical features, missing statements. To overcome these limitations, our work classifies based four theory-informed two statements, namely, (i) factual consistency, (ii) sentiment coherence, (iii) causal relation, (iv) normative relation. We demonstrate operationalization without directly training data labeled with relations, significantly better than several unsupervised baselines. further also improve supervised classifiers through representation learning.