Transformer-Based Models for Automatic Identification of Argument Relations: A Cross-Domain Evaluation

Argument Mining is defined as the task of automatically identifying and extracting argumentative components (e.g., premises, claims, etc.) detecting existing relations among them (i.e., support, attack, rephrase, no relation). One main issues when approaching this problem lack data, size publicly available corpora. In work, we use recently annotated US2016 debate corpus. largest argument corpus, which allows exploring benefits most recent advances in Natural Language Processing a complex domain like (relation) Mining. We present an exhaustive analysis behavior transformer-based models BERT, XLNET, RoBERTa, DistilBERT ALBERT) predicting relations. Finally, evaluate five different domains, with objective finding less dependent model. obtain macro F1-score 0.70 evaluation 0.61 Moral Maze cross-domain