In the latest years, natural models of argumentation and argument mining are becoming more and more important topics in the argumentation community. This increasing importance is also due to the development of frameworks combining natural language processing tools together with argumentation theory with different purposes, from the analysis of online debates up to the classification of legal arguments. Given this tendency, there is the need to produce datasets on which natural language approaches to argumentation can be evaluated.
NoDE is a benchmark of natural language arguments extracted from different kinds of textual sources. It is composed of three datasets of natural language arguments, released in two machine-readable formats, i.e., the standard XML format, and XML/RDF format adopting the SIOC-Argumentation vocabulary (extended). Arguments are connected by two kinds of relations: a positive (i.e., support) relation, and a negative (i.e., attack) relation, leading to the definition of bipolar argumentation graphs.