This is the repository for the argumentation mining experiments on the dataset of student essays collected through the "Fair Debating and Written Argumentation" project. Details on the project, data and experiments are described in our paper (currently under review, details TBD).
We introduce an annotation scheme for analysing secondary-level argumentative student essays of varying writing qualities. In this context, we introduce a fine-grained set of argument component types (AC types) that serves as a first step to a formal description of the argumentation structure within an essay. We then conduct experiments using large language models (LLMs) to automatically predict these AC types.
The latest version of our annotation guideline (at present only available in German) can be found at data/AnnoRiLi_Schueleraufsaetze-3.pdf.
Our dataset with gold-standard annotations for AC types can be found at data/gold_data/ac_types.