Argumentation Mining

The task of argumenation mining involves identifying the different aspects of the argumentation structure of a text, i.e. finding the central claim of a text, supporting reasons, possible objections and counters to those objections. For us, the overall goal is to integrate the different argumentative elements and relations in a global argumentation structure spanning over the whole text. For many practical purposes, though, only certain subtasks are needed.

A variety of applications involving automatic text processing can profit from access to the argumentative structure of text, such as: the retrieval of relevant court decisions from legal databases, the analysis of scientific papers in biomedical text mining, automatic document summarization systems, essay scoring systems, as well as opinion mining applications, not only for commercial purposes, but also as a tool for assessing public opinion in political decision-making. To make argumentation structures available for these applications, robust automatic argument recognition is required, based on resources that have been created in a reproducible fashion with a coding scheme that can be reliably applied.


Our new papers on argumentation appear in the Argumentation and Computation journal 2018, at the Coling 2018 and LREC 2018 conferences, and in a book chapter on Health Technology. Also, a book on Argumentation Mining written by Manfred Stede and Jodi Schneider will appear in the autumn of 2018. See the references at the bottom of the page.

Argumentation Structure

The argumentation structure of a text is a graph representation of the argumentative relations between the propositions expressed in the segments of the text (i.e. typically sentences or clauses). It identifies the central claim of the text, supporting premises, possible objections and counters to these objections.

We devised a scheme for annotating such argumentation structures (Peldszus Stede 2013), and worked on the automatic recognition of those structures via joint optimization (Peldszus/Stede 2015, Afantenos et al. 2018).

Argumenation structure example
Fig. 1: An example argumentation structure

Semi-automatic Analysis of Arguments in Social Media

To find recurring arguments in debates on a certain issue, we have constructed a tool with which computer-assisted argument extraction from large text collections can be carried out. The tool (i) automatically selects a subset of the text collection that contains re-occurring topics, to minimise the amount of text that the human coder has to read, and (ii) presents selected texts and extracted information in a graphical user interface that facilitates manual coding of arguments. A video demo of the tool can be found here.

Argument Synthesis

In collaboration with colleagues in Weimar, we started work on re-synthesizing new argumentative text, using the microtexts pertaining to a common topic as input. Our first step is a human annotation study, presented in (Wachsmuth et al. 2018).

Related Resources

We distribute a corpus of short argumentative texts (parallel in English and German; annotated according to the scheme mentioned above) and present results on automatically recognizing the argumentation structures (e.g., Peldszus/Stede 2015, Afantenos et al. 2018). The corpus has also been annotated with other layers of linguistic information, as explained on the corpus page.

Further, we offer the annotation tool that has been used for building the argumentation structure graphs in the microtext corpus: (GraPAT).

Annotating in GraPAT
Fig. 2: Annotating argumentation structure in GraPAT

Related publications: