skip to content

Debate Summarisation for Decision Support

Contact: Paul Piwek (P.Piwek 'at'

In recent years, online collaborative debating/argumentation tools - such as Cohere, Debategraph, and Argunet - have been used both for entertainment and serious policy making processes. These tools rely on expressing arguments as graphs. The nodes in such graphs stand for ideas, claims, opposing claims, supporting claims, rebuttals, etc. The connections between the nodes carry labels such as “oppose”, “support”, “evidence”, etc.

Though graphs provide an intuitive and easily navigable view of an argument, visual presentations are not necessarily the best means for supporting decision-making processes (Reiter, 2006). The aim of this topic is to explore the prospects of automatic generation of text summaries for argument/debate graphs. You will develop a prototype system for summarising existing argument graphs and evaluate the quality and usefulness of such summaries.

Depending on your interests and skills, the research problem could be, for example, to solve the technical problem of generating adequate linear structures (i.e., a text) from a hierarchical structure (an argument map) or, alternatively, to gain a better understanding of the role that argument maps and their summaries can play in real-world decision-making.

Ideally, the approach will be deployed and evaluated in a real-world scenario (e.g., in your working environment).

Background reading

Reading on text generation:

Reiter, E. and R. Dale (1997). Building Applied Natural-Language Generation Systems. Journal of Natural-Language Engineering, 3:57-87. URL:

Reading on argument graphs/maps:
Buckingham Shum, Simon (2003). The roots of computer supported argument visualization. In: Kirschner, Paul A.;Buckingham Shum, Simon J. and Carr, Chad S. eds. Visualizing Argumentation: Software Tools for Collaborative and Educational Sense Making. London: Springer-Verlag, pp. 3–24. URL:

Reading on text generation for decision-support
Reiter, E. (2006). Natural Language Generation for Decision Support. Technical Report AUCS/TR0602, Department of Computing Science, University of Aberdeen, UK. URL:

Examples of Online Collaborative Argument Mapping:

Website of the Special Interest Group on Natural Language Generation:

Research methods
• Collection of argument graph/maps from existing online repositories
• Creation of target texts for the maps (see Reiter & Dale, 1997)
• Development of prototype system implementing a particular approach for verbalising maps
• Evaluation of textual summaries of generated maps with human participants

Particular equipment, skills or resources needed.
• Programming skills in language of choice
• Access to 5-10 people (e.g., in the workplace) for evaluation