Political debates - a vital source of information for the humanities
Political debates present politicians with an opportunity to win people over to their side while simultaneously taking on their opponents, employing all of the rhetoric they have at their disposal. This involves building arguments to justify actions they have taken in the past, the policies they are putting forward and their social or political positions and beliefs - or, on the contrary, attacking those of their opponents.
Political debates attract huge viewing figures (according to the Commission on Presidential Debates, the first debate between Richard Nixon and John F. Kennedy on 26th September 1960 was watched by a television audience of around 66.4 million), but also provide an endless source of information for social, political and historical research. Accordingly, analysis of them - through fact checking, detecting implicit or untruthful arguments, or identifying links between arguments - is vitally important when it comes to understanding the world around us.
Using AI to extract arguments during political debates
For detailed analysis of information from political debates - which constitute a vast quantity of resources - researchers in the digital humanities now employ artificial intelligence (AI) algorithms.
As data becomes increasingly available, chiefly as a result of the digitisation of resources and the constant generation of digital data (such as social media posts), this has made AI more attractive to researchers, both in the digital humanities and in a range of other fields.
It can be used to automatically extract and analyse different aspects of arguments made in political debates, such as the ethos, how arguments are linked together, the facts that are shared and the arguments that are made, whether implicit or untruthful.
Argument extraction has attracted a lot of attention in the past decade, ever since the research carried out in 2009 by Mochales Palau and Moens, who introduced argument extraction in a legal context. Since then, argument extraction has been applied to a whole host of textual resources, ranging from essays written by students (Stab and Gurevych 2017) and clinical trials (Mayer, Cabrio and Villata 2020) to scientific articles (Teufel, Siddharthan and Batchelor 2009) and content generated by web users (Stab et al. 2018). As a natural example of argumentation, political debates have proved to be a useful field of analysis for the argument extraction community. However, there are very few results on this type of text, largely because of the difficulty of the task and the complexity of the debates themselves.
DISPUTool - a comprehensive debate analysis tool
But none of the argument extraction methods already in use had tackled the issue of developing an end-to-end solution for exploring arguments made in political debates.
This was the challenge that the team behind DISPUTool set themselves, a team made up of experts in AI from Université Côte d'Azur (Elena Cabrio, university teacher), from the CNRS (Serena Villata, CNRS research scientist, and Pierpaolo Goffredo, PhD student at 3IA Côte d’Azur) and from Wimmics project-team, in collaboration with the Digital Humanities department at the University of Luxembourg (Shohreh Haddadan, a PhD student at the University of Luxembourg). Their aim was to meet a growing need for tools with the capacity to make life easier for researchers in the humanities by developing an online tool which employs AI to carry out automatic analysis of the arguments used in political debates.
DISPUTool is made up of three main tools:
- Analysis of the arguments made in US presidential debates over the last 50 years
DISPUTool enables users to select different debates from the presidential elections (1960-2016), to choose from the various different debates held over the course of a given election year, and to analyse all of the arguments put forward by candidates in their debates. The current online version of DISPUTool has 29,811 argument components: 16,250 affirmations and 13,561 postulates.
- Extracting components from an arbitrary argument
DISPUTool can be used to automatically identify postulates and affirmations contained in a given text, whether taken from a political debate or not. The tool has the capacity to analyse a paragraph from a journal article, for instance, or a discussion on a social media platform, selected by the user
- Exploring US presidential debates from the perspective of an uninformed reader/viewer
DISPUTool also has the capacity to explore named entities taken from its corpus of presidential election debates. It is also able to filter these named entities by year, speaker or type of named entity.
Why US presidential debates?
There are two main factors that make US presidential debates, which have been televised since 1960, such a valuable resource for argument analysis:
- The debates have kept the same format since they began in 1960, making it possible to compare debates chronologically.
- The final debate normally pits the candidates from America’s two main political parties - the Republicans and the Democrats - against each other.
The evolution of arguments and propaganda in French presidential debates - new areas for DISPUTool to explore
The DISPUTool team are currently focused on three main areas of research: analysis of political debates in France, particularly French presidential debates; temporal analysis of their debates, in order to study how the arguments used by politicians have evolved over time; the automatic identification of examples of propaganda and lies in these political debates, identifying issues such as politicians playing the ‘authority’ card by claiming something they have said is true purely because an authority or an expert on the question said it, without any other proof, or ad hominem attacks aimed at the candidate as opposed to their political manifesto.
These arguments, the aim of which is to distract the public, are potentially dangerous, and it is vitally important to identify them in order to protect the most vulnerable in society.