What if you could find all arguments in a text without having to read it? Or, what if you could search a database for a controversial topic and immediately get arguments for and against it, gathered from text all around the internet? Or, imagine when writing an essay you would automatically get an estimation of how persuasive your arguments are.
Scenarios such as these could be possible with techniques developed in the field of argumentation mining. The aim of this relatively new field is to automatically retrieve and analyze arguments and argumentation in natural language, most often in text. The name is inspired from the related field of opinion mining, where the goal is to retrieve opinions. In argumentation mining, we also want to know how people argue for those opinions. Tools for retrieving argumentation would have many uses in in many areas, such as digital humanities or the social sciences. It would enable further analysis regarding how people argue and how they infer their conclusions.
Although one could argue about what argumentation consists of, it is often assumed that an argument is made up of a conclusion and one or more premises. Given this definition, a typical argument mining task would be to identify the conclusion (bold) and the premises (italic) in the text below. After this is done, another task could be to identify the relations between the retrieved arguments this text, do they support or attack each other?
”People should stop flying, it is bad for the environment. Some say that you should keep flying because it’s the airplane company’s responsibility to compensate for the emissions. I don’t agree.”
However, this is not a simple task. Whichever method is used for this, it must learn where the boundaries of the different argument components are and the relationships between them. As the field is quite new there are only a few examples of successful argumentation mining tools.
Creating resources for argumentation mining
In order to develop successful methods for argumentation mining, text annotated with argumentation is needed. As elsewhere in NLP, the main approach to argumentation mining problems is to use machine learning, especially neural networks, and such methods need lots of training examples to learn from.
To annotate argumentation is a challenge in itself. Argumentation can be complex, context-dependent and often implicit. Argumentation can be elaborate and standardized, as in for example legal documents, or informal and short like arguments from Twitter. This means that different approaches are needed, with possibly different models of argumentation and differing needs of domain knowledge and training of the annotators. As there is no consensus or standard of how to annotate argumentation, transferring knowledge and comparing results can also be a problem.
Until recently, there have been few corpora annotated with argumentation. Now, more and more corpora are being created, although still in small sizes due to the time it takes to create them. Still, this will hopefully lead to more successful argumentation mining in the future. If you want to know more about argumentation mining you can read about our initial efforts to create a Swedish corpus annotated with argumentation here or check out this survey paper.