Argument Mining

Argument and debate form cornerstones of civilised society and of intellectual life. Processes of argumentation run our governments, structure scientific endeavour and frame religious belief. Recognising and understanding argument are central to decision-making and professional activity in all walks of life, which is why we place them at the centre of academic pedagogy and practice; it's why such a premium is placed upon these skills; and it's why rationality is one of the very defining notions of what it is to be human.

Our theories of how argument is structured go back to Ancient Greece. In the past thirty years or so, the computational sciences have started to build models and engineer software based on these theories: this is the field of argument technology, and the recent surge in activity is testament to the vitality and broad applicability of the field. Though argument technology, in which the UK is a world leader, has had applications in domains as diverse as healthcare, public policy, government and the media, the focus has been squarely upon Artificial Intelligence technologies for supporting human argumentation and subsequent automated reasoning with the results. Arguments made outside such software walled gardens have been off the agenda simply because automatic machine understanding of unfettered naturally occurring reasoning has been too hard to tackle.

Before 2014, that task -- argument mining -- had been tackled only speculatively and only in specific domains by a very small number of groups such as those at Toronto, Leuven and Dundee. By the end of 2014, more than twenty research labs across the US and EU were gearing up to tackle the problem, there were several international meetings including a regular workshop series at the largest computational linguistics conference, and dozens of results being reported. The reason for this huge upswing in activity lies in maturing technology and the returns available. Opinion mining has transformed the way that market research and PR is carried out, deploying big data analysis techniques to understand the attitudes people hold towards products and brands. Sentiment analysis has had an even greater impact in predicting financial markets by analysing broad moods and perspectives that are expressed in the press. Argument mining is the natural evolution of these technologies, providing a step change in the level of detail available -- moving from not just analysing what opinions people hold, but why they hold the opinions they do. This is why major organisations such as IBM, with whom we are partnering in this project, are so interested in the technology.

The Centre for Argument Technology now curates the largest publicly accessible corpus of analysed argument in the world, and has a well known and widely used tool stack for managing datasets, conducting analyses, and visualising the results. This provides a unique platform from which we can both extend existing techniques for argument mining, but also, much more ambitiously, use insights from the philosophy of argumentation and from rhetoric to transform the reliability and applicability of argument mining technology. In particular, we will use the theory of argumentation schemes that characterises stereotypical patterns of reasoning to guide the process of searching for argument components, and the theory of rhetorical figures and tropes as the basis for developing a new class of algorithms for argument recognition. We will thus be transforming bare statistically-driven approaches with detailed theories of structure which can act to define expectations in a way that constrains the machine learning task thereby improving accuracy and applicability. By partnering with IBM and J&L Techology (a domain-specific SME), the project aims not just to radically improve performance of these techniques, establishing the UK's position at the cutting edge, but also to deliver those performance gains to end users.

The Centre for Argument Technology has a sustained, fifteen-year track record of delivering transformative research results that have enabled both engineered infrastructure (see aifdb.org) and practical applications (arg-tech.org). Argument Mining represents an area in which the UK has the potential to be the world leader if it capitalises upon its early successes in this high potential global market (estimated to be worth $10bn by 2020). This project aims to deliver foundational results in its first three years which are then scaled up and deployed in the fourth year.

One of the challenges in developing a transformative programme of research is that, on the one hand, working with too many partners quickly exhausts the effort available to manage the relationships effectively. On the other hand, working with too few risks over-specialising techniques to a particular user and, in the worst case, introducing a single point of failure. Our approach to this dilemma is to invite a number of companies (such as Amazon and the BBC with both of whom we already have a relationship) to engage in a series of meetings in year four, but more importantly to work closely throughout the project with two commercial partners -- companies that have substantially different profiles.

IBM is re-aligning much of its service provision around Watson, a cloud-based tool-suite to which Argument Mining is a natural companion. IBM thus represents a 'channel partner': a way for the results of the project to delivered to users of IBM's services. We will also work with J&L Technology, an SME working on a single product in a single domain, but a product that is likely to be highly disruptive in a domain that is traditionally woefully under-served by technology. For J&L's offering in the market, Argument Mining introduces functionality that represents an important unique selling point. J&L is currently piloting its software in one of the top ten organisations in its domain in the world (an organisation with an annual IT spend in excess of $25m). J&L thus represents an 'end user' of Argument Mining technology.

Despite their differences, both IBM and J&L also act as 'thought leaders' in their respective markets: they are seen by their customers to be establishing the trends that others will follow. If the techniques developed in this project were to be advocated by our two commercial partners, there is potential for very widespread impact as a result of this thought-leadership. Our programme of collaboration involves both formal structures (including quarterly meetings of the project's Commercial Board with representatives from both companies), and practical work practices, resting heavily on a series of bilateral exchanges of staff including over six months of residencies for University project staff with the partner organisations. The Commercial Board ensures that ad hoc, responsive communication is complemented by regular, scheduled interaction, whilst the residencies are a key way of enabling staff in the University to gain deep insight into the challenges and opportunities presented by the domains of the commercial partners, and at the same time for staff at the partners to better understand the problems and solutions investigated in the project. To ensure that the project's work is disseminated more broadly in the commercial partners (which is particularly important for such a large organisation as IBM), we will also run Demo Days on site at the partners, with open invitations to staff from across the organisation to see project results and explore opportunities.

For both IBM and J&L Technology, the project represents an area of strategic importance, which is why the two companies are together committing contribution in kind estimated to be worth over £272k, thereby significantly increasing the value for money for the public purse.