Argument Mining

Argument mining is an enormously challenging task: processing unrestricted text in natural language to recover inferential structure.

Our models of argument have matured and stabilised, and our datasets have expanded to allow work to begin in earnest.

Our own approach is to combine some statistical techniques (such as LDA) with insights that are derived form more structural approaches to both language in general and argumentation in particular.

We have worked both with monological arguments in various domains and also with transcripts of dialogical discussions. Early results on the former have been presented at the 1st Argument Mining workshop at ACL 2014 and again at the 2nd Argument Mining workshop at NAACL 2015. Equally early results on the latter have been presented at COMMA 2014 amongst other venues.

In September 2015, we won funding for a project worth £1.1m focused on Argument Mining.