Expectation Violation Mining: Detecting and Explaining Surprising Argument Outcomes

Drawing inspiration from Hunt et al. (1989) on expectancy-discrepant persuasion and Kane (2017) on processing narratives, this idea proposes a mining system trained to flag instances where outcomes (agreement, opinion change) contradict what models or theories would predict based on argument features. For example, why does a proposal with seemingly poor logical support succeed? Or why does a “strong” argument backfire? The system would use ML and explainable AI to surface these cases, analyze contributing factors (emotion, social context, argument structure), and generate hypotheses about new persuasion mechanisms. This goes beyond current argumentation mining, which tends to map what usually works, by systematically exploring exceptions—potentially leading to new theories of persuasion and negotiation.

References:

1. Don't judge a book by its author: Central and peripheral processing in narrative persuasion. K. Kane (2017).
2. Processing Effects of Expectancy-Discrepant Persuasive Messages. James M. Hunt, Michael F. Smith, J. B. Kernan (1989).