Framing in Communication: From Theories to Computation (Dagstuhl Seminar 22131)

Abstract

Framing has become recognised as a powerful communication strategy for winning debates and shaping opinions and decisions. Entman defines framing as an action of selecting “some aspects of a perceived reality and make them more salient in a communicating text, in such a way as to promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation for the item described”. Instead of engaging in costly and difficult exchanges of argument and counter-argument, a politician or a journalist can then try to reframe a dialogue on, for example, fracking from economic benefits to environmental hazards, or a dialogue on abortion from pro-life to pro-choice. Introduced in 1960’s sociology, framing has been imported into communication sciences and media studies as an attempt to address the ways in which news is reported and, thus, a way in which to tackle manipulation and fake news. The topic has spread to other disciplines such as psychology, philosophy, semantics, pragmatics, political science, journalism, and, most recently – to computational linguistics and artificial intelligence. This seminar aims to pave the way to synthesising definitions developed in these theoretically and empirically driven areas and then to operationalise them in computational and applied areas by means of cross-disciplinary hands-on exchanges in facilitated discussions. Our goal is to support the development of innovative technologies, which can help us to quantify framing phenomena, to study framing at scale, and to deploy computational techniques in order to intervene against malicious attempts to influence opinions and decisions of the general public. for humans to fill in by reading between the lines, but where computational systems struggle. We identify the relevance of commonsense knowledge and showcase that by including such knowledge resources in downstream computational argumentation tasks we can improve system performance. We then show that background knowledge a system uses to make such implicit knowledge explicit in arguments can be generated in natural languages – which helps to make the process transparent and controllable. where a number of seed words are derived from Luhmann’s books that are discriminative of each individual system compared to the others. This is done with the goal of transfersing a trained model from the domain of Luhmann’s books to more generic text domains, such as Wikipedia, news articles, or other scientific articles. Our approach shows promising results, indicating that a classification of text into social systems is indeed possible. This may give rise to quantitative analyses of social systems in social sciences, supporting social scientists in their daily work.