Computational Linguistic Approaches to Digital Conversations: the Case of Intensifiers

Presented by: Tatjana Scheffler from Ruhr-University Bochum
Date: March 17, 2023

Abstract: Intensifiers (really, so, very) are a frequent phenomenon typical of informal speech. In sociolinguistic studies, they have been shown to be highly variable and undergo constant innovation. They are also a common occurrence in casual conversations online. In this talk, I present recent research targeting medium specific, register specific, and individual variation in the use of intensifiers in German as a case study. I use a large Twitter data set as well as a new corpus of blog posts and tweets from the same authors to investigate the role of semantic differences between intensifiers, as well as the medium, register, and individual author properties to model intensifier choice. I argue that intensifiers share a core meaning component, but differ mainly in 'expressivity', a non at issue contribution that (roughly) indicates the level of emotional involvement of the author. I use information theoretic measures (surprisal) to model this expressive component. Based on this model, I test the predictions of the Uniform Information Density hypothesis to explain the existence of 'stacks' of more than one intensifier in the same phrase as well as their order. Finally, I present results on the automatic detection of intensifiers in text using ML classifiers, opening up a potential avenue to study linguistic creativity with computational means.

Location: Attend in person at J411 or via Zoom,

Time: 13:15-15:00