Tracing and classifying German intensifiers via information theory. (March 2023)

• Record Type: Journal Article
• Authors: Scheffler, Tatjana; Richter, Michael; van Hout, Roeland
• Is Part Of: Language sciences. Volume 96(2023)
• Publication Date: 2023-03
• Subjects: Intensifiers -- Surprisal -- Expressiveness -- Information theory -- German -- Corpus linguistics; Linguistics -- Periodicals; Language and languages -- Periodicals; Linguistique -- Périodiques; Langage et langues -- Périodiques; Language and languages; Linguistics; Periodicals; Electronic journals; 405
• DOI: 10.1016/j.langsci.2022.101535
• Languages: English
• ISSNs: 0388-0001

Abstract

Abstract: We apply information theoretic notions to model intensifiers in German. We show that information theory can explain that despite their common referential meaning, some intensifiers are extremely frequent ( so ), while others are uncommon ( arsch 'butt'), and seem to induce a stronger intensifying effect. We introduce two notions to model the expressivity of an intensifier in a given message: the local (paradigmatic) information content of an intensifier (surprisal), and the transitional information, based on Markov transition probabilities. Based on a large corpus of intensified predicative adjective phrases from German Twitter data, we confirm that (1) local information and transitional information are strongly anti-correlated: The more common an intensifier is, the lower its expressive value in communication and the less it constrains following adjectival heads; and (2) "stackings" of multiple intensifiers are frequent and are used to frame and strengthen the expressiveness of the intensification. We further confirm that stackings tend to occur in incremental surprisal rank order, following the Uniform Information Density hypothesis. Our analyses show that information theory is not limited to the domain of referential semantics and morphosyntactic patterns of reduction, but offers a framework to capture the expressive function of German intensifiers. Highlights: Information theory can model the large number and variety of intensifiers in digital communication. Intensifiers primarily differ in their expressivity, the more frequent an intensifier is, the less expressive (strong) it is. Intensifiers are often further strengthened by lengthening or capitalization, or by stacking several intensifiers. In intensifier stacks, more frequent/less expressive intensifiers precede the stronger, unusual ones. The preferences in intensifier ordering are predicted by the Uniform Information Density hypothesis.