Computational Linguistics & Phonetics Computational Linguistics & Phonetics Fachrichtung 4.7 Universität des Saarlandes

Wei Shi - Projects


SFB-1102 Information Density and Linguistic Encoding (funded by the DFG)
Project B2 Congnitive Modelling of Information Density for Discourse Relations
Principal Investigator: Vera Demberg

The goal of this project is to learn about whether there is support for the uniform information density hypothesis at the level of discourse relations. Specifically, the project will test whether the presence of optional discourse relation markers can be explained by information density. A second hypothesis holds that discourse relation surprisal can account for processing difficulty. These hypotheses will be tested on corpora as well as in online experimental studies. The project also proposes to build a broad-coverage model of discourse relation surprisal, where each word in a text is regarded as a potential cue conveying some information about discourse relations in the text. An important aspect for building such a computational model is the cognitive plausibility of the discourse relation inventory, which will also be addressed.