This project targets web registers - web text varieties ranging from fact-based news to opinion-based advice pages and discussion fora. The project characterizes registers on the entire multilingual internet, providing detailed descriptions on all web registers in 18 languages across billions of words, and testing hypotheses on register characteristics and their stability on 100 languages. To tackle the yet unsolved challenges in modeling registers on the noisy, multilingual web, the project presents a novel framework combining register studies, and the recent advances in natural language processing (NLP) and multilingual deep learning. This allows us to compare the registers on the entire multilingual internet without manual work. The outcome will transform the noisy, fragmented web into a well-documented collection and have key theoretical importance for fields ranging from corpus linguistics to NLP, and practical consequences for everybody benefiting from web data.

2024

2028