This paper aims to illustrate one way in which corpus-linguistic methods and specialised corpora can be combined in work on academic discourse. It reports selected findings from a study of social interaction in research articles written by German, British and US-American humanities academics, based on the 1-million-word SCEGADcorpus. While the main interest of the project was in possible cultural differences in academic discourse, statistical analysis was used to examine the influence also of linguistic background, discipline, author age, status and gender on the construction of identity and the encoding of social relations in academic writing. The findings reveal significant cultural differences, but also demonstrate the influence of variables such as discipline, gender and academic status on author-reader interaction and identity construction in scholarly texts.