Taking a cognitive approach to genre-specific language, this corpus-based study investigated the disciplinary and paradigmatic effect on the use of a specific type of attitude markers--surprise markers--with an analytical framework informed by frame semantics. A Surprise frame was generated and then used to analyze the use of surprise markers in a 2,000,000-word corpus consisting of 320 full-length empirical research articles collected from two social sciences (applied linguistics vs. clinical psychology) cutting across two research paradigms (qualitative vs. quantitative). Results from multiple binary logistic regression analyses show that the research paradigm can reliably predict the absence or presence of five categories across four frame elements of the Surprise frame. This study not only extends the application of frame semantics to discourse analysis but also has the potential to create a new direction for research in English for academic purpose.