Abstract
This paper examines the thematic structure of a corpus of insults using the inference-boundary model of Theme and Rheme. It focuses on the concept of negative inference—which must be generated for an insult to be successfully delivered—and shows how it allows us to better understand and characterize the form that insults generally take. The analysis suggests that insults are typically structured to generate backward-looking negative inferences from the decoder, much in line with how new information (in this case, the thrust of the insult) is generally located in final position. The paper also proposes a summary statement capturing the general configuration of insults and suggestions for further research.