The study explores the structure of English written scientific-technical discourse corpus in the area of civil engineering. The primary focus is on the retrieval of the essential content of the discourse. Various definitions and approaches to the concept of reference are analyzed within linguistics from a theoretical standpoint. The analysis is illustrated by examples. The study examines different statistical techniques for natural language processing applied in computational and corpus linguistics. Limitations of frequency data and challenges facing computational and corpus linguists are described. The study gives an insight into statistical methods of experimental analysis of scientific-technical discourse. The research results, some observations and conclusions can be used not only in computational discourse processing but also in reading a course of lectures on discourse analysis in higher education institutions.