The article addresses the automated extraction of terminological expressions denoting professional competencies from official-business discourse, including job descriptions and regulatory acts. The study develops a linguistic model for ranking term units that accounts for the frequency, length, and nesting of word combinations and their contextual association with competency markers. The methodology combines corpus analysis of a professional sublanguage, selection of terminological candidates by grammatical patterns of noun groups, and a quantitative measure of term significance uniting statistical and contextual components as a convex combination. The contextual channel is parameterized by the weights of the predicates «to know», «to be able to», and «to master» and by their co-occurrence with a candidate within a given window. The novelty lies in the formal mechanism for accounting for the nesting of multiword terms, the parameterization of the contextual channel, and the projection of extracted expressions onto the «to know – to be able – to master» typology with ontological tracing. The model prioritizes complete multiword terms used as requirements and filters out semantically dependent fragments. The approach is applicable to the formalization of professional discourse, the construction of domain-specific terminological resources, and the design of professional language training programs.