Abstract
The present article examines the distribution of architectural terminology across English for Specific Purposes (ESP) courses designed for future architects at three proficiency levels: elementary (A1), pre-intermediate (A2), and intermediate (B1). The empirical foundation of the study is a purpose-built corpus comprising texts extracted from three professionally oriented textbooks: Professional English in Use: Architecture (Konovalova E.N.), English for Architects (Bezruchko E.N.), and English for Construction Universities(Lukina L.V.). The corpus texts were digitised and pre-processed through case normalisation, stop-word removal, and tokenisation, then subjected to frequency and topic modelling analysis employing Python, NLTK, and gensim libraries. Frequency distributions and visualisations - bar charts, pie charts, and pyLDAvis topic maps - enabled the identification of a compact high-frequency core of architectural vocabulary (architecture, building, construction, materials, design), three thematic clusters (structural elements, materials, design principles), and one dominant LDA topic representing the conceptual nucleus of the corpus. Drawing on frequency data, topic membership, and contextual usage patterns, a level-differentiated model of terminology distribution for the A1–B1 continuum is proposed. This model informs the design of VR-based learning tasks, ranging from naming and describing basic spatial elements in a virtual environment at A1 to complex project assignments integrating form, style, and sustainability concepts at B1. The findings demonstrate that combining corpus-linguistic methods with immersive VR technologies affords a more empirically grounded trajectory for architectural vocabulary acquisition, aligning lexical selection with authentic frequency profiles and the thematic architecture of academic texts, thereby enhancing the overall effectiveness of ESP instruction.