Comparing corpus-based and GenAI-based tools: The analysis of specialised discourse using Sketch Engine and ChatGPT-5.1

This study examines the performance of Sketch Engine and ChatGPT-5.1 in identifying specialised lexical phenomena within a medical subcorpus of the MuchMore Springer Bilingual Corpus. It examines three analytically distinct layers of specialised discourse – multiword terms, adjectival collocates, and four-word lexical bundles – to assess how corpus-driven and generative approaches capture terminological, collocational, and formulaic patterns in anaesthesiology. Results show substantial convergence, particularly in lexical bundles, where 84% of units overlapped and ranked similarly. Term extraction and collocation analysis also revealed a stable conceptual core shared by both tools, while divergences reflected either corpus-specific distribution (Sketch Engine) or conceptual salience (ChatGPT-5.1). The findings suggest that the complementary use of corpus tools and generative artificial intelligence (GenAI) may offer potential applications for terminology-oriented teaching contexts, such as seminars in corpus linguistics or terminology studies, where students can critically compare corpus-based evidence with model-generated linguistic interpretations.