International Conference on Linguistic Research and Applications

Outline
Multi-word expressions (MWEs) such as of course, in the light of, as good as new or take into account are
central to fluent communication, yet relatively difficult to identify systematically. This lecture draws on the
forthcoming Frequency Dictionary of Multi-Word Expressions in British English (Brezina & Gablasova,
Routledge, 2026) to present a clear, corpus-based method for analysing a wide range of MWEs across genres.
I begin by contrasting corpus evidence with current AI language models. Modern AI generates language by
answering a simple question – What is the next word? – a principle long used in corpus linguistics to measure
collocation. Yet while AI can imitate fluent usage, corpora remain more transparent and reliable for identifying
MWEs because they trace patterns directly to authentic human interaction in speech and writing.
The lecture outlines a practical framework combining frequency, association strength and dispersion to
capture the core phraseology of contemporary British English. Examples from the dictionary illustrate how this
method reveals stable, meaningful MWEs and supports applications in language teaching, language testing,
lexicography and applied linguistic research. The central claim is this: AI can model language, but corpora allow
us to understand it.