This chapter provides an overview of corpus-based advances in Construction Grammar. After a brief introduction on kinds of data in linguistics in general and the notion of corpora in particular, I discuss a variety of corpus-based studies categorized into (i) largely qualitative studies, (ii) studies based on frequencies and probabilities, (iii) studies focusing on association strengths, and (iv) statistical as well as machine-learning studies. In each section, representative studies covering a variety of languages and questions are covered with an eye to surveying methodological as well as theoretical advantages. I conclude with an assessment of the state of the art by comparing how recent developments fare relative to Dąbrowska’s discussion of Cognitive Linguistics’s seven deadly sins.

This is a ‘quantitative methods’ chapter. It explains the consequences of more advanced developments in quantitative corpus techniques. Two themes are developed. Firstly, some advanced measures improve the range of options available to researchers studying, for example, collocation or keyness. This shows that different measures may give different results. Secondly, some applications of statistical techniques allow the contents of a corpus to be rearranged, leading to novel insights. It is argued that these bottom-up approaches amount to quantitative ‘corpus-driven’ methods. Topics discussed in the chapter are keyness, collostructions, text clustering using lexical bundles and multidimensional analysis, and word clustering using principal component analysis and topic modelling.