From a book of papers entitled, Teaching Collocation: Further Developments in the Lexical Ap-proach, (ed. Michael Lewis), comes:
I was never sure whether the context was natural or typical. Unless one knows that the collocation one is learning is absolutely characteristic of the way the word is used, more than half the value one gets from learning the word in its context disappears.
So said Michael Hoey in 2000 (p.233), the same Michael Hoey of Lexical Priming fame. In order to consider something worth adding to our learning dossiers we need to find out how typical it is, and a corpus can provide us with enough data to this end. It is also worth bearing in mind, “Authenticity does not automatically entail typicality”, (Gries 2008:425), or in other words, we cannot say that a construction is typical just because it occurs in an authentic text.
Extracted from Discovering English with Sketch Engine (Thomas 2015)
In his book, Lexical Priming: A new theory of words and language (Routledge, 2005), Michael Hoey of Liverpool University demonstrates the likelihood of chunks of language that we find in any piece of text, be it spoken or written, having occured numerous times previously.
There is an example/demonstration of this starting at Slide 18 in the presentation, Using corpora while writing.