We have already referred to discourse in relation to discourse grammar, ie. grammar which extends beyond the sentence. Let's return to it now, and look at an example of what corpora can tell us about longer stretches of text. For example, McCarthy, (1998 chapter 5) looked at grammatical patterns involving verb tenses which spanned several sentences or more of written texts, and several clauses and/or speaker turns in spoken texts. We will focus here on the pattern of 'used to' plus 'would'. Although this is most typically a written phenomenon, McCarthy found that in both spoken and written data the initial 'used to' provides a contextual frame, which allows the listener or reader to interpret the subsequent occurrences of 'would' as 'past habitual'. The following example is a spoken one:
[Speakers 1 and 2 are describing how they partook in a consumer survey which involved a remote computer automatically ringing their home telephone to collect data in the middle of the night]
It is now widely accepted that naturally occurring speech contains a large number of expressions which allow speakers to stage and monitor ongoing talk. These expressions are known as discourse markers, and include such things as 'well', 'so', 'like' and 'I mean'. Their use as discourse markers explains their extremely high frequency in spoken English. There has been more than a considerable amount of research in this area, in a huge variety of languages. As a result, corpus linguists would argue that textbook writers cannot ignore these features when they are presenting students with spoken texts, whether they are being used to present grammar or vocabulary, as listening texts, or as models for spoken language.