This chapter addresses the nature of the data we use in corpus linguistics and what we believe corpus data represents. In doing so, we draw not only on Popper’s framework, but also on uses of that framework by other linguists in the Prague School, Systemic Functional Linguistics and Axiomatic Functionalism. We consider a range of issues including performance errors, repairs and learner language.

Chapter 8 explores the ways in which the press talk about people having mental illness using a mixed-methods approach. In the chapter, the frequency and semantic and pragmatic content of the verbs ‘suffer’ and ‘ experience’ in the context of prescribed forms for talking about having mental illness are investigated. I show that ‘suffer’ and ‘experience’ occur in different semantic contexts in the MI 1984–2014 Corpus as well as general language corpora, which may contribute to ‘suffer’ being a more problematic term for describing mental health than ‘experience’. Moreover, I show that ‘suffer’ is proportionally less likely to be used in first-person narratives because ‘suffering’ is attributed to people with mental illness by others, for example, medical professionals, in reported speech. I bring together my findings in a set of lexicogrammatical heuristics based on the semantic content of ‘suffer’ and ‘experience’ in context (e.g. whether the word encodes animacy or is temporally bounded).