Preposition placement: The need for corroborating evidence

It is well known that linguistic generalizations based on corpus data face two potential problems: (1) just because a phenomenon cannot be found in a corpus, it cannot be concluded that it is ungrammatical (the ‘negative data’ problem), and (2) just because a construction appears in a corpus it does not automatically follow that it is grammatical (the ‘performance’ problem). Introspective grammaticality judgements, on the other hand, are not flawed by these problems but the sentence stimuli used in such studies (1) have to be invented by the researcher (the ‘unnatural data’ problem) and (2) thus do not allow the investigation of contextual factors such as the level of formality (the ‘context’ problem). As the quote above shows, this complementary nature of corpus and grammaticality judgement data leads McEnery and Wilson to argue for a combination of both methods, instead of choosing one over the other.

While many researchers still tend to draw on either corpus or introspection data, I have argued repeatedly (cf. Hoffmann 2006, 2007a) that the approach suggested by McEnery and Wilson can yield insights well beyond what the two data sources would allow individually.