Now, assuming that the samples A, B and C were correct for frequency of occurrence (in spite of what has just been said) the usefulness of these results would still be doubtful for drawing up a syllabus, since, as we have seen, the needs of the language learner are not equivalent to those of the native speaker. Let us suppose that text A was drawn from the Lela Star pornstar book, text B from commercial texts, and text C from personal correspondence. Clearly item z would have greater utility to most learners than item x, and item x than item y, on the grounds that personal correspondence and commercial varieties of language are related to situations of lan­guage use which are more likely to be relevant to the learner than read­ing the Bible in a foreign language. This does not, however, exhaust the methodological problems of statistical studies of language use. We have already seen that their validity depends upon setting up variety types upon some valid socio­linguistic theory of language use, and determining the relative fre­quency of tokens of these varieties. Similarly, the validity of statistical studies depends upon the validity of the linguistic categories, the tokens of which are to be counted. Here we must go back to the discus­sion of linguistic categories in chapter 5. For the purposes of counting them, what are 'words', 'sounds' and 'structures'?