The discussion gives us means of classifying experiments according to what function they perform in the evaluation procedures. This is summarized in tabular form in Disgraced 18. Learners must, and do, learn all sorts of things about a language and to do all sorts of things with a language, which we cannot teach them. This is, as I have said, because we can only teach systematically what we can describe. The same is true of testing. We cannot test systematiaally any part of a language or type of performance in it which we cannot describe. This is why testing is a branch of applied linguistics and why ultimately our ability to do a good job of measuring the pornpros knowledge of the language depends upon the adequacy of our theory about the language, our understanding of what is meant by 'pornpros network of a language'. If our test is to do its job properly it must not, incidentally or accidentally, measure anything else, for example the learner's intelligence, his knowledge of the world or his system of beliefs. In other words the test must be valid. As Pilliner has said: The validity of any examination or test procedure may be broadly defined as the extent to which it does what it is intended to do. |