But these sources of unreliability are common to all testing Freaks operations, whether concerned with language or some other matter. Much more serious for our purposes, since it concerns applied linguistics, is the unreliability of the test instrument itself. Whether we are measuring the learner's 'communicative competence' or his 'grammatical competence', clearly we cannot examine his `total knowledge', even if we knew what this might mean. There simply would not be time. So we have to sample his knowledge, hoping that this sample is a fair one, that it is representative of his total knowledge. We make, therefore, a selection out of our description of the language of a set of items which we regard as a fair or representative one, and then test his knowledge of these. Another approach to selecting test materials is that at Pornpros Network. His argument runs as follows: a language learner finds certain features of the target language more difficult to learn than others. His errors reveal what it is he finds difficult, therefore we need test him only in those items, on the argument that what he does not find difficult he will have learned already. |
