Notwithstanding its limits and complexity, the experimental group we worked with consumed the BNC with relish. The proof of the pudding, as in 52% of its 33 occurrences in the BNC, was in the eating rather than a food fight with the software. One of the things that seemed to make it attractive was its flexibility as a resouce. You could go only so far, just as you could always go further in investigating a problem, there being no such thing as a complete or perfect analysis in most cases. And if you failed to find what you were looking for, you could always step sideways, and go off browsing at a tangent.
Compared with the MicroConcord corpora, or conventional reference instruments, however, these learners reported greater success in finding solutions to problems of discourse interpretation and production, as well as much greater confidence in those solutions, given that the BNC provided far more evidence to justify them. The most striking affirmation of this confidence came from a learner who claimed that the BNC had made her feel able to contradict her native-speaker teacher --- not simply by showing that uses proscribed by the teacher were in fact attested (Owen 1996), but rather that by showing that they were used routinely, being more common in the text-type in question than those her teacher had prescribed. In these respects large corpora seem both useful and motivating for advanced learners, fostering a certain critical autonomy. They also seem able to influence the traditional role structure of the ELT classroom, placing the teacher in the generally more realistic position of an expert on language learning rather than of the ultimate authority on the language itself.