I would be interested to hear from anybody who has practical experience of making a learner corpus.