Using learner corpora for automatic error detection and correction

In this chapter we discuss the use and importance of learner corpora for the development and evaluation of automatic systems for learner error detection and correction. We argue that learner corpora are crucial in three main areas in this process. First, these corpora play an important role in identifying and quantifying common error types, in order to prioritize development of error-specific algorithms. Second, learner corpora provide valuable training data for machine-learned approaches which are dominant in the field of natural language processing today. Finally, the evaluation of error detection and correction systems is most reliable and realistic when performed on real learner data.