Using Learner Data to Improve Error Correction in Adjective-Noun Combinations

This paper presents a novel approach to error correction in content words in learner writing focussing on adjective–noun (AN) combinations. We show how error patterns can be used to improve the performance of the error correction system, and demonstrate that our approach is capable of suggesting an appropriate correction within the top two alternatives in half of the cases and within top 10 alternatives in 71% of the cases, performing with an MRR of 0.5061. We then integrate our error correction system with a state-of-the-art content word error detection system and discuss the results.