Efficient Grammatical Error Correction with Hierarchical Error Detections and Correction

Noisy text is common in semantic services and can have bad effects. Grammatical Error Correction (GEC) can be used to improve text quality, but traditional neural machine translation approaches need hundreds of milliseconds to correct a single text, which is unacceptable to time-sensitive services. To improve the efficiency of GEC, we choose to detect errors first and then make corrections. We pre...