Fig. 1From: A multi-task learning framework for efficient grammatical error correction of textual messages in mobile communicationsThe architecture of our model. The input sentence is first checked to be incorrect by the Discriminator, then located errors’ positions by the Detector, and finally corrected by the Corrector. Two errors are detected and the maximum correction length M is set to 3 in this example. N/A refers to any token as it occurs after token ‘[EOP]’Back to article page