About the NUCLE dataset

[14484...@qq.com]

In the NUCLE Release 3.3 dataset I request, the file nucle.train.gold.bea19.m2 only has about 50000 sentences. I wonder it is correct ?

[BEA 2019 Shared Task Organisers]

That is correct.

Here are the statistics for all the datasets allowed in the Restricted Track:

FCE: 33,237 sentences (train/dev/test)

Lang-8: 1,037,561 sentences

NUCLE: 57,151 sentences

W&I+LOCNESS: 43,129 sentences (train/dev/test)

Although Lang-8 is the biggest, it is also the noisiest of these datasets. All the others were professionally annotated with GEC in mind and should be of higher quality.