Link to the paper Building an English Grammar-Checker with a Recurrent Neural Network
Abstract
In the face of the numerous theoretical challenges in training a neural network to classify any given English sentence as being either grammatical or ungrammatical, we train and compare a few configurations of Recurrent Neural Networks on exactly that task. Sentences are encoded as sequences of vectors, each denoting one word’s part of speech. The resulting network is surprisingly effective at differentiating between grammatical and shuffled “word salad” sentences double the length of those on which it is trained.
$ python GrammarBrain/brown_data/util/Experiments.py
Follow the example of GrammarBrain/brown_data/experiment_scripts/gen_error.py,
which uses the dictionary from GrammarBrain/brown_data/util/Experiments.py
GrammarBrain can be downloaded, used, modified, etc., for any purpose.