The MLE model we describe in our paper exploits our alignment algorithm and the error type annotations obtained using ARETA to map erroneous words to their corrections. Running the MLE model is done using the rewriter.sh script:
train_file=/path/to/train/file
test_file=/path/to/test/file
ged_model=/path/to/ged/model
output_path=/path/to/output/file
python rewriter.py \
--train_file $train_file \
--test_file $test_file \
--ged_model $ged_model \
--mode full \
--cbr_ngrams 2 \
--output_path $output_path