linguistic-features-in-ARA

To analyze the effect of some simple linguistic features on ARA.

To extract features, run store_feature.py to get shallow features from text, and run feature_added.py to get lexical, POS and syntactic features from text.

Run these code by:

bert_with_feature.py \ python bert_with_feature.py --dataset_path=[ENTER YOUR DATASET PATH] --class_num=[THE SIZE OF YOUR LABEL SET] --train_steps=[the steps between two evaluations] --data_portion=[] --max_seq_len=[max length that the PLM could process] --test_time=1

bert_without_feature.py \ python bert_with_feature.py --dataset_path=[ENTER YOUR DATASET PATH] --class_num=[THE SIZE OF YOUR LABEL SET] --train_steps=[the steps between two evaluations] --data_portion=[] --test_time=1

longformer.py \ python longformer.py --dataset=[your dataset name] --class_num=[the size of your label set]

Please make sure that your datset is in './[dataset_name]'

Datasets

RAZ and Newsela is unavaliable due to privacy reasons. OneStopEnglish: https://github.com/nishkalavallabhi/OneStopEnglishCorpus