Hybrid4SentenceARA

This page released the datasets and codes in our BEA 2023 Workshop paper Hybrid Models for Sentence Readability Assessment.

Datasets

Wall Street Journal (WSJ)

• Please refer to Brunato et al. (2018) and their website.

OneStopEnglish (OSE)

• Please refer to Vajjala and Luˇci ́c (2018).
• The directory Data contains our processed OSE dataset.

Codes

Train models

python code/main.py \
    --model AutoModel \
    --state classification \
    --batch_size 32 \
    --epoch_num 10 \
    --embedding_dim 64 \
    --lr 1e-5 \
    --n_labels 3 \
    --dataset ./datasets/ose/ 

Obtain predicted value or probabilities

python code/pred.py \
    ./trained_models/ose/roberta_32_1e-06_10/ \
    ./datasets/traindata/ose/ \
    ./Pred_results/ose/