Chinese Idiom Paraphrasing

Chinese Idiom Paraphrasing (CIP), which goal is to rephrase the idioms of input sentence to generate a fluent, meaning-preserving sentence without any idiom:

Data in this dataset and several approaches:

• LSTM approach

• Transformer approach

• mt5-seq2seq approach

• mt5-infill approach

• mt5-knowledge approach

Dependecies

• Python>=3.6
• torch>=1.7.1
• transformers==4.8.0
• fairseq==0.10.2

Pre-trained model

you can download all pre-trained models here, and put it intomodeldirectory.

If you want train models from scratch, you need uses the pre-trained language models mt5-base (huggingface) and place the models under the model directory after downloading.

Train

train LSTM and Transformer model by fairseq, you need process data for jieba and bpe tokenize sentence, we use scripts from Subword-nmt:

git clone https://github.com/rsennrich/subword-nmt

Then run

sh prepare.sh

train LSTM, Transformer, mt5-seq2seq, mt5-fill, mt5-knowledge model

sh train_lstm.sh
sh train_transformer.sh
sh train_t5.sh
sh train_t5_fill.sh
sh train_t5_knowledge.sh

Evaluate

Run the following command to evaluate

sh evaluate_base.sh
sh evaluate_t5.sh
sh evaluate_t5_knowledge.sh
sh evaluate_t5_fill.sh

Citation

@article{qiang2022chinese,
    title={Chinese Idiom Paraphrasing},
    author={Jipeng Qiang, Yang Li, Chaowei Zhang, Yun Li and YunHao Yuan, Yi Zhu, Xindong Wu},
    journal={},
    year={2022},
}