QSTR

Code for EMNLP2023 paper "A Quality-based Syntactic Template Retriever for Syntactically-controlled Paraphrase Generation".

Data

We use the QQPPos and ParaNMT datasets for our experiments and retain the same split with SGCP. You can download the data from Data.

Due to the limitation of file size, I will email you the processed data if you need it. Please contact with me (23111135@bjtu.edu.cn).

bash fast-train-roberta.sh

bash select.sh

After obtaining the retrieved templates, you can utilize any syntactically-controlled paraphrase generation models to generate paraphrases.

If you have any questions, please create an issue or email me.

Name		Name	Last commit message	Last commit date
Latest commit History 12 Commits
Evaluation-metrics		Evaluation-metrics
Results		Results
README.md		README.md
SimFT_model.py		SimFT_model.py
cal_parascore4train-tv.py		cal_parascore4train-tv.py
cal_parascore4train.py		cal_parascore4train.py
data_utils.py		data_utils.py
diverse-select-by-TED.py		diverse-select-by-TED.py
eval_by_pearson.py		eval_by_pearson.py
fast-train-roberta.py		fast-train-roberta.py
fast-train-roberta.sh		fast-train-roberta.sh
get_pgs_from_AESOP.py		get_pgs_from_AESOP.py
get_train_true_parascore.py		get_train_true_parascore.py
parascorer_new.py		parascorer_new.py
save_trees_embeds.py		save_trees_embeds.py
select-top10.py		select-top10.py
select.sh		select.sh
test.py		test.py
top_select.py		top_select.py