PAWS: Paraphrase Adversaries from Word Scrambling

***** New August 27th, 2019: Multilingual PAWS for six languages *****

We released PAWS-X, a multilingual version of PAWS for six languages. See here for more details.

***** End new information *****

This dataset contains 108,463 human-labeled and 656k noisily labeled pairs that feature the importance of modeling structure, context, and word order information for the problem of paraphrase identification. The dataset has two subsets, one based on Wikipedia and the other one based on the Quora Question Pairs (QQP) dataset.

For further details, see the accompanying paper: PAWS: Paraphrase Adversaries from Word Scrambling

Motivation

Existing paraphrase identification datasets lack sentence pairs that have high lexical overlap without being paraphrases. Models trained on such data fail to distinguish pairs like flights from New York to Florida and flights from Florida to New York.

Below are two examples from the dataset:

	Sentence 1	Sentence 2	Label
(1)	Although interchangeable, the body pieces on the 2 cars are not similar.	Although similar, the body parts are not interchangeable on the 2 cars.	0
(2)	Katz was born in Sweden in 1947 and moved to New York City at the age of 1.	Katz was born in 1947 in Sweden and moved to New York at the age of one.	1

The first pair has different semantic meaning while the second pair is a paraphrase. State-of-the-art models trained on existing datasets have dismal performance on PAWS (<40% accuracy); however, including PAWS training data for these models improves their accuracy to 85% while maintaining performance on existing datasets such as the Quora Question Pairs.

PAWS-Wiki

This corpus contains pairs generated from Wikipedia pages, and can be downloaded here:

• PAWS-Wiki Labeled (Final): containing pairs that are generated from both word swapping and back translation methods. All pairs have human judgements on both paraphrasing and fluency and they are split into Train/Dev/Test sections.
• PAWS-Wiki Labeled (Swap-only): containing pairs that have no back translation counterparts and therefore they are not included in the first set. Nevertheless, they are high-quality pairs with human judgements on both paraphrasing and fluency, and they can be included as an auxiliary training set.
• PAWS-Wiki Unlabeled (Final): Pairs in this set have noisy labels without human judgments and can also be used as an auxiliary training set. They are generated from both word swapping and back translation methods.

All files are in the tsv format with four columns:

Column Name	Data
id	A unique id for each pair
sentence1	The first sentence
sentence2	The second sentence
(noisy_)label	(Noisy) label for each pair

Each label has two possible values: 0 indicates the pair has different meaning, while 1 indicates the pair is a paraphrase.

The number of examples and the proportion of paraphrase (Yes%) pairs are shown below:

Data	Train	Dev	Test	Yes%
Labeled (Final)	49,401	8,000	8,000	44.2%
Labeled (Swap-only)	30,397	--	--	9.6%
Unlabeled (Final)	645,652	10,000	--	50.0%

We also release source sentences that are used to generate this dataset and their mappings. Please see here for more details.

PAWS-QQP

This corpus contains pairs generated from the Quora Question Pairs corpus. We cannot directly distribute the raw PAWS-QQP data due to the license of QQP, so the examples must be reconstructed by downloading the original data and then running our scripts to produce the data and attach the labels.

To reconstruct the PAWS-QQP corpus, first download the original Quora Question Pairs dataset and save the tsv file to some location /path/to/original_qqp/data.tsv. Then download the PAWS-QQP index file from the following link:

• PAWS-QQP Index (Final)

Unpack it to some directory /path/to/paws_qqp/. Run the following commands to generate the corpus.

export ORIGINAL_QQP_FILE="/path/to/original_qqp/data.tsv"
export PAWS_QQP_DIR="/path/to/paws_qqp/"
export PAWS_QQP_OUTPUT_DIR="/path/to/paws_qqp/output/"

python qqp_generate_data.py \
  --original_qqp_input="${ORIGINAL_QQP_FILE}" \
  --paws_input="${PAWS_QQP_DIR}/train.tsv" \
  --paws_output="${PAWS_QQP_OUTPUT_DIR}/train.tsv"

python qqp_generate_data.py \
  --original_qqp_input="${ORIGINAL_QQP_FILE}" \
  --paws_input="${PAWS_QQP_DIR}/dev_and_test.tsv" \
  --paws_output="${PAWS_QQP_OUTPUT_DIR}/dev_and_test.tsv"

Note: this script requires NLTK and was tested on version 3.2.5.

The generated tsv files have the same format as PAWS-Wiki. All pairs are manually labeled, and the number of examples and the proportion of paraphrase (Yes%) pairs are shown below:

Data	Train	Dev and Test	Yes%
PAWS-QQP	11,988	677	31.3%

For the experiments in our paper, we used the train/dev/test split of the original QQP from Wang et al, 2017.

PAWS-X

This corpus contains translations of the PAWS examples in six typologically distinct languages: French, Spanish, German, Chinese, Japanese, and Korean. Please see here for more details.

Note: for multilingual experiments, please use dev_2k.tsv provided in the PAWS-X repo as the development sets for all languages, including English.

Reference

If you use or discuss this dataset in your work, please cite the following papers correspondingly:

@InProceedings{paws2019naacl,
  title = {{PAWS: Paraphrase Adversaries from Word Scrambling}},
  author = {Zhang, Yuan and Baldridge, Jason and He, Luheng},
  booktitle = {Proc. of NAACL},
  year = {2019}
}

@InProceedings{pawsx2019emnlp,
  title = {{PAWS-X: A Cross-lingual Adversarial Dataset for Paraphrase Identification}},
  author = {Yang, Yinfei and Zhang, Yuan and Tar, Chris and Baldridge, Jason},
  booktitle = {Proc. of EMNLP},
  year = {2019}
}

Contact

If you have a technical question regarding the dataset or publication, please create an issue in this repository.