--- pretty_name: 'PAWS-X: A Cross-lingual Adversarial Dataset for Paraphrase Identification' annotations_creators: - expert-generated - machine-generated language_creators: - expert-generated - machine-generated language: - de - en - es - fr - ja - ko - zh license: - other multilinguality: - multilingual size_categories: - 10K **Caveat**: please note that the dev and test sets of PAWS-X are both sourced > from the dev set of PAWS-Wiki. As a consequence, the same `sentence 1` may > appear in both the dev and test sets. Nevertheless our data split guarantees > that there is no overlap on sentence pairs (`sentence 1` + `sentence 2`) > between dev and test. ## Dataset Creation ### Curation Rationale Most existing work on adversarial data generation focuses on English. For example, PAWS (Paraphrase Adversaries from Word Scrambling) (Zhang et al., 2019) consists of challenging English paraphrase identification pairs from Wikipedia and Quora. They remedy this gap with PAWS-X, a new dataset of 23,659 human translated PAWS evaluation pairs in six typologically distinct languages: French, Spanish, German, Chinese, Japanese, and Korean. They provide baseline numbers for three models with different capacity to capture non-local context and sentence structure, and using different multilingual training and evaluation regimes. Multilingual BERT (Devlin et al., 2019) fine-tuned on PAWS English plus machine-translated data performs the best, with a range of 83.1-90.8 accuracy across the non-English languages and an average accuracy gain of 23% over the next best model. PAWS-X shows the effectiveness of deep, multilingual pre-training while also leaving considerable headroom as a new challenge to drive multilingual research that better captures structure and contextual information. ### Source Data PAWS (Paraphrase Adversaries from Word Scrambling) #### Initial Data Collection and Normalization All translated pairs are sourced from examples in [PAWS-Wiki](https://github.com/google-research-datasets/paws#paws-wiki) #### Who are the source language producers? This dataset contains 23,659 human translated PAWS evaluation pairs and 296,406 machine translated training pairs in six typologically distinct languages: French, Spanish, German, Chinese, Japanese, and Korean. ### Annotations #### Annotation process If applicable, describe the annotation process and any tools used, or state otherwise. Describe the amount of data annotated, if not all. Describe or reference annotation guidelines provided to the annotators. If available, provide interannotator statistics. Describe any annotation validation processes. #### Who are the annotators? The paper mentions the translate team, especially Mengmeng Niu, for the help with the annotations. ### Personal and Sensitive Information [More Information Needed] ## Considerations for Using the Data ### Social Impact of Dataset [More Information Needed] ### Discussion of Biases [More Information Needed] ### Other Known Limitations [More Information Needed] ## Additional Information ### Dataset Curators List the people involved in collecting the dataset and their affiliation(s). If funding information is known, include it here. ### Licensing Information The dataset may be freely used for any purpose, although acknowledgement of Google LLC ("Google") as the data source would be appreciated. The dataset is provided "AS IS" without any warranty, express or implied. Google disclaims all liability for any damages, direct or indirect, resulting from the use of the dataset. ### Citation Information ``` @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} } ``` ### Contributions Thanks to [@bhavitvyamalik](https://github.com/bhavitvyamalik), [@gowtham1997](https://github.com/gowtham1997) for adding this dataset.