Awesome
Data Augmentation Techniques for NLP
If you'd like to add your paper, do not email us. Instead, read the protocol for adding a new entry and send a pull request.
We group the papers by text classification, translation, summarization, question-answering, sequence tagging, parsing, grammatical-error-correction, generation, dialogue, multimodal, mitigating bias, mitigating class imbalance, adversarial examples, compositionality, and automated augmentation.
This repository is based on our paper, "A survey of data augmentation approaches in NLP (Findings of ACL '21)". You can cite it as follows:
@inproceedings{feng-etal-2021-survey,
title = "A Survey of Data Augmentation Approaches for {NLP}",
author = "Feng, Steven Y. and
Gangal, Varun and
Wei, Jason and
Chandar, Sarath and
Vosoughi, Soroush and
Mitamura, Teruko and
Hovy, Eduard",
booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
month = aug,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.findings-acl.84",
doi = "10.18653/v1/2021.findings-acl.84",
pages = "968--988",
}
Authors: <a href="https://scholar.google.ca/citations?hl=en&user=zwiszZIAAAAJ">Steven Y. Feng</a>,
<a href="https://scholar.google.com/citations?user=rWZq2nQAAAAJ&hl=en">Varun Gangal</a>,
<a href="https://scholar.google.com/citations?user=wA5TK_0AAAAJ&hl=en">Jason Wei</a>,
<a href="https://scholar.google.co.in/citations?user=yxWtZLAAAAAJ&hl=en">Sarath Chandar</a>,
<a href="https://scholar.google.ca/citations?user=45DAXkwAAAAJ&hl=en">Soroush Vosoughi</a>,
<a href="https://scholar.google.com/citations?user=gjsxBCkAAAAJ&hl=en">Teruko Mitamura</a>,
<a href="https://scholar.google.com/citations?user=PUFxrroAAAAJ&hl=en">Eduard Hovy</a>
Special thanks to Ryan Shentu, Fiona Feng, Karen Liu, Emily Nie, Tanya Lu, and Bonnie Ma for helping out with this repo.
Note: WIP. More papers will be added from our survey paper to this repo soon.
Inquiries should be directed to stevenyfeng@gmail.com or by opening an issue here.
Also, check out our talk for Google Research (Steven Feng and Varun Gangal) here, and our podcast episode (Steven Feng and Eduard Hovy) here and here.
Text Classification
| Paper | Datasets |
|---|
| Unsupervised Word Sense Disambiguation Rivaling Supervised Methods (ACL '95) | Paper-Specific/Legacy Corpus |
| Synonym Replacement (Character-Level Convolutional Networks for Text Classification, NeurIPS '15) | AG’s News, DBPedia, Yelp, Yahoo Answers, Amazon |
| That’s So Annoying!!!: A Lexical and Frame-Semantic Embedding Based Data Augmentation Approach to Automatic Categorization of Annoying Behaviors using #petpeeve Tweets (EMNLP '15) | twitter |
| Robust Training under Linguistic Adversity (EACL '17) code | Movie review, customer review, SUBJ, SST |
| Contextual Augmentation: Data Augmentation by Words with Paradigmatic Relations (NAACL '18) code | SST, SUBJ, MRQA, RT, TREC |
| Variational Pretraining for Semi-supervised Text Classification (ACL '19) code | IMDB, AG News, Yahoo, hatespeech |
| EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks (EMNLP '19) code | SST, CR, SUBJ, TREC, PC |
| A Closer Look At Feature Space Data Augmentation For Few-Shot Intent Classification (DeepLo @ EMNLP '19) | SNIPS |
| Nonlinear Mixup: Out-Of-Manifold Data Augmentation for Text Classification (AAAI '20) | TREC, SST, Subj, MR |
| MixText: Linguistically-Informed Interpolation of Hidden Space for Semi-Supervised Text Classification (ACL '20) code | AG News, DBpedia, Yahoo, IMDb |
| Unsupervised Data Augmentation for Consistency Training (NeurIPS '20) code | Yelp, IMDb, amazon, DBpedia |
| Not Enough Data? Deep Learning to the Rescue! (AAAI '20) | ATIS, TREC, WVA |
| Data Augmentation using Pre-trained Transformer Models LifeLongNLP @ AACL '20, code | SNIPS, TREC, SST2 |
| SSMBA: Self-Supervised Manifold Based Data Augmentation for Improving Out-of-Domain Robustness (EMNLP '20) code | IWSLT'14 |
| Data Boost: Text Data Augmentation Through Reinforcement Learning Guided Conditional Generation (EMNLP '20) | ICWSM 20’ Data Challenge, SemEval '17 sentiment analysis, SemEval '18 irony |
| Textual Data Augmentation for Efficient Active Learning on Tiny Datasets (EMNLP '20) | SST2, TREC |
| Text Augmentation in a Multi-Task View (EACL '21) | SST2, TREC, SUBJ |
| GPT3Mix: Leveraging Large-scale Language Models for Text Augmentation (arXiv '21) | SST2, CR, TREC, SUBJ, MPQA, CoLA |
| Few-Shot Text Classification with Triplet Loss, Data Augmentation, and Curriculum Learning (NAACL '21) code | HUFF, COV-Q, AMZN, FEWREL |
| Text AutoAugment: Learning Compositional Augmentation Policy for Text Classification (EMNLP '21) code | IMDB, SST2, SST5, TREC, YELP2, YELP5 |
| AEDA: An Easier Data Augmentation Technique for Text Classification (EMNLP '21) code | SST, CR, SUBJ, TREC, PC |
Translation
| Paper | Datasets |
|---|
| Backtranslation (Improving Neural Machine Translation Models with Monolingual Data, ACL '16) | WMT '15 en-de, IWSLT '15 en-tr |
| Adapting Neural Machine Translation with Parallel Synthetic Data (WMT '17) | COMMON, 1 Billion Words, dev2013, XRCE, IT, E-Com |
| Data Augmentation for Low-Resource Neural Machine Translation (ACL '17) code | WMT '14/'15/'16 en-de/de-en |
| Synthetic Data for Neural Machine Translation of Spoken-Dialects (arxiv '17) | LDC2012T09, OpenSubtitles-2013 |
| Multi-Source Neural Machine Translation with Data Augmentation (IWSLT '18) | TED Talks |
| SwitchOut: an Efficient Data Augmentation Algorithm for Neural Machine Translation (EMNLP '18) | IWSLT '15 en-vi, IWSLT '16 de-en, WMT '15 en-de |
| Generalizing Back-Translation in Neural Machine Translation (WMT '19) | ed NewsCrawl2, WMT'18 de-en |
| Neural Fuzzy Repair: Integrating Fuzzy Matches into Neural Machine Translation (ACL '19) | DGT-TM en-ml/en-hu |
| Augmenting Neural Machine Translation with Knowledge Graphs (arxiv '19) | WMT '14 -'18 |
| Generalized Data Augmentation for Low-Resource Translation (ACL '19) code | ENG-HRL-LRL, HRL-LRL |
| Improving Robustness of Machine Translation with Synthetic Noise (NAACL '19) code | EP, TED, MTNT en-fr en-jpn |
| Soft Contextual Data Augmentation for Neural Machine Translation (ACL '19) code | IWSLT '14 de/es/he-en, WMT '14 en-de |
| Data augmentation using back-translation for context-aware neural machine translation (DiscoMT @ EMNLP '19) code | IWSLT'17 en-ja/en-fr, BookCorpus, Europarl v7, National Diet of Japan |
| Improving Neural Machine Translation Robustness via Data Augmentation: Beyond Back-Translation (W-NUT @ EMNLP '19) | WMT'15/'19 en/fr, MTNT, IWSLT'17, MuST-C |
| Data augmentation for pipeline-based speech translation (Baltic HLT '20) | WMT '17 |
| Lexical-Constraint-Aware Neural Machine Translation via Data Augmentation (IJCAI '20) code | WMT '16 de-en, NIST zh-en |
| A Diverse Data Augmentation Strategy for Low-Resource Neural Machine Translation (Information '20) | IWSLT '14 en-de |
| Syntax-aware Data Augmentation for Neural Machine Translation (arxiv '20) | WMT '14 en-de, IWSLT '14 de-en |
| SSMBA: Self-Supervised Manifold Based Data Augmentation for Improving Out-of-Domain Robustness (EMNLP '20) code | IWSLT'14 |
| Data diversification: A simple strategy for neural machine translation (NeurIPS '20) code | WMT '14 en-de/en-fr, IWSLT '13/'14/'15 en-de/de-en/en-fr |
| AdvAug: Robust Adversarial Augmentation for Neural Machine Translation (ACL '20) | NIST zh-en, WMT '14 en-de |
| Dictionary-based Data Augmentation for Cross-Domain Neural Machine Translation (arxiv '20) | WMT '14/'19 |
| Sentence Boundary Augmentation For Neural Machine Translation Robustness (arxiv '20) | IWSLT '14/'15/'18 en-de, WMT '18 en-de |
| Valar nmt : Vastly lacking resources neural machine translation (Stanford CS224N) | Bible, Misc, Europarl v8, Newstest '18 |
Summarization
| Paper | Datasets |
|---|
| Transforming Wikipedia into Augmented Data for Query-Focused Summarization (arxiv '19) | DUC |
| Iterative Data Augmentation with Synthetic Data (Abstract Text Summarization: A Low Resource Challenge (EMNLP '19) | Swisstext, commoncrawl |
| Improving Zero and Few-Shot Abstractive Summarization with Intermediate Fine-tuning and Data Augmentation (NAACL '21) | CNN-DailyMail |
| Data Augmentation for Abstractive Query-Focused Multi-Document Summarization (AAAI '21) code | QMDSCNN, QMDSIR, WikiSum, DUC 2006, DUC 2007 |
Question Answering
| Paper | Datasets |
|---|
| QANet: Combining Local Convolution with Global Self-Attention for Reading Comprehension (ICLR '18) | SQuAD, TriviaQA |
| An Exploration of Data Augmentation and Sampling Techniques for Domain-Agnostic Question Answering (EMNLP '19 Workshop) | MRQA |
| Data Augmentation for BERT Fine-Tuning in Open-Domain Question Answering (arxiv '19) | SQuAD, Trivia-QA, CMRC, DRCD |
| XLDA: Cross-Lingual Data Augmentation for Natural Language Inference and Question Answering (arxiv '19) | XNLI, SQuAD |
| Synthetic Data Augmentation for Zero-Shot Cross-Lingual Question Answering (arxiv '20) | MLQA, XQuAD, SQuAD-it, PIAF |
| Logic-Guided Data Augmentation and Regularization for Consistent Question Answering (ACL '20) code | WIQA, QuaRel, HotpotQA |
Sequence Tagging
| Paper | Datasets |
|---|
| Data Augmentation via Dependency Tree Morphing for Low-Resource Languages (EMNLP '18) code | universal dependencies project |
| DAGA: Data Augmentation with a Generation Approach for Low-resource Tagging Tasks (EMNLP '20) code | CoNLL2002/2003 |
| An Analysis of Simple Data Augmentation for Named Entity Recognition (COLING '20) | MaSciP, i2b2- 2010 |
| SeqMix: Augmenting Active Sequence Labeling via Sequence Mixup (EMNLP '20) code | CoNLL-03, ACE05, Webpage |
Parsing
| Paper | Datasets |
|---|
| Data Recombination for Neural Semantic Parsing (ACL '16) code | GeoQuery, ATIS, Overnight |
| A systematic comparison of methods for low-resource dependency parsing on genuinely low-resource languages (EMNLP '19) | Universal Dependencies treebanks version 2.2 |
| Named Entity Recognition for Social Media Texts with Semantic Augmentation (EMNLP '20)code | WNUT16, WNUT17, Weibo |
| Good-Enough Compositional Data Augmentation (ACL '20) code | SCAN |
| GraPPa: Grammar-Augmented Pre-Training for Table Semantic Parsing (ICLR '21) | SPIDER, WIKISQL, WIKITABLEQUESTIONS |
Grammatical Error Correction
| Paper | Datasets |
|---|
| GenERRate: Generating Errors for Use in Grammatical Error Detection (BEA '09) | Ungram-BNC |
| Mining Revision Log of Language Learning SNS for Automated Japanese Error Correction of Second Language Learners (IJCNLP '11) code | Lang-8 |
| Artificial error generation for translation-based grammatical error correction (University of Cambridge Technical Report '16) | Several Datasets |
| Noising and Denoising Natural Language: Diverse Backtranslation for Grammar Correction. (NAACL'18) | Lang-8, CoNLL-2014, CoNLL-2013, JFLEG |
| Using Wikipedia Edits in Low Resource Grammatical Error Correction. (WNUT @ EMNLP '18) | Falko-MERLIN GEC Corpus |
| Sequence-to-sequence Pre-training with Data Augmentation for Sentence Rewriting (arxiv '19) | CoNLL-2014 , JFLEG |
| Controllable Data Synthesis Method for Grammatical Error Correction (arxiv '19) code | NUCLE, Lang-8, One-Billion, CoNLL2013, CoNLL2014 |
| Neural Grammatical Error Correction Systems with Unsupervised Pre-training on Synthetic Data. (BEA @ ACL '19) | FCE, NUCLE, W&I+LOCNESS, Lang-8 |
| Corpora Generation for Grammatical Error Correction (NAACL'19) | CoNLL-2014, JFLEG, Lang-8 |
| Erroneous data generation for Grammatical Error Correction (BEA @ ACL '19) | Lang-8,n CoNLL, JFLEG, CoNLL-2014, ABCN, FCE |
| Sequence-to-sequence Pre-training with Data Augmentation for Sentence Rewriting (arxiv '19) code | GYAFC, WMT14, WMT18 |
| A neural grammatical error correction system built on better pre-training and sequential transfer learning. (BEA @ ACL '19) | FCE, NUCLE, W&I+LOCNESS, Lang-8, Gutenberg, Tatoeba, WikiText-103 |
| Improving Grammatical Error Correction with Data Augmentation by Editing Latent Representation (COLING'20) | FCE, NUCLE, W&I+LOCNESS, Lang-8 |
| A Comparative Study of Synthetic Data Generation Methods for Grammatical Error Correction (BEA @ ACL '20) | W&I+LOCNESS, FCE, News Crawl 2, W&I+L train, FCE-train, NUCLE, Lang-8, W&I+L dev, FCE-test, Tatoeba, WikiText-103 |
| A syntactic rule-based framework for parallel data synthesis in Japanese GEC (MIT Thesis '20) | Lang-8 |
Generation
| Paper | Datasets |
|---|
| TNT-NLG, System 2: Data repetition and meaning representation manipulation to improve neural generation (E2E NLG Challenge System Descriptions) | TODO |
| Findings of the Third Workshop on Neural Generation and Translation (WNGT @ EMNLP '19) | RotoWire English-German |
| A Good Sample is Hard to Find: Noise Injection Sampling and Self-Training for Neural Language Generation Models (INLG '19) code | E2E Challenge Dataset, Laptops, TVs |
| GenAug: Data Augmentation for Finetuning Text Generators (DeeLIO @ EMNLP '20) code | Yelp |
| Denoising Pre-Training and Data Augmentation Strategies for Enhanced RDF Verbalization with Transformers (WebNLG+ @ INLG '20) | WebNLG |
Dialogue
| Paper | Datasets |
|---|
| Sequence-to-Sequence Data Augmentation for Dialogue Language Understanding (COLING '18) code | ATIS, Dec94, Stanford dialogue |
| Task-Oriented Dialog Systems that Consider Multiple Appropriate Responses under the Same Context (arxiv '19) code | MultiWOZ |
| Data Augmentation by Data Noising for Open-vocabulary Slots in Spoken Language Understanding (Student Research Workshop @ NAACL '19) | ATIS, Snips, MR |
| Data Augmentation with Atomic Templates for Spoken Language Understanding (EMNLP '19) code | DSTC 2&3, DSTC2 |
| Data Augmentation for Spoken Language Understanding via Joint Variational Generation (AAAI '19) | ATIS, Snips, MIT |
| Effective Data Augmentation Approaches to End-to-End Task-Oriented Dialogue (IALP '19) | CamRest676, KVRET |
| Paraphrase Augmented Task-Oriented Dialog Generation (ACL '20) code | TCamRest676, MultiWOZ |
| Dialog State Tracking with Reinforced Data Augmentation (AAAI '20) | WoZ, MultiWoZ |
| Data Augmentation for Copy-Mechanism in Dialogue State Tracking (arxiv '20) | WoZ, DSTC2, Multi |
| Simple is Better! Lightweight Data Augmentation for Low Resource Slot Filling and Intent Classification (PACLIC '20) code | ATIS, SNIPS, FB |
| Conversation Graph: Data Augmentation, Training, and Evaluation for Non-Deterministic Dialogue Management (TACL '21) | M2M, MultiWOZ |
| GOLD: Improving Out-of-Scope Detection in Dialogues using Data Augmentation (EMNLP '21) code | SMCalFlow, ROSTD |
| Improving Automated Evaluation of Open Domain Dialog via Diverse Reference Augmentation (ACL '21 Findings) code | DailyDialog |
Multimodal
| Paper | Datasets |
|---|
| Data Augmentation for Visual Question Answering (INLG '17) | COCO-VQA, COCO-QA |
| Low Resource Multi-modal Data Augmentation for End-to-end ASR (CoRR ’18) | TODO |
| Multi-Modal Data Augmentation for End-to-end ASR (Interspeech '18) | Voxforge, HUB4 |
| Augmenting Image Question Answering Dataset by Exploiting Image Captions (LREC '18) | IQA |
| Multimodal Continuous Emotion Recognition with Data Augmentation Using Recurrent Neural Networks (AVEC '18) | TODO |
| Multimodal Dialogue State Tracking By QA Approach with Data Augmentation (DSTC8 @ AAAI '20) | DSTC7-AVSD |
| Data augmentation techniques for the Video Question Answering task (arxiv '20) | TGIF-QA, MSVD-QA |
| Data Augmentation for Training Dialog Models Robust to Speech Recognition Errors (NLP for ConvAI @ ACL '20) | DSTC2 |
| Semantic Equivalent Adversarial Data Augmentation for Visual Question Answering (ECCV '20) | TODO |
| Text Augmentation Using BERT for Image Captioning (Applied Sciences '20) | MSCOCO |
| MDA: Multimodal Data Augmentation Framework for Boosting Performance on Image-Text Sentiment/Emotion Classification Tasks (IEEE Intelligent Systems '20) | TODO |
Mitigating Bias
| Paper | Datasets |
|---|
| Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods. (NAACL '18) code | WinoBias, OntoNotes |
| Counterfactual Data Augmentation for Mitigating Gender Stereotypes in Languages with Rich Morphology (ACL '19) code | TODO |
| CONAN - COunter NArratives through Nichesourcing: a Multilingual Dataset of Responses to Fight Online Hate Speech (ACL '19) Dataset | New Dataset Created |
| It’s All in the Name: Mitigating Gender Bias with Name-Based Counterfactual Data Substitution (EMNLP '19) code | SSA, Stanford Large Movie Review, SimLex-999 |
| Gender Bias in Neural Natural Language Processing. (Springer '20) | Wikitext-2, CoNLL-2012 |
| Improving Robustness by Augmenting Training Sentences with Predicate-Argument Structures (arxiv '20) | SWAG, CoNLL2009, MultiNLI, HANS |
Mitigating Class Imbalance
| Paper | Datasets |
|---|
| SMOTE: Synthetic Minority Over-sampling Technique (Journal of Artificial Intelligence Research '02) | Pima, Phoneme, Adult, E-state, Satimage, Forest Cover, Oil, Mammography, Can |
| Active Learning for Word Sense Disambiguation with Methods for Addressing the Class Imbalance Problem (EMNLP '07) | TODO |
| MLSMOTE: Approaching imbalanced multilabel learning through synthetic instance generation (Knowledge-Based Systems '15) | bibtex, cal500, corel5k, slashdot, tmc2007, mediamill, medical, scene, enron, emotions |
| SMOTE for Learning from Imbalanced Data: Progress and Challenges, Marking the 15-year Anniversary (Journal of Artificial Intelligence Research '18) | TODO |
Adversarial examples
| Paper | Datsets |
|---|
| Adversarial Example Generation with Syntactically Controlled Paraphrase Networks (NAACL '18) code | SST, SICK |
| AdvEntuRe: Adversarial Training for Textual Entailment with Knowledge-Guided Examples (ACL '18) code | WordNet, PPDB, SICK, SNLI, SciTail |
| Breaking NLI Systems with Sentences that Require Simple Lexical Inferences (ACL '18) | SNLI, SciTail, MultiNLI |
| Certified Robustness to Adversarial Word Substitutions (EMNLP '19) code | IMDB, SNLI |
| PAWS: Paraphrase Adversaries from Word Scrambling (NAACL '19) code | PAWS (QQP + Wikipedia) |
| Generating Natural Language Adversarial Examples through Probability Weighted Word Saliency (ACL '19) code | IMDB, AG’s News, Yahoo Answers |
Compositionality
| Paper | Datsets |
|---|
| Good-Enough Compositional Data Augmentation (ACL '20) code | SCAN |
| Sequence-Level Mixed Sample Data Augmentation (EMNLP '20) code | IWSLT ’14, WMT ’14 |
Automated Augmentation
| Paper | Datsets |
|---|
| Learning Data Manipulation for Augmentation and Weighting (NeurIPS '19) code | SST, IMDB, TREC, CIFAR-10 |
| Data Manipulation: Towards Effective Instance Learning for Neural Dialogue Generation via Learning to Augment and Reweight (ACL '20) | DailyDialog, OpenSubtitles |
| Text AutoAugment: Learning Compositional Augmentation Policy for Text Classification (EMNLP '21) code | IMDB, SST2, SST5, TREC, YELP2, YELP5 |
Popular Resources