Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Data Augmentation for Text Generation Without Any Augmented Data

About

Data augmentation is an effective way to improve the performance of many neural text generation models. However, current data augmentation methods need to define or choose proper data mapping functions that map the original samples into the augmented samples. In this work, we derive an objective to formulate the problem of data augmentation on text generation tasks without any use of augmented data constructed by specific mapping functions. Our proposed objective can be efficiently optimized and applied to popular loss functions on text generation tasks with a convergence rate guarantee. Experiments on five datasets of two text generation tasks show that our approach can approximate or even surpass popular data augmentation methods.

Wei Bi, Huayang Li, Jiacheng Huang• 2021

Related benchmarks

TaskDatasetResultRank
Machine TranslationIWSLT English-Vietnamese 2015 (tst2013)
BLEU27.12
23
Conversational Response GenerationReddit (test)
Dist-10.894
9
Neural Machine TranslationIWSLT German-to-English (De-En) 14 (tst2012)
BLEU29.16
8
Neural Machine TranslationIWSLT14 English-to-German (En-De) (tst2012)
BLEU23.42
8
Neural Machine TranslationIWSLT Vietnamese-to-English (Vi-En) 15 (tst2013)
BLEU24.74
8
Neural Machine TranslationIWSLT14 English-to-French (En-Fr) (tst2012)
BLEU41.07
8
Neural Machine TranslationIWSLT14 French-to-English (Fr-En) (tst2012)
BLEU40.46
8
Neural Machine TranslationIWSLT14 Italian-to-English (It-En) (tst2012)
BLEU29.86
8
Neural Machine TranslationIWSLT14 English-to-Italian (En-It) (tst2012)
BLEU27.15
8
Showing 9 of 9 rows

Other info

Follow for update