Generate, Annotate, and Learn: NLP with Synthetic Text
Preprint
Abstract
This paper studies the use of language models as a source of synthetic unlabeled text for NLP. We formulate a general framework called ``generate, annotate, and learn (GAL)'' to take advantage of synthetic text within knowledge distillation, self-training, and few-shot learning applications. To generate high-quality task-specific text, we either fine-tune LMs on inputs from the task of interest, or prompt large LMs with few examples. We use the best available classifier to annotate synthetic text with soft pseudo labels for knowledge distillation and self-training, and use LMs to obtain hard labels for few-shot learning. We train new supervised models on the combination of labeled and pseudo-labeled data, which results in significant gains across several applications. We investigate key components of GAL and present theoretical and empirical arguments against the use of class-conditional LMs to generate synthetic labeled text instead of unlabeled text. GAL achieves new state-of-the-art knowledge distillation results for 6-layer transformers on the GLUE leaderboard.Keywords
All Related Versions
This publication has 16 references indexed in Scilit:
- MATE-KD: Masked Adversarial TExt, a Companion to Knowledge Distillation2021
- A Knowledge-Enhanced Pretraining Model for Commonsense Story GenerationTransactions of the Association for Computational Linguistics, 2020
- Data Augmentation using Pre-trained Transformer Models2020
- Facts2Story: Controlling Text Generation by Key Facts2020
- TinyBERT: Distilling BERT for Natural Language Understanding2020
- Transformers: State-of-the-Art Natural Language Processing2020
- Document-Level Neural Machine Translation with Hierarchical Attention Networks2018
- Contextual Augmentation: Data Augmentation by Words with Paradigmatic Relations2018
- Understanding the Yarowsky AlgorithmComputational Linguistics, 2004
- Learning with a probabilistic teacherIEEE Transactions on Information Theory, 1970