Generate, Annotate, and Learn: NLP with Synthetic Text

Xuanli He, Islam Nassar, Jamie Kiros, Gholamreza Haffari, Mohammad Norouzi


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.
Anthology ID:
2022.tacl-1.48
Volume:
Transactions of the Association for Computational Linguistics, Volume 10
Month:
Year:
2022
Address:
Cambridge, MA
Venue:
TACL
SIG:
Publisher:
MIT Press
Note:
Pages:
826–842
Language:
URL:
https://aclanthology.org/2022.tacl-1.48
DOI:
10.1162/tacl_a_00492
Bibkey:
Cite (ACL):
Xuanli He, Islam Nassar, Jamie Kiros, Gholamreza Haffari, and Mohammad Norouzi. 2022. Generate, Annotate, and Learn: NLP with Synthetic Text. Transactions of the Association for Computational Linguistics, 10:826–842.
Cite (Informal):
Generate, Annotate, and Learn: NLP with Synthetic Text (He et al., TACL 2022)
Copy Citation:
PDF:
https://aclanthology.org/2022.tacl-1.48.pdf