References & Citations
Computer Science > Computation and Language
Title: MSP: Multi-Stage Prompting for Making Pre-trained Language Models Better Translators
(Submitted on 13 Oct 2021 (v1), last revised 17 Mar 2022 (this version, v2))
Abstract: Prompting has recently been shown as a promising approach for applying pre-trained language models to perform downstream tasks. We present Multi-Stage Prompting (MSP), a simple and automatic approach for leveraging pre-trained language models to translation tasks. To better mitigate the discrepancy between pre-training and translation, MSP divides the translation process via pre-trained language models into multiple separate stages: the encoding stage, the re-encoding stage, and the decoding stage. During each stage, we independently apply different continuous prompts for allowing pre-trained language models better shift to translation tasks. We conduct extensive experiments on three translation tasks. Experiments show that our method can significantly improve the translation performance of pre-trained language models.
Submission history
From: Zhixing Tan [view email][v1] Wed, 13 Oct 2021 10:06:21 GMT (126kb,D)
[v2] Thu, 17 Mar 2022 13:55:29 GMT (251kb,D)
Link back to: arXiv, form interface, contact.