Dissecting In-Context Learning of Translations in GPTs

TL;DR

This paper investigates how different demonstration attributes affect GPT's in-context translation learning, revealing the importance of output distribution and proposing a method to enhance zero-shot translation performance.

Contribution

It uncovers the impact of source and target perturbations on translation quality and introduces Zero-Shot-Context to improve zero-shot GPT translation.

Findings

Target perturbation drastically reduces translation quality

Source perturbation has little impact on results

Zero-Shot-Context improves zero-shot translation, rivaling few-shot methods

Abstract

Most of the recent work in leveraging Large Language Models (LLMs) such as GPT-3 for Machine Translation (MT) has focused on selecting the few-shot samples for prompting. In this work, we try to better understand the role of demonstration attributes for the in-context learning of translations through perturbations of high-quality, in-domain demonstrations. We find that asymmetric perturbation of the source-target mappings yield vastly different results. We show that the perturbation of the source side has surprisingly little impact, while target perturbation can drastically reduce translation quality, suggesting that it is the output text distribution that provides the most important learning signal during in-context learning of translations. We propose a method named Zero-Shot-Context to add this signal automatically in Zero-Shot prompting. We demonstrate that it improves upon the zero-shot translation performance of GPT-3, even making it competitive with few-shot prompted translations.