# Leveraging GPT-4 for Automatic Translation Post-Editing

**Authors:** Vikas Raunak, Amr Sharaf, Hany Hassan Awadallah, Arul Menezes  
**arXiv:** [2305.14878](https://arxiv.org/abs/2305.14878)  
**Date:** 2023-05-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.14878v1/

## Key points

- Neural Machine Translation (NMT) is the leading approach to machine translation.
- Even with NMT models, post-editing is still required to rectify errors and enhance quality, especially in critical settings.
- Researchers have explored various approaches to Automatic Post-Editing (APE), including context-aware models and the use of artificial training data.
- The authors formalize the task of APE with Large Language Models (LLMs) and investigate the use of GPT-4 for automatic post-editing across several language pairs.
- Their results demonstrate that GPT-4 is adept at producing meaningful edits even when the target language is not English, achieving state-of-the-art performance on WMT-22 English-Chinese, English-German, Chinese-English and German-English language pairs using GPT-4 based post-editing.
- This work represents the first investigation into using GPT-4 for automatic post-editing of translations and is related to other works exploring LLMs for translation.
- These findings suggest that GPT-4 can be a valuable tool in improving machine translation quality through automatic post-editing.

## Summary

In recent years, Neural Machine Translation (NMT) has emerged as the leading approach to machine translation. However, even with NMT models, post-editing is still required to rectify errors and enhance quality, especially in critical settings. To address this issue, researchers have explored various approaches to Automatic Post-Editing (APE), including context-aware models and the use of artificial training data. In this study, the authors formalize the task of APE with Large Language Models (LLMs) and investigate the use of GPT-4 for automatic post-editing across several language pairs. Their results demonstrate that GPT-4 is adept at producing meaningful edits even when the target language is not English. Notably, they achieve state-of-the-art performance on WMT-22 English-Chinese, English-German, Chinese-English and German-English language pairs using GPT-4 based post-editing. While previous studies have contributed significantly to the development of neural models for APE by exploring different architectures and learning strategies, this work represents the first investigation into using GPT-4 for automatic post-editing of translations. Additionally, it is related to other works exploring LLMs for translation. Overall, these findings suggest that GPT-4 can be a valuable tool in improving machine translation quality through automatic post-editing. With further advancements in LLM technology expected in the immediate future, it holds great promise for enhancing machine translation capabilities across multiple languages.