# New Trends in Machine Translation using Large Language Models: Case Examples with ChatGPT

**Authors:** Chenyang Lyu, Jitao Xu, Longyue Wang  
**arXiv:** [2305.01181](https://arxiv.org/abs/2305.01181)  
**Date:** 2023-05-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.01181v1/

## Key points

- Machine Translation (MT) has made significant progress with the emergence of large language models (LLMs) such as GPT-3 and ChatGPT.
- Lyu et al. explore interesting directions for MT using LLMs, including:
- Stylized MT: translating text into a specific style or tone that matches the intended audience or purpose.
- Interactive MT: allowing users to actively participate in the translation process by providing feedback or editing translations in real-time.
- Translation Memory-based MT: utilizing previously translated texts to improve translation quality and speed.
- Personalized MT: enabling customized translations tailored to each user's preferences and needs, which raises challenges such as collecting and storing user-specific data in a privacy-preserving manner and measuring its effectiveness accurately.
- The authors also discuss privacy concerns in using LLMs for machine translation and propose basic privacy-preserving methods to mitigate risks.
- The paper provides valuable insights into new trends in machine translation using large language models that can significantly enhance translation quality while addressing privacy concerns.

## Summary

Machine Translation (MT) has seen significant progress in recent years, particularly with the emergence of large language models (LLMs) such as GPT-3 and ChatGPT. This has opened up new opportunities and challenges for MT using LLMs. In their paper, Lyu et al. explore some interesting directions for MT using LLMs, including stylized MT, interactive MT, Translation Memory-based MT, personalized MT, and a new evaluation paradigm using LLMs. Stylized MT involves translating text into a specific style or tone that matches the intended audience or purpose. Interactive MT allows users to actively participate in the translation process by providing feedback or editing translations in real-time. Translation Memory-based MT utilizes previously translated texts to improve translation quality and speed. Personalized MT is an exciting direction for future research as it enables customized translations tailored to each user's preferences and needs. This approach can include translations adapted to the user's language proficiency, domain-specific terminology, or cultural references. One possible approach is to prompt LLMs with user-specific metadata such as search histories or social media posts. However, personalized MT raises several challenges such as collecting and storing user-specific data in a privacy-preserving manner and measuring its effectiveness accurately. The authors also discuss privacy concerns in using LLMs for machine translation and propose basic privacy-preserving methods to mitigate risks. To illustrate the potential of these proposed directions, the authors present several examples demonstrating their feasibility while highlighting opportunities and challenges for future research in this area of using LLMs for machine translation. Overall, this paper provides valuable insights into new trends in machine translation using large language models that can significantly enhance translation quality while addressing privacy concerns.