By Uwe Muegge, translation technology expert
 

Just a few days ago, OpenAI launched GPT-5.4, its most capable language model yet. With GPT-5.4, translation professionals using ChatGPT have access to stronger functions for translation-oriented terminology and style management, and automated post-editing. Below are my top tips for translation professionals who want to get better results from the latest release of ChatGPT.

 

Validate the content ChatGPT creates

---

OpenAI has announced that the new language model is significantly less likely to fabricate information than earlier models. But ChatGPT can still make serious translation errors as a result of hallucination, mistranslation, or omission.

For this reason, it is very important that you validate all content ChatGPT creates. Some translation errors can be difficult to find, so it's best to validate translated text against the source text.

When you validate all content created by ChatGPT, you combine the speed and efficiency of ChatGPT with your expert knowledge.

 

Use the Plus or Pro version

---

Note: The author is not affiliated with OpenAI, the developer of ChatGPT, and has not received or requested any compensation or other consideration from OpenAI.

One of the most powerful features GPT-5 language models offer is Thinking Mode. In Thinking Mode, ChatGPT can autonomously research terminology, create translation style guides based on sample texts or websites, and use these resources for translation or quality assurance.

However, GPT-5.4 and the Thinking Mode aren't available in the free version of ChatGPT. Users of the Go version have very limited access to GPT-5.4, but Go users can't select Thinking Mode.

For translation professionals who want to take full advantage of the advanced linguistic functions of GPT-5.4 in ChatGPT, a subscription to the Plus or Pro version is required.

 

Make sure that either 'Model improvement' is disabled or 'Temporary chat' is enabled

---

You have two options to stop ChatGPT from possibly using your data for model training. The first option is to disable the function 'Model improvement'.

Disabling 'Model improvement' prevents your chats from being used for model training but lets ChatGPT use earlier chats for reference.

The second option is to enable the function 'Temporary chat'. When 'Temporary chat' is on, ChatGPT won't use your data for model training or use earlier chats for reference.

 

Create and use project-specific dictionaries

---

ChatGPT can automate many parts of the process of creating reliable, project-specific bilingual dictionaries. ChatGPT can use a variety of methods to create bilingual dictionaries.

For example, ChatGPT can create dictionaries through term extraction from a TMX exchange file and from related source and target texts.

Using the Web search function, ChatGPT can automatically pair source terms with translated terms from websites.

With these capabilities, ChatGPT can make the creation of project-specific dictionaries faster and easier than traditional manual research.

 

Example of a bilingual dictionary with sources, created automatically with ChatGPT Plus

 

Create and use project-specific translation style guides

---

Using project-specific translation style guides can greatly improve the quality and efficiency of human translation, machine translation, and post-editing projects.

In human translation, translation style guides can provide guidance when there are no relevant matches in translation memories.

In post-editing, you can use a translation style guide in ChatGPT to apply grammatically correct changes automatically throughout a text.

In machine translation, you can use a translation style guide to customize machine translation in ChatGPT.

ChatGPT can automatically create project-specific translation style guides from TMX exchange files, translated texts, or source texts alone.

 

Example of a translation style guide with example sentences, created automatically with ChatGPT Plus

 

Use ChatGPT in combination with a translation memory program

---

Validating translated text against the source text is much easier in a translation memory program than in ChatGPT.

By using a translation memory program to edit text that ChatGPT has machine-translated, you can ensure consistency with earlier projects.

Ensuring terminological consistency is also much easier in a translation memory program than in ChatGPT.

The error-checking functions of many translation memory programs add reliability and consistency to the results ChatGPT provides.

 

Use prompts, not advanced features

---

Many translation professionals assume that taking advantage of the powerful linguistic capabilities ChatGPT offers requires understanding advanced ChatGPT functions. In practice, many multistep translation tasks can be handled simply with a well-designed prompt.

So if you're a new or casual user of ChatGPT, there is no need to learn functions such as Projects, Memories, Agents, or the API. A powerful prompt can be enough to let ChatGPT create bilingual dictionaries, translation style guides, apply dictionaries and style guides during translation, align text pairs to create TMX files, make grammatically correct global changes for post-editing, and create quality assurance reports.

This is good news for translators who want practical results without a long learning curve.

 

Conclusion

---

When used with the proper settings, processes, and expectations, ChatGPT can help translation professional work faster and more consistently. The most important step is not learning every feature. The most important step is building the right workflow for your project and validating the outputs ChatGPT creates. Translation professionals who want to improve their productivity and get step-by-step guidance for each task can find more detail in my new book 'ChatGPT Plus/Pro for Human Translation, Machine Translation, and Post-Editing'.

 

Uwe Muegge has more than ten years of experience as a freelance translator, with a focus on translation technology. He has also held leadership positions in translation agencies and large global organizations that buy translation services. Most recently, he was Head of Terminology at Facebook. In addition, Muegge was Chair of the ASTM Subcommittee on Translation and a member of ISO TC 37 on Language and Terminology. He has taught graduate-level translation technology courses in the United States and Europe.