How Prompting by Humans Improves Machine Translation

Prompts for Machine Translation

Since ChatGPT launched at the end of last year, the language services industry and academia have been in a race to investigate how prompting a large language model (LLM) can change the output of a machine translation.

While the progress of machine translation output with LLMs has been a cause for concern among some linguists, industry associations and heavyweights have been quick to highlight the continued role of the expert linguist in the loop.

As Ian El-Mokadem, CEO of RWS, told SlatorCon in October 2023, the role of the linguist will transform: “What’s our message to translators right now? Well, it’s goodbye, translators. And hello, language specialists.”

That’s because prompting machine translations has yielded promising results, and requires these ‘language specialists’ to prompt the model, and analyze the output.

Examples of prompts include defined roles or objectives, such as “act as a professional English-to-Greek translator specialized in healthcare”, or other inferred prompts to refer to information already known to the model.

Research on prompting has led to publications of the most useful prompts for translation and multilingual copy generation, and has sparked discussions on how it can help translation project managers.

The applications of prompting are vast. Possible use cases include prompting for tone of voice, gender-neutral language, creativity, or customer-specific terminology. These opportunities also open new doors for using LLMs to author content with expert-in-the-loop workflows.

Slator’s recently released Pro Guide: Translation AI provides a concise snapshot of the latest practical applications of large language models (LLMs) in translation, and includes a use case on performing machine translation with prompting.

The use case is one of ten, one-page examples of LLMs being put to use, and is drawn from research and interviews with some of the industry’s leading language technology providers.