Reducing Mistranslation Risks

50% of what I read every day has been machine-generated, and it's not from ChatGPT.

I work in a multilingual environment where people with low proficiency in each other's native languages have to write a lot. Therefore, they often rely on machine translation to write and read.

The issue of clunky writing is something I have dealt with for years, where machine translation creates nonsense, especially when dealing with the Japanese language, whether as a language to translate from or to translate to (usually English).

The problem is amplified as important communications are fragmented across multiple tools: chat, emails, online tools, intranets.

Copy-pasting content to ChatGPT, DeepL, or specific software creates additional friction. It is very tempting to write quickly, copy-paste, get a translation, send it, and then hope there are no disasters.

This makes the initial and original message even more important. Machine translation will amplify issues such as ambiguities, unclear, or uncohesive content.

Working with ChatGPT or GenAI in general implies control; you need to see what comes from you and what comes from AI. This is beneficial when you want a gap analysis whenever you request some content to be modified.

Another important point is when you have the right interface (in my case, the toolbox with preset features on top of GPT), a feedback loop with the machine becomes possible. You are literally bonding with a smart machine that strengthens your content.

Most importantly in my case, the final output before translation is content that is tailored to reduce the risk of mistranslation, which is now possible by having GPT make a 'Machine English' version.

When I write for an audience that doesn't read or understand English—usually Japanese people—the beauty of the style doesn't matter. Ensuring the message is clear and free of ambiguity is the most important part.