I'm looking for recommendations for translation services that can automatically and continuously translate texts into different languages for my client. Currently, I'm considering options like Lokalise and DeepL but would really appreciate hearing about any real-world experiences anyone has had with these services or any others that are effective.
4 Answers
Just a heads up: if you choose the AI translation route, your client will still be responsible for the content accuracy. I once worked for a company with a massive product database using AI for translations, and we faced issues with mistranslations that led to problems like false advertising. So, be sure to double-check everything or consider a dedicated translation tool instead.
I recommend trying smaller LLMs—not the top-tier ones—for translation. They can grasp context pretty well and are often cheaper too. For instance, many opt for versions like Gemini 1.5 or 2.0 because they offer good speed at around 3 cents per million input tokens. Just remember to validate the translation and keep it properly formatted!
If you're after a complete localization platform that can handle ongoing translation tasks for your app, I'd suggest checking out Tolgee. It has CLI support and native SDKs for popular web development frameworks, plus features like in-context editing. If you're just looking for an API to translate strings, I'd recommend using ChatGPT or another LLM API. You can fine-tune it for your app's context, and for smaller apps, passing a localization JSON could yield precise translations. Just an FYI, I'm actually the founder of Tolgee!
You could also consider just using an AI API for simpler tasks. What do you think about GPT-4 or similar models?
Yeah, I was thinking along those lines! Just want to make sure the outputs are reliable.
Absolutely right! If an AI translation says, "all products for free today" and it gets published, you could have a lot of happy customers but a very upset client on your hands. There have been cases where misinformation from AI led to big issues, so always verify the output before it goes live.