Ask HN: How do you handle small translations? | Hacker News Reader

Ask HN: How do you handle small translations?

2 pointsgethly10 comments

In case you are running a multilingual website, how are you handling occasional need to translate few sentences into multiple languages?

I have invested a chunk of my time and money to support 9(as of now) languages for the web UI, but as times goes on and new features and changes are being made, I sometimes have a need to add new text here or there. Usually nothing long, 2,3,4.. sentences. A word here and there.

It makes no sense to bother individual translators with such a miniscule job. Agencies are insanely expensive(from my experience 4x of a price of an individual translator) and so I have resorted to basic google translate, which is unfortunately stuck in 2010 and does not produce good results.

I have not found any "fiver" alternative for this situation, so I am wondering if people are doing anything differently?

View on HN

Use chatgpt or deepl. They're much better than Google Translate, especially if you specify the context.

Many chatbots are good at this, including Microsoft Copilot and they can answer questions about languages too if you want to understand the structure, such as "which characters mean 'new-type' in 国家新型都市化計画?"

Personally I'm interested in Japanese and Chinese culture and I frequently find Japanese speakers engage with my photos on Bluesky so I frequently use it to read responses and write replies where: (1) I can specify tone and context, and then (2) I use a different chatbot to translate my reply back to English to check that it translates properly.

I thought about that but I guess i defaulted to service dedicated to translating that i am used to. If the ML is so good with these, I wonder why google translate(or microsoft and others) have not implemented them into their existing services...

I think Google Translate has implemented some similar ML too, but you can't really prompt it and specify context the way you can with a chatbot.

I'll usually translate it with chatgpt and then run the result backward through Google Translate in order to double check. It's been excellent every time.

> It makes no sense to bother individual translators with such a miniscule job

You're not bothering them if you're paying for the job. On your part, all you need to do is to set up a web-based interface that makes it easy and fast for them to log in and add the translations.

TL;DR There are services that specialize in translating small phrases / paragraphs. Search tattoo translator services (but vet properly, there are fake ones too)

I found a few services while looking for a proper translator to turn a small phrase into Old Latin. I can't remember the services now (it was long ago) but there are professional translators, historians and college professors who moonlight on these sites doing small translations for a small fee. Some offer bulk translations of small phrases. One of the translators I got was a professor in classical text/literature/language (can't remember exactly) and another was some professor of ancient/classical history. The translations came back similar but worded a bit different because there was no way to do a 1 to 1 translation and they provided options for sightly different variants.

Machine-created translations are bad. They've become better over time, but they're still bad. If you've ever tried having your native language butchered in the way that machine-created translations butcher my native language on the daily, you'd understand why. I suspect a lot of the push for machine translations comes from monolingual developers in Silicon Valley.

I think it does it reasonably well for some major languages. What languages are you referring to here?

Finnish is a good example of a language where online translations tend to be .. subpar.

Machine translations aren't all made equal though. The LLM ones, especially ChatGPT, are many times better than older systems. They are better writers and translators (of major common languages) than most humans I know... by far.

Reply on news.ycombinator.com