We ran a test where agencies were given up to 1.000 keywords to review. They were initially given an English column and a translat. column in. with the request to fix only the real errors.
- Latin languages translat. much better. Asian languages - worse. with the worst results being with Thai.
- For Arabic. we had to jordan mobile database hire a local translator because the error rate exce.. 30%.
With the advent of ChatGPT
the situation has improv.. . but it is important to check the translation to avoid inaccuracies or fabricat. data.
And once. we took a small number of keywords. up to a thousand main ones. and gave them to agencies for translation. and not even for translation. we gave the English column a translat. column into the local language and ask. them to correct the errors. they correct. them. and we calculat. the percentage of errors for each language.
The percentage is not even errors
But the percentage of the tapes that they chang. and we ask. them to change only what was really an error. and not to paraphrase it in order to pretend to translation experiment the target language do the work. Well. actually. we paid them for the entire text. for all the symbols. so we are sure that there was no manipulation here. and in fact. our hypothesis actually came true. That Latin-speaking languages are translat. significantly
Better much worse with Asian languages. worst with Thai
We didn’t translate slovakia business directory Arabic at that time. but we later hir. a translator there too. well. there are just too many mistakes. Probably even more than 30. so with Arabic you ne. to look for a local translator. or GPT chat. GPT chat gives out. translates keys quite well. The main thing is not to lie.