The quality of translations from Google Translate can vary from good to absolutely terrible, and some language pairs are much better than other language pairs.
So, why does Google Translate work better for some languages than others?
Did you know that Google Translate can now translate from Uzbek into Zulu? And Javanese into Chichewa? Surely there aren’t many native Zulu speakers who can also speak Uzbek, or Chichewa speakers who can translate from Javanese…
Google now uses ‘Zero-shot’ translation technology, which means that it uses intermediate languages to match up content, and that no translations between the source and target were necessarily entered into the system.
For example, if English was translated to Uzbek and Zulu, then Zulu can be translated to Uzbek, and vice versa. The output is not going to be as good as for some of the other language pairs in Google Translate, however.
Frequently translated languages with many contributions to the database will produce better translations than obscure languages. By adding more phrases to the database, you increase the chance that segments are linked correctly.
Source languages with a highly structured grammar, such as Russian, often produce better translations.
Legal or highly standardised text lends itself more to machine translation than creative or marketing content.
The closer two languages are in terms of language evolution, the better the translations are likely to be. The word order will be similar, as well as the grammar. For example, translating Dutch to English produces much better translations than French to English.
Even when translating something into your own native language, with output that sounds fluent, you may not spot a mistranslation. Using Google Translate for anything apart from getting the gist of a foreign text is risky, as you can’t check it if you don’t have a good understanding of both the source and target languages.
At Web-Translations, while we wholeheartedly embrace new technology, we use only human translation.