News for hackers. Stuff that matters.

A comment on The surprisingly subtle ways Microsoft Word has changed how we use language. Parent comment

Re:The surprisingly subtle ways Microsoft Word has changed h... by ginko on Sunday November 05, 2023 @07:55AM (#38150725)
Another example I recently ran into is Google translate stripping gender from words even when translating between languages with grammatical gender.

(German) die Banditin -> (Spanish) el bandido

(Spanish) la bandida -> (German) der Bandit

So it always turns a female bandit into a male one.

[ Parent | on HN ]
Re:The surprisingly subtle ways Microsoft Word has changed h... by playingalong on Sunday November 05, 2023 @08:14AM (#38150854)
Because it goes via English?

I am not sure but I think I heard Google Translate always goes via English.

[ Parent | on HN ]
Re:The surprisingly subtle ways Microsoft Word has changed h... by ginko on Sunday November 05, 2023 @08:20AM (#38150889)
> Because it goes via English?

That's my understanding, yes.

[ Parent | on HN ]
Re:The surprisingly subtle ways Microsoft Word has changed h... by bigger_inside on Sunday November 05, 2023 @09:15AM (#38151284)
That explains SO much.
[ Parent | on HN ]
Re:The surprisingly subtle ways Microsoft Word has changed h... by mcpackieh on Sunday November 05, 2023 @09:58AM (#38151663)
Yandex Translate gets it wrong in the same way, suggesting that they also go through English instead of Russian, or something else is going on.
[ Parent | on HN ]
Re:The surprisingly subtle ways Microsoft Word has changed h... by cpeterso on Sunday November 05, 2023 @02:26PM (#38154664)
Based on this 2016 Google Brain article, “Zero-Shot Translation with Google’s Multilingual Neural Machine Translation System”, I assumed Google Translate directly translated languages:

https://blog.research.google/2016/11/zero-shot-translation-w...

But I just tested some end-to-end translations of some random simple sentences from German and Spanish Wikipedia. The translations are identical whether I translate them directly or via English as an intermediate step. This didn’t prove that Google Translate is using English (or some other interlingua), but at least demonstrates doesn’t not use one. :)

German example:

de = Wikipedia ist ein Projekt zum Aufbau einer Enzyklopädie aus freien Inhalten.

de:en = Wikipedia is a project to build an encyclopedia of free content.

en:es = Wikipedia es un proyecto para construir una enciclopedia de contenido gratuito.

de:es = Wikipedia es un proyecto para construir una enciclopedia de contenido gratuito.

Spanish example:

es = Wikipedia es una enciclopedia libre, políglota y editada de manera colaborativa.

es:en = Wikipedia is a free, polyglot and collaboratively edited encyclopedia.

en:de = Wikipedia ist eine freie, polyglotte und gemeinschaftlich bearbeitete Enzyklopädie.

es:de = Wikipedia ist eine freie, polyglotte und gemeinschaftlich bearbeitete Enzyklopädie.

[ Parent | on HN ]