That's really poor Dutch...
Hi Benjamin,
Been following your blog since only a few weeks and learnt a lot already. Keep up the writing please!!
As a native Dutch speaker and self-proclaimed linguist, I just wanted to comment on the sample phrases in this dataset. They're really... bad... π€ It looks like they were translated from English or another language by a translation engine and then "matched" to their supposed English translation.
Some of the phrases aren't even proper Dutch, full stop.
This is not targeted at you as a negative critique. Rather as a start to a conversation on where these datasets come from and if they're curated. No, I haven't googled it to find out :)
Hi,
Yes, I'm not surprised by your observations. Most of the text in OPUS that is not English is translated from English. And a very large part of this translation was not done by professional translators.
The translation process could be described like this, where X is any language:
X -> translation into English (by professional, crowd sourcing, or machine translation...) -> Translation into Dutch (by professional, crowd sourcing, or machine translation...)
So many things can go wrong in this process!
Curating the datasets would work to some extent.