DeepL experiment #90
DeepL experiment #90
Conversation
Since they are visible in the diff to this branch
Co-authored-by: zsawyer <[email protected]>
Co-authored-by: zsawyer <[email protected]>
Co-authored-by: zsawyer <[email protected]>
Co-authored-by: zsawyer <[email protected]>
Co-authored-by: zsawyer <[email protected]>
Co-authored-by: zsawyer <[email protected]>
Co-authored-by: zsawyer <[email protected]>
There was a problem hiding this comment.
I reviewed only a handful of sentences. I think both services are comparable, DeepL has the advantage of using "informal" tú conjugations over the "formal" usted that Google Translate uses, but I got the impression that DeepL was trained with content from Spain because it uses expressions/words that only Spaniards would use.
Both services produce weird translations occassionally, though as a first (automated) step they are pretty good.
| 00:00:06,060 --> 00:00:10,553 | ||
| Está mal escrito y renderizado con una resolución extremadamente baja de 28x28 píxeles, | ||
| 00:00:06,060 --> 00:00:09,964 | ||
| Está escrito de forma chapucera y representado a una resolución extremadamente |
There was a problem hiding this comment.
I think the original was better. "de forma chapucera" sounds very Spaniard (and I think I read most viewers of this channel come from Latin America) and (probably) too informal.
| 00:00:19,700 --> 00:00:23,435 | ||
| Quiero decir, esto, esto y esto también son reconocibles como 3, | ||
| 00:00:19,700 --> 00:00:23,311 | ||
| Es decir, esto, esto y esto también son reconocibles como 3s, |
There was a problem hiding this comment.
Some people would indeed add an "s", but to the best of my knowledge it should still be just "3":
| Es decir, esto, esto y esto también son reconocibles como 3s, | |
| Es decir, esto, esto y esto también son reconocibles como 3, |
| 00:00:37,520 --> 00:00:41,137 | ||
| Pero algo en esa increíblemente inteligente corteza visual tuya | ||
| 00:00:37,520 --> 00:00:42,890 | ||
| Pero algo en tu corteza visual locamente inteligente resuelve que representan la |
There was a problem hiding this comment.
"localmente inteligente" = "locally intelligent", which sounds extremely weird. Original sentence sounded better.
| 00:01:31,233 --> 00:01:34,340 | ||
| escucha acerca del aprendizaje entre comillas de una red neuronal. | ||
| 00:01:35,360 --> 00:01:38,422 | ||
| Este vídeo sólo va a estar dedicado al componente de estructura, |
There was a problem hiding this comment.
This, again, yells "Spaniard": "vídeo" is used pretty much only in Spain (reference - in Spanish). Following the logic of folks contributing to Spanish translations, "video" should be preferred as: 1. There are more viewers from Latin America than Spain, 2. There are far more Spanish speakers in The Americas than in Spain.
| Este vídeo sólo va a estar dedicado al componente de estructura, | |
| Este video sólo va a estar dedicado al componente de estructura, |
| 00:00:28,900 --> 00:00:33,085 | ||
| Las células sensibles a la luz particulares de tu ojo que se activan cuando | ||
| 00:00:28,900 --> 00:00:33,031 | ||
| Las células sensibles a la luz concretas de tu ojo que se disparan cuando |
There was a problem hiding this comment.
I find both "particulares" and "concretas" weird here. Does "particular" in the original script mean something like "specific"? If so, I think it would be fine to just drop the adjective in Spanish (to make it sound natural, the sentence would have to get longer than necessary).
While "disparan" is more literal, I feel "activan" is more accurate (though someone more familiar with eye-related terminology could correct me if I am wrong):
| Las células sensibles a la luz concretas de tu ojo que se disparan cuando | |
| Las células sensibles a la luz de tu ojo que se activan cuando |
Yes, good point this applies to the German translations, too. Switching from formal to informal German is a chore in the main branch that we are doing at the moment. Using DeepL means we can skip this transition because it is already done. |
|
Incorporated in a separate commit |
Some community members suggested DeepL was a better API to use than Google Translate, so these are a few initial tests.