Yandex translations

Ask specific questions about your target languages. Beginner questions welcome!
User avatar
sfuqua
Black Belt - 1st Dan
Posts: 1642
Joined: Sun Jul 19, 2015 5:05 am
Location: san jose, california
Languages: Bad English: native
Samoan: speak, but rusty
Tagalog: imperfect, but use all the time
Spanish: read
French: read some
Japanese: beginner, obsessively studying
Language Log: https://forum.language-learners.org/vie ... =15&t=9248
x 6299

Yandex translations

Postby sfuqua » Wed Jan 19, 2022 4:22 pm

I use google translate quite often to translate from various languages into English, and it works pretty well for most of them.
Maybe I am the last person on Earth to try Yandex.com, but lately I've been using it in addition to google. I don't know how they do their translations, but they appear to be different than google's pretty often and they are often better. https://translate.yandex.com/ :D
added:
I just tried yandex on Latin, and uh, it was bad.
2 x
荒海や佐渡によこたふ天の川

the rough sea / stretching out towards Sado / the Milky Way
Basho[1689]

Sometimes Japanese is just too much...

Kraut
Black Belt - 2nd Dan
Posts: 2599
Joined: Mon Aug 07, 2017 10:37 pm
Languages: German (N)
French (C)
English (C)
Spanish (A2)
Lithuanian
x 3204

Re: Yandex translations

Postby Kraut » Wed Jan 19, 2022 5:07 pm

I use it in the Edge Browser from time to time.
Edge Browser offers some excellent text-to-speech voices: 4 Spanish voices! (Mexican, Columbian, Spain ... a Swiss German voice .. among others).
Now when I have texts translated in Google or Deepl they don't allow me to use these voices.
When I use Yandex, I can click on "immersive reader" in Edge and the texts will appear to be read with the voice I choose.
1 x

User avatar
luke
Brown Belt
Posts: 1243
Joined: Fri Aug 07, 2015 9:09 pm
Languages: English (N). Spanish (intermediate), Esperanto (B1), French (intermediate but rusting)
Language Log: https://forum.language-learners.org/vie ... 15&t=16948
x 3631

Re: Yandex translations

Postby luke » Wed Jan 19, 2022 5:31 pm

sfuqua wrote:Maybe I am the last person on Earth to try Yandex.com.

Second to the last. :lol:

Thanks for the tip! I tried it on a sentence I was looking at yesterday and yandex had a wording that I had already settled on and I thought was better than google's suggestion:

O: Pietro Crespi entraba en puntillas al anochecer, con una cinta negra en el sombrero, y hacía una visita silenciosa a una Rebeca que parecía desangrarse dentro del vestido negro con mangas hasta los puños.
T: Pietro Crespi would tiptoe in at dusk, with a black ribbon on his hat, and would pay a silent visit to Rebeca, who seemed to be bleeding to death inside the black dress with sleeves down to her wrists.
Y: Pietro Crespi walked in on tiptoes at dusk, with a black ribbon on his hat, and made a silent visit to a Cardigan who seemed to bleed out inside the black dress with sleeves up to the cuffs.
G: Pietro Crespi would tiptoe in at dusk, wearing a black ribbon on his hat, and pay a silent visit to a Rebeca who seemed to be bleeding to death inside the black dress with sleeves down to the cuffs.
L: Pietro Crespi would enter on tiptoes at dusk, with a black ribbon on his hat, and he would make a silent visit to a Rebeca that seemed to be bleeding out inside the black dress with sleeves down to her wrists.

O = Original
T = Famous translation
Y = Yandex
G = Google
L = luke

I like Yandex and luke's (who came up with it independently) use of "bleed out" in place of "bleed to death". I like that Yandex, Google and Luke are faithful to the "a una Rebeca" to "to a Rebeca". Both of these, "bleed out", and "to a Rebeca" are more observational, fitting with the author's background as a journalist.

While I'm here giving Kudos to machines, I want to mention this great video on a famous and fantastic translator and writer that I came across this morning:

0 x
: 124 / 124 Cien años de soledad 20x
: 5479 / 5500 5500 pages - Reading
: 51 / 55 FSI Basic Spanish 3x
: 309 / 506 Camino a Macondo

User avatar
einzelne
Blue Belt
Posts: 804
Joined: Sat Mar 17, 2018 11:33 pm
Languages: Russan (N), English (Working knowledge), French (Reading), German (Reading), Italian (Reading on Kindle)
x 2882

Re: Yandex translations

Postby einzelne » Wed Jan 19, 2022 5:40 pm

I don't use Yandex translations but one thing which drives me nuts about Google translate is the fact that they always (?) use English as a mediator. So if you want to translate Russian into German, they first translate your Russian phrase into English and then into German. As a result, two Russian phrases (1. Ты красивый 2. Вы красивый) would still be translated in the same way: Du bist shön, while the second phrase should be translated as Sie sind shön. I really don't get their rationale for this approach.

At least, Yandex, it seems, doesn't have this issue.
1 x

vonPeterhof
Blue Belt
Posts: 879
Joined: Sat Aug 08, 2015 1:55 am
Languages: Russian (N), English (C2), Japanese (~C1), German (~B2), Kazakh (~B1), Norwegian (~A2)
Studying: Kazakh, Mandarin, Coptic
Language Log: viewtopic.php?f=15&t=1237
x 2833
Contact:

Re: Yandex translations

Postby vonPeterhof » Wed Jan 19, 2022 6:21 pm

einzelne wrote:I don't use Yandex translations but one thing which drives me nuts about Google translate is the fact that they always (?) use English as a mediator.

This doesn't happen for all pairs of languages. For example, translations between Russian and Ukrainian or between Japanese and Korean go directly, presumably because there's enough of a corpus for direct translations between those languages that Google Translate doesn't have to rely on a mediator language like English. I don't think it's disclosed officially which pairs of languages are translated directly, but one easy way to check a particular pair is to just put in a sentence in English. If it gets translated into the target language as if nothing's wrong then it uses English as a mediator, and if the output stays untranslated then they must be going direct.

Edit: actually now that I've played around with this a bit more this test doesn't seem to work for Irish. If you set it as the source language then the output stays in English no matter what you pick as the target language. Somehow I doubt that there's enough of an Irish/Hmong or Irish/Chichewa corpus of translations for the machine learning algorithm to be drawing upon :lol:
Last edited by vonPeterhof on Wed Jan 19, 2022 6:33 pm, edited 1 time in total.
2 x

User avatar
zenmonkey
Black Belt - 2nd Dan
Posts: 2528
Joined: Sun Jul 26, 2015 7:21 pm
Location: California, Germany and France
Languages: Spanish, English, French trilingual - German (B2/C1) on/off study: Persian, Hebrew, Tibetan, Setswana.
Some knowledge of Italian, Portuguese, Ladino, Yiddish ...
Want to tackle Tzotzil, Nahuatl
Language Log: viewtopic.php?f=15&t=859
x 7030
Contact:

Re: Yandex translations

Postby zenmonkey » Wed Jan 19, 2022 6:32 pm

einzelne wrote:I don't use Yandex translations but one thing which drives me nuts about Google translate is the fact that they always (?) use English as a mediator. So if you want to translate Russian into German, they first translate your Russian phrase into English and then into German. As a result, two Russian phrases (1. Ты красивый 2. Вы красивый) would still be translated in the same way: Du bist shön, while the second phrase should be translated as Sie sind shön. I really don't get their rationale for this approach.

At least, Yandex, it seems, doesn't have this issue.


Google does not first translate into English.
This is incorrect. This used to be the case, and it still gets repeated a lot.

But since the implementation of GNMT in 2016, Google now used contextual, "zero shot" translations (rather than the old statistical methods) with a deep neural net.

You can read about this here: https://ai.googleblog.com/2016/11/zero- ... ogles.html

As of this month, Google now does this for 100+ languages vs the original 16 or so languages.

Here is an example of the correct translation execution. From the example you gave. However, if I remove the 1. or edit the formatting, it will still make the same old mistakes. These are related to errors in the contextual mapping. Not intermediate English. In a paragraph with formal you -- you'll see that maintained. In single sentences, it still falls apart.

Screenshot 2022-01-19 at 10.28.43.png
You do not have the required permissions to view the files attached to this post.
4 x
I am a leaf on the wind, watch how I soar

User avatar
einzelne
Blue Belt
Posts: 804
Joined: Sat Mar 17, 2018 11:33 pm
Languages: Russan (N), English (Working knowledge), French (Reading), German (Reading), Italian (Reading on Kindle)
x 2882

Re: Yandex translations

Postby einzelne » Wed Jan 19, 2022 6:42 pm

zenmonkey wrote:This used to be the case, and it still gets repeated a lot.


I don't know what happens inside their black box but, before writing my reply, I tested "du bist schön/Sie sind schön" and Google failed.
0 x

User avatar
einzelne
Blue Belt
Posts: 804
Joined: Sat Mar 17, 2018 11:33 pm
Languages: Russan (N), English (Working knowledge), French (Reading), German (Reading), Italian (Reading on Kindle)
x 2882

Re: Yandex translations

Postby einzelne » Wed Jan 19, 2022 6:45 pm

vonPeterhof wrote:Edit: actually now that I've played around with this a bit more this test doesn't seem to work for Irish. If you set it as the source language then the output stays in English no matter what you pick as the target language. Somehow I doubt that there's enough of an Irish/Hmong or Irish/Chichewa corpus of translations for the machine learning algorithm to be drawing upon :lol:


Yes, I used it last fall for Latin, had the same issue, and also thought it's because of their small corpus.
0 x

User avatar
zenmonkey
Black Belt - 2nd Dan
Posts: 2528
Joined: Sun Jul 26, 2015 7:21 pm
Location: California, Germany and France
Languages: Spanish, English, French trilingual - German (B2/C1) on/off study: Persian, Hebrew, Tibetan, Setswana.
Some knowledge of Italian, Portuguese, Ladino, Yiddish ...
Want to tackle Tzotzil, Nahuatl
Language Log: viewtopic.php?f=15&t=859
x 7030
Contact:

Re: Yandex translations

Postby zenmonkey » Wed Jan 19, 2022 6:46 pm

vonPeterhof wrote:
einzelne wrote:I don't use Yandex translations but one thing which drives me nuts about Google translate is the fact that they always (?) use English as a mediator.

This doesn't happen for all pairs of languages. For example, translations between Russian and Ukrainian or between Japanese and Korean go directly, presumably because there's enough of a corpus for direct translations between those languages that Google Translate doesn't have to rely on a mediator language like English. I don't think it's disclosed officially which pairs of languages are translated directly, but one easy way to check a particular pair is to just put in a sentence in English. If it gets translated into the target language as if nothing's wrong then it uses English as a mediator, and if the output stays untranslated then they must be going direct.

Edit: actually now that I've played around with this a bit more this test doesn't seem to work for Irish. If you set it as the source language then the output stays in English no matter what you pick as the target language. Somehow I doubt that there's enough of an Irish/Hmong or Irish/Chichewa corpus of translations for the machine learning algorithm to be drawing upon :lol:


It's now being done for all languages they support since 12/2021 - there is no longer any x->English->y translation going on... and they are dealing with low corpus languages with global modeling across multiple source corpuses.

https://ai.googleblog.com/2020/06/recen ... slate.html
3 x
I am a leaf on the wind, watch how I soar

User avatar
zenmonkey
Black Belt - 2nd Dan
Posts: 2528
Joined: Sun Jul 26, 2015 7:21 pm
Location: California, Germany and France
Languages: Spanish, English, French trilingual - German (B2/C1) on/off study: Persian, Hebrew, Tibetan, Setswana.
Some knowledge of Italian, Portuguese, Ladino, Yiddish ...
Want to tackle Tzotzil, Nahuatl
Language Log: viewtopic.php?f=15&t=859
x 7030
Contact:

Re: Yandex translations

Postby zenmonkey » Wed Jan 19, 2022 7:04 pm

einzelne wrote:
zenmonkey wrote:This used to be the case, and it still gets repeated a lot.


I don't know what happens inside their black box but, before writing my reply, I tested "du bist schön/Sie sind schön" and Google failed.


But not for the reason you stated. Google will also fail for single words as the meaning is ambiguous.

Machine translation has issues and will continue to have them for a long-time but they are definitely reaching human-level quality, particularly when you provide them sufficient context. How we use the limitations of AI is on us and we need to understand how they operate to get proper results.

If you use any formal/informal 'you' L1 to German it will currently fail in these short sentences because somehow the contextual models aren't integrating that properly. It will fail less in French and I suspect French has a stricter use of the 'vous' (at least that is my experience as someone who speaks both languages).

If you'd like to see this improve, you can always downvote the translation and make a suggestion. This does become part of the operating corpus. If the process was just "English in the middle" it would not be possible to get these corrections. ;)
3 x
I am a leaf on the wind, watch how I soar


Return to “Practical Questions and Advice”

Who is online

Users browsing this forum: No registered users and 2 guests