| Pages in topic: [1 2] > | Neural translation post-editing, what to do Thread poster: Pilar Dueñas
|
|---|
Pilar Dueñas 
Costa Rica
Local time: 11:04
Spanish to English
+ ...
Hello,
I've been doing some MTPT for a company that requires that I paraphrase most of the translated segments. So even is the translation is perfect and there is no other way to say the same thing without substantially changing the meaning conveyed in the original, I'm still supposed to paraphrase.
I'm trying to understand how my work trains the software so that it produces better translations. :/
Has anybody on here come across this issue?
| | | | Elena Feriani
Italy
Local time: 19:04
Member
French to Italian
+ ...
They should pay full price for that, not your MTPE rate.
| | | | Jan Truper
Germany
Local time: 19:04
Member (2016)
English to German
You don't only have to translate, but you have to consider the MT output as a negative factor which is taking away options?
I would charge about 150 % of my regular rate for this.
| | | | Lingua 5B
Bosnia and Herzegovina
Local time: 19:04
Member (2009)
English to Croatian
+ ...
I think you are training an AI, which is a different job and more in the writing and rewriting area. Nothing to do with translation. Hourly rates for this start at $50 minimum. It doesn't make a translation better, it teaches the machine how a sentence can be rearranged and trains it into some syntax patterns.
| | |
|
|
|
Stepan Konev
Russian Federation
Local time: 20:04
English to Russian
| Don't accept what you believe is wrong | May 17, 2023 |
I don't mind using MT as a tool for translation purposes, but I only accept two categories: the translation is fine and the translation is not fine. All other criteria like "your translation sounds somewhat similar to MT" or "it resembles me of ... you name it" are just stupid to me. If they want you to use a different wording, then why on earth they pre-translate the job with their MT? I would turn down that sort of job unless their rate outweighs the stupid nature of that work.
| | | | | How do they define "most"? | May 17, 2023 |
Pilar Dueñas wrote:
I've been doing some MTPT for a company that requires that I paraphrase most of the translated segments. So even is the translation is perfect and there is no other way to say the same thing without substantially changing the meaning conveyed in the original, I'm still supposed to paraphrase
And are those segments short, everyday phrases rather than fully developed sentences with stylistic nuances and terminological intricacies, so AI is really able to produce many "perfect" translations that don't need rephrasing at all? On the other hand, if a segment contains a word or collocation that translates straightforwardly, do you still have to come up with a paraphrase in most instances? I'm joining the bunch of people who can't quite figure out what this job is about. When I'm working with MT output on my projects, I do major changes to most segments without anyone requiring me to do so. In fact, the ratio of segments with sentences that remain unchanged is usually way less than 10%. Don't know anything about AI's performance in the Eng-Esp pair, but Spanish definitely builds thoughts differently from English
| | | | Kay Denney
France
Local time: 19:04
French to English
There was a thread on this recently and someone said that in order to get paid for segments he had to change something. So the MT translation was fine, but he had to edit it into nonsense just to get paid.
| | | | | Post-editing distance stay way | May 18, 2023 |
Yes, "post-editing distance" as I think it is called is such a bullshit concept. When post-editing, you're supposed to use the MT output as much as possible. But then you get punished for working efficiently and making the fewer possible edits for achieving the required quality/result?
Just nonsense. Better pick a metric that is actually aligned with the translator's interests, not actively self-serving agencies.
Edit: but I'm not sure this applies here.
[Edited ... See more Yes, "post-editing distance" as I think it is called is such a bullshit concept. When post-editing, you're supposed to use the MT output as much as possible. But then you get punished for working efficiently and making the fewer possible edits for achieving the required quality/result?
Just nonsense. Better pick a metric that is actually aligned with the translator's interests, not actively self-serving agencies.
Edit: but I'm not sure this applies here.
[Edited at 2023-05-18 08:58 GMT] ▲ Collapse
| | |
|
|
|
Zea_Mays
Italy
Local time: 19:04
Member (2009)
English to German
+ ...
Lingua 5B wrote:
I think you are training an AI, which is a different job and more in the writing and rewriting area. Nothing to do with translation. Hourly rates for this start at $50 minimum. It doesn't make a translation better, it teaches the machine how a sentence can be rearranged and trains it into some syntax patterns.
If the MT output is "This apple is not green." you have to rewrite this, for example as "This apple has a different color from green".
If you put the latter in DeepL with for example DE as target, it shows you the former among the possible alternatives.
[Bearbeitet am 2023-05-18 10:56 GMT]
| | | | Pilar Dueñas
Costa Rica
Local time: 11:04
Spanish to English
+ ...
TOPIC STARTER
Thank you all for your answers.
I will look for the other thread, thanks for the heads-up.
I know the rate should be much higher, particularly for Spanish into English.
We are sent, say, 1000 segments ranging from a few words to 3-4 sentences. Some are badly spelled and this is where the software has the most problems. However, if the sentence is short and sweet, the translation is usually perfect, but I am required to edit at least 90% of the segments. It's not that I'm penal... See more Thank you all for your answers.
I will look for the other thread, thanks for the heads-up.
I know the rate should be much higher, particularly for Spanish into English.
We are sent, say, 1000 segments ranging from a few words to 3-4 sentences. Some are badly spelled and this is where the software has the most problems. However, if the sentence is short and sweet, the translation is usually perfect, but I am required to edit at least 90% of the segments. It's not that I'm penalised or anything, I just cannot deliver the file with over 10% of segments unchanged.
I was trying to understand how the software "thinks", so thank you for your input. ▲ Collapse
| | | |
This does ring a bell! I might be wrong but this reminds me of the infamous ISAAC/ISSAC project…
| | | | Tony Keily
Local time: 19:04
Italian to English
+ ...
| Couldn't agree more | May 19, 2023 |
Stepan Konev wrote:
I only accept two categories: the translation is fine and the translation is not fine.
It's like when revisers come back and tell you that a sentence here or there is 'too literal' or 'too like the source' when the target sentence does its job 100% and reads perfectly well. I've actually had advance instructions on a job to juggle syntax around, across the board, just so my version will look 'more translated'. It's a fetishistic approach.
And with the explosion of MTPE, all of this nonsense is about to get a lot worse.
| | |
|
|
|
| This sounds suspicious | May 19, 2023 |
They're not being straight with you, and if they're not prepared to pay something like your normal rate you should send them a link to this page and dump them.
| | | | Jeff Whittaker
United States
Local time: 13:04
Spanish to English
+ ...
[Edited at 2023-05-19 19:30 GMT]
| | | | Yaotl Altan
Mexico
Local time: 11:04
Member (2006)
English to Spanish
+ ...
Lingua 5B wrote:
I think you are training an AI, which is a different job and more in the writing and rewriting area. Nothing to do with translation. Hourly rates for this start at $50 minimum. It doesn't make a translation better, it teaches the machine how a sentence can be rearranged and trains it into some syntax patterns.
And that's a way to harm our profession.
| | | | | Pages in topic: [1 2] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Neural translation post-editing, what to do | Anycount & Translation Office 3000 | Translation Office 3000
Translation Office 3000 is an advanced accounting tool for freelance translators and small agencies. TO3000 easily and seamlessly integrates with the business life of professional freelance translators.
More info » |
| | CafeTran Espresso | You've never met a CAT tool this clever!
Translate faster & easier, using a sophisticated CAT tool built by a translator / developer.
Accept jobs from clients who use Trados, MemoQ, Wordfast & major CAT tools.
Download and start using CafeTran Espresso -- for free
Buy now! » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
