MT pretranslated segments being locked by AI Thread poster: Johan Beyens
|
|---|
Johan Beyens
Belgium
Local time: 15:56
Member (2024)
English to Dutch
+ ...
I’ve recently worked on MTPE projects where some pretranslated MT segments were locked by AI. In these cases, the AI determined that certain segments were 'good enough' to require no further human input.
I found that the quality of these locked translations was abysmal. After discussing it with the PM, it was agreed that I should edit these segments as well, and they would be included in the total word count.
I’m curious:
Is this becoming a common p... See more I’ve recently worked on MTPE projects where some pretranslated MT segments were locked by AI. In these cases, the AI determined that certain segments were 'good enough' to require no further human input.
I found that the quality of these locked translations was abysmal. After discussing it with the PM, it was agreed that I should edit these segments as well, and they would be included in the total word count.
I’m curious:
Is this becoming a common practice in MTPE workflows?
Have you worked on similar projects? How did you handle it?
Looking forward to hearing your thoughts!
[Edited at 2024-11-18 10:59 GMT] ▲ Collapse
| | | | Lieven Malaise
Belgium
Local time: 15:56
Member (2020)
French to Dutch
+ ...
Johan Beyens wrote:
I’ve recently worked on MTPE projects where some pretranslated MT segments were locked by AI. In these cases, the AI determined that certain segments were 'good enough' to require no further human input.
It's the first time I hear about this. I must say that I don't understand the concept, because AI simply isn't able to guarantee that something is correctly or even "good enough" translated. We would be all out of work instantly if it could do that.
| | | | Zea_Mays 
Italy
Local time: 15:56
Member (2009)
English to German
+ ...
| Can't imagine such a workflow working well | Nov 18, 2024 |
Relying on AI for such sensitive steps could only be a strategy used by 'naive' agencies.
Rather, we might expect some LLM to be used for an initial 'editing', with human translators doing the final checking round.
| | | | Johan Beyens
Belgium
Local time: 15:56
Member (2024)
English to Dutch
+ ...
TOPIC STARTER | so not a new trend then | Nov 19, 2024 |
Needless to say, I was just as surprised when I encountered this. I can't imagine what went through the mind of whoever thought this would be a good idea. Still, they went ahead with it... and this wasn't some small or obscure agency. At least it is reassuring to hear that this isn't a widespread practice. I have to admit, part of me wishes I had just stuck to the instructions and ignored the out-of-scope segments. You know, just to see what would have happened ... See more Needless to say, I was just as surprised when I encountered this. I can't imagine what went through the mind of whoever thought this would be a good idea. Still, they went ahead with it... and this wasn't some small or obscure agency. At least it is reassuring to hear that this isn't a widespread practice. I have to admit, part of me wishes I had just stuck to the instructions and ignored the out-of-scope segments. You know, just to see what would have happened ▲ Collapse
| | |
|
|
|
Min Xuanping
China
Local time: 22:56
Member
English to Chinese
+ ...
| Not the problem of AI, but the problem of careless human translatoin | Nov 19, 2024 |
This is a common case. I work on AI translations, also see locked segements that were translated so wrong. I simply report those segements by clicking the "report" icon aside each segement ONLY when I see them, but gradually I have to ignore them. I am paid on an hourly rate, and have no time to frequently write emails or fill their feedback form. I believe those locked segements were confirmed correct by careless editors who are not qualified to give a final go. It is strange that nobody cares ... See more This is a common case. I work on AI translations, also see locked segements that were translated so wrong. I simply report those segements by clicking the "report" icon aside each segement ONLY when I see them, but gradually I have to ignore them. I am paid on an hourly rate, and have no time to frequently write emails or fill their feedback form. I believe those locked segements were confirmed correct by careless editors who are not qualified to give a final go. It is strange that nobody cares if quality is correct when it comes with AI translations. I intentionally help AI so that it can learn fast and easily, but AI translations can never be trusted without a human's editing. The translation industruy should use AI for productivity, but not for quality — for which I never believe it will come true. ▲ Collapse
| | | | expressisverbis
Portugal
Local time: 14:56
Member (2015)
English to Portuguese
+ ...
| I think it's not a common practice, but... | Nov 21, 2024 |
... This year a translation agency contacted me for ongoing collaboration.
They assigned me the first job of just a few words, but with segments pre-translated with AI and these segments were locked (and so validated!).
The quality of the pre-translation done by the AI was poorly terrible! A massacre on the Portuguese language!
I informed them of this and they told me that their direct client wanted it that way, and didn't want to touch those segments because of the higher cos... See more ... This year a translation agency contacted me for ongoing collaboration.
They assigned me the first job of just a few words, but with segments pre-translated with AI and these segments were locked (and so validated!).
The quality of the pre-translation done by the AI was poorly terrible! A massacre on the Portuguese language!
I informed them of this and they told me that their direct client wanted it that way, and didn't want to touch those segments because of the higher costs of human intervention.
I handed over that small job, asking them never to send me anything from that client again, as I don't want to be associated with shitty jobs with AI in the middle.
Also, I added I have professional pride and a reputation to protect!
I didn't even bother to tell them that what their client was doing was very serious, and very wrong!
[Edited at 2024-11-21 10:56 GMT] ▲ Collapse
| | | | Lingua 5B
Bosnia and Herzegovina
Local time: 15:56
Member (2009)
English to Croatian
+ ...
| Saw something kind of similar | Nov 21, 2024 |
First off, I don’t do editing projects, be it human translation or MT. But I saw something similar to what you described.
I worked on a large project in online Phrase tool. They kept showing and pushing MT suggestions (some of them even locked) which I asked the PM to handle and remove as they are quite distracting and disturbing for my workflow, creating more work for me to remove all of them (there were tons). They responded and told me never mind they had cancelled them, they k... See more First off, I don’t do editing projects, be it human translation or MT. But I saw something similar to what you described.
I worked on a large project in online Phrase tool. They kept showing and pushing MT suggestions (some of them even locked) which I asked the PM to handle and remove as they are quite distracting and disturbing for my workflow, creating more work for me to remove all of them (there were tons). They responded and told me never mind they had cancelled them, they kept coming back (or the PM, who was also the Phrase project admin, didn’t know how to do it properly). Was quite disturbing.
Honestly, I was under the impression these AI/MT segments were being stubbornly pushed as a part of AI promotion (just my impression).
Your scenario might have also been something like this: somebody before you (a human translator) in the crowd translation chain, said “Yes, OK” to this segment, and then it was locked as OK.
In what tool are you working? ▲ Collapse
| | | | Johan Beyens
Belgium
Local time: 15:56
Member (2024)
English to Dutch
+ ...
TOPIC STARTER | AFAIK no humans involved | Nov 21, 2024 |
It was a proprietary tool from a major LSP (I'd rather not name them as I don't want to burn any bridges )
As far as I know this was a new project and no humans were involved in the locked pretranslated segments.
| | |
|
|
|
Ken Ho
China
Local time: 22:56
English to Chinese
+ ...
| Doesn't it sound worrying? | Nov 23, 2024 |
My feel and expierence is that the use AI translation is already putting a lot of people out of job, probably due to cost concern, or just simply demand reduction.
Some clients that were not quite serious about quality have already turned to AI, and part of them have implemented a workflow of a very light PE for AI output. Those clients who believe in human quality are still insisting on the use of human translation/proofreading/QA, but really don't know how long they can resist th... See more My feel and expierence is that the use AI translation is already putting a lot of people out of job, probably due to cost concern, or just simply demand reduction.
Some clients that were not quite serious about quality have already turned to AI, and part of them have implemented a workflow of a very light PE for AI output. Those clients who believe in human quality are still insisting on the use of human translation/proofreading/QA, but really don't know how long they can resist the lure of the AI solution for a much lower cost concern.
What should we do?
[Edited at 2024-11-23 03:18 GMT]
[Edited at 2024-11-23 03:18 GMT] ▲ Collapse
| | | | Johan Beyens
Belgium
Local time: 15:56
Member (2024)
English to Dutch
+ ...
TOPIC STARTER | agency workflow mismatch | Nov 23, 2024 |
Some clients that were not quite serious about quality have already turned to AI, and part of them have implemented a workflow of a very light PE for AI output. Those clients who believe in human quality are still insisting on the use of human translation/proofreading/QA, but really don't know how long they can resist the lure of the AI solution for a much lower cost concern.
I believe that the typical workflow used by agencies is simply inadequate for producing anything beyond "good enough". This is nothing new, but the advent of AI has made it more obvious. Normally you'd expect a human doing the core work and use technology as support, yet it seems the whole industry insists on doing it backwards.
AI can be a great tool for source analysis, keyword planning, style guides and such. It's useful for editing, too. But instead we keep feeding text into MT and use humans only to fill in the blanks, one line at a time. Sorry for the rant.
| | | | jparis (X)
Local time: 14:56
Russian to English
+ ...
| because it's cheaper | Jan 3 |
[quote]Johan Beyens wrote:
I believe that the typical workflow used by agencies is simply inadequate for producing anything beyond "good enough". This is nothing new, but the advent of AI has made it more obvious. Normally you'd expect a human doing the core work and use technology as support, yet it seems the whole industry insists on doing it backwards.
AI can be a great tool for source analysis, keyword planning, style guides and such. It's useful for editing, too. But instead we keep feeding text into MT and use humans only to fill in the blanks, one line at a time. Sorry for the rant.
Because agencies are in competition with each other, so rates get driven down if some agencies start offering clients cheaper prices for MT.
I also do editing -- copyediting and more serious in-depth editing -- and I have seen agencies offer projects at much cheaper lowball rates because they want freelancers to use "AI tools" like, I kid you not, Grammarly. Which produces a very poor outcome and the clients could simply use Grammarly themselves so as a longer term plan this seems to make little business sense.
| | | | Johan Beyens
Belgium
Local time: 15:56
Member (2024)
English to Dutch
+ ...
TOPIC STARTER | just how much cost saving? | Jan 5 |
I know money is the #1 factor driving this trend, I just have to wonder sometimes how big the cost savings actually are. I have seen plenty of MTPE jobs that simply resulted in a multitude of QA rounds to try and fix the mess - a pure waste of time (and, I presume, money) for all involved. Also there are projects where the localization cost constitutes merely a fraction of the total development budget. Saving those few cents/word (and quality risks) just doesn't seem worth it.
| | | | There is no moderator assigned specifically to this forum. To report site rules violations or get help, please contact site staff » MT pretranslated segments being locked by AI | 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 » |
| | Pastey | Your smart companion app
Pastey is an innovative desktop application that bridges the gap between human expertise and artificial intelligence. With intuitive keyboard shortcuts, Pastey transforms your source text into AI-powered draft translations.
Find out more » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
