| Pages in topic: < [1 2 3 4 5 6] > | Forget about machine translation: it still stinks, and it will stink forever Thread poster: Daniel Frisano
|
|---|
Lieven Malaise
Belgium
Local time: 23:00
Member (2020)
French to Dutch
+ ...
| No idea what you are talking about. | Nov 24, 2022 |
Kaspars Melkis wrote:
If the text contains subtle, not-easily-fixable issues and the first reviewer doesn't notice them, most likely it was because the text would seem fine to any average editor.
'Not-easily-fixable issues' should be noticed by an editor and especially by an MT post-editor. I genuinely have no idea what you are talking about. There is no such thing as a professional translator that misses mistakes, as subtle as they may be, and doesn't have to bear responsiblity for that.
This has nothing to do with machine translation as such, really, but with the core of our profession. If a post-editor doesn't correct certain mistakes, than that would be entirely his responsibility and not due to the nature of machine translation.
| | | | jyuan_us 
United States
Local time: 17:00
Member (2005)
English to Chinese
+ ...
Nobody seems to have asked an important question: does the translation agency who assigns you MTPE jobs expect you to deliver client-ready (or camera-ready, or printer ready as they called it in the past) copies?
Traditionally, an agency would have a text translated by a translator, edited by another, then proofread by a third. Or, they may have one translator do any two adjacent steps.
A translation agency may ask you to provide your translation-only service, in which... See more Nobody seems to have asked an important question: does the translation agency who assigns you MTPE jobs expect you to deliver client-ready (or camera-ready, or printer ready as they called it in the past) copies?
Traditionally, an agency would have a text translated by a translator, edited by another, then proofread by a third. Or, they may have one translator do any two adjacent steps.
A translation agency may ask you to provide your translation-only service, in which case they would explicitly make you aware that your translation copy will be edited/proofread by someone at their end. The same agency could also ask you to provide your TEP service, and expect a flawless copy from you.
Now, for those who regularly accept MTPE jobs, what are your clients' expectations? Would you think your clients will have someone else edit and/or proofread your MTPE products?
[Edited at 2022-11-24 14:39 GMT] ▲ Collapse
| | | | Kaspars Melkis
United Kingdom
Local time: 22:00
English to Latvian
+ ...
| now we are getting to the point | Nov 24, 2022 |
[quote]Lieven Malaise wrote:
'Not-easily-fixable issues' should be noticed by an editor and especially by an MT post-editor.
Now we are getting to the point. They should but do they?
I have been given MT translations on couple of occasions and I realized that the translation would be too hard to fix and offered to retranslate the text from scratch. The agency consulted the client and agreed to my offer.
But I know other translators who accept MTPE jobs. Are they super good or just don't care?
There is no such thing as a professional translator that misses mistakes, as subtle as they may be, and doesn't have to bear responsiblity for that.
All translators, even the best ones make mistakes (or misses them). I am looking at this not as individual responsibility but as the process that guarantees that issues get reported and fixed. Some rudimentary attempts are made by ISO 17100 standard. I heard that a similar standard developed for MT translation process was rejected by ISO body. I wonder why?
| | | | Lieven Malaise
Belgium
Local time: 23:00
Member (2020)
French to Dutch
+ ...
Kaspars Melkis wrote: 'Not-easily-fixable issues' should be noticed by an editor and especially by an MT post-editor. Now we are getting to the point. They should but do they?
Why is this even the point? In a perfect translation world every single translator should deliver high quality with a minimum of minor errors. In reality there are more bad translators who deliver mediocre or poor work than good ones. That's the point, and that has nothing to do with machine translation. I'll say it again: your edited machine translation will be as good as the person that has edited it. It has nothing to do with machine translation itself.
Kaspars Melkis wrote:
I have been given MT translations on couple of occasions and I realized that the translation would be too hard to fix and offered to retranslate the text from scratch. The agency consulted the client and agreed to my offer.
Yes, so? Not every text is suitable for machine translation editing. I have always agreed with that.
Kaspars Melkis wrote:
But I know other translators who accept MTPE jobs. Are they super good or just don't care?
If they don't care they are bad translators/editors. What has this to do with machine translation as such?
Kaspars Melkis wrote:
translators, even the best ones make mistakes (or misses them). I am looking at this not as individual responsibility but as the process that guarantees that issues get reported and fixed. Some rudimentary attempts are made by ISO 17100 standard. I heard that a similar standard developed for MT translation process was rejected by ISO body. I wonder why?
You are drifting from your initial point, which was that MT can be a big problem because translators wouldn't notice certain errors in machine translation because of the very nature of that machine translation. It's in no way more dangerous than an editor that overlooks certain errors made by a incompetent human translator. There is simply no such thing as errors that are so subtle that they can't be detected by an editor.
| | |
|
|
|
Kaspars Melkis
United Kingdom
Local time: 22:00
English to Latvian
+ ...
| RCT would help | Nov 24, 2022 |
[quote]Lieven Malaise wrote:
Kaspars Melkis wrote:
Why is this even the point? In a perfect translation world every single translator should deliver high quality with a minimum of minor errors. In reality there are more bad translators who deliver mediocre or poor work than good ones. That's the point, and that has nothing to do with machine translation. I'll say it again: your edited machine translation will be as good as the person that has edited it. It has nothing to do with machine translation itself.
The point is that we don't live in a perfect world. By saying that a perfect translator/post-editor would solve issues with MT text is an opt-out. First of all, I would hesitate to blame translators if the new process we have implemented produces worse outcomes than before. If they are not good enough, they should be provided necessary training.
We need to do a randomized controlled trial – randomly assign human translated and MT text to a group of translators and then evaluate the readability and comprehension of the final text with the target audience. That would provide evidence how well a randomly assigned translator can manage MT.
You are drifting from your initial point, which was that MT can be a big problem because translators wouldn't notice certain errors in machine translation because of the very nature of that machine translation. It's in no way more dangerous than an editor that overlooks certain errors made by a incompetent human translator.
I suspect that MT can be more dangerous, and it is the duty of those who push MT solutions to provide evidence that it is not.
| | | | | Objective assessment | Nov 25, 2022 |
Kaspars Melkis wrote:
We need to do a randomized controlled trial – randomly assign human translated and MT text to a group of translators and then evaluate the readability and comprehension of the final text with the target audience. That would provide evidence how well a randomly assigned translator can manage MT.
As someone who has worked in the medical area for many years, I am all in favour of RCTs, but the problem here is how to achieve an objective evaluation of the outcomes. If you rely on the subjective judgement of the "target audience" (whatever that is), then you know full well that there will be wide disagreement among them over the relative quality of the material they are asked to judge.
Or are you talking about unedited MT output? If so, then I don't think anybody here would claim that it's at the same level as that of a good human translator, so the experiment you propose would be pointless.
Kaspars Melkis wrote:
I suspect that MT can be more dangerous, and it is the duty of those who push MT solutions to provide evidence that it is not.
No. It's your "duty" to provide hard evidence to back up your suspicions. You cannot just announce a suspicion and then demand that others refute it.
I suspect that any examples you can find of bad consequences arising as a result of MT will actually prove to be the fault of a negligent human translator or editor. Like any tool, MT can be used, misused, or abused. In any case, it's not the tool that's responsible, but the user.
| | | | Lieven Malaise
Belgium
Local time: 23:00
Member (2020)
French to Dutch
+ ...
Philip Lees wrote:
If you rely on the subjective judgement of the "target audience" (whatever that is), then you know full well that there will be wide disagreement among them over the relative quality of the material they are asked to judge.
Or are you talking about unedited MT output? If so, then I don't think anybody here would claim that it's at the same level as that of a good human translator, so the experiment you propose would be pointless.
Indeed. It would be as useful as assessing human translation, 'objectively', which is pointless because the output depends on the qualities of the translators, which vary widely. That's no different for machine translation editing because the output also depends on the human who is editing. It's actually not that difficult to understand.
| | | | Kaspars Melkis
United Kingdom
Local time: 22:00
English to Latvian
+ ...
| evidence of safety should be provided by sellers | Nov 25, 2022 |
Philip Lees wrote:
No. It's your "duty" to provide hard evidence to back up your suspicions. You cannot just announce a suspicion and then demand that others refute it.
I suspect that any examples you can find of bad consequences arising as a result of MT will actually prove to be the fault of a negligent human translator or editor. Like any tool, MT can be used, misused, or abused. In any case, it's not the tool that's responsible, but the user.
I am sorry but that's not how it works. I don't claim that MTs definitely have bad consequences. I only express my worries based on plenty of anecdotes. Whereas MT proponents without any evidence claim that it definitely is as good as human translations in certain circumstances without specifying what are those circumstances.
I know I am very unpopular here with the demand that the sellers provide the evidence of effectiveness of their products. Even many doctors fail to follow evidence-based medicine but that is the only way forward.
As for evaluating translations, while it is indeed somewhat subjective, it is possible to ascertain comprehension by asking simple questions – for example, did the text instruct you to take the medicine in the morning? Yes, No or not sure.
[Edited at 2022-11-25 09:39 GMT]
| | |
|
|
|
Kaspars Melkis wrote:
MT proponents without any evidence claim that it definitely is as good as human translations in certain circumstances without specifying what are those circumstances.
I don't remember seeing any such claims - not from professional translators, anyway.
Perhaps it would be helpful if you could provide an example of the kind of claim you're referring to, with links, if necessary.
| | | | Kaspars Melkis
United Kingdom
Local time: 22:00
English to Latvian
+ ...
| Lieven Malaise
Belgium
Local time: 23:00
Member (2020)
French to Dutch
+ ...
Kaspars Melkis wrote:
Read this thread about claims that edited MT will be as good as human translated output with proper editors and reviewers.
Of course EDITED machine translation can be as good, since it is a human that is editing and that human is supposed to edit the text until you have full quality. Furthermore there is, in my experience at least, a second translator that edits the EDITED machine translation. That's basically the exact same 4-eye principle as for conventional translation. The only difference being that the first translator was ASSISTED by machine translation. To be even clearer: the second translator doesn't edit the machine translation, but the end product delivered by the first one, so he can't be bothered by the 'subtle undetectable' mistakes MT has according to you.
Perhaps you should dig into it a liitle more (meaning post-editing yourself), because I'm not under the impression that you thoroughly understand how machine translation editing in the translation business works.
| | | | | Why are people talking (writing) about peanuts? | Nov 25, 2022 |
Lieven Malaise wrote:
Of course EDITED machine translation can be as good, since it is a human that is editing and that human is supposed to edit the text until you have full quality. Furthermore there is, in my experience at least, a second translator that edits the EDITED machine translation. That's basically the exact same 4-eye principle as for conventional translation
I believe most people here should understand the method of MTPE, there's nothing special in it. But that four-eye process can by no means be cheap, or do they really pay peanuts to the MTPE specialist and even less to the MTPPE specialist? Why does anyone even mention peanuts when writing about MTPE jobs? Even non-free MT solutions are relatively inexpensive, so the portion of your standard rate that you'll have to deduct for the convenience of working with MT output will be quite small, but then why are people talking about MTPE job offerings with significant rate cuts? I don't know how the freelance MTPE market works, though. For most of my jobs, I wouldn't accept an offering where my pay rate would be reduced by more than 10% on account of me using MT output. I know how many major changes that output usually requires for the resulting text to make any sense, not to mention being pleasant to read. On the other hand, I do find modern MT engines useful, they often help me understand complicated texts in languages I don't speak, except that if the source language is, for example, Finnish, the MT output is likely to still be incomprehensible. In such instances, the 'Aha!' moment often comes after I've browsed through a few online dictionaries and found results that read nothing like what MT has suggested
| | |
|
|
|
Kaspars Melkis
United Kingdom
Local time: 22:00
English to Latvian
+ ...
| thanks for providing clear evidence (not) | Nov 26, 2022 |
Lieven Malaise wrote: Kaspars Melkis wrote:
Read this thread about claims that edited MT will be as good as human translated output with proper editors and reviewers. Of course EDITED machine translation can be as good, since it is a human that is editing and that human is supposed to edit the text until you have full quality. Furthermore there is, in my experience at least, a second translator that edits the EDITED machine translation. That's basically the exact same 4-eye principle as for conventional translation. The only difference being that the first translator was ASSISTED by machine translation. To be even clearer: the second translator doesn't edit the machine translation, but the end product delivered by the first one, so he can't be bothered by the 'subtle undetectable' mistakes MT has according to you. Perhaps you should dig into it a liitle more (meaning post-editing yourself), because I'm not under the impression that you thoroughly understand how machine translation editing in the translation business works.
The process is clear me. I just don't believe that in practice (not in imaginary ideal conditions) it achieves the same results as human initiated translation process.
Now I doubt it even more that MT assisted translation provides the same results (regardless how many eyes look at it afterwards) because when the human translation is rubbish, it is better not to try to fix it but retranslate it by a more qualified translator. In borderline cases it may still get edited and the outcome may pass all quality checks and yet won't make any linguist proud. Many choices made by the original translator will dominate the translation.
| | | | | Different claims | Nov 26, 2022 |
Kaspars Melkis wrote:
Read this thread about claims that edited MT will be as good as human translated output with proper editors and reviewers.
In other words, the circumstances in which MT should be used are clearly stated, "with proper editors and reviewers", which is not what you said. ("... certain circumstances without specifying what are those circumstances.")
I view google's claims for its own translation engine with probably just as much scepticism as you do.
In any case, that article is from 2016, which is ancient in AI terms and not relevant to today.
I think this "argument" is nothing but a straw man. Rather than "edited MT", I would prefer to call it "MT-assisted human translation". That brings it in line with "dictionary-assisted", "word processor-assisted", "spell-checker-assisted", "internet-assisted", etc., translation
Nobody would claim that a spell-checker could, on its own, produce a good translation. Does that mean we shouldn't use spell-checkers to enhance and improve our translation process?
Ridiculous! Like this entire argument.
| | | | | Different things | Nov 26, 2022 |
Kaspars Melkis wrote:
The process is clear me. I just don't believe that in practice (not in imaginary ideal conditions) it achieves the same results as human initiated translation process.
Now I see the source of the misunderstanding: you're referring to something quite different to what I, and others here, are talking about.
I use MT to speed up part of my translation process, but this is still a "human initiated" process and the MT output is checked as soon as it makes its appearance.
For the record, I wouldn't accept somebody else's raw MT output for editing - though I wouldn't criticise anybody who does. It all depends on the specific application.
But I do find MT to be a useful and time-saving tool, when used judiciously to augment my (all too) human translation work.
| | | | | Pages in topic: < [1 2 3 4 5 6] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Forget about machine translation: it still stinks, and it will stink forever | TM-Town | Manage your TMs and Terms ... and boost your translation business
Are you ready for something fresh in the industry? TM-Town is a unique new site for you -- the freelance translator -- to store, manage and share translation memories (TMs) and glossaries...and potentially meet new clients on the basis of your prior work.
More info » |
| | Trados Studio 2022 Freelance | The leading translation software used by over 270,000 translators.
Designed with your feedback in mind, Trados Studio 2022 delivers an unrivalled, powerful desktop
and cloud solution, empowering you to work in the most efficient and cost-effective way.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
