| Pages in topic: < [1 2 3 4] > | post-editing of machine translations - Are there any real time savings? Thread poster: Thomas Johansson
|
|---|
| sinks and puddings | May 21, 2022 |
Tom in London wrote: Rachel Waddington wrote:
the title of your post OFF TOPIC Mangling of the English language: I recently heard someone say "the proof is in the pudding". I rank that along with "throwing the kitchen sink". CORRECT VERSIONS The proof of the pudding is in the eating. They threw everything at it but the kitchen sink. TRANSLATED INTO ITALIAN USING DEEPL La prova del budino sta nel mangiare. Hanno messo in campo tutto, tranne il lavello della cucina. [Edited at 2022-05-21 07:34 GMT]
I don't speak Italian, so that means nothing to me.
I don't see how it's relevant either. I don't think anyone is advocating sending unedited MT back to their clients. That doesn't mean it can't usefully be incorporated into a translator's workflow in some circumstances, for some text types.
It would be nice to have this discussion without translators calling each others' abilities into question and throwing insults about.
| | | | | Adequacy of the tool | May 21, 2022 |
Hans Lenting wrote:
After all, MT is just a typing aid and potential inspiration of terminology. Nothing more, nothing less.
The usefulness of MT as a typing aid depends on how much of its output you need to delete and retype. Usually, the more complex the source text is, the more you need to change, and at a certain point this "typing aid" turns into a typing hindrance. In texts that command a premium rate, MT produces gross errors more often than not. There is also a potential quality degradation issue: a translator who is not a complete perfectionist will tend to consider the MT output "good enough", the resulting text being impoverished at least stylistically. Less qualified translators will also fail to notice many deficiencies of the output.
Regarding "potential inspiration of terminology" - this is what separates the men from the boys. Unless the translator knows the proper term off the top of his head, the inspiration is just the first shot, which is then improved by research, and if MT provides a better inspiration than a translator's own brain, I have bad news for that person.
| | | | | bad translators | May 21, 2022 |
Anton Konashenok wrote: Hans Lenting wrote:
After all, MT is just a typing aid and potential inspiration of terminology. Nothing more, nothing less. The usefulness of MT as a typing aid depends on how much of its output you need to delete and retype. Usually, the more complex the source text is, the more you need to change, and at a certain point this "typing aid" turns into a typing hindrance. In texts that command a premium rate, MT produces gross errors more often than not. There is also a potential quality degradation issue: a translator who is not a complete perfectionist will tend to consider the MT output "good enough", the resulting text being impoverished at least stylistically. Less qualified translators will also fail to notice many deficiencies of the output.
A translator who doesn't care about quality will produce bad work one way or another though, won't they? The issue here is unskilled, careless translators, not MT.
And I think we've all seen plenty of horrible howlers and stilted style that never went anywhere near MT at all.
(I'm not an MT evangalist, by the way, I just don't much like the judgemental tone of this debate).
| | | | | Running a toko | May 21, 2022 |
Anton Konashenok wrote:
The usefulness of MT as a typing aid depends on how much of its output you need to delete and retype. Usually, the more complex the source text is, the more you need to change, and at a certain point this "typing aid" turns into a typing hindrance.
True! That's why I use CafeTran Espresso, that allows me to drag over snippets in the MT suggestions, just to transfer them to the Target Segment Editor. No deleting, just dragging what is useful, ignoring the rest.
Regarding "potential inspiration of terminology" - this is what separates the men from the boys. Unless the translator knows the proper term off the top of his head.
Hahaha, very funny. And not at all true.
MT engines can give you a clue to new terms that aren't listed in any glossaries or dictionaries yet.
| | |
|
|
|
| Clues from MT??! | May 21, 2022 |
Hans Lenting wrote:
MT engines can give you a clue to new terms that aren't listed in any glossaries or dictionaries yet.
And where, may I ask, does the MT engine get such clues? Either someone has to put them in there consciously (and then why just there and not in glossaries and dictionaries?), or the engine should derive them from parallel texts and similar information occurring on the public web and in other open sources. Unfortunately, the MT engine is inherently incapable of understanding anything, so such derivation is bound to be fairly indiscriminate. If such results are a valuable clue for the translator, then sorry, under my quality criteria, this translator is direly unqualified to handle the given subject. I am speaking from experience, having seen lots and lots of translations performed by translators with no professional or semi-professional background in the subject field. If they are otherwise diligent and experienced, they may get 90% or even 99% right, but they will invariably make some truly gross and sometimes dangerous mistakes.
Rephrasing the above: For relatively simple translations, sure, feel free to use your MT if it helps you. However, for sufficiently complex ones (those commanding a premium rate, for lack of a better criterion), if you have a background in the subject field, your own clues will be better than those suggested by MT. Conversely, if you have no such background, you have no business handling that subject field at all, MT or no MT.
Rachel Waddington wrote:
A translator who doesn't care about quality will produce bad work one way or another though, won't they? The issue here is unskilled, careless translators, not MT.
You are absolutely right in that. However, many unskilled and careless translators assume that MT will enable them to handle texts beyond their understanding. It happens just because they are unskilled and careless. On the other hands, skilled and diligent ones usually see very little, if any, benefit from MT.
[Edited at 2022-05-21 16:34 GMT]
| | | | expressisverbis
Portugal
Local time: 21:30
Member (2015)
English to Portuguese
+ ...
| Be skilled in your area(s) and know how to use MTPE | May 21, 2022 |
In short, the only real time saving when using post-editing is to have an excellent command of the particular area(s) you are translating, and know how to use Machine Translation Post-Editing in a technical and professional way.
That means to ensure the correct terminology, tone of voice, consistency throughout the text, and that it will meet the intended purpose and target audience.
All the rest is bullshit. That's it.
| | | | | Not yet made available | May 22, 2022 |
Anton Konashenok wrote:
And where, may I ask, does the MT engine get such clues? Either someone has to put them in there consciously (and then why just there and not in glossaries and dictionaries?), or the engine should derive them from parallel texts and similar information occurring on the public web and in other open sources.
Exactly. They must be derived from parallel texts. But there's no 'law' that these must have been made available to the search engines' crawlers too. So, sometimes the MT systems have access to more data.
It's even stranger: I sometimes get nice translations for specific terms when they are in the context of segments ... whereas I don't get these suggestions when I feed these terms directly and isolated to the web interface of the same MT engine. Magic? Nope, just statistics.
I see no reason why I shouldn't use MT engines as a typing aid and term crawler. I stay in charge of the correctness of the translation.
But hey, if you don't like MT: by all means, don't use it. It's just that you cannot draw conclusions about the professional level of colleagues that are critical about MT too, but nevertheless aren't afraid to have a look at it.
| | | | | Further explanation | May 22, 2022 |
Hans Lenting wrote:
That's why I use CafeTran Espresso, that allows me to drag over snippets in the MT suggestions, just to transfer them to the Target Segment Editor. No deleting, just dragging what is useful, ignoring the rest.
CafeTran Espresso has this beautiful feature that allows dragging (groups of) words from resources to the Target Segment Editor.
Nice if you are not a principled opponent of using a mouse during translation. (Of course, CafeTran Espresso has keyboard shortcuts for about everything, except for this dragging and dropping.)
For me, dragging words or phrases from panes with MT suggestions is just like dragging and inserting them from my local resources (TMs, glossaries). I see no difference in principle.
I'm like a composer. I'm sitting in a comfortable chair, in front of a big, hi-res screen, and I compose my target language segments with the mouse and the keyboard. Grabbing something here, inserting something from there.
At the end I read the target segment, and when everything is okay, I navigate to the next segment.
So MT is really just another tool in my toolkit. It's for snippets, not for whole sentences. Let alone for complete texts.
| | |
|
|
|
Baran Keki 
Türkiye
Local time: 23:30
Member
English to Turkish
| @Hans Lenting | May 22, 2022 |
Hi,
You seem to know a lot about machine translation. I posted a topic the other day asking whether the use of MT integrated online CAT tools like Matecat, Smartcat etc. and also the MT plugins used with licensed CAT tools like Trados serve to improve the quality of machine translation. That is to say, if translators, by using such tools, send their translated segments directly into Google's, DeepL's databases and thus, wittingly or unwittingly, contribute to the improvement of MT engines... See more Hi,
You seem to know a lot about machine translation. I posted a topic the other day asking whether the use of MT integrated online CAT tools like Matecat, Smartcat etc. and also the MT plugins used with licensed CAT tools like Trados serve to improve the quality of machine translation. That is to say, if translators, by using such tools, send their translated segments directly into Google's, DeepL's databases and thus, wittingly or unwittingly, contribute to the improvement of MT engines or databases (or whatever you call them).
I've been wondering about this for a while. I'm not that interested in the debate between the so-called 'evangelists and naysayers'. I want to know if the translators using MT are actively contributing to MT's improvement with every segment they translate.
I'm ashamedly ignorant where MT and translation technology are concerned. I have this notion about MT based on what I've heard from here and there (they could be complete BS) and I'd like to know if there is any truth to it. I'd appreciate your and others' opinions on this matter (improvement/feeding of MT databases with the use of web based & MT integrated CAT tools).
Thanks
[Edited at 2022-05-22 10:55 GMT] ▲ Collapse
| | | |
Baran Keki wrote:
Hi,
You seem to know a lot about machine translation. I posted a topic the other day asking whether the use of MT integrated online CAT tools like Matecat, Smartcat etc. and also the MT plugins used with licensed CAT tools like Trados serve to improve the quality of machine translation. That is to say, if translators, by using such tools, send their translated segments directly into Google's, DeepL's databases and thus, wittingly or unwittingly, contribute to the improvement of MT engines or databases (or whatever you call them).
I've been wondering about this for a while. I'm not that interested in the debate between the so-called 'evangelists and naysayers'. I want to know if the translators using MT are actively contributing to MT's improvement with every segment they translate.
I'm ashamedly ignorant where MT and translation technology are concerned. I have this notion about MT based on what I've heard from here and there (they could be complete BS) and I'd like to know if there is any truth to it. I'd appreciate your and others' opinions on this matter (improvement/feeding of MT databases with the use of web based & MT integrated CAT tools).
Thanks
[Edited at 2022-05-22 10:55 GMT]
I don't know a great deal about this, but if you subscribe to Deepl your translation data is not stored, so I presume you are not helping to train the MT in that case. Other than that, I don't know.
| | | | Baran Keki
Türkiye
Local time: 23:30
Member
English to Turkish
Rachel Waddington wrote:
I don't know a great deal about this, but if you subscribe to Deepl your translation data is not stored, so I presume you are not helping to train the MT in that case. Other than that, I don't know.
Thank you for your reply. DeepL is not available in my language. I was assigned tasks with Matecat on a few occasions, and I noticed that it was automatically linked with Google Translate and all the segments were more or less translated. It was as if I was doing a PMTE job, but I was in fact getting paid for a translation job. And the quality of Google Translate was surprisingly good as well. It felt as though those segments had already been translated by human translators before (as opposed to being created/generated by a machine) and the segments I confirmed went into that database to be used by other translators using Google Translate or Matecat.
I see that people here are arguing about the quality produced by machine translation and the use of CAT tools in general, while some asserting their superiority in not resorting to them, but I don't see anybody talking about the role of translators in 'training the MT' as you call it. Isn't that where the real danger lies?
| | | |
Baran Keki wrote: Rachel Waddington wrote:
I don't know a great deal about this, but if you subscribe to Deepl your translation data is not stored, so I presume you are not helping to train the MT in that case. Other than that, I don't know. Thank you for your reply. DeepL is not available in my language. I was assigned tasks with Matecat on a few occasions, and I noticed that it was automatically linked with Google Translate and all the segments were more or less translated. It was as if I was doing a PMTE job, but I was in fact getting paid for a translation job. And the quality of Google Translate was surprisingly good as well. It felt as though those segments had already been translated by human translators before (as opposed to being created/generated by a machine) and the segments I confirmed went into that database to be used by other translators using Google Translate or Matecat. I see that people here are arguing about the quality produced by machine translation and the use of CAT tools in general, while some asserting their superiority in not resorting to them, but I don't see anybody talking about the role of translators in 'training the MT' as you call it. Isn't that where the real danger lies?
Well, yes, I suppose so. I imagine that most translators will ultimately make decisions based on what they see as in their own commercial interests though. Realistically, if something is useful, people will use it.
| | |
|
|
|
Tom in London
United Kingdom
Local time: 21:30
Member (2008)
Italian to English
Baran Keki wrote:
DeepL is not available in my language.
Turkish is available in the Web version.
| | | |
Philip Lees wrote: Ice Scream wrote:
There is also no denying that CAT and MT result in a blander, more literal translation.
I have no wish to get involved in any pissing contest, but I dispute the implication that the translations I deliver now, since I started using CafeTran's MT to save typing, are of inferior quality to those I used to deliver before. The destination is the same, a high quality target text. It's just that I take a slightly different route to get there. And I fully accept that the benefit I experience from using MT may be partly due to the type of text I normally work with and might not apply to everybody.
Sorry, I am not having a pop at anybody in particular, I was challenging the idea that MT will save time, especially after the typing/dictation stage. Every job/translator is different I know. But for the “average” text done properly, it will take longer to finalise MT than my own first draft. As I dictate, I reckon the time saving will be negligible. I also think the chances are I wouldn’t translate to the same standard with MT due to human nature. My output using CAT is certainly slightly lazier than it would be otherwise.
So on this basis I think MT on average will produce worse quality.
| | | | Lieven Malaise
Belgium
Local time: 22:30
Member (2020)
French to Dutch
+ ...
Anton Konashenok wrote:
Regarding "potential inspiration of terminology" - this is what separates the men from the boys. Unless the translator knows the proper term off the top of his head, the inspiration is just the first shot, which is then improved by research, and if MT provides a better inspiration than a translator's own brain, I have bad news for that person.
Are you able to formulate one single sentence without trying to insult other translators that in reality might be (much) better than you because you actually don't know the first thing about their skills? Every other sentence you try to imply that you are highly skilled and far superior, while as far as I know you could be nothing more than a moderate translator who refuses to accept that technology can be really helpful. You can brag as much as you want, but the only certainty that can be derived from your words is that you don't know the first thing about properly using MT and haven't got the first clue about how it can be helfpful.
It's perfectly possible to disagree on something without burying your discussion partner(s) into the ground. You should try it some time. As you can see above, the opposite is dead easy. Be civil.
[Edited at 2022-05-23 06:39 GMT]
| | | | | Pages in topic: < [1 2 3 4] > | There is no moderator assigned specifically to this forum. To report site rules violations or get help, please contact site staff » post-editing of machine translations - Are there any real time savings? | Wordfast Pro | Translation Memory Software for Any Platform
Exclusive discount for ProZ.com users!
Save over 13% when purchasing Wordfast Pro through ProZ.com. Wordfast is the world's #1 provider of platform-independent Translation Memory software. Consistently ranked the most user-friendly and highest value
Buy now! » |
| | Protemos translation business management system | Create your account in minutes, and start working! 3-month trial for agencies, and free for freelancers!
The system lets you keep client/vendor database, with contacts and rates, manage projects and assign jobs to vendors, issue invoices, track payments, store and manage project files, generate business reports on turnover profit per client/manager etc.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
