| Páginas sobre el tema: [1 2] > | Post-editing MT - What's next? Autor de la hebra: Bernhard Sulzer
|
|---|
I fear it might be the post-MT crash (of our industry, in terms of rates and quality).
"Post editing of English German machine translation, software field" - it sounds to resolute, so natural.
Pointless are any arguments against it (jobs are labeled as such already by many), "compared" it is to fuzzy matches of 50% and more percent, found are the translators that will happily take on anything for any price. Smiles all around. At least on the job board here. Or so it seems.
... See more I fear it might be the post-MT crash (of our industry, in terms of rates and quality).
"Post editing of English German machine translation, software field" - it sounds to resolute, so natural.
Pointless are any arguments against it (jobs are labeled as such already by many), "compared" it is to fuzzy matches of 50% and more percent, found are the translators that will happily take on anything for any price. Smiles all around. At least on the job board here. Or so it seems.
I don't care too much about what's on the job board, but it shows you where things are heading, IMO.
I think I'll do something else and come back in a few years.
But seriously. I don't accept the terms - MT - machine translation - to me, it's not translation; for me there must be a human involved and not just a TM, possibly created by a human or a machine, harvesting previous TMs or other word-, phrase-, and sentence data banks;
post-editing MT - to me, it's not editing a translation, in many fields it's like looking at a word salad that messes with your brain. Then you go back to the OT and take a deep breath. Well, I'm not doing it.
See: http://www.proz.com/forum/machine_translation_mt/269671-whats_your_opinion_on_machine_translation_and_quality-page5.html?action=Reply"e=1&post_id=2312813&start=60#reply
Now. I'm open to anything that helps us and if there is any good use for this whatever you want to call it stand-alone-next-to-translation-salad or TM-produced word bank or whatever, that's fine. Just don't call it (a) post-editing, (b) machine translation and expect ...
... (c) discounted prices and lower rates for it?! Please!
That's the main hook, line, and sinker here - buying the expression "post-editing machine translation;" it is used by some to imply easier work for less pay and it's the same as giving/agreeing to word-discounts based exclusively on "fuzzy" reviews of text segments by CAT tools. The machine (CAT tool) tells you there are so many 100% matches and so many fuzzy matches etc. (based on what now - a human or an MT-ed TM the quality of which is at least questionable) and, based just on that, people have come up with arbitrary (yes, arbitrary logic is applied IMO) discounts for such repetitions. As if that would be acceptable instead of or preferable to a human's review of an original text and to assessing the work involved and determining a fair price!
So how do you feel about it? Where are the colleagues who are happily embracing this new-found vehicle for rock bottom rates (do you have examples?) and where are the ones opposing it? Do you care? Should you care? Should we all care or go on with what professionals do - carry out "human" translations and editing/revision tasks using technical tools when they help us and working for adequate prices to maintain quality and livelihoods? Well, I'll do the latter or find something else altogether before the "crash."
[Edited at 2014-10-27 19:23 GMT] ▲ Collapse
| | | | | Absolutely. Total nonsense. | Oct 27, 2014 |
I am glad that I can agree with you here--it is not professional translation at all--perhaps just some sort of a game, or another means of entertainment.
The German/English MT has the reputation of being one of the best, but regardless, it is not real translation. Wait till you see other pairs--like Polish/English or Lithuanian/English (30%) of Polish words, instead of Lithuanian--the languages are not even from the same group).
It is nobody's fault, though--it is just ... See more I am glad that I can agree with you here--it is not professional translation at all--perhaps just some sort of a game, or another means of entertainment.
The German/English MT has the reputation of being one of the best, but regardless, it is not real translation. Wait till you see other pairs--like Polish/English or Lithuanian/English (30%) of Polish words, instead of Lithuanian--the languages are not even from the same group).
It is nobody's fault, though--it is just as good as it gets, since translation is a human activity, so it might have been slightly naive to assume that machines would be able to take over. I think many people had too high expectations. ▲ Collapse
| | | | | It is getting larger than better, but still... | Oct 27, 2014 |
Never accepted such a job, but played with MTs as a part of an exercises during my postgraduate studies.
The conclusion was that it completely lacked the notion of context (hence, translation is considered "human activity). It also lacked capability to "catch" idioms. I recall we used one Spanish idiom in a number of MTs: "estar en el ajo", which verbatim means "to be in garlic". However, in reality it means something like "to know something very well, to be a real pro in something... See more Never accepted such a job, but played with MTs as a part of an exercises during my postgraduate studies.
The conclusion was that it completely lacked the notion of context (hence, translation is considered "human activity). It also lacked capability to "catch" idioms. I recall we used one Spanish idiom in a number of MTs: "estar en el ajo", which verbatim means "to be in garlic". However, in reality it means something like "to know something very well, to be a real pro in something" (sorry, I am not an into-English translator). I just checked in Google Translate and it is still "to be in garlic". So, no big advances since 2010.
Yesterday I was following a thread in a Russian forum. They were talking about kind of "perle nere". Folks were posting some extracts from MT "jobs". I just could not remain seated. I was falling out of my chair. I was afraid I would suffocate, such was the intensity of my laughter.
My guess is that the "quality" of MTs remains pretty much the same. The database, though, expends. Therefore, you have more terms squeezed into the tool, but still no real "brain" to piece them together properly.
Unless there is an MT revolution any time soon, it will remain a joke, I guess. ▲ Collapse
| | | | | So how can they even consider it? | Oct 28, 2014 |
Some people here suggested that machine translation has improved, however, I struggled when I pasted a review into google translate, just to find out whether it was good or bad. Maybe there are better systems.
But how can agencies consider to ask a translator to revise that, e.g. an IT document. Could they have used TM in addition to MT, I assume that might improve it a bit, but still?
I would apply for these jobs, but I would offer my normal translation rate, explain a... See more Some people here suggested that machine translation has improved, however, I struggled when I pasted a review into google translate, just to find out whether it was good or bad. Maybe there are better systems.
But how can agencies consider to ask a translator to revise that, e.g. an IT document. Could they have used TM in addition to MT, I assume that might improve it a bit, but still?
I would apply for these jobs, but I would offer my normal translation rate, explain and just translate from scratch. ▲ Collapse
| | |
|
|
|
Samuel Murray 
Países Bajos
Local time: 13:57
Miembro 2006
inglés al afrikaans
+ ...
| Two views of MT editing | Oct 28, 2014 |
Bernhard Sulzer wrote:
I don't accept the terms "MT" or "machine translation" – to me, it's not translation. For me, there must be a human involved, and not just a TM (possibly created by a human or a machine, harvested from previous TMs or other word, phrase, and sentence data banks). To me, post-editing MT is not editing a translation – in many fields, it's like looking at a word salad that messes with your brain. Then you go back to the [original text] and take a deep breath. (edited)
In my language combination, for most texts (except extremely jargon rich texts), it is not a word salad but a phrase salad. And most of the phrases are in right order.
But you're right – editing MT matches is less about translation than about editing, if you do it quickly (which, I think, is usually the intention).
You could define MT editing in two ways:
* as correcting the errors made by a machine (I think many anti-MT translators view it that way).
* as specialist translation that uses a limited vocabulary, in which the "editor"'s duty is not to edit but to translate, using that vocabulary, fanning it out a bit with just a bit of elegant variation.
Those of us that use MT mainly as a typing aid would probably use the second definition, because to us, MT editing is not mere editing but assisted human translation.
==
That said, if the translator has no easy access to the source text, and/or if the translator is required to make only minimal changes to the target text, then it becomes a completely different type of task. Then it is not translation... and possibly not even editing, but... aw, let's give it a name... patching.
[Edited at 2014-10-28 07:49 GMT]
| | | | Thayenga
Alemania
Local time: 13:57
Miembro 2009
inglés al alemán
+ ...
Gabriele Demuth wrote:
Some people here suggested that machine translation has improved, however, I struggled when I pasted a review into google translate, just to find out whether it was good or bad. Maybe there are better systems.
But how can agencies consider to ask a translator to revise that, e.g. an IT document. Could they have used TM in addition to MT, I assume that might improve it a bit, but still?
I would apply for these jobs, but I would offer my normal translation rate, explain and just translate from scratch.
Hi Gabriele,
as one PM "accidentally" admitted several years ago, the goal is to get a perfect MT-TM by using what the translator/post-editor has corrected, rather re-translated.
Admittedly, this cannot be applied to all "real/human" translations in all fields. However, if there are recurring phrases, the use of the TM created through MT-PE can save an agency a lot of money, since it would only have to pay for those words/sentences not included in their PE-TM. Or even use MT for those words/sentences and pay 1 peanut per each 100 words (exluding discounts ) to have them post-edited.
On the bright side of the issue, here's a link (German dialect) that shows how entertaining MT can be.
http://tu-dresden.de.saxophone.parallelnetz.de/
Bernhard,
as always you have "the nail hit on the head"
This is a serious issue when applied by "not so serious" agencies. Still, MT can never be as good as a real (human) translation.
You're right, there's no point in arguing any discount demands or shameless rates. The only solution is to not accept any PE-jobs.
Happy translating!
| | | | Samuel Murray
Países Bajos
Local time: 13:57
Miembro 2006
inglés al afrikaans
+ ...
| From whose perspective, and: MT editing versus MT+TM editing | Oct 28, 2014 |
Gabriele Demuth wrote:
Some people here suggested that machine translation has improved, however, I struggled when I pasted a review into Google Translate, just to find out whether it was good or bad.
If you could not understand the source language yourself, how could you say whether the Google Translate version was good or bad? How do you know it isn't just a poor source text to begin with?
Saying that it has improved is not saying that it has been perfected. Translators who claim that machine translation has improved say so from a translator's perspective, not an end-user perspective.
But how can agencies consider to ask a translator to revise that, e.g. an IT document. Could they have used TM in addition to MT, I assume that might improve it a bit, but still?
Personally, if I were offered a choice between editing an MT-only "translation" and a translation that is a mixture of MT and TM, I would choose MT-only. That is because editing an MT match requires a completely different type of skill than editing a TM match. A TM match presumably has good grammar, but doesn't claim to say exactly what the source text says. An MT match is the opposite of that.
| | | | Samuel Murray
Países Bajos
Local time: 13:57
Miembro 2006
inglés al afrikaans
+ ...
| In our industry, terms don't mean what they mean | Oct 28, 2014 |
Bernhard Sulzer wrote:
That's the main hook, line, and sinker here - buying the expression "post-editing machine translation.
Well, in the same way as no-one would seriously consider a "post-editing machine translation readings" job as a "reading job", no-one would likely think that "post-editing machine translations" is a type of translation. That's just how English works, I'm afraid.
Besides, our industry is full of terms that don't really mean what they mean.
Take "ICE match", for example (a.k.a. "context match") – what the CAT venders call "context" is no context at all, and it certainly isn't that which any linguist outside the CAT world would call "context". And what we call "high fuzzy matches" should really be called "low-fuzzy matches", because the closer the match is to the current text, the *less* its fuzziness (we call it "high" because the percentage is high, and +not the match itself). And most people (here and in the real world) use the word "proofreading" for a type of editing that does not involve reading printer's proofs. What is called "quality assurance" in our world does not really assure quality (it just catches some of the blunders). And do you remember "uncleaned" files? Those are actually non-cleaned files.
Perhaps you will prefer it to be called "post-MT editing" or "MT post-editing" instead of "post-editing MT"? I agree that "post-editing machine translation" is a bit ambiguous. It should really be "post-editing of machine translations", right?
| | |
|
|
|
Steven Segaert
Estonia
Local time: 14:57
Miembro 2012
inglés al neerlandés
+ ...
| Not sure clients know themselves | Oct 28, 2014 |
A few agencies I work with use an MT solution to pre-translate materials. The translators are then left to fill the empty spots, and to correct the MT result. One online translation environment that seems to be particularly suited for this is Memsource.
The trick is to "train" a machine solution to give better results than it would out of the box. Microsoft has such a solution, as do other vendors. Agencies then spend a lot of time and money into "training" these machines, and expec... See more A few agencies I work with use an MT solution to pre-translate materials. The translators are then left to fill the empty spots, and to correct the MT result. One online translation environment that seems to be particularly suited for this is Memsource.
The trick is to "train" a machine solution to give better results than it would out of the box. Microsoft has such a solution, as do other vendors. Agencies then spend a lot of time and money into "training" these machines, and expect a return on investment.
Of all the agencies I work with (which is outside of the technical field), not one has managed to get a good result out of this - even if the training is done on the basis of the translation memory of a big and stable client. Still, they try to get the investment back by squeezing the rates they pay translators.
My number one problem with all this is that the end client very often has no idea what is going on. I am in a position where I get feedback from the end client, and the comments are mainly about style, and about the fact that the translation is too literal, too close to the text and stylistically lacking. That is after "human intervention".
These remarks are normal for a post-edit of a machine translation. Machine results can't be edited into a fluent and natural text, unless one re-writes the whole translation. For which one is of course not paid.
On the basis of that, I can make some observations:
1) Many agencies don't deploy custom solutions, meaning that the MT result is worse to begin with.
2) Many agencies try to pay as little as possible for MT results, which means that the translator in the end has no real economic choice but to "clean up", or "patch", as Samual calls it.
3) Clients don't know what is going on and become unhappy with the result - especially if the agency previously did use 100% human translation with a smaller profit margin. Many of these clients will go somewhere else. This explains why many of these agencies have few return clients, which leads to payment problems.
I am sure machine solutions have a future, but I think that future is in targeted QA and terminology management. I very much welcome it when a machine serves me relevant terms for the text at hand, and even subsegment fragments (e.g. the correct citation of a European Regulation, for example). The problem is that few people invest in that because it does not directly lead to visible savings, leading to sub-par results in that respect.
The other area machine translation is useful for is to take care of all the no-budget jobs. The more crappy-but-halfway-understandable machine translation is published, the more it stands out against professionally translated content. Companies who want quality will more clearly see the difference and will better understand the need to pay for it. ▲ Collapse
| | | |
Merab Dekano wrote:
I recall we used one Spanish idiom in a number of MTs: "estar en el ajo", which verbatim means "to be in garlic". However, in reality it means something like "to know something very well, to be a real pro in something" (sorry, I am not an into-English translator). I just checked in Google Translate and it is still "to be in garlic". So, no big advances since 2010.
I checked it 5 seconds ago. If you write estar en el ajo without quotation marks, the translation is 'be in garlic.' But if you write it as "estar en el ajo" with quotation markets, you get "being in the know." Weird, huh?
| | | | | Yes, but not always | Oct 28, 2014 |
Kuochoe Nikoi-Kotei wrote:
Merab Dekano wrote:
I recall we used one Spanish idiom in a number of MTs: "estar en el ajo", which verbatim means "to be in garlic". However, in reality it means something like "to know something very well, to be a real pro in something" (sorry, I am not an into-English translator). I just checked in Google Translate and it is still "to be in garlic". So, no big advances since 2010.
I checked it 5 seconds ago. If you write estar en el ajo without quotation marks, the translation is 'be in garlic.' But if you write it as "estar en el ajo" with quotation markets, you get "being in the know." Weird, huh?
Surprised, really. So, what are we supposed to do, according to Google, copy the source text between quotation marks?
Just for fun, I tried another one: "hacerse el sueco", which in Spanish (idiomatically) means "to ignore something, to look the other way or something like that". Without quotation marks Google proposed "Swedish made". With quotation marks, "get the Swedish" (what, a girlfriend?)
| | | | Inge Luus
Sudáfrica
Local time: 14:57
Miembro 2008
alemán al inglés
+ ...
| Spot on Bernhard! Who takes the final responsibility? | Oct 28, 2014 |
I can't say that post-editing MT is anything remotely like translation. Anyway, its not even called "post-translation" (which is what it should be) but "post-EDITING". A different kettle of fish entirely.
I was recently asked to join a team to do post-editing of MT and I had a look at their "good quality" example. To me, it's not human language and I lack that point of departure to understand what needs to be done. It messes with my brain and I just find it quite impossible. I wish ... See more I can't say that post-editing MT is anything remotely like translation. Anyway, its not even called "post-translation" (which is what it should be) but "post-EDITING". A different kettle of fish entirely.
I was recently asked to join a team to do post-editing of MT and I had a look at their "good quality" example. To me, it's not human language and I lack that point of departure to understand what needs to be done. It messes with my brain and I just find it quite impossible. I wish it would work - I'm all for new ways of working, etc.
My BIG question about post-editing MT is this: who takes responsibility for the final translation? As a member of a translators' association, I am bound to do my translations to the best of my ability and to render the target text in a manner loyal to the source text. In other words, a faithful translation. How faithful can a post-edited MT be when the original translation was not done by a human translator (who understands the nuances of a source and target text, etc) but by a machine (which bases its output on algorithms or whatever). So, if we still have to accept responsibility for the result, the whole process becomes an extremely elaborate checking of the MT to make sure it corresponds to the source text using a text output by a non-human. All for half or quarter the price of a translation from scratch?
And what about those clients who just want the text "tidied up" or some such "to make sense"? Who takes the responsibility when a crucial nuance is misinterpreted by the machine and not picked up by the human post-editor? The machine? The agency that used the machine? The translator who was asked to post-edit? ▲ Collapse
| | |
|
|
|
Bernhard Sulzer
Estados Unidos
Local time: 08:57
inglés al alemán
+ ...
PERSONA QUE INICIÓ LA HEBRA | The output is not good enough to be edited | Oct 28, 2014 |
Samuel Murray wrote:
Perhaps you will prefer it to be called "post-MT editing" or "MT post-editing" instead of "post-editing MT"? I agree that "post-editing machine translation" is a bit ambiguous. It should really be "post-editing of machine translations", right?
Even if you call it post-MT editing, it is still describing the same thing. As long as "MT" and "editing" are used together, it raises immediate concerns regarding the actual work involved and how little people want to pay for it.
I understand the phrase "post-editing MT"/"post-MT editing" or "MT post-editing as a "task" that means "I have to do some editing after the text has been translated by a machine." Unfortunately, the term "post-editing MT" is used to get you to do something for a lower rate. That is the first warning sign.
Secondly, the text has not been translated by the machine. What you get is matches of previously translated bits and pieces, hopefully, and not just sheer guesswork based on a dictionary.
The machine itself doesn't translate. And even if the machine would find everything that is used in the OT already used/translated exactly like it (very improbable unless it's a 100% copy), it will pick something someone has translated before in a certain way. Not necessarily what you need or want.
To me, the technology doesn't warrant calling it "translation" because a) it's not translated anew by the machine - it's a patchwork of words and phrases the machine collected based on some algorithm (and a very bad patchwork, at least in my experience, for longer, complex sentences, and even very poor with regard to the correct terminology), and b) the term implies that it is "good enough" (compared to what we usually refer to as "translation") to be edited.
So, I would suggest to stop using these terms (post-editing Mt etc.) and call it "translation." If a translator thinks he/she wants to use the information provided by a machine, that's fine. Just don't demand that I use it. And don't assume that I must agree to apply a lower fee when the outsourcer gives me a word or phrase salad tossed together by a machine.
[Edited at 2014-10-28 21:52 GMT]
| | | | Inge Luus
Sudáfrica
Local time: 14:57
Miembro 2008
alemán al inglés
+ ...
| Different task - different skill | Oct 28, 2014 |
Maybe we should rather stop asking translators to post edit and get bilingual editors to post edit.
| | | | Phil Hand
China
Local time: 20:57
chino al inglés
| MT is a thing, but it's not translation | Oct 29, 2014 |
Inge Luus wrote:
My BIG question about post-editing MT is this: who takes responsibility for the final translation?
This question is a difficult one even when working on human translations. I think Inge is exactly right that it really cuts to the heart of why PEMT is such a problematic process.
In some ways this thread is a little bit head-in-the-sand. MT is here, and it's not going away. To say it hasn't advanced is patently absurd. I don't know about the last four years, but I know for sure that ten years ago MT was nowhere; now you can quite often understand the results when you run a newspaper article through MT in some European language pairs. We can see it, clients can see it, and to claim that it doesn't exist doesn't help.
There are two points that are worth thinking about. One is the point Bernhard made: that much of the time MT has no place in a professional translation flow. Of course, that's not true all of the time - a lot of colleagues report using it with some success, so we can't claim that MT never helps. But we do have to insist that *we* get to decide when to use it and when not to use it.
The other point is about what translation is. Here's a fact about the world: in many cases, we believe that machines are more reliable than people. Machines carry our messages; they make our cars; they reproduce our music. For any client, introducing an extra human between source text and target text is a risk: the human is unpredictable. Clients *want* MT. They *put up* with us. And while that's true, we will always have a problem arguing for our relevance. If we want to change the argument, we have to present translation as something a bit different: a fundamentally human activity.
| | | | | Páginas sobre el tema: [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 » Post-editing MT - What's next? | 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 » |
| | 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 » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
