Pages in topic:   < [1 2 3] >
Post-editing of MT and MT: WHY do we have to put up with it?
Thread poster: Reea-Silvia Podeanu

Tom Hoar
United States
Local time: 03:52
English
Never is a long time Aug 16

Reea-Silvia Podeanu wrote:

...I will NEVER believe somebody building AI...


Reea-Silvia, what if you have the tools to build your own AI that's accessible only to you, would serve only you and benefit only you? Would you "NEVER believe" yourself?


 

Reea-Silvia Podeanu  Identity Verified
Romania
Local time: 10:52
Member (2011)
English to Romanian
+ ...
TOPIC STARTER
I would never... Aug 22

Tom Hoar wrote:

Reea-Silvia Podeanu wrote:

...I will NEVER believe somebody building AI...


Reea-Silvia, what if you have the tools to build your own AI that's accessible only to you, would serve only you and benefit only you? Would you "NEVER believe" yourself?


I would NEVER under NO circumstance build AI. When you are aware of the many dangers AI holds you would never do that. I am also happy I do not know how.


 

Philippe Etienne  Identity Verified
Spain
Local time: 09:52
Member
English to French
Not my invention Aug 30

Tom Hoar wrote:
...Superstition...

That's what I was told (or I understood) when I asked more info on the process we would follow. That was 10 or more years ago.
But the "Of course" is mine, as I find it very logical that a machine translation engine would somehow leverage "machine learning" to improve the raw output.

I find the fact that MT engines don't "learn" from some kind of feedback loop quite disturbing, actually. But all these machine developments fly way above my comprehension.

Philippe


 

Emma Page
United Kingdom
Local time: 08:52
Member (2017)
French to English
+ ...
Why do translators all assume that MT = shoddy, poorly-paid work? Aug 30

I work with MT. I like doing it, with the following caveats:

My combination is French > English, arguably the language combination in which MT engines are the strongest (due to the huge existing corpora of human-translated material).

AND

I only accept jobs where the MT is good enough to significantly reduce the time I spend on a project.

If the output isn't good enough, then turn down the job and insist on being paid my translation rate. If y
... See more
I work with MT. I like doing it, with the following caveats:

My combination is French > English, arguably the language combination in which MT engines are the strongest (due to the huge existing corpora of human-translated material).

AND

I only accept jobs where the MT is good enough to significantly reduce the time I spend on a project.

If the output isn't good enough, then turn down the job and insist on being paid my translation rate. If you hate the process, then tell your clients you won't do it, full stop. But don't assume that MT can never produce output which is as good as a human translator (I've read plenty of human translations which are far wore than the MT I work with), and don't assume that it's always a bad thing. I make more per hour on my MT jobs than I do on my straight translation jobs.
Collapse


Christine Andersen
Dan Lucas
Michele Fauble
Philippe Etienne
Chris Spurgin
Taner Tanrıöver
Irene McClure
 

Bernhard Sulzer  Identity Verified
United States
Local time: 03:52
Member (2006)
English to German
+ ...
Couple of thoughts and observations Aug 30

Emma Page wrote:

I only accept jobs where the MT is good enough to significantly reduce the time I spend on a project.


How do you determine that? I guess you look at a few paragraphs, compare them with the original text and then assume the rest is just as good?! Or do you just assume it's good enough for other reasons?

Emma Page wrote:
If the output isn't good enough, then turn down the job and insist on being paid my translation rate.


So you charge less when you think the output is good enough?! But isn't it you who determines that? What I mean is you, the human translator, is essential in making that determination. And not only that, you are the one who has to go over every single sentence of original text and MT output and fix what needs to be fixed. And for this your knowledge and skill, you charge less?

Emma Page wrote:
... But don't assume that MT can never produce output which is as good as a human translator (I've read plenty of human translations which are far wore than the MT I work with), ...


MT at least at this stage can certainly not be trusted to guarantee grammatically and stylistically flawless translations. MT is also not capable of making a conscious decision on how good or bad a translation is and what to change/fix. MT is a tool. I use many tools and the goal is always to deliver translations that clients can trust to be correct. That's what I sell. The tools (incl. MT) might help me get there, but there is certainly no reason to charge less. I still am responsible for the final product and that's what the client pays for.

Emma Page wrote:
and don't assume that it's always a bad thing. I make more per hour on my MT jobs than I do on my straight translation jobs.


You said you charge less if the translation is good. So are you saying you make more per hour because it takes you less time? If you use MT to speed up your work and deliver faster, then that would be another reason to either charge more (you are faster) or at least not less. If you mean by your statement that you make more money per hour because you can translate more words/hour, then that wouldn't be a good bargain, at least not IMO. Wouldn't it be selling out a bit? You use tools and your skills and brain to deliver in exceptionally shorter time?! Or, when you say you make more money, is it because you charge more per hour for post-editing than you do for translating? But I don't think so based on what you wrote above.

These are just a few questions/observations that came to mind. So I thought I would share.
B

[Edited at 2019-08-30 17:27 GMT]


 

Dan Lucas  Identity Verified
United Kingdom
Local time: 08:52
Member (2014)
Japanese to English
It's just tools Aug 31

Emma Page wrote:
I work with MT. I like doing it, with the following caveats:

This is a plucky admission from Emma, and I admire that. She is a practitioner. She has tried it and it works for her. That's all that matters.

Personally, I don't use MT for reasons enumerated elsewhere, but I am would not be averse to trialling it and using it if my clients wanted/permitted it. Rather than standing Canute-like against the wave of CAT tools, I suspect that 25 years ago I would have been an early adopter, and would have been one of those making the argument that CAT tools are not the end of the industry. (In which, as far as I can see, I would have been correct.)

I make more per hour on my MT jobs than I do on my straight translation jobs.

Ultimately, this is what matters for me. I don't accept work that I would hate to do, but I see myself primarily as a businessman providing a reliable and conscientious service. I am not some arts and crafts purist, sculpting deathless prose from a block of recalcitrant source text for awestruck clients.

So I take on translation jobs and get them done, competently, professionally, and on time, and ensure that I am well compensated. I'm prepared to put up with projects that I do not find particularly interesting because I see work as a means of supporting my family and funding my personal interests.

If you use MT, and it facilitates that for you, all power to your elbow.

Dan


Michele Fauble
Hans Lenting
Philippe Etienne
Emma Page
Christine Andersen
 

Emma Page
United Kingdom
Local time: 08:52
Member (2017)
French to English
+ ...
Some more details Sep 2

Bernhard Sulzer wrote:

Emma Page wrote:

I only accept jobs where the MT is good enough to significantly reduce the time I spend on a project.


How do you determine that? I guess you look at a few paragraphs, compare them with the original text and then assume the rest is just as good?! Or do you just assume it's good enough for other reasons?



Yes, I look at a few different representative phrases from the text. By now I have a few repeat clients who send texts which the MT always deals with fairly well, but I still check each time. Over the years I've gotten good at making this assessment quickly and accurately. It's very similar to how you might look at a potential job and decide whether it's something you are willing to work on without reading the whole thing, a skill we all use regularly.

Emma Page wrote:
If the output isn't good enough, then turn down the job and insist on being paid my translation rate.


So you charge less when you think the output is good enough?! But isn't it you who determines that? What I mean is you, the human translator, is essential in making that determination. And not only that, you are the one who has to go over every single sentence of original text and MT output and fix what needs to be fixed. And for this your knowledge and skill, you charge less?


Yes. I charge a lower per-word rate based on the source text if I agree to post-edit the machine translation instead of starting from scratch. I charge less because it is easier to read and correct than to translate each sentence from scratch. I don't charge nothing, they're still paying me for my skill. If the MT is bad enough that it is *not* easier to read and correct, then I don't charge less. Do you charge a lower rate for proofreading than you do for translation? Same principle.

Emma Page wrote:
... But don't assume that MT can never produce output which is as good as a human translator (I've read plenty of human translations which are far wore than the MT I work with), ...


MT at least at this stage can certainly not be trusted to guarantee grammatically and stylistically flawless translations. MT is also not capable of making a conscious decision on how good or bad a translation is and what to change/fix. MT is a tool. I use many tools and the goal is always to deliver translations that clients can trust to be correct. That's what I sell. The tools (incl. MT) might help me get there, but there is certainly no reason to charge less. I still am responsible for the final product and that's what the client pays for.


Yes, it's true that the client is paying for my final output. But it's a tool which saves me time and effort, and when that is the case I don't mind passing *some* of those savings on to my customers. As I've explained below, I only work with MT when it increases my margins despite the lower cost to the client.


Emma Page wrote:
and don't assume that it's always a bad thing. I make more per hour on my MT jobs than I do on my straight translation jobs.


You said you charge less if the translation is good. So are you saying you make more per hour because it takes you less time? If you use MT to speed up your work and deliver faster, then that would be another reason to either charge more (you are faster) or at least not less. If you mean by your statement that you make more money per hour because you can translate more words/hour, then that wouldn't be a good bargain, at least not IMO. Wouldn't it be selling out a bit? You use tools and your skills and brain to deliver in exceptionally shorter time?! Or, when you say you make more money, is it because you charge more per hour for post-editing than you do for translating? But I don't think so based on what you wrote above.

These are just a few questions/observations that came to mind. So I thought I would share.
B

[Edited at 2019-08-30 17:27 GMT]


I price my services so that I feel I am fairly compensated for my time, effort, and years of expertise and training. Here's an example: Say I am offered a 3000 word document to translate. When I see the MT output, I can tell that about 30% of the document is translated to 90%+ accuracy (just minor punctuation or single-word changes to be made), and the other 70% is translated fairly well (maybe 2-5 alterations per segment). If I translate it from scratch, say it would take me a day of work. If I translate it from the MT, it will take me 3 hours. I charge less for the MT, but I am able to do the work in half the time, and I can take on another job that day. My MT rate is about 60% of my translation rate, so I make more money per hour this way because I'm twice as fast.

What does "selling out" mean in the context of a professional activity? Yes, in the case above the client receives the text faster than they would have from someone else. My per-hour rate is higher, but the client pays less per word. That means they're a happy customer and I get more work from them in the future, and I'm happy because I'm making money.

It works for me, it won't work for everyone! And all of this is VERY dependent on the quality of the MT output. I never accept google translate quality output from clients asking for a lower rate, and neither should anyone else.


Dan Lucas
mughwI
 

Olly Pekelharing  Identity Verified
Netherlands
Local time: 09:52
Member (2009)
Dutch to English
MT corpora Sep 2

Emma, you say FR > EN "...MT engines are the strongest (due to the huge existing corpora of human-translated material)". Do you particularly mean the UN corpus? I work in NL>EN and so benefit from the EU corpus (though I'm not sure to what degree exactly). Is the UN corpus much bigger?

More on-topic, I do not accept MT editing projects for the simple reason that the source texts I translate are generally pretty bad. In my case, I find that MT is good at recognizing proper nouns and
... See more
Emma, you say FR > EN "...MT engines are the strongest (due to the huge existing corpora of human-translated material)". Do you particularly mean the UN corpus? I work in NL>EN and so benefit from the EU corpus (though I'm not sure to what degree exactly). Is the UN corpus much bigger?

More on-topic, I do not accept MT editing projects for the simple reason that the source texts I translate are generally pretty bad. In my case, I find that MT is good at recognizing proper nouns and can also produce fairly good terminology translations (so that saves me some typing), but I rarely see it produce a sentence I can use 90% of, let alone 100%, and this is mainly down to the fact that the source segment is badly written to start with.

Oliver
Collapse


Emma Page
 

Emma Page
United Kingdom
Local time: 08:52
Member (2017)
French to English
+ ...
More generally speaking... Sep 2

Olly Pekelharing wrote:

Emma, you say FR > EN "...MT engines are the strongest (due to the huge existing corpora of human-translated material)". Do you particularly mean the UN corpus? I work in NL>EN and so benefit from the EU corpus (though I'm not sure to what degree exactly). Is the UN corpus much bigger?

Oliver


The Canadian government produces all of its documentation in parallel French and English versions, then there are the EU and the UN bodies of texts as well. I wouldn't choose this as a hill to die on...I imagine that ESEN or even perhaps NLEN is comparable. But I do think when we talk about MT it's worth making the point that quality will differ drastically based on the volume of good translations which exist in the combination in question. Not to mention whether you are working with a plug-and-play MT or a custom one which can differentiate between different clients and learn from your TM to improve its output.


Olly Pekelharing
 

Bernhard Sulzer  Identity Verified
United States
Local time: 03:52
Member (2006)
English to German
+ ...
We should not be expected to charge less for MTPE Sep 2

Emma Page wrote:

Yes, I look at a few different representative phrases from the text. By now I have a few repeat clients who send texts which the MT always deals with fairly well, but I still check each time. Over the years I've gotten good at making this assessment quickly and accurately. It's very similar to how you might look at a potential job and decide whether it's something you are willing to work on without reading the whole thing, a skill we all use regularly.


As I said in my previous post, it takes a human brain, specifically that of a well-versed translator to determine if the MT output is good and might save you time. But in any case, you have to go over MT and original text to make sure it's accurate or to make it accurate and, if necessary, fix the style. Just because someone hands you MT output that you, based on your skills, assess as good, to me does not mean I shall be expected to charge less. Unfortunately, this is the crux of a huge problem which too many translators IMO simply ignore. By accepting or even offering lower prices for MTPE (often before even checking a job that is labeled as MTPE), they continue to make it look as if the use of MT amounts to more than a tool, something that is on a par with human brain activity. People don't realize that it's the human brain that judges, approves, disproves, corrects, and amends MT, word per word.
It's humans that use MT. Without them, MT remains words on a page without any guarantee of accuracy.

The poster of this thread asked why do we have to put up with MTPE?
No-one really has to.
But unfortunately, more and more people seem to put up with the notion that you have to charge less for a translation when MT was used in the process, no matter if it is "good" or bad.

I stand by my conviction that MT is a tool that anyone can choose to use or not to use, but it should not be treated as a reason to charge less.
It also flies in the face of the fact that we deal with different texts all the time and only the human translator can use MT responsibly every time and applies his or her knowledge and skills to arrive at the actual "translation."


Emma Page wrote:

Yes. I charge a lower per-word rate based on the source text if I agree to post-edit the machine translation instead of starting from scratch. I charge less because it is easier to read and correct than to translate each sentence from scratch. I don't charge nothing, they're still paying me for my skill. If the MT is bad enough that it is *not* easier to read and correct, then I don't charge less. Do you charge a lower rate for proofreading than you do for translation? Same principle.


You need to look at each sentence of the original text. Instead of translating it from scratch, you decide to use the MT output or amend it or correct it. It's not like you can simply gloss over 50 or 70 percent of MT because you determined at the outset that it's good. You have to check it. Even if that takes less time than translating it from scratch (and that's a big "if"), you are the one providing the correct version of each sentence. I usually charge less for proofreading a text that was translated by a competent human translator. That's completely different IMO from proofreading or editing MT output.

If MT output is bad, you write Emma, you don't charge less. Yes, but if it is good you do.Good doesn't mean good until you judge it as good and then you still have to check original and MT word for word. But I am repeating myself.

There is a fundamental difference of opinion between you and me about what to charge when MT is involved.

Emma Page wrote:

Yes, it's true that the client is paying for my final output. But it's a tool which saves me time and effort, and when that is the case I don't mind passing *some* of those savings on to my customers. As I've explained below, I only work with MT when it increases my margins despite the lower cost to the client.


It might save you time and effort, it might, but those savings are savings that are only possible because of the knowledge you bring to the table. So when you pass these time and effort savings on to the client, you actually discount your own intellectual property if you will.


Emma Page wrote:

I price my services so that I feel I am fairly compensated for my time, effort, and years of expertise and training. Here's an example: Say I am offered a 3000 word document to translate. When I see the MT output, I can tell that about 30% of the document is translated to 90%+ accuracy (just minor punctuation or single-word changes to be made), and the other 70% is translated fairly well (maybe 2-5 alterations per segment).

f I translate it from scratch, say it would take me a day of work. If I translate it from the MT, it will take me 3 hours. I charge less for the MT, but I am able to do the work in half the time, and I can take on another job that day. My MT rate is about 60% of my translation rate, so I make more money per hour this way because I'm twice as fast.


Okay. I see. Now I know why more and more agencies are successful asking for MTPE for way less than they should pay. I simply can't follow your logic here.


[Edited at 2019-09-02 18:20 GMT]


Florian Stauber
Vuwesi Chrispus
 

farolingo
Local time: 08:52
Member (2003)
German to English
+ ...
MT cannot survive Nov 7

Reea-Silvia Podeanu wrote:

The artificial intelligence grows more and more and it improves more and more with the help of our own efforts.


Not as far as I can tell.... I have just accepted a post-editing job from a firm that started using it across the board for most jobs a few years ago. I pretty much stopped working for them because of the quality of the machine translation and the fact that, as you say, it takes much longer to do than straight translation and for half the pay. Anyway, I recently accepted a new post-editing job from them again, 2 years later, mostly out of curiosity. I can honestly say, the machine has learnt absolutely nothing and the output is still as bad as when they first rolled it out. It's an absolute scandal. I also did a proofreading job for the same firm last week to see how other translators were handling this garbage. Much to my satisfaction the translation was terrible, probably because the translator simply didn't have the time to edit the output properly. Why do I find that satisfying? Because it's clear that more and more of the translators who are forced to accept these jobs are refusing to spend any more time on them than they have to - thus leading to dreadful translations for the end client. Very soon the end clients will start noticing this drop in quality and will walk away from these firms... and who knows, perhaps more and more end clients will start going direct to good freelancers. I honestly think Machine Translation will very soon be dead in the professional translation business.


 

Reea-Silvia Podeanu  Identity Verified
Romania
Local time: 10:52
Member (2011)
English to Romanian
+ ...
TOPIC STARTER
I hope you are right Nov 7

farolingo wrote:

Reea-Silvia Podeanu wrote:

The artificial intelligence grows more and more and it improves more and more with the help of our own efforts.


Not as far as I can tell.... I have just accepted a post-editing job from a firm that started using it across the board for most jobs a few years ago. I pretty much stopped working for them because of the quality of the machine translation and the fact that, as you say, it takes much longer to do than straight translation and for half the pay. Anyway, I recently accepted a new post-editing job from them again, 2 years later, mostly out of curiosity. I can honestly say, the machine has learnt absolutely nothing and the output is still as bad as when they first rolled it out. It's an absolute scandal. I also did a proofreading job for the same firm last week to see how other translators were handling this garbage. Much to my satisfaction the translation was terrible, probably because the translator simply didn't have the time to edit the output properly. Why do I find that satisfying? Because it's clear that more and more of the translators who are forced to accept these jobs are refusing to spend any more time on them than they have to - thus leading to dreadful translations for the end client. Very soon the end clients will start noticing this drop in quality and will walk away from these firms... and who knows, perhaps more and more end clients will start going direct to good freelancers. I honestly think Machine Translation will very soon be dead in the professional translation business.


Maybe they didn't update it? Just a thought. Anyway, I do hope you are right. I will be the first to rejoice.

[Edited at 2019-11-07 15:29 GMT]


farolingo
 

farolingo
Local time: 08:52
Member (2003)
German to English
+ ...
Reply Nov 7



Maybe they didn't update it? Just a thought. Anyway, I do hope you are right. I will be the first to rejoice.


Oh, you can be sure they have been updating it constantly. This firm is desparate for machine translation to work. As Emma states in previous posts, it can actually be extremely good on some texts. But only *some* where the source file is formatted perfectly and is well-written with perfect grammar etc. However, for most of the kinds of texts that we deal with as professional translators, it is completely unsuitable. Firms have an absolute cheek to apply an across-the-board policy to MT and expect their translators to simply edit the output. The worst files are ones where the source text is a pdf file and the output in Studio has a million tags. The MT doesn't care about these tags and puts them anywhere. Well at 50% of my normal rate, I don't care about the tags either. I'm certain it can't carry on as more and more translators refuse to spend anything more than the bare minimum of time on these ridiculous jobs...


 

Reea-Silvia Podeanu  Identity Verified
Romania
Local time: 10:52
Member (2011)
English to Romanian
+ ...
TOPIC STARTER
People say: „Why the fuss?” Nov 7

And I was just thinking these days... Did you notice that the number of jobs posted here are severely reduced? I am not sure if it is only my perception, and I do not know how it is for other languages, but this is my feeling that the number of jobs, at least for Romanian decreased very much! I though that maybe agencies get most of the jobs and give them to the freelancers in their list ... but the number of jobs is definitely lower for Romanian compared with the previous years.

Al
... See more
And I was just thinking these days... Did you notice that the number of jobs posted here are severely reduced? I am not sure if it is only my perception, and I do not know how it is for other languages, but this is my feeling that the number of jobs, at least for Romanian decreased very much! I though that maybe agencies get most of the jobs and give them to the freelancers in their list ... but the number of jobs is definitely lower for Romanian compared with the previous years.

Also my aunt very emphatically told my mom that anybody can be a translator... It is not such a big deal to translate if you use Google Translate... She translated the manual of a device using Google Translate and she doesn't understand why do we make such a fuss regarding out profession... And I have a feeling other people started to think like that, too.

I heard many times lately people saying they will just use Google Translate. People started to think that our profession is not much of a real profession in the last few years. I am thinking that I should change something in my life and learn to do something new before long.

I am learning to build houses now.

[Edited at 2019-11-07 15:33 GMT]
Collapse


farolingo
 

IrinaN
United States
Local time: 02:52
English to Russian
+ ...
Hail Luddites! Nov 7

I have not done any PMTE so far, I do not offer editing services at all but I have seen a few examples I wouldn't mind to edit with absolutely minimal efforts.

The other side of the problem is the same I spoke about before. Translators-turned-editors tend to rewrite no matter what. The fact that editing is not about rewriting would be breaking news for them. I have seen work of professional editors who did brilliant job saving seemingly hopeless cr*p three times faster and without c
... See more
I have not done any PMTE so far, I do not offer editing services at all but I have seen a few examples I wouldn't mind to edit with absolutely minimal efforts.

The other side of the problem is the same I spoke about before. Translators-turned-editors tend to rewrite no matter what. The fact that editing is not about rewriting would be breaking news for them. I have seen work of professional editors who did brilliant job saving seemingly hopeless cr*p three times faster and without crying rivers, than translators believing-to-be-editors while working on materials of much better quality.

Editing is a rare talent and a separate profession but everyone is an editor these days. A real disaster in my book. Nothing personal.
Collapse


 
Pages in topic:   < [1 2 3] >


There is no moderator assigned specifically to this forum.
To report site rules violations or get help, please contact site staff »


Post-editing of MT and MT: WHY do we have to put up with it?

Advanced search







SDL Trados Studio 2019 Freelance
The leading translation software used by over 250,000 translators.

SDL Trados Studio 2019 has evolved to bring translators a brand new experience. Designed with user experience at its core, Studio 2019 transforms how new users get up and running and helps experienced users make the most of the powerful features.

More info »
WordFinder Unlimited
For clarity and excellence

WordFinder is the leading dictionary service that gives you the words you want anywhere, anytime. Access 260+ dictionaries from the world's leading dictionary publishers in virtually any device. Find the right word anywhere, anytime - online or offline.

More info »



Forums
  • All of ProZ.com
  • Term search
  • Jobs
  • Forums
  • Multiple search