Pages in topic:   < [1 2]
Your Experience with Machine Translation Post-Editing: Call for Participants
Thread poster: lnvieira
Lingua 5B
Lingua 5B  Identity Verified
Bosnia and Herzegovina
Local time: 04:44
Member (2009)
English to Croatian
+ ...
Duh? Jun 2, 2017

Mario Chavez wrote:
If you'd like to contribute more to the discussion, I would advise against using LOL as a heading.


Now someone could advise you against using a condescending and sarcastic tone in each and every post?

I see fora as a place for relaxed discussions, this is not a prison. I will certainly not advise anyone on how or what they should write here, or how they should word the titles.

On topic: I did computational linguistics projects in the field (yes, I have "experience", also paid hourly). It's commonly 5-10 MT development companies competing whose software is the best and most accurate pushing you into saying it's theirs and theirs has the best output. Literally pushing you as if you grade their output with low marks, they send you back to correct it toward higher marks (so that they have better sales potential on the market).

Reality: in some language pairs it is refined and could be somewhat useful in gisting and informal conversation. IMO, in a professional context, especially if we are talking about refined and high quality copies, MT has no place in it, as a starting point or in any other way.


 
Mario Chavez (X)
Mario Chavez (X)  Identity Verified
Local time: 23:44
English to Spanish
+ ...
Professional tone, please Jun 2, 2017

Lingua 5B wrote:

Mario Chavez wrote:
If you'd like to contribute more to the discussion, I would advise against using LOL as a heading.


Now someone could advise you against using a condescending and sarcastic tone in each and every post?

I see fora as a place for relaxed discussions, this is not a prison. I will certainly not advise anyone on how or what they should write here, or how they should word the titles.

On topic: I did computational linguistics projects in the field (yes, I have "experience", also paid hourly). It's commonly 5-10 MT development companies competing whose software is the best and most accurate pushing you into saying it's theirs and theirs has the best output. Literally pushing you as if you grade their output with low marks, they send you back to correct it toward higher marks (so that they have better sales potential on the market).

Reality: in some language pairs it is refined and could be somewhat useful in gisting and informal conversation. IMO, in a professional context, especially if we are talking about refined and high quality copies, MT has no place in it, as a starting point or in any other way.



Precisely, to attain a relaxed, respectful and professional tone in these discussions, it's good to remember that this is not a water cooler stop (for verbal discussions), and writing like we're chatting with our buddies at the office or during a seminar (duh!, lol, etc.) is out of line. We speak in one way, we write differently.

I see no reason for your lecturing me on condescending or sarcastic tone in my writing. You don't know me, nor do you know how I speak in person or what facial expressions I use. Please do not make assumptions of a personal nature.

As I said elsewhere, my MT experience is different from that of others. Lucas (the original poster) has been very polite in his requests and remarks. If anyone feels or things that MT post-editing or similar activities are a 'scam' or if anyone else questions Lucas' credentials, there are more productive ways to do that than to shoot the messenger or make ad hominem references.


 
Marina Diaponon
Marina Diaponon
Cote D'ivoire
Local time: 03:44
English to French
My experience Jun 2, 2017

Tom in London wrote:

It is impossible to check/correct/edit a document that has been translated using MT unless the checker has access to the original document in the source language.

In effect, this means that the checker is going through the MT translation line by line, word for word, correcting it and more than likely, completely rewriting it.

I am occasionally asked to do this. On the very rare occasions on which I accept (only 1 so far), I charge my full rate.

[Edited at 2017-05-30 12:31 GMT]


Some clients want you to keep the MT and make it so it'll feel natural...Meaning don't switch words just change the order and it is a REAL pain.

Machine translation post editing requires you to have time since most of the time the rates are low and they going to pay you in 45 days.
So unless you don't have any jobs, don't do it.


 
Daryo
Daryo
United Kingdom
Local time: 03:44
Serbian to English
+ ...
Semantics and professionals Jun 4, 2017

Mario Chavez wrote:

Tom in London wrote:

It is impossible to check/correct/edit a document that has been translated using MT unless the checker has access to the original document in the source language.

In effect, this means that the checker is going through the MT translation line by line, word for word, correcting it and more than likely, completely rewriting it.

I am occasionally asked to do this. On the very rare occasions on which I accept (only 1 so far), I charge my full rate.

[Edited at 2017-05-30 12:31 GMT]


During my 10 months working with MT output (customized MT engine paid for by the client), it was not possible for our team of translators (Korean, French, Italian, German and Spanish) to analyze the source text since it had already been processed by the MT. In our case, we were working with an online tool, and we only had the equivalent of Trados' Concordance search to gain access to much of the context.

Our advantage was that we had final word on how the translation would look and read. Sadly, our managers had no translation management experience nor did they understand the process, or the needs in different languages. In other words, the MT output for Italian might need fewer edits than the Korean or the Spanish.


Only some indirect convoluted access to the Source Text?
Organised by managers who have only some foggy ideas about the translation process?

THAT sounds like it couldn't more professional ...

and a rock solid reference to start using MT left, right and center!

To clarify some points:

there is a priori nothing wrong with using the output of some automated process that is labelled Machine Translation.

Depending on the ST it could be one of the tools that help improve the speed or even the quality of the final result - nothing wrong with that. Ignoring available tools would be counter-productive, irresponsible, silly Luddism etc...

But there is a big "but" - arguing whether that kind of output could be called a real "translation" is more than just "playing semantics" - any human translator worth being called a "translator" [amateur or pro, doesn't matter] will first understand the source text before trying to reproduce the same meaning in another language.

As far as I know, there is not yet any artificial intelligence showing any signs of "understanding" the symbols [letters and numbers, words, especially sentences as a whole] being processed, and I doubt it's going to happen in any near future - so the MT output IS NOT a "translation". (thinking of it, Post Editing of Machine "Translation" should be banned under The Trade Descriptions Act 1968, and called something else)

The day some Artificial Intelligence manages to pass the Turing Test with flying colours - yes "Machine Translation" will be a real translation (and probably of pretty good quality) - but NOT TODAY.

But how it's used today by middlemen, it's just one more excuse for penny-pinching (like fuzzy matches ...).

PEMT is like giving someone a dictionary and then asking for lower rates - makes sense?

If you want to call the final output a "translation" a human must read from beginning to end, in any order and as much as needed the whole ST and compare it with the final output - anyone who had to check translations done by translators suffering from a serious case of Dunning–Kruger syndrome would know much fun it is when the raw material is of dubious quality... and most MT output is not much better than mechanically translating word for word.

One fact says it all: how many agencies are boasting about using MT?

About the argument that you can't discuss murder unless you've done it yourself ...

I do check from time to time the quality of Google Translate, and the results are more or less the same, only proportions vary.

Into English from French:
one part is fairly good - you know what's going one in the ST
one part is nonsense - at least you are aware that you don't know what that part is about
one part sounds perfectly OK but has in fact NOTHING TO DO with the ST - that's the really dangerous part, especially if it contains the key points of the ST!

if you want to use that raw material to produce a translation, you STILL NEED to read and analyse the whole ST from scratch as you don't know in advance which parts are total mistranslations that sound convincing, and still be wary that terms thrown at you by Google Translate might be totally out of context, not to mention that the sentence structure is mostly unusable.

Into English from Serbian: useless rubbish
Into Serbian from English or French: even more useless rubbish, not even good as a glorified dictionary!

Machine Translation worth that name? Not today.


 
Jeff Whittaker
Jeff Whittaker  Identity Verified
United States
Local time: 23:44
Member (2002)
Spanish to English
+ ...
Do "post editors" read and understand the target text or just skim the output for obvious errors? Jun 4, 2017

Let's consider Spanish to English, considered by many to be an "easy" language pair and therefore, one that should be no problem for a computer.

Here is a sentence from a recent document I completed.

El comité absolverá las consultas a las bases mediante el respectivo pliego absolutorio.

The machine translates this as

The committee shall absolve the consultations to the bases by means of the
... See more
Let's consider Spanish to English, considered by many to be an "easy" language pair and therefore, one that should be no problem for a computer.

Here is a sentence from a recent document I completed.

El comité absolverá las consultas a las bases mediante el respectivo pliego absolutorio.

The machine translates this as

The committee shall absolve the consultations to the bases by means of the respective acquittal.

(link here: https://translate.google.com/#auto/en/El%20comité%20absolverá%20las%20consultas%20a%20las%20bases%20mediante%20el%20respectivo%20pliego%20absolutorio. )

My question is, when an underpaid, and probably exhausted "post editor" reads the sentece:

"The committee shall absolve the consultations to the bases by means of the respective acquittal."

Do they say, sounds great, next sentence? Or do they stop for a moment and consider what that sentence means (despite the fact that this sentence will earn them only .30)? Or do they settle for something that just looks and sounds like English, but may 1) have no meaning or 2) have no relation to the source text?

In fact, this small sentence contains four inaccurate translations.

"The committee shall absolve [WRONG] the consultations [WRONG] to the bases [DEFINITELY WRONG] by means of the respective acquittal [NOPE].

People think that if the machine accurately translates 80% of the text, then all the translator has to do is the remaining 20%.

Problem is that the 80% the computer does get right (on a good day) requires virtualy no mental effort on our part (and is certainly not worthy of a discount).

It is the remaining 20% that requires 80% of the work, mental effort, creativity and time.

Machine pseudo translation is like cheap sausage stuffed with random pieces of rotten meat. It may look great on the plate, but it tastes awful and may be hazardous to your health.

It can be a useful tool, but is never justification for a reduced rate.

[Edited at 2017-06-04 22:58 GMT]
Collapse


 
David Sch
David Sch
Ecuador
Local time: 22:44
Dutch to German
Can be useful Jun 4, 2017

I have done quite some post editing of machine translations in the recent past. The work was done for a multinational with its own machine translation system. I was surprised about the quality of the machine translations. The software often found very good phrases; some of the translations were better than I could have done. Almost every sentence contained mistakes as well. That’s the reason the company had to pay a real translator for post editing. They paid 40 Dollars per hour, which is less... See more
I have done quite some post editing of machine translations in the recent past. The work was done for a multinational with its own machine translation system. I was surprised about the quality of the machine translations. The software often found very good phrases; some of the translations were better than I could have done. Almost every sentence contained mistakes as well. That’s the reason the company had to pay a real translator for post editing. They paid 40 Dollars per hour, which is less than my usual hourly wage. But on the other hand, they had lots of work and I could fill up gaps. Though the machine translation was surprisingly good, the final result was much worse than a human translation. Of course. But the company is completely aware of that fact and uses the edited machine translations only for internal purposes.Collapse


 
Mario Chavez (X)
Mario Chavez (X)  Identity Verified
Local time: 23:44
English to Spanish
+ ...
Keeping it professional Jun 5, 2017

Daryo wrote:

Mario Chavez wrote:

Tom in London wrote:

It is impossible to check/correct/edit a document that has been translated using MT unless the checker has access to the original document in the source language.

In effect, this means that the checker is going through the MT translation line by line, word for word, correcting it and more than likely, completely rewriting it.

I am occasionally asked to do this. On the very rare occasions on which I accept (only 1 so far), I charge my full rate.

[Edited at 2017-05-30 12:31 GMT]


During my 10 months working with MT output (customized MT engine paid for by the client), it was not possible for our team of translators (Korean, French, Italian, German and Spanish) to analyze the source text since it had already been processed by the MT. In our case, we were working with an online tool, and we only had the equivalent of Trados' Concordance search to gain access to much of the context.

Our advantage was that we had final word on how the translation would look and read. Sadly, our managers had no translation management experience nor did they understand the process, or the needs in different languages. In other words, the MT output for Italian might need fewer edits than the Korean or the Spanish.


Only some indirect convoluted access to the Source Text?
Organised by managers who have only some foggy ideas about the translation process?

THAT sounds like it couldn't more professional ...

and a rock solid reference to start using MT left, right and center!

To clarify some points:

there is a priori nothing wrong with using the output of some automated process that is labelled Machine Translation.

Depending on the ST it could be one of the tools that help improve the speed or even the quality of the final result - nothing wrong with that. Ignoring available tools would be counter-productive, irresponsible, silly Luddism etc...

But there is a big "but" - arguing whether that kind of output could be called a real "translation" is more than just "playing semantics" - any human translator worth being called a "translator" [amateur or pro, doesn't matter] will first understand the source text before trying to reproduce the same meaning in another language.

As far as I know, there is not yet any artificial intelligence showing any signs of "understanding" the symbols [letters and numbers, words, especially sentences as a whole] being processed, and I doubt it's going to happen in any near future - so the MT output IS NOT a "translation". (thinking of it, Post Editing of Machine "Translation" should be banned under The Trade Descriptions Act 1968, and called something else)

The day some Artificial Intelligence manages to pass the Turing Test with flying colours - yes "Machine Translation" will be a real translation (and probably of pretty good quality) - but NOT TODAY.

But how it's used today by middlemen, it's just one more excuse for penny-pinching (like fuzzy matches ...).

PEMT is like giving someone a dictionary and then asking for lower rates - makes sense?

If you want to call the final output a "translation" a human must read from beginning to end, in any order and as much as needed the whole ST and compare it with the final output - anyone who had to check translations done by translators suffering from a serious case of Dunning–Kruger syndrome would know much fun it is when the raw material is of dubious quality... and most MT output is not much better than mechanically translating word for word.

One fact says it all: how many agencies are boasting about using MT?

About the argument that you can't discuss murder unless you've done it yourself ...

I do check from time to time the quality of Google Translate, and the results are more or less the same, only proportions vary.

Into English from French:
one part is fairly good - you know what's going one in the ST
one part is nonsense - at least you are aware that you don't know what that part is about
one part sounds perfectly OK but has in fact NOTHING TO DO with the ST - that's the really dangerous part, especially if it contains the key points of the ST!

if you want to use that raw material to produce a translation, you STILL NEED to read and analyse the whole ST from scratch as you don't know in advance which parts are total mistranslations that sound convincing, and still be wary that terms thrown at you by Google Translate might be totally out of context, not to mention that the sentence structure is mostly unusable.

Into English from Serbian: useless rubbish
Into Serbian from English or French: even more useless rubbish, not even good as a glorified dictionary!

Machine Translation worth that name? Not today.


Daryo, I find your sarcastic references unpalatable and unpleasant. Please stop. You don't know the people I've worked with nor the nature of the project I was involved in. You are assuming that the translators in my team were fed short sentences as MT output. I am done trying to reason with you.

And if we're going to use proper definitions and names for things, it's the Dunning-Kruger effect, not syndrome. Here's a reference I found (I didn't know what it was): https://www.thisamericanlife.org/radio-archives/episode/585/transcript


 
Mario Chavez (X)
Mario Chavez (X)  Identity Verified
Local time: 23:44
English to Spanish
+ ...
Similarities with my project Jun 5, 2017

David Sch wrote:

I have done quite some post editing of machine translations in the recent past. The work was done for a multinational with its own machine translation system. I was surprised about the quality of the machine translations. The software often found very good phrases; some of the translations were better than I could have done. Almost every sentence contained mistakes as well. That’s the reason the company had to pay a real translator for post editing. They paid 40 Dollars per hour, which is less than my usual hourly wage. But on the other hand, they had lots of work and I could fill up gaps. Though the machine translation was surprisingly good, the final result was much worse than a human translation. Of course. But the company is completely aware of that fact and uses the edited machine translations only for internal purposes.


David, for 10 months a team of translators I was part of worked on a similar project. The documentation was not for public consumption but for decision-makers in key vertical markets. Contrary to what one or two people here assumed, we had access to the source text and thus we could judge whether the MT output was worth rewriting, tweaking or otherwise modifying it.


 
David Sch
David Sch
Ecuador
Local time: 22:44
Dutch to German
It was a good job Jun 5, 2017

Here is one sentence from the instructions for the post editors:

“If possible, make sure the translation is making sense (this is optional when you think you have time).”

These people know what they are doing. They do not expect high quality translations for low rates, but they buy low quality translations for low rates, because this is enough for their purpose.

And they did not put pressure on their translators. It was quite a nice job.


 
Arianne Farah
Arianne Farah  Identity Verified
Canada
Local time: 23:44
Member (2008)
English to French
Separating the 'post-editing' from the 'MT' Jun 7, 2017

I've never post-edited a machine translated text (I've edited text that was obviously text that had gone through a round of MT & editing rather than translating... but that's a story for another day), however I do use MT on a daily basis - I've found it's wonderful as an additional auto-suggest dictionary.

I think we're all looking at the technology from the wrong end : editing MT text is bad for translation quality and for translators, no question about that; I agree with David Sch
... See more
I've never post-edited a machine translated text (I've edited text that was obviously text that had gone through a round of MT & editing rather than translating... but that's a story for another day), however I do use MT on a daily basis - I've found it's wonderful as an additional auto-suggest dictionary.

I think we're all looking at the technology from the wrong end : editing MT text is bad for translation quality and for translators, no question about that; I agree with David Sch's assessment in that matter; but, like him I've also been sometimes surprised by a perfect turn of phrase in MT generated text... but I don't generate MT and then edit it on the off chance I'll find a diamond in the rough, that would be counter-productive, rather I start translating and insert bits of MT text here and there, 1 word, 2 words, 5 words, like an enormous auto-suggest dictionary crossed with a worldwide concordance motor, it's actually quite elegant and saves me loads of time typing!
Collapse


 
David Sch
David Sch
Ecuador
Local time: 22:44
Dutch to German
Which MT provider? Jun 7, 2017

Arianne, may I ask you, which MT provider is the best in your eyes for the purpose you described?

 
Daryo
Daryo
United Kingdom
Local time: 03:44
Serbian to English
+ ...
Human translated in plain English Jun 14, 2017

Mario Chavez wrote:

Daryo wrote:

Mario Chavez wrote:

Tom in London wrote:

It is impossible to check/correct/edit a document that has been translated using MT unless the checker has access to the original document in the source language.

In effect, this means that the checker is going through the MT translation line by line, word for word, correcting it and more than likely, completely rewriting it.

I am occasionally asked to do this. On the very rare occasions on which I accept (only 1 so far), I charge my full rate.

[Edited at 2017-05-30 12:31 GMT]


During my 10 months working with MT output (customized MT engine paid for by the client), it was not possible for our team of translators (Korean, French, Italian, German and Spanish) to analyze the source text since it had already been processed by the MT. In our case, we were working with an online tool, and we only had the equivalent of Trados' Concordance search to gain access to much of the context.

Our advantage was that we had final word on how the translation would look and read. Sadly, our managers had no translation management experience nor did they understand the process, or the needs in different languages. In other words, the MT output for Italian might need fewer edits than the Korean or the Spanish.


Only some indirect convoluted access to the Source Text?
Organised by managers who have only some foggy ideas about the translation process?

THAT sounds like it couldn't more professional ...

and a rock solid reference to start using MT left, right and center!

To clarify some points:

there is a priori nothing wrong with using the output of some automated process that is labelled Machine Translation.

Depending on the ST it could be one of the tools that help improve the speed or even the quality of the final result - nothing wrong with that. Ignoring available tools would be counter-productive, irresponsible, silly Luddism etc...

But there is a big "but" - arguing whether that kind of output could be called a real "translation" is more than just "playing semantics" - any human translator worth being called a "translator" [amateur or pro, doesn't matter] will first understand the source text before trying to reproduce the same meaning in another language.

As far as I know, there is not yet any artificial intelligence showing any signs of "understanding" the symbols [letters and numbers, words, especially sentences as a whole] being processed, and I doubt it's going to happen in any near future - so the MT output IS NOT a "translation". (thinking of it, Post Editing of Machine "Translation" should be banned under The Trade Descriptions Act 1968, and called something else)

The day some Artificial Intelligence manages to pass the Turing Test with flying colours - yes "Machine Translation" will be a real translation (and probably of pretty good quality) - but NOT TODAY.

But how it's used today by middlemen, it's just one more excuse for penny-pinching (like fuzzy matches ...).

PEMT is like giving someone a dictionary and then asking for lower rates - makes sense?

If you want to call the final output a "translation" a human must read from beginning to end, in any order and as much as needed the whole ST and compare it with the final output - anyone who had to check translations done by translators suffering from a serious case of Dunning–Kruger syndrome would know much fun it is when the raw material is of dubious quality... and most MT output is not much better than mechanically translating word for word.

One fact says it all: how many agencies are boasting about using MT?

About the argument that you can't discuss murder unless you've done it yourself ...

I do check from time to time the quality of Google Translate, and the results are more or less the same, only proportions vary.

Into English from French:
one part is fairly good - you know what's going one in the ST
one part is nonsense - at least you are aware that you don't know what that part is about
one part sounds perfectly OK but has in fact NOTHING TO DO with the ST - that's the really dangerous part, especially if it contains the key points of the ST!

if you want to use that raw material to produce a translation, you STILL NEED to read and analyse the whole ST from scratch as you don't know in advance which parts are total mistranslations that sound convincing, and still be wary that terms thrown at you by Google Translate might be totally out of context, not to mention that the sentence structure is mostly unusable.

Into English from Serbian: useless rubbish
Into Serbian from English or French: even more useless rubbish, not even good as a glorified dictionary!

Machine Translation worth that name? Not today.


Daryo, I find your sarcastic references unpalatable and unpleasant. Please stop. You don't know the people I've worked with nor the nature of the project I was involved in. You are assuming that the translators in my team were fed short sentences as MT output. I am done trying to reason with you.

And if we're going to use proper definitions and names for things, it's the Dunning-Kruger effect, not syndrome. Here's a reference I found (I didn't know what it was): https://www.thisamericanlife.org/radio-archives/episode/585/transcript



all your reactions to what you don't like hearing can be summarised by "please keep away your needles from my bubbles..."

I would've imagined that reacting to factual arguments by emotive reactions is more suited to marketing and PR BS, not to "professional" discussions, but then it's just my opinion.

As for PEMT, I made my opinion clear enough, and I'm not in the mood for some "Exercices de style" ["MT" that and try to "post-edit" it ...]


 
delphinetran (X)
delphinetran (X)
France
English to French
+ ...
Hi Lucas, Jun 21, 2017

lnvieira wrote:

Dear all

I'm working on a research project in a partnership between the University of Bristol (UK) and Universidad Pablo de Olavide (Spain), and we are interested in speaking to companies, translators and other language professionals who in one way or another are involved in post-editing of machine translation. The purpose of the project is to gather information on different practices, experiences and perceptions relating to the use of machine translation in human translation workflows. We are conducting short Skype interviews to gather the data. If you think you might be interested in getting involved, please drop me a line at l.nunesvieira[at]bristol.ac.uk and I can send you more details. The Skype chats normally last for about 30-45min, and this can be booked at a time that suits you. The only requirement for taking part is having had some professional experience with machine translation post-editing. This may be as a translator/post-editor (freelance or in-house), as a project manager, as the owner of a translation company, as a technology specialist, etc. We cannot offer payment for this, but participating should be an interesting experience nonetheless!

I look forward to hearing from you.
Lucas

Webpage: http://www.bristol.ac.uk/sml/people/lucas-nunes-vieira/index.html


 
delphinetran (X)
delphinetran (X)
France
English to French
+ ...
Sounds [email protected] Jun 21, 2017

lnvieira wrote:

Dear all

I'm working on a research project in a partnership between the University of Bristol (UK) and Universidad Pablo de Olavide (Spain), and we are interested in speaking to companies, translators and other language professionals who in one way or another are involved in post-editing of machine translation. The purpose of the project is to gather information on different practices, experiences and perceptions relating to the use of machine translation in human translation workflows. We are conducting short Skype interviews to gather the data. If you think you might be interested in getting involved, please drop me a line at l.nunesvieira[at]bristol.ac.uk and I can send you more details. The Skype chats normally last for about 30-45min, and this can be booked at a time that suits you. The only requirement for taking part is having had some professional experience with machine translation post-editing. This may be as a translator/post-editor (freelance or in-house), as a project manager, as the owner of a translation company, as a technology specialist, etc. We cannot offer payment for this, but participating should be an interesting experience nonetheless!

I look forward to hearing from you.
Lucas

Webpage: http://www.bristol.ac.uk/sml/people/lucas-nunes-vieira/index.html


 
Pages in topic:   < [1 2]


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


Your Experience with Machine Translation Post-Editing: Call for Participants







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 »
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! »