| Pages in topic: [1 2] > | Post-Processing of Machine Translations, the New Money Maker of “the Translation Industry” Thread poster: Jeff Whittaker
|
|---|
Jeff Whittaker 
United States
Local time: 15:52
Member (2002)
Spanish to English
+ ...
| Jean Lachaud
United States
Local time: 15:52
English to French
+ ...
| Nothing new here | Feb 9, 2015 |
Although interesting (as in: there's no fault in the reasoning), this is a too-wordy rehash of a beaten-to-death issue, which is of no interest to anyone who is NOT a translator (more precisely, a translator of into-Japanese patents), and useless to those who are.
Just some run-of-the-mill ranting and raving.
| | | | Robert Forstag
United States
Local time: 15:52
Spanish to English
+ ...
| The whole notion seems like a hoax | Feb 9, 2015 |
I don't have a lot of experience with reworking machine translation, but on those occasions when I was asked to edit such material (sometimes by PMs who did not identify it as such) the text in question was so flawed that I could have produced a translation more quickly by simply working from scratch. Agencies "make money" on post-editing by requiring the translator to do more work for less pay. There will not be a lot of takers in developed countries for such a scheme.
Now perhaps ... See more I don't have a lot of experience with reworking machine translation, but on those occasions when I was asked to edit such material (sometimes by PMs who did not identify it as such) the text in question was so flawed that I could have produced a translation more quickly by simply working from scratch. Agencies "make money" on post-editing by requiring the translator to do more work for less pay. There will not be a lot of takers in developed countries for such a scheme.
Now perhaps some agencies generate their machine translations on the basis of memories developed and vetted over the course of many years. In such cases, perhaps the original output is something better than disastrous, and could actually (rather than "mythically") save time. One international medical organization was trying to do this some 7 years ago. I wonder if they or others have made any progress in such endeavors. ▲ Collapse
| | | | Samuel Murray
Netherlands
Local time: 21:52
Member (2006)
English to Afrikaans
+ ...
| Nothing wrong with fuzzy matches per se | Feb 10, 2015 |
Jeff Whittaker wrote:
Exhibit A: How could translation agencies get away with the monstrosity that they call “fuzzy matches”?
I have no objection to the principle of fuzzy matching, but it must be applied right.
The translator must be able to see both the original segment's source text and target text, and preferably the CAT tool should point out to him how the previous segment and the current segment differ from each other. If this information is not present, the fuzzy match often takes the translator longer to review than a complete new translation would have taken, because the translator is forced to carefully evaluate the entire suggestion and not just the portion that had changed (because it is not know which portion had changed).
Another oddity is fuzzy match discounts for editing/proofreading, because the original source text and the original translation is in most cases not present during such a situation, and again it takes the reviewer longer to deal with fuzzy matches because the reviewer has to evaluate not only the translation but also any potential blunders that are so hard to see that the translator didn't even see them himself.
I've skim-read your blog post, but apart from the fact that you are against fuzzy matches I was unable to figure out *why* you believe that fuzzy matching is a flawed concept.
| | |
|
|
|
Samuel Murray wrote:
I've skim-read your blog post, but apart from the fact that you are against fuzzy matches I was unable to figure out *why* you believe that fuzzy matching is a flawed concept.
Unfortunately, agencies often understand fuzzy matching differently from translators. I am all in favour of fuzzy matching within the text and with my own TM. On the other hand, when an agency is trying to reduce my fee because it finds some fuzzy matches in its own TM indiscriminately created from (unreconciled) work of ten different translators, only five of whom are worth their salt and none of whom is a dedicated terminologist, that's where I draw the line. A TM, like a dictionary, is only as good as the worst entry in it.
[Edited at 2015-02-10 11:48 GMT]
| | | | Phil Hand
China
Local time: 03:52
Chinese to English
Anton Konashenok wrote:
I am all in favour of fuzzy matching within the text and with my own TM. On the other hand, when an agency is trying to reduce my fee because it finds some fuzzy matches in its own TM indiscriminately created from (unreconciled) work of ten different translators, only five of whom are worth their salt and none of whom is a dedicated terminologist, that's where I draw the line. A TM, like a dictionary, is only as good as the worst entry in it.
Could not agree more. I love agreeing to fuzzy matches because in my pair the agencies don't usually have big existing TMs, and I have never yet run into the kind of problem that Anton refers to here. So I cheerfully agree to the agency's request for fuzzy discounts, make it sound like a concession, and try to get a higher base rate in return. But if I was presented with a TM full of the work of other translators... well, I don't do editing any more, precisely because I can't take reading bad translations.
[Edited at 2015-02-10 15:09 GMT]
| | | | | Fuzzy matches useless for Slavic languages | Feb 10, 2015 |
From my perspective, this is the typical workflow for any match lower than 95%: have a look at it, sigh, delete it, re-translate from scratch.
With languages where nouns and adjectives get 36 different suffixes, verbs get different suffixes depending on whether the statement is said by a woman or a man, and plurals differ based on whether you count 2-4 of relevant objects or more than 4, fuzzy matching is of no use whatsoever.
Why should I get paid less for the same am... See more From my perspective, this is the typical workflow for any match lower than 95%: have a look at it, sigh, delete it, re-translate from scratch.
With languages where nouns and adjectives get 36 different suffixes, verbs get different suffixes depending on whether the statement is said by a woman or a man, and plurals differ based on whether you count 2-4 of relevant objects or more than 4, fuzzy matching is of no use whatsoever.
Why should I get paid less for the same amount of work done? This might or might not be true for other languages - but applying the logic indiscriminately makes no sense.
[Edited at 2015-02-10 15:30 GMT]
[Edited at 2015-02-10 15:30 GMT] ▲ Collapse
| | | | | fuzzy matches showing differences? | Feb 11, 2015 |
Samuel,
Are there CATs that display the differences between the present source text and the fuzzy match source text? I have only ever worked with WFC, and I do not think this is possible with that tool. I own a WFP license, because it basically came with WFC (cost something like 50 EUR extra), but I have never actually got around to trying it out.
I always wondered why this function could not be provided, but just assumed that it involved some kind of insurmountable programmin... See more Samuel,
Are there CATs that display the differences between the present source text and the fuzzy match source text? I have only ever worked with WFC, and I do not think this is possible with that tool. I own a WFP license, because it basically came with WFC (cost something like 50 EUR extra), but I have never actually got around to trying it out.
I always wondered why this function could not be provided, but just assumed that it involved some kind of insurmountable programming difficulty that was beyond the grasp of my puny mind. ▲ Collapse
| | |
|
|
|
Samuel Murray
Netherlands
Local time: 21:52
Member (2006)
English to Afrikaans
+ ...
Michael Wetzel wrote:
Are there CATs that display the differences between the present source text and the fuzzy match source text? I have only ever worked with WFC, and I do not think this is possible with that tool.
Press Ctrl+Alt+M.
Also investigate the PB option ShowMemoryIf.
However, since about the middle of the version 5 range of WFC, it's buggy and unreliable.
I own a WFP license, because it basically came with WFC (cost something like 50 EUR extra), but I have never actually got around to trying it out.
I think WFP does, in a separate window located on the other side of the room, but I'm not sure, because I use WFP mainly as a TXML viewer.
I know that OmegaT has this function, and you can choose to have the differences shown either in colour or in pseudo-"tracked changes". I'm confident many other tools do it, too.
| | | | Richard Foulkes (X)
United Kingdom
Local time: 20:52
German to English
+ ...
| You mean like in SDL Studio Michael? | Feb 11, 2015 |
The concordance/match window shows the 'match' segments, highlighting parts of the 'match' that aren't in the source segment. Although, to my knowledge, it doesn't have a way of telling you which parts of the source segment are NOT in the match, which you have to be diligent about.
On the original topic, Dell saves 40% on its translation costs (which I imagine are substantial) using MT/post-editing according to this article, so I think it's fair to say it's a market reality.
<... See more The concordance/match window shows the 'match' segments, highlighting parts of the 'match' that aren't in the source segment. Although, to my knowledge, it doesn't have a way of telling you which parts of the source segment are NOT in the match, which you have to be diligent about.
On the original topic, Dell saves 40% on its translation costs (which I imagine are substantial) using MT/post-editing according to this article, so I think it's fair to say it's a market reality.
http://www.economist.com/blogs/prospero/2014/06/computer-aided-translation
Few translators 20 years ago would have foreseen the rise of online resources / CAT / MT so I think it's fair to assume the trend is going to continue / intensify as we head towards AI.
[Edited at 2015-02-11 11:08 GMT] ▲ Collapse
| | | | Dan Lucas
United Kingdom
Local time: 20:52
Member (2014)
Japanese to English
| AI was far more popular 30 years ago | Feb 11, 2015 |
Richard Foulkes wrote:
Few translators 20 years ago would have foreseen the rise of online resources / CAT / MT so I think it's fair to assume the trend is going to continue / intensify as we head towards AI.
Using machine intelligence for translation was a hot topic in the 1980s and, yes, there were sophisticated MT systems available back then too. There is nothing new under the sun...
Dan
| | | | | Sam, you're wonderful! | Feb 11, 2015 |
Sam, that's beautiful. I'll look into the Pandora's Box settings when I have time, but the shortcut works fine and made me very happy. I guess I'll get to know the bugginess as time goes on.
That's not enough to make me want to deal with figuring out a "real" CAT like WFP, because I basically never work with anything but word documents or the occasional PDF that I have to convert, but it's good to know that the function is probably waiting there for me.
I don't want to ... See more Sam, that's beautiful. I'll look into the Pandora's Box settings when I have time, but the shortcut works fine and made me very happy. I guess I'll get to know the bugginess as time goes on.
That's not enough to make me want to deal with figuring out a "real" CAT like WFP, because I basically never work with anything but word documents or the occasional PDF that I have to convert, but it's good to know that the function is probably waiting there for me.
I don't want to use OmegaT, because of the potential formatting issues going back and forth between Word and Open Office.
@Richard, like I just said, I don't have much use for a conventional CAT and I like staying within Word with WFC, but that's useful information to have posted here.
And I agree with you to the extent that I consider genuine MT (= devoted systems developed for specific contexts and involving major investments) important and believe that it will become more and more important in the immediate future. As I see it, the more an MT system functions like a TM and glossary the more relevant it is and the more it can serve a real purpose - pre-editing and controlled language are also key here.
However, I consider fantasy MT (= à la Google Translate, i.e., covering every field and text type and dealing with entirely uncontrolled input) completely irrelevant to our work.
I would think that patents contain a lot of material that could be very effectively dealt with using genuine MT and a lot of material where the best MT would be no real help at all.
@Dan Lucas
It goes back more than 30 years. Here is an interesting press release from IBM: https://www-03.ibm.com/ibm/history/exhibits/701/701_translator.html ... Now this deals with a different fundamentally flawed strategy (rule-based) than that pursued by Google (primarily stastics-based), but the - to my mind - absurd optimism and disconnection with reality are the same. ▲ Collapse
| | |
|
|
|
Richard Foulkes (X)
United Kingdom
Local time: 20:52
German to English
+ ...
| AI in the 80s? | Feb 11, 2015 |
Dan Lucas wrote: Richard Foulkes wrote:
Few translators 20 years ago would have foreseen the rise of online resources / CAT / MT so I think it's fair to assume the trend is going to continue / intensify as we head towards AI. Using machine intelligence for translation was a hot topic in the 1980s and, yes, there were sophisticated MT systems available back then too. There is nothing new under the sun... Dan
I'm no expert in the area Dan so, for clarification, I meant computers being able to fully simulate the human brain.
Ray Kurzweil, who has apparently been pretty accurate in predicting the exponential progress of technology, believes a $1,000 computer will have the full capacity of a human mind by 2020 and the ability to process the knowledge of all human minds combined by 2030. Who knows what that will mean for translation and translators?
| | | |
Richard Foulkes wrote: Dan Lucas wrote: Richard Foulkes wrote:
Few translators 20 years ago would have foreseen the rise of online resources / CAT / MT so I think it's fair to assume the trend is going to continue / intensify as we head towards AI. Using machine intelligence for translation was a hot topic in the 1980s and, yes, there were sophisticated MT systems available back then too. There is nothing new under the sun... Dan I'm no expert in the area Dan so, for clarification, I meant computers being able to fully simulate the human brain. Ray Kurzweil, who has apparently been pretty accurate in predicting the exponential progress of technology, believes a $1,000 computer will have the full capacity of a human mind by 2020 and the ability to process the knowledge of all human minds combined by 2030. Who knows what that will mean for translation and translators?
Is this about the time Skynet becomes self-aware?
| | | | Richard Foulkes (X)
United Kingdom
Local time: 20:52
German to English
+ ...
| Probably...let me have a think. | Feb 11, 2015 |
I'll be back .
| | | | | Pages in topic: [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-Processing of Machine Translations, the New Money Maker of “the Translation Industry” | 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! » |
| | 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
