MT: An Increasing Trend? Thread poster: Sam Berner
|
|---|
Sam Berner 
Australia
Local time: 01:00
Member (2003)
English to Arabic
+ ...
Hi all,
I am not into MT very much, but over the past few months I have noticed that I am increasingly given pieces for review that smack of serious machine translation (wrong grammar, style, use of terms, etc.) I was wondering if I am the only person with such an experience, or if my colleagues are also seeing an increase in such texts.
| | | | Tadzio (X)
English to Spanish
| Declined for obvious... | Mar 22, 2005 |
Hey Sam,
Fortunately, it's not a trend. You just happened to be there, that's all.
I once received such a request, and when I found out it was a machine translation the client wanted to pay as "proofreading", I just rejected the job saying that IT WAS a MT. If client wanted a good translation, he would have to first hire a professional, and next I would gladly do the proof job.
Some times we translators need to educate clients. Nothing to worry about. It happens from time to ti... See more Hey Sam,
Fortunately, it's not a trend. You just happened to be there, that's all.
I once received such a request, and when I found out it was a machine translation the client wanted to pay as "proofreading", I just rejected the job saying that IT WAS a MT. If client wanted a good translation, he would have to first hire a professional, and next I would gladly do the proof job.
Some times we translators need to educate clients. Nothing to worry about. It happens from time to time...
Kind regards from Mexico,
Tadzio. ▲ Collapse
| | | | mac9lxx (X)
Local time: 21:00
English to Bosnian
+ ...
| I fully agree | Mar 22, 2005 |
I fully agree with Tadzio, it is not a trend, however some companies are trying to cut costs and are trying to pawn off MTs as ITs.
| | | | 00000000 (X)
English to French
+ ...
| Enough of a trend to start worrying | Mar 22, 2005 |
There are two different types of MT problem appearing. The first is not too worrisome and can be easily resolved by educating the client. The second one is more nefarious.
I've had two cases of the first type in the last two years or so. One day a regular client for whom I do small translations paid by the hour because they involve creative adaptation sends me text to "edit" instead of text to translate. And it's easy to see that the translation was done either by MT or by some seco... See more There are two different types of MT problem appearing. The first is not too worrisome and can be easily resolved by educating the client. The second one is more nefarious.
I've had two cases of the first type in the last two years or so. One day a regular client for whom I do small translations paid by the hour because they involve creative adaptation sends me text to "edit" instead of text to translate. And it's easy to see that the translation was done either by MT or by some second-language beginner. Being paid by the hour, it's easy to make them understand that it's harder for me to try to correct a bad translation than simply translating from scratch -- all the more so since I have all their previous stuff in my translation memories and I honestly bill by the time it actually takes me.
The second type involves multi-million dollar contracts. One of my main sources of income recently vanished when the agency for which I've been working on certain manuals for years lost the contracts to a company that offered a much lower price based on post-editing the output of machine translation.
At the same time, now having more time on my hands, I responded to an ad for translators in the same field. I took the translation tests and gave my rates. My tests were fine and, by and large, my rates were accepted.
All was fine, you'd think. Well, except for one thing. There was one extra line in the rate schedule that raised a red flag: a rate for post-machine editing (a little less than half the rate for translating new words).
I cheerfully responded that everything was fine but please note that I don't do post-machine editing. I haven't heard from them since and don't really expect to.
Esther
[Edited at 2005-03-22 06:17] ▲ Collapse
| | |
|
|
|
Marc P (X)
Local time: 17:00
German to English
+ ...
| Machine Translation (not) | Mar 22, 2005 |
It might help if we reserved the term "translation" for processes which involve interpretation of meaning. This might go some way to dispelling the misconception that a machine translation application works in principle in the same way as a human translator. Notwithstanding the advanced grammatical parsing of some MT systems, this is not the case.
Personally, I wouldn't call raw MT output a "translation".
MT = machine transliteration?
Marc
| | | | RobinB
United States
Local time: 10:00
German to English
smack of serious machine translation (wrong grammar, style, use of terms, etc.)
Forgive me if I sound cynical, but that seems awfully like human translation to me. MT can certainly get the grammar wrong, but when it does so, it tends to get it consistently wrong, i.e. you can recognise a pattern. But when human translators get the grammar wrong (assuming they know what grammar is in the first place), it tends to be wrong all over the place. At least MT tends to have grammar hard-wired into the system.
And as far as bad style and incorrect terminology are concerned, I've yet to see an MT system that can beat a really bad human translator.
(sigh)
Robin
| | | | | Although some would like it to be a trend, not yet | Mar 22, 2005 |
There is one member of this site who seems to show a great deal of enthusiasm for MT. You may have seen one of his recent postings: http://www.proz.com/topic/30577
His claim from doing studies is that translators can work more efficiently by "editing" raw MT output, in comparison to translating from scratch.
Although one may be able to work faster by editing MT output, I personal... See more There is one member of this site who seems to show a great deal of enthusiasm for MT. You may have seen one of his recent postings: http://www.proz.com/topic/30577
His claim from doing studies is that translators can work more efficiently by "editing" raw MT output, in comparison to translating from scratch.
Although one may be able to work faster by editing MT output, I personally don't need the stress of having to re-do sentence after sentence. Proofreading a poorly done translation is enough stress already - I can't imagine what it would be like having to proof a machine-read translation.
Fortunately, there are very few agencies that take MT seriously. Occasionally I have seen postings from an agency or two offering editing jobs for MT, but they are few and far between.
Russell ▲ Collapse
| | | | dgmaga (X)
English to Spanish
| Yes, it's a trend and it will grow | Mar 22, 2005 |
I was wondering if I am the only person with such an experience, or if my colleagues are also seeing an increase in such texts.
Hi,
Yes, there is a trend to use MT more and more and it will grow, as computers get more powerful and the MT systems get more sophisticated.
Nevertheless, the idea of "machine translation" covers too many different situations, from the 20 euro package which covers 12 languages mentioned in another post to high-end server systems.
The combination of machine translation plus human editing is already more coste-effective and provides the same quality as human-only translation for some language combinations, in certain subject fields and for a certain types of documents.
Automative companies have used a combination of controlled language (limited grammar rules and terminology) and machine translation plus human editing successfully for some years now.
Of course, using machine translations without proper up-front preparation or in language combinations which are not so well supported by MT engines is counterproductive.
Is it a thread to translators? I don't think so. I just think that there will be more editing jobs but, at the same time, an increase in productivity will mean that globally more translations will be produced and there will be more work. I think the effect will be similar as that of CAT tools. Maybe the price per unit (word or line) will decrease but the increase productivity should hopefully compensate for that.
Daniel
| | |
|
|
|
Aleksandr Okunev (X)
Local time: 18:00
English to Russian
| The increasing trend. | Mar 22, 2005 |
Yes, there is a trend to use MT more and more and it will grow, as computers get more powerful and the MT systems get more sophisticated.
Power of computers does not necessarily mean greater sophistication of machine translation software. In essence all MT tools are going forever remain stupid brutes, just because of the differences mentioned above. Their presence, similarly to presence of CAT tools creates an illusion that a translator can be reduced to a sweatshop worker, churning out words like hot dogs or khaki underpants. No, it is not possible, and will never be.
Another thing is that the trend at the market is set by the most aggressive, "technically advanced" and knowledgeable agencies who get carried away and fail to see that by offering those fuzzy match discounts and things like that, including MT they are killing the goose that lays the golden eggs, when they are trying to turn translators into labourers. The ace up their sleeve is the technology, lots of empty words about algorithms, approaches, databases and cool stuff. What they fail to notice is that even now, behind every translation there is man, and it has been like this for several thousand years. The reports of new successful versions are just a smokescreen aimed at attracting more customers.
Here's my guess of what is going to happen. Things will be going worse. Thanks to Internet, CAT and MT there will be non-native translations and machine translations going public, and decent folks will drift away from the translation business. The rates will continue to fall until one day, some of those happy end clients will realize that they have hundreds of gigabytes of bullsh*t in their databases, and there's not a single decent human translator around, because MT and attitudes have driven them all out, to businesses which experienced revolutionary improvements some 30-2000 years before and had time to settle after that.
I remember a time when it cost me a pack of Marlboro filters to get from Sheremetievo-2 airport to the southern edge of Moscow (that's about 25 km to reach the town and drive through the whole of it, North to South). Despite the fact that I had a huge supply of Marlboro packs and beloved greenbacks, I was not completely happy, because the situation was not normal at all. It resulted in much turmoil, then a very slow recovery and now a taxi in Moscow is available for a competitive but realistic price. Most of The trend-setting folks are happy with the changes taking place, because they still have more Marlboro in the briefcase and do not see too far and do not understand, that taxis will soon get increasingly scarce and bad if nothing changes.
Against stupidity gods themselves fight in vain. Throw in greed and blindness caused by the so-called technology achievements and you get the energy of several atomic bombs. The trend is here all right and I don't know where it's going to reverse.
Stay well
Alex
P.S. It has just occurred to me that neither Einstein nor Shakespeare could send SMS. Morons!
| | | | Sam Berner
Australia
Local time: 01:00
Member (2003)
English to Arabic
+ ...
TOPIC STARTER | Efficiency is relative, Russell | Mar 22, 2005 |
Russell Gillis wrote:
His claim from doing studies is that translators can work more efficiently by "editing" raw MT output, in comparison to translating from scratch.
Russell
It all depends what you are translating. We do monthly brochures for a local council, and over the past three years have aquired a sizeable glossary of their terms. Last month we tried using MT and although it limped a tat, a few strokes of lipstick made the monster a lassy It should be remembered, however, that this is ultra simple text we are discussing. I would not attempt it on, say, an exposition about philosophy.
| | | | Sam Berner
Australia
Local time: 01:00
Member (2003)
English to Arabic
+ ...
TOPIC STARTER | Bad Translators vs MT | Mar 22, 2005 |
RobinB wrote:
And as far as bad style and incorrect terminology are concerned, I've yet to see an MT system that can beat a really bad human translator.
(sigh)
Robin
I have had that experience while proofreading, too. Not to mention that the translator was a "native-speaker" who advertised him/herself as caring for the clients image, and then produced a piece of marketing translation that reminded me of a washing line after a windy day But then, I strongly suspect that (s)he bid for the job on very low price, whacked the text into a CAT tool, spawned it out without re-reading and sent it off to the poor antipodians thinking, "Hell, they don't have anyone to review it for them." So again, it was MT plus an unscrupulous tranlator.
| | | | Jeff Allen
France
Local time: 17:00
Multiplelanguages
+ ...
| more than just enthusiasm for MT | Mar 23, 2005 |
Russell Gillis wrote: There is one member of this site who seems to show a great deal of enthusiasm for MT. You may have seen one of his recent postings: http://www.proz.com/topic/30577
Russell,
Thanks for your pointer to my recent post. It's one of many on Proz.
For example:
Machine Translation: yes used for professional translation work
http://www.proz.com/post/177427#177427
MT: how much time to refine
http://www.proz.com/post/177428#177428
The easiest way to access all of my posts on Proz with practical info about MT is to do a search on the following word: mtpostediting
All of my MT postings will display, each with the relevant topic title.
Russell Gillis wrote:
His claim from doing studies is that translators can work more efficiently by "editing" raw MT output, in comparison to translating from scratch.
I've reproduced the same high level of productivity several times:
* in several domains
* on several types of texts (marketing brochures, automotive, heavy-machinery, legal texts, telecom software acceptance customer test plan, etc)
* on real-production texts in real production mode (not academic research mode)
And, the methodology has been reproduced by another translator for another language pair.
GUERRA, Lorena. 2003. Human Translation versus Machine Translation and Full Post-Editing of Raw Machine Translation Output. Master's Thesis. Dublin City University.
http://www.geocities.com/mtpostediting/lorena-guerra-masters.pdf
and I'm about ready to train another professional translator on still another language direction to do the same. Over the past year, he took the time to read all of my posts on several translation discussion lists + all of my articles, presentations, and papers produced on this topic over the past 10 years. He came to me and asked how I could help him produce the same results.
Russell Gillis wrote:
Although one may be able to work faster by editing MT output, I personally don't need the stress of having to re-do sentence after sentence. Proofreading a poorly done translation is enough stress already - I can't imagine what it would be like having to proof a machine-read translation.
But if you do the dictionaries in the right way and very efficiently from the beginning, then there is little clean-up aftwards. Just minor post-editing changes here and there. That's all shown in my AMTA2004 paper that is cited with a URL in my Proz postings.
Just today I finished the full analysis and extraction of nearly 1000 complete terms from large sets of newsline texts for a project that I'm working on. All initial steps done in 7 logged hours (working on a Pocket PC during train and subway/metro daily commuting rides to the office) I am about ready to import these terms into a custom MT dictionary. These were manually analyzed and extracted (and all time spent was carefully logged) to provide baseline statistics on productivity.
Now I'm going through and sifting out and sorting the entries to create a minimal size dictionary that will provide maximum coverage of well-translated terms of both the analyzed texts + other non-analyzed sample texts. That will take me another 1-2 hours. Then enter in the translations will take me about another 3-5 hours. Then test and retest and adapt dictionary entries. Another 3-5 hours.
And this for approximately 6-8 years of newsline texts produced every 2-3 weeks with approximately 10 entries of 100 words each. Just do the calculation of the figures and you'll see that this is a lot of text. And it's newsline marketing materials, not technical manuals. So it's supposed the impossible stuff to apply MT to. We'll, I'll be proving that wrong in about a week with another MT case study article that provides all the statistics (timelogs, number of entries created, amount of text processed, etc).
So estimate of 15-20 hours to create custom dictionary, test the dictionary on all text test sets, review and improve the dictionary, and then postedit clean-up the translations.
All this to do about 1/6 of about 100,000 words.
Even if I add another 10 hours of time to my estimate to cover all the formatting steps and postediting stages. And I can also add an estimated 25% time error factor (in case I underestimate the workload) and come to about 40 hours of time for the project. If it is possible to produce 20,000 words of high-quality translated text in that amount of time with a tool, and it is possible to produce such an amount on an ongoing basis in a non-stressed out mode production setting, then it seems really worth the upfront investment.
You cannot compare translations run through an free online MT portal -- that has just a general dictionary and no means to connect a custom dictionary -- to a Professional or Expert MT software package with well-designed dictionary manager modules. I use 3 different software package brands to do such translation projects, and I aim at creating techniques that can be applied to any of the software packages.
Just like MS Excel or MS Project or Quark Xpress, etc, if you don't master the software, then it will just slow you down in your tasks and waste time. And sometimes other software packages will do your task for you more efficiently (such as MS Monday or Intuit Quicken). But if you really master MS Excel and know how to do multi-axis tables and complex graphs, and integrated publi-posting with MS Word, then it can save you tons of time on many tasks.
That's what I've managed to do with several MT software packages.
Jeff
http://www.geocities.com/mtpostediting/
| | |
|
|
|
Jeff Allen
France
Local time: 17:00
Multiplelanguages
+ ...
| upfront work needed for good MT results | Mar 23, 2005 |
dgmaga wrote:
Automative companies have used a combination of controlled language (limited grammar rules and terminology) and machine translation plus human editing successfully for some years now.
Of course, using machine translations without proper up-front preparation or in language combinations which are not so well supported by MT engines is counterproductive.
Is it a thread to translators? I don't think so. I just think that there will be more editing jobs but, at the same time, an increase in productivity will mean that globally more translations will be produced and there will be more work. I think the effect will be similar as that of CAT tools. Maybe the price per unit (word or line) will decrease but the increase productivity should hopefully compensate for that.
Yes, I was involved in evaluating the software interface interactive process with writers and translators and evaluating the writing rules and postediting rules for the General Motors project several years ago. I was also the trainer/mentor of controlled language writing, translation tools and postediting at Caterpillar for all their technical writers and translators. As for limited sets of grammar rules and terminology, this varies from system to system. GM had 62 writing rules. Caterpillar had 140. Caterpillar cleaned up their English terminology from 1 Million down to 50,000 technical terms at beginning of the projet. Then progressively built up the terminology database up to 70,000 source terms during the first 4 years of the project. And it kept growing with time.
And I fully agree with you on upfront work. All tools require upfront work. Let's just look at all the forums dedicated on Proz just to CAT tool support. There are literally hundreds of postings just about pre- and post-formatting issues, how to create special filters, or how to fix alignment issues. The same is true of the business intelligence performance reporting software that I am involved with at my day job. It requires upfront work, yet when it is done correctly, then good quality at high-volume makes it worth the investment.
As for a threat to translators, I've always said it is the opposite. It is an opportunity for professional translators to use tools that can help them to be more productive. The main issue is that no MT software or systems comes with a really excellent task-focused user guide and tutorial. So, this makes it hard for any professional translator to become really productive with the tools.
Jeff
http://www.geocities.com/mtpostediting
| | | | Jeff Allen
France
Local time: 17:00
Multiplelanguages
+ ...
| MT tools: as intelligent as their programmers & users | Mar 23, 2005 |
Yes, there is a trend to use MT more and more and it will grow, as computers get more powerful and the MT systems get more sophisticated.
Aleksandr Okunev wrote:
Power of computers does not necessarily mean greater sophistication of machine translation software. In essence all MT tools are going forever remain stupid brutes, just because of the differences mentioned above. Their presence, similarly to presence of CAT tools creates an illusion that a translator can be reduced to a sweatshop worker, churning out words like hot dogs or khaki underpants. No, it is not possible, and will never be.
Sorry, I don't agree about MT tools remaining stupid. The tools are as intelligent as those who program them and those who know how to master them by way of dictionary entries, additional entered custom rules, and the use of the integrated Postediting features.
Just like some people write macros in Excel to improve and speed up their use of the tool, I do the same with several MT software programs by way of using the dictionary modules.
I also helped design an automated postediting module (my co-authored CLAW2000 paper) on automatically cleaning up inappropriate translations.
The problem with productivity with many software tools (any field) today is usually not simply a problem of the software, but often that there lacks a decent training methodology available to show users how to really master the use of the tool to its full potential and to the user's greatest benefit.
A year or so ago I even introduced a new concept in the dictionary making process to include the use of alternate stylistic choices. This was a jump forward in allowing translation post-editors to integrate stylistic variation into their postedited texts.
Jeff
http://www.geocities.com/mtpostediting
[Edited at 2005-06-26 07:53]
| | | | Aleksandr Okunev (X)
Local time: 18:00
English to Russian
| Any CAT and MT can beat human translator, BUT 8) | Mar 23, 2005 |
RobinB wrote: And as far as bad style and incorrect terminology are concerned, I've yet to see an MT system that can beat a really bad human translator.
I have ProMT XT which is a nice piece of software, I use it only as terminology server, because it contains around 20 huge dictionaries, a lot of my custom glossaries and I can change their numbers and priority.
I did experiment a lot with ProMT and I can say that I can teach it to translate literary prose nearly well, but the amount on effort required to acieve that outweighs the benefits by imeasurable factor.
You can get 100% matches from CAT, can't you? All you need to do is to put all translations of all phrases in all the languages into the database and live happily ever after. It will take you several thousand years to do it, and it will be all in vain once teenagers round the corner invent a slang word or phrase.
CAT and MT are good *tools* wise are the developers who took this direction, poor are the ones who have entered the blind alley and try to replace human translator with a computer.
Stay well
Aleksandr
| | | | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » MT: An Increasing Trend? | Trados Studio 2022 Freelance | The leading translation software used by over 270,000 translators.
Designed with your feedback in mind, Trados Studio 2022 delivers an unrivalled, powerful desktop
and cloud solution, empowering you to work in the most efficient and cost-effective way.
More info » |
| | CafeTran Espresso | You've never met a CAT tool this clever!
Translate faster & easier, using a sophisticated CAT tool built by a translator / developer.
Accept jobs from clients who use Trados, MemoQ, Wordfast & major CAT tools.
Download and start using CafeTran Espresso -- for free
Buy now! » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
