| Pages in topic: < [1 2 3 4] > | Is this the future? Automatic simultaneous translation within 5 years? Thread poster: Barnaby Capel-Dunn
|
|---|
| Thanks, Jeff, for the interesting summary | Oct 14, 2005 |
Jeff Allen wrote:
Hi Victor,
Sure, let me explain a few points.
...
Jeff
Thanks for this very interesting and informative summary and the associated links. I have only had time for a cursory scan of some of the links, but they certainly provide food for thought.
As far as I can see, the projects outlined in the links tend to be very focussed on the role of the expert guiding the individual project, and the roles of pre-processing and post-editing. To some extent, controlled source language also seems to be an issue.
This thread started with a rather different idea, i.e. a quote summarised from Bill Gates: "... Microsoft's founder thinks that within 5 years, the keyboard-voice-wiring on screen combination should revolutionise our way of working,in the same way as SIMULTANEOUS TRANSLATION WHICH SHOULD COME INTO EFFECT AT THIS TIME".
You are at the cutting edge of MT. Do you regard such visions (which are by no means new) as hopelessly simplistic, or do you feel it is simply a matter of time?
There are of course many other issues involved, such as technical intelligence issues (cf. my rather tongue-in-cheek "typo" posting earlier in this thread) and social and professional issues (are we a dying profession after all, or will we simply be downgraded to half-price post-editors?). And of course the wider questions of the relative roles of computers and humans in our future world (and a new twist in the debate on the question "What are humans here for anyway?").
A fascinating subject, and thanks for bringing your expertise into the debate.
| | | | Jeff Allen 
France
Local time: 06:01
Multiplelanguages
+ ...
| Controlled Language, Speech translation, etc | Oct 14, 2005 |
To use Controlled language or not?:
Victor Dewsbery wrote:
As far as I can see, the projects outlined in the links tend to be very focussed on the role of the expert guiding the individual project, and the roles of pre-processing and post-editing. To some extent, controlled source language also seems to be an issue.
All "controlled language only" and "controlled language + MT" projets are listed in one of my conference talks last year:
Introductory overview of Controlled Languages. Invited talk presented at the Society for Technical Communication meeting of the Paris, France chapter. 2 April 2004.
http://www.geocities.com/controlledlanguage/Allen-Barthe-CL-intro-STC-France-v1.01-2Apr2004.ppt
Multilingual Controlled Languages. Invited talk presented at the Society for Technical Communication meeting of the Paris, France chapter. 2 April 2004.
http://www.geocities.com/controlledlanguage/Allen-multiling-CL-STC-Fr-v1.00-2apr2004.ppt
Conference keynote talk on Controlled Language and Machine Translation, May 5-6, 2004
http://www.geocities.com/jeffallenpubs/Allen-besancon-6may2004_v1.01.ppt
My projects over the past few years on MT dictionary development + postediting do not include controlled language (CL). I've been demonstrating that MT can be productive without CL. When any type of input source language standardization efforts (writing guidelines, spell-checking, CL, terminology standardization, etc) are performed, this will tend to favorably impact the translation process. However, I will cite one of my former colleagues, the late Eric Adolphson, who stated at the CLAW98 conference
that the problem with CL is that it is possible to write in good CL and in bad CL, depending on how well people are trained, and if a CL editing process is conducted. So, yes, it is possible for authors to write in bad CL and negatively impact the target translations. Yet, we see this already in writing/translation workflows that do not including CL.
Simultaneous speech translation:
Victor Dewsbery wrote: This thread started with a rather different idea, i.e. a quote summarised from Bill Gates: "... Microsoft's founder thinks that within 5 years, the keyboard-voice-wiring on screen combination should revolutionise our way of working,in the same way as SIMULTANEOUS TRANSLATION WHICH SHOULD COME INTO EFFECT AT THIS TIME".
You are at the cutting edge of MT. Do you regard such visions (which are by no means new) as hopelessly simplistic, or do you feel it is simply a matter of time?
My background in fact includes speech-enabled translation with both speech recognition and speech synthesis modules. I organized, conducted and supervised the speech data collection projects for several hundred participants for the EngHaitian Creole and EngKorean speech-to-speech translation projects at Carnegie Mellon University (CMU) (http://www.lti.cs.cmu.edu/Research/Diplomat/). I was also Technical Director of speech system language resources for a couple of years at ELDA around the year 2000. Several of my articles on speech translation are at http://www.geocities.com/jeffallenpubs/speechtech.htm
At CMU, we developed complete birectional speech-to-speech translation systems for several language pairs (different language typologies) and even provided conference papers/posters which indicated comparative charts on the timeframe needed to get the different systems up and running. At that time (1997-1998), we were creating what I would refer to as "interactive bilingual speech/text translation" systems. These systems were successfully demo'ed to sponsors and were used at least for periods of time in real-user contexts for which they were designed (basically military checkpoints and humanitarian organization efforts in war zones).
The current thread that started with the Bill Gates quote seems to be a different level, and I interpret this quote to mean: "Full-scale simulataneous speech translation will be ready in about 5 years"
Do I believe that? No.
Yet I do believe that in another 5 years there will be more improved speech-enabled and speech-interactive systems that could be combined with text-based translation systems, and which could be used in various contexts and probably for industrial and business purposes. These will probably be in limited domain and limited vocabulary/terminology areas. For example, weather forecasts are written in more or less defined sublanguage in many languages. Providing speech translation for announcing weather forecasts at the Olympics might be a likely application. Another might be announcing arrival and departure times and information in multiple languages at major transport hubs (airports, train stations).
But such speech translation systems, in my opinion, would not be used in the near future for phone-conversation simultaneous interpretation or conference simultaneous interpretation.
Victor Dewsbery wrote:
There are of course many other issues involved, such as technical intelligence issues (cf. my rather tongue-in-cheek "typo" posting earlier in this thread) and social and professional issues (are we a dying profession after all, or will we simply be downgraded to half-price post-editors?). And of course the wider questions of the relative roles of computers and humans in our future world (and a new twist in the debate on the question "What are humans here for anyway?").
The typo example is the never-ending "garbage-in/garbage out (GIGO)" debate. And the simple fact is that there is always input garbage. People simply don't use spell-checkers because half the time they provide more noise than necessary. Some of the new MT systems are able to correctly recognize words that do not have accent marks. Even at CMU, a colleague and I developed an accent mark reinsertion module. This was because most of the electronic texts that we were using had been stripped of accent marks 10-15 years ago when issues in text conversion and display which caused the hieroglyphics E10, E11 and other characters for language that had accent marks.
Humans are truly the added-value:
My perspective as stated in conferences and articles has been that computers are binary-restricted, number and code crunching machines that are only as good as what you program them to do. Our computer experts love to try and resolve the 25th century Star Trek language barrier issues. Yet, computers are very good at recognizing and working with repetitive things. And many humans hate to redo the same operation 25 million times in a single day, unless they already see themselves as a low-value player in the system and don't care about when processes can be improved and optimized. So why not use computer-based tools in an optimal way to deal with the mid and high-frequency full-match and partial match repetitive sequences, and for content gisting of what you don't understand, and allow the human translators the opportunity to fill the added-value roles as pre-processors, key processors, and post-processors within the translation workflow.
Pure computer programmers do not necessarily agree with me, but then the majority of them do not have many years of experience in human-based translation that I had before I joined the ranks of computational linguistics and natural language processing.
Jeff
[Edited at 2005-10-15 23:48]
[Edited at 2006-01-31 22:51]
| | | | Jeff Whittaker
United States
Local time: 00:01
Spanish to English
+ ...
| Even professionals don't always agree | Oct 15, 2005 |
If professional translators disagree on the correct translation of a word, term or phrase (see KudoZ for numerous examples), then how will a computer (or the person who programs the computer) ever "decide" which translation to use?
| | | | | That's true but.... | Oct 15, 2005 |
Jeff Whittaker wrote:
If professional translators disagree on the correct translation of a word, term or phrase (see KudoZ for numerous examples), then how will a computer (or the person who programs the computer) ever "decide" which translation to use?
at a professional level, isn't it also the case that a) many (most?) of us handle the same sort of text over and over again, and b) (this is a terrible thing to say but nonetheless true) in many cases the quality, as opposed to the accuracy, of the translation is not of primary importance? If it IS true, then surely we can expect "machines" to play an increasing role in the years ahead?
| | |
|
|
|
Jeff Whittaker
United States
Local time: 00:01
Spanish to English
+ ...
How can you provide "quality" without "accuracy"?
Barnaby Capel-Dunn wrote:
at a professional level, isn't it also the case that a) many (most?) of us handle the same sort of text over and over again, and b) (this is a terrible thing to say but nonetheless true) in many cases the quality, as opposed to the accuracy, of the translation is not of primary importance? If it IS true, then surely we can expect "machines" to play an increasing role in the years ahead?
| | | |
Jeff,
When I read your message I happened to be looking at a Microsoft text about Word Viewer (in French).
I ran an automatic translation and got this result:
"Word Viewer enables you to open documents Word 2003 and documents created with all the preceding versions of Microsoft Word for Windows® and Microsoft Word for Macintosh. In addition to the files of document Word (doc), you can open the files recorded with the following formats:...."
Not perf... See more Jeff,
When I read your message I happened to be looking at a Microsoft text about Word Viewer (in French).
I ran an automatic translation and got this result:
"Word Viewer enables you to open documents Word 2003 and documents created with all the preceding versions of Microsoft Word for Windows® and Microsoft Word for Macintosh. In addition to the files of document Word (doc), you can open the files recorded with the following formats:...."
Not perfect by any means but pretty accurate. Compare with Microsoft's own original English text:
"Word Viewer 2003 lets you open Word 2003 documents and documents created with all previous versions of Microsoft Word for Windows® and Microsoft Word for Macintosh. In addition to Word document files (.doc), you can also open files saved in the following formats:..."
Obviously, no self-respecting translator would be happy with the first version (but don't forget this is MT and not CAT) but surely it is accurate enough for many situations? ▲ Collapse
| | | | Jeff Allen
France
Local time: 06:01
Multiplelanguages
+ ...
| alternative translation feature in MT packages | Oct 16, 2005 |
Jeff Whittaker wrote:
If professional translators disagree on the correct translation of a word, term or phrase (see KudoZ for numerous examples), then how will a computer (or the person who programs the computer) ever "decide" which translation to use?
The alternative translation option feature and the hide/show alternative translation feature in the "professional level versions" of MT software allow you to code and then use the options during the translation process. I've described in the PROMT, Reverso and SYSTRAN sections of my MT tips site (http://www.geocities.com/jeffallenpubs/MT-tips.htm) how such features are available in those software packages.
Jeff
http://www.geocities.com/jeffallenpubs/about-jeffallen-htm
| | | | | Can of worms | Oct 17, 2005 |
This whole debate brings up a number of issues.
1. Redefining the idea of the "native speaker"
An old chestnut in the trade: should we translate into or from our native language?
How good must we be in our foreign language before we translate into it on a professional basis?
Is it good enough to translate into our foreign language and then get a native speaker to proofread the result and repair the damage?
However, with MT, the concept of the "native speaker"... See more This whole debate brings up a number of issues.
1. Redefining the idea of the "native speaker"
An old chestnut in the trade: should we translate into or from our native language?
How good must we be in our foreign language before we translate into it on a professional basis?
Is it good enough to translate into our foreign language and then get a native speaker to proofread the result and repair the damage?
However, with MT, the concept of the "native speaker" loses much of its force. Or the process become similar to drafting by a non-native and proofreading (sorry, post-editing) by a native.
2. The expert and the quack (and perhaps the pragmatist)
The systems that Jeff presents seem to be very much at the high end - i.e. heavily dependent on hands-on work by an expert linguist for pre-processing and post-processing, and with a hefty emphasis on "teaching" the system as you go along (and the associated need for a computer expert to oversee or carry out the process). This is probably only viable for the "global big players" in business communication, because the investment in equipment, in continuity of (human) support expertise and the long-term gains of "teaching" the system are likely to be unviable for anyone else.
The quack is the computer nerd who thinks that you can feed in any old language and the computer will do its stuff, irrespective of context and without any pre-processing and post-editing. Long live Babelfish, because without it, a comical literary genre (the hopeless mistranslation) would vanish.
The pragmatist would probably be somewhere in between, someone like us who uses an off-the-peg MT program occasionally but realises that you can't rely on the product, or can only do so with a big pinch of salt for a very limited range of uses. Or in another variant, the translator using a CAT tool mainly to increase productivity.
3. The "little-old-me" factor
What will become of my earning power? Will the development force me to stop translating and move over to post-editing at half my present hourly rate?
This is not a new concern and not restricted to translating (it hit manufacturing trades long before it came to us).
This can of course be compounded into a whole range of economic polarisation theories (all the world economy needs is a handful of global business owners and a few million cheap labourers, and we fall into the latter category. Everybody is involved in the wealth-creation process, but nobody ever gets to see any wealth.)
In other words, just a compounded variant of the price squeeze that some agencies and other clients already try to employ without MT (similar to the price squeeze on suppliers which enables your friendly corner supermarket to offer you rock-bottom prices).
4. Translation - repetitive task or creative work?
In his long reply a couple of days ago, Jeff wrote "computers are very good at recognizing and working with repetitive things". Does that describe our work? To put it another way, if we process a sentence with a CAT tool or MT tool, do we sometimes feel that it is best to scrap the suggested translation and start all over again?
In fact, this concept comes up in the thesis by Lorena Guerra which Jeff referred us to at:
http://www.geocities.com/mtpostediting/lorena-guerra-master s.pdf
She writes that MT post-editors must actually be trained not to do a complete re-write of the MT translation, but only to edit it as far as is really necessary.
This is perhaps an important question for the ongoing discussion - is language creative or repetitive? There are obviously elements of both, but what is predominant in the work we do?
Oh well, must get some work (!?!?!) done. Perhaps someone will come up with interesting comments to push the discussion forward. ▲ Collapse
| | |
|
|
|
| I am not worried | Oct 18, 2005 |
I am not worried aobut this, simply because I think that it is another one of many attempts to oversimplify something that is very complex. Translation and interpretation both require not only the ability to speak two or more languages, but they also require a whole array of senses, for example, compassion, the ability to adjust to constantly changing moods in a conversation, the ability to put yourself in the other person's shoes, the ability to think with and ahead of the person you are transl... See more I am not worried aobut this, simply because I think that it is another one of many attempts to oversimplify something that is very complex. Translation and interpretation both require not only the ability to speak two or more languages, but they also require a whole array of senses, for example, compassion, the ability to adjust to constantly changing moods in a conversation, the ability to put yourself in the other person's shoes, the ability to think with and ahead of the person you are translating or interpreting for and many pther things that a machine simply cannot and will never be able to do with the accuracy that a human can. Sure attempts can be made to emulate these abilities in a machine, but the machine will be limited to whatever is programmed into it. As for voice recognition, if they have not been able to fine-tune voice recognition in cell phones, (and that has been around for a while now), to the voice changes of the humans using this technology, I wonder if theya are far enough in their research to be able to even apporach the subject at a much greater scale. To give an example of the problems I have encountered with voice recognition on cell phones. I have tried using it, say for dialing my home number (I say "home")...but if I say it in a tone that is not the same as the tone I used when I recorded the word on the phone, it simply does not dial. I can't have a sore throat, or am in a crowded and loud place because it will not work. So I think there is still a lot of work to be done, if they do plan to comeout with something like that. Besides, if they do "revolutionize the market" with that type of technology, it will probably have a peak in sales at first, but then people will realize that there are indeed things that cannot substitute "human" translation or interpretation.
anyway, I hope I have not bored you with all this and I hope it makes sense. These are just my thoughts on the whole thing.
Greetings everyone,
alexandra123 ▲ Collapse
| | | | Angus Woo
Local time: 12:01
Chinese to English
+ ...
| No matter what, a computer is not a human, at least not for the next ten years. | Oct 19, 2005 |
The question in fact is age-old: Can a computer think like a human brain? Make decisions the way we do?
The answer can only be too obvious. The so called artificial intelligence technology so far has been miles from satisfactory. And though it is still painstakingly inching forward, nothing fundamental is likely to happen in the foreseeable future.
If, just if, a computer can indeed think like a human brain, then the issue will, with mathematical certainty, concern not ... See more The question in fact is age-old: Can a computer think like a human brain? Make decisions the way we do?
The answer can only be too obvious. The so called artificial intelligence technology so far has been miles from satisfactory. And though it is still painstakingly inching forward, nothing fundamental is likely to happen in the foreseeable future.
If, just if, a computer can indeed think like a human brain, then the issue will, with mathematical certainty, concern not only translators and interpreters but the entire human race.
Our lives will no longer be the same ever. The world would become something we can't dream about even in our wildest dreams.
Think how scary that can be.
[Edited at 2005-10-19 04:01] ▲ Collapse
| | | | | We won't see machine translations in the foreseable future | Oct 20, 2005 |
It is utterly unrealistic to think that we will see software that can translate correctly in the foreseable future. It will take more than one or two geniuses to solve the linguistic code that governs human speech - and certainly not Microsoft, who are good at one thing: operating systems for machines, which are slightly less complicated, I fear,than the human brain.
Voila
| | | | Jeff Allen
France
Local time: 06:01
Multiplelanguages
+ ...
| expert linguists not needed for using MT systems | Oct 24, 2005 |
Victor Dewsbery wrote:
2. The expert and the quack (and perhaps the pragmatist)
The systems that Jeff presents seem to be very much at the high end - i.e. heavily dependent on hands-on work by an expert linguist for pre-processing and post-processing, and with a hefty emphasis on "teaching" the system as you go along (and the associated need for a computer expert to oversee or carry out the process). This is probably only viable for the "global big players" in business communication, because the investment in equipment, in continuity of (human) support expertise and the long-term gains of "teaching" the system are likely to be unviable for anyone else.
Expert Linguists:
There is not really a need for an expert linguist. In July 2003, I trained 2 technical writers and 1 Human Resource Assistant in the same company (all together in the same room) to use a professional/expert version of a commercially available MT tool. Within 3 hours they were all creating customized dictionaries based on their own documents that had been used to prepare the course exercises. None of them are expert linguists. One has a degree in translation, another one a diploma in technical writing.
A couple of months before I gave a 1-2 hour training session to a bilingual secretary.
Computer Experts:
Also, no need for computer geeks/nerds/experts to monitor the process for PC and Mac workstation based systems. Several of the modern versions of MT systems have user-friendly interfaces in the dictionary editor modules and the translation environement modules.
About 10 years ago, I did work on a system that was much more linguistically complex and which required knowing and using more linguistic meta-language to use the system. It took me much time to develop training courses that simplified the information presented by that system to the level of understanding of linguistics of the user audience. Yet, it was a customized system. Most of the commercial MT products on the market today aim at the level of non-linguist users (who use MT for business and personal correspondence), which requires providing more user-friendly words and terms in the interface that will be understandable to the users.
Equipment:
As for equipment, it varies from MT program to program. Yet I still do test all of them on a Windows 98 First Edition with 64Mb RAM, 350Mhz processor and then from there move to testing on other plaforms of various hardware configurations and OS configurations such as Win 2000, WinXP, etc. A couple of the current MT software packages still work fine on the Win98 platform.
Training/Teaching the system:
I completed a fully documented MT project a few months ago with fully documented log sheets on what was done. This allows to provide benchmarked baseline information on productivity in the using the system. For a sample of approx 75 marketing press releases (approx 100 words each), most of them coming from different product and service announcements from different customer, in 1145 minutes I was able to manually hand analyze all of the text, identify all potential dictionary candidates (1000 entries), create an Excel-based glossary of all candidate terms, sort out which candidates were not necessary and streamline the dictionary down to half the size, translate all terms to be entered, code all entries + all variant and alternative entry forms, test all dictionary entries on the specific context text as well as alternate texts, retranslate and retest, and post-edit the final resulting text.
And the majority of this was done using MS Word and MS Excel on a Pocket PC or on a laptop sitting in a train or standing in a metro/subway during 15-30 minute trips. Only a few of the sessions were performed at a desk or in the yard with a computer for periods of longer than 1 hour.
All the work, the amount of entries created, and other tasks were carefully logged to provide benchmarked information on what can be done with a real translation project in time-limited and location-constrained contexts.
I even logged MT software bugs during that time, including 1 hour lost due to conversion issues between Pocket 2002 Excel and MS Excel 2003.
One of my next tasks is to measure the amount of time lost due to a missing feature (that I logged) which required me to hand code at least a hundred entries which should have been importable semi-automatically if the feature had been available for the usage that I reported (since it is available for a parallel usage in the same product).
These 75 marketing press releases represent 3 single date publications among press releases that have been published every 2-3 week for the past 10 years.
Another next step would be to run the entire set of 10 years of press release through the optimized minimal dictionary of about 500 terms and see how it fares for extended coverage.
All of these techniques can be taught to anyone who has an interest in words and terminology, understands the essential basics of word grouping, and has an interest in translation.
Jeff
http://www.geocities.com/mtpostediting/
| | |
|
|
|
Jeff Allen
France
Local time: 06:01
Multiplelanguages
+ ...
| new events in speech-to-speech translation systems | Oct 31, 2005 |
Jeff Allen wrote: Simultaneous speech translation: My background in fact includes speech-enabled translation with both speech recognition and speech synthesis modules....( http://www.lti.cs.cmu.edu/Research/Diplomat/).... At CMU, we developed complete birectional speech-to-speech translation systems for several language pairs .... used at least for periods of time in real-user contexts for which they were designed (basically military checkpoints and humanitarian organization efforts in war zones). ... "Full-scale simulataneous speech translation will be ready in about 5 years" Do I believe that? No. Yet I do believe that in another 5 years there will be more improved speech-enabled and speech-interactive systems that could be combined with text-based translation systems...
See below links to a press release and a follow-up article of the event that just appeared during the past week on speech-to-speech translation systems.
I will comment on this in the coming week as time permits.
Jeff
http://jeffallen.chez.tiscali.fr/about-jeffallen.htm
======= begin paste ===========
http://www.cmu.edu/PR/releases05/051013_alex.html
October 18, 2005
Carnegie Mellon and University of Karlsruhe To Demonstrate Breakthroughs In Cross Lingual Communication and Speech-to-Speech Translation
and the follow-up article ......
http://www.phillyburbs.com/pb-dyn/news/103-10272005-561361.html
Lost in translation? New technology comes to the rescue
By RAMESH SANTANAM
The Associated Press
October 27, 2005
======= end paste ===========
[Edited at 2005-10-31 17:24]
| | | | | | | Pages in topic: < [1 2 3 4] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Is this the future? Automatic simultaneous translation within 5 years? | 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! » |
| | 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 » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
