| Pages in topic: < [1 2 3] > | Machine Translation - anyone using it successfully? Thread poster: gianfranco
|
|---|
Yasiri
Local time: 00:27
Arabic to English
+ ...
It's useful for me when I use it in translating short expressions and idioms rather than a paragraph or even long sentences , and almost gives me the right equivalent collocation ,
and I offten use MT to translate from "Latin", Chinese languages into English then to mine and it works fine .
[Edited at 2009-10-26 21:08 GMT]
| | | | Susan Welsh 
United States
Local time: 17:27
Russian to English
+ ...
| Different kinds of MT | Nov 2, 2009 |
Jeff Allen has posted a lot on this already, in another forum.
I just want to add that at the American Translators Association conference this past week, there was a panel discussion on MT, Crowd Sourcing, and Cloud Sourcing. One important point that was made, was that the MT systems that are developing really fast are the STATISTICAL ones, rather than the RULE-BASED ones. Google Translate is a statistical system. Large corporations also formed TAUS, to pool TMs and MT data, but thi... See more Jeff Allen has posted a lot on this already, in another forum.
I just want to add that at the American Translators Association conference this past week, there was a panel discussion on MT, Crowd Sourcing, and Cloud Sourcing. One important point that was made, was that the MT systems that are developing really fast are the STATISTICAL ones, rather than the RULE-BASED ones. Google Translate is a statistical system. Large corporations also formed TAUS, to pool TMs and MT data, but this involves millions of pages of documents, and is therefore way out of reach of ordinary mortals (translators). The MT systems that you can buy and train as a freelancer are RULES-BASED. These systems are not developing at nearly so rapid a rate, although the presenter said they are improving slowly. And, as others have noted, they take a lot of work on the translator's part to input dictionaries, declensions, etc.
Susan ▲ Collapse
| | | |
I started using Power Translator many years ago for my biomedical French-English translation work. The results were awful, but it was still quicker editing the "franglais" than typing the entire translation with two fingers.
To improve the results I developed a few macros (mini-programs) in Word, whereby I could copy, with a single keystroke (Ctrl + 1, 2, 3, 4 or 5) frequently recurring mistranslations (1, 2, 3 , 4 or 5 words) into a separate Word file consisting of a two-column table. O... See more I started using Power Translator many years ago for my biomedical French-English translation work. The results were awful, but it was still quicker editing the "franglais" than typing the entire translation with two fingers.
To improve the results I developed a few macros (mini-programs) in Word, whereby I could copy, with a single keystroke (Ctrl + 1, 2, 3, 4 or 5) frequently recurring mistranslations (1, 2, 3 , 4 or 5 words) into a separate Word file consisting of a two-column table. Once a month or so I would add the correct translations to the right-hand column and use another macro to convert the table into a list of "search-replace" commands.
Over the years I've built up a database of about 2500 mistranslations and their corrections.
When I start a new translation I run it through Power Translator (same old version...) then execute the database-macro that corrects most of the frequent mistranslations. The whole process takes between 5 and 10 minutes, and the result is pretty good - better than the latest Systran release, anyway.
I'm extremely impressed by Google Translate, although the mistranslations can be dangerous.
In my view, customized MT can be very useful for translators working in a narrow subject area, but less so for "generalists".
Having followed this subject closely for some 20 years, I'm pretty sure that things are poised to accelerate quickly in the field of MT, especially using corpus-based approaches -- and the Web is a pretty massive digital corpus. ▲ Collapse
| | | | Jeff Allen
France
Local time: 23:27
Multiplelanguages
+ ...
| statement that rule-based MT systems are developing slowly | Nov 22, 2009 |
Susan Welsh wrote:
Jeff Allen has posted a lot on this already, in another forum.
I just want to add that at the American Translators Association conference this past week, there was a panel discussion on MT, Crowd Sourcing, and Cloud Sourcing. One important point that was made, was that the MT systems that are developing really fast are the STATISTICAL ones, rather than the RULE-BASED ones. Google Translate is a statistical system. Large corporations also formed TAUS, to pool TMs and MT data, but this involves millions of pages of documents, and is therefore way out of reach of ordinary mortals (translators). The MT systems that you can buy and train as a freelancer are RULES-BASED. These systems are not developing at nearly so rapid a rate, although the presenter said they are improving slowly. And, as others have noted, they take a lot of work on the translator's part to input dictionaries, declensions, etc.
Thanks Susan,
I've actually posted a lot on this in a number of fora here on ProZ. Easy to find them by just searching on my posting alias (mtpostediting).
As for the ATA conf in NY last week, always take a very detailed look at the background of anyone giving a talk or participating in a panel discussion about MT.
The statement that Rule-based systems are not developing is not quite true. They are developing regularly with bug fixes released in patches probably every quarter. I've been the program manager for releases for some of those products, so I know this from real experience.
Some of the mainline Commercial MT vendors that develop the mainline MT software have combined statistical-based approaches to their primarily rule-based engines and have announced these hybrid solutions over the past year.
As for the part of creating the dictionaries for the rule-based MT system and it taking a lot of time, it is important to ask yourself who is stating this. If it is people who have never been trained on such a system and who are stating things 2nd hand, then of course if it easy to say it takes lots of time.
What is the real practical and pragmatic experience with the tools in a Translation/Edit/Proof (TEP) production cycle?
Were there any tutorials on MT systems at the ATA 50th conf?
Were any MT vendors present at the ATA 50th conf?
I can take any of the MT software programs sitting on my shelf (several different brand names, with a range of problably 20+ different product models and versions) and for most of them can start immediately creating dictionary entries in a software interface. Several of them offer batch mode import from text file and Excel sheet formats. several of my published software reviews from 2001 through 2006 indicate that that there are feature to improve the dictionary creation process.
As can been seen from my Proz profile, I have many years of experience as a professional translator, and also as a sales/account/project manager of translation and localization services, which means that I know what the real production cycle is and what the expectations are for language service providers (both companies and freelance translators). And yet also have been involved in the development, testing, deployment and user training for about 10 different MT systems/software (all types: stat-based, example-based, rule-based, knowledge-based, multi-engine, as well as the development, enhancement and deployment of Translation memory, Authoring Memory, Terminology Management and Translation management systems) for the past 20 years.
See my my tutorial at AMTA2004 on MT Postediting (at http://www.mt-archive.info/AMTA-2004-Allen-tutorial.pdf , but an original, non-scanned, more easily readable version of it up on my Linked Profile at http://www.linkedin.com/in/jeffallen > files > MT ), I clearly stated that almost all MT vendors have a lot of computer geeks who are trying to break the "communication barrier" (the Star Trek syndrome), and nearly all of the Translation Memory vendors are people coming from the professional translation services field who developed a tool to solve the needs of "professional translation" based on concrete needs for translators, and then made it available as a general product.
Who has been more successful to the professional translation community? The TM tools.
Why? because they knew how to talk to the professional translators.
Stat-based MT is mainly aimed at large enterprise installation.
Rule-based MT also focuses on catering to that same audience (for sales reasons), but also have the tools to provide productivity tools to the smaller job audiences. Do they really do it? That's another story.
It is possible to tap into both stat-based and rule-based MT software through various TM tools.
thanks for the info on what was said.. If I had been at the conference, I would have immediately corrected a number of statements that seem to have been made.
Maybe I should create a 1-2 page Powerpoint slide that shows all the linked relationships between the commercially sold TM and MT tools.
Jeff
[Edited at 2009-11-22 15:43 GMT]
| | |
|
|
|
Susan Welsh
United States
Local time: 17:27
Russian to English
+ ...
| ATA Conference | Nov 22, 2009 |
Jeff Allen wrote:
Were there any tutorials on MT systems at the ATA 50th conf?
Were any MT vendors present at the ATA 50th conf?
No and no. This was an hour-long, apparently impromptu, forum (wasn't on the program) in the slot for the Language Technology Division's annual meeting (which they dispensed with in about 5 minutes).
There was a guy there from the Assn for MT in America (http://www.amtaweb.org), Alain Lavie [phonetic], who introduced himself from the floor and said that the AMTA has planned its conference next year to be in Denver, right after the ATA conference there, so as to try to attract translators. They will start on Sunday with tutorials oriented toward translators. I don't recall that he intervened in the discussion otherwise.
Susan
| | | | Jeff Allen
France
Local time: 23:27
Multiplelanguages
+ ...
| MT participation at ATA conf | Nov 22, 2009 |
Jeff Allen wrote:
Were there any tutorials on MT systems at the ATA 50th conf?
Were any MT vendors present at the ATA 50th conf?
Susan Welsh wrote: No and no. This was an hour-long, apparently impromptu, forum (wasn't on the program) in the slot for the Language Technology Division's annual meeting (which they dispensed with in about 5 minutes). There was a guy there from the Assn for MT in America ( http://www.amtaweb.org), Alain Lavie [phonetic], who introduced himself from the floor and said that the AMTA has planned its conference next year to be in Denver, right after the ATA conference there, so as to try to attract translators. They will start on Sunday with tutorials oriented toward translators. I don't recall that he intervened in the discussion otherwise.
Alon Lavie is the current president of AMTA. He is one of my former colleagues (I worked at the Center for Machine Translation / Language Technologies Institute of Carnegie Mellon Univ a number of years ago). He is one of the professors at that institute, and they work a lot on statistical-based MT systems. Others from that institute work on Knowledge-based MT systems. Also a lot of work in speech-MT (which I also did there).
Yes, I know that AMTA is scheduling its conference in coordination with ATA in Denver next year. I was part of the org committees for 2 of the AMTA conferences and worked on trying to get professional translator participation. It was too early at that point in time. The interest is now picking up with GT and all the hype about stat-MT and hybrid systems.
But professional translators are not on the academic/industrial research timeline. The economomic crisis is still on and professional translators need solutions now to be able to produce more content in less time, and face the issues of declining price rates from customers. This is reality.
End of next year to have a conference to give tutorials is a bit of a wait.
Some of these tutorials take 20-30 min to do in an online session. It's not really that complicated at all, and isn't rocket science to organize it.
Gee, maybe I should just talk with Henry D and see what we can organize through ProZ in the short term. Something like a virtual ProZ Pow-wow on the topic of MT software/systems.
how many Proz translators can we get to participate in a live online intro tutorial on MT software?
Maybe this deserves a ProZ poll.
Jeff
| | | | Jeff Allen
France
Local time: 23:27
Multiplelanguages
+ ...
| comments on TM macros tricks + MT for subject area specializations vs generalists | Nov 22, 2009 |
David,
Thanks for your very interesting post above. Some comments on it.
david young wrote:
I started using Power Translator many years ago for my biomedical French-English translation work.
.... To improve the results I developed a few macros (mini-programs) in Word,
.... use another macro to convert the table into a list of "search-replace" commands
* Macros:
You are using Word in a way that is the same foundation on which TM tools were designed. And good to see that many translators are doing that.
I have stated in 2 other posts that TM tools are just simply multilingual advanced search-replace mechanisms that provide users with a software interface to guide them through this in a task-oriented way, rather than having to design, maintain and update between-the-scenes macros to achieve the same results:
http://www.proz.com/post/1055587#1055587
http://www.proz.com/post/1240258#1240258
Yet, one can get stuck with exact match patterns (partial or total exact segments--- not exact or fuzzy matches-- depending how you process it).
With several MT software products, however, you can set the part-of-speech category on each word within multi-word terms, so that these can be produced correctly in their declined and conjugated forms in the target language.
Some MT tools have special categories to handle short-segment expressions, and 2 brands have created internal TM modules as pre-processing filters to handle exact match segments like you do with your macros.
Here's an additional tip for your macros. I've used Regular Expressions in MS Word to customize the search/replace patterns in even more depth. Yet, I think Microsoft changed this a bit in recent versions of Word, or maybe they moved it. I haven't needed to use it for a couple of years because I've found other ways to do things faster and better with my set-up.
* Subject area specializations vs generalists
david young wrote:
In my view, customized MT can be very useful for translators working in a narrow subject area, but less so for "generalists".
MT can actually cover both. I have many areas of subject area expertise (formal degrees, professional certifications and work experience, all indicated in my Proz profile) and have used MT in all of them. Also have used MT in quite a few other areas as specific needs arose, and which successfully met the requirements of the translation request.
Below is a consolidated list of areas and topics for MT-related projects that I have personally translated or have been heavily involved in to coordinate, advise and/or execute the project with other participants. I may have left out a few topic, but I think this one provides a good range.
Marketing/sales:
- Company web site localization
* consumer electronics (product descriptions and fact sheets)
* travel/tourism/accomodations
* software companies (different fields)
- calls for bids/proposals and response documents (software/systems)
- Software application descriptions and marketing brochures
News:
- press releases (successful pilot project on sample of 75 texts to extend by 100x to reach 1M source words)
- Newswire (17 separate main subject areas, 2 year project covering 2-3M source words, 3 target languages)
- newspaper texts
Customer Support:
- online real-time customer support via instant messaging for troubleshooting software installation problems
- customer support FAQs and Knowledge base information
Telecommunications Wireless Networks documentation
- software test acceptance plan
- marketing/sales RFI/RFQ documents
- user documentation
- training course materials
Military/para-military
- wayfinding and topic detection texts for speech-to-speech MT systems
- content of aerospace sector
Specific industries
- telecom (see above)
- general computer science field dictionary
- automotive sector dictionary
- construction heavy-machinery and engines (multilingual terminology database, user manuals, service manuals)
Human Resources related topics
- job announcements
- resumes/CVs
- Reduced Work week local law information
- various specific topics on local laws
Software localization projects
Training
- various training course materials (various software topics)
- also user manuals of various software products
Linguistics
- published peer-reviewed journal and magazine articles and conference papers
- masters and doctoral theses
Corporate business
- corporate project reports and presentations
- weekly program / project management meeting minutes
Legal documents (Reduced Work week doc, Child Adoption documents, legal statutes for organizations, corporate legal disputes and mediation)
Theological texts
- series of presentations on book of Revelation
- set-up an entire workflow to scan documents, check source text, and MT texts on New Testament content and then extract out (sub)sections for a specific need
- create translation of software review article of Bible software product
- article on the background of Bible translation as related to the translation industry
- conducted a 1-day project on 4 years of content of monthly newsletters sent out to key supporters of mission organizations
- children's book
Other misc topics
- set-up and installation instructions for music/sound systems within a new building
- general communication e-mails on many topics
And I've even been investigating using MT dictionary building on a few other topics that could provide interesting derivative dictionaries for the related industries.
- recipes (food)
- children's toys/games
note: several other previous posts on Proz indicate partial lists of what is indicated above:
http://www.proz.com/post/177427#177427
http://www.proz.com/post/192815#192815
http://www.proz.com/post/1209811#1209811
http://www.proz.com/post/213454#213454
http://www.proz.com/post/1205999#1205999
MT software/systems can be used in many more fields and on types of content than most people are willing to admit and would even try it out on. It just requires mastery of the tools, knowing the process and steps to follow to be successful for any given analyzed project, and availability of subject matter experts to make it possible.
Jeff
| | | | Susan Welsh
United States
Local time: 17:27
Russian to English
+ ...
| Proz tutorial | Nov 22, 2009 |
Jeff Allen wrote:
Gee, maybe I should just talk with Henry D and see what we can organize through ProZ in the short term. Something like a virtual ProZ Pow-wow on the topic of MT software/systems.
how many Proz translators can we get to participate in a live online intro tutorial on MT software?
Maybe this deserves a ProZ poll.
I think that's a great idea!
But I never participate in these Proz polls, and wish they would take them off the home page. "What kind of music do you listen to while you translate?" Humbug!
I say just go for it, Proz should publicize it, and see who shows up.
Susan
| | |
|
|
|
| Machine Translation - anyone using it successfully?" | Nov 22, 2009 |
gianfranco wrote:
I would like to know if there is anyone successfully using MT tools in their translation cycle, in-house or as a freelancer.
I know, I know... this is likely to raise a hell of a negative criticism, but this is not exactly what I\'m aiming at.
Please, don\'t flood this topic with all the obvious comments about the rubbish that you have seen coming out from MT. We all agree that any self-respecting translator wouldn\'t deliver the output straight from the MT tool.
I\'m just curious to know if MT tools have been integrated in a translation cycle, that includes all the necessary steps in editing, reviewing, spell-checking and proof-reading, according to the quality requirements.
Any experience to share?
[ This Message was edited by: on 2001-06-18 02:14 ]
Keep it simple. MT is useful:
a) if it is used with other CAT tools
b) translation pairs are not of etymologically remote languages
c) you can setup them to a great extent for your translation pair
According to my poor knowledge, this trend began with the integration of MT in Wordfast and his last expression is GT4T. But, if MT does not fulfill some of the three previous premises, I believe it will be not useful at all. Just my 2 cents...
[Editado a las 2009-11-22 23:20 GMT]
| | | | Jeff Allen
France
Local time: 23:27
Multiplelanguages
+ ...
| forgot banking sector website localization | Nov 23, 2009 |
david young wrote:
In my view, customized MT can be very useful for translators working in a narrow subject area, but less so for "generalists".
Jeff Allen wrote:
* Subject area specializations vs generalists
MT can actually cover both ...
Below is a consolidated list of areas and topics ...
Marketing/sales:
- Company web site localization
* consumer electronics (product descriptions and fact sheets)
* travel/tourism/accomodations
* software companies (different fields)
I forgot to add company website localization for the banking sector as well as technical documentation from various companies (software and others).
Jeff
| | | | Jeff Allen
France
Local time: 23:27
Multiplelanguages
+ ...
| MT can even be subject to ISO9001 quality management | Nov 23, 2009 |
gianfranco wrote:
I'm just curious to know if MT tools have been integrated in a translation cycle, that includes all the necessary
steps in editing, reviewing, spell-checking and proof-reading, according to the quality requirements.
Lori Thicke of Lexcelera just made the following statement in a post on LinkedIn: "By the way, we hold our MT translations to the same ISO 9001 quality standards that we hold our human translations to. "
Automated Language Translation Group on LinkedIn
http://www.linkedin.com/groupAnswers?viewQuestionAndAnswers&discussionID=9773035&gid=148593&commentID=8435410&trk=view_disc
As ISO9001 requires independent audits by an external auditor, including the tracking of customer problems, corrective measures, and preventive measures, that is a significant statement about their commitment to implementing MT in optimal ways.
Jeff
(much experience in ISO9001 audits having obtained Quality Auditor certification and 2 years of organizing and running audits)
[Edited at 2009-11-24 06:08 GMT]
| | | | Jeff Allen
France
Local time: 23:27
Multiplelanguages
+ ...
| when is MT useful? | Nov 24, 2009 |
Pablo Bouvier wrote:
Keep it simple. MT is useful:
a) if it is used with other CAT tools
b) translation pairs are not of etymologically remote
languages c) you can setup them to a great extent for your translation pair
According to my poor knowledge, this trend began with the integration of MT in Wordfast and his last expression is GT4T. But, if MT does not fulfill some of the three
previous premises, I believe it will be not useful at all.
Hi Pablo,
a) MT doesn't need other CAT tools because the MT tools are creating such environments in their tools as well as the TM tools creating links to the MT tools.
After numerous presentations during the late 90s about how MT, TM, and terminology systems were integrated together in various industrial sectors by translation buyers, there were a few attempts by MT and TM vendors to partner together but did not go very far.
Slide 33 of of my AMTA200 MT tutorial (http://www.mt-archive.info/AMTA-2004-Allen-tutorial.pdf) gives the workflow diagram from a presentation given at the Society for Automotive Engineering (SAE)/Localization Industry Standards Association (LISA) Toptec in 1999 Amsterdam.
By 1999, Deja Vu was already implementing the Assemble feature.
Artile on Deja Vu in the 2 issues of the ELRA/ELDA quarterly newsletter in 1999 focusing on TM tools (http://www.elra.info/Newsletters-from-1996.html#1999)
Wordfast was simply the first lter on to put it into a different and more user-familiar translation workstation environment (ie, MS Word).
By 2003, a few MT vendors were already making proprietary internal Translation Memory modules available within their own software (PROMT called it associated memory, and Systran called it just translation memory), and also providing plug-in adaptors for the major TM products on the market (PROMT has a Trados4PROMT plugin)
b) MT only good for related langages. There have been successful implementations of non-related languages. Take for example Symantec which has implemented it not just between EN and other FIGES languages, but also with Japanese and Chinese:
first page of PDF under the topic "Solutions" of:
http://www.systran.co.uk/systran/corporate-profile/translation-case-studies/symantec
section 4.3 of:
http://www.mt-archive.info/MTS-2009-Roturier.pdf
slide 3 of
http://www.mt-archive.info/Translingual-Europe-2009-Roturier-ppt.pdf
c) you are very right. These hybrid solutions can be set up for specific language pairs. Much depends on which specific MT and TM tools are able to link up.
However, it's better to link up a TM tool with an MT "software package" or "internally installed system" rather than pointing out to Google or other free online systems (the free systems are the lowest-end of the quality for MT use).
It's good to see that many TM tools are starting to consider setting up connections with MT.
Jeff
| | |
|
|
|
Tim Drayton
Cyprus
Local time: 00:27
Turkish to English
+ ...
| Cognate languages v. others | Nov 24, 2009 |
b) MT only good for related langages. There have been successful implementations of non-related languages. Take for example Symantec which has implemented it not just between EN and other FIGES languages, but also with Japanese and Chinese:
first page of PDF under the topic "Solutions" of:
http://www.systran.co.uk/systran/corporate-profile/translation-case-studies/symantec
section 4.3 of:
http://www.mt-archive.info/MTS-2009-Roturier.pdf
slide 3 of
http://www.mt-archive.info/Translingual-Europe-2009-Roturier-ppt.pdf
[...]
[/quote]
The links you have quoted are all to pieces written in praise of a certain company's software on that company's own website. That is hardly the most objective source of information. I have recently glanced at quite a few academic artiles about developing statistical MT translation software to translate from Turkish into English, and they all seem to be suggesting that a significant number of transformations need to be applied to a sentence before it can be submitted to statistical processing. These are all tentative suggestions at the moment and I do not feel that anybody is close to cracking this particular 'nut' soon.
The day that I see with my own eyes that MT can produce translations from Turkish into English of documents which the parties to a complex lawsuit feel comfortable about submitting in evidence is the day that I will throw in the towel. This day may come, but probably not in the next ten years.
I think it is obvious to anyone with a good grounding in language and linguistics and at least a passing knowledge of the huge diversity that exists in natural languages around the world that it is going to be much easier to develop MT software between cognate languages than between those with radically different structures. MT is going to have to succeed convincingly with closely related languages and then move on to conquer other territory.
| | | | Susan Welsh
United States
Local time: 17:27
Russian to English
+ ...
| The idea is not to elminate the human being! | Nov 24, 2009 |
Tim Drayton wrote:
The day that I see with my own eyes that MT can produce translations from Turkish into English of documents which the parties to a complex lawsuit feel comfortable about submitting in evidence is the day that I will throw in the towel.
If that is your criterion for a useful product for translators, then you're asking a bit much! I have used a rudimentary MT program (Google Translate) between Russian and English, which are not "cognate languages" in any meaningful sense, and for some texts it did poorly, and for others it was extremely helpful. That does NOT mean it's "ready" for legal briefs that no human eye need peruse! ALL MT translations need to be edited by a human being, and always will, since no machine can detect all the puns, jokes, ambiguities, and ironies that the human mind is capable of creating (perhaps, admittedly, not in a legal brief!)
It's silly to say, "If it's not perfect, I won't use it." The same could be said of my word processor. Or my food processor, for that matter.
There are zillions of posts on the site from the past two weeks, notably by Jeff Allen, which will make people who are interested better informed. Look in the MT forum.
Susan
| | | | Tim Drayton
Cyprus
Local time: 00:27
Turkish to English
+ ...
| They are distant relatives | Nov 24, 2009 |
ussian and English, which are not "cognate languages" in any meaningful sense
[...]
[/quote]
To split hairs, they are both members of the Indo-European language family, so in at least one sense they are 'cognate'. They may well be quite different languages, but a they are still a great deal more similar than, say, Turkish and English. I have a postgraduate diploma in Russian language, so I can comment with some degree of certainty here.
| | | | | Pages in topic: < [1 2 3] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Machine Translation - anyone using it successfully? | 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 » |
| | TM-Town | Manage your TMs and Terms ... and boost your translation business
Are you ready for something fresh in the industry? TM-Town is a unique new site for you -- the freelance translator -- to store, manage and share translation memories (TMs) and glossaries...and potentially meet new clients on the basis of your prior work.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
