Pages in topic:   [1 2] >
Adaptive machine translation is coming? Or it's already here?
Thread poster: Spence Green
Spence Green
Spence Green  Identity Verified
United States
Local time: 09:30
Member (2014)
English to Arabic
+ ...
Aug 24, 2016

Online learning for MT (aka, "incremental training", aka "adaptive MT") was first investigated in 2006 by Liang et al. Various academic systems have been built over the years, but the technology is finally in production, as I announced in this forum in February. Adaptation addresses the singular complaint of translators from user studies dating to at least the 1970s: that the MT system doesn't *immediately* learn from feedback. This technology should help MT systems produce more relevant results... See more
Online learning for MT (aka, "incremental training", aka "adaptive MT") was first investigated in 2006 by Liang et al. Various academic systems have been built over the years, but the technology is finally in production, as I announced in this forum in February. Adaptation addresses the singular complaint of translators from user studies dating to at least the 1970s: that the MT system doesn't *immediately* learn from feedback. This technology should help MT systems produce more relevant results while the translator works.

It's great to see that SDL is now promoting the technology, which they will release this fall. However, their announcement contains a creative interpretation of the history and availability of this type of system.

Nevertheless, we're excited that they believe in the technology, too, and we hope that this development catalyzes widespread deployment of MT.
Collapse


 
Michael Beijer
Michael Beijer  Identity Verified
United Kingdom
Local time: 16:30
Member (2009)
Dutch to English
+ ...
(copying my reply from over at your LinkedIn post) Aug 24, 2016

Spence Green wrote:

Online learning for MT (aka, "incremental training", aka "adaptive MT") was first investigated in 2006 by Liang et al. Various academic systems have been built over the years, but the technology is finally in production, as I announced in this forum in February. Adaptation addresses the singular complaint of translators from user studies dating to at least the 1970s: that the MT system doesn't *immediately* learn from feedback. This technology should help MT systems produce more relevant results while the translator works.

It's great to see that SDL is now promoting the technology, which they will release this fall. However, their announcement contains a creative interpretation of the history and availability of this type of system.

Nevertheless, we're excited that they believe in the technology, too, and we hope that this development catalyzes widespread deployment of MT.



Michael J.W. Beijer
Dutch to English translator (IT/legal/patents), e-terminologist and translation software consultant.

Hi Spence. Interesting, but I think the main problem with your approach (which I only really tried for a few minutes, I must admit) is the fact that your CAT tool (or "TEnT") is online and extremely limited. I would be MUCH more interested in your offering if it were integrated into e.g. memoQ or Studio (or CafeTran)via an API/plugin.

Like Reply


 
Robert Rietvelt
Robert Rietvelt  Identity Verified
Local time: 17:30
Member (2006)
Spanish to Dutch
+ ...
@Michael Aug 24, 2016

Excuse me my curiosity, but you are a.o. 'e-terminologist' and 'translation software consultant'. Wow, sounds great. What is it?

[Edited at 2016-08-24 19:58 GMT]


 
Michael Beijer
Michael Beijer  Identity Verified
United Kingdom
Local time: 16:30
Member (2009)
Dutch to English
+ ...
ha ha, they're just words Aug 24, 2016

Robert Rietvelt wrote:

Excuse me my curiosity, but you are a.o. 'e-terminologist' and 'translation software consultant'. Wow, sounds great. What is it?

[Edited at 2016-08-24 19:58 GMT]


… that look good to my clients. But more seriously:

An "e-terminologist" is someone who is basically a terminologist, but whose work is to a very large extent based on digital and online resources, rather than on old-fashioned paper volumes, although I also have a huge collection of paper dictionaries on my bookshelves.

A "translation software consultant" is someone who can help both freelance translators and LSPs with anything related to CAT tools/TEnTs (I like to test them all), term extraction, corpora, web search tools, file prepping (CodeZapper/TransTools, etc.), PDF conversion, QA, translation memory and termbase consulting, OS optimisation, scripts, voice recognition, etc. Basically, all the stuff we use in our work these days.

And in my spare time, I translate

Michael

[Edited at 2016-08-24 20:09 GMT]

[Edited at 2016-08-24 20:09 GMT]


 
Robert Rietvelt
Robert Rietvelt  Identity Verified
Local time: 17:30
Member (2006)
Spanish to Dutch
+ ...
Aha! Aug 24, 2016

Michael Joseph Wdowiak Beijer wrote:

Robert Rietvelt wrote:

Excuse me my curiosity, but you are a.o. 'e-terminologist' and 'translation software consultant'. Wow, sounds great. What is it?

[Edited at 2016-08-24 19:58 GMT]


… that look good to my clients. But more seriously:

An "e-terminologist" is someone who is basically a terminologist, but whose work is to a very large extent based on digital and online resources, rather than on old-fashioned paper volumes, although I also have a huge collection of paper dictionaries on my bookshelves.

A "translation software consultant" is someone who can help both freelance translators and LSPs with anything related to CAT tools/TEnTs (I like to test them all), term extraction, corpora, web search tools, file prepping (CodeZapper/TransTools, etc.), PDF conversion, QA, translation memory and termbase consulting, OS optimisation, scripts, voice recognition, etc. Basically, all the stuff we use in our work these days.

And in my spare time, I translate

Michael

[Edited at 2016-08-24 20:09 GMT]

[Edited at 2016-08-24 20:09 GMT]




 
DLyons
DLyons  Identity Verified
Ireland
Local time: 16:30
Spanish to English
+ ...
Who is we? Aug 24, 2016

Spence Green wrote:

"we're excited that they believe in the technology


We as in the good folk at Lilt selling e.g. a version for project managers at LSPs and agencies starting at
$99/mo.


 
Mirko Mainardi
Mirko Mainardi  Identity Verified
Italy
Local time: 17:30
Member
English to Italian
MT Aug 24, 2016

Spence Green wrote:

Nevertheless, we're excited that they believe in the technology, too, and we hope that this development catalyzes widespread deployment of MT.



On the other hand, we(*) believe MT, PEMT, adaptive MT and the like will "catalyze" even worse work conditions and rates for translators at large and contribute to further devaluing their profession and creativity.

*Liberal usage of the royal we, ofc


 
DLyons
DLyons  Identity Verified
Ireland
Local time: 16:30
Spanish to English
+ ...
Lilt Aug 24, 2016

Just looked at their cringe-making page. Among other things it attributes "You're a translator. Lilit can help" to Mildred Harnack https://en.wikipedia.org/wiki/Mildred_Harnack

Something similar for Georgia O'Keeffe.

Rather throws the CEO's academic credibility into question IMHO.




[Edited at 2016-08-24 20:51 GMT]

[Edited at 2016-
... See more
Just looked at their cringe-making page. Among other things it attributes "You're a translator. Lilit can help" to Mildred Harnack https://en.wikipedia.org/wiki/Mildred_Harnack

Something similar for Georgia O'Keeffe.

Rather throws the CEO's academic credibility into question IMHO.




[Edited at 2016-08-24 20:51 GMT]

[Edited at 2016-08-24 20:56 GMT]
Collapse


 
Bernhard Sulzer
Bernhard Sulzer  Identity Verified
United States
Local time: 12:30
English to German
+ ...
If adaptive MT is coming, let's talk shop. Aug 24, 2016

Spence Green wrote:

Online learning for MT (aka, "incremental training", aka "adaptive MT") was first investigated in 2006 by Liang et al. Various academic systems have been built over the years, but the technology is finally in production, as I announced in this forum in February. Adaptation addresses the singular complaint of translators from user studies dating to at least the 1970s: that the MT system doesn't *immediately* learn from feedback. This technology should help MT systems produce more relevant results while the translator works.

It's great to see that SDL is now promoting the technology, which they will release this fall. However, their announcement contains a creative interpretation of the history and availability of this type of system.

Nevertheless, we're excited that they believe in the technology, too, and we hope that this development catalyzes widespread deployment of MT.



Don't mean to offend Spence but this almost sounds like an ad for the miraculous adaptive power of SDL's forthcoming "adaptiveMT" engine (fancy!) = "adaptive machine translation" which, as a technology, as you say, was investigated 10 years ago and has now seemingly reached critical mass and promises to be another fantastic tool we can BUY making things super easy for us and will give some people reason enough to demand even more discounts I am sure. I work with a CAT tool and although it often helps greatly with consistency and terminology - which is based on my human input -, it's no magic tool. It's a bilingual library from which translated segments that "seem" to match new untranslated text segments can be extracted quickly. I'll give you the algorithm that allows this for segments that don't match word for word - which also always means they might have to be translated quite differently or need very important adjustments. Of course, any TM will be more helpful the more text I input, and as long as my new texts really are similar in subject and the way they NEED to be translated; thus it will be better adapted if you will (I tend not to treat a TM like a digital proxy of myself cause it isn't, by a long shot [it's not the human or a human proxy that is actually translating ...], and thus don't describe myself as giving IT [=the TM] feedback).
Enter the "power" of new adaptive MT which is coming - SDL advertises adaptiveMT for Trados 2017 and promises a highly individualized translation experience where the new engine will be capable of providing my style and class at my fingertips, as I move from one word or segment (which is it?) to the next, imitating my style and terminology so perfectly that I will doubtlessly blush (sorry, I got carried away here).
Where the new magic is supposed to come from I don't know because TMs already update asap and you can use that in your next segment. What are we talking about here? Speed as in seconds? It should already be immediate. How this new engine will not only "adapt" more quickly but seek out our most fitting style and terminology for the new text isn't really explained yet.

So we get (more!!??) immediate and more relevant results than ever before? How so?

I am all for technology because it can be very helpful, but let's not make it sound again like the MT androids will take over soon.
And as far as ever more or "seemingly ever more often updated/upgraded" technology is published and sold successfully, the demands for discounts seem to be reinvigorated because we can provide translations in the blink of an eye (well, let's say impressively fast).

When did it ever become cheaper to send a letter by express mail? Think about it.

I'll be back to post a link to a previous thread that contains more info related to this topic.

Here is that link:
http://www.proz.com/forum/sdl_trados_support/305435-sdl_trados_studio_2017_is_coming_soon.html



and the questions I raised there:
http://www.proz.com/forum/sdl_trados_support/305435-sdl_trados_studio_2017_is_coming_soon-page4.html#2581591

[Edited at 2016-08-24 21:45 GMT]


 
Spence Green
Spence Green  Identity Verified
United States
Local time: 09:30
Member (2014)
English to Arabic
+ ...
TOPIC STARTER
Re: MT Aug 24, 2016

Mirko Mainardi wrote:

On the other hand, we(*) believe MT, PEMT, adaptive MT and the like will "catalyze" even worse work conditions and rates for translators at large and contribute to further devaluing their profession and creativity.



This is a reductive analysis of downward pressure on per-word pricing in the industry. At least as powerful as technology are the low barrier to entry to the industry, outsourcing by agencies to low-wage countries, and the rise of commoditized translation platforms (Gengo, Unbabel, etc.) that emphasize convenience.

Absent collective action, freelance translators have no control over these three market forces.

The single variable that translators can control is hourly productivity and thus the *effective* hourly rate. And that is where automation---be it TM, MT, or any other linguistic resource---is obviously useful.


 
Spence Green
Spence Green  Identity Verified
United States
Local time: 09:30
Member (2014)
English to Arabic
+ ...
TOPIC STARTER
Re: Lilt Aug 24, 2016

DLyons wrote:

Rather throws the CEO's academic credibility into question IMHO.


I'm not sure about the link between our marketing campaign and my reputation as a computer scientist....

FWIW, we (as in our team) were in Berlin last week and visited a memorial to Harnack. It was inspiring. The other figures in our campaign are people who both inspired us personally and loved language.


 
Michael Beijer
Michael Beijer  Identity Verified
United Kingdom
Local time: 16:30
Member (2009)
Dutch to English
+ ...
hmm Aug 24, 2016

Spence Green wrote:

Mirko Mainardi wrote:

On the other hand, we(*) believe MT, PEMT, adaptive MT and the like will "catalyze" even worse work conditions and rates for translators at large and contribute to further devaluing their profession and creativity.



This is a reductive analysis of downward pressure on per-word pricing in the industry. At least as powerful as technology are the low barrier to entry to the industry, outsourcing by agencies to low-wage countries, and the rise of commoditized translation platforms (Gengo, Unbabel, etc.) that emphasize convenience.

Absent collective action, freelance translators have no control over these three market forces.

The single variable that translators can control is hourly productivity and thus the *effective* hourly rate. And that is where automation---be it TM, MT, or any other linguistic resource---is obviously useful.


I suspect there might be a few more variables we can control.

Michael


 
Mirko Mainardi
Mirko Mainardi  Identity Verified
Italy
Local time: 17:30
Member
English to Italian
Re: Re: MT Aug 24, 2016

Spence Green wrote:

Mirko Mainardi wrote:

On the other hand, we(*) believe MT, PEMT, adaptive MT and the like will "catalyze" even worse work conditions and rates for translators at large and contribute to further devaluing their profession and creativity.



This is a reductive analysis of downward pressure on per-word pricing in the industry. At least as powerful as technology are the low barrier to entry to the industry, outsourcing by agencies to low-wage countries, and the rise of commoditized translation platforms (Gengo, Unbabel, etc.) that emphasize convenience.

Absent collective action, freelance translators have no control over these three market forces.

The single variable that translators can control is hourly productivity and thus the *effective* hourly rate. And that is where automation---be it TM, MT, or any other linguistic resource---is obviously useful.


Your adaptive auto-correct feature must have somehow replaced "quality" with "hourly productivity and thus the *effective* hourly rate". Weird.

No, really, Spence, translation is not assembly line work, not even in more "technical" fields (where there arguably is more standardization in terminology/syntax). And even if it were, in what universe this new and wonderful technology would only benefit translators, without agencies taking advantage of it to push rates further down (as they did/do with fuzzy matches and PEMT, for instance)?


 
DLyons
DLyons  Identity Verified
Ireland
Local time: 16:30
Spanish to English
+ ...
I am quite sure about the status of misattribution in academia. Aug 24, 2016

Spence Green wrote:

DLyons wrote:

Rather throws the CEO's academic credibility into question IMHO.


I'm not sure about the link between our marketing campaign and my reputation as a computer scientist....

FWIW, we (as in our team) were in Berlin last week and visited a memorial to Harnack. It was inspiring. The other figures in our campaign are people who both inspired us personally and loved language.


If you actually had any respect for her, that pseudo-quote would not have been on your website!


[Edited at 2016-08-25 00:26 GMT]


 
Spence Green
Spence Green  Identity Verified
United States
Local time: 09:30
Member (2014)
English to Arabic
+ ...
TOPIC STARTER
Re: lots of questions! Aug 24, 2016

Bernhard Sulzer wrote:

...promises to be another fantastic tool we can BUY making things super easy for us and will give some people reason enough to demand even more discounts I am sure.



I think that the general discussion about MT would be greatly enhanced if the technology and the concept of rate discounting were de-coupled.

Were I a translator---and I'm not, for sure!---I wouldn't accept a flat-rate discount for a basic post-editing job. The literature shows high variance (both negative and positive, but mostly positive in recent years) in terms of the effect of machine assistance on translator productivity. Some of this variance can be ascribed to input MT quality, but some of it is also due to individual variation. Uninformed, flat rate discounting simply isn't justified by the literature.

The empirical solution is to measure yourself on the types of content that you typically translate so that you can see if assistance improves productivity. It may not. But if it does, then you can compute your effective hourly rate. From that you can make an informed decision on how to modify (if at all) your per-word price so that you maintain your effective hourly rate.


 
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 »

Adaptive machine translation is coming? Or it's already here?






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 »
Protemos translation business management system
Create your account in minutes, and start working! 3-month trial for agencies, and free for freelancers!

The system lets you keep client/vendor database, with contacts and rates, manage projects and assign jobs to vendors, issue invoices, track payments, store and manage project files, generate business reports on turnover profit per client/manager etc.

More info »