Machine translation specialized softwares
Thread poster: Bruno Depascale

Bruno Depascale  Identity Verified
Italy
Local time: 15:36
Member (2009)
English to Italian
+ ...
Aug 1, 2014

Hi everybody,
I am writing to ask you about specialized machine translation softwares to be used within Trados 2014.
I have installed microsoft translator, but it is very complicated to train it, and I haven't understood if I can train it with translation memories.
I have also talked with Trados representative about SDL beglobal, but this service is aimed at corporations, and they won't let you test it before buying it.
For what concerns the other plugins, like mymemory, I don't know if it can be trained and I haven't been able to import my TMs in it.
I am looking for machine translation softwares (paying a fee) that can be trained with TMX and termbases, because I have a client who sends me regularly a high amount of easy medical texts, so MT could speed my work greatly in this particular case.
Thank you very much for your input, I wish you a nice day.
Bruno


Direct link Reply with quote
 

Radian Yazynin  Identity Verified
Local time: 16:36
Member (2004)
English to Russian
+ ...
Promt Aug 1, 2014

Hi Bruno,
Promt, as standalone software, can be used as a plug-in in Trados. You need to choose a proper version for yourself. By the way, plugging in MT and training it with translation memories within a CAT is really unnecessary because you would use a rules-based MT which uses TMs from CAT as is, while CAT would use such TMs itself, in this case nothing would change. Of course, there is another variant, a hybrid MT, which can use a statistical scheme, and this is where you can train the system, but such option is more expensive, though available too.
Good luck!


Direct link Reply with quote
 

Bruno Depascale  Identity Verified
Italy
Local time: 15:36
Member (2009)
English to Italian
+ ...
TOPIC STARTER
Systran? Aug 1, 2014

Radian Yazynin wrote:

Hi Bruno,
Promt, as standalone software, can be used as a plug-in in Trados. You need to choose a proper version for yourself. By the way, plugging in MT and training it with translation memories within a CAT is really unnecessary because you would use a rules-based MT which uses TMs from CAT as is, while CAT would use such TMs itself, in this case nothing would change. Of course, there is another variant, a hybrid MT, which can use a statistical scheme, and this is where you can train the system, but such option is more expensive, though available too.
Good luck!


Thank you for your help Radian!
Indeed, I am not at all practical about MT.
I have seen that another company, Systran, uses a hybrid MT, and I have just asked them for more informations. I am also trying to train microsoft translator, but it takes very long (3 hours until now, and it's still processing the TM).
I'll check Prompt and let you know.
Have a nice day!


Direct link Reply with quote
 

Michael Joseph Wdowiak Beijer  Identity Verified
United Kingdom
Local time: 14:36
Member (2009)
Dutch to English
+ ...
PangeaMT + KantanMT Aug 1, 2014

Hi Bruno,

I suggest taking a look at these two:

http://pangeamt.com/en/# (Manuel Herranz)
http://www.kantanmt.com/index.php

I have heard good things about them both.

Michael

[Edited at 2014-08-01 12:52 GMT]

PS: I wouldn't give up on the Microsoft Translator Hub too soon. I haven't had time to test it yet, but it looks great (and it's free!). Microsoft also seem to be very eager to get freelance translators involved, which is about time if you ask me.

-> https://hub.microsofttranslator.com/SignIn?returnURL=/Home/Index

[Edited at 2014-08-01 12:56 GMT]


Direct link Reply with quote
 

Patrick Porter
United States
Local time: 09:36
Spanish to English
+ ...
training a stat MT engine can take a long time Aug 1, 2014

Bruno Depascale wrote:

...I am also trying to train microsoft translator, but it takes very long (3 hours until now, and it's still processing the TM)....


I don't have direct experience training engines on the MS translator hub, but the training step in general can take a lot of time. It depends on how long your training corpus is, both the number of segment pairs and the average length of each pair. It also depends on the amount of memory and processing power you have available.


I am looking for machine translation softwares (paying a fee) that can be trained with TMX and termbases


A free option is the open source Moses project. You need a linux machine to use it (I use a virtual machine on my Windows PC). The nice thing is that it's free and you get to keep all your data privately on your own machine. It's definitely more technically involved than an off-the-shelf solution, but the user guide is fairly easy to follow if you have basic familiarity with linux. Even if you don't want to use the program, the user guide contains a lot of information about the different types of stat MT in general and is probably worth reading.

I've had some success using Moses to train engines using my own TMs. To use the trained models in Trados, I've created my own Studio plugin. Although it's in a very early stage, I'm thinking of releasing a version of it on SDL's app store at some point in the future. If you would be interested in trying out a pre-release version of the plugin and giving feedback, email me through Proz and I'd be glad to send you a copy.


Direct link Reply with quote
 

Radian Yazynin  Identity Verified
Local time: 16:36
Member (2004)
English to Russian
+ ...
Thanks to Patrick Aug 1, 2014

Reading with interest about Moses, and have just found out it can also run on Windows under Cygwin. No idea if it's better than a virtual machine but am thinking about trying it out.

[Edited at 2014-08-01 15:18 GMT]


Direct link Reply with quote
 

Patrick Porter
United States
Local time: 09:36
Spanish to English
+ ...
Moses decoder under cygwin Aug 1, 2014

Radian Yazynin wrote:

Reading with interest about Moses, and have just found out it can also run on Windows under Cygwin. No idea if it's better than a virtual machine but am thinking about trying it out.


The Moses decoder works under cygwin and in fact that's how I use it after having trained models inside a Ubuntu virtual machine. Also, if I'm not mistaken, the project downloads include a pre-built cygwin binary of the decoder.

There might be an issue with training under cygwin, and I think the Moses user manual mentions something about this, or at least that it's not tested. In any event, I've never bothered to try training that way.


Direct link Reply with quote
 

Radian Yazynin  Identity Verified
Local time: 16:36
Member (2004)
English to Russian
+ ...
Thanks Patrick - Aug 1, 2014

for the advice! This is what I suspected.

Direct link Reply with quote
 

Bruno Depascale  Identity Verified
Italy
Local time: 15:36
Member (2009)
English to Italian
+ ...
TOPIC STARTER
Thank you for your inputs! Aug 2, 2014

I am trying to train microsoft translator, and the first stages took more than 16 hours!
I have now requested microsoft to "deploy" the results and it will take 2 other business days.
However, it is pretty complicated to set up microsoft translator in Trados 2014.
I have followed the tutorial found at the following address: http://www.translatum.gr/forum/index.php?topic=357465.0
According to such tutorial, a translator should first create an account in microsoft translator hub.
Then, after installed the Trados plugin, one should go to the Windows Azure Marketplace, where you need to sign in and register your Microsoft Translator application.
Here you will create/get your "Client ID" and "Client Secret".
Indeed, this last step is not quite clear, because I don't get any key named "Client Secret", but only 2 keys named client ID and primary client key.
Trados plugin requests to include a key named "Client ID" and another named "Client Secret".
I have inserted the Client ID, but since I don't know what the "Client Secret" is, I have inserted the Primary client key.
In such way, Trados plugin is activated, but I cannot get any MT results due to an error message in Trados.
I'll keep you updated about my progress with this tool.
In the meanwhile, I have emailed all the companies you have listed, in order to obtain more information from them.
However, for what concerns the Moses project, unfortunately, I don't have a linux environment, and don't want to spend too much time in installing it.
By the way, I have also discovered by chance that SDL is launching a new product: https://languagecloud.sdl.com/translators

[Edited at 2014-08-02 12:29 GMT]


Direct link Reply with quote
 

neilmac  Identity Verified
Spain
Local time: 15:36
Spanish to English
+ ...
Do the right thing Aug 4, 2014

No matter how "easy" you may think the medical texts are, my advice is to forego MT and do the translations yourself rather than seeking a cheat or shortcut. I certainly wouldn't want my medical records tended to by a machine.

Direct link Reply with quote
 

Bruno Depascale  Identity Verified
Italy
Local time: 15:36
Member (2009)
English to Italian
+ ...
TOPIC STARTER
should we go back to the '70s? Aug 4, 2014

neilmac wrote:

No matter how "easy" you may think the medical texts are, my advice is to forego MT and do the translations yourself rather than seeking a cheat or shortcut. I certainly wouldn't want my medical records tended to by a machine.


Dear Neilmac,
thank you for your contribution, which is really important to me, because gives me the opportunity to explain better my opinion about MT.
I don't think you are as narrow-minded as to really think that I am looking for a way to cheat my clients.
As I had already written in my first post, I am looking for a specialized MT, not a generic MT, like google translate. And I am willing to pay for it.
Such specialized MT software should be trained to an acceptable degree in the medical field; moreover, I should be able to train it to the phraseology and terminology I commonly use in my medical translations.
In my translation process, I use high quality devices and softwares, which speed up my translation process, like: Trados 2014, up-to-date dictionaries, DNS, a new Imac that doesn't freeze every now and then like windows computers and so on, to mention only a few. Apart from this, I type very quickly, because I have followed a typist course.
Would you say that all such things are a way to cheat my clients?
Following your reasoning, we should all go back in the 70s and use a typewriter instead of a computer. But I am sure I have misunderstood your reasoning.
MT is an instrument, and like every instrument it should be used in a thought out manner.
I like translating and don't want that an instrument takes my place.
I have a high reputation of my competences in the medical field, and I want to give my clients always 100% accurate and competent translations.
For such reasons, I think specialized MT could help me in optimizing my time, giving me more time to perform in-depth terminological searches and reducing the time spent on typing elemental sentences like: "Seventy two female undergraduate students with no prior experience in basketball volunteered to participate in the study".
So MT, like Trados or DNS, maybe could help me in reducing by 5% the time spent on an average job, but I could use that 5% time saved to improve the final document and to heighten the quality I can offer my clients.
These are the reasons I am considering to use specialized MT softwares in my workflow.
At the moment, I don't use any, because google translate and the like are free services and as such offer low-quality products, which don't fasten the translation process, but rather hinder it.
I hope I have explained myself clearly and look forward to receiving your opinions about it.
Thank you


Direct link Reply with quote
 
Oleg Vigodsky
Russian Federation
Local time: 16:36
English to Russian
Today there is no PROMT_Trados-2014 plugin Aug 6, 2014

Radian Yazynin wrote:

Hi Bruno,
Promt, as standalone software, can be used as a plug-in in Trados. You need to choose a proper version for yourself. By the way, plugging in MT and training it with translation memories within a CAT is really unnecessary because you would use a rules-based MT which uses TMs from CAT as is, while CAT would use such TMs itself, in this case nothing would change. Of course, there is another variant, a hybrid MT, which can use a statistical scheme, and this is where you can train the system, but such option is more expensive, though available too.
Good luck!


Hello, Radian.
Unfortunately, today there is no plugin between PROMT and SDL TRADOS 2009/2011/2014.

To integrate these tools into your translation workflow, you need to make some steps (see my workflow here:

http://argonaut.spb.ru/files/Trados_Studio_Promt_eng.pdf (in English)
http://argonaut.spb.ru/files/Trados_Studio_Promt.pdf (in Russian)

There is a hope, PROMT will release a specific plugin this year.


Direct link Reply with quote
 

Radian Yazynin  Identity Verified
Local time: 16:36
Member (2004)
English to Russian
+ ...
Yes, right Aug 6, 2014

Hi Oleg! Thanks for reminding me. I did read your files before, very interesting indeed. And yes, we must defferentiate between a plugin and (what I initially meant) adding MT via Option > Translation Memory and Automated Translation, which allows working in the Editor and only with a single file, without batch processing using MT.

Direct link Reply with quote
 

neilmac  Identity Verified
Spain
Local time: 15:36
Spanish to English
+ ...
DIY Aug 8, 2014

Bruno Depascale wrote:

I don't think you are as narrow-minded as to really think that I am looking for a way to cheat my clients.
As I had already written in my first post, I am looking for a specialized MT, not a ...




You are quite right - I was not suggesting you were trying to cheat your clients; I simply used "cheat" in the gaming sense, as a synonym of "shortcut".

However, we obviously have a different approach to translation. I tend to think there is too much reliance on doing everything automatically nowadays, so perhaps I do have slightly retro attitudes (technofear?) to the plethora of CAT tools around today, as, yes, I did grow up in the 70s. All I really meant to say is that if I was getting "easy" medical texts I'd usually be only too pleased to do them myself, "manually", without resorting to an automated solution, other than my customary WF Classic, which I try to use "sparingly"....


Direct link Reply with quote
 


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 specialized softwares

Advanced search






Across v6.3
Translation Toolkit and Sales Potential under One Roof

Apart from features that enable you to translate more efficiently, the new Across Translator Edition v6.3 comprises your crossMarket membership. The new online network for Across users assists you in exploring new sales potential and generating revenue.

More info »
SDL Trados Studio 2017 Freelance
The leading translation software used by over 250,000 translators.

SDL Trados Studio 2017 helps translators increase translation productivity whilst ensuring quality. Combining translation memory, terminology management and machine translation in one simple and easy-to-use environment.

More info »



Forums
  • All of ProZ.com
  • Term search
  • Jobs
  • Forums
  • Multiple search