Need help to understand how DeepMiner works with Google Translate Thread poster: Bjørnar Magnussen
|
|---|
Dear advanced Deja Vu users
I read the following on Atril's web page, which prompted me to download a trial version of Deja Vu:
"You can activate Google Translate on three different levels: Pretranslation, Assembling and specific segments in the translation grid. Thus it allows you to set DeepMiner to query MT systems including Google Translate for portions it does not find in your databases, extending its unprecedented ability to generate usable full translations even... See more Dear advanced Deja Vu users
I read the following on Atril's web page, which prompted me to download a trial version of Deja Vu:
"You can activate Google Translate on three different levels: Pretranslation, Assembling and specific segments in the translation grid. Thus it allows you to set DeepMiner to query MT systems including Google Translate for portions it does not find in your databases, extending its unprecedented ability to generate usable full translations even when it finds partial or no matches in your databases."
Then I did the following experiement:
Source text: If you enter your card details for the first time
GT: Hvis du skriver inn kortdetaljer for første gang
Termbase: card details = TEST
In my theory, this should produce something like:
DeepMiner: "Hvis du skriver inn TEST for første gang"
However, there's no way I can get this or anything like it no matter how much I play with the settings. Did I misunderstand how DeepMiner works with Google Translate? ▲ Collapse
| | | | | | | no info in video | Jan 11, 2012 |
Dear Epameinondas
Thanks for your answer. I don't find any information on how DeepMiner works with Google Translate in the video. Why do you think this video would be helpful?
| | | | | Yes an no... | Jan 11, 2012 |
Bjørnar Magnussen wrote:
I read the following on Atril's web page, which prompted me to download a trial version of Deja Vu:
"You can activate Google Translate on three different levels: Pretranslation, Assembling and specific segments in the translation grid. Thus it allows you to set DeepMiner to query MT systems including Google Translate for portions it does not find in your databases, extending its unprecedented ability to generate usable full translations even when it finds partial or no matches in your databases."
In theory, it's right but, IMO, DeepMiner is still at some beta stage.
I.e. it's very promising, really good for AutoWrite (i.e., translating to Trados dialect, AutoSuggest) but the Assemble is often poor.
Then I did the following experiement:
Source text: If you enter your card details for the first time
GT: Hvis du skriver inn kortdetaljer for første gang
Termbase: card details = TEST
In my theory, this should produce something like:
DeepMiner: "Hvis du skriver inn TEST for første gang"
However, there's no way I can get this or anything like it no matter how much I play with the settings.
In this case, i would use simply AutoAssemble without DeepMiner.
It should give better results if a similar sentence is found in the TM.
You can also simply hit Ctrl+G, then go to TEST, it's likely you'll get TEST from AutoWrite.
Did I misunderstand how DeepMiner works with Google Translate?
For a moment, forget it.
I.e. I like it a lot because it throws me a lot a "riddles" from my old translations but the real results may be dubious.
PS
Provide us some Norvegian target example from GT... some of us will be able to understand some chunks...
Try to ask the same question at [email protected]
You'll spot there some skilled Scandinavian guys...
Catspeed
GG
[Edited at 2012-01-11 18:28 GMT]
| | |
|
|
|
| Have the DV support people answer this question HERE | Jan 21, 2012 |
Hi Bjørnar,
I think this question is extremely interesting, and think you should contact DV support and provide them with the link to this question and URL, so that they can give a better explanation of this issue here. If it really works like you assume, Bjørnar, this is really ground braking technology, which would take DV further steps ahead of Trados Studio. But already DeepMiner is really cutting edge, exploring our previous translations automatically, saving a lot of work.... See more Hi Bjørnar,
I think this question is extremely interesting, and think you should contact DV support and provide them with the link to this question and URL, so that they can give a better explanation of this issue here. If it really works like you assume, Bjørnar, this is really ground braking technology, which would take DV further steps ahead of Trados Studio. But already DeepMiner is really cutting edge, exploring our previous translations automatically, saving a lot of work.
And I think the answer to this would be very interesting for all of us using DV, boosting our productivity.
Regards
S. H. Djaerff ▲ Collapse
| | | | | It's not how I understand it | Jan 21, 2012 |
Bjørnar Magnussen wrote:
Dear advanced Deja Vu users
I'm not an advanced user. In fact, I'm not a user at all. However, it's not how I understand it:
Bjørnar Magnussen wrote:
I read the following on Atril's web page, which prompted me to download a trial version of Deja Vu:
"You can activate Google Translate on three different levels: Pretranslation, Assembling and specific segments in the translation grid. Thus it allows you to set DeepMiner to query MT systems including Google Translate for portions it does not find in your databases, extending its unprecedented ability to generate usable full translations even when it finds partial or no matches in your databases."
As I understand it, this says DV will "repair" something found in the TM, or assembled by it from various bits, using Google Translate to query the missing bits. What you are trying to do is the opposite: taking a translation served by Google Translate and using DV's own resources (in your case, the term base) to "repair" the GT translation.
Bjørnar Magnussen wrote:
Then I did the following experiement:
Source text: If you enter your card details for the first time
GT: Hvis du skriver inn kortdetaljer for første gang
Termbase: card details = TEST
In my theory, this should produce something like:
DeepMiner: "Hvis du skriver inn TEST for første gang"
However, there's no way I can get this or anything like it no matter how much I play with the settings. Did I misunderstand how DeepMiner works with Google Translate?
The scenario covered by DeepMiner (again, as I understand it) is the following:
You have translated the following segment (now in your TM):
If you enter your card details for the first time
You need to translate the following new source segment:
If you enter your login information for the first time
Then DeepMiner would query the missing bit ("login information"), using GT.
| | | |
Bjørnar Magnussen wrote:
I read the following on Atril's web page, which prompted me to download a trial version of Deja Vu:
"You can activate Google Translate on three different levels: Pretranslation, Assembling and specific segments in the translation grid. Thus it allows you to set DeepMiner to query MT systems including Google Translate for portions it does not find in your databases, extending its unprecedented ability to generate usable full translations even when it finds partial or no matches in your databases."
(...)
However, there's no way I can get this or anything like it no matter how much I play with the settings. Did I misunderstand how DeepMiner works with Google Translate?
It's fixed now and basically should work as expected.
The former algorithm was somehow "leftie" i.e. the left part of a group of words was interpreted incorrectly, now the coverage is taken in account and MT fills the gaps.
Cheers
GG
| | | |
Unfortunately my Deja Vu trial license has expired. Could you provide some examples of how it works (can be English to Polish).
| | |
|
|
|
Bjørnar Magnussen wrote:
Unfortunately my Deja Vu trial license has expired. Could you provide some examples of how it works (can be English to Polish).
I just modified your example as:
Unfortunately my AAA BBB trial license has expired. Could you provide some CCC DDD of how it works (can be English to Polish).
I put in my termbase "AAA BBB" = "XXX YYY" and "CCC DDD" = "ZZZ".
The NB-NO result is:
Dessverre min XXX YYY prøvelisensen er utløpt. Kunne du gi noen ZZZ av hvordan det fungerer (kan være engelsk til polsk).
The original GT is:
Dessverre min Deja Vu prøvelisensen er utløpt. Kan du gi noen eksempler på hvordan det fungerer (kan være engelsk til polsk).
As I see, quite nice although it's some difference because of the length of the portions provided to Google i.e. Google has less context and translates more literally.
It's just the easiest example, now DVX is also able to replace strings not found in fuzzy matches using portions found by DeepMiner or provided by a machine translation engine.
Cheers
GG
[Edited at 2012-07-05 13:52 GMT]
| | | | | Thanks for the example! | Jul 5, 2012 |
Your NB-NO result with DVX termbase is indeed not worse than the original GT result!
This is really interesting. For a long time, I have felt that the future of translation lies in an integration between statistically-based MT, rules-based MT and user-specified termbases.
| | | | | Has anyone started using this feature for production? | Nov 27, 2012 |
I am considering buying Deja Vu in order to get access to this feature.
| | | | | Yes, depending of the project, language... | Nov 28, 2012 |
Bjørnar Magnussen wrote:
I am considering buying Deja Vu in order to get access to this feature.
Yes, depending of the project and of the language pair.
E.g. I tend to use GT more for my secondary languages and for smaller projects (no additional GT related costs when one restarts DVX i.e. when GT cache is cleared in DVX).
Generally, I use heavily GT but the "pure Google"suggestions are almost always unusable and/or the adaptation need too much work.
When Google fills the gaps between the portions taken from the databases, it's clearly better.
GT may also propose better solutions for portions found in the databases (via AutoWrite), e.g. when the subject differs.
Nonetheless, if you have very large TMs (e.g. DGT), the Assemble with GT is very often identical (or almost identical) to the Assemble without GT if DeepMiner is activated.
The funny fact is GT is sometimes better than DM 'cause DM often inserts incorrectly short Polish words as "się", "nie", "lub" (reflexive pronoun, nie, or) instead of articles (as Polish has no articles, DM is fooled and selects instead some frequent short word but it happens also in GT although it's less frequent...).
As the Assemble may be slow for this kind of environment with large TMs, I rather open the same project in two DVX instances and pretlanslate simultaneously.
Usually I make my first pretranslations without resolving portions not found in the TMs, I only select to repair the fuzzies with DM/GT.
I.e. you'll tune in your setup experimentally.
The Google suggestions are also used in AutoWrite, so it may speed up a lot the translation although sometimes DVX proposes too large chunks.
BTW, unlike most (all?) CAT tools I'm aware, you can invoke Google only for selected segments/portions (Ctrl+G, Ctrl+Shift+G) wjhen GT is disabled globally.
E.g. in Tradis Studio or memoQ GT can be enabled/disabled only globally.
PS
Wait for some sales 
Cheers
GG
| | | | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Need help to understand how DeepMiner works with Google Translate | Trados Business Manager Lite | Create customer quotes and invoices from within Trados Studio
Trados Business Manager Lite helps to simplify and speed up some of the daily tasks, such as invoicing and reporting, associated with running your freelance translation business.
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 
