Pages in topic:   [1 2] >
Google Translate. As good as it got?
Thread poster: Cilian O'Tuama

Cilian O'Tuama  Identity Verified
Local time: 23:06
German to English
+ ...
Apr 19, 2016

We're all familiar with GT and the 'quality' it spews out, though I can imagine it might have some merits in some scenarios as a provider of risky and very-rough-gist attempts at transsomething. And in that respect, it serves its understood purpose.

I'm wondering if the 'quality' situation has improved over the years? I mean, Gbooks have scanned and OCR'd so much stuff, surely there must be 'a source text of some sort with a good translation' somewhere in their database. I've been trying to find one in DE-ENG, and I've been making it easy for them by searching for the obvious prayers/quotations/poems, to little avail.

After all that scanning wouldn't you expect to come across a good GT proposal of appreciable length at least occasionally, of more than a few words?

Has anyone noticed any improvement over the years? What's the longest piece of text that GT has been known to 'process' well?

The reason I'm asking is that I saw a 3-line text and its 'translation' recently and thought to myself, a machine wouldn't do worse. Sure enough, it was GT.

Nite nite, C


 

Jean Lachaud  Identity Verified
United States
Local time: 17:06
English to French
+ ...
More complicated than that Apr 20, 2016

I have now and then used GT, mostly for legal texts.

There is no consistency. Some sentences may come out as needed little editing, then another apparently-close sentence will come out as gibberish.

Even very specialized technical terminology will, in a single document, come out different between one sentence and the next.

GT can't even spew out correct grammar in the target language. One would think that, into English, grammar wouldn't be a problem. Word order is all mixed up.

Originally, I had hoped to gain some time by using GT for the non-technical parts of sentences in what I translate, expecting to have only to deal with technical terms and some mild editing or re-ordering words, but the output is so bad in terms of syntax that not only did it not improve my productivity but made it worse: proofreading for overlooked incorrect words left over from GT required way too much time.

Every so often, I try it again, because I think that maybe one document or another may "work." Nope, it never does.

Even 98% TM matches (usually, a punctuation difference, or some spelling change) cause GT to produce a translation completely different from what it suggested for the original segment.

GT is hopeless in my language pairs (FR>EN & EN>FR).


 

Lincoln Hui  Identity Verified
Hong Kong
Local time: 05:06
Member
Chinese to English
+ ...
Works better for European languages, certainly Apr 20, 2016

I've fed a Norwegian Wikipedia article into it and got a reasonably plausible output. I also use it when I need to do a word-for-word translation of song lyrics from German, provided that I exercise discretion on which of the suggested translations I follow - basically a more flexible dictionary.

Now, GT with Chinese or Japanese - speakth of it thou shalt not.


 

neilmac  Identity Verified
Spain
Local time: 23:06
Spanish to English
+ ...
Yes/No Apr 20, 2016

Yes, that's as good as it got. No, it hasn't improved as far as I can see. I've been using it for a few years now on snippets of text (clauses, shortish sentences, names of institutions and organisations) and I think I've learnt to recognise the kind of thing that it may be able to handle reasonably accurately.

Sometimes the howlers it comes up with are good for a laugh - my favourite is when it turns a negative statement into a positive or vice versa, which happens quite a lot.

File under 'for entertainment value only'.


 

Balasubramaniam L.  Identity Verified
India
Local time: 02:36
English to Hindi
+ ...
For Hindi Apr 20, 2016

I mostly use it as a dictionary/thesaurus.

I don't think the quality has improved over the years.

The longest text that it generates accurately is just one word! More than one word confuses it terribly and it generates rubbish.

Also, it is completely foxed by idiomatic terms. It handles only simple nouns satisfactorily.

Here is an illustrative example of the translation of a Hindi idiom:

नौ-दो-ग्यारह होना (literally, become nine two eleven, which means, to scoot, or run away) is translated correctly when fed as it is, to "escape".

But when placed in a sentence, as
चोर नौ-दो-ग्यारह हो गया (the thief beat it), it outputs "The thief was nine-two-Eleven", (with the e of eleven capitalized), which is a literal word-for-word translation of the source, and just gibberish.

It also has great difficulty in correctly reading punctuation marks and diacritical annotations.

Here is an example of the reverse, ie., English to Hindi translation of a simple sentence this time:

The elephant is the largest land mammal.
हाथी सबसे बड़ा भूमि स्तनपायी है।

भूमि स्तनपायी is a literal translation of land mammal (land - भूमि (भूमि however also means Earth, which distinction GT has not been able to make), and mammal - स्तनपायी)

In this example, land mammal means "mammal living on land". However, GT has not been able to make this simple connection.

And, भूमि is never used in Hindi for land in this sense. The usual term would be स्थलजीवी.

This example illustrates the limitations of GT excellently - it handles technical terms (mammal here) well, but any word which is not used in its straight forward dictionary meaning, completely floors it.


 

Samuel Murray  Identity Verified
Netherlands
Local time: 23:06
Member (2006)
English to Afrikaans
+ ...
It got better for a while, a while ago Apr 20, 2016

Cilian O'Tuama wrote:
I'm wondering if the 'quality' situation has improved over the years?


For a while, it did, but then the quality went down a bit again. For my language combination, currently the quality is about 8/10 of the best quality that it ever used to be.

I took the first paragraph of your post, put it through Google Translate, and then made the absolute minimum edits required to fix the translation. It's not that bad, actually:



Added: it also misspelt "scenarios" in Afrikaans.

[Edited at 2016-04-20 08:33 GMT]


 

Daryo
United Kingdom
Local time: 22:06
Serbian to English
+ ...
two different cases Apr 20, 2016

(1) for someone who doesn't understand a world of the source language, Google translate can be EXTREMELY dangerous -

The problematic part is not the part of the translation that is patent nonsense, there you are aware that you are missing something, but you don't know what. THAT is not so bad.

The really nasty part is that part of the translation that sounds perfectly understandable and plausible. I've done several tests - (EN => FR and vice versa) and regularly you get bits of translated texts that sounds perfectly fine but are in fact completely wrong - nothing to do with the source text.

Now, if you don't understand a word of the source text, how are you going to know which part of the nicely sounding translation is the good one, and which one is totally wrong? You stand absolutely no chances of spotting it, and if you are really unlucky the nicely sounding mistranslation might be the key part of the whole document - the one you MUST get right.

You can as well try to rely on someone who is in the habit of randomly throwing porkies in his report ...

(2) For a translator, it can be useful as a kind of advanced dictionary, nothing much more. And even then, it's more a starting point for more serious research/verification. Calling GT output a "translation" is no more than marketing BS, at least for now.

Machine Translation certainly has the potential to get better and better, but as it is now it's no more that a product in development NOTHING MORE

[Edited at 2016-04-20 11:18 GMT]


 

Gabriele Demuth  Identity Verified
United Kingdom
Local time: 22:06
Member (2014)
English to German
What about style and target audience? Apr 20, 2016

I don't think it has learned to respond to that yet.

 

Dan Lucas  Identity Verified
United Kingdom
Local time: 22:06
Member (2014)
Japanese to English
You'd think German would be better Apr 20, 2016

Cilian O'Tuama wrote:
Has anyone noticed any improvement over the years? What's the longest piece of text that GT has been known to 'process' well?

I don't know, but I use GT to translate emails from the Zahlungspraxis email list and while it's good for getting the gist, it surprises me how poor the translation is. One would think that if GT was going to do well, it would be in a major European language pair like this. But it doesn't. It would need a lot of massaging to produce something useable in a professional context.

Incidentally, GT itself acknowledged a couple of years back that its statistical data mining approach had run out of steam. I expect some further slow improvements, but nothing dramatic.

Dan


 

Michael Beijer  Identity Verified
United Kingdom
Local time: 22:06
Member (2009)
Dutch to English
+ ...
haters gonna hate Apr 20, 2016

I actually find GT produces very useful stuff in my pair: Dutch to English.

As I translate in memoQ, I have MT output from …

1. Google Translate,
2. Bing Translator, and now my own custom
3. Slate Desktop (see: http://slate.rocks/ )

… in three little boxes next to my src/trgt boxes. There is always something useful in one of them, if not all.

Michael

PS: Slate Desktop is basically a Windows GUI for Moses, which allows you to build your own MT engines from your own TMs, on a Windows PC.


 

Michael Beijer  Identity Verified
United Kingdom
Local time: 22:06
Member (2009)
Dutch to English
+ ...
I suppose that's where Lilt comes into the picture … Apr 20, 2016

Dan Lucas wrote:

Cilian O'Tuama wrote:
Has anyone noticed any improvement over the years? What's the longest piece of text that GT has been known to 'process' well?

I don't know, but I use GT to translate emails from the Zahlungspraxis email list and while it's good for getting the gist, it surprises me how poor the translation is. One would think that if GT was going to do well, it would be in a major European language pair like this. But it doesn't. It would need a lot of massaging to produce something useable in a professional context.

Incidentally, GT itself acknowledged a couple of years back that its statistical data mining approach had run out of steam. I expect some further slow improvements, but nothing dramatic.

Dan


"Founders John DeNero and Spence Green met while working on Google Translate during the summer of 2011. They believed that better machine assistance could make translation more enjoyable for translators and more available for those who seek information. By 2014 a research prototype of an interactive, machine-assisted translation system called Predictive Translation Memory had been built. Lilt was incorporated in March 2015 with the mission of making fast, high-quality translation available to everyone."

(https://lilt.com/about )

Not that I'm actually using it, or know anything about it really. Just thought I'd mention it as I think it's one of Google's ways of evolving its MT offering.


 

Daryo
United Kingdom
Local time: 22:06
Serbian to English
+ ...
when, not if ... Apr 20, 2016

Gabriele Demuth wrote:

What about style and target audience?

I don't think it has learned to respond to that yet.


by the time MT gets to THAT point not only translators, but most of humankind will have become surplus to requirement.

But no point worrying about that today.

Deluded people imaging that GT is already good enough for real-life translation are more to be worried about now.


 

Michael Wetzel  Identity Verified
Germany
Local time: 23:06
German to English
closely related languages Apr 20, 2016

I would think that closely related languages, such as Dutch/Afrikaans with English or Spanish with Italian or with Portuguese, might produce significantly better results than other pairs.

I've never really experimented with GT, but German and English are just constructed in very different ways. Maybe the lack of cognates is not all that relevant for GT, but the words and sentences are constructed very differently.

I've always also assumed that translations out of English are difficult, because English syntax provides so little visible help. With the excepion of most adverbs, there is often no formal means to recognize whether an English word is an adjective, noun, or verb. English also fails to provide a lot of information about case, gender, etc. that is needed to produce a translation in another language.
I don't know about other languages, but English words also tend to have a whole range of meanings that goes way beyond what is typical for German.

Regardless of the absolute level of GT's quality, I also think that it is important to dispel the constantly repeated myth or mantra that GT is somehow getting better. Why would it be getting better? Does someone really believe that the statistical analysis of 20 mega-tetra-ultra-gigabytes of infomation will produce substantially better results than the analysis of 10?


PS: If anyone has a couple million worth of venture capital that they are looking to get off their hands, I can start up a start-up for them that is sure to be the next big thing.

PPS: (just saw Michael's posts) It's not along the lines of Lilt.

[Edited at 2016-04-20 10:58 GMT]


 

PAS  Identity Verified
Local time: 23:06
English to Polish
+ ...
Reversing the meaning Apr 20, 2016

I have found a significant number of instances where GT actually reverses the meaning of the translated text.

It will say "There are no parts available", where the original says "There are parts availabe."

I don't understand why it does it, but this follows the "rule" that GT sometimes messes up seemingly simple sentences. Now, does this mess things up or what?

Funny though, looking at the responses, how many of us here actually do use this "toy", knowing its fallacies.

I use GT when I have no time to do a normal translation. I google the text and then clean it up so it's "fit for purpose". BUT, I do this for internal purposes only (I work in-house), never for a client.


 

Lingua 5B  Identity Verified
Bosnia and Herzegovina
Local time: 23:06
German to Serbian
+ ...
Dangerous, more dangerous, most dangerous. Apr 20, 2016

PAS wrote:

I have found a significant number of instances where GT actually reverses the meaning of the translated text.

It will say "There are no parts available", where the original says "There are parts availabe."

I don't understand why it does it, but this follows the "rule" that GT sometimes messes up seemingly simple sentences. Now, does this mess things up or what?

Funny though, looking at the responses, how many of us here actually do use this "toy", knowing its fallacies.

I use GT when I have no time to do a normal translation. I google the text and then clean it up so it's "fit for purpose". BUT, I do this for internal purposes only (I work in-house), never for a client.


If the error twisted the meaning into "no meaning", it is less dangerous than when it twists it into a new/different meaning, especially if that new meaning could actually (logically) fit into the context. When I receive non-native English source texts, those gems are really the most dangerous, as they can mislead you easily.

And back to the original question, they will polish and improve the output once they invest in linguistic projects (ie. expert linguists and linguist-consultants). For some languages they already do (I am surprised about bad reports re. German), while for some other languages that don't have any market or commercial value for them, they don't care to do it at this point.


 
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 »

Google Translate. As good as it got?

Advanced search







SDL Trados Studio 2019 Freelance
The leading translation software used by over 250,000 translators.

SDL Trados Studio 2019 has evolved to bring translators a brand new experience. Designed with user experience at its core, Studio 2019 transforms how new users get up and running and helps experienced users make the most of the powerful features.

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 »



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