Example of neural MT versus human translation
Thread poster: Jeff Whittaker
Jeff Whittaker
Jeff Whittaker  Identity Verified
United States
Local time: 22:21
Member (2002)
Spanish to English
+ ...
Apr 2, 2019

Google Translate
https://translate.google.com/#view=home&op=translate&sl=auto&tl=en&text=Zutrittskontrolle,%20Zugangskontrolle,%20Zugriffskontrolle
... See more
Google Translate
https://translate.google.com/#view=home&op=translate&sl=auto&tl=en&text=Zutrittskontrolle,%20Zugangskontrolle,%20Zugriffskontrolle

DeepL:
https://www.deepl.com/en/translator#de/en/Zutrittskontrolle,%20Zugangskontrolle,%20Zugriffskontrolle

Human (from 13 years ago):
https://www.proz.com/kudoz/german-to-english/computers-systems-networks/1589142-zutrittskontrolle-zugangskontrolle-zugriffskontrolle.html

[Edited at 2019-04-02 16:55 GMT]
Collapse


 
Jean Lachaud
Jean Lachaud  Identity Verified
United States
Local time: 22:21
English to French
+ ...
Shocking! Apr 2, 2019

To be filled under "What else is new?"

 
Maria Teresa Borges de Almeida
Maria Teresa Borges de Almeida  Identity Verified
Portugal
Local time: 02:21
Member (2007)
English to Portuguese
+ ...
Interesting... Apr 2, 2019

Thanks!

 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 03:21
Member (2006)
English to Afrikaans
+ ...
@Legal Apr 2, 2019

LegalTranslatr2 wrote:
Google Translate...
DeepL...
Human (from 13 years ago)...


So, the MT systems got it right. The human in this example desired a translation for a very specific application, and so obviously the translation he might use would be of a specialist nature. Google and DeepL give you the translation for generic purposes. All three these terms mean "access control" and it is likely that translations of texts that use these terms would all use "access control" in English. Only someone who doesn't know anything about languages will believe that three distinct terms in German should always have three distinct equivalents in English. Don't you agree? We are not terminologists. We are translators.

[Edited at 2019-04-02 16:54 GMT]


 
Jeff Whittaker
Jeff Whittaker  Identity Verified
United States
Local time: 22:21
Member (2002)
Spanish to English
+ ...
TOPIC STARTER
Neural MT Apr 2, 2019

You are focusing too much on cutting down the specific example and missing the point. Perhaps this is true if these terms were used in isolation, but all three terms are used in EU-mandated Data Protection documents. Is the difference significant? That's beside the point. The overall point is that MT translation will only take you so far and only humans can know when and if further distinction is required. The monolingual reader of the MT output will be left without this choice.

Samuel Murray wrote:

LegalTranslatr2 wrote:
Google Translate...
DeepL...
Human (from 13 years ago)...


So, the MT systems got it right. The human in this example desired a translation for a very specific application, and so obviously the translation he might use would be of a specialist nature. Google and DeepL give you the translation for generic purposes. All three these terms mean "access control" and it is likely that translations of texts that use these terms would all use "access control" in English. Only someone who doesn't know anything about languages will believe that three distinct terms in German should always have three distinct equivalents in English. Don't you agree? We are not terminologists. We are translators.

[Edited at 2019-04-02 16:54 GMT]


[Edited at 2019-04-02 17:01 GMT]

[Edited at 2019-04-02 17:01 GMT]

[Edited at 2019-04-02 17:02 GMT]


 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 03:21
Member (2006)
English to Afrikaans
+ ...
@Legal Apr 2, 2019

LegalTranslatr2 wrote:
You are focusing too much on cutting down the specific example and missing the point. ... The overall point is that MT translation will only take you so far...


I did not realise that that was the point you were trying to make. Did you think it was obvious? (-:


 
Heinrich Pesch
Heinrich Pesch  Identity Verified
Finland
Local time: 04:21
Member (2003)
Finnish to German
+ ...
Very nice exemple Apr 3, 2019

It shows the semantic strength of German. It distinguishes between taking a step into somewhere, moving somewhere or take a grip at something.
But there is no reason why MT should not learn to understand the meanings and apply appropriate translations.


Jeff Whittaker
Jorge Payan
 
Philip Lees
Philip Lees  Identity Verified
Greece
Local time: 04:21
Greek to English
Context Apr 3, 2019

Samuel Murray wrote:

LegalTranslatr2 wrote:
Google Translate...
DeepL...
Human (from 13 years ago)...


So, the MT systems got it right.


The MT systems don't appear to have been given the extra information that was provided to the human translator: namely, that three different target phrases were required for the three terms.

The ProZ asker acknowledges that humans would probably have come up with the same result as the MT if he had adhered strictly to KudoZ rules.

I'm not a fan of MT either, but this is hardly a fair comparison.



[Edited at 2019-04-03 05:28 GMT]


Giovanni Guarnieri MITI, MIL
Jorge Payan
 
Jeff Whittaker
Jeff Whittaker  Identity Verified
United States
Local time: 22:21
Member (2002)
Spanish to English
+ ...
TOPIC STARTER
This example Apr 3, 2019

Of course, this data, for this particular example, could be entered into the system (but it hasn't been in 12 years). However, there will always be things like this and it would be impossible to cover every situation.

That's the problem with pointing out defects in MT, as soon as you try to come up with an example as to why the system won't work, people attack the specific example and how the system could be tweaked to rectify this particular error, rather than the big picture that
... See more
Of course, this data, for this particular example, could be entered into the system (but it hasn't been in 12 years). However, there will always be things like this and it would be impossible to cover every situation.

That's the problem with pointing out defects in MT, as soon as you try to come up with an example as to why the system won't work, people attack the specific example and how the system could be tweaked to rectify this particular error, rather than the big picture that there are an unlimited number of situations where the system does not have the data or real-world experience to come up with a suitable translation.



Philip Lees wrote:

The MT systems don't appear to have been given the extra information that was provided to the human translator: namely, that three different target phrases were required for the three terms.

The ProZ asker acknowledges that humans would probably have come up with the same result as the MT if he had adhered strictly to KudoZ rules.

I'm not a fan of MT either, but this is hardly a fair comparison.



[Edited at 2019-04-03 05:28 GMT]
Collapse


Kaspars Melkis
 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 03:21
Member (2006)
English to Afrikaans
+ ...
@Legal Apr 3, 2019

LegalTranslatr2 wrote:
That's the problem with pointing out defects in MT, as soon as you try to come up with an example as to why the system won't work, people attack the specific example...


I think the problem here was that you did not make any point yourself, but simply gave three examples, and left it to us to draw our conclusions about what it was that you were hoping we would discuss. Had I known at the time when I read your first post what point you were trying to make, my reply would have been different.


Wilsonn Perez Reyes
Jorge Payan
 
Philip Lees
Philip Lees  Identity Verified
Greece
Local time: 04:21
Greek to English
Better examples Apr 4, 2019

LegalTranslatr2 wrote:
That's the problem with pointing out defects in MT, as soon as you try to come up with an example as to why the system won't work, people attack the specific example and how the system could be tweaked to rectify this particular error


Then it would perhaps be better to use examples that really do illustrate the inherent defects of MT, rather than ones that don't.

They're not difficult to find, after all. There are some in another forum thread, along with a discussion of MT versus human errors.



[Edited at 2019-04-04 06:27 GMT]


Thomas Pfann
 


To report site rules violations or get help, contact a site moderator:


You can also contact site staff by submitting a support request »

Example of neural MT versus human translation







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 »
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 »