How a German company (DeepL) managed to be better than Google
Iniziatore argomento: Hans Lenting

Hans Lenting  Identity Verified
Paesi Bassi
Membro (2006)
Da Tedesco a Olandese
Jun 13

http://www.spiegel.de/plus/deepl-der-deutsche-unternehmer-ist-besser-als-google-a-00000000-0002-0001-0000-000157181383

(German, paid)

[Subject edited by staff or moderator 2018-06-13 19:28 GMT]


 

Daniel Frisano
Monaco
Local time: 01:41
Membro (2008)
Da Inglese a Italiano
+ ...
Anyone tried Pro? Jun 13

Can you do this directly in Excel with DeepL Pro?

Cell A1 = text
Cell B1 = text translated via DeepL

(Without having to copy back-and-forth, of course. That I can do with Goo).


 

Gerard de Noord  Identity Verified
Francia
Local time: 01:41
Membro (2003)
Da Tedesco a Olandese
+ ...
Don't send us to a paid article Jun 13

Don't send us to a paid article, DeepL deserves better.

Cheers,
Gerard


 

Jean Dimitriadis  Identity Verified
Francia
Local time: 01:41
Membro (2015)
Da Inglese a Francese
+ ...
DeepL Pro Jun 13

Hello Daniel,

Apart from an optimized privacy conscious Web translator (texts are not stored online), DeepL Pro - https://www.deepl.com/pro.html - offers API access.

Note that Google Translate and Microsoft Translator also include API access.

Your CAT tool may already have a connector for DeepL’s API.

With some translation software, you can take Excel cell A1, and export its translation to B1, and so forth.

Example: I know for a fact that CafeTran offers both a DeepL Pro connection and an Excel filter with that export capability.

With the API access, you may also be able to create or use a script (or something similar, depending on your OS) to take advantage of the clipboard while querying the API.

Otherwise, for a ready-to-use solution, you might want to check out GT4T - https://gt4t.net/en/ : It brings MT engines (including DeepL) at your fingertips, irrespective of the application you use, as long as you work on a Windows platform (which is not my case).

Jean

[Edited at 2018-06-13 20:59 GMT]


 

Hans Lenting  Identity Verified
Paesi Bassi
Membro (2006)
Da Tedesco a Olandese
AVVIO ARGOMENTO
Paper edition Jun 14

Gerard de Noord wrote:

Don't send us to a paid article, DeepL deserves better.

Cheers,
Gerard


Too bad that it’s hidden behind the paywall. Will scan the printed edition for any interesting quotes. To be continued...


 

Daniel Frisano
Monaco
Local time: 01:41
Membro (2008)
Da Inglese a Italiano
+ ...
Tried Jun 15

OK, I gave DeepL a honest try for DE>IT and it's thumbs way down. A few inexplicable issues:

- Omission of whole phrases, like 10 or 12 words at a time.

- Inconsistent terminology, i.e. same technical term translated in 3 or 4 different ways.

- Mixing definite and indefinite articles (how hard is it to get "der" and "ein" right?).

- Terms in English, as DE>IT is still done with DE>EN>IT.

All in all, to produce the same result it took at least the same time/effort or more.

EN>IT was perhaps a bit better, but still with heavy editing needed that took almost as long as translating anew.

If an artificial intelligence translates within the same (patent!) document 40 times "Bauteil" as "componente" and 2 or 3 times as "parte", that sounds more like artificial dumbness.

Conclusion, humans are safe (for now, and provided that they use natural intelligence).

[Edited at 2018-06-15 04:56 GMT]


 

Hans Lenting  Identity Verified
Paesi Bassi
Membro (2006)
Da Tedesco a Olandese
AVVIO ARGOMENTO
It's not about DeepL taking over from us professionals ... Jun 15

It's not about DeepL taking over from us professionals: it's about how we, qualified and experienced translators, can benefit from the typing aid and the nice trouvailles that it offers. Nice wordings that other human beings in the past have created and that were harvested, processed and made available by DeepL. No need to panic, just relax and enjoy how elegantly others have formulated. I'm not afraid to admit that (especially) DeepL has made my daily work even nicer (and it was already very nice).

Don't expect DeepL to translate perfectly, just harvest as much as you can from its suggestions and enjoy DeepL's lucky finds. Use it as a typing aid that also offers translations for long words that miss one letter etc. and that aren't recognised by your own term base.

Daniel Frisano wrote:

OK I gave DeepL a honest try for DE>IT and it's thumbs way down. A few inexplicable issues:

- Omission of whole phrases, like 10 or 12 words at a time.


Never experienced that in my language pair.

- Inconsistent terminology, i.e. same technical term translated in 3 or 4 different ways.


It's you who is the captain on your ship. You decide which one of the offered (interesting?) alternatives you'll use. Choose the right one, and use it. Save it for future reference: like you said, chances are that these perfect suggestions/nice wordings aren't offered in the next segment to be translated, take it from your own glossary then, that you've quickly built with DeepL's help. Use DeepL to improve your style (yes, I'm really saying that).

On a side note: A really smart CAT tool, like CafeTran Espresso 2018, can override DeepL's suggestion with your preferred term.


- Mixing definite and indefinite articles (how hard is it to get "der" and "ein" right?).


I translate from German 'all day' and as far as I remember, I never experienced this. (I'm not trying to say that you did experience this problem, I'm just stating my experience.)

All in all, to produce the same result it took at least the same time and effort or more.


For me DeepL is a real time saver and a smart typing aid.


EN>IT perhaps a bit better, but still with heavy editing needed that took almost as much time as translating anew.


This might be caused by the fact that DeepL translates everything to English first. Which can lead to very 'funny' (okay, at least on the first 1000 occasions) translations, where 'Dichtungen' are translated with 'seals' (in the sense of sea animals). But nothing easier than to prevent this occasional stupid glitch (CafeTran!) that indeed can blur the positive image.


If an artificial intelligence translates within the same (patent!) document 40 times "Bauteil" as "componente" and 2 or 3 times as "parte", that's more like artificial dumbness.


Use a CAT tool that allows you to prevent this nonsense.

For me, DeepL and CafeTran Espresso 2018 is a winning combination.

[Edited at 2018-06-15 05:12 GMT]

Edit: I forgot to mention that if DeepL occasionally offers a suggestion that sounds very strange to you, you'd better investigate this suggestion on the web. Chances are very high that this (in your ears) strange sounding translation is ... indeed the correct translation for your source term. I've had this nice experience several times now, mostly related to terms from EU directives.

So, DeepL can be a detective and investigation aid too icon_smile.gif.

[Edited at 2018-06-15 05:22 GMT]


 

Daniel Frisano
Monaco
Local time: 01:41
Membro (2008)
Da Inglese a Italiano
+ ...
No thanks Jun 15

As a true language lover, this is what pisses me the most:

joxkakq8gjvfaujyrjmg.jpg

I mean turning a subjunctive into a lame indicative because AD (artificial dumbness) apparently can't be bothered with verb moods. Come on, even stupid Google does better:

zlszprluntd3la9mnsqx.jpg

For good measure I'll stay away from all of this, lest I get slowly sucked into such mediocrity...


 

Hans Lenting  Identity Verified
Paesi Bassi
Membro (2006)
Da Tedesco a Olandese
AVVIO ARGOMENTO
For clarity Jun 15

Daniel Frisano wrote:

As a true language lover, this is what pisses me the most:

joxkakq8gjvfaujyrjmg.jpg

I mean turning a subjunctive into a lame indicative because AD (artificial dumbness) apparently can't be bothered with verb moods. Come on, even stupid Google does better:

zlszprluntd3la9mnsqx.jpg

For good measure I'll stay away from all of this, lest I get slowly sucked into such mediocrity...



I do understand what you are saying (although I don't speak Italian). What I was trying to say is that you shouldn't expect perfect sentences from DeepL (or from any other MT system). But that you should use it on a word level. Then it can be a great help. And still, you stay in control.


 

Daniel Frisano
Monaco
Local time: 01:41
Membro (2008)
Da Inglese a Italiano
+ ...
Doesn't get much dumber than this Jun 15

Here's an indefinite article turning definite:

wnlrnxm0f0lnuvr9t9eg.jpg

Articles are the most common words in most languages (at least languages that have articles), and they can't even get THOSE right? Again, even Google does better.

Note also that "Detektieren" and "Erkennung" are both rendered as "rilevamento". This smells a lot like laziness and clients HATE it, and justly so. Why do I need a tool that keeps giving me confusing suggestions?

I'll stop here, but I could go on all day.

[Edited at 2018-06-15 06:56 GMT]


 

Stepan Konev  Identity Verified
Russia
Local time: 02:41
Da Inglese a Russo
Sounds ridiculous Jun 15

How DeepL can be better if it only supports 7 languages?
Google supports 100+
Better for whom?

May be this is reason why your dsicussion of the best ever MT engine is that extremely limited in terms of participants and generally sluggish? No?

Not interesting.


 

Hans Lenting  Identity Verified
Paesi Bassi
Membro (2006)
Da Tedesco a Olandese
AVVIO ARGOMENTO
Only 7 but more to come Jun 15

Stepan Konev wrote:

How DeepL can be better if it only supports 7 languages?


Because of the quality of these 49 language pairs?

Better for whom?


For those who notice the quality of the suggestions? Probably because they work in the domain where DeepL harvested its source texts (www.linguee.com)?


 

Phil Hand  Identity Verified
Cina
Local time: 07:41
Da Cinese a Inglese
If only MT tools actually worked this way... Jun 18

Hans: "...the quality of the suggestions...use it on a word level..."

I would very much like a tool that did this - offered the power of big data and/or AI on the word level, without presenting the results as "sentences" in the target language. Unfortunately MT has to present something to casual users/gisters that looks a bit like a natural text. That's a very useful function when you're trying to get a general understanding of something in a language you don't know. But I've yet to find it any help at all when doing a precise translation from a language that I do know.

I remain in the MT skeptics camp, though I admit I was impressed by the naturalness of the English sentences when I used DeepL to help me understand a bit of German background.


 

Hans Lenting  Identity Verified
Paesi Bassi
Membro (2006)
Da Tedesco a Olandese
AVVIO ARGOMENTO
Some quotes Jun 19

Hans Lenting wrote:
Will scan the printed edition for any interesting quotes. To be continued...


mddoejgyimpfkl8d4dy8.pnghlpnkoa7lfyaly0bwold.pngtx5ycld7t0ggdtunhgak.png


 


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


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

How a German company (DeepL) managed to be better than Google

Advanced search






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 »
BaccS – Business Accounting Software
Modern desktop project management for freelance translators

BaccS makes it easy for translators to manage their projects, schedule tasks, create invoices, and view highly customizable reports. User-friendly, ProZ.com integration, community-driven development – a few reasons BaccS is trusted by translators!

More info »



Forums
  • All of ProZ.com
  • Cerca un termine
  • Lavori
  • Forum
  • Multiple search