Pages in topic:   < [1 2 3 4 5 6 7] >
DeepL
Thread poster: Michael Mangelsdorf

Cilian O'Tuama  Identity Verified
Local time: 23:27
German to English
+ ...
And I wasn't even trying to trick it... Sep 1, 2017

In the one and only test that I did, with a randomly chosen sentence from wikipedia, Google Translate was actually better than DeepL.

Wikipedia:
Im März 1995 wurde den beiden Forschern S und H ... ein Patent ... erteilt.

DeepL:
In March 1995, S and H granted a patent ....
(should be “were granted”, not “granted” - pretty serious IMO)

Google Translate got that bit right:
In March 1995, two researchers S and H wer
... See more
In the one and only test that I did, with a randomly chosen sentence from wikipedia, Google Translate was actually better than DeepL.

Wikipedia:
Im März 1995 wurde den beiden Forschern S und H ... ein Patent ... erteilt.

DeepL:
In March 1995, S and H granted a patent ....
(should be “were granted”, not “granted” - pretty serious IMO)

Google Translate got that bit right:
In March 1995, two researchers S and H were granted a patent....

So maybe it's swings and roundabouts...
Collapse


 

Tom in London
United Kingdom
Local time: 22:27
Member (2008)
Italian to English
swings and roundabouts Sep 1, 2017

Schaukeln und Kreisel? That's what DeepL thinks!

[Edited at 2017-09-01 14:39 GMT]


 

Tom in London
United Kingdom
Local time: 22:27
Member (2008)
Italian to English
New balls Sep 2, 2017

Amusingly, if you input "Rompicoglioni" into DeepL, it returns "replace them".



 

Robert Rietvelt  Identity Verified
Local time: 23:27
Member (2006)
Spanish to Dutch
+ ...
Mixed feelings Sep 2, 2017

Let me start by saying that I am amazed about the development of translation technology, and albeit not perfect yet, it is here to stay.

I also played around with DeepL, and what I noticed is that the more 'emotions' in a text, the less the performance. It scored way less on subjects as Tourism and/or Marketing then it did on IT and/or Legal.

Furthermore (imho) it produces flat texts without a recognisible style and the results in all cases needed editing.

... See more
Let me start by saying that I am amazed about the development of translation technology, and albeit not perfect yet, it is here to stay.

I also played around with DeepL, and what I noticed is that the more 'emotions' in a text, the less the performance. It scored way less on subjects as Tourism and/or Marketing then it did on IT and/or Legal.

Furthermore (imho) it produces flat texts without a recognisible style and the results in all cases needed editing.

Like somebody put it, DeepL repeats words like a parrot, but doesn't know what they mean.

So till sofar a nice extra tool, nothing more, nothing less. In the future? Who knows.






[Edited at 2017-09-02 12:01 GMT]

[Edited at 2017-09-02 12:02 GMT]
Collapse


 

Laurent Chevrette  Identity Verified
Canada
Member (2011)
English to French
Maybe they won't be any API :\ Sep 5, 2017

CafeTran Training wrote:

Let's wait for the API. I'm eager to have this new MT system integrated in my favourite CAT tool .


 

Jean Dimitriadis  Identity Verified
France
Local time: 23:27
Member
English to French
+ ...
API already announced Sep 5, 2017

It's been announced: https://www.deepl.com/press.html

Do you have information to the contrary?


 

Tom in London
United Kingdom
Local time: 22:27
Member (2008)
Italian to English
API or no API, DeepL is useless Sep 6, 2017

I've been using DeepL over the past few days, to see if it can help with a long translation I'm doing (a rather tedious academic report about how a particular university course was organised, and how the students performed).

The language of the source document is very flat and straightforward, not particularly well-written, and somewhat impenetrable due to its tediousness and inelegance. There is no stylistic complexity and no technical terminology is used.

I'm finding
... See more
I've been using DeepL over the past few days, to see if it can help with a long translation I'm doing (a rather tedious academic report about how a particular university course was organised, and how the students performed).

The language of the source document is very flat and straightforward, not particularly well-written, and somewhat impenetrable due to its tediousness and inelegance. There is no stylistic complexity and no technical terminology is used.

I'm finding that as in the case of Google Translate, going through a document with DeepL paragraph by paragraph, sentence by sentence, Deepl can be useful as a first hit, for the simple reason that less mental effort is required to grasp the general meaning of a very boring document without having to think about it too hard. But apart from that initial advantage, extensive rewriting/rephrasing is then required to make a presentable translation. And there are also mistakes in the MT - some of them quite serious.

So in the case of this simple, straightforward source text I don't find any significant difference between the performance of GT and that of DeepL. Both are equally useful/useless. And a text of this type should be the easiest for DeepL, because it's very mechanical. Heaven knows how DeepL would manage with anything more complicated.

I predict that there will soon be an avalanche of future "proofreading and editing jobs" of translations that an agency has just run through DeepL....

[Edited at 2017-09-06 08:39 GMT]
Collapse


 

Jean Dimitriadis  Identity Verified
France
Local time: 23:27
Member
English to French
+ ...
Usefulness variations Sep 6, 2017

Seemingly straightforward texts can pose serious problems to MT engines, especially if they suffer from ambiguous or clumsy phrasing, weird syntax choices and long-winded periods. No wonder significant use of MT requires not only post-editing, but also pre-editing, controlled natural language writing and subject/corpus specific MT training.

From the point of view of a translator who uses a general MT engine as another resource within his CAT environment (and not an MT post-editor wh
... See more
Seemingly straightforward texts can pose serious problems to MT engines, especially if they suffer from ambiguous or clumsy phrasing, weird syntax choices and long-winded periods. No wonder significant use of MT requires not only post-editing, but also pre-editing, controlled natural language writing and subject/corpus specific MT training.

From the point of view of a translator who uses a general MT engine as another resource within his CAT environment (and not an MT post-editor who requires a different mindset as he needs to work directly with both the source text and the MT output), the assessment of its usefulness can vary wildly between projects, depending on several factors. It's up to the individual translator (or should I say linguist ?) to manage his different resources per project in a way that helps him be efficient in his work.

In a recent project, I found DeepL's usefulness to be clearly superior to other MT engines, but I'm pretty sure its performance is not consistent and depends on the source material. For now, I consider it as a useful addition to the already available options. An API will make things easier in that respect.

And, selecting a word in DeepL's target box lets you access translation variations, which may be helpful when reviewing the rephrasing options available in the target language.

Jean
Collapse


 

Robert Rietvelt  Identity Verified
Local time: 23:27
Member (2006)
Spanish to Dutch
+ ...
Exactly! Sep 6, 2017

Tom in London wrote:

I predict that there will soon be an avalanche of future "proofreading and editing jobs" of translations that an agency has just run through DeepL....

[Edited at 2017-09-06 08:39 GMT]


That is what I am afraid of too. Not so much about the "proofreading and editing jobs", they are part of our profession, but about the offered rates!


 

Susan Welsh  Identity Verified
United States
Local time: 17:27
Member (2008)
Russian to English
+ ...
wrong forum Sep 9, 2017

This should be moved to the Machine Translation forum, where more users who are interested would find it.

 

Ligeti  Identity Verified
Netherlands
Local time: 23:27
English to Dutch
+ ...
Breakthrough Sep 11, 2017

Georgie Scott wrote:

"Nonetheless, in my language pairs I still produce a faster, better translation on my own.

Perhaps it will push all the lousy translators out of the bottom of the market and project managers will stop telling me they know lots of people who'll translate for 5 cents a word.


It will push you out of the market as well. You just dont know it yet. This is what is called a breakthrough.


 

Tom in London
United Kingdom
Local time: 22:27
Member (2008)
Italian to English
In your dreams Sep 11, 2017

Maurice Koopman wrote:

Georgie Scott wrote:

"Nonetheless, in my language pairs I still produce a faster, better translation on my own.

Perhaps it will push all the lousy translators out of the bottom of the market and project managers will stop telling me they know lots of people who'll translate for 5 cents a word.


It will push you out of the market as well. You just dont know it yet. This is what is called a breakthrough.



I congratulate you on your amazing ability to see into the future.


 

Shouguang Cao
China
Local time: 06:27
English to Chinese
+ ...
CAT tool makers are eager Sep 12, 2017

DeepL creates such a fuss, that CAT tool makers are eager to integrate it now, even before the official API is released.

 

Tom in London
United Kingdom
Local time: 22:27
Member (2008)
Italian to English
Why? Sep 12, 2017

Shouguang Cao wrote:

DeepL creates such a fuss, that CAT tool makers are eager to integrate it now, even before the official API is released.


I don't see why. I've been experimenting with it for a week now, on a couple of long translations I'm doing, and finding it ultimately useless.

At first it seems good in the sense that it seems accurate, but when you go through the text you realise that it has completely failed to reproduce the "movement" and "intention" of the source text (two characteristics that no algorhythm can perceive) and that it chooses the most obvious words every time, rather than the *best* words. It reads, searches, selects, and replaces, but it does not *understand what it is reading*.


 

CafeTran Training (X)
Netherlands
Local time: 23:27
Use the resources that you consider to be useful Sep 12, 2017

Shouguang Cao wrote:

DeepL creates such a fuss, that CAT tool makers are eager to integrate it now, even before the official API is released.


Why wouldn't one use the available resources that can be useful for a certain language combination and a specific subject field?

E.g. when translating manuals, the aligned corpus of DeepL (Linguee) can give very nice terminological pointers. Just don't expect the translation to be correct. That's where we pros come in.


 
Pages in topic:   < [1 2 3 4 5 6 7] >


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


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

DeepL

Advanced search






CafeTran Espresso
You've never met a CAT tool this clever!

Translate faster & easier, using a sophisticated CAT tool built by a translator / developer. Accept jobs from clients who use SDL Trados, MemoQ, Wordfast & major CAT tools. Download and start using CafeTran Espresso -- for free

More info »
WordFinder Unlimited
For clarity and excellence

WordFinder is the leading dictionary service that gives you the words you want anywhere, anytime. Access 260+ dictionaries from the world's leading dictionary publishers in virtually any device. Find the right word anywhere, anytime - online or offline.

More info »



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