Pages in topic:   < [1 2 3 4 5] >
What do you think about DeepL?
Thread poster: kakapo77
kakapo77
kakapo77  Identity Verified
France
Local time: 15:42
French to Italian
+ ...
TOPIC STARTER
Well, the day you have the understanding, you will say goodbye to any kind of human job... Sep 11, 2017

Kay-Viktor Stegemann wrote:

kakapo77 wrote:

Hello Maurice,

Thanks for your reply. I totally agree with you, especially when you say that many comments on DeepL - I am not necessarily speaking of the ones included in this thread, I am speaking by a general point of view - are incredibly naive.

I mean, let's face it, this tool works in an excellent way and it outperforms a big chunk of the translators on Proz. Just take a look to the terminology section, and you will see that a lot of people need help about really basic stuff, as if they were not even translators.

In my opinion, most translators only have a humanistic background, which leads them to think that translation will never be mastered by computers, and that only human beings will ever be able to provide good results. But this is just a sort of cognitive bias.

I am sure that poetry and literature need humans, that is obvious. But this kind of translation just represents a part of the market. I guess that most people here work on technical stuff, and I think that this type of texts can already be largely automated. Only limited post-editing is be required.

Regards,

Marco


Marco, I believe that the naivete is on the side of the MT-believers. (Let me add that I have a technical and IT background, not a "humanistic" one, at least not in a professional sense ...)

The claim that DeepL "outperforms a big chunk of Proz translators" is just that, a claim. I think that this is far from the truth. Please prove it. I can only say that in my pair DeepL outperforms other MT engines only, which is a good thing in itself, but the DeepL results are still a far cry from anything a native translator would produce.

And the reason for the still lousy results of DeepL are simple. What DeepL does is not translating. The process of translating includes the understanding of the source text and the reproduction of the meaning in the target text. No MT engine yet is able to do this. There is no such thing as artificial intelligence, which would be required for this. (Note that not even the Turing test, a rather crude idea to confirm the existence of artificial intelligence, has ever beed passed yet by any system. If that happens, there might be reason for concern.)

What DeepL does, like any other MT system, is applying algorithms and heuristics on huge amounts of data in order to create a target text that has a certain probability to imitate the result of a translation. It seems that this probability has been improved, compared with other MT systems. But that's about it. It is obvious that the rate of errors and outright embarrassments this MT produces makes it still unusable for any serious purpose. Algorithmic MT (and you can call it neural all day long, it's still algorithmic) will never be able to translate accurately. As you know, the same sentence can have more than one translation, depending on context. Understanding the context is one requirement for an accurate translation. You need AI for that, and AI does not exist. AI will probably exist at some point in the future, but right now it is nothing but science fiction.

For example, a truly intelligent MT system would simply not be able to translate right away a sentence you enter. A truly intelligent MT system would ask questions about the context, before suggesting a translation.

[Edited at 2017-09-11 14:34 GMT]


 
Giovanni Guarnieri MITI, MIL
Giovanni Guarnieri MITI, MIL  Identity Verified
United Kingdom
Local time: 14:42
Member (2004)
English to Italian
Good... Sep 11, 2017

Mirko Mainardi wrote:

Same thing with low rates, "discounts" for fuzzies, "mandatory" cat tools, absurd clauses in SLAs/NDAs, no minimum/rush fees, etc. etc. In other words, PEMT is here to stay, DeepL, ShallowL or whatever are just details.


I like a sprinkle of optimism from time to time...


 
kakapo77
kakapo77  Identity Verified
France
Local time: 15:42
French to Italian
+ ...
TOPIC STARTER
These arguments do not prove that I am wrong Sep 11, 2017

Kay-Viktor,

Thanks for your reply, you raised some very interesting points.

Yes, I cannot prove my statement about the terminology section on ProZ. I should collect data and analyzed them, that is right. But I could also say that your answer is an easy way to dismiss my argument.

I know that DeepL only uses algorithms, and of course it does not understand what it translates.

But the truth is that, in many cases, that is not relevant. Try it you
... See more
Kay-Viktor,

Thanks for your reply, you raised some very interesting points.

Yes, I cannot prove my statement about the terminology section on ProZ. I should collect data and analyzed them, that is right. But I could also say that your answer is an easy way to dismiss my argument.

I know that DeepL only uses algorithms, and of course it does not understand what it translates.

But the truth is that, in many cases, that is not relevant. Try it yourself, and you will see that it works extremely well. Make no mistake, the day you have a software able to understand the context, you will say goodbye to 99% of jobs on earth.

Marco

****

Giovanni,

You say that "it's up to us to say no". Do you really think that things work this way?

Marco
Collapse


 
Giovanni Guarnieri MITI, MIL
Giovanni Guarnieri MITI, MIL  Identity Verified
United Kingdom
Local time: 14:42
Member (2004)
English to Italian
@Marco Sep 11, 2017

kakapo77 wrote:

Giovanni,

You say that "it's up to us to say no". Do you really think that things work this way?



No. But that doesn't mean that the present and future decline of our profession is not our fault...

edited for typo

[Edited at 2017-09-11 16:20 GMT]


 
Kay-Viktor Stegemann
Kay-Viktor Stegemann
Germany
Local time: 15:42
English to German
In memoriam
Extremely unwell ... Sep 11, 2017

kakapo77 wrote:
Try it yourself, and you will see that it works extremely well.


No, Marco, it doesn't. At least not in my pair (EN>DE). I have started to use DeepL and there are cases where it looks good at first sight, but the rate of nonsense it produces is still appallingly high. Using it for commercial purposes would be disastrous. Maybe there are other pairs where it works better, but in EN>DE it's still a laugh, and when I use it for reading other languages I also get a lot of nonsensical output in German. Nonsense with quite good grammar at times, I give you that. The grammar imitation technique must be somewhat advanced in DeepL, probably because it draws from big amounts of language data.


Swati Kolazhy Vishwanthan
 
Maurice Koopman
Maurice Koopman  Identity Verified
Netherlands
Local time: 15:42
English to Dutch
+ ...
Random sample from a job Sep 13, 2017

Kay-Viktor Stegemann wrote:

kakapo77 wrote:
Try it yourself, and you will see that it works extremely well.


No, Marco, it doesn't. At least not in my pair (EN>DE). I have started to use DeepL and there are cases where it looks good at first sight, but the rate of nonsense it produces is still appallingly high. Using it for commercial purposes would be disastrous. Maybe there are other pairs where it works better, but in EN>DE it's still a laugh, and when I use it for reading other languages I also get a lot of nonsensical output in German. Nonsense with quite good grammar at times, I give you that. The grammar imitation technique must be somewhat advanced in DeepL, probably because it draws from big amounts of language data.


From EU

This Directive lays down health rules for the production and placing on the market of fresh meat intended for human consumption from domestic animals of the following species: bovine animals (including the species Bubalus bubalis and Bison bison), swine, sheep and goats, and domestic solipeds.
2. This Directive shall not apply to the cutting and storage of fresh meat performed in retail shops or in premises adjacent to sale points, where the cutting and storage are performed solely for the purpose of supplying the consumer directly on the spot.
3. This Directive shall apply without prejudice to specific Community rules on minced meat.
4. This Directive shall not affect any restrictions imposed, in compliance with the general provisions of the Treaty, on the retail sale of meat from solipeds.
Article 2
For the purposes of this Directive:
(a) ‘meat’ means all parts of domestic bovine animals (including the species Bubalus bubalis and Bison bison), swine, sheep, goats and solipeds which are suitable for human consumption;
(b) ‘freshmeat’ means meat, including meat vacuum-wrapped or wrapped in a controlled atmosphere, which has not undergone any treatment other than cold treatment to ensure preservation;
(c) ‘mechanically recovered meat’ means meat obtained by mechanical means from flesh-bearing bones apart from the bones of the head, the extremities of the limbs below the carpal and tarsal joints and, in the case of swine, the coccygeal vertebrae, and intended for establishments approved in accordance with Article 6 of Directive 77/99/EEC (2);

DeepL:

Diese Richtlinie enthält Hygienevorschriften für die Herstellung und Inverkehrbringen von frischem Fleisch, das für den menschlichen Verzehr bestimmt ist
von Haustieren folgender Arten: Rinder (einschließlich der Arten Bubalus bubalis und Bison bison), Schweine, Schafe und Ziegen und Hauseinhufer.
Zwei. Diese Richtlinie gilt nicht für das Zerlegen und Lagern von frischem Fleisch, das in Einzelhandelsgeschäften oder in an Verkaufsstellen angrenzenden Räumlichkeiten verzehrt wird, wenn die Zerlegung und Lagerung ausschließlich zu folgenden Zwecken erfolgt Versorgung der Verbraucher direkt vor Ort.
3. Diese Richtlinie gilt unbeschadet spezifischer gemeinschaftlicher Regeln für Hackfleisch.
4. Diese Richtlinie berührt nicht etwaige Beschränkungen, die in folgenden Fällen auferlegt werden Einhaltung der allgemeinen Bestimmungen des Vertrags über den Einzelhandel Verkauf von Fleisch aus Einhufern.
Artikel 2
Im Sinne dieser Richtlinie:
a)"Fleisch": alle Teile von Hausrindern (einschließlich Arten Bubalus bubalis und Bison bison), Schweine, Schafe, Ziegen und für den menschlichen Verzehr geeignete Einhufer;
b)"frisches Fleisch" Fleisch, einschließlich vakuumverpacktes oder jede andere Behandlung als Kaltbehandlung, um die Haltbarkeit zu gewährleisten;
c)"Separatorenfleisch": Fleisch, das durch mechanische Verfahren gewonnen wird aus fleischtragenden Knochen außer den Knochen des
Kopf, die Extremitäten der Gliedmaßen unterhalb des Handwurzel- und Tarsalgelenks und bei Schweinen die Steißbeinwirbel, und bei Schweinen die Steißbeinwirbel, und bestimmt für gemäß Artikel 6 der Richtlinie zugelassene Betriebe
77/99/EWG (2 );

Orginal translation from EU:

(1) In dieser Richtlinie werden die gesundheitlichen Bedingungen für die Gewinnung und das Inverkehrbringen von zum Verzehr
bestimmtem frischem Fleisch von Haustieren der Gattungen Rinder (einschließlich „Bubalus bubalis“ und „Bison bison“), Schweine, Schafe und Ziegen sowie von als Haustiere gehaltenen Einhufern festgelegt.
(2) Sie gilt nicht für die Zerlegung und Lagerung von frischem Fleisch im Einzelhandel oder in Räumlichkeiten, die an Verkaufsstellen angrenzen und in denen das Fleisch ausschließlich zum Zwecke des an Ort und Stelle stattfindenden Direktverkaufs an den Verbraucher zerlegt und gelagert wird.
(3) Diese Richtlinie gilt unbeschadet der besonderen Gemeinschaftsvorschriften für Hackfleisch.
(4) Diese Richtlinie berührt nicht etwaige Einschränkungen für das Inverkehrbringen von Fleisch von Einhufern auf der Stufe des Einzelhandels,
soweit sie mit den allgemeinen Bestimmungen des Vertrags in Einklang stehen.
Artikel 2
Im Sinne dieser Richtlinie sind:
a) „Fleisch“: alle genußtauglichen Teile von Haustieren der Gattungen Rinder (einschließlich „Bubalus bubalis“ und „Bison bison“), Schweine, Schafe, Ziegen und von als Haustiere gehaltenen Einhufern;
b) „frisches Fleisch“: Fleisch (einschließlich im Hochvakuum oder in definierter Atmosphäre umhülltes Fleisch), das nicht zum Zweck
der Haltbarmachung — außer mit Kälte — behandelt worden ist;
c) „Separatorenfleisch“: mechanisch von fleischtragenden Knochen, ausgenommen Kopfknochen, Röhrenknochen, Gliedmaßenenden unterhalb Karpal- bzw. Tarsalgelenk sowie Schweineschwänzen, gewonnenes und für die gemäß Artikel 6 der Richtlinie 77/99/ EWG (3) zugelassenen Betriebe bestimmtes Fleisch;

The tender was maybe not such a good idea. Therefor change of text. It is not as good as the tender translation, but in my opinion still sufficient to go straight to PE. I am impressed with it because most of the sentences are grammatically correct. The terminology may be faulty at some point, but the structure of the sentence is most of the time good.

So you believe that a translation agency will hire you to translate this text from scratch when DeepL delivers this?

[Edited at 2017-09-13 09:02 GMT]

[Edited at 2017-09-13 09:24 GMT]

[Edited at 2017-09-13 09:36 GMT]


 
Maurice Koopman
Maurice Koopman  Identity Verified
Netherlands
Local time: 15:42
English to Dutch
+ ...
No it wont Sep 13, 2017

Giovanni Guarnieri MITI, MIL wrote:

Maurice Koopman wrote:

Translation agencies and companies will not opt for a translator if they can have the text translate for free by DeepL and have it checked for a small amount because the quality is just too good to not use it.


It's up to us, isn't it? Just say no to the checking... if we all refuse, the "problem" goes away...


No it wont. Because this is what is called a market economy. It is based on competition. I am not a great fan of it, but that is how it works. That is also why there is DeepL now. DeepL found a way to deliver translations faster and in the end I am sure better. And I am not pessimistic for the sake of being pessimistic. I think DeepL is a real breakthrough - read the reactions from for example LeMonde, Stampa, Wired, RTL-Z - and we have to adapt. The sooner we do that, the better.

[Edited at 2017-09-13 08:48 GMT]

[Edited at 2017-09-13 09:26 GMT]


 
kakapo77
kakapo77  Identity Verified
France
Local time: 15:42
French to Italian
+ ...
TOPIC STARTER
What Maurice says is 100% correct Sep 13, 2017

That's 100% correct,

The problem is competition. As Maurice states, competition is the force driving down prices. And, by the way, competition indeed has already driven down translation prices in a dramatic way. Just compare where fees were ten years ago and today.

I mean, just take a look to Upwork or other platforms. Yes, I know, then you can tell me that Upwork is just for low entry jobs... but do yo know how many translators actually rely on website like that? And d
... See more
That's 100% correct,

The problem is competition. As Maurice states, competition is the force driving down prices. And, by the way, competition indeed has already driven down translation prices in a dramatic way. Just compare where fees were ten years ago and today.

I mean, just take a look to Upwork or other platforms. Yes, I know, then you can tell me that Upwork is just for low entry jobs... but do yo know how many translators actually rely on website like that? And do you know which are the fees over there?

As Maurice says, we are not pessimistic for the sake of being pessimistic. It's just that you have to wake up to reality. And yes, again, Kay-Viktor is right, a big amount of post-editing is needed.

But this means that you work as a post-editor. Which, by the way, is likely to be the future of translation - and its present as well.

Marco
Collapse


 
gayd (X)
gayd (X)
what are you going to do? Sep 13, 2017

Maurice Koopman wrote:

we have to adapt. The sooner we do that, the better.

[Edited at 2017-09-13 08:48 GMT]

Are you going to leave the translation business?


 
Maurice Koopman
Maurice Koopman  Identity Verified
Netherlands
Local time: 15:42
English to Dutch
+ ...
Maybe Sep 13, 2017

David GAY wrote:

Maurice Koopman wrote:

we have to adapt. The sooner we do that, the better.

[Edited at 2017-09-13 08:48 GMT]

Are you going to leave the translation business?


Perhaps, I dont know yet. As for now I will also mention to my (new) customers that they can contact me for PE.


 
gayd (X)
gayd (X)
in any case Sep 13, 2017

1) In my opinion, the first translation you had posted wasn't good: it was kind of a word for word translation which is not acceptable.
2) if DeepL is as good as you say it is, agencies are going to disappear as well.
3) Don't you think there's still a premium market for which human translation is still needed?


 
kakapo77
kakapo77  Identity Verified
France
Local time: 15:42
French to Italian
+ ...
TOPIC STARTER
Yes, but again think about the law of demand and supply Sep 13, 2017

Yes,

There is and there will be a premium market. Literature and philospohy, for instance.

But again, because of the forces which drive markets, then the fees will decrease. Because the offer will be on the increase.

By the way, that is likely to be a structural problem for the market of translation. There is too much offer.

Marco


 
gayd (X)
gayd (X)
not acceptable (at all) Sep 13, 2017

Diese Richtlinie gilt nicht für das Zerlegen und Lagern von frischem Fleisch, das in Einzelhandelsgeschäften oder in an Verkaufsstellen angrenzenden Räumlichkeiten verzehrt wird, wenn die Zerlegung und Lagerung ausschließlich zu folgenden Zwecken erfolgt Versorgung der Verbraucher direkt vor Ort.

 
Kay-Viktor Stegemann
Kay-Viktor Stegemann
Germany
Local time: 15:42
English to German
In memoriam
I get a different result from DeepL Sep 13, 2017

Maurice Koopman wrote:

Kay-Viktor Stegemann wrote:

kakapo77 wrote:
Try it yourself, and you will see that it works extremely well.


No, Marco, it doesn't. At least not in my pair (EN>DE). I have started to use DeepL and there are cases where it looks good at first sight, but the rate of nonsense it produces is still appallingly high. Using it for commercial purposes would be disastrous. Maybe there are other pairs where it works better, but in EN>DE it's still a laugh, and when I use it for reading other languages I also get a lot of nonsensical output in German. Nonsense with quite good grammar at times, I give you that. The grammar imitation technique must be somewhat advanced in DeepL, probably because it draws from big amounts of language data.


...

The tender was maybe not such a good idea. Therefor change of text. It is not as good as the tender translation, but in my opinion still sufficient to go straight to PE. I am impressed with it because most of the sentences are grammatically correct. The terminology may be faulty at some point, but the structure of the sentence is most of the time good.

So you believe that a translation agency will hire you to translate this text from scratch when DeepL delivers this?

[Edited at 2017-09-13 09:02 GMT]

[Edited at 2017-09-13 09:24 GMT]

[Edited at 2017-09-13 09:36 GMT]


Marco, when I copy the text above to DeepL, I get a slightly different translation with more errors, particularly with unnecessary strange word repetitions and omissions.
Anyway, if you take a text from the Internet that already has a translation (like the EU content), it is well possible that Linguee/DeepL knows this existing translation because the working model of Linguee/DeepL consists of collecting tons of existing translations, therefore the results might be better than if you feed other text to DeepL. I agree with you that the DeepL result here could be used as a base for post editing. But my experience is different. I am trying DeepL again and again and in most cases I receive a result that is more misleading than helpful. This might be particularly because the grammar seems so good at first sight.

You might also notice in your example that DeepL changes formatting and numbers in the translation, an absolute no-go even for MT. But this of course could be fixed by the DeepL developers.

But the bottom line is, yes, I am convinced that a translation agency will hire me for translating texts. It happens every hour of my working day. It might be that in the middle term DeepL could indeed be improved to the point that it delivers a result usable as a base for PE, and then I will probably adapt to the market and accept PE projects. As long as my resulting hourly rates are not affected, I have no problem with that. If it one day is so good that it does boost my productivity, I will use it and be able to produce more output. But this day has not come yet.


 
Mirko Mainardi
Mirko Mainardi  Identity Verified
Italy
Local time: 15:42
Member
English to Italian
Corpora Sep 13, 2017

Maurice Koopman wrote:

The tender was maybe not such a good idea. Therefor change of text. It is not as good as the tender translation, but in my opinion still sufficient to go straight to PE. I am impressed with it because most of the sentences are grammatically correct. The terminology may be faulty at some point, but the structure of the sentence is most of the time good.

So you believe that a translation agency will hire you to translate this text from scratch when DeepL delivers this?


AFAIK, a great deal of Linguee's corpora is made up of EU stuff, so it's only natural it may deal better with that kind of texts, also considering that its language is quite formulaic and repetitive, therefore naturally more suited to a MT engine than, for instance, creative content.

Also, about a past post of yours:

You say that "it's up to us to say no". Do you really think that things work this way?


They WOULD definitely work IF there was consensus. It's not unthinkable. Take as an example unions, associations, and every group that is able to pursue a common goal... The "market economy" is not an unquestionable truth and is definitely not infallible (as we all know quite well, I believe...), and needs correctives in order not to trample everything and everyone in its path, otherwise it would just be down to the 'survival of the fittest' or 'the law of the jungle'...


 
Pages in topic:   < [1 2 3 4 5] >


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


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

What do you think about DeepL?






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 »
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 »