| Pages in topic: < [1 2 3 4 5] > | DeepL Thread poster: Gerard Barry
|
|---|
Tony Keily
Local time: 03:05
Italian to English
+ ...
| It can be pretty good and pretty bad | Jun 24, 2021 |
The bottom line is, like all tools, it's as good as the person using it.
If you're translating regulatory documents inside the EU it can be pretty good (mainly thanks to the EU institutions making all of their multilingual texts available for such apps to use) and save you some time. However, you need to keep awake because it will play some strange tricks on you now and again.
On certain more periphrastic legal or tricky financial/accounting documents, it's basically cr... See more The bottom line is, like all tools, it's as good as the person using it.
If you're translating regulatory documents inside the EU it can be pretty good (mainly thanks to the EU institutions making all of their multilingual texts available for such apps to use) and save you some time. However, you need to keep awake because it will play some strange tricks on you now and again.
On certain more periphrastic legal or tricky financial/accounting documents, it's basically crap, but that's to be expected. ▲ Collapse
| | | | Daniel Frisano 
Italy
Local time: 03:05
Member (2008)
English to Italian
+ ...
| Translators, really? | Jun 24, 2021 |
If you use DeepL or any other equivalent engine, don't call yourself a translator: you're a post-editor.
No skilled translator would ever use a tool that produces a much crappier version of their work.
Glossaries, termbases, etc. are useful for terminology, but MT just drags your natural language ability (that you should always cultivate and enhance) into utter mediocrity.
| | | | Stepan Konev
Russian Federation
Local time: 04:05
English to Russian
| Why one-word assistance is useful but multiword one is not? | Jun 24, 2021 |
Daniel Frisano wrote:
Glossaries, termbases, etc. are useful for terminology, but MT just drags If glossaries/termbases that help you type one or two words are useful, then why MT engines that can be trained to use not only your glossaries/termbases but also TMs to build phrases and even sentences in your language/style are not? A glossary is a system that saves you time to remember and type one or two words. An MT engine is a system that saves you time to remember and type many words. From this standpoint, a glossary is just a single element in a system that uses glossaries and other resources. Have you ever tried a trained MT engine in action? Many CAT tools these days use fuzzy repair, or uplift, or whatever it is called, a technology that builds sentences based on your own previous translations and glossaries. This is exactly what trained MT engines do. It is merely a tool for professionals. You must know how to use it. Like a knife: you can cut a beautiful sculpture or fingers depending on your mastery.
[Edited at 2021-06-24 23:37 GMT]
| | | |
Stepan Konev wrote:
Why one-word assistance is useful but multiword one is not?
It is merely a tool for professionals. You must know how to use it. Like a knife: you can cut a beautiful sculpture or fingers depending on your mastery.
MT is inherently unable to understand the text, it can only fake such understanding, and is bound to make blunders. These blunders are much less predictable than human ones, so one has to check all the output against the source very carefully. For a seasoned professional it will normally take longer than producing a translation from scratch.
I am not saying MT has no use, I merely consider it counterproductive for top-quality translation.
| | |
|
|
|
Adieu
Ukrainian to English
+ ...
| I experimented with it | Jun 25, 2021 |
It is certified crap.
When it is good, it IS pretty good, but alas it has a nasty habit of just VANISHING entire clauses.
Forget mistranslating them with memeworthy MT. Nope. No bad translation here, just NO TRANSLATION (and no indication that anything was ever there) right in the middle of your paragraph.
| | | | Gerard Barry
Germany
Local time: 03:05
German to English
TOPIC STARTER | I think DeepL is very good | Jun 25, 2021 |
I've just ran another text (a newspaper article) through DeepL and the result was again very good. I think it's possible some translators are in denial about how good it is. I admit that factors like source/target language and so on may make a difference but still.
| | | | Stepan Konev
Russian Federation
Local time: 04:05
English to Russian
| Different people, different opinions | Jun 25, 2021 |
Anton Konashenok wrote:
MT is inherently unable to understand the text, it can only fake such understanding, and is bound to make blunders. These blunders are much less predictable than human ones, so one has to check all the output against the source very carefully. For a seasoned professional it will normally take longer than producing a translation from scratch. Well, thank you for your explanation, but I have been using translation tools, including MT engines, for many years now and I perfectly know how they work. And I remember translators who complained that CAT tools drastically lower their efficiency. It is ok because all people are different and have different level of skills. If you can't use a tool, it takes longer for you to use that tool for sure. Isn't it obvious? What you consider normal and what takes longer for your, can be different for others. Obviously, MT cannot understand context. And obviously again you have to double-check any translation, both MT and human-made. I never said that you can blindly rely on MT and leave it uncontrolled or unedited. But do you really think that everybody who use MT in their work are idiots or not seasoned professionals. Are you serious in stating this?
*Generally I am strongly against PEMT projects proposed by many agencies these days. Not because I am against MT, but because MT is a tool for translators to facilitate their work, and not for agencies to rip off translators. I always kept this position regarding CAT tools (CAT tools are designed for higher efficiency, not for CAT-grid discounts). And I still keep the same position with regard to MT.
[Edited at 2021-06-25 14:07 GMT]
| | | | Erik Freitag
Germany
Local time: 03:05
Member (2006)
Dutch to German
+ ...
| Newspaper articles | Jun 25, 2021 |
Gerard Barry wrote:
I've just ran another text (a newspaper article) through DeepL and the result was again very good. I think it's possible some translators are in denial about how good it is. I admit that factors like source/target language and so on may make a difference but still.
Newspaper articles are one of the easiest kind of texts to translate. They're usually very well written, coherent texts in full sentences, with limited complexity of the content. They do not rely too much on external context. Not much jargon, no errors. I wish I would get this kind of text in my daily work, but I rarely do.
I have tried out DeepL extensively and found it to be not very helpful at all, sometimes even counterproductive, for the kind of texts I mostly work with. I'm doing better without it so far.
| | |
|
|
|
Tom in London
United Kingdom
Local time: 02:05
Member (2008)
Italian to English
Here's a typical example of the mess Deepl makes with a perfectly ordinary piece of Italian:
ITALIAN: un processo identitario quanto mai problematico e, ancora oggi, a tratti inconcluso
DEEPL: an identity process that has never been so problematic and, even today, at times inconclusive
This is horrendous! I hope nobody relies too much on Deepl.
[Edited at 2021-06-27 12:11 GMT]
| | | | | MT translation detection tool - anyone? | Jun 28, 2021 |
Hi,
Is there any software program, app, or web tool that can help differentiate between MT and Human-translated text?
| | | | Elena Feriani
Italy
Local time: 03:05
Member
French to Italian
+ ...
| The other way around works better | Jun 28, 2021 |
Tom in London wrote:
Here's a typical example of the mess Deepl makes with a perfectly ordinary piece of Italian:
ITALIAN: un processo identitario quanto mai problematico e, ancora oggi, a tratti inconcluso
DEEPL: an identity process that has never been so problematic and, even today, at times inconclusive
This is horrendous! I hope nobody relies too much on Deepl.
[Edited at 2021-06-27 12:11 GMT]
I guess EN-IT MT works better because English uses recognisable (sensible) structures, while Italians tend to mix up the word order as much as possible and can make long sentences without any verb.
Anyway, it turns out that the British English translation of your sentence is less horrendous on DeepL:
A very problematic and, even today, sometimes unfinished identity process.
| | | | Tom in London
United Kingdom
Local time: 02:05
Member (2008)
Italian to English
Elena Feriani wrote:
....
Anyway, it turns out that the British English translation of your sentence is less horrendous on DeepL:
A very problematic and, even today, sometimes unfinished identity process.
I didn't translate it like that. One of the traps is "identitario" (which does not mean "identity"; it means "identitiarian", an unusual word that is, however appropriate in the wider context of the paper) and another is "a tratti" (which in this context could mean something like "here and there" or "in some parts" etc.).
It took me quite a long time to capture the writer's intended meaning (so far; I'm still working on it).
My point is (was) that when is come to intelligent academic texts that rely on allusion, nuance, and a lexicon that uses terms that might not be common in everyday texts but are perfectly à propos in the academic world, Deepl is completely.....out of its depth. Or deepth.
| | |
|
|
|
Jan Truper
Germany
Local time: 03:05
English to German
| My thoughts on DeepL | Jun 28, 2021 |
First of all, I assume that quality and benefits of DeepL vary greatly between languages and types of application.
For me, when translating games from English to German in memoQ, it’s a wonderful tool.
—It doesn’t make me faster, but it makes me better. —
I use it as follows:
I pretranslate the next 50 segments; anything under a TM threshold of 70% gets filled in by DeepL.
I read the source segment and translate in my head;... See more First of all, I assume that quality and benefits of DeepL vary greatly between languages and types of application.
For me, when translating games from English to German in memoQ, it’s a wonderful tool.
—It doesn’t make me faster, but it makes me better. —
I use it as follows:
I pretranslate the next 50 segments; anything under a TM threshold of 70% gets filled in by DeepL.
I read the source segment and translate in my head; this happens rather quickly, somewhat like a simultaneous interpreter, as I have a lot of experience. By the time my eyes wander to the target column, I have the segment readily translated in my head.
Now, very importantly, I always assume the pretranslation is shit (obviously, since the machine is too dumb to consider basic requirements like context, applicable form of address, character limits of a segment, separate notes, etc.). But gazing over the MT output, I quickly spot terms or phrases that I might prefer to the ones I translated in my head.
I then type or dictate my translation incorporating MTs usable improvements and go on to the next segment.
All in all, this process does not make me faster than if I were not to use DeepL, but it makes me better, since I incorporate usable improvements; it’s sort of like consulting a second opinion.
Also, game translation sometimes involves fields that I only have a basic understanding of (genetic engineering or rocket science, for example), and I find DeepL’s output helpful as a starting point for such passages.
And lastly, I like to use the DeepL standalone application as a thesaurus (you click on a word and get presented with numerous options). ▲ Collapse
| | | | Elena Feriani
Italy
Local time: 03:05
Member
French to Italian
+ ...
Tom in London wrote:
Elena Feriani wrote:
....
Anyway, it turns out that the British English translation of your sentence is less horrendous on DeepL:
A very problematic and, even today, sometimes unfinished identity process.
I didn't translate it like that. One of the traps is "identitario" (which does not mean "identity"; it means "identitiarian", an unusual word that is, however appropriate in the wider context of the paper) and another is "a tratti" (which in this context could mean something like "here and there" or "in some parts" etc.).
It took me quite a long time to capture the writer's intended meaning (so far; I'm still working on it).
My point is (was) that when is come to intelligent academic texts that rely on allusion, nuance, and a lexicon that uses terms that might not be common in everyday texts but are perfectly à propos in the academic world, Deepl is completely.....out of its depth. Or deepth.
A tratti means many things and DeepL got it wrong. As for idenditarian, I see that it can have a negative/racist connotation in English, while in Italian I understand it as the process (of) creating one's identity. Sorry for the off-topic! I agree DeepL is not helpful when translating this kind of sentences.
| | | | Jeff Whittaker
United States
Local time: 21:05
Spanish to English
+ ...
| The problem is that you never know when there is a crucial error | Aug 18, 2021 |
For example: DeepLcom translates "la Partie créancière de l'obligation" as "the Party owing the obligation." But in fact it means the exact opposite: the party to whom the obligation is owed (=the "obligee").
DeepL translates "Autonomie des stipulations" in a contract as "Autonomy of stipulations."
But this clause is called "Severability" in English.
[Edited at 2021-08-18 13:03 GMT]
| | | | | Pages in topic: < [1 2 3 4 5] > | There is no moderator assigned specifically to this forum. To report site rules violations or get help, please contact site staff » DeepL | Trados Business Manager Lite |
|---|
Create customer quotes and invoices from within Trados Studio
Trados Business Manager Lite helps to simplify and speed up some of the daily tasks, such as invoicing and reporting, associated with running your freelance translation business.
More info » |
| | LinguaCore |
|---|
AI Translation at Your Fingertips
The underlying LLM technology of LinguaCore offers AI translations of unprecedented quality. Quick and simple. Add a human linguistic review at the end for expert-level quality at a fraction of the cost and time.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
