Pages in topic:   < [1 2 3]
The Future of Translation
Thread poster: Isely Mills

Guofei_LIN  Identity Verified
Australia
Local time: 17:52
Chinese
Market force Dec 8, 2017

Michael Wetzel wrote:

MT is nowhere near being able to translate relatively simple natural language (this is an article from a newspaper, not an academic journal, and it's not slang or poetry or advertising-speak or dialogue).
I accept that there are a lot of other things DeepL probably can translate much better, and there are a lot of more or less incompetent translators out there who can't translate natural language either, so I'm not saying MT doesn't have a place, but I am saying that it continues to be irrelevant for my work and my business.


I just took this paragraph and feed it into Google Translate. The result (into Chinese) is astonishingly good:

MT远不能翻译相对简单的自然语言(这是一篇来自报纸的文章,而不是一本学术期刊,也不是俚语,诗歌或广告说话或对话)。
我同意DeepL可能翻译得好多了,而且还有很多翻译者或多或少都不能翻译自然语言,所以我不是说MT没有一个地方,但我说的是,这对我的工作和我的事业仍然是无关紧要的。

Not ready for publishing, but it is mostly smooth and fluent, easy to understand. I bet most of the clients would be happy with this quality. Given time, I think we will see the translation industry involving from the current 'machine-assisted (human) translation' into 'human-assisted machine translation', and when that day comes, the 'learn anything but translation' advice will be even more sound as the job will be taken over by industry experts in the subject materials who may not have any translation skills at all.

When we discuss machine translation vs human translation, we often subconsciously compare machines with the very good or the best human translators, and come to the conclusion that since machines will not be able to beat the best human translators in quality, they will not beat human translators at all.

But the market doesn't work like that. If machines can be as good as 60% of human translators in quality, they will take away all entry-level jobs from the market. When the entry-level jobs are gone, human translators will not be able to hone their skills with a paid entry-level job for 20 years till they become very good at their craft. When entry-level jobs are gone in an industry, the top jobs will be gone next.

[Edited at 2017-12-08 22:44 GMT]


 

Mario Chavez (X)  Identity Verified
Local time: 02:52
English to Spanish
+ ...
I ain't no ostrich :) Dec 8, 2017

Hans Lenting wrote:

Gerard de Noord wrote:

Things have changed this year. NMT is so much better than anything we had before that I'm not even restricting this opinion to my working languages. The technology is better, the method is better, and the outcomes are better. Old translators can go on ignoring NMT, middle aged translators should start worrying, young translators should rethink their future.


I completely agree with Gerard. You can replace 'this year' with last September, when DeepL was introduced. The waiting is for DeepL adding more languages and for Google's response and Amazon's implementation of their MT system, etc.

The perception of the translation process and of our profession will change dramatically over the next few years. We will have to raise our prices per word to compensate for the decrease in volume that is to be expected and for us filling in the gaps between parts (sentences, paragraphs) that already have been translated in an acceptable of even elegant form by MT. And for the process of making our human translations (the gaps) and the MT's output consistent.

For this activity we will have to find a compensation that is acceptable for us translators and for our clients. One thing is for sure: we won't be able to make a living anymore when we are getting paid our current word prices for filling in the gaps that the MT system cannot fill itself. And rest assured that it'll quickly learn to fill in the gaps in it's output by harvesting our translations of these gaps.

Unknown


I don't believe in all-or-nothing scenarios. There are always solutions and adaptations.

I use Déjà Vu X3 for most of my technical and medical translations. Although DVX3 gives me the option to add one or more MT engines, I don't use them. Why? For two reasons: my intellectual work isn't free (I think we can all safely agree on this) and, consequently, I'm not giving it to Google Translate and their ilk for nothing. The second reason is less idealistic but more important in the here and now: I'm not going to risk divulging confidential information, whether my client specifically labels it as confidential or not, by sending it to the Internet Ether.

Most online MT engine models rely on translators and ad hoc translators (bilingual people who act like translators) to build their database with better versions of a sentence or paragraph. Since I've been proven wrong before, I'll be happy to hear about an MT engine that does not harvest nor use third-party translations without due compensation or without confidentiality protections (an oxymoron, I know).


 

Hans Lenting  Identity Verified
Netherlands
Member (2006)
German to Dutch
Based on Linguee Dec 8, 2017

DeepL is trained with data from the huge Linguee corpus, so it should be tested with manuals/instructions. How often does one get paid to translate news paper articles? Apples and pears.

 

Alistair Gainey  Identity Verified
United Kingdom
Local time: 07:52
Member (2009)
Russian to English
Our old friend Dec 8, 2017

Google Translate actually gets this bit

Michael Wetzel wrote:

"Die SPD will auf ihrem Parteitag der privaten Krankenversicherung den Weg in den Tod bereiten.


right. Sort of

"The SPD wants to prepare the way to death at its private health insurance congress."


 

Michael Wetzel  Identity Verified
Germany
Local time: 08:52
German to English
I was trying to be generous. Dec 8, 2017

My point was that it can't even translate newspaper articles, not that it is specifically unable to translate newspaper articles. I can feed text from an exhibition catalogue into it if you like, but I figured people would complain that is unfair.

That's what I get paid for and that's why I don't really care if DeepL can or can't translate manuals and instructions. Those would also be some of the things that I could imagine it might be significantly better at doing.

Co
... See more
My point was that it can't even translate newspaper articles, not that it is specifically unable to translate newspaper articles. I can feed text from an exhibition catalogue into it if you like, but I figured people would complain that is unfair.

That's what I get paid for and that's why I don't really care if DeepL can or can't translate manuals and instructions. Those would also be some of the things that I could imagine it might be significantly better at doing.

Comparing translators with translators is like comparing apples with oranges or pears: I thought you were generalizing about translators in general with your assertions about DeepL/MT, if you were only generalizing about manual-and-instruction translators, then I don't have any opinion about that.
Collapse


 

Mario Chavez (X)  Identity Verified
Local time: 02:52
English to Spanish
+ ...
Linguee Dec 9, 2017

Hans Lenting wrote:

DeepL is trained with data from the huge Linguee corpus, so it should be tested with manuals/instructions. How often does one get paid to translate news paper articles? Apples and pears.


I remember resorting to Linguee for some civil engineering expressions, but not as my go-to resource or reference but for comparison purposes with other resources I had. Ew, I write “-Linguee” every time I run a Google search for an expression.


 

Hans Lenting  Identity Verified
Netherlands
Member (2006)
German to Dutch
Quite the opposite Dec 9, 2017

Michael Wetzel wrote:

My point was that it can't even translate newspaper articles, not that it is specifically unable to translate newspaper articles.


I'd say that translating newspaper articles is highly difficult for DeepL, since it has no training data for this. Feed DeepL with thousands of (translations of) weather reports or soccer reports and it'll be able to translate new reports very well. At least, that's my expectation. Feed it with thousands of (translations of) articles about the conflict between the Israelis and Palestinians and it'll be capable to translate tomorrow's news about this conflict.


 

Yolande Hivart
Austria
Local time: 08:52
Member (2016)
German to French
Linguee and co Dec 9, 2017

I think linguee and co is giving a false sense of security to young translators at best. The translation is mostly literal - maybe something you learn at school - but does not have the sense of creativity that my clients expects from me. Copying linguee or google translation would get me at least a complaint from France for it not to be French (event if you could prove with a dictionary that it is the official translation). Translation is not translating words, not terms, it is making yourself a... See more
I think linguee and co is giving a false sense of security to young translators at best. The translation is mostly literal - maybe something you learn at school - but does not have the sense of creativity that my clients expects from me. Copying linguee or google translation would get me at least a complaint from France for it not to be French (event if you could prove with a dictionary that it is the official translation). Translation is not translating words, not terms, it is making yourself a picture of what the writer is trying to say and writing it in the target language. As long as MT is not able to translate this "imaged bubble" of implied content in any other way that reusing what a translator of another style of writing might have been writing under different conditions for maybe a different client for a different product, I do not see the chance of MT replacing seasoned translator having a specific style.
I even know for instance of a client for DE-EN who is ready to pay 3 times the price and wait 10 times the normal writing time of a translator to have THIS translator and not anyone else because THIS translator is writing the way that fits HIS purpose.
As for linguee and experience with it, i think it is slightly better than google translator but not the way if you want to settle yourself as an irreplaceable translator.
You should not use lingue at such without double check. Maybe it helps but most of the time it is a translation of a translator that might have settle for the first approximate term that might halfway fits the purpose.
Linguee might help as a starter and only there the search might start if you are stuck. You have then to research the definitions of terms in the target language, the source language, read technical documentation around the term you found, find synonyms, similar products, see what they used in the target language for the same semantic field and only then you will come to deliver a translation that will make you a stock of clients that do see the difference and will not try to replace you with MT. And i am talking for technical terms. This might be even worse for creative terms, for legals term in austria, i had to learn as specific list of terminology that the courts are usually using and it might not always be what is written in linguee - you have to regularly sit next to the judge to understand the subtilities.
Using automated translation for a marketing text just using MT to get it cheaper is for you to jump out of the cliff out someone's else untested parachute. If you are not rewriting you get a complaint for it not be fitting. If you are rewriting you are getting a complaint for not using the TM. When i am making myself a base of better clients, I have the opportunity to develop my style on the clients that respect my style and pay accordingly for such and not on the agency who keep pressing and try to find a way to pay less, later, and in the end not at all.

So far on the line i am going I do not worry about MT, I notice that the clients try more and more to send me jobs without MT because they feel the difference when i am free from the bondages of a modular system - even it is is still interesting for me, even for my learning to look in a TM to see what other good translators wrote, this sometimes helps me to improve my own style but this is not pressed on me and if the client does not want to pay for 100% he should lock them and takes the responsibility for it which he usually does.
Collapse


 

Ligeti  Identity Verified
Netherlands
Local time: 08:52
English to Dutch
+ ...
More bad news Jan 15, 2018

- http://money.cnn.com/2018/01/15/technology/reading-robot-alibaba-microsoft-stanford/index.html

The robots are coming, and they can read.
Artificial intelligence programs built by Alibaba (BABA) and Microsoft (MSFT) have beaten humans on a Stanford University reading comprehension test.

"This is the first time
... See more
- http://money.cnn.com/2018/01/15/technology/reading-robot-alibaba-microsoft-stanford/index.html

The robots are coming, and they can read.
Artificial intelligence programs built by Alibaba (BABA) and Microsoft (MSFT) have beaten humans on a Stanford University reading comprehension test.

"This is the first time that a machine has outperformed humans on such a test," Alibaba said in a statement Monday.

The test was devised by artificial intelligence experts at Stanford to measure computers' growing reading abilities. Alibaba's software was the first to beat the human score.

Related: Google is opening an artificial intelligence center in China

Luo Si, the chief scientist of natural language processing at the Chinese company's AI research group, called the milestone "a great honor," but also acknowledged that it is likely lead to a significant number of workers losing their jobs to machines.

The technology "can be gradually applied to numerous applications such as customer service, museum tutorials and online responses to medical inquiries from patients, decreasing the need for human input in an unprecedented way," Si said in a statement.

Alibaba has already put the technology to work on Singles Day, the world's biggest shopping bonanza, by using computers to answer a large number of customer service questions.

In a tweet, Pranav Rajpurkar, one of the Stanford researchers who developed the reading test, called Alibaba's feat "a great start to 2018" for artificial intelligence.

Related: These three countries are winning the global robot race

The Stanford test generates questions about a set of Wikipedia articles.

For example, a human or AI program reads a passage about the history of British TV show Doctor Who and then answers questions like, "What is Doctor Who's space ship called?" (Spoiler alert: It's the TARDIS, for non-Doctor Who fans out there.)

Alibaba's deep neural network model scored 82.44 on the test on January 11, narrowly beating the 82.304 scored by the human participants. A day later, Microsoft's AI software also beat the human score, with a result of 82.650.

"These kinds of tests are certainly useful benchmarks for how far along the AI journey we may be," said Andrew Pickup, a spokesman for Microsoft. "However, the real benefit of AI is when it is used in harmony with humans," he added.

Facebook (FB), Tencent (TCEHY) and Samsung (SSNLF) have also previously submitted AI models to the Stanford project.

Related: Jack Ma: We need to stop training our kids for manufacturing jobs

Artificial intelligence is already causing disruption in industries around the world -- replacing warehouse workers with robots, operating self-driving cars and even helping farmers grow better crops.

Russian President Vladimir Putin predicted in September that whoever becomes the leader in artificial intelligence "will become the ruler of the world."

China is making a big push to be a dominant force.

Beijing said it wants the country to be a leader in artificial intelligence by 2020. In July, government officials set out goals to build a domestic artificial intelligence industry worth nearly $150 billion in the next couple years.

[Edited at 2018-01-15 13:10 GMT]

[Edited at 2018-01-15 13:10 GMT]
Collapse


 
Pages in topic:   < [1 2 3]


There is no moderator assigned specifically to this forum.
To report site rules violations or get help, please contact site staff »


The Future of Translation

Advanced search







Wordfast Pro
Translation Memory Software for Any Platform

Exclusive discount for ProZ.com users! Save over 13% when purchasing Wordfast Pro through ProZ.com. Wordfast is the world's #1 provider of platform-independent Translation Memory software. Consistently ranked the most user-friendly and highest value

More info »
SDL MultiTerm 2019
Guarantee a unified, consistent and high-quality translation with terminology software by the industry leaders.

SDL MultiTerm 2019 allows translators to create one central location to store and manage multilingual terminology, and with SDL MultiTerm Extract 2019 you can automatically create term lists from your existing documentation to save time.

More info »



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