Pages in topic:   [1 2 3] >
Why translators have nothing to fear from machine translation
Thread poster: ViktoriaG

ViktoriaG  Identity Verified
Canada
Local time: 03:12
English to French
+ ...
Jul 21, 2009

I just read this today: http://www.google.com/hostednews/afp/article/ALeqM5iwxC1-zObGGmevHxhtGcwlqm10Gw

I thought I would bring this to my colleagues' attention, seeing as many seem to be scared of being replaced by machines. The machine can use all the algorythms it can and be the best computer engineering has to offer, it will always lack something: intelligence. The machine will never be intelligent.

As we all know, translation is not about words--it's all about meaning. In order to be able to convey meaning, one has to understand it first. Without intelligence, it can't be done.

Machine translation has its uses, but it will only ever work combined with human intelligence. So, the worst thing that can ever happen is that translators are asked to calibrate machine translation engines, to feed machine translation dictionaries and to correct machine translation output. And we all know how reticent most of us are to do that--unless we are improving our own subject-specific machine translation engines installed on our work computers.

See? Nothing to fear!


 

Shripal Sharma
India
Local time: 12:42
Member (2009)
English to Hindi
+ ...
Business Issue Jul 21, 2009

In Hindi machine translation is not perfect. It is sense based. Before translation we have to understand the concept, purpose and even educational standard of the end users. We have to translate the matter keeping in view all these.

So machine translation is not suitable in case of language pair English to Hindi or Hindi to English.


 

Heinrich Pesch  Identity Verified
Finland
Local time: 10:12
Member (2003)
Finnish to German
+ ...
Not for natural languages Jul 21, 2009

To make MT work the source text should be written in an artificial, unambiguous language. Only simple facts can be translated properly by MT. But nobody would like to read the stuff.

 

Carla Guerreiro  Identity Verified
France
Local time: 09:12
French to Portuguese
+ ...
I agree! Jul 22, 2009

Hello Viktoria,

I agree with you. No machine can replace human language and thoughts.
Maybe one day someone will create an amazing, great and complete translation software, which includes all subtilities of human language and a huge therminological database.
However, I don't think this might happen in a near future.


 

Samuel Murray  Identity Verified
Netherlands
Local time: 09:12
Member (2006)
English to Afrikaans
+ ...
That is precisely what they fear Jul 22, 2009

ViktoriaG wrote:
So, the worst thing that can ever happen is that translators are asked to calibrate machine translation engines, to feed machine translation dictionaries and to correct machine translation output.

See? Nothing to fear!


Isn't being reduced to proofreaders of machine translated content exactly what most translators who fear machine translation fear?


 

Laurent KRAULAND  Identity Verified
France
Local time: 09:12
French to German
+ ...
Yes, at least for me Jul 22, 2009

Samuel Murray wrote:

Isn't being reduced to proofreaders of machine translated content exactly what most translators who fear machine translation fear?


When that day comes, I will turn to something else... definitively! My main point being that people (administrations, you name it!) nowadays are conditioned/inclined to take for granted or correct what comes out of a machine. On a general level, I don't foresee any mental or intellectual improvements for the future, plus you don't absolutely need to be a translator to amend MT outputs.

Laurent K.

[Edited at 2009-07-22 08:36 GMT]


 

Susan Welsh  Identity Verified
United States
Local time: 03:12
Member (2008)
Russian to English
+ ...
MT software that you install on your computer, not Google Translate Jul 22, 2009

Hi Viktoria,

In this thread from some months back, you mentioned that there is MT software (not Google Translate) that you can "train" for your own purposes and that you have found useful in translating. I'd be interested to know what it is.

http://www.proz.com/forum/business_issues/132815-does_google_translate_render_the_human_mind_unnecessary-page3.html

Thanks,
Susan


 

Richard Bartholomew  Identity Verified
Germany
Local time: 09:12
Member (2007)
German to English
Intelligence first, then translation Jul 22, 2009

Whether or not machines will ever be intelligent (in the Turing test sense), I will not venture to guess. However, I agree that no non-intelligent machine can translate as well as a human translator.

On the other hand, just imagine a conversation between two intelligent (not in the Turing sense) machines discussing human translation.

Machine 1: What's this?
Machine 2: It's a human translation.
Machine 1: Oh, no, not another one of those. Some of the stuff those human translators write is completely absurd. Half the time I have no idea what they're talking about. Not only that, but they're slow and expensive.
Machine 2: Yeah, I guess humans have a way to go before catching up with us machines. I'll get a mechanical proofreader to go over it.

BTW, do you think you could produce a convincing machine translation? In other words, do you think you could pass a reverse Turing test in which another human would fail to identify your work as human with 0.5 probability?


[Edited at 2009-07-22 11:39 GMT]


 

Laurent KRAULAND  Identity Verified
France
Local time: 09:12
French to German
+ ...
A great LOL at this... Jul 22, 2009

Richard Bartholomew wrote:

BTW, do you think you could produce a convincing machine translation? In other words, do you think you could pass a reverse Turing test in which another human would fail to identify your work as human with 0.5 probability?


and a great thank you for putting this into words, Richard!

Laurent K.


 

Samuel Murray  Identity Verified
Netherlands
Local time: 09:12
Member (2006)
English to Afrikaans
+ ...
I once failed it Jul 22, 2009

Richard Bartholomew wrote:
BTW, do you think you could produce a convincing machine translation? In other words, do you think you could pass a reverse Turing test in which another human would fail to identify your work as human with 0.5 probability?


A proofreader once sent a piece of work of mine back to me, saying that she doesn't proofread machine translations. The translation was rather direct (as per the style of the relevant publication) but it was done by me alone (okay, and a CAT tool). The thing that made the proofreader identify the work as machine translated was the directness of the translation, and I was unable to convince her that it was human made (even telling her that there are no cheap machine translation engines in my language combination didn't convince her).


 

RoyMarie
United States
Local time: 00:12
An example of MT that can be trained Jul 22, 2009

How do you build a high quality customized translation engine?
There are several factors that contribute to a high quality custom engine. You may need different engines or domains built for different purposes. An engine that was built for high quality translation of financial reports would probably do a terrible job of translating content on healthcare, medicine or an article on the English Premier League football. For this reason we create high quality customized translation engines in key domains that are focused on a specific subject matter or areas of knowledge. NOTE: This section uses a number of terms that have specific definitions. The Common Definitions of Terms Used by Asia Online page provides additional clarity.

Quality does not mean perfect. Typically, system output begins at a “good enough” quality for accurate comprehension, and then evolves to higher levels of quality over a period of time. Asia Online has built a series of tools and processes that enable the rapid creation of a custom engine, and then facilitates the ongoing refinement, fine tuning and evolution of that engine, until it reaches a level that is able to greatly reduce the time, cost and effort of producing ongoing translations.

There are three key stages to building a high quality custom translation engine.
1 Build a custom engine foundation.
2 Fine tune the foundation to the desired level of quality.
3 Correct translation errors as they occur and feedback into the Asia Online platform as new corrective learning data.

Cycling through these three stages delivers significant increases in productivity, and ensures that the human feedback and corrections are applied as knowledge driving continued improvements in the translation engine.

http://www.asiaonline.net/translation.aspx


 

ViktoriaG  Identity Verified
Canada
Local time: 03:12
English to French
+ ...
TOPIC STARTER
The fear isn't convincing them into doing it Jul 22, 2009

Samuel Murray wrote:

Isn't being reduced to proofreaders of machine translated content exactly what most translators who fear machine translation fear?

Well, it seems that, for the time being, most of us are reluctant to work as MT output reviewers and proofreaders. So, as long as most of us refuse to do the fixing of maimed text, I would think we are all safe.

What I was pointing at more specifically when I was talking about fear was people fearing to be replaced by machines, not people fearing to be forced to work with them. Being replaced by machines is what I say we should not fear, because it is not likely to happen. Not now, not ever. Unless we somehow manage to create intelligent machines, but that is an oxymoron to me (I've called machines 'stupid' many more times than I did people).


 

ViktoriaG  Identity Verified
Canada
Local time: 03:12
English to French
+ ...
TOPIC STARTER
Promt Jul 22, 2009

Susan Welsh wrote:

...you mentioned that there is MT software (not Google Translate) that you can "train" for your own purposes and that you have found useful in translating. I'd be interested to know what it is.

I have tried Promt a few times (a colleague uses it and really likes it). I used to be totally against MT, and that's why I wanted to get first-hand experience with it. I was rather impressed, but not by the MT technology or this particular software. Rather by the realization that a local, customizable MT engine can produce some pretty handsome translations (with some mistakes here and there all the same), provided the person who uses it constantly teaches it how to handle phrases and keeps feeding it with terms.

As I recall, Promt allows you to create subject-specific dictionaries (we're a long way from Google here), much like you would create subject-specific termbases. So, for instance, if a certain word has several translations, depending on the dictionary you load, the machine will pick the right translation for the context. The colleague who uses it also told me that he sees MT software as a productivity tool rather than a tool to produce translation, which is also my impression (based on my limited experience). It comes down to having the machine pretranslate a sentence for you and then reviewing the sentence yourself before inserting it into the target document. This is different from using online MT engines, as with a local engine, you can customize everything. It more or less comes down to getting high-quality fuzzy matches, if you are familiar with CAT technology. I remember Promt had several interfaces depending on what you wanted to do, but one allowed pretty straightforward integration with CAT tools.

The Pro version also allows you to translate a document and automatically create a TM for it, which you can then use to finish the translation using a CAT tool. It also has a module that lets you import a termbase into a Promt dictionary. I haven't tried Systran (I think I should), so I don't know if it has similar capabilities. All I know is that Promt is surprisingly efficient as a productivity tool for a translator. I am actually considering adding it to my productivity arsenal. But with all my CAT tool misadventures lately, I may end up thoroughly changing my entire software library (read: switching Trados out for something else). So, I have no idea what I will come up with. But if any of you tries or buys MT software, let me know what you think.

I believe that MT should be a tool for translators. I strongly disagree with translators being asked to review MT output, as those who produce that output are often totally ignorant about translation and therefore can't create an acceptable MT output either. The translator should be the one to create the MT output to begin with. And that would be added value, worth a price. The only advantage of MT to the end users should be faster turnaround. The use of MT shouldn't affect rates at all. Sadly, some people tend to reduce their rates every time they can improve their turnaround, so if the use of such software were widespread, I am afraid rates would suffer further, and those who refuse to use MT software would be the first to quit. But that is a topic for another thread...

It seems you can even get the Pro version as trial software at http://www.promt.com/

I think that MT can be very useful with human intervention and translators could benefit from its use. However, without human intervention...


 

Susan Welsh  Identity Verified
United States
Local time: 03:12
Member (2008)
Russian to English
+ ...
Thanks, Viktoria! Jul 23, 2009

I will definitely try it out.
(I recently ran an RU-EN job through Google Trans, to see if it would help me. Mostly it was garbage, but every so often it would come up with a word that I couldn't find elsewhere, but that seemed just right. Yet it would miss the most ordinary sorts of words and phrases. If I understand correctly, this is because MT mostly uses government documents, which are freely and easily available in large volumes, to construct its databases. And govt documents don't often say things like, "The dog ate my sandwich.")

Susan


 

George Hopkins
Local time: 09:12
Swedish to English
Too true Susan ... Jul 23, 2009

As Susan stated: And govt documents don't often say things like, "The dog ate my sandwich".

And govt. documents don't ever say things like, "We're taking big chunks of your income in tax".
In other words they contain little straight to the point information.

In Sweden, the government and other authorities take about 70% of a person's income in the form of direct and indirect taxation.


 
Pages in topic:   [1 2 3] >


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


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

Why translators have nothing to fear from machine 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 »
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 »



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