I don’t offer Machine "Translation" Post-Editing, Here's Why... Thread poster: Jeff Whittaker
|
|---|
Jeff Whittaker 
United States
Local time: 08:07
Member (2002)
Spanish to English
+ ...
"Why editing machine "translation" is a hybrid approach that mixes the disadvantages of MT & human translation"
[Note: This is not my blog post, but with all the seminars and blogs about this, I believe that counterpoints are still relevant. Otherwise, some new translators may start believing that this is truly a viable and wise business practice just because it's the latest "trend"].... See more "Why editing machine "translation" is a hybrid approach that mixes the disadvantages of MT & human translation"
[Note: This is not my blog post, but with all the seminars and blogs about this, I believe that counterpoints are still relevant. Otherwise, some new translators may start believing that this is truly a viable and wise business practice just because it's the latest "trend"].
http://www.at-it-translator.com/i-dont-offer-machine-translation-post-editing-here-is-why/
[Edited at 2015-04-02 15:02 GMT] ▲ Collapse
| | | |
Patrick Porter
United States
Local time: 08:07
Spanish to English
+ ...
| agree, but... | Apr 2, 2015 |
I generally agree with the linked post for the most part and never take so-called "post-editing" jobs mainly for the same reasons. But I disagree with the author that it never saves time and always creates inefficiencies. I use machine-assisted translation in my work very frequently (not Google Translate or any others like it, mind you, but my own tools trained with my own linguistic resources). For some types of texts it's very useful and I've had significant productivity gains. For other j... See more I generally agree with the linked post for the most part and never take so-called "post-editing" jobs mainly for the same reasons. But I disagree with the author that it never saves time and always creates inefficiencies. I use machine-assisted translation in my work very frequently (not Google Translate or any others like it, mind you, but my own tools trained with my own linguistic resources). For some types of texts it's very useful and I've had significant productivity gains. For other jobs it's better not to use at all and translate from scratch.
The linked post is an opinion piece seemingly based on the author's anecdotal experience with machine-assisted translation and personal findings that it isn't worthwhile. Meanwhile there are actual empirical studies that show it can increase productivity (here is one example). Now, you could argue that academic research like this doesn't mean anything in a real production situation, but that doesn't disprove anything either. I think the blog author's points are valid and need to be raised, but the conclusion could be a little more nuanced.
In my opinion the problem with the way things are playing out right now is LSPs and clients are approaching the issue of MT from the wrong angle. If they can offer linguistic tools and resources that increase a translator's productivity then market forces will take their natural course and they will achieve savings that way. Instead, my experience to date has been some companies trying to foist so-called post-editing jobs on me and asking for an unjustifiably large discount in proportion to the real potential productivity gain and real value added. Fortunately most of my clients don't worry about the exact details of my production process; they are just happy that I can provide them an expert level of quality with good turnaround times at fair rates. I hope it stays that way.
[Edited at 2015-04-02 21:48 GMT] ▲ Collapse
| | | |
| Star Trek is coming | Apr 2, 2015 |
About 30 years years ago my brother, who was by then a professional system analyst, writing computer programs for big companies, told me that the Internet, which was just starting up, was just a "toy" with no chance of surviving. Look now!
The way we are talking about MT makes me remind those visionary words of my brother then. No, it can't replace us translators of flesh and blood yet, but what in about 5, 10 or 15 years?
I fear that the sympathetic self thinking Star ... See more About 30 years years ago my brother, who was by then a professional system analyst, writing computer programs for big companies, told me that the Internet, which was just starting up, was just a "toy" with no chance of surviving. Look now!
The way we are talking about MT makes me remind those visionary words of my brother then. No, it can't replace us translators of flesh and blood yet, but what in about 5, 10 or 15 years?
I fear that the sympathetic self thinking Star Trek computer will become reality at the end, and that we will be out of a job, it is just a question of time.
My grandfather grew up with horse and cart, and saw the first man landing on the moon. All I want to say, a lot can happen in a lifetime. So, is it that unbelievable that we are going to be 'outjobbed' by the first intelligent computer?
Or maybe it is just a family attitude?
[Edited at 2015-04-02 21:44 GMT]
[Edited at 2015-04-02 21:45 GMT] ▲ Collapse
| | | |
Dan Lucas
United Kingdom
Local time: 13:07
Member (2014)
Japanese to English
| MT goes back a long, long way | Apr 2, 2015 |
Robert Rietvelt wrote:
No, it can't replace us translators of flesh and blood yet, but what in 5, 10 or 15 years?
The internet, to which you refer in your comment, proved itself indispensable in a very short time - it over-delivered massively. In contrast, machine translation has been around for half a century and has under-delivered for most of that time.
The versions of MT that most of us see are still little more than a gimmick for serious work. It's useful for anybody just wanting to get the gist of something - I use GT myself - but given a choice, would you trust it to translate something important? Your child's medical report, shall we say?
And even Google admits that the approach it has used thus far has gone about as far as it can go:
Andreas Zollmann, who has been researching in the field for many years and working at Google Translate for the last year, suggests, along with Blunsom, that the idea that more and more data can be introduced to make the system better and better is probably a false premise. "Each doubling of the amount of translated data input led to about a 0.5% improvement in the quality of the output," he suggests, but the doublings are not infinite. "We are now at this limit where there isn't that much more data in the world that we can use," he admits.
So the law of diminishing returns kicked in some time ago for GT. Sure, GT will get better incrementally, but it needs a dramatic improvement to impress with the kind of texts most of us are translating.
If you're a translator who's just churning out short strings of text in a language pair that is more amenable to MT then I concur that you may be justified in feeling worried.
Having said all that, personally I agree with Patrick's earlier comment and feel that some form of MT could offer significant advantages to those translators willing to invest time in the proper training and feeding of a system.
We should be thinking about how to co-opt new tools rather than assuming that they're the 21st century equivalent of the flying shuttle or the spinning jenny.
Regards
Dan
| | |
|
|
|
| Not quit the answer I expected, but never mind | Apr 2, 2015 |
I never said MT doesn't offer significant advantages, and yes we should invest time in the proper training to understand the (new) system(s) and learn it to use it for our own advantage. I totally agree with you, but translation will never be the same. On the other hand, the future is inevitable, and 'Star Trek' (sorry for the metaphor) will track us down eventually, that is what I think (sorry). So.....? What does the future for a translator looks like?
[Edited at 2015-04-02 22:23 GMT]... See more I never said MT doesn't offer significant advantages, and yes we should invest time in the proper training to understand the (new) system(s) and learn it to use it for our own advantage. I totally agree with you, but translation will never be the same. On the other hand, the future is inevitable, and 'Star Trek' (sorry for the metaphor) will track us down eventually, that is what I think (sorry). So.....? What does the future for a translator looks like?
[Edited at 2015-04-02 22:23 GMT]
[Edited at 2015-04-02 22:24 GMT]
[Edited at 2015-04-02 22:25 GMT]
[Edited at 2015-04-02 22:25 GMT] ▲ Collapse
| | | |
Jeff Whittaker
United States
Local time: 08:07
Member (2002)
Spanish to English
+ ...
TOPIC STARTER | Star Trek also has the transporter... | Apr 2, 2015 |
but I don't see that invention coming true anytime soon (if fact, you would have to disintegrate your existing body and recreate yourself on the other end). We will come up with a new way to travel, but it may be something we never envisioned. (remember all the push buttons in Star Trek - they never thought about flat touch screens, or the internet).
Besides, if computers ever do reach the point where they could understand context and make emotional inferences (and let's remember th... See more but I don't see that invention coming true anytime soon (if fact, you would have to disintegrate your existing body and recreate yourself on the other end). We will come up with a new way to travel, but it may be something we never envisioned. (remember all the push buttons in Star Trek - they never thought about flat touch screens, or the internet).
Besides, if computers ever do reach the point where they could understand context and make emotional inferences (and let's remember that even human translators disagree about the translations of some terms), then most other people would also be out of a job. At this point computers would have to understand what it means to be human and what it means to interact and live in a human society in order to understand when a suggested translation doesn't make sense.
Add advanced robotics to that, and pretty much everyone is out of a job - actors, painters, doctors, lawyers, check-out clerks, chefs, managers, teachers... even website designers and computer programmers.
If you read the book Future Hype, it becomes evident that we cannot predict the future based on the perceived further advancement of existing technology (just because a system operates at level "x" now, does not mean that it will continue to evolve exponentially in the future). Technology is usually only evident in hindsight. In fact, there is evidence that statistical based translation (such as GT) will actually get worse over time as more and more machine translated content gets mixed in with and eventually statistically supercedes the human translations.
What we could see in the future are better human translator interfaces that would connect us with a wider range of terminology tools and existing translations (think a very sophisticated and extremely comprehensive (global) Linguee that will also remember our past translation choices and preferences and allow us to collaborate with others in real time).
Right now there is a lot of redundant and wasted space in the TM area in the misguided attempt to save every possible form of every possible sentence in the crazy hope that at some point, no sentence will be left unsaved, even though as linguists we know that this is an impossible, futile and infinite task, and 95% or more of these matches will never be used again (and in another 20 years, the "memories" that we've wasted decades creating, will no longer be compatible with the then-technology. It will be like having translations on micro-fiche or punched data cards).
Robert Rietvelt wrote:
I never said MT doesn't offer significant advantages, and yes we should invest time in the proper training to understand the (new) system(s) and learn it to use it for our own advantage. I totally agree with you, but translation will never be the same. On the other hand, the future is inevitable, and 'Star Trek' (sorry for the metaphor) will track us down eventually, that is what I think (sorry). So.....? What does the future for a translator looks like?
[Edited at 2015-04-02 23:37 GMT] ▲ Collapse
| | | |
Phil Hand
China
Local time: 20:07
Chinese to English
| Star Trek and Microsoft | Apr 3, 2015 |
I pretty much agree with everyone on the thread.
Jeff's point:
if computers ever do reach the point where they could understand context and make emotional inferences...pretty much everyone is out of a job
is one I've made many times. In a way, translation is among the jobs which are best shielded from machine takeover, because on the day a computer can reliably understand a text, it will be able to do virtually any human job better than a human.
I like the linked blogger's way of putting it very much: "mixes the disadvantages of both MT and human translation". It's a pithy way of saying what I often feel about PEMT jobs.
But Patrick's point is a good one, too. Our CAT tools could definitely be smarter. At the moment, most CATs offer only two levels of analysis: the segment and the word. A smarter type of CAT tool could mine TMs for phrases, and try to recognise where similar phrases or rhetorical structures are being used. And there is definitely more room for importing smarter dictionaries and resources - I understand this is happening in Europe with EURLEX databases, but for many language pairs, there's nothing. A semi-smart tool should mine Amazon for the names of translated books; index standard/famous translations of classics; official translations of laws; etc. I wouldn't call this MT, I'd call it smart CAT, but it doesn't really matter what name you use, it would be a big help.
Having said that, I do think that Star Trek will come, and I actually think that Microsoft's Skype interpreter is the start of it. Not because it's any good now, but because it will learn better how to understand language as she is spoken. In particular, it will learn better how to analyse sentences into phrases. More voice interactions with robots will force robots to learn better how to extract meaning from the phrases, and eventually you get to proper understanding of language... but it might never happen. We might end up never really talking to our robots, just giving them one word commands. In which case, they'll never learn. Hard to predict!
| | | |
Samuel Murray
Netherlands
Local time: 14:07
Member (2006)
English to Afrikaans
+ ...
Basically the author of that blog is saying that he refuses to do PEMT because he believes it is unethical, because it produces a lower quality translation, and clients deserve better. Oh, and he's afraid that he might miss a mistake, which would result in liability for him.
But generally, he doesn't refuse because it pays less, or because it is tiring, or because it hurts his business, or because it hurts the industry. He mostly feels that it is immoral to offer clients a slight reduction in price for a disproportionately large reduction in quality.
The problem with this argument is that PEMT jobs that involve Google Translate typically involve very low rates anyway, and that says nothing about how "good" the translators are who accept such jobs (only optimistic translators believe that the better the translator, the more and the better paying jobs he'll get).
However, I do like the quotable quote he's trying to make (which you are also quoting): In an ideal world, PEMT does not give much lower prices for slightly lower quality, but slightly lower prices for much lower quality. Will this discourage clients from using PEMT? No, some clients who do not believe that premium quality is required will always be looking to save money, even if it's only pennies. They may be wrong about whether premium quality is required, but that will take more convincing than a quotable quote.
[Edited at 2015-04-03 08:47 GMT]
| | |
|
|
|
Alex Lago
Spain
Local time: 14:07
English to Spanish
+ ...
I always find this subject fascinating and it is really interesting to see everyone’s different points of view.
Personally I am of the opinion that one day (problem is when?) we will in fact reach the singularity (for those of you that don't know what this is have a look at http://en.wikipedia.org/wiki/The_Singularity_Is_Near ), once this happens all bets are off for ev... See more I always find this subject fascinating and it is really interesting to see everyone’s different points of view.
Personally I am of the opinion that one day (problem is when?) we will in fact reach the singularity (for those of you that don't know what this is have a look at http://en.wikipedia.org/wiki/The_Singularity_Is_Near ), once this happens all bets are off for everyone.
On a more "current" basis, there are new technologies and new algorithms that are changing the world on a daily basis, just look at this TED video (http://www.ted.com/talks/fei_fei_li_how_we_re_teaching_computers_to_understand_pictures ) and you will see that at the end of 2014 they managed to teach a computer how to start understanding what it sees in a photo, they had been trying for 15 years. Just six months ago a computer could not do this and now it can and it will only get better at it.
As regards MT, sure current MT "methodologies" seem to have reached a plateau as mentioned by the Google exec but that will just make people look for different approaches.
The main reason I believe this will happen is because we need it, the world is increasingly connected, companies and individuals are generating content at an astronomical rate, just look at Youtube, according to their own statistics (https://www.youtube.com/yt/press/statistics.html ) people upload 300 hours of video every minute, sure not all those hours include speech but you get the picture (pun intended), and that's just Youtube, never mind Facebook, Twitter, Pinterest, and on and on and on and on and on....
People want to connect to people and there aren't enough translators and interpreters in the world to handle this enormous amount of data, MT would help enormously which is why enormous amounts of money are being (and will continue to be) spent on it, the only question in my opinion is not if but when, but like I said once the singularity comes along all bets are off. Just hope it comes along soon enough so I can upload myself.
[Edited at 2015-04-03 14:38 GMT] ▲ Collapse
| | | |
Login to reply/comment 
