| Pages in topic: [1 2] > | "Stanford system combines software with human intelligence to improve translation" - article Thread poster: Astrid_H
|
|---|
Astrid_H 
Germany
Local time: 23:59
German to English
+ ...
Link:
http://news.stanford.edu/news/2014/october/translate-human-machine-10-29-14.html
The gist: machine translation combined with human translation as developed at Stanford speeds up productivity.
I'm not familiar with current similar tools like Trados AutoSuggest, but if this new system is so much better, would you t... See more Link:
http://news.stanford.edu/news/2014/october/translate-human-machine-10-29-14.html
The gist: machine translation combined with human translation as developed at Stanford speeds up productivity.
I'm not familiar with current similar tools like Trados AutoSuggest, but if this new system is so much better, would you think this is good news for translators or bad?
I can imagine both positive and negative outcomes for translators. With higher productivity for a start, less translators will be needed for the same amount of words translated in the world. Will this hit the more qualified
translators harder than the less qualified or vice versa?
Thoughts?
Has someone by any chance tried out the online demo offered? ▲ Collapse
| | | | | Not a "revolution" | Oct 31, 2014 |
Just another "toy". Some like them, others do not.
The article does not give much details, but I am having hard time figuring out how different this new interface is from, say, Studio environment with proper (populated) term-base attached to it. AutoSuggest is something I have not used either. Moreover, you cannot even create those dictionaries in Freelance edition (you need Studio Professional).
In my case, properly maintained TM and term-base do the job just fine. H... See more Just another "toy". Some like them, others do not.
The article does not give much details, but I am having hard time figuring out how different this new interface is from, say, Studio environment with proper (populated) term-base attached to it. AutoSuggest is something I have not used either. Moreover, you cannot even create those dictionaries in Freelance edition (you need Studio Professional).
In my case, properly maintained TM and term-base do the job just fine. However, open mind is never a bad thing. ▲ Collapse
| | | | | A waste of time. | Oct 31, 2014 |
Really qualified, fully bilingual or trilingual translators do not need any of that. It will still be viewed as nothing but a hinderance to the translation flow.
| | | | | Participating ProZ members, please report | Oct 31, 2014 |
Merab Dekano wrote:
Just another "toy".
I'm not so sure. The article doesn't provide a lot of information, but it does refer to the research: http://www.spencegreen.com/pubs/green%20wang%20chuang%20heer%20schuster%20manning.emnlp14.pdf
And some of us gave it a try (even though the pay was waaaay too low):
For each language pair, we recruited 16 professional, freelance translators on Proz, which is the largest online translation community. We posted ads for both language pairs at a fixed rate of $0.085 per source word, an average rate in the industry. In addition, we paid $10 to each translator for complet- ing the training module. All subjects had significant prior experience with a CAT workbench.
Perhaps they can share their thoughts here.
Cheers,
Hans
[Edited at 2014-10-31 12:46 GMT]
| | |
|
|
|
| And by the way, | Oct 31, 2014 |
it's not Auto-Suggest, Auto-Write, Auto-Complete, or whatever the CAT tool vendors call it. It's more like Auto-Assemble as available in DejaVu and CafeTran, but on a much higher level. I think it's frightening.
Cheers,
Hans
| | | | Michael Beijer
United Kingdom
Local time: 22:59
Member (2009)
Dutch to English
+ ...
| A waste of your time maybe, but not of mine. | Oct 31, 2014 |
Thanks for the tip Astrid! Looks very interesting.
As usual, I will be analysing their approach to see if anything useful can be borrowed for use in CafeTran, which already has a rather interesting example of machine translation + human translation integration.
In fact, as far as I know, CafeTran is the only CAT tool apart from DVX2/3 to try to use MT output to improve its auto-assembly system.
See also: http://cafetranhelp.com/the-mt-services-tab
Michael
| | | | Jean Lachaud
United States
Local time: 17:59
English to French
+ ...
In plain English, it means: "Our Department needs financing."
Astrid_H wrote:
Thoughts?
| | | | Astrid_H
Germany
Local time: 23:59
German to English
+ ...
TOPIC STARTER | that's interesting! | Nov 1, 2014 |
JL01: Yes, which department doesn't nowadays?
Meta Arkadia wrote: And some of us gave it a try (even though the pay was waaaay too low): For each language pair, we recruited 16 professional, freelance translators on Proz, which is the largest online translation community. We posted ads for both language pairs at a fixed rate of $0.085 per source word, an average rate in the industry. In addition, we paid $10 to each translator for complet- ing the training module. All subjects had significant prior experience with a CAT workbench. Perhaps they can share their thoughts here. Cheers, Hans [Edited at 2014-10-31 12:46 GMT]
Ah, that is interesting. I didn't have enough time yesterday to read the research paper itself. I would be curious to hear about that - or request an online demo at some point as suggested in the article.
I think I'll also have a look at CafeTran, thanks for pointing it out!
I'm always curious about these things, so after thinking about it for a while, I don't think this new development is not necessarily just scary or dangerous, but it will quite certainly bring change to the business, if it is as good as it sounds. Maybe like horse carts being complemented by cars and then later replaced by them?
| | |
|
|
|
Astrid_H wrote:
I didn't have enough time yesterday to read the research paper itself.
Unfortunately, I have more than enough time. From the article (and I think this sort of summarises it):
2.1 UI Overview and Walkthrough
We categorized interactions into three groups: source comprehension: word lookups, source coverage highlighting; target gisting: 1-best translation, real-time target completion; target generation: real-time autocomplete, target reordering, insert complete translation. The interaction designs are novel; those in italic have, to our knowledge, never appeared in a translation workbench.
Some CAT tools are a bit more advanced than the authors think they are, but in general, they are right.
In my previous posting I sneered at the low rates they offered, after reading the article, I'd have participated for free. Very interesting.
Cheers,
Hans
[Edited at 2014-11-01 09:19 GMT]
| | | |
Astrid_H wrote:
...I don't think this new development is not necessarily just scary or dangerous, but it will quite certainly bring change to the business, if it is as good as it sounds.
It's a bit frightening, because it'll eat a bit more of our rates. But it's evolutionary. Tomorrow - or in a few years - there will be a revolution. It can be based on hardware, software, a better understanding on how the brain and language work, but we can wait for it. It will happen. Yes, Lilian, it will happen. Resistance is futile.
Cheers,
Hans
| | | | | Aware, but never worried | Nov 2, 2014 |
Meta Arkadia wrote:
It's a bit frightening, because it'll eat a bit more of our rates. But it's evolutionary. Tomorrow - or in a few years - there will be a revolution. It can be based on hardware, software, a better understanding on how the brain and language work, but we can wait for it. It will happen. Yes, Lilian, it will happen. Resistance is futile.
Believe me: despite the big advances since the likes of Saussure already a hundred years ago, Linguistics is really only scratching the surface of the very nature of language and how it works in the mind. Hundreds of magnificent scholars still don't know how a 3-year old child produces plenty of gramatical sentences with no effort at all. To sum up, what is not known about language cannot be added to automated translation, can it?
To me it seems that, faced with the reality that there could still be centuries before a machine can think like a person, those developing automatic translation are trying to solve the problem by brute force.
While I do think that an increasing number of work could potentially be done by automatic translation, professional translators will always be needed for the tricky stuff. Automatic translation can only make our life more interesting! Who wants to translate sentences like "To open a file, choose File + Open."?
| | | | Giles Watson
Italy
Local time: 23:59
Italian to English
In memoriam | Carts and horses | Nov 2, 2014 |
There is space for machine translation even in areas beyond the sort of grunt work where they are currently proving relatively useful.
However the trick will be to get MT to mimic the strategies used by human translators and not, as the article suggests, to employ humans to patch up the guff that algorithms churn out.
[Edited at 2014-11-02 21:51 GMT]
| | |
|
|
|
| Included in the study | Nov 3, 2014 |
Giles Watson wrote:
However the trick will be to get MT to mimic the strategies used by human translators and not, as the article suggests, to employ humans to patch up the guff that algorithms churn out.
Which is why:
Our backend innovations support the UI and enable feature-based learning from human corrections. In contrast, most previous work on incremental MT learning has focused on extracting new translation rules, language model updating, and modifying translation model probabilities (see: Denkowski et al. (2014a)).
Cheers,
Hans
| | | | Giles Watson
Italy
Local time: 23:59
Italian to English
In memoriam | why doesn't it work, then? | Nov 3, 2014 |
Meta Arkadia wrote: Giles Watson wrote:
However the trick will be to get MT to mimic the strategies used by human translators and not, as the article suggests, to employ humans to patch up the guff that algorithms churn out. Which is why: Our backend innovations support the UI and enable feature-based learning from human corrections. In contrast, most previous work on incremental MT learning has focused on extracting new translation rules, language model updating, and modifying translation model probabilities (see: Denkowski et al. (2014a)). Cheers, Hans
Hi Hans,
Your cart is still firmly in front of the horse if you are thinking in terms of "learning from human corrections". The difference between an effective translator and an indifferent one (or MT) is that good translators don't just tweak phrases: they reformulate thought in its entirety in new contexts, and generally have a wide range of non-obvious options to choose from.
So instead of getting human translators to correct MT, why not get an MT program to ask good translators about their translations while they are translating? You will need a computer that can handle multiple coffee breaks and endless mind-changing but the results might be interesting.
| | | | | Irreplaceable | Nov 3, 2014 |
Giles Watson wrote:
Your cart is still firmly in front of the horse if you are thinking in terms of "learning from human corrections".
That's how I learned my native tongue. My parents and others corrected me. Humans.
...good translators don't just tweak phrases: they reformulate thought in its entirety in new contexts, and generally have a wide range of non-obvious options to choose from.
I bet you haven't done much proofreading lately. Anyway, that's exactly what the software provides, options. And now it's learning how humans select them. I hope it will do better than the average texts I read on the Internet and even in quality newspapers, but I have little doubt.
Anyway, the consensus here is that human translators cannot be replaced, we are vastly superior, we charge at least € 0.20/word, we never ever give discounts, and we have lots of work. Me no.
Cheers,
Hans
[Edited at 2014-11-03 09:19 GMT]
| | | | | Pages in topic: [1 2] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » "Stanford system combines software with human intelligence to improve translation" - article | 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 » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
