Comparing machine translation and fuzzy matches Autor wątku: pcovs
|
|---|
pcovs
Dania
Local time: 05:53
angielski > duński
It is really interesting to read how many translators fiercely fight the mere thought of post-editing machine translated texts, while we see more and more translators letting their clients force them into exploiting fuzzy match schemes, which - looking at it differently - may in some language pairs and text types compare more or less directly to the quality of a machine translated text.
Just yesterday, I received yet another offer to letting myself be robbed blind by a client reques... See more It is really interesting to read how many translators fiercely fight the mere thought of post-editing machine translated texts, while we see more and more translators letting their clients force them into exploiting fuzzy match schemes, which - looking at it differently - may in some language pairs and text types compare more or less directly to the quality of a machine translated text.
Just yesterday, I received yet another offer to letting myself be robbed blind by a client requesting a fuzzy grid including these two discounts:
a- 75-84% at 50% of my per word rate
b- 50-74% at 70% of my per word rate
(to me, the latter is more or less directly comparable to MT in quality, although the things to correct may be of a different kind, but I think the workload is just about the same.)
To me, this is obvious robbery, and I turned it down flat.
I don't accept fuzzy matches.
It just started me thinking about how I should be calculating rates for post-editing, should I come across a sample actually worth working on, and how this should perhaps affect my take on fuzzy match discounts. ▲ Collapse
| | | | | There is market for fuzzies | Jun 10, 2011 |
I just wanted to point out that there are markets where fuzzies are a natural part of the process. Imagine for instance a big technical manuals about machinery: when a customer makes some upgrades in an existing model, as well as some corrections in the source text in a new edition of the manual, they don't want to pay a retranslation of the whole manual. It is either with a discount in the fuzzies, or no order...
In ... See more I just wanted to point out that there are markets where fuzzies are a natural part of the process. Imagine for instance a big technical manuals about machinery: when a customer makes some upgrades in an existing model, as well as some corrections in the source text in a new edition of the manual, they don't want to pay a retranslation of the whole manual. It is either with a discount in the fuzzies, or no order...
In technical translation we have plenty of manuals like this, and a lower rate for the fuzzies just feels natural. Of course we must be vigilant so that the effort spent adapting the fuzzies matches the rate you are getting for that work. ▲ Collapse
| | | | cranium
francuski > angielski
+ ...
| Excellent point | Jun 10, 2011 |
Thanks for providing an additional argument to use when explaining my refusal of sliding fees.
| | | | pcovs
Dania
Local time: 05:53
angielski > duński
NOWY TEMAT | Yes, I know there is a market and why | Jun 10, 2011 |
My point was more along the lines of:
I consider the quality of MT at this stage to be more or less directly comparable to a fuzzy match of 50-74% in my language pair and my areas of expertise. This is based on many years of experience.
On the assumption that this is a rather common comparison to be made, howcome some would be willing to take a steep discount for the fuzzy match but refuse the post-editing?
Is the "willingness" to take the steep rate cuts only bas... See more My point was more along the lines of:
I consider the quality of MT at this stage to be more or less directly comparable to a fuzzy match of 50-74% in my language pair and my areas of expertise. This is based on many years of experience.
On the assumption that this is a rather common comparison to be made, howcome some would be willing to take a steep discount for the fuzzy match but refuse the post-editing?
Is the "willingness" to take the steep rate cuts only based on the fact that our clients have made us believe, that our investment in sometimes rather expensive translation software should only benefit them and the end clients financially and not ourselves, and that machine translation is "the new kid on the block" that noone really wants to play with yet? ▲ Collapse
| | |
|
|
|
Jaroslaw Michalak 
Polska
Local time: 05:53
Członek ProZ.com
od 2004
angielski > polski
SITE LOCALIZER
PCovs wrote:
My point was more along the lines of:
I consider the quality of MT at this stage to be more or less directly comparable to a fuzzy match of 50-74% in my language pair and my areas of expertise. This is based on many years of experience.
On the assumption that this is a rather common comparison to be made, howcome some would be willing to take a steep discount for the fuzzy match but refuse the post-editing?
In my opinion your point of view is a bit distorted by the offer you quote. The terms are not usual by any means - out of curiosity I've checked a few offers I've been sent myself (i.e. not those I've made) and typically agencies propose 70-80% of rate for 75-99% fuzzies and 80-100% for 50-74% (most of them did not include them in the grid, so it's 100%).
On the other hand, posters of MT "editing" jobs really expect the rate to be close to that for editing/proofreading of human translation, i.e. 30-50%. Obviously, there is no middle ground between 30-50 they expect and 80+ you might expect based on your assesment of the MT quality...
| | | | Tom Fennell
USA
Local time: 22:53
rosyjski > angielski
+ ...
| Concur with Jabberwock | Jun 10, 2011 |
His analysis seems most sound.
The issue is that those pushing MT cannot just assert the equivalence of editing machine and human translation - they need to prove it statistically (or through the market).
One must also factor in quality.
Those ordering MT post editing at 70 pct discount may only want "bare intelligibility" not "a fine translation" as the result.
The thing is that this result may be obtained by someone with very little source l... See more His analysis seems most sound.
The issue is that those pushing MT cannot just assert the equivalence of editing machine and human translation - they need to prove it statistically (or through the market).
One must also factor in quality.
Those ordering MT post editing at 70 pct discount may only want "bare intelligibility" not "a fine translation" as the result.
The thing is that this result may be obtained by someone with very little source language experience who will work for cheap cheap cheap. For them this is good work.
An experienced translator simply cannot produce such a crude translation - it is too psychologically burdensome. It is like asking a gourmet chef to serve up greasy sausages and dry potatoes with a really pasty sauce while at the same time producing masterpieces for other clients.
The mind doesn't work that way.
The bottom line is words per hour. An experienced translator may be able to produce a fine translation 20% faster with MT for certain texts, especially repetitive simple lists. For more complex language the additional mental strain of restructuring faulty MT syntax just isn't worth it.
The thing to remember is that in our global society, the volume of translations is growing so enormously that there is plenty of room for the bottom feeders, including crowdsourcers who do certain work for free.
But spot on translations will always be needed for contracts, important manuals, annual reports, medical studies, marketing materials etc etc. And the volume of these is growing, growing, growing.
MT will become more and more integrated into CAT tools, but this will take many more years.
The critical thing is for all translators to track their hourly earnings vigilently!
[Edited at 2011-06-10 11:03 GMT]
[Edited at 2011-06-10 11:04 GMT] ▲ Collapse
| | | | Samuel Murray
Holandia
Local time: 05:53
Członek ProZ.com
od 2006
angielski > afrikaans
+ ...
| Sighted or non-sighted fuzzies | Jun 10, 2011 |
PCovs wrote:
...we see more and more translators letting their clients force them into exploiting fuzzy match schemes, which ... may in some language pairs and text types compare more or less directly to the quality of a machine translated text.
Yes, this may be true for unsighted fuzzy matches (i.e. fuzzy matches in which the client does not show you what the original source text for the matching segment is). But for sighted fuzzy matches (i.e. you can see both the source text of the fuzzy match, so that you can compare it with your current source text), one has the assurance (generally) that the fuzzy match is a good translation of its own source text, so determining which parts need fixing is easier.
a- 75-84% at 50% of my per word rate
b- 50-74% at 70% of my per word rate
No, that is just silly. Anything below 75% should be paid in full. In fact, ideally anything below 85% should be paid in full. A 51% match may be useful sometimes but it won't save you 1/3 of the time to translate the text by hand.
It just started me thinking about how I should be calculating rates for post-editing, should I come across a sample actually worth working on, and how this should perhaps affect my take on fuzzy match discounts.
Good thinking... but if the fuzzy matches are unsighted, then I think you should charge your full per-word rate for it (or even more, since the client is expecting you to evaluate the entire segment *and* remain consistent with it).
| | | | | The days of TM are numbered | Jun 18, 2011 |
I was just going to post a new entry on this forum and I saw PCovs entry on more or less the same subject! amazing
anyway my entry was:
Exceptionally this week I had a document to translate with Trados that was completely unconfidential as the information was already published on the web and also accompanied by a giant translation memory whichI was asked to use . This gave me an opportunity to do a small test. Which works faster and is more useful - Google translate or a big ... See more I was just going to post a new entry on this forum and I saw PCovs entry on more or less the same subject! amazing
anyway my entry was:
Exceptionally this week I had a document to translate with Trados that was completely unconfidential as the information was already published on the web and also accompanied by a giant translation memory whichI was asked to use . This gave me an opportunity to do a small test. Which works faster and is more useful - Google translate or a big TM. I found that Google translate allowed me to go faster than using the TM. This might not have been the case if there had been lots of 100%matches, though I am one of those translators who fiddles with 100% matches. For this document there were hundreds of fuzzies. What I did find annoying though, as you have all mentioned above, is how bad 70-80 % fuzzy matches with a TM are, confirming my refusal to accept any reduction in my rate for them. Goggle translate gave more "useful" and accurate translations than the translation memory for a large number of fuzzies.
My conclusion is that there is not really such a big difference between a translation memory and Google translate as they both use algorithms to suggest translations . Moreover it seems that the Google algorithm is far better than the Trados one and also that the vast number of translations harvested by Google give it an edge over even the largest translation memory.
NB: This was for the FR-EN language pair. In my opinion GT does not cope with other language pairs (IN-EN for example) as well.
and my conclusion is that for an experienced translator MT is more useful than TM
I don't use google but use a desk-based system. A TM is only useful for retaining all your own translations
[Edited at 2011-06-18 06:50 GMT] ▲ Collapse
| | |
|
|
|
Neil Coffey
Wielka Brytania
Local time: 04:53
francuski > angielski
+ ...
| Interesting observation that may apply in *some* cases | Jun 18, 2011 |
According to one line of thinking, there would ultimately be little difference between a "fuzzy match" algorithm and a "machine translation" algorithm, although we're not at that stage yet.
I know this depends heavily on the particular text and field, but my observations would just be:
- people don't half bang on about "fuzzy matches" so they must be terribly terribly useful
- in many cases, machine translation is extremely unuseful (though I grant it does vary)... See more According to one line of thinking, there would ultimately be little difference between a "fuzzy match" algorithm and a "machine translation" algorithm, although we're not at that stage yet.
I know this depends heavily on the particular text and field, but my observations would just be:
- people don't half bang on about "fuzzy matches" so they must be terribly terribly useful
- in many cases, machine translation is extremely unuseful (though I grant it does vary)
So I infer that either (a) fuzzy matches really aren't quite as useful as people bang on about them being, or (b) MT and fuzzy matches are still quite different beasts, and the latter tend to be more useful in a significant number of cases.
[Edited at 2011-06-18 17:49 GMT] ▲ Collapse
| | | | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Comparing machine translation and fuzzy matches | Protemos translation business management system | Create your account in minutes, and start working! 3-month trial for agencies, and free for freelancers!
The system lets you keep client/vendor database, with contacts and rates, manage projects and assign jobs to vendors, issue invoices, track payments, store and manage project files, generate business reports on turnover profit per client/manager etc.
More info » |
| | 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 » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
