| Pages in topic: < [1 2 3] > | Agencies wanting discounts for post editing machine translation Thread poster: Jeff Whittaker
|
|---|
Phil Hand 
China
Local time: 11:08
Chinese to English
Great quote in the blog post that Emma linked to (http://signsandsymptomsoftranslation.com/2012/12/11/post-analysis/):
"...I, or a trusted colleague, has already checked it. This, in a nutshell, is the justification for repetition discounts: when TUs have already been checked by a human being..."
Trust gets to the heart of proofreading, TM discounts and post-editing.
You have to trust the translator to be able to proofread effectively, otherwise you put in as much effort checking as you would translating.
You have to trust the TM, otherwise you have to go over every 100% match or 95% match to make sure its right, and there's no time saving.
For post-editing to work, you'd have to trust the computer, trust that when the MT target segment is in good English, it actually means what the source segment means. Otherwise you have to check everything, and there's no time saving.
That's why I don't do proofreading and rarely offer TM discounts: in my pair the average quality is terrible, and I have no trust in my colleagues.
But no-one trusts a computer. They still get it wrong all the time, really basic stuff. So it's absurd to suggest that we could accept lower rates for post-editing.
| | | | | My own observations | Dec 22, 2012 |
Phil Hand wrote: John Fossey wrote:
...The main problem I find with MT is that when my mind should be clear to capture the meaning and render it faithfully, a MT text tends to circumscribe my thinking and make it difficult to think outside the box it has already presented. Even when the MT translation is completely wrong it sometimes takes a second look at the source text to realize that it's wrong.
I can hardly see where the savings is with MT for serious translation... So true. There have been a couple of studies done on the use of MT, but they're very limited, and didn't involve professional translators, so far as I can tell. I'd love to see a properly controlled study done, because I honestly believe that MT will make a real translator worse and slower. If someone does the empirical work, and I'm wrong, then I'll listen. But my personal experience is that it's a hindrance, not a help. I'm sure that there is a place for MT. But at the moment, the place is not in the workflow of professional translators.
I'm not sure about larger studies, but I have noticed in my own work that when ever I work with MT (willingly or otherwise) my quality drops dramatically, as does my speed. I used to think that MT was a great new tool, but I now avoid it like the plague.
| | | | Gyula Erdesz
Hungary
Local time: 05:08
Member (2009)
English to Hungarian
+ ...
| The right product for the right purpose | Dec 22, 2012 |
Heike Behl, Ph.D. wrote:
- what texts don't deserve the highest quality of translation? If a text is that unimportant, why bother at all? Do you have any examples of texts that don't deserve the best translation quality? How do you determine the quality level they do deserve?
If I say that (machine translation + post editing) has the right place in our business, I naturally don't think of academic papers, patterns, marketing articles or user manuals of power plants. Where you need polished, 99% error-free translation, MT+PE is out of the question.
But there are some situations where approximated translations do have justification for existence. Just imagine a global company that is represented in 100 countries around the world and receive customer complaints in 80 languages. How can you convert these information to a format that is understandable for the central development team? You need to find an easy, partly automated solution which is able to convert the core message of these complaints, but you are not interested in the style of the writer at all. You need information and not accurate translation.
In some cases some information with potential errors is much more than no information at all.
Heike Behl, Ph.D. wrote:
The end customer (and maybe also the agencies selling post-edited MT tanslations) believe that because the translation has been touched by a human, it is "acceptable".
You are totally right.
Clients need to know what they get from their money and what could expect from a machine translation. But this is true for practically any products in our times: if you buy something for the wrong purpose, you will be upset, although the product itself might be appropriate.
Regards,
Gyula
[Módosítva: 2012-12-22 05:57 GMT]
| | | |
Gyula Erdész wrote: Heike Behl, Ph.D. wrote:
- what texts don't deserve the highest quality of translation? If a text is that unimportant, why bother at all? Do you have any examples of texts that don't deserve the best translation quality? How do you determine the quality level they do deserve? If I say that (machine translation + post editing) has the right place in our business, I naturally don't think of academic papers, patterns, marketing articles or user manuals of power plants. Where you need polished, 99% error-free translation, MT+PE is out of the question.
Novels, poems, marketing, etc. - texts that are creative, use metaphoric language and can jump across a range of areas - are the worst candidates for MT. Texts that are semantically relatively unambiguous, syntactically straight forward and focus on a limited subject area are best suited. This is particularly true if a subject-specific and well-maintained dictionary is provided. Patents, user manuals - even for power plants -, certain types of academic papers, etc. are much, much better suited for MT than you think.
Without any data to back it up but based on my extensive experience with MT (I used to be "on the dark side," working for one of the companies developing MT), I would say that medical texts are probably among the best suited candidates for MT. Prerequisite is an excellent dictionary, if possible for the particular subject area within the medical field. Most words tend to have a very specific meaning, the language tends to be straight forward, the writing clear. Of course, this is also a field where MT errors could be fatal and first-rate human post-editing would be a must.
Particularly, if a source text is written with MT - and even better, a particular specialized MT system and dictionary - in mind, MT does actually have a pretty good chance of delivering useful translations. In such an environment, maybe also with the goal of improving the MT system and dictionary in collaboration with engineers or R&D, I wouldn't mind working together with a company, as a writer, post-editor and/or consultant. This would be a great challenge.
Gyula Erdész wrote:
But there are some situations where approximated translations do have justification for existence. Just imagine a global company that is represented in 100 countries around the world and receive customer complaints in 80 languages. How can you convert these information to a format that is understandable for the central development team? You need to find an easy, partly automated solution which is able to convert the core message of these complaints, but you are not interested in the style of the writer at all. You need information and not accurate translation.
In some cases some information with potential errors is much more than no information at all.
Your example of customer complaints and similar texts such as survey answers, emails, etc. are actually the worst suitable candidates for MT. The writers of these types of texts are generally no schooled writers, not necessarily linguistically up to par, don't usually care about typos or, for example, proper capitalization in German (without which dictionary lookup, disambiguation and parsing would be nightmare for any MT system), grammatically complete and correct sentences, punctuation, etc. They use colloquial language or even make up their own words. Generally, not a chance for MT. These types of texts lack everything that MT needs.
The problem of most agencies offering MT "solutions" to their customers is that they only use a one-size-fits-all system. They might offer some kind of subject-specific dictionaries, but even those are hit or miss if you want to have a specific text translated. If I'm not mistaken, at least one company offers user-editable subject dictionaries, but in order to create useful dictionaries for MT, you need to know at least your parts of speech, some grammar, depending on language more or less extensive inflection patterns, etc. I doubt that many customers go through the trouble or are even capable of properly editing their own user dictionaries.
And the texts usually offered to translators for post-editing are the results of this one-size-fits-all approach. Of course, the MT results are at their worst, the work for the post-editor pretty much impossible and extremely time-consuming, and probably the end customer - should they ever get feedback on their MT translation - will not be happy with the results. No translator worth their money and proud of delivering first-rate translations should consider working in such a context.
| | |
|
|
|
Thayenga
Germany
Local time: 05:08
Member (2009)
English to German
+ ...
| Well said, Jeff! | Dec 22, 2012 |
Your reply to that company was down to the point. I couldn't have said it better!
When I started out "ages" ago, I did do a few post editing jobs - granting no discounts! - until I got so tired of editing this, well, yes, at times absolute nonsense, aside from the hearty laugh it provided me with from time to time, that I simply stopped.
And, Jeff, that the company called you "narcissistic" seems to be rude only at first glance. Their "term" for you does, in fact, only... See more Your reply to that company was down to the point. I couldn't have said it better!
When I started out "ages" ago, I did do a few post editing jobs - granting no discounts! - until I got so tired of editing this, well, yes, at times absolute nonsense, aside from the hearty laugh it provided me with from time to time, that I simply stopped.
And, Jeff, that the company called you "narcissistic" seems to be rude only at first glance. Their "term" for you does, in fact, only confirm that you take pride in your hard work and in all the studies that have brought you to where you are today: being a professional translator with a strong sense of responsiblity toward your clients. Simple as that. ▲ Collapse
| | | | Gyula Erdesz
Hungary
Local time: 05:08
Member (2009)
English to Hungarian
+ ...
| Different anwers to the same phenomenon | Dec 22, 2012 |
Heike Behl, Ph.D. wrote:
Your example of customer complaints and similar texts such as survey answers, emails, etc. are actually the worst suitable candidates for MT.
My example was not fictive:
http://www.youtube.com/watch?v=h1qm7_ILTX8
Heike Behl, Ph.D. wrote:
No translator worth their money and proud of delivering first-rate translations should consider working in such a context.
As Emma Goldsmith wrote:
...we should acknowledge this and not bury our heads in the sand, hoping it will go away, because it won't.
So, the direction is more or less given and it is up to you how to adapt to it. Being proud is one of your options. Getting the most out of it is another one. I personally vote for the latter.
Regards,
Gyula
[Módosítva: 2012-12-22 08:43 GMT]
| | | | Jeff Whittaker
United States
Local time: 23:08
Member (2002)
Spanish to English
+ ...
TOPIC STARTER | What customers want | Dec 22, 2012 |
From another thread:
"I receive more and more calls every day from individuals actively seeking out individual translators rather than agencies because they are very concerned about their confidential documents being put through Google or sent to some unknown translator in cyberspace. Today's customers want to know who is translating their documents. I think soon agencies will have to adapt by posting profiles of their translators on websites, asking translators to sign translations, etc. ... See more From another thread:
"I receive more and more calls every day from individuals actively seeking out individual translators rather than agencies because they are very concerned about their confidential documents being put through Google or sent to some unknown translator in cyberspace. Today's customers want to know who is translating their documents. I think soon agencies will have to adapt by posting profiles of their translators on websites, asking translators to sign translations, etc. In short, customers want more personal service and more control over the process."
And to add a bit of levity to the discussion since some of us are now on vacation:
Here is the song Jingle Bells translated into Italian by Google:
http://www.youtube.com/watch?feature=player_embedded&v=xwV5iOcHf8k#!
[Edited at 2012-12-22 14:54 GMT] ▲ Collapse
| | | | | A little math | Dec 23, 2012 |
Thinking about these issues, I’ve been observing myself closely during my translation work recently, trying to work out what exactly it is that I do for a living. It’s pretty variable from text to text, but, on average, I’d say it breaks down into four roughly equal shares (in terms of word count, not work):
1) Lowest level – more or less word for word, typing over the source text in a straight line. Here fragments from MT could potentially be useful (save me typing), and ap... See more Thinking about these issues, I’ve been observing myself closely during my translation work recently, trying to work out what exactly it is that I do for a living. It’s pretty variable from text to text, but, on average, I’d say it breaks down into four roughly equal shares (in terms of word count, not work):
1) Lowest level – more or less word for word, typing over the source text in a straight line. Here fragments from MT could potentially be useful (save me typing), and apart from some language-specific differences there’s a 1-to-1 correspondence between Source and Target.
2) Lower middle level – Encountering areas where things are said rather differently in the Target compared to the Source – e.g. different word order, different idioms, constructions that do not map 1-to-1 because Source uses one type of grammatical construction where Target prefers another etc. More advanced MT could deal with a lot of this (i.e. if it could move on from linearity and 1-to-1). Very occasionally, I notice that GT will come up with a whole phrase of 2 or 3 words in Target which map an analogous phrase in Source, so there’s potential here (albeit years in the future).
3) Upper middle level – The text I’m producing in Target hardly maps onto Source at all. If you broke down Source and Target into semantic or grammatical subsegments, the mapping from Source to Target would have a lot of crossing lines, and the corresponding subsegments would not be direct translations of each other because of the fragmented syntax. There is still some correspondence, though, because by letting my eyes jump up and down, left and right in the Source text, I can still clearly proofread that my Target text has all the salient points in it.
4) Highest level – I don’t know what happens here. All I can say is that I swallow the meaning of the Source whole, sit back for a minute, then write it completely differently (with the same sense) in the Target text. This is beyond algorithmic processing, there are pretty much no corresponding subsegments in Source and Target, only isolated terms, and the only way I can check that the translation is correct is by “mystically” stepping into the perspective of the Source, then the Target, and checking that I am in the “same position” in both cases. This type of translation can occur at sub-sentence level in routine texts, but may involve whole paragraphs in marketing or journalistic texts. The “feel” of this work is akin to interpreting, where you absorb a whole 10 second utterance and re-create it without remembering the 1-to-1 (word for word) details.
So, I think many conflicts and confusions occur because the naïve idea about languages is that translation mostly involves level (1), with a bit of level (2) thrown in by the really “fluent”. Customers tend to be naive in this way by nature (thinking that translation = typing), and agencies feign naivety as a business strategy. Sadly, only by actually doing a lot of translation work, and really trying hard to do a good job, do you learn that 80% of the time you’ll be doing (3) and (4) – level (3) being well beyond MT at the present time, and level (4) for any imaginable future. And (3) and (4) involve high-level cognitive processing, and are consequently really tiring and time-consuming. I would say that level (4) is about a quarter of my texts in terms of words but at least a half in terms of time required.
Now, the concept of post-editing-as-business-proposal seems to be this:
“We’ll give you a (rather poor) level (1) version, and we want you to do some of your magic (2), (3) and (4) on it to make it right. Because (1) is 25% of the work (as agencies we think in word counts), we want a 25% reduction in your price.”
BUT, (1) is not 25% of the work! It may, potentially, be 25% of the text as typed, but it is only about 5% of the work, if that. Due to the increasing demands of each level, you actually have something more like this:
Level (1): Easy = 5% of the work
Level (2): Not so easy = 15% of the work
Level (3): Fairly hard = 30% of the work
Level (4): Very hard = 50% of the work
i.e. even if it were directly usable in some way, an MT text would justify only a 5% reduction in price. In reality, since the “good bits of MT” cannot automatically be slid into place at the right time, the MT parts are more or less unusable, except at the beginning of sentences, if correct (i.e. deleting the remainder and getting on with the other 95% of the translation job).
--
However, maybe this is just my view, my way of working. God knows, when I see the work of colleagues in my language pairs, I often think that they don’t go beyond level (2), or even level (1) for that matter. In which case, they should, of course, accept post-editing jobs, and do it at a 25%, or, respectively, 100% discount.
s ▲ Collapse
| | |
|
|
|
Jeff Whittaker
United States
Local time: 23:08
Member (2002)
Spanish to English
+ ...
TOPIC STARTER | Machine translation | Dec 23, 2012 |
Thank you septima for that marvelous post.
I completely agree with everything you wrote!
If only there were a way for us to offer tiered services based on the difficulty level of a text and the required deadline. I would not mind, for example, offering a small discount for work that I find "easy" and that I can translate at a pace that would make the process more enjoyable for me. However, "easy" is subjective and not easily quantifiable for customers who want translat... See more Thank you septima for that marvelous post.
I completely agree with everything you wrote!
If only there were a way for us to offer tiered services based on the difficulty level of a text and the required deadline. I would not mind, for example, offering a small discount for work that I find "easy" and that I can translate at a pace that would make the process more enjoyable for me. However, "easy" is subjective and not easily quantifiable for customers who want translations done yesterday and do not want to take the time to shop for the best price and the best quality for each individual project. They prefer a one size fits all rate resulting in overpayment for some projects and massive underpayment for others.
Thanks again for taking the time to write that post.
[Edited at 2012-12-23 19:36 GMT] ▲ Collapse
| | | | Phil Hand
China
Local time: 11:08
Chinese to English
| Thanks, Septima | Dec 24, 2012 |
That is very interesting. I don't have any idea how much of my texts fit into your categories 1, 2, 3 and 4, but I certainly recognise all of those experiences.
One interesting difference is that when I want to work really fast - when I'm up against a deadline - I go completely into mode 4. I stop thinking about the words at all. At that pace, there are some sentences which I probably could do at a lower level, but I don't take the time to make the word- or phrase-level correlations... See more That is very interesting. I don't have any idea how much of my texts fit into your categories 1, 2, 3 and 4, but I certainly recognise all of those experiences.
One interesting difference is that when I want to work really fast - when I'm up against a deadline - I go completely into mode 4. I stop thinking about the words at all. At that pace, there are some sentences which I probably could do at a lower level, but I don't take the time to make the word- or phrase-level correlations, I just write whatever English I think of to carry the meaning.
I do interpret as well, and I think your comparison with interpreting is very apt. ▲ Collapse
| | | | Vladimír Hoffman
Slovakia
Local time: 05:08
Member (2009)
English to Slovak
+ ...
| That's great, Phil! | Dec 24, 2012 |
I experienced such a state of mind only several times (also when I was up against a deadline) and it was the best possible translating as could be reached. I am not talking about quality, although it was surprisingly good, but about feelings in the course of such wok. Ambient world is "switched off", you are fully concentrated and entire sentences emerge in your head and are typed by your hands. It is much more like a meditation state than common work.
Unfortunately, I was never abl... See more I experienced such a state of mind only several times (also when I was up against a deadline) and it was the best possible translating as could be reached. I am not talking about quality, although it was surprisingly good, but about feelings in the course of such wok. Ambient world is "switched off", you are fully concentrated and entire sentences emerge in your head and are typed by your hands. It is much more like a meditation state than common work.
Unfortunately, I was never able to evoke it by my own will.
Phil Hand wrote:
One interesting difference is that when I want to work really fast - when I'm up against a deadline - I go completely into mode 4. I stop thinking about the words at all. At that pace, there are some sentences which I probably could do at a lower level, but I don't take the time to make the word- or phrase-level correlations, I just write whatever English I think of to carry the meaning.
▲ Collapse
| | | | | Such an interesting debate | Apr 19, 2014 |
I am currently trying out MT post editing for a large translation company as a freelancer. First you have to spend hours studying piles of documents about all the computer setups you have to go through, then they expect you to provide a most perfect translation (no sloppy approximation here), and because they tell you how much more productive you are going to be, they can justify impossible deadline like 7000 words in 1.5 days. On top of that, they also expect you to spend more hours generating ... See more I am currently trying out MT post editing for a large translation company as a freelancer. First you have to spend hours studying piles of documents about all the computer setups you have to go through, then they expect you to provide a most perfect translation (no sloppy approximation here), and because they tell you how much more productive you are going to be, they can justify impossible deadline like 7000 words in 1.5 days. On top of that, they also expect you to spend more hours generating QA reports. And then the real chock comes: they pay you exactly half your regular rate. In other words, they assume that you will double your productivity with MT. It will never happen.
On the other hand, I found one use of MT which is helpful, using your own MT system in house. That way you can customize the MT for a particular application or customer, couple that with a user dictionary that the MT system can connect to. I find it helps me provide a decent base for the translation as the MT product is no longer at the garbage level. I have been using Systran for a few years with helpful results. However, I still charge my regular rate, the client does not have to know what tools I use.
Otherwise, I think translation agencies are going to have to move to an hourly rate for MT post edit, this is the only fair way to compensate the translator. ▲ Collapse
| | |
|
|
|
Jeff Whittaker
United States
Local time: 23:08
Member (2002)
Spanish to English
+ ...
TOPIC STARTER
but they will want to pay the hourly rate of $15-25 an hour instead of $50-60 an hour.
Thank you for reviving this thread. I had completely forgotten about it.
Update: I just checked and it seems that the "trans-#$#@%" company in the initial post is now no longer in business. Is Schadenfreude too narcissistic?"
Or as MT would say: "Is glee also excessive?"
Jean-Claude Haas wrote:
I think translation agencies are going to have to move to an hourly rate for MT post edit, this is the only fair way to compensate the translator.
[Edited at 2014-04-19 14:01 GMT]
| | | | | The only way to avoid these kind of problems is to do a translation as perfect as possible!!! | Apr 19, 2014 |
Heike Behl, Ph.D. wrote:
It can take much more time to unscramble MT-translated text and compare soure and target to decide what is salvagable and what not than doing the translation from scratch.
On top of that, if somebody says it doesn't have to be perfect, just understandable, my main problem becomes: How do I know what exactly might be understandable for a person that doesn't have the source text for comparison, that is not familiar with certain syntactical or semantical quirks of the source language? How do I decide which error is important enough to fix and which I can ignore? How can I be sure that what I can understand of an MT-translated text is exactly the same what somebody else might understand? People differ to some extent in the reading comprehension skills for texts composed in their mother language. How can a post-editor be sure that their target audience will indeed be able to understand enough once their done?
Heike Behl, Ph.D. wrote:
The only way to avoid these kind of problems is to do a translation as perfect as possible. No more, no less.
I agree 100%!
Heike Behl, Ph.D. wrote:
If the target audience is really just interested in the gist of a text, why not ask a qualified translator to provide a short summary of the main items?
Keeping in mind that to do that, an apprpriate charge should be applied as well.
Heike Behl, Ph.D. wrote:
I think the biggest problem with MT is the idea of saving money. People become so obsessed with "saving money" that they don't even care if they end up losing money in the end. Companies offering MT and promise that a little post-editing will work wonders are basically scamming their customers.
Agree again. And they try to dupe the translator. The reasoning (by the agency/client who orders the post-editing work) that the translator's work has been made easier by supplying him/her with a text created by a machine and therefore warrants a discount (compared with translating from scratch) is a sure sign of ignorance or, more likely, lack of business skills, and often just simply a sign of greed.
It's often a trick that is now possible since MT is available and does supply "lots of words" that then could be rearranged and altered to actually be acceptable.
The problem is as long as there are people out there falling for this trickery or believe they have to accept anything that is offered to them, this kind of thing will go on for a while at least.
It's important to talk about it and let everyone know how unrealistic, unprofessional and wrong such an approach (>> accepting MT translations for "post-editing" at discounted rates) really is.
Heike Behl, Ph.D. wrote:
Only a company that specializes on particular predictable subjects, has prepared extensive MT dictionaries for these subjects and is fully aware of MT's shortcomings can employ post-editing as a useful step of the translation process. And these preparations are expensive in time and money and are a considerable investment in the future.
Anybody else better be happy with what they get without post-editing.
Dictionaries and TMs are one thing, machine translation of entire texts another.
Even in the first case, the translator will always have to be careful and spend considerable time weighing the options.
The sometimes frustrating thing (when I hear of such cowboys - agencies and "translating people" alike) is that you and I know all about these unacceptable practices but that there are so many people out there who just haven't got a clue and hurt our business sector in general.
So to all those who accept such "discount and other unprofessional demands, terms and conditions," you're in the same category as someone who knows a second language "good" and is completely and utterly naive.
Here it is, your moment of zenity:
http://tinyurl.com/n75eaep
B
[Edited at 2014-04-19 20:48 GMT]
| | | | DLyons
Ireland
Local time: 04:08
Spanish to English
+ ...
| | 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 » Agencies wanting discounts for post editing machine translation | 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 » |
| | 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
