| Pages in topic: < [1 2 3 4 5 6] > | Forget about machine translation: it still stinks, and it will stink forever Thread poster: Daniel Frisano
|
|---|
Kaspars Melkis 
United Kingdom
Local time: 20:19
English to Latvian
+ ...
| “a tool is as good as its user” | Nov 26, 2022 |
I find those arguments “a tool is as good as its user” disingenuous at best. Obviously, the tool is something that is intended to help people to do things faster and better and people using them should be appropriately trained. And yet this phrase is used to shut down any criticism you may have about any particular tool.
Don't like its UI? Suck it up and learn to use it better!
It seems to be very slow? Use it with smaller files!
It doesn't have the feature X that i... See more I find those arguments “a tool is as good as its user” disingenuous at best. Obviously, the tool is something that is intended to help people to do things faster and better and people using them should be appropriately trained. And yet this phrase is used to shut down any criticism you may have about any particular tool.
Don't like its UI? Suck it up and learn to use it better!
It seems to be very slow? Use it with smaller files!
It doesn't have the feature X that is very important in my language? Program it yourself!
Or more closely to the topic of this tread: MT translation seems off? You just need to use it with the right type of texts! Or have two editors to fix it!
Translation is a teamwork. I guess that many freelance translators working alone often forget about this point and being perfectionists think that everything depends on them. I am here not to criticize anyone for using or not using MT but about the process where translator often is just a small wheel in the whole process. Especially when a translator no longer can select the texts or file sizes they work with.
I tried to describe certain issues with MT, and it is not only me who sees them. There are reasons why some clients clearly specify that MT should not be used at any step of translation process and it is not only for confidentiality.
[Edited at 2022-11-26 06:12 GMT] ▲ Collapse
| | | | | В огороде бузина, а в Киеве дядька... (see below for translation) | Nov 26, 2022 |
Я ему про Фому, а он мне про Ерему. I guess the closest English version of these two idioms is apples and oranges, but I like my language's version more. This discussion is clearly leading you nowhere, guys. There's better ways to spend your time and words.
| | | | Stepan Konev
Russian Federation
Local time: 22:19
English to Russian
| RE: a tool is as good as its user | Nov 26, 2022 |
The difference between tool users and tool haters is that tool users always have two options to choose from: use a tool or not use a tool. On the other hand, tool haters don't have this choice which makes them suggest loads of arguments and "proofs" to justify why their inability to use a tool is rather a benefit than a shortage. When you can use a tool, you just use it. When you can't use a tool, you have to prove why you are still good even without being able to use the tool. This is how all t... See more The difference between tool users and tool haters is that tool users always have two options to choose from: use a tool or not use a tool. On the other hand, tool haters don't have this choice which makes them suggest loads of arguments and "proofs" to justify why their inability to use a tool is rather a benefit than a shortage. When you can use a tool, you just use it. When you can't use a tool, you have to prove why you are still good even without being able to use the tool. This is how all this discussion is seen form my standpoint. Take it or leave it.
[Edited at 2022-11-26 09:20 GMT] ▲ Collapse
| | | | | On the acrane ways of knowing how | Nov 26, 2022 |
People who dismiss other people's concerns with 'you just don't know how to do it' usually won't even hint at what it means to know how (maybe that silence is a means of safeguarding trade secrets). How many setups am I supposed to create in a customizable MT solution to make it perform adequately in every job for which it can technically be used such as highly standardized, repetitive tasks? This setup is for welding procedure specifications, this one for supply contracts, this one for concreti... See more People who dismiss other people's concerns with 'you just don't know how to do it' usually won't even hint at what it means to know how (maybe that silence is a means of safeguarding trade secrets). How many setups am I supposed to create in a customizable MT solution to make it perform adequately in every job for which it can technically be used such as highly standardized, repetitive tasks? This setup is for welding procedure specifications, this one for supply contracts, this one for concreting process flows, this one for material test reports, this one for engineering proposals, this one for OHSE manuals... What if some fields of those documents contain individual letters or lists of indiviual letters that need to be spelled out depending on what a particular standard says? What if I have no separate TMs for such tasks or my dedicated TMs are too small to train AI? Am I supposed to split up my huge TBs to feed them into individual setups according to the task at hand? Am I supposed to set exceptions manually because my TBs contain a lot of translations that I know will keep misleading AI? How much code am I supposed to write to create all those setups? What if I don't enjoy tweaking software settings for hours especially if I'm not sure this will reduce my human workload by a meaningful amount? Hours spent teaching AI things I can grasp within seconds and produce a streamlined translation manually using filters and regex, with near-complete control. So many questions, so few answers ▲ Collapse
| | |
|
|
|
Tom in London
United Kingdom
Local time: 20:19
Member (2008)
Italian to English
| And what's more---- | Nov 26, 2022 |
Denis Fesik wrote:
People who dismiss other people's concerns with 'you just don't know how to do it' usually won't even hint at what it means to know how (maybe that silence is a means of safeguarding trade secrets). How many setups am I supposed to create in a customizable MT solution to make it perform adequately in every job for which it can technically be used such as highly standardized, repetitive tasks? This setup is for welding procedure specifications, this one for supply contracts, this one for concreting process flows, this one for material test reports, this one for engineering proposals, this one for OHSE manuals... What if some fields of those documents contain individual letters or lists of indiviual letters that need to be spelled out depending on what a particular standard says? What if I have no separate TMs for such tasks or my dedicated TMs are too small to train AI? Am I supposed to split up my huge TBs to feed them into individual setups according to the task at hand? Am I supposed to set exceptions manually because my TBs contain a lot of translations that I know will keep misleading AI? How much code am I supposed to write to create all those setups? What if I don't enjoy tweaking software settings for hours especially if I'm not sure this will reduce my human workload by a meaningful amount? Hours spent teaching AI things I can grasp within seconds and produce a streamlined translation manually using filters and regex, with near-complete control. So many questions, so few answers
Not to mention when the source text is badly written and includes errors, jumbled-up acronyms, misspelled words imported from other languages, tradenames of products that have been "localised" by the writer, etc.
| | | | Stepan Konev
Russian Federation
Local time: 22:19
English to Russian
Denis Fesik wrote:
What if I don't enjoy... Leave it then. You simply count points to persuade yourself and others that you have made the right choice to reject tools. What if... I don't know how to turn on my PC? How many steps on the learning curve am I supposed to take to make my every job perfect? There is no end in sight to excellence. It's all up to you whether you enjoy it or not.
[Edited at 2022-11-26 11:43 GMT]
| | | | | Apples and oranges | Nov 26, 2022 |
It strikes me that many of the differences here are simply down to the types of text we translate.
In what I do, subtle nuances have to be conveyed. Sentences are complicated. The end-product needs to be well-written. MT can only provide a first draft.
But if someone wrote the user manual for a car in nice simple English, I have no doubt MT would produce a perfectly good translation into most other languages without editing.
And while we can all think of th... See more It strikes me that many of the differences here are simply down to the types of text we translate.
In what I do, subtle nuances have to be conveyed. Sentences are complicated. The end-product needs to be well-written. MT can only provide a first draft.
But if someone wrote the user manual for a car in nice simple English, I have no doubt MT would produce a perfectly good translation into most other languages without editing.
And while we can all think of things MT will struggle with, they are almost always bad writing at the end of the day.
So perhaps those feeling under threat from MT should move up the chain into writing for MT of pre-editing for MT. Logically there should be a big and lucrative market there. ▲ Collapse
| | | |
If you cared to read my other posts in this thread, you'd see I'm not rejecting tools. I use them all the time (except for MT, which I can't use anymore for a reason other than believing it's completely useless). What I do question is that the time and effort I'd have to spend training the most advanced translation AI to do the kinds of jobs I do in my language pair will eventually pay off. I'm not trying to push my views onto anyone working on different jobs in different language pairs. Why are... See more If you cared to read my other posts in this thread, you'd see I'm not rejecting tools. I use them all the time (except for MT, which I can't use anymore for a reason other than believing it's completely useless). What I do question is that the time and effort I'd have to spend training the most advanced translation AI to do the kinds of jobs I do in my language pair will eventually pay off. I'm not trying to push my views onto anyone working on different jobs in different language pairs. Why are you trying to dismiss my experience without knowing much about it? To me, excellence in translation and excellence in the use of tools are almost unrelated things, and I'd rather pursue the former. The types of standardized, repetitive documents I mentioned are what I can now translate in a breeze with no MT involved, simply relying on prior work that didn't involve MT either ▲ Collapse
| | |
|
|
|
IrinaN
United States
Local time: 14:19
English to Russian
+ ...
| On a lighter side | Nov 26, 2022 |
Let's remember that on thousands of paintings of the greatest artists of all times we see draperies, backgrounds and horses' behinds painted by the apprentices. It was a conveyor belt back then. Anyone has a problem with it? Prado alone houses over 300 Murillo's works. Can you imagine all that physical work, talent aside?!? So be a Murillo but cherish apprentices, ... See more Let's remember that on thousands of paintings of the greatest artists of all times we see draperies, backgrounds and horses' behinds painted by the apprentices. It was a conveyor belt back then. Anyone has a problem with it? Prado alone houses over 300 Murillo's works. Can you imagine all that physical work, talent aside?!? So be a Murillo but cherish apprentices, they save a ton of time and energy.
Personally, I am a pretty poor user of CATs, even used to be a hater, and do my best to avoid it still but I can't deny many benefits and advantages that my more inquisitive and modern colleagues are enjoying by using CATs with all its broad range of new features. Unfortunately or not, I was just too lazy and too busy to spend more time and effort on it. As the Russian saying goes, a machine in the hands of a caveman is but a piece of iron. I hope there is enough sense of humor to go around, and I certainly refer to myself as a CAT caveman.
[Edited at 2022-11-26 15:14 GMT] ▲ Collapse
| | | | | Paid adaptive neural MT engines are quite good | Dec 1, 2022 |
For those who can read Russian, just have a look at the couple of raw outputs from one of the paid adaptive NMT engines. It has been used to process about two dozen lengthy documents on the same topic, so it has learned the relevant terminology by now.
English
Regularly meet with Children’s Representatives in a safe space to listen to their views about what helps and hinders access to the ombuds and their suggestions for improvements.
Russian
Рег... See more For those who can read Russian, just have a look at the couple of raw outputs from one of the paid adaptive NMT engines. It has been used to process about two dozen lengthy documents on the same topic, so it has learned the relevant terminology by now.
English
Regularly meet with Children’s Representatives in a safe space to listen to their views about what helps and hinders access to the ombuds and their suggestions for improvements.
Russian
Регулярно встречайтесь с представителями детей в безопасном месте, чтобы выслушать их мнения о том, что помогает и мешает доступу к омбудсмену, и их предложения по улучшению.
English
Use role-play for children to practice ways of sharing information about the ombuds with their peers, ensuring they have clear and realistic expectations about the ombud’s role.
Russian
Используйте ролевые игры для детей, чтобы практиковать способы обмена информацией об омбудсмене со своими сверстниками, гарантируя, что они имеют четкие и реалистичные ожидания о роли омбудсмена.
As you may see, the Russian text requires minimal post-editing effort. It explains why I've been using this plugin in my Trados Studio over the last two or three years. Does it make me a poor translator? No, at least dollarwise ... ▲ Collapse
| | | | | Different people have different ideas of quality | Dec 1, 2022 |
Vladimir Pochinov wrote:
Регулярно встречайтесь с представителями детей в безопасном месте, чтобы выслушать их мнения о том, что помогает и мешает доступу к омбудсмену, и их предложения по улучшению.
Используйте ролевые игры для детей, чтобы практиковать способы обмена информацией об омбудсмене со своими сверстниками, гарантируя, что они имеют четкие и реалистичные ожидания о роли омбудсмена.
As you may see, the Russian text requires minimal post-editing effort.
Do you really consider these sentences good translations? They are semantically correct but stylistically awful. They may be OK in internal correspondence, but certainly not in publications (for which they are obviously intended). To become well-readable, they need to be completely rewritten.
| | | | | Perfect translations? No, but helpful drafts. | Dec 1, 2022 |
Anton Konashenok wrote:
Do you really consider these sentences good translations? They are semantically correct but stylistically awful. They may be OK in internal correspondence, but certainly not in publications (for which they are obviously intended). To become well-readable, they need to be completely rewritten.
Anton, I consider these to be workable MT output. Generally, this adaptive NMT engine helps reduce turnaround time by 20-40%, so it makes sense to use it on my projects.
Over the last 25 years, I've heard people rejecting CAT tools, cloud-based platforms (for storage, translation, project management, etc.), MT, you name it. Personally, I find all of them very useful, if and when used properly.
| | |
|
|
|
jyuan_us
United States
Local time: 15:19
Member (2005)
English to Chinese
+ ...
| I'm talking about the English to Chinese pair | Dec 1, 2022 |
For this language pair, MT outputs are generally very poor. On the first glance, they seem OK but if you go deeper into/look closer at them, you will find most sentences need substantial rephrasing/re-structuring, which can cost more time than translating them from scratch.
I remember I had a job of translating around 10,000 words. The MT gave a seemingly fluent output, and it looked as if I could finish the job within 10 hours. However, it actually took me 30 hours to make it look ... See more For this language pair, MT outputs are generally very poor. On the first glance, they seem OK but if you go deeper into/look closer at them, you will find most sentences need substantial rephrasing/re-structuring, which can cost more time than translating them from scratch.
I remember I had a job of translating around 10,000 words. The MT gave a seemingly fluent output, and it looked as if I could finish the job within 10 hours. However, it actually took me 30 hours to make it look like a human translation product.
[Edited at 2022-12-02 01:12 GMT] ▲ Collapse
| | | | | Jyuan, are you talking about a paid adaptive NMT engine, or some free MT engine? | Dec 1, 2022 |
I'm talking about the English to Chinese pair.
For this language pair, MT outputs are generally very poor.
| | | | Stepan Konev
Russian Federation
Local time: 22:19
English to Russian
| DeepL translation as is. Is it awful too? Can you edit it then please? | Dec 1, 2022 |
Anton Konashenok wrote:
Do you really consider these sentences good translations? They are semantically correct but stylistically awful. They may be OK in internal correspondence, but certainly not in publications (for which they are obviously intended). Вы действительно считаете эти предложения хорошим переводом? Они семантически правильны, но стилистически ужасны. Они могут быть приемлемы во внутренней переписке, но точно не в публикациях (для которых они, очевидно, предназначены).
| | | | | Pages in topic: < [1 2 3 4 5 6] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Forget about machine translation: it still stinks, and it will stink forever | 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 » |
| | 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 
