| 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
|
|---|
Daniel Frisano 
Italy
Local time: 03:17
Member (2008)
English to Italian
+ ...
Anyone needing one more proof of how MT stinks?
Recently I was assigned a proofreading job, and when I loaded the original package with pre-translated segments to be edited by the translator, here is what I saw:
Four different translations for the same term, within the same document, same context, in four consecutive segments.
And you call this dumb thing an artificial *intelligence*? You'd think that MT at least could be trusted with terminological consistency. After all, machines exist exactly to perform repetitive tasks in a consistent manner.
Well, there you go. You're been led to believe that MT is a useful tool for your work, and the reality is that it makes mistakes no 12-year old would make. You're constantly drawn towards mediocrity while struggling to escape towards excellence.
It still sucks, and if it hasn't been fixed it by now, you can be sure that it will suck forever.
By the way, by using MT, you’re constantly feeding it back its own faults, in a sort of diabolical error self-propagation.
Don't use MT. Unless you're OK with mediocrity, of course.
| | | | | Preferred target terms | Nov 19, 2022 |
You probably don’t want to hear this but in many engines and tools you can set preferred target terms.
Let us assume that this was done. How is the rest of the translation?
| | | | | Post removed: This post was hidden by a moderator or staff member for the following reason: Empty post | Stepan Konev
Russian Federation
Local time: 05:17
English to Russian
| Try 115 to 110 remove 156 confusing 230 numbers 180 | Nov 20, 2022 |
It took several seconds for me to understand what all those numbers mean... Probably they are some part numbers or drawing positions, but they interrupt the normal text flow. If you remove all numbers or put them into parentheses, you get this (DeepL):
1. Il sistema combinato di produzione additiva e lavorazione (110) può includere una o più navette portautensili (230).
2. Le navette dell'utensile (230) possono consentire all'utensile di produzione additiva (120) e al... See more It took several seconds for me to understand what all those numbers mean... Probably they are some part numbers or drawing positions, but they interrupt the normal text flow. If you remove all numbers or put them into parentheses, you get this (DeepL):
1. Il sistema combinato di produzione additiva e lavorazione (110) può includere una o più navette portautensili (230).
2. Le navette dell'utensile (230) possono consentire all'utensile di produzione additiva (120) e all'utensile di lavorazione (180) di muoversi in qualsiasi direzione desiderata.
3. L'utensile di produzione additiva (120) e l'utensile di lavorazione (180) possono condividere una navetta (230) o ogni utensile può avere una propria navetta (230).
4. L'utensile per la produzione additiva (120) e l'utensile per la lavorazione (180) possono essere scambiati tramite le navette dell'utensile (230).
Well, ok, you made it to a good example of bad application of MT. In a text artificially disrupted with meaningless numbers. However, it is not natural for most texts to go like this.
[Edited at 2022-11-20 01:08 GMT] ▲ Collapse
| | |
|
|
|
Tom in London
United Kingdom
Local time: 02:17
Member (2008)
Italian to English
| Excellent example | Nov 20, 2022 |
Daniel Frisano wrote:
Anyone needing one more proof of how MT stinks?
Recently I was assigned a proofreading job, and when I loaded the original package with pre-translated segments to be edited by the translator, here is what I saw:
Four different translations for the same term, within the same document, same context, in four consecutive segments.
And you call this dumb thing an artificial *intelligence*? You'd think that MT at least could be trusted with terminological consistency. After all, machines exist exactly to perform repetitive tasks in a consistent manner.
Well, there you go. You're been led to believe that MT is a useful tool for your work, and the reality is that it makes mistakes no 12-year old would make. You're constantly drawn towards mediocrity while struggling to escape towards excellence.
It still sucks, and if it hasn't been fixed it by now, you can be sure that it will suck forever.
By the way, by using MT, you’re constantly feeding it back its own faults, in a sort of diabolical error self-propagation.
Don't use MT. Unless you're OK with mediocrity, of course.
Thanks Daniel - that's an excellent example.
| | | | | There are two types of people | Nov 20, 2022 |
It's always good to realize there are two types of people: it's potential meme material. I had an editing project that had been totally botched, the customer didn't even want to pay for it because they couldn't understand anything in the translation. And there was a guy in my company who did nothing but train translation AI. I gave him a list of terms with correct translations I wanted to see everywhere in the project; he spent a few weeks on it (the project was really big) and still screwed the... See more It's always good to realize there are two types of people: it's potential meme material. I had an editing project that had been totally botched, the customer didn't even want to pay for it because they couldn't understand anything in the translation. And there was a guy in my company who did nothing but train translation AI. I gave him a list of terms with correct translations I wanted to see everywhere in the project; he spent a few weeks on it (the project was really big) and still screwed the job. And that's what happens every time they use AI so trained: the job gets screwed. MT adepts had better use their best effort to make MT output good enough to be used without editing (they believe it's possible, don't they?); I won't bother helping them to make that output any better
[Edited at 2022-11-20 08:39 GMT] ▲ Collapse
| | | |
This year I worked on an MT project for a set of technical documents from Italian into Turkish. Some machines components had been listed as "X legge 1", "Y legge 2", etc.. It didn't take me long to figure out that the Italian text was in fact the MT output from French to Italian.
| | | | Michael Davies
Denmark
Local time: 03:17
Member (2009)
Danish to English
+ ...
This item is lacking some important information. What MT tool was used in the case example? There is, in my experience, a world of difference in the quality of the output from previous generations of MT (often free of charge) and the newest generation of NMT (usually subject to subscription charges).
Without this additional information, the item does not present a true picture of machine translation, which in my view (and experience) can be a useful tool to improving translation ef... See more This item is lacking some important information. What MT tool was used in the case example? There is, in my experience, a world of difference in the quality of the output from previous generations of MT (often free of charge) and the newest generation of NMT (usually subject to subscription charges).
Without this additional information, the item does not present a true picture of machine translation, which in my view (and experience) can be a useful tool to improving translation efficiency - and final quality if followed by suitable post-editing (PEMT).
However, using MT (or NMT) alone is a sure way of producing poor quality and unsatisfactory translations - and getting MT a bad name! ▲ Collapse
| | |
|
|
|
Tom in London
United Kingdom
Local time: 02:17
Member (2008)
Italian to English
| Waiting for the PEMT software | Nov 20, 2022 |
Michael J.H. Davies wrote:
.... machine translation ..... can be a useful tool to improving translation efficiency - and final quality if followed by suitable post-editing (PEMT).
So Michael - when are we going to have PEMT software?
| | | | David GAY
Local time: 03:17
English to French
+ ...
Yes, but how long does it take you to fix those mistakes and how much can you charge for it? Can you make a living fixing those mistakes?
| | | |
And even if there had been no mistakes, it still conditions the translator as it produces a word for word translation and the result is that the translation is stilted, not natural and it's then very difficult to fix. But agencies only care about having to pay the translator much less. Agencies that want and appreciate a good, publishable translation do not send MTPE jobs. I recently had one such task and it was an instruction manual with lots of warnings so there were a lot of repetitions which... See more And even if there had been no mistakes, it still conditions the translator as it produces a word for word translation and the result is that the translation is stilted, not natural and it's then very difficult to fix. But agencies only care about having to pay the translator much less. Agencies that want and appreciate a good, publishable translation do not send MTPE jobs. I recently had one such task and it was an instruction manual with lots of warnings so there were a lot of repetitions which is suitable for MTPE, not like when the task involves marketing, but still, the end result was stilted, unnatural and I had to turn round the parts for them to make sense and sound natural. So, there you have it. ▲ Collapse
| | | | David GAY
Local time: 03:17
English to French
+ ...
| patent translation | Nov 20, 2022 |
Josephine Cassar wrote:
it produces a word for word translation
I think it s not a problem as far as patent translation is concerned.
| | |
|
|
|
finnword1
United States
Local time: 21:17
English to Finnish
+ ...
| legal briefs | Nov 20, 2022 |
Google still thinks that is means "judicial underpants" in Finnish
| | | | Stepan Konev
Russian Federation
Local time: 05:17
English to Russian
finnword1 wrote:
Google still thinks that is means "judicial underpants" in Finnish Which one is "judicial underpants"?
Translators do not translate single words. As a seasoned translator you didn't try to translate "legal briefs" alone without any context, did you?
[Edited at 2022-11-20 21:39 GMT]
| | | | finnword1
United States
Local time: 21:17
English to Finnish
+ ...
| What if there is no context? | Nov 21, 2022 |
What if "legal briefs" appears in a list of miscellaneous items?
Besides, in your example, legal briefs has been incorrectly translated as "legal documents". Not the same thing.
[Edited at 2022-11-22 06:02 GMT]
[Edited at 2022-11-22 06:04 GMT]
| | | | | 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 » |
| | CafeTran Espresso | You've never met a CAT tool this clever!
Translate faster & easier, using a sophisticated CAT tool built by a translator / developer.
Accept jobs from clients who use Trados, MemoQ, Wordfast & major CAT tools.
Download and start using CafeTran Espresso -- for free
Buy now! » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
