| Pages in topic: < [1 2] | How does machine translation get better? Thread poster: Baran Keki
|
|---|
Stepan Konev 
Russian Federation
Local time: 20:35
English to Russian
Ice Scream wrote:
I would want to know FOR CERTAIN whether my translations were being used to train the MT engine Even though I have already answered this question above, let’s try again: no, your translations are not used for MT training and not even sent anywhere unless you intentionally do it.
Ice Scream wrote:
isn’t MT going to eat itself Until these days, MT is getting better and better. Even those who don’t use MT admit that from time to time. Statements like “I personally never use MT but I tried it recently just for a couple of words and it was surprisingly good” become common on forums. I don’t like it either, but MTPE jobs become common too which actually proves that there is no soon end to MT in sight.
| | | | | This is vast... | May 24, 2022 |
Novian Cahyadi wrote:
Hi, Philippe.
Do you mind sharing some papers explaining about this in more details? That sounds like an interesting read for this weekend.
Cheers, bud.
This topic is massive and it's constantly evolving. There is a vast number of papers available online, just google paper + machine translation + a brand name to find some. Now be prepared to some serious lingo and data science / AI references...
To get started, I'd recommend looking at Wikipedia etc. for some history. Then for current trends, you can check https://slator.com/machine-translation/ (bearing in mind that Slator is a pay-to-display-your-article sort of publication...).
For those going or attending on-line, at the Wordfast Forward conference Jamie Lucero will show how you can train your own MT engine, that's an interesting application of knowledge. (WFF link for streamed/recorded program: https://wff22.eventbrite.com/?aff=pl)
I'm thinking of making a video on Machine Translation (I have a few already) as an update to what's happening. For those interested, feel free to add questions/topics in comments below my videos (that'll be easier to find than on ProZ forums where such topic is likely to be nuked JJJJ) here's one https://youtu.be/p1RIVC3II3w
That's what I can think of, off the top of my head. Have fun!
| | | | Lieven Malaise
Belgium
Local time: 19:35
Member (2020)
French to Dutch
+ ...
| No stopping. | May 24, 2022 |
Stepan Konev wrote:
Until these days, MT is getting better and better. Even those who don’t use MT admit that from time to time. Statements like “I personally never use MT but I tried it recently just for a couple of words and it was surprisingly good” become common on forums. I don’t like it either, but MTPE jobs become common too which actually proves that there is no soon end to MT in sight.
My sister in law works in a large bank where they use Deepl internally, my accountant works in an accounting firm where they also use Deepl ...
Even if you are a sceptic it's hard to deny that MT already has its wide-spread use and that it will probably only increase in the future.
| | | | | Misunderstanding | May 24, 2022 |
Lieven Malaise wrote:
Even if you are a sceptic it's hard to deny that MT already has its wide-spread use and that it will probably only increase in the future.
I think Stepan misunderstood me. MT is undeniably getting better and its use will inevitably spread, but unless someone is feeding corrected translations back into the engine, surely its quality will stagnate or deteriorate?
As I understand it, MT learns by analysing stuff translated or written by humans, much of which is flawed even before the MT applies its lack of intelligence.
In future there will be less human output to use, so presumably MT will increasingly be feeding on its own even more flawed output. Hence my comment about eating itself. Chasing its own tail. Vicious circle. Etc etc.
How can it learn from its mistakes if nobody corrects it?
| | |
|
|
|
Mr. Satan (X)
English to Indonesian
| @Philippe Locquet | May 24, 2022 |
True, but not all research papers created equally. Or should I say, competently. You seem to be very knowledgeable about this topic, so I was hoping to get a hint about some good sources where the authors actually know what they are talking about. Gotta separate real science from propaganda and shills as well.
Anyway, thanks for giving a starting point and sharing your videos discussing this. I highly appreciate it. 👍
[Edited at 2022-05-24 13:45 GMT]
| | | |
Novian Cahyadi wrote:
True, but not all research papers created equally. Or should I say, competently. You seem to be very knowledgeable about this topic, so I was hoping to get a hint about some good sources where the authors actually know what they are talking about. Gotta separate real science from propaganda and shills as well.
Anyway, thanks for giving a starting point and sharing your videos discussing this. I highly appreciate it. 👍
[Edited at 2022-05-24 13:45 GMT]
You're right. I'll do this quick since I'm swamped today.
This is my yearly go-to: https://try.inten.to/machine-translation-report-2021/
Good authors of papers: Kyunghyun Cho, Université de Montréal (Yoshua Bengio...), Transperfect AI, Bruno Kessler Institute (FONDAZIONE BRUNO KESSLER). Remember papers will look at something very specific and evaluate outcomes.
| | | | Lieven Malaise
Belgium
Local time: 19:35
Member (2020)
French to Dutch
+ ...
Ice Scream wrote:
In future there will be less human output to use, so presumably MT will increasingly be feeding on its own even more flawed output. Hence my comment about eating itself. Chasing its own tail. Vicious circle. Etc etc.
How can it learn from its mistakes if nobody corrects it?
I'm not entirely sure, but I think MT uses everything that is published and freely accessible on the internet. So it can't use the direct input from translators, but it can use the zillions of documents that are first translated with help of MT and afterwards published on the internet in one way or another. If I'm right MT is eventually using corrected versions of it's previous content.
I have to admit, though, that my reasoning here is a bit of a long shot.
| | | | | Meaning of "correct" | May 25, 2022 |
Ice Scream wrote:
How can it learn from its mistakes if nobody corrects it?
I'm not an expert on this, but I don't think the mistakes made by MT need to be explicitly corrected in order for it to improve. There just has to be some kind of feedback loop via which it is informed whether its output is successful or not, in the broadest possible terms.
The AI that plays Go well enough to beat world champions doesn't "understand" the game of Go. It's just played millions (or billions) of games against itself and noted which ones it won. That was enough for it to develop a strategy that lets it win at the highest levels. Even the programmers don't know why the software picks a particular play in a given situation.
Think of it as a process of evolution. Evolution doesn't care about the individual creatures that lived out their lives and died over thousands and millions of years. It only cares which ones produced the most offspring in a given environment. That rough statistic was enough to produce the incredible range of living things with which we share our planet (and which human beings are now busily destroying).
Nobody told any of those critters whether they were adopting the "correct" lifestyle or not, yet somehow, over the aeons, they got "better" at surviving.
| | |
|
|
|
Mr. Satan (X)
English to Indonesian
Philippe Locquet wrote: This is my yearly go-to: https://try.inten.to/machine-translation-report-2021/ Good authors of papers: Kyunghyun Cho, Université de Montréal (Yoshua Bengio...), Transperfect AI, Bruno Kessler Institute (FONDAZIONE BRUNO KESSLER). Remember papers will look at something very specific and evaluate outcomes.
Thank you! Thank you, thank you, thank you!
I'm trying to wrap my head around AI in general, and MT in particular. Indeed, AI is a very complex topic, but also a very popular one. Everyone is talking about it and I feel kinda overwhelmed by the varying degrees of POVs people have. I wanted to cut down some static noises and focus on what matters. Because AI is one of the next big things in history, right? I know it's stupid to predict anything about tech, but I always prefer to be prepared and devise a plan accordingly. Especially since this may affect my livelihood.
| | | | expressisverbis
Portugal
Local time: 18:35
Member (2015)
English to Portuguese
+ ...
| Great source! | May 25, 2022 |
Thanks for sharing it, Philippe!
Muito obrigada!
| | | | | 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 » How does machine translation get better? | Wordfast Pro | Translation Memory Software for Any Platform
Exclusive discount for ProZ.com users!
Save over 13% when purchasing Wordfast Pro through ProZ.com. Wordfast is the world's #1 provider of platform-independent Translation Memory software. Consistently ranked the most user-friendly and highest value
Buy now! » |
| | 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 » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
