| Pages in topic: < [1 2 3 4] > | Poll: Would you be willing to work on a project that contributes to development of AI translation? Thread poster: ProZ.com Staff
|
|---|
Dan Lucas 
United Kingdom
Local time: 13:28
Member (2014)
Japanese to English
| I suspect we passed that point... | Mar 4 |
Michele Fauble wrote:
In five years most translations will be done by AI.
...several years ago.
Dan
| | | | | Post removed: This post was hidden by a moderator or staff member for the following reason: Referencing a post which has been removed as per site rules. | Michele Fauble
United States
Local time: 05:28
Member (2006)
Norwegian to English
+ ...
| AI could pass human tests in five years | Mar 4 |
Christopher Schröder wrote: Michele Fauble wrote:
In five years most translations will be done by AI.
None of mine will.
Nvidia (NVDA.O), Chief Executive Jensen Huang on Friday said that artificial general intelligence could - by some definitions - arrive in as little as five years.
Huang, who heads the world's leading maker of artificial intelligence chips used to create systems like OpenAI's ChatGPT, was responding to a question at an economic forum held at Stanford University about how long it would take to achieve one of Silicon Valley's long-held goals of creating computers that can think like humans.
Huang said that the answer largely depends on how the goal is defined. If the definition is the ability to pass human tests, Huang said, artificial general intelligence (AGI) will arrive soon.
"If I gave an AI ... every single test that you can possibly imagine, you make that list of tests and put it in front of the computer science industry, and I'm guessing in five years time, we'll do well on every single one," said Huang, whose firm hit $2 trillion in market value on Friday.“
https://www.reuters.com/technology/nvidia-ceo-says-ai-could-pass-human-tests-five-years-2024-03-01/
[Edited at 2024-03-04 20:04 GMT]
| | | | Lieven Malaise
Belgium
Local time: 14:28
Member (2020)
French to Dutch
+ ...
Michele Fauble wrote:
Huang said that the answer largely depends on how the goal is defined.
That says it all, doesn't it? If and when...
| | |
|
|
|
Alex Lichanow
Germany
Local time: 14:28
Member (2020)
English to German
+ ...
Michele Fauble wrote:
Nvidia (NVDA.O), Chief Executive Jensen Huang on Friday said that artificial general intelligence could - by some definitions - arrive in as little as five years.
[Edited at 2024-03-04 20:04 GMT]
OK, guess I'll stick to AMD GPUs for now.
| | | | | MT might be AI Translation | Mar 5 |
I think MTPE might already be assisting with the development of AI translation...
| | | |
ProZ.com Staff wrote:
This forum topic is for the discussion of the poll question "Would you be willing to work on a project that contributes to development of AI translation?"
I'm an art translator, so translate books, catalogues, articles, museum texts, et cetera. Italian to English. When a respected colleague or two mentioned using DeepL and ChatGTP, I was curious and tried out both. For some things they were shockingly good. For most things, shockingly bad. But the worst thing was that I was not using my brain to translate the text. And the other worst thing was that I was now having to deal not only with source language interference, but also with machine translation and AI translation interference.
When I read a beautifully translated text, I am humbled and awed, but it also just feels amazing. It feels good to read a good translation, no matter what the subject. And one thing that comes to mind immediately is that many of these translations were done before the internet. And almost all, if not all, before AI, et cetera. In a certain way, it's a reminder of what the human mind is capable of. It's exciting and inspiring. Look at what we are capable of! On the other hand, when I read a poorly translated text, it feels depressing. A bit like a lot of what passes for writing in this new everything-goes, everyone's a writer world. Empty, mechanical, homogeneous, superficial. And that's the side DeepL and ChatGTP are on. At least for things in my areas of interest.
When I was experimenting, I would put in a sentence I was having trouble with, to see what DeepL or ChatGTP would do with it. Sometimes, I was impressed with the result. But the next time I read a beautifully translated text, inevitably far and away better than anything DeepL or ChatGTP could have come up with, I would think: this person did this with their brain. And I would be inspired to push my own farther. And the next challenging sentence I came across would be easier to handle.
| | | | Tom in London
United Kingdom
Local time: 13:28
Member (2008)
Italian to English
Abdulsamad Attayee wrote:
I think MTPE might already be assisting with the development of AI translation...
Yes, we are probably all collaborating without actually knowing about it. Just using Deepl probably implicates us.
But there's a difference between collaborating because it's not possible not to, and collaborating intentionally and enthusiastically in one's own demise.
| | |
|
|
|
Tom in London wrote:
Yes, we are probably all collaborating without actually knowing about it. Just using Deepl probably implicates us.
But there's a difference between collaborating because it's not possible not to, and collaborating intentionally and enthusiastically in one's own demise.
I would argue that using DeepL would be collaborating intentionally and enthusiastically in our own demise.
| | | | | I thought you knew all that | Mar 5 |
I think MTPE might already be assisting with the development of AI translation...
This is well known, since the beginning. Every entry in agency-owned MT, or Google Trans, or DeepL etc, is entered into the co-called "corpus" (the large database). Agency-owned MTs are the most accurate, because they follow glossaries and discipline-specific databases.
And every time you change the suggested sentence with something better, your new sentence is entered, and it will pop-up at some point, depending on the majority of choices that are accepted (re-entered) by translators.
You must have seen that if it doesn't have a specialized term, but you enter it a couple of times, the term then appears in all your subsequent sentences. It "learns" fast, because you're teaching it with every entry. If it doesn't have enough entries in a particular topic, it fails miserably.
Translation technology was deemed almost impossible (in terms of accuracy), until IBM decided to just feed a gigantic database with lots of data (already translated sentences, which internally are also split into segments).
Twenty five years ago we were doing similar things with Trados TMs, "Find-Replace", and clean 2-3 versions in the TM to have alternatives. We were also manually aligning approved/officially published texts and then reversing the language pairs and create different TM sets in both combos, etc. (what serious translator had not automated on his computer the H-phrases, P-phrases, create his own TMs from the gigantic freely circulated back then TMX files with the translated EU laws/regulations, etc).
AI will not improve the MT result, but it will provide a friendly interface so that end-clients can use MT without special tools or APIs. In many cases they won't need agencies at all. Yeah, they'll still need proofreaders.
| | | | | Doesn’t work like that IME | Mar 5 |
Lefteris Kritikakis wrote:
And every time you change the suggested sentence with something better, your new sentence is entered, and it will pop-up at some point, depending on the majority of choices that are accepted (re-entered) by translators.
You must have seen that if it doesn't have a specialized term, but you enter it a couple of times, the term then appears in all your subsequent sentences. It "learns" fast, because you're teaching it with every entry.
I thought it would work like that but both times I’ve mistakenly accepted jobs that have been run through MT, matches have been taken from the TM and the MT has then translated new text without any reference to what was in the TM. There was no learning at all. Which was a good thing really because the TM was even worse than the MT. This was for one of the biggest agencies.
| | | |
Michele Fauble wrote:
Nvidia (NVDA.O), Chief Executive Jensen Huang on Friday said…
ChatGPT runs on Nvidia's hardware. Of course, he is going to say that. The more people adopting ChatGPT, the more profit Nvidia makes.
Huang also said that programming is dead as a profession, by the way. Anyone with the slightest understanding of software development would concur this is utter nonsense.
| | |
|
|
|
Liena Vijupe
Latvia
Local time: 15:28
Member (2014)
French to Latvian
+ ...
Lefteris Kritikakis wrote:
You must have seen that if it doesn't have a specialized term, but you enter it a couple of times, the term then appears in all your subsequent sentences. It "learns" fast, because you're teaching it with every entry.
I have seen the results changing sometimes, but it's not very different from a new TM match simply replacing a previous one, it is not learning anything. I am working with MT for EU documents and, specifically, a certain type of documents where one specific term is repeated in every other paragraph and I am already sick of having to correct it every time and at every new assignment. Even Find+replace would work better, at least then I would only have to adjust the endings which are different for each grammar case in my language, not the entire term.
I was quite impressed and slightly terrified by the MT quality when my EU client switched to it a few years ago (I even worried it could make me too lazy for "old-fashioned" translation and I might lose that skill completely!) but I have not seen any improvements since then and having to clean up the same mess every time makes me despise it. I believe it has reached the peak already and will only go down from now with the input quality worsening.
| | | |
Aren't they written in a language that can be machine-generated and machine-translated seamlessly? Everyone who's had a taste of EU language should have a similar feeling about it, or am I wrong? Maybe that's one of the reasons why the set of translators who have EU documents as their primary workload and the set of translators who have been noticing a striking improvement in the quality of machine translations may be said to overlap. My experience with EU texts has been limited to ECHR document... See more Aren't they written in a language that can be machine-generated and machine-translated seamlessly? Everyone who's had a taste of EU language should have a similar feeling about it, or am I wrong? Maybe that's one of the reasons why the set of translators who have EU documents as their primary workload and the set of translators who have been noticing a striking improvement in the quality of machine translations may be said to overlap. My experience with EU texts has been limited to ECHR documents, and, as I think I've mentioned earlier, they weren't all that boring to translate because most of them dealt with life and crime stories ▲ Collapse
| | | | Liena Vijupe
Latvia
Local time: 15:28
Member (2014)
French to Latvian
+ ...
Denis Fesik wrote:
Aren't they written in a language that can be machine-generated and machine-translated seamlessly?
In theory, yes. The same is often said about technical manuals which are seen as ideal candidates for MT, but I hear a lot of complaints about those too. Also in EU documents I see wrong translations, translations which are not necessarily wrong but do not correspond to the officially approved terminology, nonsense translations and translations which are completely opposite to the intended meaning with particularly disastrous results in the context of human rights (I've lost count on how many times I've had to make corrections to clarify that somebody was not supporting mass killings or contributing to child abuse etc.) due to misread syntax, unidiomatic style or poorly written source.
I think it shows the limits of MT, it can never be perfect because it does not understand the message. The whole idea of "learning" implies it is being fed better and better translations of the same thing and is also able to select the best and most appropriate among millions of matches each time which is neither true, nor realistic.
| | | | | Pages in topic: < [1 2 3 4] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Poll: Would you be willing to work on a project that contributes to development of AI translation? | 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 Studio 2022 Freelance | The leading translation software used by over 270,000 translators.
Designed with your feedback in mind, Trados Studio 2022 delivers an unrivalled, powerful desktop
and cloud solution, empowering you to work in the most efficient and cost-effective way.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
