LLMs and translators, what can they do that MT can’t? Test with free Bing Chat Thread poster: Philippe Locquet
|
Hi, I’ve seen anything and everything said about LLMs these days. But in real-world applications, what do LLMs offer compared to MT? How to leverage that and in which scenario? I decided to create a video to explain that here: https://youtu.be/u4nROnmnIxI This video will demonstrate 2 situations: _You want to translate some test keeping to a specific target style (your own, that of... See more Hi, I’ve seen anything and everything said about LLMs these days. But in real-world applications, what do LLMs offer compared to MT? How to leverage that and in which scenario? I decided to create a video to explain that here: https://youtu.be/u4nROnmnIxI This video will demonstrate 2 situations: _You want to translate some test keeping to a specific target style (your own, that of the customer etc.). _You want to use a specific glossary and have terms inserted automatically where they should land. These are tasks where LLMs are good at and where if you wanted to do this with MT you would need to pay through the nose for such tasks. Off-course I’m not comparing this to human translation here. A seasoned translator can create the correct writing style and adhere to a glossary quite easily with what CAT tools do. I’m just explaining how LLMs capabilities as robots differ from regular MT. Regarding leveraging these LLM functions directly in a CAT, AFAIK the plugins available at the moment don’t do this yet. To make this work, they would need to support proompting (yes two “oo”s). I found to research and test this quite interesting! My bests to all. ▲ Collapse | | | Mr. Satan (X) English to Indonesian Impressive, but… | Jun 29, 2023 |
…isn’t this going to be detrimental to the plain NMT industry? I guess the translators who want to see DeepL dead finally would have their wish granted, although not in the way they expected.
[Edited at 2023-06-29 12:10 GMT] | | | Philippe Locquet Portugal Local time: 16:54 English to French + ... TOPIC STARTER
Mr. Satan wrote: …isn’t this going to be detrimental to the plain NMT industry? I guess the translators who want to see DeepL dead finally would have their wish granted, although not in the way they expected.
[Edited at 2023-06-29 12:10 GMT] I see what you mean. Performing specific pre-translation tasks such as I show in the video is not something that can be done in a user-friendly from a CAT plugin or dialogue at the moment AFAIK. So, it’s either for tech-savvy users or for tiny jobs, like a problematic segment for example. Then there’s the question of who do you trust to use your (and the client’s) data. Do we offer exclusive MT providers and LLM providers the same amount of trust? Last, but not least, there’s the language pair to consider. Maybe a specific MT will always be better for a certain language pair that an LLM which will endeavour to target broad audiences? That’s just a few things that came up to mind when considering this, there may be more. We’ll probably get a clearer picture of this in a year or two. I’m curious, what’s MT like for Indonesian? | | | Mr. Satan (X) English to Indonesian MT for Indonesian as a Target Language | Jul 1, 2023 |
Philippe Locquet wrote: I’m curious, what’s MT like for Indonesian? Many of my Indonesian colleagues would probably disagree with me. But I find Google Translate to be somewhat better than DeepL for certain translation sub-fields I dealt with. There was one time when I translated an academic essay on anthropology and history, and I used GT for that. Indeed, the results were mostly usable. I only needed to make minor adjustments for corrections and style consistency. Having said that, I don't typically work on the types of texts which are suitable for MT. On the contrary, MT couldn't handle colloquialisms, in my humble experience. For one, they were unable to differentiate between the formal, informal, singular, and plural forms of the pronoun “you”, or the inclusive and exclusive forms of the pronoun “we”. Output style also became exponentially problematic. Irrespective of what I fed to the MT, it tends to be formalized into a presentation akin to boilerplate texts. The outcome was simply devoid of any contextual emotions. This gets even more complicated in audiovisual translation, as we are bound by the technical constrains which MT completely heed no attention to. There's the issue of multidimensionality as well. And don't get me started on slangs, idioms, and wordplay.
[Edited at 2023-07-01 09:50 GMT] | |
|
|
Philippe Locquet Portugal Local time: 16:54 English to French + ... TOPIC STARTER
Mr. Satan wrote: Output style also became exponentially problematic. Irrespective of what I fed to the MT, it tends to be formalized into a presentation akin to boilerplate texts. The outcome was simply devoid of any contextual emotions. I did a quick search, and it seems that in MT Indonesian is still considered a Low-Resources language (not a big amount of parallel corpus available). For MT output to be good in a language pair, besides the MT technology itself you need copious amounts of data and of the highest quality if possible. When faced with something it finds difficult, MT often has the tendency to output the same line it has somewhere it it's training "translation memory" regardless of the source meaning. What you describe makes me think about this. So, I suppose most clients will stick with pure human translations for Indonesian these days. My bests | | | Mr. Satan (X) English to Indonesian 'Tis interesting. | Jul 3, 2023 |
Philippe Locquet wrote: I did a quick search, and it seems that in MT Indonesian is still considered a Low-Resources language (not a big amount of parallel corpus available). That might be so. Indonesian is supported by DeepL only since last year. It would make sense if GT has the edge in this particular case. I should mention that I don't deal with legal, medical, financial, and engineering translation sub-fields. Maybe DeepL is better suited for these tasks. When faced with something it finds difficult, MT often has the tendency to output the same line it has somewhere it it's training "translation memory" regardless of the source meaning. What you describe makes me think about this. Ah, so it works like fuzzy matches in a CAT program, I presume? (I know nothing!) So, I suppose most clients will stick with pure human translations for Indonesian these days. Sadly, that doesn't stop some annoying agencies from dumping the source documents into an untrained MT engine and expect the translators to clean up the mess, while demanding nothing more than a light MTPE rate. Source: personal experience.
[Edited at 2023-07-03 14:36 GMT] | | | Philippe Locquet Portugal Local time: 16:54 English to French + ... TOPIC STARTER Back to Momma | Jul 5, 2023 |
Mr. Satan wrote: Ah, so it works like fuzzy matches in a CAT program, I presume? Well, to train an MT you need parallel data (not exclusively, monolingual data can be used in conjunction with bilingual). So, when an MT (and an LLM, they behave the same) struggles to translate something, they go back to Momma and spit out word-for-word (or slightly modified) content they have in their databank (memory?). | | | There is no moderator assigned specifically to this forum. To report site rules violations or get help, please contact site staff » LLMs and translators, what can they do that MT can’t? Test with free Bing Chat 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... | | | | | |