Fóruns sobre o proz.com » Poll Discussion »
Poll: Are Large Language Models ready for legal translation?
Autor da sequência: ProZ.com Staff
ProZ.com Staff
ProZ.com Staff
STAFF DO SÍTIO
Jan 7, 2025

This forum topic is for the discussion of the poll question "Are Large Language Models ready for legal translation?".

View the poll results »
 
Alex Lichanow
Alex Lichanow
Alemanha
Local time: 13:06
Inglês para Alemão
+ ...
No Jan 7, 2025

Legal translation requires expert knowledge of the target country's legal system. i.e. the source cannot simply be replaced by statistically selected target words. Of course, an LLM can provide a word-by-word "translation", which will then require MAJOR edits by a human expert who would probably not save any time vs. a human translation workflow that includes their expert knowledge from the get-go.

In other news: LLMs are not ready for ANY area of expertise. A computer algorithm tha... See more
Legal translation requires expert knowledge of the target country's legal system. i.e. the source cannot simply be replaced by statistically selected target words. Of course, an LLM can provide a word-by-word "translation", which will then require MAJOR edits by a human expert who would probably not save any time vs. a human translation workflow that includes their expert knowledge from the get-go.

In other news: LLMs are not ready for ANY area of expertise. A computer algorithm that simply replaces words cannot convey meaning.

[Edited at 2025-01-07 09:40 GMT] ▲ Collapse


Maria Laura Curzi
Maria Teresa Borges de Almeida
Daryo
Xin Yi
expressisverbis
 
Lieven Malaise
Lieven Malaise  Identity Verified
Bélgica
Local time: 13:06
Membro (2020)
Francês para Holandês
+ ...
Translation Jan 7, 2025

I notice a certain laziness in formulating poll questions. Is it that hard to be a little more specific? What is meant by "ready"? That they can translate legal texts without human intervention? That they are useful as an aid for translating legal texts?

I know little about large language models, because every single one of my customers that offers MTPE assignments uses neural machine translation instead. That probably implies that there is no good reason (yet) to prefer ChatGPT and... See more
I notice a certain laziness in formulating poll questions. Is it that hard to be a little more specific? What is meant by "ready"? That they can translate legal texts without human intervention? That they are useful as an aid for translating legal texts?

I know little about large language models, because every single one of my customers that offers MTPE assignments uses neural machine translation instead. That probably implies that there is no good reason (yet) to prefer ChatGPT and the like.

[Bijgewerkt op 2025-01-07 10:07 GMT] ▲ Collapse


Maria Teresa Borges de Almeida
Chris Spurgin
Zea_Mays
neilmac
Iulia Parvu
Edith van der Have
Ian Keith Jones Williams
 
Giovanni Guarnieri MITI, MIL
Giovanni Guarnieri MITI, MIL  Identity Verified
Reino Unido
Local time: 12:06
Inglês para Italiano
+ ...
I suspect... Jan 7, 2025

this might be Henry trying to leverage the survey tool to gather information on AI...

Ian Keith Jones Williams
Maria Teresa Borges de Almeida
 
Denis Fesik
Denis Fesik
Local time: 14:06
Inglês para Russo
+ ...
"Legal translation" makes no sense Jan 8, 2025

Unless you specify if it involves English. Not just any kind of English, but the one and only English legalese we all know and love. There are talks about simplifying legal English, but people who make money on legal writing should know better than to allow this to happen. If you do English to Russian, you'll have the problem of dealing with long chains of verbs followed by a specification of objects they relate to. My take on this problem is to find a way to phrase the whole thing so that each ... See more
Unless you specify if it involves English. Not just any kind of English, but the one and only English legalese we all know and love. There are talks about simplifying legal English, but people who make money on legal writing should know better than to allow this to happen. If you do English to Russian, you'll have the problem of dealing with long chains of verbs followed by a specification of objects they relate to. My take on this problem is to find a way to phrase the whole thing so that each verb is linked to the related object in a grammatically correct way. A gargantuan task, if you know what I mean. How do MT engines handle it? Well, they just ignore grammatical connections. They translate the verb chain and link the last verb to the first object that follows it. But wait! Many hooman translators do the same thing. And I even know that one of the old-time proz here is guilty of this sin. So, if he had bouts of hiccups over a dozen years ago (way before proz.com was first plagued by my presence), they can probably be attributed to my reactions to his legal translations uploaded in a TM on a language forum for everyone to use. Awful translations, believe me. There's a way and a way to have our job done. Most translators are okay with writing something that's grammatically disconnected. I've read one too many good books in my time to be able to do something like that ▲ Collapse
 


To report site rules violations or get help, contact a site moderator:

Moderador(es) deste fórum
Jared Tabor[Call to this topic]

You can also contact site staff by submitting a support request »

Poll: Are Large Language Models ready for legal translation?






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! »
PerfectIt
Check translations faster for consistency, quality and style guide compliance

PerfectIt helps professional translators deliver error-free documents in Microsoft Word and PowerPoint. It helps improve consistency, ensure quality, and enforce style guides. ProZ members receive a 30% discount off individual purchases of PerfectIt.

More info »