Amazing language translation with Google Gemini Advanced and ChatGPT

Looking for translations, or for help with translations and transliterations? This is the place.
Post Reply
abcdef
Posts: 7
Joined: Tue Jun 22, 2021 9:22 am

Amazing language translation with Google Gemini Advanced and ChatGPT

Post by abcdef »

Google released their AI product Gemini Advanced today which is a competitor to ChatGPT. I did a few tests and it seems really good at translating Tibetan texts to English. Also it can summarize and analyze sutras at 84000.co really well, just by pasting the URL to the sutra.

I tried to translate some shorter Tibetan comments and simple prayers and it did it really well.

Here is an example from pasting the Tibetan text from "The Array of Virtues of Mañjuśrī’s Buddha Realm" at https://read.84000.co/translation/toh59 ... 41-004-123:

Image
Image

And for Sanskrit the dharani here in ChatGPT and Gemini Advanced: https://read.84000.co/translation/toh46 ... 039-002-80
Image
Image

It's not perfect and there are still many mistakes, but it definitely seems like AI already can be of help both for translators and practitioners.
User avatar
Queequeg
Former staff member
Posts: 14493
Joined: Tue Jul 03, 2012 3:24 pm

Re: Amazing language translation with Google Gemini Advanced and ChatGPT

Post by Queequeg »

abcdef wrote: Thu Feb 08, 2024 7:32 pm
Very cool.

I'm interested to see what kind of further information these machine learning programs will be able to uncover. For instance, determine authorship, or perhaps determine influences and "geneologies" of texts. For instance, there are often texts suspected of being apocryphal but attributed to well known authors. The texts could be analyzed side by side against known texts to determine if the same person wrote them based on terminology, structure, content, etc. Or perhaps look at a group of texts - Prajnaparamita texts for instance - to see if they might have been written by the same person.

Years ago I read a study that purported to analyze Pali texts to sift out the oldest layers of the texts and determine if they were spoken by a single person. I imagine these machines will be able to do such analysis with much more depth and precision.

The information that will be available to researchers is going to be amazing.
There is no suffering to be severed. Ignorance and klesas are indivisible from bodhi. There is no cause of suffering to be abandoned. Since extremes and the false are the Middle and genuine, there is no path to be practiced. Samsara is nirvana. No severance achieved. No suffering nor its cause. No path, no end. There is no transcendent realm; there is only the one true aspect. There is nothing separate from the true aspect.
-Guanding, Perfect and Sudden Contemplation,
User avatar
Kim O'Hara
Former staff member
Posts: 7091
Joined: Fri Nov 16, 2012 1:09 am
Location: North Queensland, Australia

Re: Amazing language translation with Google Gemini Advanced and ChatGPT

Post by Kim O'Hara »

Queequeg wrote: Fri Feb 09, 2024 4:24 am
abcdef wrote: Thu Feb 08, 2024 7:32 pm
Very cool.

I'm interested to see what kind of further information these machine learning programs will be able to uncover. For instance, determine authorship, or perhaps determine influences and "geneologies" of texts. For instance, there are often texts suspected of being apocryphal but attributed to well known authors. The texts could be analyzed side by side against known texts to determine if the same person wrote them based on terminology, structure, content, etc. Or perhaps look at a group of texts - Prajnaparamita texts for instance - to see if they might have been written by the same person.

Years ago I read a study that purported to analyze Pali texts to sift out the oldest layers of the texts and determine if they were spoken by a single person. I imagine these machines will be able to do such analysis with much more depth and precision.

The information that will be available to researchers is going to be amazing.
Yes and no. The more recent of the existing studies already relied on computer analysis, so improvements are likely to incremental rather than revolutionary.
And AI makes stuff up and doesn't admit to doing so, which I reckon is a fatal flaw in anything purporting to be serious research.
A quick search will get you lots of results about it - here's one from from local news organisation. https://www.abc.net.au/news/2023-08-02/ ... /102678968

:coffee:
Kim
User avatar
Queequeg
Former staff member
Posts: 14493
Joined: Tue Jul 03, 2012 3:24 pm

Re: Amazing language translation with Google Gemini Advanced and ChatGPT

Post by Queequeg »

Kim O'Hara wrote: Fri Feb 09, 2024 9:19 am
Queequeg wrote: Fri Feb 09, 2024 4:24 am
abcdef wrote: Thu Feb 08, 2024 7:32 pm
Very cool.

I'm interested to see what kind of further information these machine learning programs will be able to uncover. For instance, determine authorship, or perhaps determine influences and "geneologies" of texts. For instance, there are often texts suspected of being apocryphal but attributed to well known authors. The texts could be analyzed side by side against known texts to determine if the same person wrote them based on terminology, structure, content, etc. Or perhaps look at a group of texts - Prajnaparamita texts for instance - to see if they might have been written by the same person.

Years ago I read a study that purported to analyze Pali texts to sift out the oldest layers of the texts and determine if they were spoken by a single person. I imagine these machines will be able to do such analysis with much more depth and precision.

The information that will be available to researchers is going to be amazing.
Yes and no. The more recent of the existing studies already relied on computer analysis, so improvements are likely to incremental rather than revolutionary.
And AI makes stuff up and doesn't admit to doing so, which I reckon is a fatal flaw in anything purporting to be serious research.
A quick search will get you lots of results about it - here's one from from local news organisation. https://www.abc.net.au/news/2023-08-02/ ... /102678968

:coffee:
Kim
Thanks. I have not been keeping up, but I imagine machine learning will produce increasingly refined results with larger data sets. I suspect the advantages going forward is that currently, those computer analyses are human programmed to look for specific things the researcher identifies. The ideal with machine learning is that they'll be able to ask questions independently and find patterns that people never thought to consider.

Of course the work needs to be "peer" reviewed, whatever that may mean, but if machines are just making things up, that's no different than their human minders. Scholars and researchers are getting busted for making stuff up all the time, sometimes upending whole fields. Those scholars don't admit they make the stuff up either. I don't think that's a unique problem. If we're designing these machines, its likely we're just amplifying ourselves. Machines will need Dharma to correct their delusions, too.
There is no suffering to be severed. Ignorance and klesas are indivisible from bodhi. There is no cause of suffering to be abandoned. Since extremes and the false are the Middle and genuine, there is no path to be practiced. Samsara is nirvana. No severance achieved. No suffering nor its cause. No path, no end. There is no transcendent realm; there is only the one true aspect. There is nothing separate from the true aspect.
-Guanding, Perfect and Sudden Contemplation,
tingdzin
Posts: 1968
Joined: Fri Feb 15, 2013 7:19 am

Re: Amazing language translation with Google Gemini Advanced and ChatGPT

Post by tingdzin »

Why should anyone bother to study Tibetan at all now? :stirthepot:
User avatar
Kim O'Hara
Former staff member
Posts: 7091
Joined: Fri Nov 16, 2012 1:09 am
Location: North Queensland, Australia

Re: Amazing language translation with Google Gemini Advanced and ChatGPT

Post by Kim O'Hara »

Queequeg wrote: Fri Feb 09, 2024 1:23 pm
Kim O'Hara wrote: Fri Feb 09, 2024 9:19 am
Queequeg wrote: Fri Feb 09, 2024 4:24 am
Very cool. ...
Yes and no. The more recent of the existing studies already relied on computer analysis, so improvements are likely to incremental rather than revolutionary.
And AI makes stuff up and doesn't admit to doing so, which I reckon is a fatal flaw in anything purporting to be serious research.
A quick search will get you lots of results about it - here's one from from local news organisation. https://www.abc.net.au/news/2023-08-02/ ... /102678968

:coffee:
Kim
Thanks. I have not been keeping up, but I imagine machine learning will produce increasingly refined results with larger data sets. I suspect the advantages going forward is that currently, those computer analyses are human programmed to look for specific things the researcher identifies. The ideal with machine learning is that they'll be able to ask questions independently and find patterns that people never thought to consider.
That's certainly possible and could be useful.
Of course the work needs to be "peer" reviewed, whatever that may mean, but if machines are just making things up, that's no different than their human minders. Scholars and researchers are getting busted for making stuff up all the time, sometimes upending whole fields. Those scholars don't admit they make the stuff up either. I don't think that's a unique problem. ...
The crucial difference is that real live people can ask other real live people to produce their sources. That, in fact, is how they get busted. And then they do admit their faults, or they lose so much reputationally that they just back quietly away from the whole scene and maybe stick to lower-level jobs.
None of that can happen with AI.
Well, not really. I do remember one researcher going after ChatGPT for unsubstantiated statements. The AI prevaricated and obfuscated like ... a human until it was finally nailed. https://www.theguardian.com/books/2023/ ... if-batuman is a long read but quite funny.
This one - https://www.nature.com/articles/d41586-023-03635-w - is more serious but paywalled.
But what happens when ChatGPT is nailed? Will "ChatGPT" be banned from publishing anything ever again? Or even forced to issue a retraction? Nope.

:coffee:
Kim
User avatar
Queequeg
Former staff member
Posts: 14493
Joined: Tue Jul 03, 2012 3:24 pm

Re: Amazing language translation with Google Gemini Advanced and ChatGPT

Post by Queequeg »

Well, this is why we need to remain in charge. That's going to be harder and harder as machine learning gets smarter than us. I don't see how we stay in charge except to keep training the machines to have ethics.
There is no suffering to be severed. Ignorance and klesas are indivisible from bodhi. There is no cause of suffering to be abandoned. Since extremes and the false are the Middle and genuine, there is no path to be practiced. Samsara is nirvana. No severance achieved. No suffering nor its cause. No path, no end. There is no transcendent realm; there is only the one true aspect. There is nothing separate from the true aspect.
-Guanding, Perfect and Sudden Contemplation,
Mirror
Posts: 329
Joined: Thu Oct 11, 2018 9:53 am

Re: Amazing language translation with Google Gemini Advanced and ChatGPT

Post by Mirror »

Dharma is very precious. So the correct translation is very important and the best translator is a practitioner, who through his/her experience and understanding of the mind, can translate correctly. It's not perfect, unexperienced translators can make bigger mistakes than AI, but AI can never be better than a skilled translator, because AI lacks a mind. I'm speaking only about translating buddhist texts. In worldly translation AI is going to be much better than a human.
Memento mori
Remember that you die
User avatar
Zhen Li
Posts: 2770
Joined: Sun Apr 07, 2013 8:15 am
Location: Tokyo
Contact:

Re: Amazing language translation with Google Gemini Advanced and ChatGPT

Post by Zhen Li »

I use these LLM tools on a daily basis for various tasks. One has to know "how" and "when" to use them as a tool, but never as an authority. These are not more intelligent than humans, only faster at retrieving information it has in its LLM. Any translator worth their salt would never just rely on an LLM translation, for modern or classical languages. A few hours of experience will show you how much nonsense it spews out alongside useful data. Every word and every sentence needs to be sifted through, and we still find major errors, in a process that can be just as laborious as translating oneself from scratch.

Just using the OP's examples, for instance. When you get beyond the fact that a computer is able to write something comprehensible and convincing, if you look at it as a translator, the examples are both inaccurate and unusable. The verses are split wrong, basic vocabulary is misidentified, and it doesn't have knowledge of the nuances of Buddhist Hybrid Sanskrit grammar as used in Dharanis and mantras that a scholar who works on these would know about. Moreover, it just hallucinated passages not in the source text, which a Buddhist human translator would view as a very serious fault.

How is this useful? Well, it is essentially a very fast dictionary look up tool. Suppose you are competant in Tibetan and Sanskrit grammar and you have a good sense of the general vocabulary, but you need a quick jog to the memory on words you haven't seen in a while, having these translations can help you to have an immediate look up, for what would take a few minutes with a regular online dictionary, or a half hour with a paper dictionary.

The second use case is in sorting out unusual syntax. It also isn't always right, but sometimes it can give you an idea of how syntax might fit together in a more complex sentence with subclauses and so forth.

Finally, I would say that even though Gemini is an advance over Bard, it is still less reliable than GPT4. For Tibetan, both are less reliable than MonlamAI. A translator would never "rely" on these, but might "refer" to them as they would to a dictionary.

How is it not useful? Basically, if people who don't know these languages rely on these translations as authoritative, they are guaranteed to be relying on wrong translations most of the time. If you don't read Buddhist languages, support translators who can interpret more accurately by purchasing their translations or donate to translation projects. Without the human translators, we won't know the rubbish from the Dharma.
Post Reply

Return to “Language”