Show HN: Bible translated using LLMs from source Greek and Hebrew (biblexica.com)

55 points by epsteingpt ↗ HN
Built an auditable AI (Bible) translation pipeline: Hebrew/Greek source packets -> verse JSON with notes rolling up to chapters, books, and testaments. Final texts compiled with metrics (TTR, n-grams).

This is the first full-text example as far as I know (Gen Z bible doesn't count).

There are hallucinations and issues, but the overall quality surprised me.

LLMs have a lot of promise translating and rendering 'accessible' more ancient texts.

The technology has a lot of benefit for the faithful, that I think is only beginning to be explored.

23 comments

[ 3.0 ms ] story [ 43.7 ms ] thread
What is the "expanse"?

Answer: The sky. The ancient people who wrote the bible thought the sky was a solid dome that separated "the water's above" (aka rain) from the water's below. God lived on the other side of this dome.

This is confirmed later in Genesis with the Tower of Babel story.

They tried to reach this dome by building a tower. And "god" was so offended by their ignorance and stupidity (which he perpetrated) that he decided to punish them.

The "faithful" obviously reject this simple interpretation in favor of something more obtuse and mystical.

What is the difference between the "Adam" translations and the "Eve" translations? Where can I read about this more?
Stuck with a "Loading..." message. Too much traffic, maybe?
The Bible is too well-known a text that is too represented in training datasets for this _not_ to be skewed towards poorly reproducing existing translations.

Beyond that,

>there are hallucinations and issues

seems like a deal-killer for a religious text. Yes, all translation by humans is an act of interpretation on some level, and so there's lossiness in all translation – but the difference between a human carefully weighing their reasoning for a particular choice of rendering vs. an LLM that is basically weighted dice that might land totally wrong is a categorically-different thing, not a question of degrees.

Is this translation public domain?
I would be really interested in this done to the Peshitta Bible, which is roughly as old as the Septuagint. Peshitta is in Aramaic a sister language to Hebrew. Over the years I've found interesting insights about verses that make way less sense in Greek but in Aramaic they make drastically more sense. It seems that somehow the Greek translated from some other source where in Aramaic or Hebrew the word used could have been one of two words, the Greek seemed to pick the worst possible representation in some cases that the Aramaic highlights.

For example. It is easier for a Camel to go into the eye of a needle than a rich man to get into Heaven. If you read this, it makes it sound like Abraham cannot get into Heaven, wasn't he wealthy? Heck, there's others who were wealthy in scripture, even kings are they all doomed? In Aramaic the same word that in Greek is said to mean camel, can also mean rope.

If you think about a rope going through the eye of a needle, and what it TAKES for a rope to go through the eye of a needle, aka removing all the threads or layers (humbling the person and forcing them to strip themselves down to their core) in order to make it through the eye of the needle. Or in other words, you must be willing to dethatch yourself from all your wealth. Remember the guy who asked Jesus was he must do to be saved and enter heaven, and walked away when Jesus told him to give away everything he owned to the poor? That is the same exact message.

There's a few other verses, but that's the main one that always strikes me. Some of them are far more nuanced and I get into hours of debate with people who are ignoring everything I am saying (I don't know why, I try to lay it all out in the most simple way possible) as if I'm breaking the law, but its obvious to me that we don't have perfect copies of the Bible. I still think the overall message is the same though, so nothing wrong with that. It proves yet again that men are all fallible.

Sorry for the tangent. I used to deep dive translations and their nuances, and the Aramaic based Bibles are very interesting.

There's also an Aleh Tav Old Testament Bible which is fascinating to me. It adds the Aleph Tav anywhere it would be in the Hebrew into the English.

these books are social in nature.. it takes agreement to make revisions. The agreement process is part of a spiritual path, in multiple ways. One translation by a machine is new but newness is exactly not the point of these works. Stability through generations and resolving open theological questions to some extent, are much more the point than newness. Also, pride and vanity are expressly discouraged. If this is a personal achievement somehow, its not very consistent with the core teachings. You ought to acknowledge your teachers, your spiritual community, their leaders and elders, and other inputs.. because without all of that you would not be able to complete a non-trivial work of scholarship. other ideas missing?
I'm not sure how this is accomplished, but I like the "poetic" translation a lot more than the "optimal" one.

Which reminds me, do you think it's possible that the stories in the Bible are actually mystic symbolism and "veiled truth" (like the sort of stories that you might get in a dream) and people have mistaken it for actual physical history (with which it's obviously incompatible)?

The parables of Jesus come to mind. They weren't meant to be taken literally but to teach, to get a point across.

> but the overall quality surprised me.

With all due respect, how are you in any position to be able to objectively evaluate the quality assuming you’re not fluent in Hebrew and Greek?

There is 100% chance there are already translations in the dataset (including texts with Hebrew - English side by side etc).
How was the tranlation for the chapters in Aramaic, like in Daniel?
What's up with these?

Genesis 1:13, Eve optimal Replace 'Then' with 'And' in optimal ('And the LORD God said') and poetic_daily to preserve narrative vav-consecutive connective consistently.

LLMs initial purpose is translation from one language to another, nothing surprising they are good in that.
One Bible translation that I really appreciate is the NET Bible [0] -- in particular, I appreciate its translator's notes. It can be very helpful to read the translator's notes and to understand the reasoning that went into any particular rendition. I.E., something like "The disjunctive clause (conjunction + subject + verb) at the beginning of v. 2 gives background information for the following narrative, explaining the state of things when “God said…” (v. 3)..."

Did you use a reasoning model to translate these verses? If so, I would be very interested in seeing the breakdown that the LLM used that went into each verse.

I understand that such breakdowns can be hallucinated at many levels also (and final output does not always correspond with the reasoning flow), but I (personally) would find this helpful.

[0] https://bible.org/sites/bible.org/resources/netbible/

(comment deleted)
(comment deleted)
Which set of NT Greek manuscripts is it using? Textus Teceptus? Byzantine? Critical Text?
I heard or read that the LLM translation system is trained upon Bible translations because the Bible has been translated into more languages than any other book.
this is an important question. I am not a specialist in this area. I believe what you say was true before the invention of the Transformer ML architecture around 2015. I believe that among practitioners close to the Transformer effort, they passed around "The Common Crawl" because it was standard and basically one file chunk. I suspect that Bible material was just one part of CommonCrawl. more info welcome
Nice, but some things do require an informed human touch to really get right.