Handwriting / Character Recognition for Cosmere Writing

[YungDankBlast]

I'm in a deep learning class in my master's program, and I've been considering working on a project where I train or fine-tune an AI model to do character recognition or handwriting recognition on cosmere writing (steel alphabet, women's script, etc.) to "translate" the characters to english. I'm also aware that some of the writing systems do not have a one-to-one relationship with english characters (ex. No "q" in steel alphabet, "u"/"w" are the same in women's script, etc.), so it would be interesting to apply some method of converting from a direct character translation to actual english writing. Does anyone know whether someone has done a project like this already or whether datasets of handwritten cosmere languages exist? Is this something people would be interested in following? Maybe I can get together some kind of repository where people can submit their own cosmere writing if there is enough interest.

[OOKLA_the LIV She/They/Lesbian/hungry]

17 hours ago, YungDankBlast said:

I'm in a deep learning class in my master's program, and I've been considering working on a project where I train or fine-tune an AI model to do character recognition or handwriting recognition on cosmere writing (steel alphabet, women's script, etc.) to "translate" the characters to english. I'm also aware that some of the writing systems do not have a one-to-one relationship with english characters (ex. No "q" in steel alphabet, "u"/"w" are the same in women's script, etc.), so it would be interesting to apply some method of converting from a direct character translation to actual english writing. Does anyone know whether someone has done a project like this already or whether datasets of handwritten cosmere languages exist? Is this something people would be interested in following? Maybe I can get together some kind of repository where people can submit their own cosmere writing if there is enough interest.

THAT IS AWESOME.

I have actually been learning to write in Alethi Womens script for a while now.

I think that would be a super cool project. I dont know of data sets.....

but I am not an expert.

Here, i shall summon people who have been here longer.

@Through The Living Glass @Doomslug Doomslug Doomslug @kpShadowFox @Vyzkel

 

[kpShadowFox He/Him]

19 minutes ago, LIV_SB-DustBringer said:

THAT IS AWESOME.

I have actually been learning to write in Alethi Womens script for a while now.

I think that would be a super cool project. I dont know of data sets.....

but I am not an expert.

Here, i shall summon people who have been here longer.

@Through The Living Glass @Doomslug Doomslug Doomslug @kpShadowFox @Vyzkel

 

look, I joined beginning of october. I haven't been around as long as it seems I have.

As of my (very limited) knowledge, no, this hasn't been done yet.

Also, I would most definitely be interested in this

[OOKLA_the LIV She/They/Lesbian/hungry]

2 minutes ago, kpShadowFox said:

look, I joined beginning of october. I haven't been around as long as it seems I have.

As of my (very limited) knowledge, no, this hasn't been done yet.

Also, I would most definitely be interested in this

youve been around longer than me!

and thanks anyway

[Through the Living Elan He/Him]

19 hours ago, YungDankBlast said:

I'm in a deep learning class in my master's program, and I've been considering working on a project where I train or fine-tune an AI model to do character recognition or handwriting recognition on cosmere writing (steel alphabet, women's script, etc.) to "translate" the characters to english. I'm also aware that some of the writing systems do not have a one-to-one relationship with english characters (ex. No "q" in steel alphabet, "u"/"w" are the same in women's script, etc.), so it would be interesting to apply some method of converting from a direct character translation to actual english writing. Does anyone know whether someone has done a project like this already or whether datasets of handwritten cosmere languages exist? Is this something people would be interested in following? Maybe I can get together some kind of repository where people can submit their own cosmere writing if there is enough interest.

This could be a pretty good start:

https://coppermind.net/wiki/Category:Language

It has at least a few, each offshoot might have languages.

I know there are a few on Scadrial, obviously there's Aons which are logographic and thus more difficult, and there's... more other ones. 

[Through The Living Glass She/They]

1 hour ago, LIV_SB-DustBringer said:

THAT IS AWESOME.

I have actually been learning to write in Alethi Womens script for a while now.

I think that would be a super cool project. I dont know of data sets.....

but I am not an expert.

Here, i shall summon people who have been here longer.

@Through The Living Glass @Doomslug Doomslug Doomslug @kpShadowFox @Vyzkel

 

Yo.

*reads*

Yeah, I've never seen anything like that done before. It's a really cool idea!

I know there's a bit on the Coppermind, like Elan oh-so-helpfully linked above, but are you looking for more organic-seeming writing? Is that what you're asking? I may need a bit more clarification on what you're looking for.

[YungDankBlast]

30 minutes ago, Through The Living Glass said:

Yo.

*reads*

Yeah, I've never seen anything like that done before. It's a really cool idea!

I know there's a bit on the Coppermind, like Elan oh-so-helpfully linked above, but are you looking for more organic-seeming writing? Is that what you're asking? I may need a bit more clarification on what you're looking for.

I think my basic idea right now is to use an image-to-text encoder-decoder, which would take in a picture of the script and output the translation in english. I think it would be cool to eventually have some handwritten organic writing; however, that type of data is potentially difficult to collect, especially with accurate translations, and so I may start by using one of the already created custom fonts that I have seen for women's script or the steel alphabet to convert an existing text dataset into one of those languages. Eventually, it would be cool to refine that training with actual handwritten data if it is available.

Honestly, I'm just looking for any related data. I'll start with Coppermind, then go from there.

Edited November 14, 2024 by YungDankBlast

[Treamayne]

On 11/13/2024 at 5:30 PM, YungDankBlast said:

I'm in a deep learning class in my master's program, and I've been considering working on a project where I train or fine-tune an AI model to do character recognition or handwriting recognition on cosmere writing (steel alphabet, women's script, etc.) to "translate" the characters to english. I'm also aware that some of the writing systems do not have a one-to-one relationship with english characters (ex. No "q" in steel alphabet, "u"/"w" are the same in women's script, etc.), so it would be interesting to apply some method of converting from a direct character translation to actual english writing. Does anyone know whether someone has done a project like this already or whether datasets of handwritten cosmere languages exist? Is this something people would be interested in following? Maybe I can get together some kind of repository where people can submit their own cosmere writing if there is enough interest.

Deconstruction and methods of decryption were primaily in this thread (to include Glyph basics): SA Translation

Technically, all of Navani's Journal is handwritten Women's Script (just by Isaac, and done with paint brushes, so more of a calligraphic style)

Evolution of the Steel Alphabet will make characters dependant upon the era from which the text originates - with things like Alendi's Logbook likely to be an entirely separate alphabet system (based on how Sazed needed to go about translation - and Isaac's comments on Evolution of Scadrian Glyphs)

Hope that helps.