search

jincheng-yang / shorthand

Gregg shorthand recognition and translation
8 stars 2 forks source link

Informative: other Gregg software and a modest proposal #2

Open rebcabin opened 5 years ago

rebcabin commented 5 years ago

Hi, Teddy ... I really like your dictionary https://teddyyjc.github.io/shorthand.html. I wanted to let you know about another English-to-Gregg resource, http://steno.tu-clausthal.de/Gregg.php. That one does ligatures, phrasing, and some brief forms. The author, Sarman, of that software has also published slides from a talk wherein he shows his adventures with metafont, but I do not currently have a link to those slides.

My proposal to you is two-fold: (1) to work with me and with him to build a Gregg-to-English translator. I believe such will be possible by domain-randomization on the output of Sarman's translator to simulate the variations that would be generated by human hands. In theory, one might run 10^6+ books from the public domain to work up a corpus for training Gregg OCR neural nets a-la Yann LeCun 1988. I wrote a letter to Sarman asking him to open-source his software and data so that one might build new software from it or on it, but Sarman did not respond to me. If you became interested and wrote to him as well, perhaps he might be more motivated. (2) To open-source your software and data for similar reasons: so that one might build new software from it and on it. I see that you have some software here, but it doesn't look like your dictionary. Perhaps I didn't look hard enough.

jincheng-yang commented 1 year ago

Hello Rebcabin,

I am very sorry that I did not see your message until today, more than three years late. I had no idea that you left me a message, and I am not sure if you will receive some notification from my reply.

I am very interested in your proposal. In fact, I have been thinking about this for quite a while, but I am not a programmer, and my knowledge about neural networks or OCR is very limited. If you are still interested (I understand it is quite late now), please email me, and we can discuss more.

Best regards, Jincheng.

rebcabin commented 1 year ago

I'm going to be looking into ML in a few weeks via a new book called "The Little Learner." I have some rough ideas how to go about Gregg -> English, but rough ideas are a long way from "code that works." It is a serious shame that Sarman cannot release his great English -> Gregg ( https://tug.org/tug2008/abstracts/sarman.pdf). It's a great achievement but of no value to the World because we can't use it or modify it! A tragic loss!

On Mon, Feb 6, 2023 at 2:20 PM Jincheng Yang @.***> wrote:

Hello Rebcabin,

I am very sorry that I did not see your message until today, more than three years late. I had no idea that you left me a message, and I am not sure if you will receive some notification from my reply.

I am very interested in your proposal. In fact, I have been thinking about this for quite a while, but I am not a programmer, and my knowledge about neural networks or OCR is very limited. If you are still interested (I understand it is quite late now), please email me, and we can discuss more.

Best regards, Jincheng.

— Reply to this email directly, view it on GitHub https://github.com/jincheng-yang/shorthand/issues/2#issuecomment-1419857705, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABSRR6PABNPD732OPYZWLLWWF2JXANCNFSM4INYAY5Q . You are receiving this because you authored the thread.Message ID: @.***>

jincheng-yang commented 1 year ago

I agree. Have you tried to contact Sarman?

rebcabin commented 1 year ago

Yes, I wrote to Sarman and he said his employer owns the code and will not release it to open source. The work is interesting but useless. I estimated it would take me 3 to 6 months full-time to reproduce the code from the high-level description in the paper, which leaves out all details. The mathematical description is understandable, but it's just a lot of manual labor to reduce it to code.

On Tue, Apr 18, 2023 at 5:30 PM Jincheng Yang @.***> wrote:

I agree. Have you tried to contact Sarman?

— Reply to this email directly, view it on GitHub https://github.com/jincheng-yang/shorthand/issues/2#issuecomment-1513952167, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABSRR56B2PFPQX7J4DKXLTXB4W3XANCNFSM4INYAY5Q . You are receiving this because you authored the thread.Message ID: @.***>

Zireael07 commented 5 months ago

@rebcabin That's sad that he can't/won't release the code.