Open-Source Code for Hand Gesture Recognition – Sign Language Translation | Hacker News Reader

Open-Source Code for Hand Gesture Recognition – Sign Language Translation

github.com

91 pointsbgrynol11 comments

There is nothing here close to sign language translation. Hand gesture are a very small and quite easy part of sign language who involve all the upper part of the body and face and use a lot of context and placement in space.

Yep this is fingerspelling, a small part, used for names and things there aren't signs for.

Here's an instructor running at medium speed for clarity. Experts go faster, drop more letters bcs cn undrstd wthout thm, and combine motions, like dipthongs.

https://www.youtube.com/watch?v=yizRk2CP9gs

What amazes me it’s that a relatively new language easy to use (no voice, no accent distortion) it’s not a global standard and we need tools like this to translate the different language signs.

This situation feels very stupid IMO.

It is a common misconception (dare I say prejudice) that gestures are "universal" and therefore easy to learn/unify sign languages. Sign languages are a full-fledged natural language with nuances and complexities, just like spoken languages. Their grammar tends to be spatial as well as temporal, and a typical signer can express abstract concepts and its relationships using different placement/movement of signs. The reason why it's diverse is that most sign languages have been organically developed (just like spoken languages pre radio/TV era) in mostly isolated deaf communities.

Even just a simple head movement depends on the culture, as nodding for "yes" and shaking the head for "no" is not universal [0][1]. And this doesn't even account for something like the "indian head bobble". [2]

Another example is the use of index finger and thumb to create an O. [3]

So if these very simple examples depend on the culture, I expect signing to be much more localized.

[0]: https://en.wikipedia.org/wiki/Nod_(gesture)

[1]: https://en.wikipedia.org/wiki/Head_shake

[2]: https://en.wikipedia.org/wiki/Head_bobble

[3]: https://en.wikipedia.org/wiki/OK_gesture#Negative_connotatio...

The challenge with having universal sign language, is that people communicate in over 300 different sign languages around the world.

Even in English speaking countries, like the United States vs. New Zealand, both countries have two different forms of sign language (ASL vs. NZSL) which are communicated amongst the Deaf and Hard of Hearing communities.

Sign languages are not new but just not as well documented and researched as spoken languages because pre film/video where was no easy way to record language use. Most sign languages have no writing system. Additionally for the past 150 years the deaf community have been oppressed. Some researchers think some sign languages date back further than the first spoken languages.

Looks like you have a few misconceptions there. There are many different sign languages, just like there are many different spoken languages, and they developed in much the same way. Sign languages definitely have accents and dialects, and are not inherently easier (they're full complex languages with all kinds of nuances). This tool doesn't translate sign language signs, or even detect them. Your argument basically amounts to "everyone should speak Esperanto" or similar.

Can someone change the title? It seems to be Google open sourcing the code, not GitHub.

There's also a fair distance between hand detection and "Sign Translation". To begin with, most sign has both hands overlapping.

Ha, I read the headline and thought it was "git by gesture".

This is because I'd recently watched "Perl out loud", a talk this year by Emily Shea (2shea): https://www.youtube.com/watch?v=Mz3JeYfBTcY

Reply on news.ycombinator.com