#DRRS2019@jahochcam in her favorite(?) topic: the tyranny of glossing! Glosses are not accessible without a visual representation — glosses+images should be the norm!
We may think that SL glossing is mostly a thing of the 80s, but it's still being done today. With non-transparent glosses, our documentation of #signlanguages will be lost: we cannot access the sign(ing) being described. #DRRS2019
Things that can be done (from NOW🤙🤙 on): 1. Stable ID glosses 2. Link to lexical database 3. Link to original data 4. Presenting data in publications with images/videos (alongside glossing)
Last night I was playing a little with Openpose data in #RStats. I realized it's not too hard to wrangle the Openpose output and plot signing directly using #ggplot2 and #gganimate, like so:
But I decided to make some tweaks so you can change the color of the signer+clothes, which makes seeing the hands a bit easier (contrast!)...
But also, why not give your signer a pretty turtleneck to wear?
You guys know that IKEA products are basically just #Swedish words and place names, right? Walking around an IKEA store is like walking through a dictionary.
This is a script simulating the idea in Swedish and other places/languages: github.com/borstell/fakea
So you can now input a video and it outputs it slower and/or repeated. Here's an example of a sign for 'deaf' in STS rendered with a repeated 30% speed playback!
(Oh, and passed to the make_gif() function as well!)
And the automatic face blurring works great! Even with multiple people in the image (or, like here, multiple repetitions of the same person in one composite image)!
So, it's like *very* easy to process and reconstruct actual images with only a few lines of code. As in plotting software redrawing the image, pixel by pixel.
Here's is a gif of me made with #ggplot2 and #gganimate. Sunday = fun day!