Adding "BEST: BERT Pre-training for Sign Language Recognition with Coupling Tokenization"

[cleong110]

My notes on the process: cleong110#19

A few things:

• No I don't know what BEST stands for, it's not called out explicitly in the paper but I believe it is BERT pre-training for Sign language recognition with coupling Tokenization
• Might be good to double-check some of the stuff about coupling tokenization, I might not have understood it properly
• extremely similar in a number of ways to Adding SignBERT+ #54, to the point where I could not resist referencing it, and I think we need to merge that one first. Similar approach, evaluated on the same datasets with similar results. Even the table layouts are similar, with both of them calling their method "Ours +R" in the "from video" results. Neither one offers any code.

TODO:

• rewrite for conciseness
• Merge Adding SignBERT+ #54 first

[cleong110]

Merged latest, working on rewrites

[cleong110]

All right did some rewrites. https://chatgpt.com/share/f42199e8-bd03-4ec3-9913-e5e832bb2885 had various suggestions, some of which I incorporated. It consistently struggles with citations that start with [dataset:, and will sometimes write out the rendered citation, and has a few other issues, but generally seems helpful