Quelle: Including Signed Languages in Natural Language Processing Kanal: Amit Moryossef Erschienen: 2021-05-24 Gesehen am: 2023-01-04 → linguistics

  • 01:32 There is no single sign-language, they also don’t rely on spoken languages

  • 02:50 Sign Language heavily depends on body movement, facial expressions and the surrounding environment

  • 04:59 Words sometimes spelled out in letter alphabet

  • 04:37 Since sign language is not the dominant language, it is subject to the same changes that happen to other minority languages → loan words

  • 06:46 SignWriting is a way to convert sign language into simple pictograms, HamNoSys does the same but is more complex

  • 07:45 Resources for NLP tasks are relatively rare since video-based analysis is subject to privacy concerns, it’s hard to anonymize

  • 13:45 Writing down sign language is not only useful for NLP and to faciliate data sharing but also for archival purposes

  • 15:15 Continuous SignWriting ~= ASCII videos of a stick figure signing

    • How de tokenize that? How do we parse it?
    • Deconstructing written signs → map to words
  • 22:52 Using StyleGAN to adjust images based on text prompts → raising eyebrows

  • See also Including Signed Languages in Natural Language Processing - Full Talk at ETH - YouTube

  • Holistic | mediapipe