lexicon-induction
is a project aimed at inducing a lexicon from continuous sign language corpora. The project leverages large datasets of sign language videos, pose estimation technologies, sign language segmentation tools, and machine learning models to analyze and categorize sign language data.
For each video in corpus/videos
, we run pose estimation:
video_to_pose -i sign.mp4 --format mediapipe -o sign.pose
Output pose files are stored in poses
.
Pose sequences are automatically segmented using the sign language segmentation tool:
pose_to_segments -i sign.pose -o sign.eaf --video sign.mp4
Segmentation outputs (ELAN files) are stored in segments
.
For each segment in the ELAN file:
We focus on the softmax layer output for label distributions.
[^1]: Prillwitz, Siegmund, et al. "DGS Corpus project--development of a corpus based electronic dictionary German Sign Language/German." Sign-lang at LREC. 2008. [^2]: Crasborn, Onno, and Inge Zwitserlood. "The Corpus NGT: An online corpus for professionals and laymen." 2008. [^3]: Schembri, Adam, et al. "Building the British sign language corpus." Language Documentation & Conservation, vol. 7, 2013, pp. 136-154.