Automated Extraction of Prosodic Structure from Unannotated Sign Language Video

This repository contains the code for the titular article, presented at LREC-COLING 2024.

Apart from the methodology itself, a command line tool is included that makes reproducing our experiments (hopefully) easy, and allows using our code to extract some information from unannotated video.

This code is free and open source software, licensed under the EUPL.

You can now try it in Colab:

Features

• Command line interface
• Install with pip: pip install sign-prosody-extraction
• Plot the prosodic structure (velocity and direction profile of articulators along the temporal axis) of a sign language video.
• Extract thumbnails at "target points": points of low velocity, i.e. high static visual significance.
• Extract a short clip that includes only the sign articulation, ignoring preparation and relaxation. This can help dictionary makers prepare high significance animated thumbnails of their videos.
• Use either cotracker or mediapipe as outlined in Börstell (2023) for the articulator tracking.
• Export the data as CSV to use it in ELAN or similar.

Usage

pip install sign-prosody-extraction
# To compute everythig for VIDEO.mp4, and save in output/
sign-prosody-extraction VIDEO.mp4 --output-dir output/ --everything
# To see the different options
sign-prosody-extraction --help

You can modify some further runtime options with environment variables. Set CACHE_DIR and optionally CACHE_LIMIT to cache some long computations. To use the mediapipe algorithm, the pose_landmarker.task model must be downloaded, and placed in the data folder or in a path pointed to by the env var POSE_LANDMARKER. The cotracker model will be loaded from torchhub.

Article Abstract

As in oral phonology, prosody is an important carrier of linguistic information in sign languages. One of the most prominent ways this reveals itself is in the time structure of signs: their rhythm and intensity of articulation. To be able to empirically see these effects, the velocity of the hands can be computed throughout the execution of a sign. In this article, we propose a method for extracting this information from unlabeled videos of sign language, exploiting CoTracker, a recent advancement in computer vision which can track every point in a video without the need of any calibration or fine-tuning. The dominant hand is identified via clustering of the computed point velocities, and its dynamic profile plotted to make apparent the prosodic structure of signing. We apply our method to different datasets and sign languages, and perform a preliminary visual exploration of results. This exploration supports the usefulness of our methodology for linguistic analysis, though issues to be tackled remain, such as bi-manual signs and a formal and numerical evaluation of accuracy. Nonetheless, the absence of any preprocessing requirements may make it useful for other researchers and datasets.

• Read more at the LREC-COLING 2024 proceedings

How to cite

If you use our code, please do cite us! Also don't hesitate to get in touch.

@inproceedings{sevilla-etal-2024-automated-extraction,
    title = "Automated Extraction of Prosodic Structure from Unannotated Sign Language Video",
    author = "Sevilla, Antonio F. G. and Lahoz-Bengoechea, Jos{\'e} Mar{\'\i}a and Diaz, Alberto",
    editor = "Calzolari, Nicoletta and Kan, Min-Yen and Hoste, Veronique and Lenci, Alessandro and Sakti, Sakriani and Xue, Nianwen",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.161",
    pages = "1808--1816",
}

Authors

Related software

• cotracker
• mediapipe
• SciPy
• Scikit-learn
• Rye
• Click