NGT-LEX-SignBeach_W_part3

Read full text Summarise with VIRAT
Authors: Gomèr Otterspeer, Floris Roelofsen, Annika Schiefner, Beyza Sümer
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This dataset is part of the data collection NGT-LEX-SignBeach (https://doi.org/10.21942/uva.c.8233876).About the content of this data collectionThe NGT-LEX-SignBeach data collection contains video recordings of 1400 frequently used signs in Dutch Sign Language (NGT). Each sign has been recorded simultaneously with 5 cameras from different angles (M = Middle, L = Left, R = Right, A, B). Each sign was recorded with four different signers. Two of them faced the middle camera while signing, the other two oriented their body toward camera R while looking into camera M while signing. In total, 6 signers participated in the data collection (because not every sign could be performed by the same four signers, for reasons of limited availability).Video recordings are organized alphabetically into 67 datasets. Each video file is named using the pattern: word_signer_orientation_camera_video.mp4.The data collection also includes:HaMeR (Hand Mesh Recovery) derivative files (hamer.zip)SAM 3D Body segmentation derivative files (sam3dbody.zip)Pose estimation derivative files (pose.zip)Enriched metadata CSV with derivative availability (has_hamer, has_sam3dbody, has_pose)A codebook documenting all variables (PDF and TXT)About the creation of this data collectionThe data collection was created as part of the SignBeach project, a collaboration between the University of Amsterdam (UvA) and Prowise Learn. The main aim of the project was to create a serious game for learning NGT vocabulary. The game was integrated in the online learning environment that Prowise Learn provides for elementary schools in the Netherlands. The current data collection was primarily created to serve as content for the learning game.All videos were recorded in the recording studio of UvA’s SignLab in 2024-2026. The data collection methods are described in detail in the following paper: https://aclanthology.org/2026.signlang-1.46.pdfLicenseAll data is shared under a CC-BY-NC 4.0 license.FundingThe creation of this data collection was supported by the University of Amsterdam through the research programme Responsible Digital Transformations.Related datasetsFor each sign in the current data collection, we also collected iconicity and transparency ratings from different populations. These are available at https://osf.io/u7rkf/ and are discussed in the following paper: https://link.springer.com/article/10.3758/s13428-026-02969-3In the codebook, we provide a Signbank Gloss ID for each sign in the current data collection. This Gloss ID corresponds to an entry in the NGT Lexicon on Global Signbank (https://signbank.cls.ru.nl/datasets/NGT/). Each such entry in turn contains semantic and phonological information about the sign in question, and in some cases it contains example sentences as well.About this datasetSign language video recordings from the SignBeach project. Contains Dutch Sign Language (NGT) videos recorded from left (L), middle (M), and right (R) camera angles. Each video is named using the pattern: word_signer_orientation_camera_video.mp4.

Fields of Study & Subjects

Hand Gesture Recognition Systems Hearing Impairment and Communication Human Pose and Action Recognition

Publication details

Journal / source
Universiteit van Amsterdam
Publication date
28 March 2026
Language
EN
Document type
Dataset