SignCollect: A 'Touchless' Pipeline for Constructing Large-scale Sign Language Repositories

The projectteam of the Signbank project at the University of Amsterdam intends to substantially extend the NGT lexicon in Global Signbank within a limited timespan. To make this possible, the signCollect platform was developed to automate a major part of the workflow. The signCollect system includes a 'touchless' interface which enables a signer to control the system through simple gestures (recognized using computer vision) to (i) prompt the display of the next lexical entry, (ii) start a new recording, and (iii) approve/disapprove a recording. This capability allows a signer to record between 60 to 120 signs per hour, without the need for any assisting staff to be present. The approved recordings immediately become visible in the signCollect database, so that other members of the team can add metadata. With feedback from workshop participants we intend to further optimize the signCollect platform and make it available as an open-source tool for all sign language research teams.