Developing a Sign Dictionary using Continuous Turkish Sign Data via Sign Spotting
Büşra Küden, Saliha Nur Şahin, Özge Can Özel, Lale Akarun, Murat Saraçlar
2026 34th Signal Processing and Communications Applications Conference (SIU)
Abstract
Sign language recognition and translation require large-scale signed data. In this study, I3D, DINO, and SHu-BERT architectures were used to automatically capture signs corresponding to target words in subtitles on the BUTID dataset compiled from television programs interpreted in Turkish Sign Language (TİD) and their capture performances were analyzed comparatively. It was shown that the system's performance was increased with an bootstrapping procedure supported by human validation.