Sign Language Recognition

Sign language recognition studies how to identify isolated signs, fingerspelled words, and signs within continuous signing, often from video or body and hand keypoints. Common approaches include CNNs and Transformers for visual features, temporal models and CTC for continuous sequences, and self-supervised or contrastive learning to use limited annotations; few-shot and cross-lingual methods aim to extend recognition to new signs and languages. Benchmarks such as WLASL, PHOENIX-2014T, and ASL Citizen support comparison, but strong results on familiar signers or fixed vocabularies do not guarantee performance on unseen people, viewpoints, or real conversations. A central challenge is building broader, well-annotated datasets and models that capture linguistic structure and generalize to low-resource languages and natural signing.