Brief

Researchers have introduced a new task and dataset for emotion recognition in sign language conversations, addressing the limitations of existing models that struggle with conversational context. The eJSL Dialog dataset, comprising 1,920 video samples from 480 dialogues, was benchmarked using various models, revealing a significant domain gap for generic emotion recognition systems. The findings highlight the need for context-aware visual extractors specifically designed for sign language and suggest that larger conversational datasets are crucial for future pre-training efforts. AI