CNSL-bench: New Benchmark Evaluates MLLMs on Chinese Sign Language Understanding
Researchers have introduced CNSL-bench, the first comprehensive benchmark designed to evaluate multimodal large language models (MLLMs) on their ability to understand Chinese National Sign Language. The benchmark is grounded in the officially standardized National Common Sign Language Dictionary, ensuring authoritative and consistent semantic definitions. It provides multimodal coverage with aligned textual descriptions, illustrative images, and sign language videos. CNSL-bench also supports fine-grained analysis across key manual articulatory forms, addressing the underexplored area of sign language understanding in LLMs.
Key facts
- CNSL-bench is the first comprehensive benchmark for evaluating MLLMs on Chinese National Sign Language understanding.
- It is anchored to the officially standardized National Common Sign Language Dictionary.
- The benchmark includes aligned textual descriptions, illustrative images, and sign language videos.
- It supports fine-grained analysis across key manual articulatory forms.
- The research highlights the underexplored intrinsic ability of LLMs to understand sign language in multimodal contexts.
- The benchmark aims to mitigate ambiguity from regional or non-canonical variants.
- The work is published on arXiv with identifier 2604.22367.
- The research is motivated by advances in large language models.
Entities
—