Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition | Synapse