Video Content Representation with Grammar for Semantic Retrieval