Richly annotationed model

This is a general question.

Given that beta testing has been limited to existing KGs I've been unable to determine the extent that model annotations (descriptions for classes, properties, etc) contribute to generating queries. Once Voicebox is commercially available I would take a populated KG with a richly annotated model and seek answers to natural language questions then would ask the same questions of a KG with the same model and data, only with a less richly described model, and see if the answers changed.

Is there any guidance that Stardog can provide regarding a richly annotated model versus a less richly annotated model and the impact that would have in generating SPARQL and answers to natural language questions?

Best regards...