Presumably, they're referring to chunking up the data into discrete semantic units—smaller vectorizable subsections (e.g., paragraphs) more precisely capturing

Presumably, they're referring to chunking up the data into discrete semantic units—smaller vectorizable subsections (e.g., paragraphs) more precisely capturing different parts of the data.