I think the best way to understand it is that you're creating an AST (a tree describing the program), for a bottom up parser you are recognising and making litt

I think the best way to understand it is that you're creating an AST (a tree describing the program), for a bottom up parser you are recognising and making little treelets and stitching them together into larger treelets until you reach the top/root, while for recursive descent you're building them from the top down from the root. The main difference is that LR (bottom up) family grammars can describe more languages than LL (top down) ones can largely because bottom up has slightly more contextual information available when it puts stuff together (often small treelets rather than just the next token)