Well, it was transformations. In generative semantics, transformations are/were taken to mediate between logical form and speech. The problem was, there turned out not to be any transformations. So much for that idea, then.
Probably amongst you all there are some readers who have not yet given up on TG. So I'll just say very briefly how I became disillusioned with transformations. It was Gerald Gazdar's account using phrase structure grammar of Ross's across-the-board condition for the coordinate structure constraint. That hardly even makes sense in TG (how can things move out of several places to wind up in just one place?), while it's predicted in Gazdar's treatment from the condition on coordination that only like categories can be conjoined.
I'm very fond of McCawley's The Syntactic Phenomena of English, and I think McCawley was headed in a good direction when he formulated the purpose of transformations as fitting conceptual structures into the templates that permit them to be pronounced in the way we humans have learned to do. It's just that the transformations don't work. Just look at that the very strange way Conjunction Reduction has to apply so as to yield structures that could have been produced directly by phrase structure rules. It's awful.