A Data-Driven Model For The Generation Of Prosody From Syntactic Sentence Structures