We present a system for identifying the semantic relationships, or semantic roles, lled by constituents of a sentence within a semantic frame. Various lexical and syntactic features are derived from parse trees and used to derive statistical classi ers from hand-annotated training data.