We introduce the theory of Frame Semantics, in which word senses are defined in relation to semantic frames. The prototypical semantic frame is a conceptualization of a type of event, with its participant roles (frame elements, FEs), props, and circumstances, but frames can also represent classes of relations, states, and even entities. A lexical unit (LU = word sense) is defined as a relation between a lemma and a semantic frame ; we give three examples of frames, each with FEs, LUs, and annotated examples, and introduce the FrameNet (FN) Project (http://framenet.icsi.berkeley.edu), which is building a frame semantic lexicon of English with 960 + frames and 11,700 + LUs. We show how FN displays the valence  information for an LU, and how semantic dependency graphs can be derived from  FNannotated sentences. We discuss relations between frames and FEs, the structure of the FN database, FrameNets for other languages, the limitations of FN, and a new project to annotate syntactic constructions in a similar way.