Semantic graph is a structured representation of document meaning,
It combines coreference, WSD, and SRL information to determine which entities are mentioned in the document, which actions or events are described, and how they relate (i.e. which entities participate in which actions/events).

E.g., for the text: Mary says that Peter Smith bought a car that runs fast. Peter loves Mary. you would get a graph with three entities (Mary, Peter, and a car) and four events (say, buy, run, love), and you will also get relations between them (who says what, who buys what, who runs, who loves who).

This may be encoded in a RDF database in the form of triples and used for text mining applications. Or it may be used to compare documents and decide whether they describe the same events (even if they are in different languages), or many other applications.