Since the specifications of most of the existing context-sensitive graph grammars tend to be either too intricate or not intuitive, a novel context-sensitive graph grammar formalism, called context-attributed graph grammar（CAGG）, is proposed. In order to resolve the embedding problem, context information of a graph production in the CAGG is represented in the form of context attributes of the nodes involved. Moreover, several properties of a set of confluent CAGG productions are characterized, and then an algorithm based on them is developed to decide whether or not a set of productions is confluent, which provides the foundation for the design of efficient parsing algorithms. It can also be shown through the comparison of CAGG with several typical context-sensitive graph grammars that CAGG is more succinct and, at the same time, more intuitive than the others, making it more suitably and effortlessly applicable to the specification of visual languages.
