A method and system for generating driving directions composes computer-based instructions that emulate a human driving perspective. Language-based instructions guide a driver along a route that encompasses a sequence of roads and intersections. An algorithm applies rules based on human perception to route components. The algorithm diagnoses road name changes so that instructions are clear and concise. The algorithm analyzes road and intersection geometry at confusing areas, such as road forks, roundabouts, ferry crossings, cloverleaf interchanges, and ramps, to generate instructions that conform to a driver's natural perspective. The algorithm analyzes the configuration of an intersection with respect to a driver's field of view to compose a clear and concise instruction. When appropriate, the algorithm produces compound instructions for adjacent intersections.