I am a PhD student working in the SiGHTSee Project. Our aim is to create better walking directions by synchronously generating verbal directions and 3D scenes from a unified knowledge representation. In this setting, I am especially interested in landmarks: Which objects are used as landmarks by humans? How do people refer to them in different situations, and how can we use this knowledge for automatically generating useful referring expressions? And of course, how can we find the optimal combination of verbal instructions and 3D graphics, allowing people to create a mental map of the route and easily reach their goal?