This paper is the first full elaboration of use of our dynamically-updated verbal descriptions to support indoor spatial learning, cognitive mapping, and wayfinding behavior. In addition to manipulating information density of the descriptions, e.g. the verbal view depth, human performance is compared to a random walk model which shows that access to the verbal descriptions is significantly better than acting at chance. Importantly, there were no reliable differences between navigating real buildings with dynamically-updated verbal descriptions and with vision, suggesting the former can be as effective at supporting spatial learning and navigation behavior. These results opened the door to the use of virtual verbal displays described in several subsequent papers (above).