The textual geographical information is frequently organized around spatial named entities. Such entities have intrinsic ambiguities and Named Entity Recognition and Classification methods should be improved in order to handle this problem. This article describes a knowledge-based method implementing a full process with the aim of annotating in a more precise way the spatial information in the textual documents. This gain in accuracy guarantees a better analysis of the spatial information and a better disambiguation of places. The backbone of our proposal is a construction grammar and a cascaded finite-state transducers. The evaluation shows that the introduced concept of hierarchical overlapping, is very helpful to detect a local context associated with Named Entities.