In this paper we describe a markup language for semantically annotating raw texts. We define a formal representation of text documents written in natural language that can be applied for the task of Named Entities Recognition and Spatial Role Labeling. The proposal relies on a multi-layer annotation process based on a core generic layer, which can be freely adapted into more specific layers depending on the intended goal. Our markup language is based on the TEI Guidelines1 to propose a generic and extensible markup language. This language is particularly dedicated for the text mining task and ready to use to be layered with more semantic relationships between elements of the text. We show the feasibility of this proposal from a generic annotation of texts describing itineraries toward a geospatial semantic annotation.