Referring to objects or entities in the space, as well as to relations holding among them, is one of the most important functionality in natural language understanding. As a result, the detection of spatial utterances finds many applications, such as in Spatal Relation Extraction, GPS navigation systems, or Human-Robot Interaction (HRI). In this presentation a system that participated to the Spatial Role Labeling SemEval task will be presented. The spatial roles classification is addressed as a sequence-based word classification problem: the SVM-hmm learning algorithm is applied, based on a simple feature modeling and a robust lexical generalization achieved through a Distributional Models of Lexical Semantics. In the identification of spatial relations, all roles found in a sentence are combined to generate candidate relations, later verified by a SVM classifier. The Smoothed Partial Tree Kernel is here applied, i.e. a convolution kernel that enhances both syntactic and lexical properties of the examples.