In this paper we present a family of kernel functions, named Syntagmatic Kernels, which can be used to model syntagmatic relations. Syntagmatic relations hold among words that are typically collocated in a sequential order, and thus they can be acquired by analyzing word sequences. In particular, Syntagmatic Kernels are defined by applying a Word Sequence Kernel to the local contexts of the words to be analyzed. In addition, this approach allows us to define a semi supervised learning schema where external lexical knowledge is plugged into the supervised learning process. Lexical knowledge is acquired from both unlabeled data and hand-made lexical resources, such as WordNet. We evaluated the syntagmatic kernel on two standard Word Sense Disambiguation tasks (i.e. English and Italian lexical-sample tasks of Senseval-3), where the syntagmatic information plays a crucial role. We compared the Syntagmatic Kernel with the standard approach, showing promising improvements in performance.