The purpose of this report is to present the study of a linguistic network based on the relation of synonymy. The network has been extracted from a dictionary of synonyms in French. Due to errors and ambiguities in the data, the synonyms are not always grouped into distinct components. To refine the results, we apply a clustering algorithm to the network. Each densely connected cluster is considered to correspond to a general concept in the language. Finally we develop a new general method, based on the introduction of noise in the network, to identify nodes which lie on the border between clusters and to evaluate the robustness of the clustering. This method can be applied with any clustering algorithm, provided that the algorithm works on weighted networks.