Several challenges face the interlinking of large RDF datasets. Perhaps some of the most salient are the polysemy of natural language labels and the lack of reference anchors or starting points. In this paper we propose a novel interlinking method, called TERA, driven by text annotation. The proposed method consists in annotating the same textual context with two different datasets and comparing the selected RDF resources. We discuss the adequacy of the annotation methods with this particular task and evaluate the precision of the interlinking method in preliminary experiments.