The number of datasets on the web of data increases continuously. However, the knowledge contained therein cannot be fully utilized without finding links between the entities contained in these datasets. Equivalent entities cannot be identified solely by checking the equivalence of IRIs because of the different origins and naming schemes of different data providers. Yet, such equivalences can be discovered by computing the similarity of their attributes. In this paper we propose Glide, an approach that links entities from two different datasets by embedding a joint model of these datasets enriched by additional relations describing the similarity of literals. The joint model is embedded into a latent vector space while paying attention to juxtaposing similar literals. We evaluate our approach against state-of-the-art algorithms using real-world datasets commonly used in link discovery literature. The results show that Glide outperforms all baselines on 5 of 7 datasets with perfect or near-perfect accuracy. Our approach achieves its best performance on datasets that feature several literals with similarities. Our experiments indicate that researchers should not only pay attention to equal literals in knowledge graph embedding but should also be aware of the distance between similar literals.