This dissertation develops methods to discover and disambiguate named entities, thus linking texts to knowledge bases. The first contribution is a robust disambiguation method using a graph algorithm that makes use of the coherence among entities in the input. The second contribution is a novel model to compute the coherence among entities that works especially well for lesser known entities and is applicable to newly emerging entities. The third contribution addresses the discovery of emerging entities by modeling the entities not present in the knowledge base in an explicit manner. Finally, two applications using the developed entity disambiguation methods are presented.