are essential steps for many linguistic tasks such as information extraction and text categorization. A variety of
named entity disambiguation methods have been proposed, but most of them focus on Wikipedia as a sole
knowledge resource. This focus does not fit all application scenarios, and customization to the respective application domain is crucial.
This dissertation addresses the problem of building an easily customizable system for named entity disambiguation.
The first contribution is the development of a universal and flexible architecture that supports plugging in different
knowledge resources. The second contribution is utilizing the flexible architecture to develop two domain-specific
disambiguation systems. The third contribution is the design of a complete pipeline for building disambiguation
systems for languages other than English that have poor annotated resources such as Arabic. The fourth contribution
is a novel approach that performs fine-grained type classification of names in natural language text.