Topic of the thesis is the design and implementation of the Piggyback framework that supports natural language processing (NLP) scientists in experimenting with features for solving different kinds of statistical NLP problems. Special attention is given to the usage of web search results in feature functions, as substitute for world knowledge, which alleviates data sparseness.

This thesis makes the following contributions: The design and implementation of an extensible framework from a software engineering point of view, providing corpus parsers, statistical classifiers, access to web search engines and measures to reduce processing time, as well as an evaluation component. The evaluation of the framework as well as the piggyback approach for exemplary tasks such as Named Entity Recognition with focus on Named Entity Recognition in Query, Language Detection, and Coreference Resolution.