Abstract: The present invention provides more accurate natural language searching capabilities by generating contextual phrases that are representative of the key words in a given query and uses those key contextual phrases to locate relevant documents through a search engine or database management system. The present invention generates such contextual phrases by first tagging the text using tagging assumptions and learning methods derived from the comparison of a domain specific and naively annotated corpus. Once tagged, the system then applies matrix rules to the tagged text to create a structural representation of the text. After the generation of the structural representation of the text, the system then applies phrase generation rules, which identify the relationships of the values in the matrix and from those relationships builds a concept phrase table that represents a pattern of contextual phrases derived from the query request.