In this paper we present our ongoing work on integrating large-scale terminological information into NLP tools. We focus on the problem of selecting and generating a set of suitable terms from the resources, based on deletion, modification and addition rules. We propose a general framework in which the raw data of the resources are first loaded into a knowledge base (KB). The selection and generation rules are then defined in a declarative way using query templates in the query language of the KB system. We illustrate the use of this framework to select and generate term sets from a UMLS dataset.