Semantic Problems of Thesaurus Mapping
With networked information access to heterogeneous data sources, the problem of terminology provision and interoperability of controlled vocabulary schemes such as thesauri becomes increasingly urgent. Solutions are needed to improve the performance of full-text retrieval systems and to guide the design of controlled terminology schemes for use in structured data, including metadata. Thesauri are created in different languages, with different scope and points of view and at different levels of abstraction and detail, to accommodate access to a specific group of collections. In any wider search accessing distributed collections, the user would like to start with familiar terminology and let the system find out the correspondences to other terminologies in order to retrieve equivalent results from all addressed collections. This paper investigates possible semantic differences that may hinder the unambiguous mapping and transition from one thesaurus to another. It focusses on the differences of meaning of terms and their relations as intended by their creators for indexing and querying a specific collection, in contrast to methods investigating the statistical relevance of terms for objects in a collection. It develops a notion of optimal mapping, paying particular attention to the intellectual quality of mappings between terms from different vocabularies and to problems of polysemy. Proposals are made to limit the vagueness introduced by the transition from one vocabulary to another. The paper shows ways in which thesaurus creators can improve their methodology to meet the challenges of networked access of distributed collections created under varying conditions. For system implementers, the discussion will lead to a better understanding of the complexity of the problem.