Udo Hahn

Medical knowledge reengineering—converting major portions of the UMLS into a terminological knowledge base

We describe a semi-automatic knowledge engineering approach for converting the human anatomy and pathology portion of the UMLS metathesaurus into a terminological knowledge base. Particular attention is paid to the proper representation of part-whole hierarchies, which complement taxonomic ones as a major hierarchy-forming principle for anatomical knowledge. Our approach consists of four steps. First, concept definitions are automatically generated from the metathesaurus, with LOOM as the target language. Second, integrity checking of the emerging taxonomic and partonomic hierarchies is automatically carried out by the terminological classifier. Third, terminological cycles and inconsistencies are manually eliminated and, in the last step, the knowledge base built this way is incrementally refined by a medical expert. Our experiments were run on a terminological knowledge base which is composed of 164,000 concepts and 76,000 relations. Empirical evidence for the lack of logical consistency, adequacy and improper granularity of the UMLS knowledge source is given, and finally, assessments of what kind of efforts are needed to render the formal target representation structures complete and empirically adequate.

Parts, Locations, and Holes - Formal Reasoning about Anatomical Structures

We propose an ontology engineering framework for the anatomy domain, focusing on mereotopological properties of parts, locations and empty spaces (holes). We develop and formally describe a basic ontology consisting of the mutually disjoint primitives solid object, hole and boundary. We embed the relations part-of and location-of into a parsimonious description logic (ALC) and emulate advanced reasoning across these relations - such as transitivity at the T-Box level - by taxonomic subsumption. Unlike common conceptualizations we do not distinguish between solids and the regions they occupy, as well as we allow solids to have holes as proper parts. Concrete examples from human anatomy are used to support our claims.

Joint knowledge capture for grammars and ontologies

We introduce a methodology for automating the maintenance and growth of domain-specific concept taxonomies and grammatical class hierarchies simultaneously, based on knowledge capture from natural language texts. The assimilation process is centered around the linguistic and conceptual `quality of various forms of evidence underlying the generation, assessment and on-going refinement of lexical and concept hypotheses. On the basis of the strength of evidence, hypotheses are ranked according to plausibility, and the most reasonable ones are selected for assimilation into the given lexical class hierarchy and domain ontology.

Context-Based Ambiguity Management for Natural Language Processing

We introduce a formal context mechanism, embedded into a description logics framework, which allows to uniformly represent and manage different kinds of natural language ambiguities as they occur in the course of text understanding. Multiple lexical, syntactic and semantic interpretations are separated by assigning each of these alternatives a single context space for local reasoning and incremental disambiguation.

A Search Engine for Morphologically Complex Languages

Document retrieval on natural languages with a rich morphology -- particularly in terms of derivation and (single-word) composition -- suffers from serious performance degradation with the direct query-term-to-text-word matching paradigm that underlies the vast majority of current search engines. We propose an alternative approach in which morphologically complex word forms, which appear in the query as well as in the documents, are segmented into relevant subwords (such as stems, named entities, acronyms) and are subsequently submitted to the matching procedure. We evaluate our approach with the Alta Vista? Search Engine on a large medical document collection.

Semantic Interpretation of Medical Language - Quantitative Analysis and Qualitative Yield

We report on results from an empirical analysis of the semantic interpretation of medical free texts. Our approach to semantic interpretation is based on a lean collection of interpretation rules which are triggered by well-defined configurations in dependency graphs in order to compute a conceptual representation of the texts contents. We evaluate the accuracy of semantic interpretation for three types of syntactic dependency patterns, viz. genitives, auxiliary and modal verb complexes, and prepositional phrases. Besides quantitative considerations, we focus on the heuristic guidance, as provided by patterns underlying the semantic interpretation of prepositional phrases, for monitoring the quality of the medical domain knowledge base.

Coping with Different Types of Ambiguity Using a Uniform Context Handling Mechanism

We introduce a uniform context mechanism which is able to adequately represent and manage different forms of ambiguities as they occur in the course of text understanding. Different lexical, syntactic and semantic interpretations are clearly separated by assigning each alternative a single context space for local reasoning. The mechanism we propose directly supports the task of disambiguation at all levels of text analysis, since it also incorporates constraints from the discourse context, as text understanding proceeds.