A hypernymic proposition in a text is a statement such as “modafinil is a novel stimulant,” in which one concept is identified as a subtype of another. Rindflesh and Fiszman apply a heuristic method of identifying such propositions in biomedical text to enhance SemRep, a general semantic interpretation system that can be used in applications such as information retrieval, information extraction, and ontology engineering.

Three types of syntactic structures that potentially indicate hypernymic propositions were considered: verbs, appositives, and the modifier-head relationship in simple noun phrases. The method was implemented in SemSpec, which is a module of SemRep. Using the output of the underspecified parser in SemRep, and the domain knowledge in the unified medical language system (UMLS), SemSpec applies some semantic indicator rules to interpret hypernymic propositions in the text. Medline citations were used in the experiment. The results were evaluated, and sources of errors are discussed in detail.

The paper addresses an important relation in medical text. The examples shown in the development of the method, as well as in the detailed error analysis, provide interesting insight into the complexity of the problem.

The method described in the paper extends the original SemRep system by incorporating several simple rules. Some of the rules, and how they are applied, are not clearly explained, however. For example, in the description of appositive structures, it is mentioned that some syntactic and semantic conditions must be met to interpret a hypernymic proposition, but the conditions are not specified.