Information Retrieval (IR) is the field of computer science that is concerned with general methods of processing text so as to facilitate its subsequent search. Large bibliographic databases such as MedLine (and ISI's own Current Contents) use this technology to allow users to search these databases by keywords. In most cases, keywords (or search terms) are the individual words that are part of the abstract (or the full text of the article). The drawback of using individual words is the problem of synonyms: a user must specify all synonyms of a particular word to make sure the search does not miss articles of interest.

Controlled vocabularies are really thesauri, which contain CONCEPTS in a particular domain (e.g., medicine, and life sciences) and their synonyms. If one is able to scan the text of an article and match the text to concepts in the thesaurus, then the IDs of the concepts can be used for indexing, instead of words. This way, the user, having specified a single keyword for searching, can have the thesaurus expand the query by matching the keyword to a concept and then searching for all articles indexed by those concepts. In practice, concept-indexing is not foolproof. One problem is polysemy-words that have multiple meanings, and can match to multiple concepts. For example, "anesthesia" can be a procedure ancillary to surgery, or a loss of sensation in part of the body (e.g., following a nerve injury). Also, cryptic abbreviations, neologisms, and elisions, all of which occur in dictated medical text, can foil the process of concept recognition. Our work concluded that concept indexing by itself could not substitute for traditional word-indexing, but could be ancillary to the latter.

Prakash Nadkarni, MD
Associate Professor
Yale University School of medicine,
New Haven, CT