This paper attacks the problem of constructing a large phrasal thesaurus for an information retrieval system. Experiments by Fox and Wang have convinced us that it is important to identify semantic relationships between index terms rather than relation ships based only on statistical co-occurrence. Therefore we have turned to a machine readable dictionary as a source for semantic information linking phrases with other potential index terms. Defining formulae seem to provide valid clues to

more »

... onships in many dictionary definitions of phrases. Where no defining formulae are present we have relied on parsing using Sager's Linguistic String Parser.