The idea that loosely defined simulation models of organizational behavior can yield more significant information than conventional precisely defined ones, has been explored. Natural language has been utilized as a medium for this purpose. This has allowed for the values of the variables to be linguistic rather than numerical, and for causal relations between the variables to be formulated verbally rather than mathematically. Such models have been called verbal models. A generative grammar is presented which restricts the set of allowed linguistic values and relations in a model specification. This makes it possible to formulate a semantical model based on fuzzy set theory of the words in the vocabulary. The semantical model can be used to calculate the dynamic behavior of verbal models. Thus it becomes possible to infer future behavior of a verbal model, given its linguistically stated initial state. This process was greatly facilitated by implementing the semantical model in an APLworkspace, thus making it possible to write linguistic values and relations directly on a terminal, using a syntax very close to that of natural language. The semantical model would then be automatically activated and respond with the linguistic values of output variables.
A simulation study is presented which shows that verbal models indeed may yield significant information based on rather general premises. This indicates that they may, under certain circumstances, be superior to corresponding conventional simulation models. It is generally concluded that the present approach towards modelling the behavior of complex organizations is not without interesting potentialities.