Zalles, D. R., (1986). The California History-Social Sciences Framework: A Case Study in Implementation Process Analysis. Stanford University. June 1986.
Imagine you’re faced with the following problem: someone – no doubt someone with lots of money – has given you a specification for a system. One component of this system is to be a medium of representation: a language for expressing what needs to be expressed in order that the system can do whatever it’s supposed to do. The client is an expert about what needs to be done and hence, perhaps, about what needs to be expressed. Still, you may want to try for a certain generality – both to cover the client’s changes of mind and to accommodate, in one blow, a diversity of prospective clients. (The first contingency is as sure as death and taxes; the second – ALAS! – is not.) Notice that I’ve said nothing to delimit the purposes of the client, nothing to delimit the potential uses to which the system might be put or the environments in which it might have to operate. Is there any real chance of there being a single representational formalism that would be best along the weighted sum of all the reasonable dimensions of evaluation for all foreseeable – let alone possible – uses? Surely one has only to ask the question to get the right answer.
Now I’m going to ask you to imagine something else. Imagine you once learned a whole lot of logic and you’d like, for once in your life, to make some use of that knowledge. Some of the things you learned were that there are many logical languages, each of which can be given (typically a variety of) systematic semantic accounts. You have learned, that is, that one can state, in various ways, precisely what the sentences of such languages mean, or at least under what conditions they are true. (After looking at various AI representational formalisms, this point should seem like a MAJOR WIN.) Moreover, these semantic accounts all follow a pattern: they proceed by an induction tied directly to the inductive definition of the class of sentences of the language. So the semantic values (e.g., truth values) of complex sentences are determined by the semantic values of the simpler sentences out of which they are composed and the semantic values of simple sentence are themselves determined by the values of the subsentential expressions, (e.g., terms and predicates), out of which they are composed. Finally, how are the semantic values of the terms and predicates determined? Well, that depends on the kind of semantic account you want to give. For an excellent introduction to the standard – indeed, so called classical-model theoretic account and to the relation between such an account and the theory of relational databases, see [REIT841. To continue, you’ve learned that such languages, especially, those built on the language of first order quantification theory, are very powerfully expressive indeed. So, you reason as follows: within very wide bounds, whatever my clients want/need to say, I can say in (some extension, perhaps, of) some first order language. I can teach the client enough logic to be able to read and understand my formulation of his domain – remember the point about intelligible, systematic semantic accounts – and, after a few iterations (and assuming the client doesn’t change his mind about his requirements too often), I will have both formulated and analyzed the job’s representational requirements.