This section added by HassanAitKaci - on Dec. 19, 2005.
The criteria proposed above form a possible initial set that can be used toward forming a rule system ontology, although it follows a "top-down" approach. Namely, it is proposes a classification system for rules based on "semantic" issues. That is, a rule system classified according to it is characterized only by the nature of its semantics. While such an ontology is definitely useful for such a purpose, it may be not immediately needed for what is planned as an objective for the first phase of the RIF effort as exposed in the Charter.
Be that as it may, it bears to reason also to consider a dual bottom-up classification scheme. Viz., rather than starting with lofty semantic criteria such as listed above and working our way down toward specific systems, we could as well start with a set of representative systems and derive a classification scheme from specific criteria identifying actual operational features by working our way up the abstraction ladder.
This, for several members of this group, is likely to be a much easier and more immediately useful exercise, because:
- contrary to the top-down approach, members are generally most familiar with their own system; in other words, each ought to be able to describe their own system in simple terms (using elementary notions familiar to most);
- starting from concrete systems, we would have a greater chance to synthesize a set of criteria that are relevant to this WG's members' systems obtaining a classification by comparing and contrasting specific system descriptions;
- semantics aspects for the later phase on semantical issues could thus be eligible for consideration as per their relevance to the materialized bottom-up "ontology".
The heuristic for favoring a bottom-up approach here is the same as that pertaining to top-down vs. bottom-up computational derivations. Namely, a top-down approach (i.e., starting from a potiential large or even open-ended set of roots and trying to reach to leaves) is more likely to incur useless work than a bottom-up approach (i.e., going from leaves to relevant roots).
This also has the advantage of forcing everyone to present to the rest of the WG a core description of their rule language, be it concrete syntax and object model, as well as its basic requirements as for representation of rules. This would be the first step toward achieving what Harold Boley appropriately called "structure homomorphisms". Indeed, once each system is capable of describing in simple terms its core concepts, we can then start trying to map such concepts to more general abstract ones expressive for the larger set. Thus, a common format can naturally emerge delineating a "language space" (à la Landin's) in which all parishes could in principle map their own specific gospel irrespective of creed.
Working in semantics should then be a matter of projecting (so to speak) the bottom-up classification onto semantic schemes according to their desired operational behaviors. Doing so, the representation-bound structure homomorphisms may be completed into semantic homomorphisms (i.e., mapping of representations able to relate their interpretations as well as their concrete structures)