- From: pat hayes <phayes@ai.uwf.edu>
- Date: Thu, 12 Dec 2002 10:41:34 -0600
- To: "Wagner, G.R." <G.R.Wagner@tm.tue.nl>
- Cc: Sandro Hawke <sandro@w3.org>, www-rdf-rules@w3.org, timbl@w3.org, pfps@research.bell-labs.com
> >>> Here one of the problems with N3. There is no indication in N3 of >whether >>>> log:implies is supposed to be material implication, a proof rule, or >some >>>> other logical connective (such as, for example, some sort of relevance >>>> implication). >>> >> In the sense of 'rule' I think you want (the one in which RuleML and >> rule engines are about rules) there is no fundamental difference. The >> use of "rule" rather than "implication" is a signal that you are not >> planning to use it in the context of a logically complete reasoner > >I think that we prefer the term "rule" over "implication" since >we consider a rule not in the narrow sense of mathematical logic, >but in the broader sense of a knowledge representation (KR) expression, >and KR is not a branch of classical/mathematical logic but rather >the other way around. My goodness, you do have a way of making large assertions that are highly debateable. I absolutely do not accept your premis here. >In real world domains, and in natural language, >we use the term "rule" for denoting many different things with very >different semantics. I take it then that 'rule' in Rule ML can have no single semantics, and that (very much in the XML spirit) , RuleML is not intended to convey a uniform semantic framework, but instead to be all things to all men? That is fine, let me emphasize, but we need to be clear on the point. >The same holds for the term "business rule", which >is a popular concept in enterprise modeling and information system >design. > >That's why, in RuleML, we use a basic distinction between at least >three different types of rules: derivation rules, integrity rules, >and reaction rules (a possible fourth kind of rule may be called >"deontic assignment rules" subsuming permission, prohibition, >duty assignment and empowerment rules). > >The discussion here (at rdf-rules) seems to be limited to certain >types of derivation rules. Yes, the discussion on this thread has been about rules in two senses: inference rule as in logic, and 'rule' as in logic programming, where it refers to a Horn clause. >But notice that there may be many different >types of such rules, and indeed there are many different derivation >rule systems out there: normal/extended logic programs with one/two >negations, intuitionistic logic programs, temporal/fuzzy/etc. logic >programs, OO rule systems, etc. Most of these systems have some >underlying non-classical logic, which does often not have any >(genuine) implication connective. I find it hard to generalize over such a range, but most logics that I am acquainted with have some kind of implication connective (other than logics which are deliberately restricted in some syntactic way so as to deliberately exclude it, obviously.) I note that all the examples you cite are systems which are considered to be a kind of programming language, and those, of course, typically do not have a normal assertional semantics. I think it might be worth emphasizing that all the SW languages so far contemplated are not programming languages in this sense and do have conventional logical semantics. >Rules are simply more fundamental >than implications as KR expressions. I profoundly disagree. In this open sense 'rules' can mean almost any specification of any process, so they are not fundamental at all. A Fortran interpreter could be implemented as 'rules', but it would hardly thereby acquire the status of being knowledge representation. The microcode in my G4 chip can be seen as a set of 'rules'. This is a deep issue in what might be called the philosophy of KR, as I expect you know. The SOAR architecture for example has been much argued over on just these grounds: is it a genuine theory of a cognitive architecture, or 'just' a rule interpreter? >This also holds, e.g., for SQL >which allows a type of derivation rule, called "views", but does not >support an implication (in the query language). > >> (or, possibly, that you are using it relative to a more >> 'computational' semantics, such as minimal-model semantics, relative >> to which it is complete; but then you ought to think hard about >> whether you still want to call it the material conditional, maybe.) > >Under the minimal model semantics, a rule does no longer have the >same intended models as the corresponding implication. This is easy >to see: consider the rule q :- ~p. It has only one intended (i.e. >minimal) model, which may be expressed by the set {q}, whereas the >corresponding material implication ~p -> q, which is equivalent to >q v p, has two intended/minimal models: {q} and {p}. Right, precisely my point. As a larger point, the fact that this is a 'rule' in both cases seems to me merely a shallow analogy; the alteration in the semantics changes the fundamental role as KR, since it changes the K that gets R'd. Pat -- --------------------------------------------------------------------- IHMC (850)434 8903 home 40 South Alcaniz St. (850)202 4416 office Pensacola (850)202 4440 fax FL 32501 (850)291 0667 cell phayes@ai.uwf.edu http://www.coginst.uwf.edu/~phayes s.pam@ai.uwf.edu for spam
Received on Thursday, 12 December 2002 11:41:46 UTC