next up previous
Next: Mv-Atom Completeness Up: No Title Previous: Partial deduction of rules

FORMALISATION OF A MANY-VALUED SPECIALISATION CALCULUS FOR RULE BASES

 

In this section we present a parametric family of many-valued calculi for rule specialisation. Each calculus is determined by a particular algebra of truth-values belonging to a parametric family of algebras that is described next.

Throughout this paper, an Algebra of truth-values tex2html_wrap_inline1366 tex2html_wrap_inline1368 will be a finite linearly ordered residuated lattice with a negation operation, that is:

  1. ( tex2html_wrap_inline1370 , tex2html_wrap_inline1372 ) is a chain of n elements: tex2html_wrap_inline1376 where 0 and 1 are the booleans False and True respectively.
  2. The negation operation tex2html_wrap_inline1386 is a unary operation defined as tex2html_wrap_inline1388 , the only definable order-preserving involutive mapping in tex2html_wrap_inline1390 , i.e. it holds:
  3. The conjunction operator T is a binary operation such that the following properties hold tex2html_wrap_inline1400 :
  4. The implication operator tex2html_wrap_inline1416 is defined by residuation with respect to T, i.e.

    displaymath1420

    Such an implication operator satisfies the following properties tex2html_wrap_inline1400 :

As it is easy to notice from the above definition, any of such truth-values algebras is completely determined as soon as the set of truth-values tex2html_wrap_inline1370 and the conjunction operator T are chosen. So, varying these two characteristics we generate a family of different multiple-valued logics. For instance, taking tex2html_wrap_inline1444 or tex2html_wrap_inline1446 we get the well-known Gödel's and tex2html_wrap1678 ukasiewicz's semantics (truth-tables) for finitely-valued logics [6, 7, 8, 9].

In the following we describe the language, the semantics and the deduction system (specialisation calculus) of a particular logic corresponding to a given algebra tex2html_wrap_inline1448 .

The propositional language tex2html_wrap_inline1450 is defined by:

Notation conventions. We shall commonly use:

tabular143


Further, a, b, ... will denote truth-values from tex2html_wrap_inline1370 while V, W, ... will denote intervals of truth-values. For simplicity we shall also write tex2html_wrap_inline1520 to denote the mv-formula tex2html_wrap_inline1522 .

The semantics is obviously determined by the connective operators of the truth-value algebra tex2html_wrap_inline1448 . Interpretations are defined by valuations tex2html_wrap_inline1526 mapping the (propositional) sentences to truth-values of tex2html_wrap_inline1370 fulfilling the following conditionsgif:

tabular151


Having truth-values explicit in the sentences enables us to define a classical satisfaction relation in spite of the models being multiple-valued assignments. The satisfaction relation between interpretations and mv-formulas is defined as tex2html_wrap_inline1542 iff tex2html_wrap_inline1544

and it is extended to a semantical entailment between sets of mv-formulas and mv-formulas as usual: tex2html_wrap_inline1546 iff tex2html_wrap_inline1542 for all tex2html_wrap_inline1526 such that tex2html_wrap_inline1552 , for all tex2html_wrap_inline1554 .

Taking into account the motivations introduced in the previous section, the deduction system we consider for rule specialization in our many-valued logical framework is the following one.

  definition160

Remark: In the above description of the Specialization inference rule we are assuming tex2html_wrap_inline1602 . It is understood that if n = 1 then the specialization rule of inference turns out into the following modus ponens inference rule: from (p, V) and tex2html_wrap_inline1608 infer tex2html_wrap_inline1610 .

The notion of proof inside Mv-SC is defined as usual.

definition184

It is easy to check that the above specialisation calculus is sound.

theorem189

Axioms A1 and A2 are trivially satisfied by any interpretation. Weakening, not-introduction, not-elimination and composition inference rules also trivially preserve truth. Let us check the truth preservation of Specialization rule for the simplest Modus Ponens case, i.e. when n = 1. We shall prove that

displaymath1632

By definition, tex2html_wrap_inline1634 is the minimal interval containing all the solutions of the following family of functional equations:

displaymath1636

for any tex2html_wrap_inline1638 , and tex2html_wrap_inline1640 . Suppose then that tex2html_wrap_inline1642 and let tex2html_wrap_inline1526 a model of (p, U) and of tex2html_wrap_inline1648 . Let tex2html_wrap_inline1650 and tex2html_wrap_inline1652 . Then tex2html_wrap_inline1638 and tex2html_wrap_inline1640 . By hypothesis, any solution for tex2html_wrap_inline1658 satisfying the equation tex2html_wrap_inline1660 must be in W, so tex2html_wrap_inline1526 is also a model of (q, W).

On the other hand, it is obvious that the logic Mv-SC is not complete. For instance, if we consider the two-valued case, i.e. tex2html_wrap_inline1668 , we have tex2html_wrap_inline1670 but tex2html_wrap_inline1672 . It is also the case that the language is not complete for literal deduction in general. For instance, we have tex2html_wrap_inline1674 but tex2html_wrap_inline1676 . However, it can be shown that the system is complete for mv-atom deduction provided we further restrict the language basically by not allowing negated literals in the language. This restricted mv-atom completeness can be seen as a many-valued counterpart of the completeness of classical modus ponens for atom deduction with propositional Horn clauses. This will be shown in the next subsection.




next up previous
Next: Mv-Atom Completeness Up: No Title Previous: Partial deduction of rules

Josep Puyol-Gruart
Wed Jun 11 15:38:47 MET DST 1997