190 likes | 293 Views
Measuring Ranks via the Complete Laws of Iterated Contraction. Wolfgang Spohn Dagstuhl Seminar, Aug. 26 - 30, 2007. Ranking Functions. is a negative ranking function for the algebra A over W iff is a function from A into R + = R { } such that for all A , B A :
E N D
Measuring Ranksvia the Complete Laws of Iterated Contraction Wolfgang Spohn Dagstuhl Seminar, Aug. 26 - 30, 2007
Ranking Functions is a negative ranking function for the algebra A over W iff is a function from A into R+ = R {} such that for all A, B A: (a) (A) ≥ 0, (W) = 0, and (A) = iff A = , (b) (A B) = min {(A), (B)}. (Thus, for simplicity we assume regularity.) For AAthe conditional rank of B given A is defined by (B | A) = (A B) – (A). Spohn, Dagstuhl, 26.-30.8.07
Belief Sets, Ax-Conditionalization A belief setK is a filter in A, i.e., a subset of A not containing and closed under intersection and the superset relation. The belief set K() associated with is defined as K() = {A A | (¬A) > 0}. For AA and xR, the Ax-conditionalizationAx of is defined by Ax(B) = min {(B | A), (B | ¬A) + x}. Spohn, Dagstuhl, 26.-30.8.07
Revisions and Contractions The single (AGM) revisioninduced by is defined as the function assigning to each A the belief set (A) = K(Ax) for some x > 0. For AA the contraction÷AofbyA is defined by ÷A = , if (¬A) = 0, and ÷A = A0, if (¬A) = 0. The single (AGM) contraction÷ induced by is defin-ed as the function assigning to each A the belief set ÷(A) = K(÷A). Spohn, Dagstuhl, 26.-30.8.07
Iterated Contractions of Ranking Functions The iterated contraction÷A1, …, Anof by A1, …, An is defined as = (…(÷A1)…) ÷An; this includes the iterated contraction ÷ = by the empty sequence . The iterated contraction÷induced by is defined as that function which assign to any finite sequence A1, …, An the belief set ÷A1, …, An = K(÷A1, …, An ). Spohn, Dagstuhl, 26.-30.8.07
Potential Iterated Contractions Let A denote the set of all finite (possibly empty) sequences of propositions from A. Then is a potential iterated contraction, a potential IC, for A iff is a function from A into the set of belief sets. A potential IC is an iterated ranking contraction, an IRC, for A iff there is a negative ranking function such that = . Spohn, Dagstuhl, 26.-30.8.07
Potential Disbelief Comparisons For any ranking function we have: (A) ≤ (B) iff ¬A¬A¬B. [This corresponds to Gärdenfors’ definition of an entrenchment relation.] Let be a potential IC for A. Then the potentialdisbelief comparisonassociated with is the binary relation on A such that for all A, BA: A ÷B iff ¬A¬A¬B. Strict comparison ÷ and equivalence ÷are defined analogously. Spohn, Dagstuhl, 26.-30.8.07
Reasons Expressed by Iterated Contractions A is a reason forB or positively relevant to B w.r.t. iff (¬B | A) > (¬B | ¬A) or (B | A) < (B | ¬A). [Analogous notions are defined analogously.] Then we have for any ranking function : A is not a reason against B, or non-negatively relevant to B, given C w.r.t. iff (ABC) – (A¬BC) ≤ (¬ABC) – (¬A¬BC), or iff neither (CA) ¬B nor (C¬A) B is a member of CA, C¬A, CB, C¬B. Spohn, Dagstuhl, 26.-30.8.07
Potential Disjoint Difference Comparisons Let be a potential IC for A. Then the potential disjoint difference comparison (potential DisDC) associated with is the relation defined for all quadruples of mu-tually disjoint propositions in A such that for all such propositions A, B, C, D: (A - B) (C - D) iff ¬A, ¬D ¬A ¬B, ¬C ¬D, ¬A ¬C, ¬B ¬D. Similarly strict comparison ÷ and equivalence ÷. is the potential DisDC associated with the IRC ; we have (A - B) (C - D) iff (A) – (B) ≤ (C) – (D). Spohn, Dagstuhl, 26.-30.8.07
Potential Doxastic Difference Comparisons We need to extend a potential DisDC from quadruples of mutually disjoint propositions in Ato arbitrary quadruples of propositions. Such an extension is called the potential doxastic difference comparison (potential DoxDC) associated with ÷ and also denoted by . Such an extension can be produced, e.g., by assuming that for each non-empty proposition A there are four mutually disjoint propositions A1, A2, A3, A4 with A ÷Ai(i = 1,2,3,4). (But one might think of other methods.) Spohn, Dagstuhl, 26.-30.8.07
Doxastic Difference Comparisons is a doxastic difference comparison (DoxDC) for A (with being the associated equivalence and the associated strict comparison) iff is a quarternary relation on A such that for all A, B, C, D, E, FA: (a) is a weak order on AA [weak order], (b) if (A - B) (C - D), then (D - C) (B - A) [sign reversal], (c) if (A - B) (D - E) and (B - C) (E - F), then (A - C) (D - F) [monotonicity], (d) if (A - W) (B - W), then (A - W) (AB - W) [law of disjunction]. Spohn, Dagstuhl, 26.-30.8.07
Supplementary Axioms The DoxDC is Archimedean iff, moreover, for any sequence A1, A2, … in A: • if A1, A2, … is a strictly bounded standard sequence, i.e., if for all i (A1 - A1) (A2 - A1) (Ai+1 - Ai) and if there is a DA – N such that for all i (Ai - A1) (D - W), then the sequence A1, A2, … is finite. Finally, the DoxDC is full iff for all A, B, C, DA: (f) if (A - A) (A - B) (C - D), then there exist C', D'A such that (A - B) (C' - D) (C - D'). Spohn, Dagstuhl, 26.-30.8.07
The Representation Theorem Let be a full Archimedean DoxDC for A. Then there is a regular negative ranking function for A such that for all A, B, C, DA: (A - B) (C - D) iff (A) – (B) ≤ (C) – (D). If ' is another negative ranking function with these properties, there is an x > 0 such that ' = x. [Cf. D.H. Krantz et al., Foundations of Measurement, vol I, Academic Press 1971, p151.] [Weak order, sign reversal, monotonicity, law of disjunction, and the Archimedean property are necessary axioms, fullness is a sufficient structural axiom.] Spohn, Dagstuhl, 26.-30.8.07
Preliminary Conclusion We have seen how potential iterated contractions in-duce potential disbelief comparisons and, via the in-tuitively accessible reason or relevance relation, po-tential doxastic difference comparisons (namely just the way iterated ranking contractions would do it). And now we have seen how such potential DoxDC measure ranking functions on a ratio scale, pro-vided they satisfy the mentioned six axioms. The only remaining question is: How must a potential iterated contraction behave so that the potential DoxDC generated by it satisfies the six axioms? Spohn, Dagstuhl, 26.-30.8.07
Iterated Contractions is an iteratedcontraction (IC) for A iff is a potential IC for A such that for all A, B, CA and SA: (IC1) the function A A is a single (AGM) contraction [single contraction], (IC2) if A, then A, S = S [strong vacuity], (IC3) if A B = , then A, B, S = B, A, S [restricted commutativity], (IC4) if AB and A BA, then AB, B, S = A, B, S [path independence], (IC5) if AC or A, BC and A ÷B, then A ÷CB, and if the inequality in the antecedent is strict, that of the consequent is strict, too [order preservation], (IC6) S is an IC [iterability]. Spohn, Dagstuhl, 26.-30.8.07
Some Explanations Single contraction(IC1) is clearly required. Strong vacuity(IC2) is stronger than AGM vacuity, but the intention is the same: vacuous contraction leaves not only the belief set, but even the belief state un-changed. Iterability(IC6) is again clearly required (and does not make the definition circular). (IC3) - (IC5) are the proper laws of iterated contractions. Order preservation(IC5) could, of course, be expressed entirely in terms of iterated contractions. It is equi-valent to the Darwiche-Pearl postulates. Thus, it is precisely (IC4) and (IC5) that go beyond the so far known and accepted axioms of iterated contraction. Spohn, Dagstuhl, 26.-30.8.07
Restricted Commutativity Ranking contractions do not always commute: A, BB, A if and only if A, BK(), (B | ¬A) = 0 or (A | ¬B) = 0, and (¬B | ¬A) < (¬B | A). The latter condition says that A is positively relevant to B. However, (IC3) claims restricted commutativity only under the condition that A logically excludes B, i.e., is unrevisably negatively relevant to B. That is, if two disbeliefs are logically incompatible, there can be no interaction between giving up these disbe-liefs, and hence it seems intuitively convincing that the order in which they are given up should not matter. Spohn, Dagstuhl, 26.-30.8.07
Path Independence Path independence (IC4) says this in terms of disbelief: Suppose you disbelieve two logically incompatible pro-positions, and you have to contract both of them. Then you can either contract one after the other. Or you can first contract their disjunction, and if you still dis-believe one of them, you then contract it as well. (IC4) says that both ways result in the same doxastic state. Spohn, Dagstuhl, 26.-30.8.07
The Completeness of Iterated Contractions Let be an IC for A. Then is called Archimedean iff the DoxDC induced by is Archimedean. And is called full iff the DoxDC induced by is full. Then we have the following completeness theorem:For any IC for A, the potential DoxDC induced by is a DoxDC for A. And any full Archimedean IC for A is an IRC for A, i.e., there is a ranking function for A with = . Moreover, for each ranking function ' with = ' there is an x > 0 such that ' = x. Spohn, Dagstuhl, 26.-30.8.07