> Y\"bjbjWWHz=="#]JJJ$nnnnPR\n""rtttttt$#%-JpP
JJrnnJJrrJJr@YRUnn2rPreference Logic and Radical Interpretation
Kanger meets Davidson
Wlodek Rabinowicz
The primary purpose of this paper is to trace the intellectual effects of an encounter between two very different philosophers working in two seemingly unconnected areas. As will be seen, Stig Kangers meeting with Donald Davidson led the latter to modify his influential theory of radical interpretation and gave the former an inspiration for a rather striking paradox in preference logic. While the paradox can be dissolved, radical interpretation confronts some serious difficulties.
1. A paradox in preference logic
Stig Kangers paradox can be found in A Note on Preference Logic. This characteristically short two-page note was his contribution to a Festschrift for Thorild Dahlquist, published in Uppsala in March 1980. To introduce the paradox, let us suppose that }" is a preference relation on a set of states of affairs (propositions) that is assumed to be closed under Boolean operations. I.e., }" is the type of relation that is studied in preference logic. We may read A }" B either evaluatively, as State A is at least as good as state B , or descriptively, as State A is at least as preferred as state B . We use {" to refer to strict preference: A {" B is to be read as A is strictly preferred to (or, better than) B . Now, consider two conditions that one might want to impose on }":
Interpolation of Exclusive Disjunction (IED): For all states A, B,
if A {" B, then A }" (A(B) }" B.
Here, A(B stands for the symmetrical difference of A and B, respectively, i.e., for the Boolean analogue of exclusive disjunction ( either A or B, but not both ).
Four Levels (4L): There are some states A, B such that
A, B, -A and -B all occupy different levels in }".
That is, either A {" B or B {" A, and similarly for all the other pairwise comparisons between the four states. -A and -B stand for the complements of A and B, respectively. Strict preference {" is immediately definable in terms of weak preference }":
A {" B =df A }" B and it is not the case that B }" A.
Similarly, we can define indifference as follows:
A ( B =df A }" B and B }" A.
The two conditions above seem evident , says Kanger, but they cannot be upheld in tandem:
Kanger s impossibility result: There is no weak (= transitive & complete) ordering }" that satisfies both the Four Levels and the Interpolation of Exclusive Disjunction.
The assumption of completeness for }" is very strong. In many cases, we would like to allow for gaps in the preference ordering. If states A and B are significantly different from each other, we may well insist that neither A }" B nor B }" A are the case. Can we avoid to assume completeness and still prove that Kanger s two conditions are mutually incompatible?
Fortunately, we can. It is enough to assume that }" is transitive. In fact, as may be seen from the proof below, it would be enough just to assume that ( is a transitive relation. While the transitivity of indifference is relatively uncontroversial when indifference is interpreted evaluatively, the transitivity requirement is more problematic on the descriptive interpretation. As is well known, if preferences of a subject are determined by multidimensional pairwise comparisons, then intransitivities of indifference are to be expected. Still, these well-known phenomena might also be interpreted in a different way - as showing that a subjects preference ordering should be determined holistically rather than by a series of independently conducted pairwise comparisons. Transitivity will then function as an apriori constraint on any adequate holistic determination of preference.
Here, then, is the paradox in its final version:
Strengthened impossibility result: There is no transitive relation }" that satisfies both the Four Levels and the Interpolation of Exclusive Disjunction.
The proof that follows slightly simplifies Kanger s original version. By the transitivity of }", {" is transitive. Consequently, (4L) implies the existence of states A, B, -A and -B that are linearly ordered by strict preference, in one of the twenty-four possible ways. In Kangers somewhat sadistic version of the proof, we are supposed to go through all these possible cases, one by one, and for each case demonstrate that (IED) is violated. Here, we shall be more economical.
Proof: Assume (IED). We first show that, given (IED), the exclusive disjunction of unequally valued states must be equal in value to one of its disjuncts:
Lemma: If A {" B, then A ( (A(B) or B ( (A(B).
Proof of Lemma: Assume that A {" B. By (IED),
(i) A }" (A(B) }" B.
Thus, in particular, A }" (A(B). We consider two cases:
Case 1: (A(B) }" A. Then A ( (A(B), by the definition of (, and we are home.
Case 2: It is not the case that (A(B) }" A. Then A {" (A(B), by the definition of {". Thus, applying (IED) once again, this time to A and A(B,
(ii) A }" (A((A(B)) }" (A(B).
Since
A((A(B) = (A(-(A(B)) ( ((A(B)(-A) = (A(B) ( (B(-A) = B,
it follows from (ii) that B }" (A(B). Since we already know from (i) that (A(B) }" B, it follows that B ( (A(B), by the definition of (. And so we are home again.
We next show that:
Lemma + the transitivity of ( ( (4L) is violated.
Proof by reductio: Suppose (4L) holds for some A and B. Then we have either A {" B or B {" A. In each case, Lemma implies:
(i) A ( (A(B) or B ( (A(B).
We also have either -A {" -B, or -B {" -A. In each case, Lemma implies:
(ii) -A ( (-A(-B) or -B ( (-A(-B).
Since A(B = -A(-B, (ii) is logically equivalent to:
(iii) -A ( A(B or -B ( A(B.
(i) and (iii) imply that two of the four states A, B, -A, -B are on the same level as A(B. But given the transitivity of (, this means that these two states occupy the same level, contrary to the hypothesis. (
2. The paradox dissolved
How are we to deal with this paradox? To begin with, one might argue that the two conditions are not as evident as Kanger suggests. That (4L) is not uncontroversial will be seen below. But what is especially important is that the seemingly strong intuitive appeal of (IED) is deceptive. In the first place, the value of the exclusive disjunction may simply be incomparable with the values of the disjuncts. The evaluation of disjunctive states is notoriously contested. But even if we ignore this possibility, we might still want to reject (IED). If the disjuncts are mutually compatible with each other, the value of their exclusive distinction need not lie somewhere in-between the values of the disjuncts. After all, in the exclusive disjunction of A and B, the alternatives envisaged are not simply A and B but A-and-not-B and B-and-not-A. Thus, the value of such a disjunction should lie somewhere in-between the values of the mutually incompatible alternatives. These alternatives coincide with A and B only if A and B themselves are incompatible with each other. This suggests that (IED) is intuitive only when it is weakened:
Interpolation of the Exclusive Disjunction of Incompatibles (IEDI):
For all mutually incompatible states A and B, if A {" B, then A }" (A(B) }" B.
Unlike (IED), (IEDI) is fully compatible with the Four Levels, as will be seen below.
Does it mean, then, that (IED) is a condition without any appeal whatsoever? I do not think so. The Interpolation of Exclusive Disjunction does have some independent plausibility. Even when two unequally valued states are mutually compatible, it is not easy to see how their exclusive disjunction could be preferred or dispreferred to both of them at the same time. In order to finally dissolve the paradox, therefore, we should explain how it is possible that the Four Levels and the Interpolation of Exclusive Disjunction, taken separately, appear to be rather plausible (if not quite evident), even though, as we have seen, they cannot be upheld together. Such an explanation is not hard to come by. (4L) and (IED) are both plausible, because each of them is separately satisfied by some relatively plausible interpretations of }". Examples of such interpretations will be presented below.
Case 1: The Four Levels is satisfied while the Interpolation of Exclusive Disjunction is violated. Example: the expected utility interpretation.
On this interpretation, A }" B iff U(A) ( U(B), where U measures the expected value of a state. Following Jeffrey (1983), the expected utility of a state is here taken to be the weighted sum of the expected utilities of its different possible realisations, with weights being the conditional probabilities of the realisations in question. Thus, if P is the underlying probability function, we assume that the following holds for all states A and B:
(EU) If P(A(B) = 0 and P(A(B) > 0,
U(A(B) = P(A/A(B)U(A) + P(B/A(B)U(B)
It is easy to see that (EU) immediately implies (IEDI): If A and B are incompatible, then A(B = A(B. Consequently, U(A(B) = U(A(B) equals the weighted sum of the utilities of the disjuncts. Thus, the utility of A(B must lie somewhere between the utilities of A and B.
(IED), on the other hand, may well be violated by this expected utility interpretation, as is shown by the following example. Let C, D and E be three equiprobable states that are mutually exclusive and jointly exhaustive. Thus, P(C) = P(D) = P(E) = 1/3, and P(C(D) = P(D(E) = P(C(E) = 0. Let the U-values for C, D and E be 0, 2 and 3, respectively. Consider A = C(D and B = C(E. Note that A(B = D(E. Using (EU), we can calculate the utilities of A, B and A(B:
U(A) = P(C/A)U(C) + P(D/A)U(D) = 1/2 ( 0 + 1/2 ( 2 = 1.
Similarly, U(B) = 1/2 ( 0 + 1/2 ( 3 = 1,5,
while U(A(B) = 1/2 ( 2 + 1/2 ( 3 = 2,5.
Thus (IED) is violated but the Four Levels holds; states A, B and their both complements have all different utilities:
U(-A) = U(E) = 3, U(-B) = U(D) = 2.
Case 2: The Interpolation of Exclusive Disjunction is satisfied while the 4-Level Condition is violated. Examples: extremal preference (maximin or maximax), or ceteris paribus preference.
Suppose that preferences between states are derived from preferences between possible worlds. The preference between states is defined as the preference between selected world-representatives of the states in question. To be more precise, assume a weak preference ordering " on the set U of possible worlds. We identify states with sets of worlds: a state is identified with the subset of U that consists of all worlds in which that state obtains. Let c be a choice function from states to worlds, such that c(A) is defined iff A ( (, and for each such non-empty A, c(A) ( A. Then, define A }" B as c(A) " c(B). (The relation }" does not obtain between A or B, if either of them is empty.)
Suppose that the choice function c is based on some underlying linear ordering " of U: c(A) picks out that world in A that comes highest in ". (If such a world is to exist, for any non-empty A, whether finite or not, it is not enough that " is linear; every subset of U, and not just the finite ones, must contain the maximal "-element. In other words, U must be well-ordered by the converse of ".)
Possible interpretations of ":
(i) " is a linearisation of " on U (with equally good worlds coming in an arbitrary order). That is, " is any linear ordering such that for all A and B, if A comes above B in ", then A " B. Then }" is maximax.
(ii) " is a linearisation of the converse of " (with equally good worlds again coming in an arbitrary order). Then }" is maximin.
(iii) " is the ordering of worlds with respect to their similarity to the actual world (= the status quo, the reference world). For any worlds x and y, x " y iff x is more similar to the actual world than y is. Then A }" B stands for: Ceteris paribus, it would be better that A than that B. In other words: It would be better that A rather than that B, other things being equal (to what they actually are).
It can now be shown that if }" is based on a preference relation " between world-representatives of states, selected by means of a choice function c that is based on a linear ordering " of U, then (IED) is satisfied and (4L) is violated.
Proof: Clearly, since " is transitive on worlds, the derived ordering }" is transitive on states. Therefore, in view of the impossibility result, it is enough to prove (IED). Suppose that A {" B, i.e., c(A) comes higher than c(B) in the "-ordering. Let x be that element of {c(A), c(B)} that comes highest in the "-ordering. Clearly, x ( A(B, since otherwise we would have c(A) = c(B) and so it would not be the case that c(A) comes higher than c(B) in ordering ". Since x ( A(B, it follows that x ( A(B. Consequently, c(A(B) = x. But then c(A(B) = c(A) or c(A(B) = c(B). In each case, A }" (A(B) }" B.
3. What s this got to do with Davidson? Problem of Radical Interpretation
Here is Davidson s problem of radical interpretation: What does it take to understand another person, more or less from scratch? Davidsons goal is to elicit what a subject means by what he says, what he believes and what he prefers. The elicitation should be based on some relatively unproblematic set of empirical data.
While in his earlier work, collected in Truth and Interpretation (Davidson 1986), the goal of radical interpretation was just to disentangle meaning from belief, the new project involves an additional task: determination of the subjects desires. In this way, understanding a person allows us to understand not just what he thinks but also what he does, given that what he does is a function of what he believes and desires.
This project of a three-way elicitation - of meaning, belief and preference - has been pursued by Davidson over two decades in a series of publications starting with his 1978 lecture Towards A Unified Theory of Meaning and Action (published two years later, see Davidson, 1980) and continuing with such papers as Expressing Evaluations (Davidson, 1982), A New Basis for Decision Theory (Davidson, 1985), The Structure and Content of Truth (Davidson 1990), and The Folly of Trying to Define Truth (Davidson 1996).
In order to understand Davidsons project, it is instructive to compare it with Quines well-known conception of radical translation (cf. Quine, 1960). Here is how Davidson himself describes Quines undertaking:
Noting that, while there is no direct way to observe what speakers mean, all the evidence required to implement communication must be publicly available, Quine surveys the relevant available evidence, and asks how it could be used to elicit meanings. [...] For Quine, the key observables are acts of assent and dissent, as caused by events within the ambit of the speaker. From such acts it is possible to infer that the speaker is caused by certain kinds of events to hold a sentence true. [Added in footnote: The step from observed assents to the inferred attitude of holding true is not, I think, explicit, in Quine.] Just here a basic challenge arises. A speaker holds a sentence true as a result of two considerations: what he takes the sentence to mean, and what he believes to be the case. [...] How can the roles of these two explanatory factors be distinguished and extracted from the evidence? [...] Quines key idea is that the correct interpretation of an agent by another cannot intelligibly admit certain kinds and degrees of difference between interpreter and interpreted with respect to belief. As a constraint on interpretation, this is often called by the name Neil Wilson gave it [Wilson (1959)], the principle of charity. (Davidson, 1990, pp. 318f)
Quines problem is then that different hypotheses about the speakers meaning can be defended by adjustments in the hypotheses about his beliefs. Quines solution of this problem is that the speakers beliefs must obey constraints imposed by the principle of charity that requires a far-reaching consensus in beliefs between the speaker and us. Given these constraints on beliefs, the available data about the speakers assents to and dissents from sentences (including the external circumstances of such assents and dissents), we can fix the speakers meaning (up to remaining indeterminacies).
The differences between Quine and Davidson are at least sixfold:
(i) While Quine pursues a project of (radical) translation, Davidson is interested in interpretation.
While Quine is concerned with the conditions of successful translation from a speakers language into the interpreters, I emphasise that the speaker needs to know of the semantics of the speakers language, that is, what is conveyed by the T-sentences entailed by a theory of truth [for the speakers language]. (Davidson, 1990, p. 319).
(ii) Quine takes the circumstances that prompt assents to (or dissents from) observation sentences to be patterns of stimulation of nerve endings rather than external objects or events. These proximal stimuli are taken to determine the empirical content of such sentences in the speakers language, which implies that their translation is relatively unproblematic: they are translated into the sentences in the interpreters language that are correlated with the similar stimulation patterns. (Relatively unproblematic, that is. For some second thoughts on this issue, cf. Quine, 1990, sections 15 and 16.) Davidsons view on this issue is more in line with common sense:
... interpretation depends (in the simplest and most basic situations) on the external objects and events salient to both speaker and interpreter, the very objects and events the speakers words are then taken by the interpreter to have as subject matter. (Davidson, 1990, p. 321)
(iii) For Quine, charity is much less inevitable than it is for Davidson. While we normally assume that the speaker shares our beliefs to a large extent, this assumption of consensus is not inescapable: it might be overturned by considerations of simplicity.
The linguist assumes that the natives attitudes and ways of thinking are like his own, up to the point where there is contrary evidence. He accordingly imposes his own ontology and linguistic patterns on the native wherever compatible with the natives speech and other behavior, unless a contrary course offers striking simplifications. (Quine, 1990, pp. 48f., my italics.)
(iv) Furthermore, Quines version of charity requires us to assign to the speaker not so much the beliefs we actually hold but rather the beliefs we imagine we would have held in the speakers shoes.
[The translator] will favor translations that ascribe beliefs to the native that stand to reason or are consonant with the natives observed form of life. [...] Practical psychology is what sustains our radical translator all along the way, the method of his psychology is empathy: he imagines himself in the natives situation as best he can. (Quine, 1990, p. 46.)
Davidson appears to be less prepared to make such allowances for expected divergences and, when he does make them, he takes empathy to uncover divergences in needs and valuations rather than divergences in beliefs. The interpreter is counselled
to interpret agents he would understand as having, in important respects, beliefs that are mostly true and needs and values the interpreter shares or imagines himself sharing if he had the history of the agent and were in compatible circumstances. (Davidson, 1985, p. 93)
(v) While Quines project is strictly behaviourist - the basic data concern an outward assent and dissent behaviour of the speaker - Davidson is fully prepared to allow as empirical data the speakers mental attitudes to sentences, such as holding a sentence to be true. To be sure, such a mental attitude is manifested in an assent behaviour, but the attitude and the behaviour are not the same thing.
(vi) However, Davidson thinks that knowing which sentences are assented to or held true by a subject under which external circumstances would be insufficient as the data basis for elicitation. To fix the meaning, and especially to fix the meaning of theoretical sentences, we need to determine evidentiary relations between the different sentences in the speakers language: what he counts as evidence for what, what sorts of evidence would make him consider a given sentence as more probable. We need probability assignments rather than simple yes-or-no attitudes of holding true. As Davidson illustrates this point:
... a sentence [tentatively] interpreted as meaning that there is a patter on the roof, if held true (given a high probability), ought to increase the probability of the sentence [tentatively] interpreted as meaning that it is raining. In this way, by marking what the speaker takes as evidence for the truth of a sentence, it is possible to interpret sentences and words of an increasingly abstract and theoretical nature. (Davidson, 1982, p. 15)
But how are we to access the subjects probabilities? Here it is time to introduce another important source of inspiration for Davidson: Frank Ramseys program for decision theory. The task Ramsey put himself was to simultaneously determine a subjects probability assignments (quantitative degrees of belief) and his quantitative degrees of preference for different outcomes. (Cf. his paper Truth and Probability, posthumously published in Ramsey, 1931) The data for this elicitation were to be the subjects ordinal preferences over various gambles, as revealed by his (actual and hypothetical) choices. Being prepared to gamble on an event shows something about an agents probability for the event in question and about his valuation of the possible outcomes.
As Quine, Ramsey encountered the problem of compensatory adjustments: changes in the hypotheses about the subjects probabilities could be compensated for by adjustments in the hypotheses about his degrees of preference for various outcomes. He solved that problem by imposing a number of constraints on the subjects preferences over gambles (including such constraints as transitivity, completeness, etc.). Given the constraints, the preferences could be seen as going by the expected utility of gambles and the constraints made it possible to uniquely determine the probability assignments that underlied these expected utilities. When event probabilities were determined, it was then easy to determine the degrees of preference for outcomes (up to the positive linear transformations).
Davidson accepts Ramseys idea of taking ordinal preferences as basic data, but rejects using preferences over gambles as base. As he points out, when the subject is given a choice between gambles, each gamble is presented to a subject as a proposition of the form:
A if C, B if not C.
Thus, preference over gambles is an intensional attitude - such preferences are attitudes towards propositional objects. As such, they are unfit to function as empirical data for radical interpretation.
The same objection immediately applies to Ethan Bolkers and Richard Jeffreys approach to expected utility (cf. Jeffrey, 1983): their base for the elicitation of probabilities and degrees of preference are a subjects ordinal preferences over propositions:
that A is more desirable than that B.
Again, such a base is unfit to function as an empirical point of departure for radical interpretation.
But from Davidsons point of view, Bolker-Jeffrey constraints on preferences over propositions have one important advantage as compared with Ramseys: they are formulated in such a way as to allow the objects of preference to be any set of entities whatsoever, as long as the Boolean operations are definable on that set. In view of the close connection between Boolean operations and truth-logical sentential connectives, it thus becomes possible to replace propositions as objects of preference with linguistic entities.
This leads to Davidsons own proposal. Provided we can identify the truth-functional (Boolean) connectives in a subjects language, why not replace Bolker-Jeffrey preferences over propositions with preferences over (otherwise) uninterpreted sentences as basic data for radical interpretation?
The subject prefers the truth of p to the truth of q (in symbols, p }" q).
This is a mental attitude towards linguistic objects. As such, it still is an intentional attitude, but no longer an intensional one. The objects of the attitude are not propositions, but sentences that obey well-defined identity conditions. As Davidson puts it:
... the objective was not to avoid intentional states; it was to avoid individuative intentional states, intensional states, states with (as one says) a propositional object. A preference for the truth of one sentence over another is an extensional relation that relates an agent and two sentences (and a time). Because it can be detected without knowing what the sentences mean, a theory of interpretation based on it can hope to make the crucial step from the nonpropositional to the propositional. (Davidson 1990, p. 323)
Constraints on the preferences among sentences are the same as Bolker-Jeffrey conditions on preferences among propositions, with one extra constraint added:
If p ( q is a truth-functional tautology, then p ( q.
For the subject, his sentences are meaningful. But their meaning is originally not given to the interpreter. To replicate Bolker-Jeffrey project of elicitation,
(i) we first need to identify the truth-functional connectives in the subjects language.
Otherwise, simple sentences may be treated as black boxes. Then, using Bolker-Jeffrey methods,
(ii) we elicit the subjects degrees of preference and probabilities for sentences.
Unlike in Ramseys case and in the case of other standard expected utility theories, utilities in Bolker-Jeffrey approach are elicitable only up to so-called fractional linear transformations (with four free parameters instead of the usual two), and the probability assignment is only determined within certain limits. But this remaining indeterminacy is a cost that Davidson is quite willing to pay.
In the last step, given the subjects probabilities for uninterpreted sentences (plus our knowledge of the external circumstances), and making use of the Principle of Charity,
(iii) we determine the subjects intensional attitudes: his degrees of (intensional) belief and the meaning of the sentences in his language.
With the meaning in place, the subjects degrees of preference for sentences allow us to determine the third factor in the subjects mental life, his degrees of intensional preference. Which completes the task of radical interpretation.
4. In search of truth-functional connectives
But how are we to achieve task (i)? How can we identify truth-functional connectives, in an otherwise uninterpreted language, just on the basis of its users preferences over sentences? This presupposes, of course, that we have already managed to identify the sentences of the language in question, and that we have found some way to test, for each pair of such sentences, whether the subject prefers one sentence to the other, or vice versa, or is indifferent between them. Davidson does not discuss how this can be done. Nor does he discuss the obvious objection that Bolker-Jeffrey constraints on preferences may be much too exacting for any subject to be able to obey them in full. But let us suppose that we can ignore these problems. What then? How are the truth-functional connectives to be identified?
The first try was made by Davidson in the summer of 1978, at the Wittgenstein seminar in Kirchberg. In the published version of this lecture (Davidson, 1980), he suggests that we first should find the connective ? that stands for neither ... nor in the subjects language. We can do it, he claims, by examining the subjects preferences for sentences. Since, as is well known, all the other truth-functional connectives are definable in terms of that one, the rest is easy. For example:
A connective n stands for negation iff for all sentences p, np ( p?p.
And then,
A connective c stands for conjunction, iff for all sentences p and q, pcq ( np?nq.
And so on.
But how can we determine that ? is what we have been looking for? Well, Davidson suggests the following test: ? stands for neither... nor iff:
For all sentences p and q, p {" q iff q?q {" p?p.
Note that, if the hypothesised interpretation of ? is correct, q?q and p?p stand for (q and (p, respectively. Thus, what Davidson relies on here is the preference principle according to which p is preferred to q iff (q is preferred to (p.
As it stands, this proposal is unsatisfactory, for at least two reasons, one of which is partly recognised by Davidson himself:
(1) The principle p {" q iff (q {" (p does not generally hold for the intended expected utility interpretation of the preference relation. On this interpretation, the equivalence only holds for probabilistically independent p and q. This means, that we must devise a way of telling, from preferences among sentences, that two sentences are independent (ibid., p. 11). While Davidson thinks it likely that such a test for probabilistic independence can be developed, he does not provide it in the paper.
In fact, it can be shown that the needed test may be impossible to obtain. As we have seen, in Bolker-Jeffrey theory, the subjects probability assignments are not uniquely determinable. In particular, as Levi (forthcoming) has proved, if p and q are probabilistically independent on one of the probability assignments that are compatible with the evidence, then there will be infinitely many other such admissible assignments on which p and q are not independent. This means that the test of probabilistic independence is unavailable on the Bolker-Jeffrey approach!
Levi (ibid.) suggests that such a test could still be available to Davidson, who takes the interpreter to rely on more information than the mere preference data about the subject. The interpreter can access his own beliefs and values and thus might be able to reduce the indeterminacies in his interpretation by invoking the Principle of Charity. I am not convinced, however, that this solution can be of help at the present early stage of interpretation. Charity comes into play at stage (iii), when the interpreter tries to determine the subjects intensional attitudes and in this process is supposed to maximise the consensus between himself and the subject. At present, we are still at the very beginning of the interpretation process, when the interpreter has not yet managed to fully determine the subjects extensional attitudes (= attitudes to sentences). He still has to find out the subjects degrees of preference and probability for uninterpreted sentences. At this preliminary stage, there is no room for charity.
(2) Still, I may be too pessimistic. But even if a test of probabilistic independence could be developed, Davidson would not yet be in the clear. There are other truth-functional connectives that ? might stand for and p?p will still correspond to negation. An example is the Sheffer stroke (= not both). So even if we found ? such that, for all probabilistically independent p and q, p {" q iff q?q {" p?p, we would still be unable to tell whether ? stands for neither-nor or for not both.
Davidson made a second try, which he never published, in his lecture in Oslo, in the fall of 1979. This time, instead of neither-nor, he proposed to start with a search for exclusive disjunction in the subjects laguage. In fact, the latter looks like a right connective to start with as far as preference data are concerned. Note that Ramseys gambles may be understood as such disjunctions: If it is read truth-functionally, a gamble description A if C, B if not C is equivalent with the exclusive disjunction Either (C and A) or (not-C and B). Admittedly, this truth-functional reading of if in the gamble description is quite unsatisfactory: it ignores the subjunctive connection that is supposed to obtain between the gamble event (C) and the prizes (A and B). What is even more important, unlike neither-nor, exclusive disjunction does not suffice for the definition of all the remaining truth-functional connectives. However, if the exclusive disjunction could be identified in the subjects language, we would at least make some progress in the process of interpretation.
Davidson took it for granted that exclusive disjunction satisfies the interpolation principle (IED), possibly because he did not clearly distinguish it from the (relatively) innocuous condition (IEDI). Therefore, he suggested that connective ? stands for either ... or iff it satisfies the interpolation condition:
For all p and q, if p {" q, then p }" (p?q) }" q.
It is here that Kanger comes in: In Oslo 1979, Kanger showed Davidson that, on the intended expected utility interpretation of }", (IED) does not generally hold. Kanger s paradox from 1980 is, as far as I can see, an indirect result of this exchange, even though Davidsons name is never mentioned in that paper. As Davidson conjectures (personal communication), this polite silence might have been partly motivated by Kangers feeling that, whatever might be said against it, (IED) is still an internally plausible principle, worth considering in its own right.
Kanger had also been helpful in connection with Davidsons last and final try to solve the connective problem. This time, Davidson proposed to start with the Sheffer stroke. He presented this idea in his Hgerstrm lectures in Uppsala, with Kanger as a host, in the Spring of 1980, and he kept to it in all his later work on the subject (cf. Davidson 1982, 1985, 1990 and 1996).
Thus, the idea is to first identify ? that stands for not both and then to identify all the remaining truth-functional connectives in terms of ?. The latter task is easy since all such connectives are definable by means of the Sheffer stroke. (The identification method to be used is thus the same as the one sketched above in connection with neither - nor.) But what about the Sheffer stroke itself?
The expected utility interpretation implies that, for all p,
(a) if p {" T, then T }" (p,
where T is an arbitrary tautology. We also have,
(b) if T {" p, then (p }" T.
We note first that if ? stands for the Sheffer stroke, then
p?p stands for (p, while q?(q?q) stands for a tautology: (q (((q.
Consequently, Davidson asserts, ? stands for the Sheffer stroke iff,
(i) for all p and q, (() if p {" q?(q?q), then q?(q?q) }" p?p,
(() if q?(q?q) {" p, then p?p }" q?(q?q);
and
(ii) for some p and q, it is not the case that p?p ( q?q.
According to Davidson, no truth-functional connective apart from the Sheffer stroke satisfies both (i) and (ii).
(I have somewhat simplified Davidsons condition (i). Instead of q?(q?q), he makes use of a more complicated expression in which he substitutes q?r for q in q?(q?q). This is unnecessary, since the simpler q?(q?q) already stands for a tautology if ? is the Sheffer stroke.)
Condition (ii) is first added in Davidson (1990), probably in order to avoid the obvious objection that (i) by itself would be insufficient to pick out the Sheffer stroke. It is easy to check that (i) is satisfied even if ? stands for the material implication. It is also satisfied if ? stands for the tautological connective that for any two sentences yields truth as value. In both cases, (i) translates into the trivially valid:
(() if p {" T, then T }" T, and (() if T {" p, then T }" T.
In fact, pace Davidson, conditions (i) and (ii) are still insufficient to pick out the Sheffer stroke. Given just (i) and (ii), ? might stand for the contradictory connective that for any two sentences yields falsity as value. Since on the intended interpretation, the expected utility of contradiction is undefined, the contradictory sentences do not belong to the field of }". Consequently, if ? were such a connective, (i) and (ii) would be vacuously satisfied.
As a matter of fact, given just (i) and (ii), ? might also stand for neither... nor. Since on this interpretation q?(q?q) is again contradictory, (i) would be vacuously satisfied and (ii) would be true.
To exclude these remaining interpretive possibilities, we should add an extra condition, such as:
(iii) for some p and q, p }" q?(q?q).
But are we now in the clear? Not quite. Some objections, not mentioned by Davidson, still remain:
Objection 1: What if ? is not a truth-functional connective at all? For example: What if it stands for it is impossible that both? Or it is improbable that both?
The Sheffer stroke is the only truth-functional connective that satisfies conditions (1) - (iii). This much can be shown. But the subjects language may contain several non-truth-functional connectives that also satisfy these conditions. This possibility has not been excluded by Davidson; it is a potentiality he never even considers!
The non-truth-functional interpretation could be excluded if Davidsons interpreter had access to some additional information about the subjects sentential attitudes, apart from the evidence about the subjects preference ordering on sentences. Suppose the interpreter also knows which sentences are held to be true and which are held to be false by the subject. (While it is a requirement of rationality that the two sets of sentences be disjoint, they will normally not be jointly exhaustive.) Then the task of identifying the truth-functional connectives is easy. Thus, ? stands for the Sheffer stroke iff:
(() Whenever the subject holds p or q to be false, he holds p?q to be true;
(() Whenever the subject holds p?q to be true, he does not hold both p and q to
be true;
(() The subject holds p?q to be false iff he holds both p and q to be true.
But, as we have seen, Davidson wants to make do without the information about which sentences are held to be true/false by the subject. In this respect, Davidsons approach to radical interpretation has changed in recent years, as compared with his earlier work collected in Truth and Interpretation. The reasons for this austerity are not quite clear: After all, the attitudes of holding true/false, if directed to sentences, are just as extensional as the attitudes of preferring one sentence being true rather than another. Possibly, Davidsons motivation was aesthetic: it is clearly more elegant to use only one kind of data (preference data) instead of two. But just as possibly, under Jeffreys influence, he might have come to suspect that the simple yes-or-no attitudes of holding true/false are not easily ascribable to a person. In particular, their relationship to a persons probability assignments is notoriously unclear. As is well known, holding true cannot be identified with assigning high probability: the former, unlike the latter, is supposed to be closed under conjunction.
Objection 2: What is there to guarantee that the preference data reflect the subject s expected utility comparisons?
As we have seen in the first section, there are other plausible interpretations of }". To be sure, if the interpreter knew that the subject s preferences satisfied Bolker-Jeffrey constraints, then he would know that they can be interpreted as in expected utility terms. But in order to know that they satisfy the constraints, he must first identify the truth-functional connectives in terms of which these constraints are formulated. At the same time, the interpreters procedure for the identification of the connectives assumes that the subjects preferences do obey the relevant constraints, which the interpreter cannot yet know at that stage.
I do not think that the two objections mentioned above are unanswerable. Given Davidsons general holistic approach to theorising, they might both be met by a standard recipe: Why not try it out and see how it works? Thus, suppose we find a connective ? in the subjects language that obeys conditions (i) - (iii). We can then start the process of interpretation with the hypothesis that ? does indeed stand for the Sheffer stroke. If given this hypothesis, the subjects preferences over sentences do turn out to satisfy the Bolker-Jeffrey constraints, then we may conclude that ? was the Sheffer stroke and that the preference data in fact reflected the subjects expected utility comparisons. On the other hand, if the hypothesis turns out to be unworkable, we might try to look for some other candidate for the Sheffer stroke, and eventually, if all such attempts would fail, for some wholly different set of preference data. So our two objections can be met. But it is somewhat surprising that Davidson never even considers them in his essays.
5. Problem of limited language resources
A more serious problem arises in connection with the second stage of interpretation - the stage at which the interpreter, after having identified the truth-functional connectives, moves on to the task of elicitation of the subjects degrees of preference and probabilities for sentences. At this stage, the interpreter is supposed to make use of the Jeffrey-Bolker methods of elicitation, as described in Jeffrey (1983). But are these methods applicable for Davidsons purposes?
Unlike Davidson, Bolker and Jeffrey start from a preference ordering on a Boolean algebra of propositions. Davidson seems to think that sentences would do just as well, provided the language contains truth-functional connectives. To be sure, truth-functional connectives are not quite Boolean operations, but there is a close correspondence between the two. For each sentence p, we can determine its equivalence class [p] consisting of all the sentences that are truth-functionally equivalent to p. Then, in terms of the truth-functional connectives, it is easy to define the Boolean operations on such equivalence classes. For example, if A = [p] and B = [q] are such equivalence classes, then the complement of A = [(p] and the intersection of A and B = [p(q]. We get in this way the so-called Lindenbaum algebra, which is an example of a Boolean algebra.
However, we are not home yet. Bolker-Jeffrey representation theorem, on which Davidson relies, rests on a very strong presupposition. It presupposes that the Boolean algebra on which the preference ordering is defined is both atomless and complete. To explain these notions, let us first define the notion of implication: we shall say that a state A implies a state B iff A = A(B.
An algebra is atomless, if for any non-zero state A (i.e., A ( 0 = A(-A), A contains a non-zero state B such that B implies A but is not implied by it (we say that such a state is strictly smaller than A).
If states of the algebra are propositions, then the algebra is atomless if for every consistent proposition there is a stronger consistent proposition.
If X is a set of states, a lower (upper) bound of X is any state that implies (is implied by) every member of X. The infimum inf(X) of X is the greatest lower bound of X, i.e., every other lower bound of X is strictly smaller than inf(X). The supremum sup(X) of X is the least upper bound of X, i.e., it is strictly smaller than every other upper bound of X.
An algebra is complete if for any set X of states, whether finite or not, it contains the infimum and the supremum of X.
As a matter of fact, if a Boolean algebra contains the infima for all its subsets, it will of necessity contain the suprema as well. It can be shown that
sup(X) = -inf({-A: A ( X}).
Consequently, if the states of a Boolean algebra are propositions, then the algebra is complete if for every infinite set of propositions, it contains the infimum, i.e., the proposition that corresponds to the conjunction of the propositions in that set.
Now, what about the Lindenbaum algebra? Does it satisfy the two requirements of Bolker-Jeffrey representation theorem? Roughly, this would mean (i) that for any (truth-functionally) consistent sentence p, the language contains another consistent sentence that is stronger than p (atomlessness), and (ii) that for any set of sentences, whether finite or not, the language contains a sentence that is exactly as strong as the conjunction of the sentences in the set (completeness). It is easy to see that these two requirements pull in opposite directions. The number of sentences in any language is countable. The same applies, therefore, to the number of equivalence classes in the corresponding Lindenbaum algebra. If that number is finite, the language will be complete but it obviously will not be atomless. And if the number of such classes is denumerable (= countably infinite), the algebra will be atomless but it will not be complete. In general, no countable Boolean algebra can be both atomless and complete. For a sketch of the proof, due to Sten Lindstrm, see Appendix.
Thus, the Bolker-Jeffrey theorem presupposes a non-denumerable algebra of propositions. The number of propositions needed for the representation theorem exceeds by far the sentential resources of any language that we might encounter. This means that Davidsons project of elicitation, with its strong dependence on Bolker-Jeffrey elicitation methods, is doomed from the start. A move from propositions to sentences is necessary if the elicitation is to build on extensional data. But it is precisely this move that makes the Bolker-Jeffrey elicitation impossible.
One might try to resist this objection. Even if full elicitation is not possible, a partial one may be all that is needed. If one could show that the Lindenbaum algebra for the language that we are trying to interpret is rich enough to allow for an approximate elicitation of the language users preferences and beliefs, Davidsons project would still be viable. After all, Davidson is quite prepared to accept that interpretation cannot resolve all indeterminacies. However, to show that approximate elicitation is possible, one would have to prove that the Bolker-Jeffrey theorem would still approximately hold for a sufficiently rich algebra even if that algebra falls short of Bolker-Jeffrey requirements. Proving this result is certainly not a trivial exercise. It is surprising that Davidson has never even touched upon this problem in his long series of papers on radical interpretation.
Appendix
Theorem: No countable Boolean algebra is both atomless and complete.
Proof: Since all countable atomless Boolean algebras are isomorphic (cf. Bell and Slomson, 1969, Corollary 7.7 in ch. 1, p.30), and since Lindenbaum algebras are Boolean and countable, it is enough to prove that an infinite Lindenbaum algebra is atomless and incomplete. Then it will follow, by isomorphy, that every countable atomless Boolean algebra is incomplete.
We first prove that any infinite Lindenbaum algebra LA is atomless. If LA is infinite, its underlying language L must contain infinitely many atomic sentences (= sentences that are not built up from other sentences in L by means of truth-functional connectives). Therefore, for every A in LA, where A = [p] for some consistent sentence p of L, we can always find an atomic sentence q that does not appear in p. Since p(q is stronger than p and consistent (as far as truth-functional sentential logic is concerned), [p(q] is a non-zero element of LA that is strictly smaller than [p]. Consequently, [p] is not an atom of LA.
Given that LA is atomless, it must be incomplete. For let X be any smallest set of states in LA such that for each atomic sentence p either [p] or [(p], but not both, belongs to X. Suppose for reductio that LA is complete. Then LA contains inf(X). This, however, is impossible, since the infimum of X, if it existed, would have to be an atom of the algebra. To see this, let q be any sentence such that [q] = inf(X). By the definition of X, q must be (truth-functionally) consistent; and since for every atomic sentence p, q implies either p or (p, q cannot be extended to a stronger consistent sentence.
References
Bell, J.L., and A.B. Slomson: 1969, Models and Ultraproducts: An Introduction, North Holland Publ. Comp., Amsterdam - London.
Broome, John: 1990, Bolker-Jeffrey Expected Utility Theory and Axiomatic Utilitarianism, Review of Economic Studies 57, pp. 477-502.
Davidson, Donald: 1980, Towards A Unified Theory of Meaning and Action, Grazer Philosophische Studien 11, pp. 1-12.
Davidson, Donald: 1982, Expressing Evaluations, Lindley Lectures, Lawrence, Kansas, University of Kansas.
Davidson, Donald: 1985, A New Basis for Decision Theory, Theory and Decision 18, pp. 87-98.
Davidson, Donald: 1986, Truth and Interpretation, ed. by E. LePore, Blackwell, New York.
Davidson, Donald: 1990, The Structure and Content of Truth, Dewey Lectures, The Journal of Philosophy 87, pp. 279-328.
Davidson, Donald: 1996, The Folly of Trying to Define Truth, The Journal of Philosophy 93, pp. 263-278.
Hansson, Sven Ove: 1998, Structures of Value. An Investigation of the Statics and Dynamic of Values and Norms, Lund Philosophy Reports, Department of Philosophy, Lund University.
Jeffrey, Richard C.: 1983, The Logic of Decision, University of Chicago Press, Chicago and London, second revised edition; first edition 1965.
Kanger, Stig: 1980, A Note on Preference Logic, ThD 60 - Philosophical essays dedicated to Thorild Dahlquist on his sixtieth birthday, Philosophical Studies published by Philosophical Society and Department of Philosophy, University of Uppsala, vol. 32, Uppsala, pp. 37-38.
Levi, Isaac, forthcoming, Representing Preferences, to be published in the Schilpp volume on Davidson.
Quine, W. V.: 1960, Word and Object, MIT, Cambridge, Mass..
Quine, W. V.: 1990, Pursuit of Truth, Harvard University Press, Cambridge, Mass., and London.
Rabinowicz, Wlodek: 1998, Preference Logic and Radical Interpretation: Kanger Meets Davidson, in Lars Lindahl, Jan Odelstad and Rysiek Sliwinski (eds.), Not Without Cause - Philosophical Essays Dedicated to Paul Needham on the Occasion of His Fiftieth Birthday, Uppsala Philosophical Studies 48, Uppsala University, Department of Philosophy.
Ramsey, Frank: 1950, Truth and Probability, in his Foundations of Mathematics and Other Logical Essays, ed. by R.B. Braithwaite, Routledge & Kegan Paul, London, pp. 156-198.
Rawling, Piers: 1996, Davidsons Measurement-Theoretic Reduction of the Mind, draft.
Stoutland, Fredrick: forthcoming, Davidson on Truth and Interpretation, a book chapter.
Wilson, Neil: 1959, Substances Without Substrata, Review of Metaphysics 12, pp. 521-539.
von Wright, Georg Henrik, 1963, The Logic of Preference, Edinburgh University Press, Edinburgh.
PAGE
PAGE 23
Notes
This paper is a revised version of Rabinowicz (1998). I wish to thank several people who have helped me with comments, references and suggestions: John Broome, Thorild Dahlquist, Sven Danielsson, Dagfinn Fllesdal, Sren Halldn, Bill Harper, Per-Erik Malmns, Paul Needham, Jan Odelstad, Rysiek Sliwinski, Howard Sobel, Fredrik Stjernberg, Fredrick Stoutland, Gran Sundholm, and Folke Tersman. I am especially grateful to Donald Davidson, who has kindly supplied historical information, and to Sten Lindstrm, who has proved a theorem I needed for my argument, apart from being supportive in many other ways.
Kanger had to impose completeness because of his choice of primitive: instead of }", he took {" to be his only primitive and then he simply defined }" as the complement of {". Such a definition immediately yields completeness for }". Which shows, by the way, that }" should not be defined in this way if we want to uphold a distinction between indifference and incomparability.
Proof: Suppose, for reductio, that (i) A {" B, (ii) B {" C, but (iii) not A {" C. By the definition of {" and by the transitivity of }", (i) and (ii) imply (iv) A }" C. Again by the definition of {", (iii) and (iv) imply (v) C }" A. Since }" is transitive, (v) and (i) imply (vi) C }" B, which contradicts (ii).
Since A and B are incompatible, we could just as well replace the exclusive disjunction A(B with the inclusive one: A(B. A word of warning: That (IEDI) is an intuitive condition does not mean that it is unassailable. A counter-example is provided in the next footnote.
In fact, a slight strengthening of (IEDI) is an axiom in the Bolker-Jeffrey theory of expected utility:
(Averaging) For all mutually incompatible states A and B,
(i) if A {" B, then A {" (A(B) {" B; and (ii) if A ( B, then A ( (A(B) ( B.
Since A(B = A(B, averaging implies that an exclusive disjunction can be strictly interpolated between unequally valued disjuncts, if these are mutually incompatible.
The role of the averaging axiom in Bolker-Jeffrey theory is somewhat similar to the function of the axiom of independence in other axiomatisations of expected utility. Even though independence is a much stronger assumption than averaging, the two axioms express essentially the same idea. Consequently, the well-known Allais-type and Ellsberg-type objections to independence can be re-formulated as objections to averaging and to (IEDI). Thus, (IEDI) is not quite as innocent as it might seem to be! Here is an Allais-type counter-example: Suppose that, being risk-aversive, you prefer receiving one million dollars to a high chance of five million dollars, even though the chance is as high as, say, .9. Let A be the state to the effect that you receive one million but dont get the .9 chance for five millions, and let B be the state in which it is the other way round: instead of one million, you get the .9 chance for five millions. A and B are thus incompatible and A {" B. Suppose, further, that A and B are equiprobable. Then the state A(B may be seen as a prospect of getting one million with the probability of .5, five millions with the probability of .45, and of getting nothing with the remaining probability of .05. It may well happen that you prefer B to A(B (if there is some risk of getting nothing, you may be prepared to take an additional risk in order to win a larger prize), even though you prefer A to B. Which violates (IEDI).
In this case, not only the exclusive but also the inclusive disjunction of A and B has a higher expected utility than A and B: U(A(B) = U(C(D(E) = 5/3. But unlike (IED), the Interpolation of (Inclusive) Disjunction:
(ID) For all states A, B, if A {" B, then A }" (A(B) }" B,
is a principle that is consistent with (4L). To see that, consider the expected utility model just as the one we have described, in which the three equi-probable, mutually exclusive and jointly exhaustive states C, D, E are assigned U-values 1/5, 2/5, 2/5, respectively. (4L) is still satisfied and if C, D and E are assumed to be the atoms of this state algebra, it can be checked that the model satisfies (ID).
Maximin and maximax are two examples of what might be called extremal preferences. Such preferences are discussed at length in Sven Ove Hansson (1998), Ch. 7.
For another interpretation of ceteris paribus preference cf. Sven Ove Hansson (1998), Ch. 6. Following von Wright (1963), Hansson interprets such a preference in a very demanding way: A is ceteris paribus preferred to B iff for any possible realisations of these states, the A-realisation is preferred to the B-realisation, provided that other things are equal between them. More precisely, A is ceteris paribus better than B iff every complete alternative x (= Hanssons analogue of a possible world) that instantiates A but not B is preferable to a complete alternative y that instantiates B but not A, and is otherwise as similar to x as possible.
While I take ceteris paribus preference to be interpretable in terms of the dyadic relation of comparative similarity of alternatives (worlds) to an assumed status quo, Hansson interprets this kind of preference in terms of the degrees of similarity between pairs of alternatives: we are supposed to look for a pair of alternatives that are as similar to each other as possible. Thus, the underlying comparative similarity relation is four-place rather than two-place: x is more similar to y than x to y. It is easy to show that the preference principles validated by the two interpretations differ from each other (cf. ibid., p.83f). (IED) is a case in point: it is valid on my interpretation but invalid on Hanssons. Both interpretations of ceteris paribus preference are, I think, legitimate, but the proposal I favour is considerably less demanding and therefore more common in everyday use.
Note that we also have it that c(A(B) = x. Thus, the present interpretation validates both (IED) and (ID).
For an excellent overview of this early work, see Stoutland (forthcoming).
These limits are delineated as follows (cf. Jeffrey, 1983, sections 6.1 and 6.6.): For any probability-utility pair (P, U) that provides an expected utility representation for a preference ordering }" on the algebra of propositions (with the logically false proposition removed), (P , U ) is also an expected utility representation of }" iff for some parameters a, b, c, d such that
(i) ad - bc > 0,
(ii) cU + d is positive for all the arguments of U,
and
(iii) if T is the logically true proposition, cU(T) + d = 1,
U is a fractional linear transformation of U with respect to the parameters in question, i.e.,
U = (aU + b)/(cU +d),
while
P = P(cU + d).
For simplicity, I am here ignoring the distinction between use and mention. I hope this systematic ambiguity will not confuse the reader.
Even though charity considerations directly apply only at the third stage of the interpretation process, they might, when applicable, make us re-consider the interpretive hypotheses we have started with. Thus, if applying charity at the third stage to the material obtained at the earlier stages turns out to be difficult, the interpreter might at that stage come to suspect that he has made some mistakes at the outset of the process. Thus, he might decide to go back to the first stage again and try out a new interpretation of the sentential connectives. In this sense, it could be argued, charity is relevant to all the interpretive stages. But whether such a back-and-forth process of charity-driven interpretation would allow us to identify all pairs of probabilistically independent sentences is by no means clear.
For this conjecture, I am indebted to Howard Sobel. Sobel has also reminded me that Ramseys gambles cannot be described by means of truth-functional statements.
Davidson, (1985), fn. 5, and (1990), fn. 68: I am indebted to Stig Kanger for showing me why an earlier attempt at a solution to this problem [= identification of the connectives] would not work. The details of their encounter have been confirmed by Davidson, in personal communication. Dagfinn Fllesdal, who was their host in Oslo, has informed me that he arranged the 1979 meeting because he felt that Kangers logical expertise might be helpful in setting straight the formal framework for radical interpretation.
[Kanger] also added some needed refinements to the present proposal. (Davidson, 1985, fn. 5, and 1990, fn. 68.)
Strictly speaking, since it is a rationality constraint that a sentence is never held to be both true and false at the same time, condition (() is redundant given (().
See Axiom 2 in Jeffrey (1983), Ch. 9. The preference ordering is defined on such an algebra with its zero element removed. As Jeffrey points out, the assumptions of atomlessness and completeness are used in the proof of the existence part of the representation theorem. They are not needed for the uniqueness part. For an excellent short presentation of the theorem, cf. Broome (1990). That the requirements of completeness and atomlessness might pose a problem for Davidson is suggested in Rawling (forthcoming).
,ABCV34=^
x
(J "24xz
NPVXHJ
jCJCJNH
jCJCJOJQJ6CJNHCJ5CJj0J5:CJU
5:CJO,CUV=^(X$ddd$d$dxd$d
$dx$dd,CUV=^]
(X^5L J!!">#J##$(%%&&Z''(.())--8..35I7q778t:::;};;^<r=@>AbDDNFVGJdLQQSTVW\^9__`cde*ggaiWjgkleo%q&tb>@^`qr=fhln|~+,5; $ & : < B D !
!!!^!`!f!h!~!!!!!!("*"0"2"H"J"P"R"""""####"#$#,#.#
jCJ6CJCJNH
jCJj0JCJUCJOJQJCJU5L J!!">#J##$(%%&&Z''(.())--8.$dd$d$d.#4#6#L#N#R#T#b#d#j#l#t#v#~##############R$T$Z$\$$$$$$$b%d%j%l%2&4&D&F&&&&&&&&&&&''t'v'~''''''''''(((('(((*(+((((((((6CJ
jCJ
jCJCJOJQJ
jCJ
jCJ
jCJCJS(())))**R+e+----....$.&.,...4.6.90:000X3Z3l3n3334444F5H5`5b555O6P666#7$7X7Y7g7h7t7u777777777
8888k8l8888899999
jCJ
jCJ
jCJj0JCJU
jCJCJOJQJCJNH6CJCJ
jCJN8..35I7q778t:::;};;^<>AbDDNFVGJdLQQ$d$d$d
$d$d$d99::: :.:/:4:5:p:q:::::::::::::::;;<<B<Q<W<X<<<>>t??@@@@T@V@|@~@@@@@AA\B^B"C$CNNNOOO
OOPPPPPFPHPLPNPvPxPPPPPP
jCJ
jCJ
jCJ
jCJ
jCJCJNHj0JCJU6CJ5CJCJOJQJCJLPPQQ
QQQQQQQ
RRRRRRRES^SSSJTKT,U-USVTVWlXmXXXXXo\\\|]]]]h_s______`&b'bbbbbbbcdeeeeefffg*ggggggai$j%jNH6CJNH6CJNH6CJCJNHCJCJNH6CJj0JCJU
jCJCJCJOJQJMQSTVW\^9__`cde*ggaiWjgkleo%q&t>wKxdx
$77d$d$d$d
/%jWjgkkk0lAlMlNlmm!n6nn o?o@oeo%q^rarttIuKuLuUuuuvvwwwwxxxxzy{yyyzz||||}}}&~,~J~p~r~"$0fpr~EԁՁR!"
jCJ
jCJCJNH6CJ
56CJCJOJQJ5CJ6CJNH6CJCJNHCJCJN&t>wKx_x*y.zUzz|}|~ER&˃%jƈۍ#.;&|?ΞΥ,IǪ6xڭ& |qRly(ud?0DXZ[cKx_x*y.zUzz|}|~ER&˃%jƈۍ#
$77d$d$dd$d$d" ~ɄτЄ
CDYyAbƈ^_ȌɌڌ&'!";<uvxyz{|}
CJOJQJ
jCJ5CJ6CJNHj0JCJU6CJCJNHCJT#.&|ΞΥ,IǪ6xڭ&$d$d$dАҐސ֑ؑ02@BFHJLғԓDa>A×jw ! xzʞƟޟ-x|<>Vp~5CJj0JCJU6CJCJOJQJCJNH
jCJCJV¥"$,l*,ln YZB\RTlnprHJʮ̮ЮҮ
Z\دگ
:<
jbCJ
jaCJ
jCJ
jCJj0JCJU6CJ5CJCJOJQJCJNHCJO<HJNPVXlnrtxz~:;=>Ʋǲɲʲ RTt~̵εxy(*DdVX\^b
jbCJ
jaCJ6CJCJNH
jCJCJOJQJ5CJCJU |Rly(ud?0$d$d$dbd'01;K AQ_o<=STz{)*?@st|}(H45!*CDoq7=67AMlujkCJOJQJj0JCJU
jgCJ
jbCJ
jaCJCJNH6CJCJ5CJP/0$,19:; Na&/(1mn@K?@}~+,!+HIZ[
jCJ5CJ
jCJ
jCJj0JCJU6CJCJNH
jCJCJ
jCJO0DXZ[flO
$Vd$d$d$d$d$d$d[flO+?BLYFFc 2
Y
nPPv4Q-7V Z"["\"
C[fj@[^_TefnqrX]
<=*+I2Hy)*FGCJOJQJj0JCJU6CJ0J
j0JUCJNH6CJNH5CJ6CJCJ0JCJN+?BLYh&`#$Vd
$VdFFc 2
Y
nPP$d$d$d$ddd&`#$"$dfv|46:<FHrLNln\^cd "(*NPfhnpvx Vb()
jCJ
jCJ
jCJj0JCJUCJNH6CJCJCJOJQJS ` a i j k l
"
(
*
Y
Z
z
)hinoPR?@QRkl-. !"#78; < E F R
jbCJCJOJQJCJNH
jCJ6CJj0JCJUCJSPv4Q-7V Z"["\"
$Vddddx$d$ddR S V W Z"["\"CJNHj0JCJUCJ
jgCJ"
00. A!"#$%"
00. A!"#$%
[2@2NormaldhxCJmH HHRubrik 1
$<5CJKHOJQJkH@@@Rubrik 2$$d@&6CJBA@BStandardstycketeckensnitt22Formatmall1$CJ.@.FotnotstextCJ2&@2FotnotsreferensH*0@"0Sidhuvud
p#")@1" Sidnummer, @B,Sidfot
p#8>@R8Rubrik$dh8
5;CJ"(@a" Radnummer2
@2Index 1
o#
6B@6Brdtext$dhCJ2+@2
SlutkommentarCJ@*@@SlutkommentarsreferensH*A>#<&,N.13$e/m)x~>Al3k!e
9#
!!##Tz
$z
!!!!!!!!!!$.#(9DP%j"<b[R \"8.QKx#0P\"&t[\"
$!!ck CE
<D"* #ae+.C F !!K"M"d"h"E#H#$$%% ))c+d+W-Z-////57<77788G;N;<<A=H==?D?H?I?FFJGQGHHKKVVjYrYZZo]u]__!a1aaabbbbzccHdNdggjj[kdkmm&m(m+m.mcmfmmmmm'n*nMnPnQnTnWnZnnnnn0o3ooo&r,rssxxy yyyyyyyzzm{y{~~~~AD!$@HAI|ry5<̄ӄąӅօFMUX^_ƆɆ
T[stćǇ #PS07Z[ފ()wxˌΌŎ]dlo7:GJИ֘X_Ɯ̜
agY_IOߪ!\]CIֱou;Cahehost{ļż̼(/VY$+CJel
#(ELRU^cdjtz|
%/:CEJKSY^_f<=z~4:>BLP26pst{ kn9<]fAJVX/;6Bu| CHA
G
i
"%z|IM^cKQG I !!
""""##/$1$$$%%V%X%y%~%%%&&((* *h+j+3,4,<,>,,,W-Z--N. 00R0T000F1H146@ACC!FGF'G)GeJgJLLwOxOR&RYXZXmwx|*.DI|Ӆԅ367=be͆Іچ݆ćNPy{ostzJL
Ɍˌύٍ&*+-ڏJMNTt͒8:7=>B(,BɸƻȻRU59:KM_a9<%'ewlodek Rabinowicz4C:\windows\TEMP\terskapningsinfo fr KANGERLMPS.asdwlodek RabinowiczE:\Text\KANGERLMPS.dotwlodek Rabinowicz4C:\windows\TEMP\terskapningsinfo fr KANGERLMPS.asdwlodek RabinowiczE:\Text\KANGERLMPS.dotwlodek RabinowiczE:\Text\KANGERLMPS.dotwlodek RabinowiczE:\Text\KANGERLMPS.dotwlodek RabinowiczE:\Text\KANGERLMPS.dotwlodek RabinowiczE:\Text\KANGERLMPS.dotwlodek RabinowiczE:\Text\KANGERLMPS.dotwlodek RabinowiczE:\Text\KANGERLMPS.dot@(]]]]]]] ]
] ()*+,-./012^_InIoIpyz~]]22222@
0@<@ "$&P@.`@24l@>@BDFHJLNP@~@(@@@P@d@l@x@@@@@
@0@GTimes New Roman5Symbol3&ArialO&?Lucida Sans Unicode"s9FIA&"3
S[?!a0d0What s Stig Got to Do with ItWlodzimirez Rabinowiczwlodek Rabinowicz
Oh+'0 (4
P\h
tWhats Stig Got to Do with ItMihatWlodzimirez RabinowiczilodKANGERLMPS wlodek Rabinowiczwi10dMicrosoft Word 8.0i@p! @\+-f@N]@SQU
՜.+,D՜.+,`hp
Filosofiska institutioneno[S0jWhats Stig Got to Do with ItTitle 6>
_PID_GUIDAN{F28A7A2A-F8BB-11D2-848D-00201855257A}
!"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~Root Entry Fl:`RU1Tableq&WordDocumentHzSummaryInformation(DocumentSummaryInformation8CompObj`ObjectPool`RU`RU
FWord-dokument
MSWordDocWord.Document.89q