You are on page 1of 9

The Pragmatics of Explanation Author(s): Bas C. Van Fraassen Source: American Philosophical Quarterly, Vol. 14, No. 2 (Apr.

, 1977), pp. 143-150 Published by: University of Illinois Press on behalf of North American Philosophical Publications Stable URL: http://www.jstor.org/stable/20009661 Accessed: 28/02/2010 23:24
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=illinois. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

University of Illinois Press is collaborating with JSTOR to digitize, preserve and extend access to American Philosophical Quarterly.

http://www.jstor.org

American Volume

Philosophical 14, Number

Quarterly 2, April 1977

Paper American Portland,

presented

at

the

51st Annual

Association, Philosophical March 1977 Oregon,

Meeting Pacific

of

the

Division,

VII.

THE

PRAGMATICS
BAS C. VAN

OF EXPLANATION
FRAASSEN
that we can intelligibly say that the theory explains, that people can explain by means and not merely is not very of the theory. But this consequence restrictive, because the former could be an ellipsis for the latter. There are questions of usage here. I am happy to report that the history of science allows systematic use of both idioms. In Huygens the and Young seemed to be that phenomenon typical phrasing laws and may be explained by means of principles, a or to the other view.1 On according hypotheses, in 1815 "Tous ces hand, Fresnel writes to Arago
ph?nom?nes . . . sont r?unis et expliqu?s par la

are two problems scientific T'HERE about * to it: when is first The is describe explanation. something explained? The second is to show why is a virtue. Pre? sense) explanation (or in what
sumably we have no explanation unless we have a

good theory ; one which is independently worthy of acceptance. But what virtue is there in explanation over and above this? I believe that philosophical concern with the first problem has been led on views the mistaken astray by thoroughly
second. I. False Ideals

three ideas about begin I wish to dispute a seem to that have influ? subliminal explanation ence on the discussion. The first is that explanation is a relation simply between a theory or hypothesis or facts, just like truth for and the phenomena is that explanatory The second power example. cannot be logically separated from certain other virtues of a theory, notably truth or acceptability. And the third is that explanation is the overriding the end of scientific virtue, inquiry. When is something explained? As a foil to the above three ideas, let me the simple propose answer: when we have a theory which explains. Note To
first that "have" is not "have on the books"; I

m?me and Lavoisier th?orie des vibrations," says that the oxygen hypothesis he proposes explains the
phenomena of combustion.2 Darwin also speaks in

idiom: "In scientific investigations it is to invent any hypothesis, and if it permitted classes of explains various large and independent facts it rises to the rank of a well-grounded theory" ; though elsewhere he says that the facts of geo? graphical distribution are explicable on the theory of the latter
migration.3

cannot claim to have such a theory without imply? ing that this theory is acceptable all told. Note also
that both "have" and "explains" are tensed; and

My answer did separate acceptance of the theory from its explanatory power. Of course, the second can be a reason for the first ;but that requires their have held that separation. Various philosophers true explanation logically requires (or acceptable)
theories as premises. Otherwise, they hold, we can

at most
explanation.

that I have allowed that we can have a theory which does not explain, or "have on the books" an
unacceptable one that does. Newton's theory ex?

mistakenly

believe

that

we

have

an

plained
perihelion

the

tides

but
;we

not
used

the advance
to have an

in

the

This is also a question of usage, and again usage is quite clear. Lavoisier said of the phlogiston that it is too vague and consequently hypothesis
"s'adapte on veut a toutes les le faire entrer."4 explications Darwin dans explicitly lesquelles allows

of mercury

accept?

able theory, provided by Newton, which bore (or bears timelessly?) the explanation to relationship some facts but not to all. My answer also implies

explanations hardly be

by false theories when he says "It can that a false theory would supposed

11 owe these and following to my student Mr. Paul references see C. Huygens, For instance Treatise on Light, tr. by Thagard. S. P. Thompson (New York, Peacock ed. by George 1962), pp. 19, 20, 22, 63; Thomas Works, Young, Miscellaneous (London, I> PP- 168, 170. 1855), Vol. 2 Oeuvres Compl?tes Fresnel, Augustin Oeuvres (Paris, (Paris, I, p. 36 (see also pp. 254, 355); Antoine 1866), Vol. Lavoisier, II, p. 233. 1862), Vol. 3Charles The Variation of Animals and Plants Darwin, (London, I, p. 9; On the Origin of the Species (Facs. of first 1868), Vol. edition, Mass., Cambridge, 1964), p. 408. 4 Antoine Lavoisier, op. cit., p. 640.

H3

144 in so

AMERICAN

PHILOSOPHICAL

QUARTERLY

explain, theory

satisfactory selection,

a manner the several

as

does

the

believing
or does,

that the phenomenon


indeed occur."7 That

to be explained
information has

did,
two

of natural

large

classes

of facts above specified."5 More recently, Gilbert that a theory has argued Harman similarly: the evidence is of certain part phenomena explains that the that leads us to accept it. But that means is visible beforehand. Finally, explanation-relation we criticize theories selectively :a discussion of celes? around the turn of the century would tial mechanics the assertion that Newton's contain theory surely does explain many planetary phenomena, though not the advance in the perihelion of Mercury. I consider is a third false ideal, which There is the summum bonum and worst: that explanation exact aim of science. A virtue could be overriding in one of two ways. The first is that it is a minimal Such is consistency with criterion of acceptability. the facts in the domain of application (though not are all if these with data, dubitable!). necessarily is not like that, or else a theory would Explanation at all unless it explained all facts not be acceptable in its domain. The second way in which a virtue is that of being required when it may be overriding that if two theories can be had. This would mean tests other pass simplicity) (empirical adequacy, then the one which explains more equally well, must be accepted. As I have argued elsewhere,6 and as we shall see in connection with Salmon's of this demand views below, a precise formulation for indeterministic variables hidden requires theories. But of course, hidden variables are rejected
in scientific practice as so much "metaphysical

one supplied by components, the other consisting of auxiliary of providing The relationship as explicated (a) implying conferring a high probability not lowered by the addition
evidence.

the scientific theory, factual information. is good grounds or (D-N case), (b) is (I-S case), which of other (available)

As Hempel is not a points out, this criterion for explanation: sufficient condition the red shift that distant gives us good grounds for believing galaxies are receding from us, but does not explain classic case is the barometer they do. The why come exactly storm the will if the baro? example: meters if do the atmos? fall, which they exactly pheric conditions are of the correct sort ; yet only a is the criterion the last factor explains. Nor case for this is the classic the necessary condition;
paresis example. We explain why the mayor, alone

the townsfolk, contracted among paresis by his of latent, contracted history syphilis; yet such histories are followed by paresis in only a small
percentage of cases.

of testa? second criterion is the requirement bility ;but since all serious candidates for the role of scientific theory meet this, it cannot help to remove the noted defects. The
2. Beckner, The Putnam, criterion and of Salmon explanatory relevance was

baggage" empirical

they predictions.

when

make

no

difference

in

II. A Biased History to characterize the attempts outline I will of the past three decades, with no explanation the On the contrary, of objectivity. pretense to illustrate the diagnosis, and selection is meant point to the solution, of the next section. i. Hempel his summarized In 1966, Hempel two main for criteria explanation. listing
is the criterion of explanatory relevance: "the

revised in one direction, informally by Beckner and Putnam and precisely by Salmon. Morton Beckner, in his discussion of evolution theory, pointed out that this often explains a phenomenon only by showing how it could have happened, given certain do this by con? possible conditions.8 Evolutionists of processes which utilize only structing models
and genetic the outcome natural agrees selection with the mechanisms, actual in which phenomenon.

views by The first


explana?

Parallel conclusions were drawn by Hilary Putnam are in which celestial phenomena about the way theory of gravity : celestial explained by Newton's motions could indeed be as they are, given a certain possible (though not, known) distribution of masses
in the universe.9

tory information

adduced

affords good grounds

for

We may take the paresis example to be explained the theory is of consistency with similarly. Mere

5 Origin (sixth ed., New York, 1962), p. 476. 6 "Wilfrid vol. 14 (1975), pp. 606-616. Sellars on Scientific Realism," Dialogue, 7G. G. Cliffs, New Jersey, 1966), p. 48. Hempel, Philosophy of Natural Science (Englewood 8 The in 1959. Biological Way of Thought (Berkeley, 1968), p. 176; this was first published 9 In a a summary is found in Frederick paper of which Suppe (ed.), The Structure of Scientific

Theories

(Urbana,

111., 1974).

THE

PRAGMATICS

OF

EXPLANATION

I45

course much too weak, since that is implied by Hence Salmon made irrelevance. Wesley logical this precise as follows: to explain is to exhibit (the) relevant factors.10 (I shall leave till statistically about later the qualifications "screening off.") Since this sort of explication discards the talk about and mechanisms of Beckner and Putnam, modelling itmay not capture enough. And indeed, I am not that his criterion satisfied with Salmon's arguments a He condition. sufficient gives the example provides of Uranium of an equal mixture 238 atoms and Polonium the Geiger 214 atoms, which makes counter click in interval (t, t+ m). This means that one of the atoms disintegrated. Why did it ? The correct answer will be: because it was a Uranium the probability of 238 atom, if that is so?although is much its disintegration higher relative to the that the atom belonged to the previous knowledge described mixture.11 The problem with this argu?
ment is that, on Salmon's criterion, we can

in different ways by Michael and was developed Friedman and James Greeno. Friedman says kind of under? in his that "the view, explicitly standing provided by science is global rather than in the simplification and local" and consists on our world That unification picture.14 imposed of two facts: St Sx explains S2 1S a conjunction knowledge implies S2 relative to our background (and/or belief) K, and ?\ unifies and simplifies the set of its consequences relative to K. Friedman will no doubt wish to weaken the first condition in view
of Salmon's work.

Friedman The precise explication gives of the second condition does not work, and is not likely
to have a near variant that does.15 But here we may as

look at Greeno's
statement

proposal.16 His abstract and closing


to the same general view

subscribe

Friedman. But he takes as his model which specifies a single probability


correct one, plus two partitions

of a theory one space Q as the


(or random

explain not only why there was a disintegration, but also why that atom disintegrated just then.And surely that is exactly one of those facts which atomic physics leaves unexplained ?
But Whatever there the is a more phenomenon serious general can is, we criticism. amass the

variables)
and the

of which
other

one

is designated
An example:

explanandum
sociology

explanans.

statistically
does "What Salmon have an not more

relevant
rule out could

factors,
the one But

as long as the theory


altogether. explanation?" as soon as we we have an

phenomenon ask of an in that case, theory, as we

inquires.12 empirically

adequate

explanation
claim the an

of every
can

fact
as be soon

in its domain. We
have shown embedded is, in some does not

may
that model throw But

explanation

phenomenon allowed the by theory?that on doubt the theory's empirical

cannot lives in San who explain why Albert, and whose Francisco father has a high income, to. But it does explain steals a car. Nor is itmeant terms in of such other factors as delinquency and parental residence of income. The degree power is measured explanatory by an ingeniously devised quantity which measures the information / the theory provides of the explanandum variable M on the basis of explanans S. This measure takes its maximum if all conditional value probabilities are zero or one (D-N P{M{?S^ case), and its minimum value zero if S and M are statistically
independent. Unfortunately, tion imposed on this way of measuring our data abandons the unifica? Friedman's

adequacy.13

surely that is too sanguine

3. Global Properties power cannot be identified with Explanatory but it may still reside in the empirical adequacy;
performance of the theory as a whole. This view is

insight that scientific understanding identified as a function of grounds For if we let S and M expectation.
behavior of the barometer and coming

cannot be for rational describe the


storms, with

that science does accompanied by the conviction not explain individual facts but general regularities

P(barometer falls) =P(storm comes) =0.2, P (storm = P (storm comes/ and 1, comes/barometer falls) barometer does not fall)=o, / then the quantity

10 "Statistical in R. G. Colodny pp. 173-231 Explanation," (ed.), The Nature and Function of Scientific Theories (Pittsburgh, 1970) ; also in Salmon's book cited below. reprinted 11 has further, unpublished, Ibid., pp. 207-209. Nancy Cartwright to the necessity and sufficiency of Salmon's counter-examples criterion. 12 Ibid., p. 222. 13 are discussed These in my "To Save the Phenomena," concepts The Journal vol. 73 (1976), of Philosophy, forthcoming. 14 and Scientific The Journal of Philosophy, "Explanation vol. 71 (1974), pp. 5-19. Understanding," 15 See Philip Kitcher, and Unification," The Journal of Philosophy, vol. 73 (1976), pp. 207-212. "Explanation, Conjunction, 16 and Information," in Wesley Salmon "Explanation and Statistical Relevance pp. 89-103 (ed.), Statistical Explanation (Pitts? with a different burgh, title in Philosophy 1971). This paper was originally published of Science, vol. 37 (1970), pp. 279-293.

I46

AMERICAN

PHILOSOPHICAL

QUARTERLY

takes itsmaximum
we designate M

value.
or S as

Indeed,
explanans.

it does sowhether

bound
temporal

into causal
continuity

chains by two relations:


and statistical

spatio

relevance.

It would seem that such asymmetries as exhibited by the red shift and barometer examples must
necessarily restraints on remain recalcitrant for any attempt to

Explanation requires the exhibition of such chains. Salmon's of departure is Reichenbach's point
principle of the common cause: every that a relation of

strengthen Hempel's
theories

or Salmon's
alone.

criteria by global

statistical
causal

relevance

ought
means

to be explained be explained
gives two

by one of
of

relevance.

This

correlation

simultaneous 4. The Major Difficulties There are two main difficulties, illustrated by the old paresis and barometer examples, which none of the examined positions can handle. The first is that
there the are request cases, for clearly explanation in a domain, theory's is nevertheless where rejected. common cause.

values must
Salmon

by a prior
condi?

statistical

tions that must be met events A and B:

by a common

cause C of

= (a) P(A & B?C) P{A?C)P(B?C) = B from (b) P{A/B &O P(A/C) "C screens off
A."

We can explain why John, rather than his brothers contracted paresis, for he had syphilis ;but not why he, among all those syphilitics, got paresis. Medical science is incomplete, and hopes to find the answer
some day. But the example of the uranium atom

If P(B?C) t?0 these are equivalent, and symmetric in A and B. is typically the demand Suppose that explanation
for a common cause. Then we still have the prob? and

then rather than is just later, disintegrating the theory to be formally similar and we believe as the also reject such questions complete. We :why does a body Aristotelians asked the Galileans free of impressed forces retain its velocity? The of this sort of case, and its pervasive importance character, has been repeatedly discussed by Adolf
Gr?nbaum.

lem :when
explain the

does this arise ?Atmospheric


correlation between

conditions

barometer

say; but are still prior causes required to the correlation between explain atmospheric conditions and falling barometers ? storm,
In the is violated quantum because domain, "causal Salmon influence says, causality trans? is not

mitted
situation

with
is

spatio-temporal
worse. To

continuity."
assume

But

the

to be satisfiable, aside, is to principle continuity rule out all genuinely indeterministic theories. As let a theory say that C is invariably does, it may be that it explains the one in terms of example, events A, B, or followed not versa. the other and vice An which by one of the incompatible example combines both the first and second difficulty is this : D, each with probability 1/3. Let us suppose the and its probabilities to atomic chemical element each irreducible, theory complete, according physics, a charac? with C the complete specification of state. Then we has a characteristic structure atomic and will find a correlation for which only C could be the teristic spectrum (of light emitted upon excitation). common but it is not. Assuming that A, B, D Yet the spectrum is explained cause, by the atomic are a and the substance has structure, always preceded by C and that they have low question why but equal prior probabilities, there is a statistical that structure does not arise at all (except in the = = or correlation between trivial sense that the questioner may need to have <f> (A D) and if* (B or D), = for P(?/0)=P(^/?) to him). the terms explained i/2^P(?). But C, the only available from 0 : candidate, does not screen off <j>
one condition obtains when and only when another

The second difficulty is the asymmetry revealed : even the barometer if the by theory implies that

Reichenbach's

5. Causality Why are there no longer any Tasmanian


Well, they were a nuisance, so the white

natives
settlers

just kept shooting them till there were none The request was not for population statistics, for the story; though in some truncated way,
statistics "tell" the story.

left. but the to


are

this may sound complicated, the con? Although struction is so general that almost any irreducibly situation will give a similar example. probabilistic Thus Reichenbach's principle of the common cause is in fact a demand for hidden variables. Yet we retain the feeling that Salmon has given
an essential clue to the asymmetries of explanation.

P(0/C&0)=P(0/0)

i/2^PWC)

which

is 2/3.

In a later paper Salmon


causal 17 mechanisms "Theoretical in

gives a primary

place

explanation.17 pp. 118-145

Events in Stephan

For surely the crucial point about


(ed.), Explanation (Oxford, 1975).

the barometer

is

Explanation,"

K?rner

THE

PRAGMATICS

OF

EXPLANATION

147

screen off the that the atmospheric conditions fall from the storm? The general point barometer are totally bound up with that the asymmetries
causality was argued in a provocative article by

correct answer: because of his latent syphilis. But the question why he did in contrast to the other syphilitics in his country club, has no true correct
answer. Intuitively we may in say: contrast Q is a correct

B. A. Brody.18 Aristotle certainly discussed examples : the planets do not twinkle because of asymmetries they are near, yet they are near if and only if they do not twinkle {Posterior Analytics, I, 13). Not all
explanations a second use The spectrum are causal, says Brody, but that : sodium the of others essence. has that Aristotelian angle notion, case is a clear

answer
reasons members

to Why P
to

in contrast toX?
proposal

only
to for

if Q gives
the a other precise

that P, expect of X. Hannson's

it has this atomic because structure, spectrum is its essence. which that Brody's account has the further advantage
he can say for when questions of do not arise: other

of P given Q is higher criterion is : the probability than the average of the probabilities of R given Q, for members R of X. Hannson points out that the set X of alternatives is often left tacit; the two questions about paresis might well be expressed by the same sentence in contexts. The is that different important point
explanations are not requested of propositions, and

properties
request

are explained
an explanation

in terms of essence, but the


the essence does not

consequently
answered However, and

a distinction

can be drawn
in correct a

between
way. to answer

I do not see how he would dis? arise. However, the questions why the uranium between tinguish atom disintegrated and why it disintegrated just then. In addition there is the problem that modern
science is not formulated in terms of causes and

Hannson

requests rejected a makes Q

clear

that these concepts essences, and it seems doubtful can be redefined in terms which do occur there. 6. Why- Questions
A why-question is a request for explanation.

Why P in contrast to X? when Q is statistically irrelevant, when P is already more likely than the rest ; or when Q implies P but not the others. I do not see how he can handle the barometer (or red On his precise shift, or spectrum) asymmetries.
criterion, that the barometer fell is a correct answer

to why it will storm as opposed to be calm. The : are necessarily R is if and P very deep difficulty
equivalent, according to our accepted theories, how

called P the presupposition of Sylvain Bromberger the question Why-P? and restated the problem of as that of explanation giving the conditions under which proposition Q is a correct answer to a why P.19 However, question with presupposition Bengt Hannson has pointed out that "Why was it John who ate the apple?" and "Why was it the apple
which John ate ? : are different why-questions, al?

can Why P in contrast toX? Why R in contrast toX? III. The i. Prejudices Two convictions
of explanation, substantive.

be distinguished

from

Solution the discussion


and one

have
one

prejudiced

is the same.20 The though the comprised proposition difference can be indicated by such phrasing, or by emphasis ("Why did John . . . ?") or by an auxiliary clause than rather ("Why did John ...?").
Hannson says that an

methodological

The first is that a philosophical


to produce necessary and sufficient

account must
conditions

aim
for

of a proposition
proposition. As is at least

explanation

or fact,
suggested

is requested,

not

but
by

of an
Hannson,

E. A similar theory T explaining phenomenon prejudice plagued the discussion of counter-factuals


for twenty years, requiring the exact conditions

aspect of a
we can

under tions

which, are

if A were
liberating insight

the case, B would


was that these

be.

Stalnaker's

condi?

these cases by saying that we wish an of why P is true in contrast to other explanation members of a set X or propositions. This explains all
the The tension in our reaction the mayor, contracted to the question why townfolk generally, paresis-example. in contrast to other paresis has a true

cover

and largely determined by context interest. This central the speaker's brings question to light: whatybrm can these conditions take? The second conviction is that explanatory power is a virtue of theories by themselves, or of their to the world, relation like simplicity, predictive
pp. 20-31.

18 an Aristotelian "Towards of Scientific Theory Explanation," of Science, vol. 39 (1972), Philosophy 19 in R. G. Colodny Mind and Cosmos (Pittsburgh, "Why-Questions," pp. 86-108 (ed.), 1966). 20 What?" Stanford University, "Explanations?Of (mimeographed: 1974).

I48

AMERICAN

PHILOSOPHICAL

QUARTERLY

strength,
again an

truth,
analogy

empirical
with

adequacy. or directly
cases,

There
: it used to be

is

counterfactuals

thought the

that science
In all

contains,
but

implies,
however,

causality, suggests that we should look not only to the worlds theory T allows as possible, but also to those it rules out as impossible, and speaks of
counterfactuals which are counterlegal. Relevant

counterfactuals.

limiting

proposition and dependent,


relative as to the speakers'

expressed the implication

is

context highly is there at most


factors, such

determining interest.

contextual

draw distinctions between logic sentences and their semantics logically equivalent devised by Routley and Meyer use both inconsistent and incomplete worlds. I believe such approaches to be totally inappropriate for the problem of
explanation, tions of detours for when phenomena through we look at we or actual do events not by circumstances theories, explana? see any ruled out

and

entailment

2. Diagnosis The earlier accounts lead us to the format: C explains E relative to theory T exactly if (a) T has certain global virtues, and (b) T implies a certain in the language of proposition E) expressible <?>{C, accounts and Different logic probability theory. to the specification directed themselves of what should go into (a) and (b).We may add, following Beckner and Putnam, that T explains E exactly if C consistent with T (and there is a proposition such that C beliefs) background presumably, explains E relative to T. were proposed by The significant modifications Hannson and Brody. The former pointed out that E cannot be reified as a proposi? the explanadum tion :we request the explanation of something F in contrast to its alternatives X (the latter generally is tacitly specified by context). This modification
absolutely necessary to handle some of our puzzles.

as impossible by the theory. A further approach, developed by Rolf Schock, and Romane sentences Clark, myself distinguishes them true. The idea is by the facts that make simple. That it rains, that it does not rain, that it
snows, and that it does not snow, are four distinct

facts. The disjunction that it rains or does not rain true equally by the first and second, and ismade not by the third or fourth, which distinguishes it from the logically equivalent that it disjunction
snows or does not snow.21 The distinction remains

even

if there is also a fact of its raining or not raining, distinct or identical with that of its snow?
or ing This not snowing. can Such work for the of asymmetries are possible approach

explanation.

asymmetries

It requires that in (b) above we replace "<?(C, E)" by the formula form "^(C, F, X)." But the problem
of implies atomic if T remains because recalcitrant, asymmetries the necessary of F and F' equivalence (say, structure and characteristic spectrum),

because,
is emitted make different structure. metries how up

for example,
with the facts So

the distinct
A, /?,

facts
...

that

light

wavelengths characteristic conjoin we have

spectrum, to make up shown that in how

conjointly while quite the such atomic asym? showed can

then T will implies


The only

also
account

imply
we

F', X) ift(C,
have

if and only
grapples

if it
at

ifj(C,F, X).
seen which

can arise, failures of

in the way transitivity we still comes so of far cause

Stalnaker counterfactuals

arise. But while we have


asymmetrically, problem: whence

the distinct
have the is

all successfully with this, isBrody's. For Brody points out that even properties which we believe to be
constantly can be related asymmetries conjoined into divided as cause were and no in all possible essences and effect. problem In circumstances, accidents, this sense, or the

facts to classify

the non-logical classification? The it essence comes ; but from if so,

only suggestion Aristotle's concepts

that and

modern

science will

not supply

it.

for Aristotle.

3. The logical problem We have now seen exactly what logical problem is posed by the asymmetries. To put it in current terms: how can we distinguish propositions which are true in exactly the same possible worlds?
There are several known approaches that use

4. The Aristotelian Sieve I believe that we should return to Aristotle more thoroughly, and in two ways. To begin, I will state without how I understand Aristotle's argument theory of science. Scientific activity is divided into
two parts, demonstration and explanation, the former

impossible worlds.

David

Lewis,

in his discussion

of

treated mainly by the Posterior Analytics and the latter mainly by Book II of the Physics. Illustrations
pp. 477-487 in Milton and in A. R. Anderson, Munitz (ed.), Logic and

21Cf. my "Facts and Tautological et al, (ed.), Entailment (Princeton, Ontology (New York, 1973).

1975);

Entailments," and "Extension,

The Journal

of Philosophy, vol. 66 (1969), and Comprehension" Intension,

THE

PRAGMATICS

OF

EXPLANATION

I49

in the former are mainly examples of explanations are applied; in which the results of demonstration contain premises and this is why the examples
conclusions which are not necessary and universal

although principles, from such principles.


to our pure such words versus

is only to and demonstration Thus the division corresponds


applied science. There is no

drew my attention to the following : If John asked his father for argument spurious not have quarreled then they would money, (because John is too proud to ask after a quarrel). Also if John asked and they hadn't quarreled, he Professor Geach
would receive. By the usual logic of counter?

reason
have

to think that principles


as "cause" and

and demonstrations
"essence" in them,

looking at pure science from outside, although state causes Aristotle could say that its principles and essences. In applications, the principles may be filtered through a conceptual sieve originating
outside science.

factuals, it follows that if John asked his father for he would receive. But we know that he money, would not, because they have in fact quarreled. because "what was The fallacy is of equivocation, in the constant" middle of the changed kept if the you like, aspects by which monologue. (Or worlds are graded as more or less similar to this
one.) decide factuals. explanations. Because to "keep By science cannot dictate it contains science what mo constant" exact parallel, speakers counter? contains no

The
for the

doctrine
systematic

of the four "causes"


ambiguity or

(aitiai) allows

context-dependence

of why-questions.22 Aristotle's example (Physics II, 3; 195a) is of a lantern. In a modern example, the question why the porch light is on may be answered "because I flipped the switch" or "because we are
expecting company," and the context determines

5. The Logic of Why-Questions What remains of the problem of explanation is to study its logic, which is the logic of why
questions. This can be put to some extent, but not

which

is appropriate. Probabilistic relations cannot these. Which factors are explanatory is distinguish decided not by features of the scientific theory but from outside. This is true by concerns brought even ifwe ask specifically for an "efficient cause," for how far back in the chain should we look, and which factors are merely auxiliary contributors? Aristotle would not have agreed that essence is The essence is what the thing context-dependent. its sum of classificatory is, hence, properties. Realism has always asserted that ontological distinctions determine the "natural" classification. But which property is counted as explanatory and
as explained seems to me clearly context

totally,
and

in the general
and others.23

form developed of three classes


and

by Harrah of response,
A pre?

Belnap

A question
direct answers,

admits

corrections,

comments.

which

For consider Bromberger's dependent. flagpole the shadow is so long because the pole example: has this height, and not conversely. At first sight,
no contextual factor could reverse this asymmetry,

the pole's height is a property it has in and is a very accidental by itself, and its shadow feature. The general principle linking the two is that its shadow is a function f(x, t) of its height x and the time t (the latter determining the sun's elevation). But imagine the pole is the pointer on a the values of f have desired giant sundial. Then for each time t, and we appeal to these properties to explain why it is (had to be) such a tall pole. because
We may again draw a parallel to counterfactuals.

is any proposition supposition, it has been held, or equivalently, implied by all direct answers, I believe we must add that denied by a correction. the question "Why P, in contrast to X?" also pre? of X, (b) P is true supposes that (a) P is a member and the majority of X are not. This opens the door not be that a question may to the possibility set answers. its of direct determined by uniquely itself should decompose The question into factors which determine that set : the topicP, the alternatives the doctrine X, and a request specification (of which of the four "causes" is perhaps the first description). involved in We have seen that the propositions be individuated and answer must by question I something more than the set of possible worlds.
propose that we use the facts that make them true

(see footnote
asymmetric

21). The
relation

context

will

determine
of

an

among

these

facts,

explana?

the theory or tory relevance; it will also determine beliefs which determine which worlds are possible, and what is probable relative to what.
We must now determine what direct answers are

and how

they are evaluated.


pp. 3-18. to Arise,"

They must

be made

22 Cf. Julius Moravcik, on Adequate "Aristotle Explanations," 23Cf. N. D. Belnap, Their Jr., "Questions: Presuppositions, Lambert Things, ed. by Karel (New Haven, 1969), pp. 23-39.

Synthese, vol. 28 (1974), and How They Can Fail

The Logical Way

of Doing

I50

AMERICAN

PHILOSOPHICAL

QUARTERLY

true by facts (and only by facts forcing


are topic explanatorily true. Moreover, relevant these to those facts which must

such) which
make be the statistic?

When
advocated

scientist
theory,

campaigns
point

on behalf
out how our

of an
situa?

he will

tion will

change

if we

accept
relevant, of an

it. Hitherto
known intricate web,

unsus?
relations some

ally relevant, telling for the topic in contrast to the to be this part I believe alternatives generally; Salmon's combining by probabilities, explicable
and Hannson's account. How strongly the answers

factors become pected are revealed to be strands

count
tion

for the topic should be part of their evalua?


as better or worse answers.

The main difference from such simple questions as "Which cat is on the mat?" lies in the relation of a why-question to its presuppositions. A why question may fail to arise because it is ill-posed (P is false, or most of X is true), or because only answers tell probabilistically for question-begging P in contrast to X generally, or because none of the relevant factors that do tell for P are explanatorily Scientific in the question-context. theory enters in the evaluation of possibilities and mainly probabilities, which is only part of the process, and it has in common with other applications which such as prediction and control.

terribly puzzling questions are laid to rest as not arising at all. We shall be in a much better position to explain. But equally, we shall be in a much better position to predict and control. The features of the theory that will make this possible are its empirical and logical strength, not special "ex? adequacy
planatory power" and "control power." On the

IV.

Simple

Pleasures

There
philosophers semantic

are no explanations
come rather to mislocate than pragmatic

in science. How
explanation relations?

did

to say explanatory other hand, it is also a mistake power is nothing but those other features, for then we are defeated no by asymmetries having basis in science. "objective" are new predictions so much more to the Why credit of a theory than agreement with the old ? Because they tend to bring to light new phenomena which the older theories cannot explain. But of in doing course, so, they throw doubt on the empirical adequacy of the older theory : they show that a pre-condition for explanation is not met. As said of the radiometer, Boltzmann "the theories based on older hydrodynamic experience can never describe" these phenomena.24 The failure in
explanation is a by-product.

among This

Scientific explanation.
supremacy theories. of For

was certainly in part because the positivists tended to identify the pragmatic with subjective psycho? measures by which logical features. They looked for
to evaluate are a such, theory theories. but they is watered Truth and empirical are weak, being measure Some down. adequacy when preserved of "good?

is inference to the best inference That does not rule at all for the
we explanation evaluate among how good the an virtues explanation of

ness of fit" was also needed, which did not reduce to a purely internal criterion such as simplicity, but
concerned studies giving of us the theory's relation have to the world. The explanation a measure, such some gone it was but toward way a mistake to

is given by how good a theory is used to give it, how close it fits to the empirical facts, how internally the explanation. There is a simple and coherent in terms of a prior judgment further evaluation of
which this other peculiar instead, and kinds further of factors evaluation are took explanatorily precedence, But relevant. If overriding would be the this is not so as so : to

considerations, virtue sought science schools

explanation above all. our

call this explanatory


to confirm have an this explanation error

power. The
was unless that we we do have

fact that seemed


not a say theory Theories that we which

imagination

revise just those prior judgments


eliminates wonder. power to dismiss is something as Explanatory

of what
we value

satisfies
and

is acceptable,
alternatives,

and victorious
whereby we can

in its competition
explain.

with
are

desire. But we are as ready, for the sake of scientific


progress, questions not really arising

but the peculiar in explanation, and applied are of features supplied by explanation puzzling I shall now redescribe involved. other factors several familiar subjects from this point of view. University of Toronto and University of Southern California

at
still,

all.
less

Explanation
a virtue

is
than

indeed
an

virtue;

but

anthropocentric

pleasure.25

Received
1964), with p. 25. Professors

September
N.

i, igj?
B. Hannson, K.

24 Boltzmann, Ludwig 25The author wishes Lambert, and W.

tr. by S. G. Brush Lectures on Gas Theory, (Berkeley, to acknowledge and correspondence helpful discussions and the financial Council. support of the Canada Salmon,

Cartwright,