Explaining Explanation - PhilPapers · Explaining Explanation ... original and challenging...

Explaining Explanation

David-Hillel Ruben offers a discussion of some of the main historicalattempts to explain the concept of explanation, examining the worksof Plato, Aristotle, John Stuart Mill, and Carl Hempel. Building onand developing the insights of these historical figures, he introducesan elaboration and defense of his own solution.

In this volume, Ruben relates the concept of explanation to bothepistemological and metaphysical issues. Not content to confine theconcept to the realm of philosophy of science, he examines it within a farmore broadly conceived theory of knowledge. He concludes with his ownoriginal and challenging explanation of explanation.

Explaining Explanation will be read with interest by students of generalphilosophy as well as those specializing in the philosophy of science andscholars with a more advanced level of interest.

Private OwnershipReligious Belief and the Will*Rationality*The Rational Foundations of Ethics*Moral Knowledge*Mind-Body Identity Theories*Practical Reasoning*Personal Identity*The Infinite*Thought and LanguageHuman Consciousness*Explaining ExplanationThe Nature of ArtThe Implications of DeterminismWeakness of the WillKnowledge of the External WorldIf P, Then Q: Conditionals and

the foundations of reasoningPolitical Freedom*ScepticismKnowledge and BeliefThe Existence of the WorldNaming and Reference: From

word to object

*Also available in paperback

James O.GrunebaumLouis P.PojmanHarold J.BrownT.L.S.SpriggeAlan GoldmanCynthia MacdonaldRobert AudiHarold W.NoonanA.W.MooreJulius MoravcsikAlastair HannayDavid-Hillel RubenA.L.CotheyRoy WeatherfordJustin GoslingBruce Aune

David H.SanfordGeorge G.BrenkertChristopher HookwayFrederick F.SchmittReinhardt Grossman

R.J.Nelson

The Problems of PhilosophyTheir Past and Present

General Editor: Ted HonderichGrote Professor of the Philosophy ofMind and LogicUniversity College, London

Each book in this series is written to bring into view and to deal witha great or significant problem of philosophy. The books are intendedto be accessible to undergraduates in philosophy, and to other readers,and to advance the subject, making a contribution to it.

The first part of each book presents the history of the problem inquestion, in some cases its recent past. The second part, of a contemporaryand analytic kind, defends and elaborates the authors preferred solution.


David-Hillel RubenSenior Lecturer in PhilosophyThe London School of Economics and Political Science

London and New York

First published 1990 by Routledge

This edition published in the Taylor & Francis e-Library, 2004.

First published in paperback in 1992by Routledge

11 New Fetter Lane, London EC4P 4EE

Simultaneously published in the USA and Canadaby Routledge

a division of Routledge, Chapman and Hall, Inc.29 West 35th Street, New York, NY 10001

1990, 1992 David-Hillel Ruben

All rights reserved. No part of this book may be reprintedor reproduced or utilized in any form or by any electronic,

mechanical, or other means, now known or hereafterinvented, including photocopying and recording, or in anyinformation storage or retrieval system, without permission

in writing from the publishers.

British Library Cataloguing in Publication Data

Ruben, David-Hillel.Explaining explanation.(The problems of philosophy)

1. ExplanationI. Title II. Series

160

Library of Congress Cataloging in Publication Data

also available

ISBN 0-203-16930-1 Master e-book ISBN

ISBN 0-203-26475-4 (Adobe eReader Format)ISBN 0-415-08765-1 (Print Edition)

For my parentsBlair S. Ruben

Sylvia Ginsberg Ruben

Hear, my son, the instruction ofthy father,

And forsake not the teaching of thy mother

vii

Contents

Preface and Acknowledgements ix

I Getting our Bearings 1Some explanations 3Process and product 6The methodology of explaining explanation 9Restricting the scope of the analysis 15Scientific and ordinary explanation 16Partial and full explanation 19Bad explanations and no explanations 21Some terminology 23Theories of explanation 25Dispensing with contrastives 39

II Plato on Explanation 45The Phaedo 47Platonic explanantia and explananda 51Problems for the physical explainers 53Some terminology 56Platos Principles 58Platos (PP2) 64Platos (PP1) 66The Theaetetus 72Summary 75

III Aristotle on Explanation 77The doctrine of the four causes 77Does Aristotle have a general account of explanation? 83Incidental and per se causes 87Necessitation and laws in explanation 93Aristotle on scientific explanation 95Aristotles demonstrations 101Summary 108

viii


IV Mill and Hempel on Explanation 110Mills account: laws of coexistence and succession 115Mills account: the symmetry thesis 123Mill on ultimate explanations 125Mill on deduction and explanation 129Hempels account of scientific explanation 138Hempels methodology 141Hempel on the symmetry thesis 145Hempel on inductive-statistical explanation 149Hempel on epistemic ambiguity 152Summary 154

V The Ontology of Explanation 155Explanation and epistemology 155Extensionality and the slingshot 156The relata of the explanation relation 160Explaining facts 168The non-extensionality of facts 171Facts: worldly or wordy? 172The co-typical predicate extensionality of facts 173The name transparency of facts 177

VI Arguments, Laws, and Explanation 181The standard counterexamples: irrelevance 183The standard counterexamples: symmetry 191A proposed cure and its problems: the causal condition 192Generalizations get their revenge 205

VII A Realist Theory of Explanation 209Are all singular explanations causal explanations? 211What would make an explanation non-causal? 217Identity and explanation 218Are there other non-causal singular explanations? 222Disposition explanations 225Again: determinative, high and low dependencyexplanations 230

Notes 234

Bibliography 256

Name Index 262

Subject Index 264

ix

Preface and Acknowledgements

This book is written in the conviction that the concept of explanationshould not be exclusively hijacked by the philosophy of the naturalsciences. As I repeat often in the following, like knowledge,explanation is an epistemic concept, and therefore has a philosophicallocation within the theory of knowledge, widely conceived. Thephilosophy of science has great relevance for a theory of explanation,just as it does for discussions of knowledge. But it is not the soleproprietor of either concept.

It is a pleasure to acknowledge the many debts I have incurred in thewriting of this book. A Nuffield Foundation Fellowship for the period ofJanuary-April 1988, and a grant from the Suntory-Toyota InternationalCentre for Economics and Related Disciplines which funded a period ofleave from January to April 1989, were both invaluable in providing mewith time to write the book. I am extremely grateful for their help, andwish to thank them publicly for it. In addition to funding leave, both alsoprovided me with a small sum of money for the purchase of books, whichI found immensely helpful in ensuring that I had all that I needed to workand write efficiently.

My intellectual debts are many. Peter Milne read ancestors of chaptersII and V, and generously helped me with some of the more technical partsof chapter II. Jonathan Barnes read and commented on an ancestor ofchapter III. Graham Macdonald and Mark Sainsbury commented on, andmade many helpful suggestions for the improvement of, early versions ofchapters I and V. Peter Lipton provided me with many fruitful discussionsof explanation generally, and also commented in detail on chapters I, IV,V, and VI. Gary Clarke and Paul Noordhof read over the whole manuscriptin an almost final form; both made many useful suggestions throughoutthe manuscript, and saved me from numerous errors. It would, perhaps,

x


not be inappropriate in a paragraph on intellectual debts to mention mydeep respect for the literature I discuss (even when I argue with it), andthe extent to which I have learned and profited from it. This is obvious inthe case of the historical figures, but, obvious or not, it is similarly thecase with the contemporary literature on explanation which I cite (andsome which I do not have space or time to cite). Whatever I have beenable to discern has only been by standing on their shoulders. I have learneda great deal from everything I have read, but perhaps the greatest singleinfluence on my thinking has been the work of Peter Achinstein.

It is so self-evident that only the writer himself can be responsible forany remaining mistakes and errors, that writers often attempt to discoverincreasingly novel or amusing ways in which to say this. I shall not try; Iknow that the philosophical influence of all these people made the. bookmuch better than it would otherwise have been, and it cannot be the faultof any of them that they were unable to detect all of the errors I made, orunable to ensure that I was capable of making good every error they pointedout to me.

In each of my previously published books and articles, I have thankedMark Sainsbury for philosophical conversation, whichall too oftenhas been one-sided, with him as teacher and me as pupil. I, like mostphilosophers, cannot work without constant philosophical discussion, andI have him principally to thank for bringing it about that I live in aphilosophically acceptable environment.

The strategy of the book is almost, but not quite, straightforward. Inthe historical portion of the book, chapters II, III, and IV, I discuss thetheories of explanation of Plato, Aristotle, John Stuart Mill, and CarlHempel. Although there is little explicit philosophical work on explanationbetween Aristotle and Milla gap of over two thousand yearsthere ismuch implicit in the writings of Bacon, Berkeley, and many otherphilosophers that is relevant to explanation, but which considerations ofspace have forced me to neglect. I discuss and state my view on someissues as I move through these historical chapters, but in the main I reservechapters V, VI, and VII for the elaboration of my own views on explanation.

I have not yet mentioned the purpose of chapter I. The placement ofthis chapter has given me some pause. As I began my discussions of thehistorical figures, I found myself in constant need of a technical vocabularywith which to make the issues they treat clear and precise. I thereforedecided to devote an opening, non-historical chapter to questions ofterminology, and to classification of kinds of theories of explanation. Thedanger in this strategy is that the reader will not really see the point of

xi

Preface and Acknowledgements

chapter I, until much later in the book. I might suggest, for readers whobegin to tire of chapter I, that they proceed to chapter II, and return tochapter I only when they find a need for a discussion of the issues it dealswith. I decided not to relocate chapter I to a later position in the book, butto leave it in place, allowing readers to decide when the reading of thechapter would be appropriate.

David-Hillel RubenLondon, 1990

1

CHAPTER I

Getting our Bearings

The series in which this book is appearing is called The Problems ofPhilosophy: Their Past and Present; this volume, since it is about theconcept of explanation, discusses some of the philosophical problemsabout explanation, as they arise in the writings of past philosophers.

It is necessary to introduce certain distinctions, and settle a fewsubstantive matters, before beginning the discussion of explanation inthe succeeding chapters. One possible consequence of this approach isthat readers will not always see the motive for the distinction or decision;I can only ask them to be patient, for the discussion in the followingchapters returns to these issues time and time again; I engage in a separate,introductory treatment of these common and recurring themes, rather thanweave them into the body of the ensuing text. But perhaps a map ofwhat this chapter contains will help.

First, it is essential to identify more precisely the concept I shall bediscussing. Which concept does the term explanation designate? Theliterature is somewhat remiss in this respect. Usually, the authorpresupposes that the audience will have no difficulty in identifying whichconcept it is, about which the author wishes to raise certain problems.This may be an acceptable presupposition in discussions of concepts likecausation and knowledge. It does not seem to me to be an acceptablepresupposition in the case of explanation (or, for that matter, in the caseof the concept of a person). Hence, it is not a presupposition that I shallmake. One of my main motives, in the sections entitled Someexplanations, Process and product, Restricting the scope of theanalysis, Scientific and ordinary explanation, Partial and fullexplanation, and Bad explanations and no explanations, is to specify

2


as precisely as I can which concept it is that I shall be discussing, bydistinguishing it from others with which it might easily be confused.

I also use this chapter to introduce some terminology and draw variousdistinctions that I need for my later discussion. One needs a perspicuousterminology in which to raise the central questions properly. Thephilosophical implications (for surely there are such) of choice ofterminology are not always apparent to the writer; it is therefore especiallyincumbent on the writer to be as clear about this as possible, so that othersmay be able to see those implicit and unnoticed ramifications, which mayescape notice. Introduction of terminology, and drawing of pertinentdistinctions, occur in the sections mentioned above, but also in the sectionsentitled Some terminology, Theories of explanation, and Dispensingwith contrastives. In the last section, Dispensing with contrastives, Idiscuss a certain view about what it is that one explains in an explanation.I discuss explanation in a traditional terminology, which the contrastiveview seeks to overturn; hence, my motive for taking on the contrastiveview in this introductory chapter.

The section on theories of explanation is the longest in the chapter. Itoffers a typology by which to identify and describe specific theories ofexplanation. In order to help the reader see what is going on in that section,I introduce its own map at the beginning of the section. But I wouldstress that the motive for drawing the distinctions in the way I do can onlyemerge in the subsequent chapters, in which the distinctions are appliedto specific theories.

Many writers on explanation fail to make the ground rules of thediscussion of explanation at all clear. One is presented, in the literatureon explanation, with many extremely plausible but competing accountsof explanation. In virtue of what features is one account better thananother? What acceptance tests should an account of explanation beprepared to meet? I address this question in the section entitled Themethodology of explaining explanation.

Throughout the book, I make use of a contrast between epistemologyand metaphysics, and the various concepts whose analyses belong to oneor the other of these two branches of philosophy. For example, a themethat recurs throughout the book is that explanation is an epistemologicalconcept, but one which requires a metaphysical backing.

I am content for this contrast to be understood in a rough and readyway. Metaphysics is the study of what there is, and what it is like,quite apart from questions about our knowledge of these matters.Typical metaphysical questions include: are there universals?; what is

3


an event?; does every event have a cause?; is the concept of causationa deterministic concept? Epistemology is the study of knowledge,belief, reasons, and evidence. Typical epistemological questionsinclude: must all beliefs be justified by other beliefs?; is all knowledgecertain?; which, if any, non-deductive arguments with true premissesprovide reasons for belief in their conclusions? I am quite prepared toadmit that there are some concepts which do not fit easily into onecategory rather than the other (perhaps the concepts of truth and offact are examples), but this does not, I think, detract from the usefulnessof the distinction.

I do occasionally refer to the views of Carl Hempel throughout thischapter. I discuss Hempel fully in chapter IV. However, since his writingson explanation have proved to be so central to contemporary discussions,reference to him here is intended to be merely a useful illustration ofwhatever specific question is at hand.

Some explanations

Giving explanations is a common activity, engaged in by layman andscientific specialist alike. Most books about explanations begin bygiving examples of scientific explanation. The following arerepresentative cases of the sort of explanations that scientists offer:

(a) Two kilograms of copper at 60 degrees C are placed in three kilogramsof water at 20 degrees C. After a while, water and copper reach thesame equilibrium temperature, 22.5 degrees C, and then cool downtogether to the temperature of the surrounding atmosphere. Why isthe equilibrium temperature 22.5 degrees C? Since the specific heatsof water and copper are 1 and 0.1 respectively, and since theconservation of energy requires that the total amount of heat be neitherincreased or diminished, the heat loss of copper, namely, 0.1x 2x(60-T) must be the same as the heat gain of water, namely, 1x3x(T-20),where T is the final equilibrium temperature. And this yields 22.5degrees C as the value of T.

(b) Two nerve impulses, I1 and I2, in close physical proximity in a neuron,arrive within 0.3 milliseconds of each other at the synapse of thatneuron. Neither has a local potential quite strong enough to fire acertain adjacent dendrite. Nevertheless, the dendrite in question fired.Why? Because the local potentials of I1 and I2 have summated to adegree high enough to evoke a spike potential in the adjacent dendrite,a phenomenon that will occur in the described circumstances provided

4


that the arrival time of the distinct nerve impulses does not exceed0.5 milliseconds.

(c) It is observed that certain human beings suffering from extreme fatigueand lengthy food deprivation show little or no desire to eat whenpresented with food. The explanation for this is that extreme fatigueinhibits the rhythmic contractions in the duodenum that initiate bloodchemistry changes which in turn trip off the central mechanismsleading to eating behaviour.1

The three examples of scientific explanation cited above are pickedalmost at random from many equally good ones with which the readerwould be presented in any adequate book on the concept of scientificexplanation.

The third example is an example of the explanation of a generalization(well, almost a generalization; the point is that it does not concern a specificor particular case): certain human beings suffering from extreme fatigueand lengthy food deprivation show little or no desire to eat when presentedwith food. The first and second examples are examples of the explanationof particular cases: two specific nerve impulses which fire a dendrite, anda specific sample of copper weighing two kilograms placed in a containerwith three kilograms of water. Carl Hempel cites a particular caseexplanation in the opening pages of his Aspects of Scientific Explanation:

John Dewey describes a phenomenon he observed one day whilewashing dishes. Having removed some glass tumblers from the hotsuds and placed them upside down on a plate, he noticed that soapbubbles emerged from under the tumblers rims, grew for a while, cameto a standstill and finally receded into the tumblers. Why did thishappen? Dewey outlines an explanation.2

I have relatively little to say in this book about the explanation of lawsand generalizations; I concentrate on what I call singular explanation.Some writers, for instance Michael Friedman, have claimed thatexplanation in science is almost always explanation of laws:

what is explained is a general regularity or pattern ofbehavioura law if you like. Although most of thephilosophical literature deals with the explanation of particularevents, the type of explanation illustrated by the account aboveseems much more typical of the physical sciences. Explanations

5


of particular events are comparatively rarefound only perhapsin geology and astronomy.3

I think Friedmans claim is exaggerated. Two of the cases which Icited above, which are taken from scientific journals, are examplesof the explanation of particular events (and neither is from geologyof astronomy). It is true that science oftenperhaps alwayshasexplanatory interest in particular cases only in so far as they areexamples of a general sort. It would not really have mattered if theabove explanations had been of two similar impulses firing a similardendrite, or of a similar sample of copper placed in a similar amountof water. As Raimo Tuomela says: Singular facts, events, etc., arenot per se of any interest to at least pure science. All interest in themis ultimately interest in their being instantiations of some universalrather than another, for indeed there are no bare particulars.4

This may have something to do with the nature of explanation itself.Whenever a particular case is explained, perhaps the same explanationcould be given for any relevantly similar example, and so the explanatoryinterest is never in the particular case as such, but only in it in so far as itis a particular case of a general sort. But this, if true, is not the same thingas having little or no interest in particular cases. In any event, if my neglectof the explanation of laws is a weakness of the book, at least I can claimthat what I have to say is consistent with the truth about the explanationof laws or generalizations, whatever it may be.

A theory of explanation does not only address itself to cases of explanationin science. It must address itself to other cases as well, in which non-specialistsexplain things to one another. I am not thinking of explanations of humanaction, about which I will have very little to say in this book. Rather, I have inmind the perfectly acceptable ordinary explanations we are able to give oneanother of natural occurrences: the onset of warm weather explains the meltingof the snow; overexposure to the sun explains my painful burn; my match litbecause I struck it. The person who explains the melting of the snow by theonset of warm weather may not be able to explain how or why higher airtemperature causes the snow to melt; for this latter, they may need amicrotheory which only scientific specialists possess. But inability to explainhow or why an air temperature increase leads to the melting of the snow doesnot imply inability to explain the snows melting on the basis of an increasein the air temperature. Nearly everyone, whether or not they have a degree ina natural science, knows that the snow melts because spring has come.

6


The analysis of explanation, then, belongs to general epistemology, inthe same way as the analysis of knowledge does, and not just to thephilosophy of science, narrowly conceived. Scientific explanation, likescientific knowledge, has a special importance and pride of place in ageneral theory of knowledge. But just as there is more to knowledge thanscientific knowledge, so too there is more to explanation than scientificexplanation. The knowledge that I now have that I am sitting at my deskand writing is not scientific knowledge. The explanation that I can give ofthe snows melting in terms of the warmer weather is not scientificexplanation. Ordinary explanations, like ordinary knowledge, are notimpervious to error, and it may sometimes happen that science overturnswhat we wrongly took to be an example of ordinary knowledge or ofacceptable ordinary explanation. But, when not so overturned, suchordinary explanation or knowledge is not, per se, scientific explanationor knowledge. I do not intend these introductory remarks to beg anyquestions about the nature of the distinction between ordinary and scientificexplanation, nor to suggest that there is some hard and fast contrast betweenthem. I deal with these issues in the course of the chapter. Rather, theseremarks are intended only to serve as a reminder about the scope of ourtopic. Far too many discussions of explanation assume that what can besaid about scientific explanation exhausts what of interest there is thatcan be said about explanation tout court, and this is, in my view, simplynot so.

For the present, I shall move rather cavalierly between explanationand scientific explanation. I ask the readers temporary indulgence. Ideal with this (alleged) distinction later in the chapter.

Process and product

Explanation itself is susceptible to a well-known process-productambiguity, as are many other words ending in -ion. In the process-product shift, a word, often one ending in -ion or -tion, maysignify an activity or its result.5 A simple example is this: I saw thedestruction at Rotterdam. The sentence might mean either that I sawthe act of Rotterdam being destroyed or that I saw the results of suchan act.

Explanation is ambiguous in the same way as destruction. AsBromberger points out, in one sense

7


an explanation may be something about which it makes sense toask: How long did it take? Was it interrupted at any point? Whogave it? When? Where? What were the exact words used? Forwhose benefit was it given?6

On the other hand, an explanation may be something about whichnone of [the previous] questions make sense, but about which it makessense to ask: Does anyone know it? Who thought of it first? Is it verycomplicated?

The linguistic evidence points to two different senses of explanation.The first suggested by Brombergers evidence is the process or act sense;the second, the product sense. Other examples of words which have thisambiguity range from philosophically uninteresting ones like simulationand destruction (Clark and Welshs example) to ones which raisephilosophical issues similar to those raised by explanation: prediction,deduction, derivation, proposition, argument, statement, andanalysis (although the last three do not end in -ion).

So, in speaking of an explanation, one might be referring to an act ofexplaining, or to the product of such an act. How are these two sensesrelated? There seem to be just four possibilities:

(1) The idea of an explanatory act can only be analysed by using theidea of an explanatory product, but not vice versa.

(2) The idea of an explanatory product can only be analysed by usingthe idea of an explanatory act, but not vice versa.

(3) The ideas of explanatory act and explanatory product mutually dependon one another.

(4) The ideas of explanatory act and explanatory product are independentof one another.

Most of the literature on explanation, and certainly the four writerson explanation whom I shall be discussing, Plato, Aristotle, Mill,and Hempel, were interested only in the idea of an explanatoryproduct. They believed (and I agree with them) that an explanatoryproduct can be characterized solely in terms of the kind ofinformation it conveys, no reference to the act of explaining beingrequired. Hence, each would have rejected (2) and (3). Their questionwas this: what information has to be conveyed in order to haveexplained something?

One recent writer, Peter Achinstein, has advanced (2).7 If (2) weretrue, then the idea of an explanatory act would have a far more central

8


position in the analysis of explanation than it has previously been given.According to Achinstein, an explanatory product is neither just anargument (the Hempelian view) of a certain sort nor just a proposition ofa specific kind, nor any other entity which can be characterized solely interms of its syntactic form and/or the type of information that it conveys.Rather, according to him, an explanatory product is an ordered pair, inpart consisting of a proposition, but also including an explaining act type(e.g. the type, explaining that such-and-such). For example, onAchinsteins view, the explanation of why Nero fiddled might be theordered pair, {Nero fiddled because he was happy seeing Rome burn;the act type, explaining why Nero fiddled}.

Why does Achinstein think that an explanatory product cannot becharacterized solely in terms of its information content? His argumentrests on the uncontroversial fact that the same information content mightbe conveyed by both an act of explaining and an act of another type,8 e.g.an act of criticizing. For instance, in saying that Nero fiddled because hewas happy seeing Rome burn, Achinstein claims that I could be eithercriticizing Nero or explaining his action (Achinstein 1983:889).Achinstein reasons that since the same information content can beconveyed in two different kinds of acts, and since no product of anexplaining act could be identical with, for example, the product of acriticizing act, the explanation product (and the criticism product) mustbe more than just the information conveyed.

But why cant the product of an act of criticizing and an act of explainingbe identical? Achinstein relies on the following sorts of principles to showthat they cannot be:

(5) The product of Ss act is an explanation only if S explained.(6) The product of Ss act is a criticism only if S criticized.

These principles will lead to the conclusion that Achinstein wants.If I am explaining Neros actions but not criticizing them, and youare criticizing them but not explaining them, then the explanationproduct of my act cannot be identical with any criticism productand the criticism product of your act cannot be identical with anyexplanation product. If explanation and criticism products are to bedistinguished in this case, the products ought to be distinguishedeven in the case in which one person is engaging in two or moreacts9 at one and the same time. Each of the acts will have its owninternal product.

9


But what reason is there to think that (5), (6), and other analogousprinciples, which claim that a necessary condition for something to be apersons product10 of a certain kind is that he has actually produced it inan act of that kind, are true? These principles simply presume what theyare used to prove. I can see no good reason to deny that objectivelyspeaking, quite apart from whatever intention you (or anyone else) mayhave had in acting, the information you impart in criticizing (explaining)Nero may also be an explanation (criticism) of what he did, in the productsense. One can, in criticizing Nero, convey information which is also anexplanation (in the product sense) of why he fiddled, whether the criticizeror indeed anyone else has ever engaged in an act of explaining what hedid. There can be explanations (in the product sense), even if no one hasever explained anything; (5) and (6) are false.

Explanatory products can be fully characterized in terms of theirinformation content independently of explanatory acts, so (2) and (3) arefalse. (I do not wish to pronounce on the choice between (1) and (4).) Ofcourse, we may tend to call such information an explanation (in theproduct sense) as opposed to a criticism or an argument only if it figuresas the product of an explaining act. But that gives us no more reason todeny that an explanation product may be the same as a criticism productthan there is to deny that the Morning Star=the Evening Star on the groundsthat we tend to call the heavenly body the latter only when it appears inthe evening, and the former only when it appears in the morning.

The methodology of explaining explanation

The title of this book is Explaining Explanation. The suspicious mightthink that there is something self-defeating in such a title. How, onemight ask, if one were genuinely in need of enlightenment about theconcept of explanation, could one undertake to explain whatexplanation is? Would it not be rather like trying to pull oneself upby ones own bootstraps?

Of course, there is no real difficulty here. In offering a philosophicalanalysis of any concept, one must attribute to oneself an (at least partial)implicit understanding of that concept, which the analysis is attemptingto make explicit. Some sophistication or other of this basic idea of what itis to offer an analysis is necessary if one is to escape the paradox ofanalysis. The alleged paradox asserts: if one knew what was involved in

10


the concept, one would not need the analysis; if one did not know whatwas involved in the concept, no analysis could be forthcoming. The escapeis through some implicit/explicit distinction. One can know implicitly,but need the analysis to make the knowledge explicit.

Moreover, there is a second reason why explaining explanation offersno difficulty. What we are explaining is the concept of explanation, asemployed not only in science, but also in ordinary life. But the explainingthat we are undertaking is specifically philosophical explication or analysis(I use these terms interchangeably) of a concept. Carl Hempel, for example,repeats in several passages that what he is doing is offering an explication,and that the purpose of an explication or analysis is to lay bare the logicalstructure of the concept. Hempel often speaks of the analysis of a conceptas a model, as he does when he says that there are three models ofexplanation.11 On his view, there are really three distinct concepts ofexplanation, and the three models make clear in what ways the threediffer.

The literature abounds with competing and incompatible explications,or analyses, of explanation. Optimally, we should like to be able to chooserationally one from amongst them. We need to know, then, what constraintsthere are on such a choice. Should the best analysis fit the way in whichwe ordinarily use the term explanation? Should it, rather, meet somemore technical requirements of science or philosophy? I wish below todraw a contrast between two different ways of answering these questions.

I intend the following general remarks on methodology to be as anodyneand uncontroversial as possible, for they are not intended as an excursusinto the philosophy of language or philosophical logic. I do not think oneneeds to take them as a serious contribution to the understanding of thenature of concepts. I will also assume in the discussion the view of analysisor explication adopted explicitly by Hempel, when he said that he isengaged in laying bare the logical structure of a concept. Although Hempeldoes not say so, it would seem to follow from this view that the truths soexposed about a concept have the status of analytic or necessary truths. Ido not here distinguish between analyticity and necessity, for nothing ofimportance for my discussion hangs on that distinction.

Perhaps a brief comparison with the philosophical literature on theanalysis of knowledge will help us to understand the idea of the twodifferent ways of judging competing analyses. I gesture to this otherliterature, only as a way of drawing the contrast in philosophical methodthat I will then apply to the case of the analysis of explanation. Thesedifferent ways of proceeding philosophically, whether in discussing

11


knowledge or explanation, arise out of different traditions of what it is todo philosophy.

First, some discussions of knowledge proceed in this way. Variouscomplicated situations are described, for instance a situation in which aperson has justified true belief but there is no causal connection betweenthe fact the belief is about and the belief itself. We are then asked whetherwe would apply knows in such a situation. So, all three conditions forknowledgeare fulfilled; but we still do not want to say that S knows.Surely we do not want to say that his friends wild guess endows S withknowledge.12

The idea is that our analysis of knowledge should capture all and only,or (in a weaker and more plausible version) most of, those situations inwhich we would prephilosophically be prepared to use the term knows.I am, of course, thinking of the vast literature inspired by Gettiers famousarticle.13 For better or worse, I call this method the language usersapproach. Notice that this language users approach might not be weddedto everyones use of the word at all times (it is important to see that thisapproach need not be wedded to the idea of ordinary language). It is opento an exponent of this view to designate some subset of users of the wordas having a special status. For instance, the philosopher of knowledgemight only be interested in how scientists employ the concept ofknowledge, and perhaps only while they are engaged in some specificscientific activity. I still think of this as the same view, but with the classof users cut down in size and scope.

Michael Friedman adopts this language users approach in his accountof explanation:

most, if not all, scientific the orie s tha t we all co ns explanatoryshould come out as such according to our theory. Although it isunreasonable to demand that a philosophical account of explanationshould show that every theory that has ever been thought to beexplanatory really is explanatory, it must at least square with most ofthe important, central cases.14

Friedman does not say why his requirement is plausible. Isnt itlogically possible that all or most of the central and important casesof theories we thought were explanatory fail really to be so? PerhapsFriedman has in mind here some version of the paradigm caseargument; if so, the prospects for his view seem dim.15

12


It is more difficult to give a succinct general characterization of thealternative method which I wish to describe. I call it the technicalapproach. In one way or another, it dispenses with such reliance on the.way in which terms are actually used or employed. As far as the analysisof knowledge is concerned, a good example of this approach is KarlPoppers Epistemology Without a Knowing Subject: scientificknowledge simply is not knowledge in the sense of the ordinary usage ofthe words I know.

ordinary languagehas no separate terms fo corresponding twosenses of knowMy quoting The Oxford English Dictionary shouldnot be interpreted as either a concession to language analysis or as anattempt to appease its adherents. It is not quoted in an attempt to provethat ordinary usage covers knowledge in the objective sense of mythird world.16

Poppers characterization of knowledge must meet some constraints,and his article goes on to specify just what they are. But whateverthey are, they do not include fit with the way in which we (or evenjust scientists) employ or use the term knows.

My names for these two positions, the language users approach andthe technical approach, are not especially happy ones, but they do at leastsuggest the sort of position intended. Different philosophical orientationshave tended to favour one or the other of these positions, but of coursethese are ideal types, and the actual practice of many philosophers ismore complicated, combining elements of both of these approaches, andperhaps others besides.

Even on the language users approach, one might regard usage as vague,ambiguous, imprecise, even inconsistent or incoherent; the philosophermay say that there is no single concept that is expressed by all of theordinary uses of some term. He then may single out a subset of those usesas a way to disambiguate and to focus on one concept at a time. Or theconcept as used may be vague (it is not strictly true that one can speak ofthe concept in such a circumstance); there may be general agreementabout the paradigm cases, but dispute about cases in the conceptspenumbra. The language users approach should permit us to depart fromordinary usage at least to the extent of eliminating vagueness in a conceptsapplication and disambiguating. Lets call this tidying up a discourse.But, in all versions of this approach, actual language-use is where one atleast begins ones analysis.

13


The other approach I called the technical approach. A philosopherengaged on some project might eschew interest in the concepts used bythe speakers of a language. The philosopher might see as part of his taskthe introduction of some quite novel concept, whose criteria are given bystipulative specification. Examples of this include some of the technicalconcepts of philosophy: sense data, the distinction between essence andaccident, the ideas of a metalanguage, and material implication. The greatphilosophical systems, e.g. the Platonic, Kantian, and Hegelianphilosophies, provide examples of this technical concept introduction:Forms, the distinction between reason and understanding, the synthesisof the understanding, noumena and phenomena, transcendental arguments,the Absolute in and for itself. The philosopher might think that such aconcept plays an important role in coming to understand something thatwe simply failed to understand before. I call this simple introduction.

The above examples of the technical approach are cases ofstraightforward concept introduction. But there are other cases in whichthe introduced concept is intended to replace or improve upon one alreadyin use by the common man. Humes reformed concept of the self,Berkeleys idea of a physical object which excludes the commitment tounperceived existence, Hobbess redefinition of desire and aversion interms of internal motions, and the idea of truth-in-a-language, are examplesof concept replacement. Many scientific reductions involve conceptreplacement in the reduced science; arguably, the pre-reduction conceptof water is not the same as the concept of water after its identificationwith H20. The latter would then be a replacement for the former.

Suppose a philosopher practising this technical approach decides thatthere are good reasons for the introduction of a new concept of X, toreplace the old one. The new replacing and old replaced concepts willsometimes have very similar extensions, and their analyses (or, models)may have many features in common. But this is hardly essential. Replacingconcepts might differ dramatically in intension from the concepts thatthey replace.17 Moreover, the new and old concepts of X may differ inextension. We might even come to believe that nothing correctly calledX before, when the old concept was in use, can be correctly so-callednow that it is the new concept that is in service, and vice versa.

I have stressed the intensional and extensional discontinuities theremight be between replacing and replaced concepts of something. Butsurely there are limits here. There must be some difference between (a)replacing the old concept of X with the new concept of X; (b) eliminatingthe concept of X and simply introducing the concept of Y, as two separate

14


exercises in improving our discourse. How could we account for thisdifference if not by introducing some sort of continuity between the oldand new concepts?

Indeed, it is true that there must be some sort of continuity. There is,by and large, a point in having the concepts we do. For example, at leastpart of the point and purpose of explanation is that we should come tounderstand why things happen.18 That is the function that explanation hasfor us. If a replacing concept of explanation is a replacement for thestandard or ordinary concept of explanation, it surely must serve at leastthis function. The requisite continuity between the old and new conceptof X might be provided by continuity of function.

Some philosophers may believe that these functional facts about aconcept have no place in its logical analysis. They will say, in the caseof explanation, that although it is true that explanations do or shouldlead us to understand, that this is so is not a logically or conceptuallynecessary truth about explanation. Hempel, for example, says that suchexpressions as realm of understanding and comprehensible do notbelong to the vocabulary of logic, for they refer to psychological orpragmatic aspects of explanation.19 These facts about what explanationsdo for us have, on his account, no place within the analysis of explanationitself. For such philosophers, there could be a complete intensional andextensional discontinuity between the old and new concepts ofexplanation, with only the sameness of contingent functional factslinking the two concepts as two concepts of explanation.

Other philosophers will find room for these functional facts within theanalysis of the concept.20 The analysis of explanation will include somemention of understanding. For these philosophers, there will after all haveto be some, at least minimal, intensional continuity between replaced andreplacing concepts of explanation.

I have great sympathy for the technical approach, rather than thelanguage users approach in any of its possible refinements. However, ifthe technical approach is adopted, one needs to consider arguments whichattempt to justify the new replacing concept one has introduced. Manynew concepts might be introduced, which could be said to have the samepoint as the old replaced one. How can we justify one candidate over theothers as the replacing concept, if language use does not constrain thatchoice? How can we show that the replacing concept we select is not justarbitrary, ad hoc?

Suppose concepts a, bn are all put forward by different philosophersas competing new replacement concepts of explanation. Each might

15


plausibly be thought of as a concept of explanation. For each, itsphilosophical champion can produce a set of necessary or analytic truths.That by itself is wholly uninteresting. Which of a, b, cn is the bestreplacement for the old concept of explanation? It is only when we cananswer that question that we will know which set of analytic truths hasany real claim to be of interest to us, and what it is that we are trying to dowhen we offer an analysis of explanation. This is an issue which I willwant to raise when I look at Aristotle, Mill, and Hempel, and which willprovide a thread of continuity that runs throughout the book.

Restricting the scope of the analysis

The Hempelian models are not intended as models of all explanations.Hempel contrasts the cases of explanation covered by his models ofscientific explanation with others in which we do not explain whysuch-and-such or that such-and-such: explaining the rules of acontest, explaining the meaning of a cuneiform inscription or of acomplex legal clause or of a passage in a symbolist poem, explaininghow to bake a Sacher torte or how to repair a radio (Hempel1965:41213). In the cases of explaining the meaning of something,the explanandum will be specified by means of a nounphrasewhereas explanations of the kind we have beenconsideringare characterized by means of a sentence (Hempel1965:414). Hempel would consider none of these above mentionedsorts of explanation as scientific in his sense, and none constitutes areasonable objection to his account of explanation:

Similarly, to put forward the covering-law models of scientificexplanation is not to deny that there are other contexts in whichwe speak of explanation, nor is it to assert that the correspondinguses of the word explain conform to one or another of ourmodels. Obviously, those models are not intended to reflect thevarious senses of explain that are involved. Hence to deplore,as one critic does, the hopelessness of the deductive-nomological model on the ground that it does not fit the case ofexplaining or understanding the rules of Hanoverian successionis simply to miss the intent of our model.

(Hempel 1965:41213)

16


Hempel indicates two ways by which to delimit the explanations forwhich he seeks to offer an analysis. The first is grammatical: explanations of the kind we have been considering are concernedwith [whatever] is properly characterized by means of a sentence(Hempel 1965:414). Elsewhere, he speaks, rather circularly, of theexplanations in which he is interested as being answers toexplanation-seeking why-questions (Hempel 1965:412). Fully andcompletely explaining how to ride a bike is not a case of explanationto which Hempel would consider his models of scientific explanationappropriate; it fails both the grammatical and the why-question tests.In chapter III, I return to the question of the adequacy of these twoways of characterizing the subset of explanations to which Hempelrestricts his analysis.

Since Hempel, in the above quotation, speaks of the various senses ofexplain, he seems to commit himself to the thesis that explain inexplain that p, explain how, and explain the meaning is ambiguous.That thesis seems to me dubious, but we do not need to decide the matterone way or the other, in order to delimit the instances of explain inwhich Hempel is interested.

Scientific and ordinary explanation

As my opening remarks suggested, there are, or are thought to be,such things as scientific explanations. The contrast is usually withordinary explanations. What does this contrast come to? Is scientificexplanation anything more than a pleonasm for explanation?

There are at least two possible senses of scientific explanation. In thefirst sense, it refers to explanations which are actually given in science.As we shall see, this is not the sense of the expression in which Hempel isprimarily interested. In the second sense, the meaning of scientificexplanation is commendatory, or honorific, in some way. In any event,in this second sense, it is an open question whether any of the explanationsactually given in science are scientific explanations at all.

There is without doubt a distinction between ordinary explanationsand scientific explanations in the first sense, since it is simply a fact thatsome explanations are given in the course of lifes ordinary affairs andothers are given by scientists when they do science. But Hempel usesscientific explanation in the second sense. Consequently, the question I

17


address in this section is whether there is a distinction between ordinaryexplanations (and also scientific explanations in the first sense) on theone hand and scientific explanations in this second sense on the other.

In my view, the only distinction that can usefully be drawn is thatbetween full and partial explanations, and the distinction between scientific(in the second sense) and ordinary explanations is either that distinctionor no distinction at all. As I indicated before, although I discuss Hempelsviews on explanation fully in a later chapter, I use him here as a way ofsharpening the issue (and in this case, actually stating my own position).

To begin with, Hempel does not think of scientific explanations asexplanations actually given by scientists: these models are not meant todescribe how working scientists actually formulate their explanatoryaccounts.21 The practising scientist may use explanation in as loose orvague a way as the ordinary man on the street. What the scientist calls anexplanation, and indeed his actual explanatory practices too, how heactually goes about explaining things, may fall woefully short of whatHempel requires of an explanation. Actual explanations in science maysuffer from the same deficiencies as do explanations offered by the non-scientist on the Clapham bus.

Perhaps, then, the term scientific explanation is meant to conjure upthe fact that there is a goal or ideal of precision and completeness,explicated by Hempels models, which explanations in science can aspireto and can actually meet if so required. The construction of our modelstherefore involves some measure of abstraction and of logicalschematization (Hempel 1965:412); we have foundthat theexplanatory accounts actually formulated in science and in everydaycontextsdiverge more or less markedly from the idealized andschematized covering-law models (Hempel 1965:424). Hempel compareshis models of explanation with the ideal (this is his term)metamathematical standards of proof theory (Hempel 1965:414). So, themodels are ideals, in some sense. Actual explanations in science mayfall short of the ideal by being elliptic, incomplete, partial, or mere sketchesof an explanation. Hempel describes these various forms of incompletenessat some length (Hempel 1965:41225).

In what sense does Hempel use the terms ideal and idealized? Themodels are surely not ideals for Hempel in the sense that explainers shouldalways strive to do their best to make their explanations complete; thereis no doubt that circumstances can justify explainers in explaining onlyincompletely, by omitting information known by their audience. In normalcircumstances in which no one doubts the prevalent atmospheric

18


conditions, a scientist would be a bore if he attempted to explain the fireby adducing both the short circuit and the presence of oxygen. It is nottrue that even scientists always ought to give as full an explanation as ispossible.

Rather, the models Hempel introduces are ideals for him simply in thesense that they are complete; they specify a type of complete or fullexplanation. In fact, Hempel believes that such complete explanations?are rarely given even in science. It is possible, and it would not matter tohis argument if it were so, that no one, not even a scientist, actually everoffers such a complete and full explanation, which includes exceptionlesslaws needing no further qualification and all relevant initial conditions.Moreover, it could even be that every actual explanation ever given wasjustifiably incomplete, due to the pragmatic constraints on providingexplanations. However, and this is surely the important point for him,incomplete explanations explain only in virtue of there being suchcomplete explanations, whether or not anyone ever gives or should giveone. One might draw the necessary distinction in Kantian terms. Hempelsrequirements provide a constitutive ideal for full explanation; they arenot intended as a regulative ideal.

A consequence of this interpretation of what Hempel has in mind isthat, if these models provide an ideal or goal for explanations in science(scientific explanations in the first sense), there is no reason why theyshould not equally provide an ideal for explanations in ordinary life too.The ideal sets a standard for explanation, tout court. Indeed, Hempeldiscusses quite explicitly the application of his model to historical events,to the actions of agents, and to functional systems. In science, a scientistmight give some explanation that, because of the constraints of time orthe interests of his audience, fails to live up to Hempelian standards.However, exactly the same is true in ordinary life. We normally are happyto explain why the chicken crossed the road by saying that it wanted toget to the other side, but if required, we could impose all of the Hempelianrequirements, to obtain a full explanation of what the chicken did. Onseveral occasions, Hempel explicitly couples scientific (in the first sense)and everyday explanations together, as both being subject to the sameidealized and schematized models (Hempel 1965:4245). As he toldus above, explanations in science and everyday contexts diverge moreor less markedly from the ideals set by his models.

So a scientific explanation (in the second sense) doesnt seem to beeither an explanation actually offered in science, or an ideal appropriateonly for explanations offered in science or by scientists. The truth is that,

19


as far as Hempel is concerned, the Hempelian models of scientificexplanation, if they provide an ideal for any explanations, provide anideal for all explanations (subject only to the restriction described in thepreceding section on the range of cases for which the analysis is offered).They are models of complete explanation, in science and in ordinaryaffairs. By scientific explanation (in the second sense), Hempel meansonly a complete or full explanation, and nothing more.

I have developed my discussion of scientific explanation around remarksof Hempels. But I think that the lesson is general. Scientific explanation isan expression that repeatedly occurs in most discussions of explanation. Ifscientific explanation does not mean explanation actually offered in science,the sense of the expression is far from obvious, and needs to be made clear.Many philosophers of explanation use it merely in the sense of an ideallycomplete explanation. Much of the potentially mesmerizing mystique ofscientific explanation will vanish, if this is kept in mind.

Partial and full explanation

The key, then, to unlocking the idea of a scientific explanation (in thesecond sense) is the distinction between complete or full andincomplete or partial explanation.22 The distinction between partialand full explanations is a distinction between two different sorts ofexplanatory products; presumably, the activity of explanation-givingcan at least sometimes justify giving partial rather than full ones.

It is not possible to draw the distinction between full and partialexplanations in a neutral way, equally agreeable to all theories ofexplanation. Different theories disagree about what counts as a fullexplanation. Some will hold that explanations, as given in the ordinaryway, are full explanations in their own right; others (like Hempel) willargue that full explanations are only those which meet some ideal, rarelyif ever achieved in practice. A partial explanation is simply a fullexplanation (whatever that is) with some part of it left out. On any theoryof explanation, we sometimes do not say all that we should say if wewere explaining in full. Sometimes we assume that the audience is inpossession of facts which do not stand in need of repetition. At othertimes, our ignorance does not allow us to fill some of the explanatorygaps that we admit occur. In such cases, in which we omit information forpragmatic or epistemic reasons, we give partial explanations.

20


Partiality is sometimes related to falsity. Laws may be omitted entirelyfrom a partial explanation. Sometimes they are not omitted, but rather aregiven an incomplete formulation which ignores certain exceptions. If alaw is an exceptionless generalization, an incompletely formulated law isa generalization with exceptions, and which is therefore not strictly true.Something not strictly true is just false. On other occasions, strictly relevantinitial conditions might be too marginally relevant to the explanandumoutcome to include in the explanans, and so the explanation, in order topresent itself as if it were complete rather than only partial, may make aclosure assumption about the environment in which the outcome occurswhich is not strictly true.

Of course, whether some particular explanation is partial or not maybe contentious. Since theorists will disagree on standards for fullexplanation, they are bound to disagree about which explanations arepartial. All I assert is that every theory of explanation must draw somedistinction between full and partial explanation, and that the idea of apartial explanation is parasitic on the idea of a full one.

Recall that in the first sense of the term, scientific explanation refersto the explanations actually given in science. Most or all of theseexplanations are, like their ordinary counterparts, merely partialexplanations for Hempel. It is consistent with my interpretation of Hempelthat the way in which explanations actually given in science are partialmay generally differ from the way in which actual ordinary explanationsare partial. For example, typically ordinary explanations omit all mentionof laws, and this may not be so in at least some areas of science. Forexample, in the first example of a scientific explanation given at thebeginning of this chapter, even if it were to count as partial on somegrounds, it does mention the law of the conservation of energy. In thesecond example, although a law is not explicitly mentioned, it proffers allthe materials for the formulation of one in the concluding sentence. Ireturn to the question of the place of laws in explanation, and the idea ofa full explanation, in chapter VI.

In what follows, unless I otherwise indicate, I mean to be speakingof full explanation. If I want to speak of partial explanation, I explicitlyuse the qualifying adjective. I sometimes add full as a qualification, ifthe qualification is especially important and stands in need of emphasis.

21


Bad explanations and no explanations

Is the concept of explanation for which we are seeking an explicationthe same as the concept of a good explanation? This question is highlycontentious (e.g. it involves the distinction between semantics andpragmatics), and is inextricably bound up with other questions aboutexplanation. I will have something more to say about this in chapterV. Whatever the right answer, it is important for a philosopher to beclear about how he would answer it.

Consider the following remarks by Hilary Putnam:

Explanation is an interest-relative notionexplanationexplanationhas to be partly a pragmatic concept. To regard the pragmatics ofexplanation as no part of the concept is to abdicate the job of figuringout what makes the explanation good. More precisely: the issue is notwhether we count the pragmatic features as part of the meaningthat is a silly kind of issue in the case of such notions as explanationbut whether our theory does justice to them or relegates them to merepsychology.23

Lets call Putnam an explanatory pragmatist. I take that to meanthat what counts, for him, as a full explanation of something (and notjust as a good explanation of that thing) is audience-variant; theinterests of audiences differ, and therefore what counts as a fullexplanation differs as a function of differences in interest. Everytheorist of explanation can admit that the idea of a good explanationis audience-variant. Putnam is refusing to draw a sharp distinctionbetween explanation and good explanation, and therefore argues thatthe idea of full explanation, not just that of good explanation, isaudience-variant.

From my point of view, Putnam unjustifiably conflates the analysis ofexplanation with the pragmatics of giving explanations (or the pragmaticsof information giving, for, following David Lewis,24 I think that therequirements for explaining well are included in the requirements forconveying information well). Nor do I see why Putnam thinks this is asilly kind of issue. In this, I follow Hempel and others, in thinking thatthere is a clear distinction between the analysis of explanation and thepragmatics of explanation-giving. It will be my view that we can markout what counts as an explanation by the information content of what is

22


said. For example, on one specific sort of non-pragmatic view, a causaltheory of explanation (this is not my view, but I use it for the purpose ofillustrating the point), an explanation of an event e is always in terms ofits cause, c. Perhaps not just any true statement of the form, c is the causeof e, would be an explanation. But to try to explain e in terms of someevent that is not its cause would be, on this view, to produce no explanationat all. It would be to cite something simply irrelevant from the point ofview of explanation. Such a requirement for explanatory relevance wouldnot be audience-variant.

What a causal theorist, indeed any non-pragmatist about explanation,can concede is that how we select from the full list of explanatory relevantfeatures in order to obtain the ones required in a particular (partial)explanation we may offer is a pragmatic and audience-variant question.A partial explanation is one that omits certain relevant factors; a fullexplanation is one that includes all relevant factors. In c causally explainse, one might be citing what is in fact only part of the cause (or, the causeonly partially described, if one prefers). The cause of the matchs lightingwas its being struck. But if I say this, I assume that my audience knows orassumes that the match was dry and that oxygen was present, and that myaudience has no further interest in having the dryness of the match or thepresence of the oxygen mentioned. That is a matter of pragmatics. A partialexplanation may be good relative to one set of circumstances, but badrelative to another, in which interests, beliefs, or whatever differ.

There are additional ways in which an explanation can be bad otherthan by being partial in its selection of relevant factors in the wrong way.A full explanation can be bad too, if it conveys more information than isrequired (suppose it sends the listener to sleep). A partial explanation canalso be bad for other reasons. The cause could be described in a causallyrelevant but too general or too specific a way. In a history textbook, theoccurrence of a plague can explain a population decline, but theexplanation might be bad if it included a detailed microbiologicaldescription of the disease. Putnam himself contrasts the goodness of thesimple explanation of why a 1 inch square peg will not pass through a 1inch round hole in terms of geometry, compared with the awfulness ofthe far more complex and detailed explanation in terms of a completeenumeration of all the possible trajectories of the elementary particlesmaking up the peg, obtained by applying forces, and the fact that nocombination of them takes the peg through the round hole.25

But the non-pragmatist will insist that all of these remarks are aboutthe goodness of explanations, and relate to ill-advised choices concerning

23


selection from or description of relevant features. None of theseconcessions shows that there are no audience-invariant constraints on whatcould count as a relevant feature (for the purposes of explanation) andhence on what could count as an explanation.

In this book, I take it that the topic is the analysis or explication of theconcept of explanation. I have nothing to say directly about pragmaticissues. That one can produce, contrary to Putnams remarks, an accountof explanation that distinguishes between explanations (whether good orbad) and non-explanations on the basis of information content, is bestargued for not in the abstract, but by producing just such an account. It isthis that I hope to do in chapter VII.

Some terminology

The expressions explanans (i.e. that which does the explaining) andexplanandum (i.e. that which is explained)and their pluralsexplanantia and explanandaoccur repeatedly in this book. Theyalso occur ambiguously, and this is intentional on my part.

If explanation is a relation, one can refer to its relata, whatever they maybe, as the explanans and the explanandum. What ontological sort ofentities are these explanantia and explananda? We shall discuss this issuefully in chapter V. Obvious candidates include: phenomena, events, facts,and true propositions (or beliefs or statements). Whichever candidate isselected, we can call this non-sentence explanation. If events can explainevents, then chunks or bits of reality (like the matchs striking and the matchslighting) literally explain and are explained. Or perhaps it is the fact thatsome event occurred which explains the fact that some other event occurred.

On the other hand, another possibility is that it is true statements whichexplain true statements rather than events which explain events.(Propositions and statements are not sentences.) Even if this is so,statements explain and are explained only in virtue of the way the thingsin the world which they are about really are. If it is the statement thatthere is a short circuit that explains the statement that there is a fire, theexplanation only works in virtue of the real short circuit bringing aboutthe real fire (and although it would not be true, strictly speaking, that it isthe short circuit that explains the fire).

On one well known theory we will be examining, we explain only ifwe can deduce a sentence describing the explained phenomenon from a

24


sentence that describes the explaining reality and a lawlike generalization.In this way, then, one might also think of explanantia and explanandaas sentences (which should be sharply distinguished from statements orpropositions); e.g. the explanans entails the explanandum. We can callthis sentence explanation.26

But if there is sentence explanation, it is conceptually dependent onthe primary idea of non-sentence explanation (whether the right choiceof relata for that relation is events, or facts, or true statements). This is, Ihold, uncontroversial.27 Even a theory that seeks to analyse explanationin terms of the logical form of, and logical relations between, varioussentences is analysing the idea of explanation in the primary, non-sentencesense. The theory may attempt to reduce the idea of non-sentenceexplanation to some facts about sentences, but it does not reduce non-sentence explanation to the idea of sentence explanation.

In this intentional ambiguity of explanans and explanandum, Ifollow Hempel himself (except that he conflates sentence andstatement):

The conclusion E of the argument is a sentence describing theexplanandum-phenomenon; I will call E the explanandumsentence, or explanandum statement; the word explanandumalone will be used to refer either to the explanandum-phenomenon or the explanandum-sentence: the context will showwhich is meant.28

Context will also determine whether I am using explanans orexplanandum in the sentence or non-sentence sense.

So, I variously employ these expressions to refer to sentences,statements (or beliefs), the facts and the actual worldly events thestatements are about. The ambiguity is harmless; it often lets me say lessclumsily what would otherwise involve cumbersome expression. In anyevent, even if we wished, it would not be possible to sort out fully theambiguity, beyond what I have said here, in advance of the discussion inchapter V concerning the relata of the explanation relation.

Salmon introduces all three obvious non-sentence categories:statements, events, and facts:

It is customary, nowadays, to refer to the event-to-be-explainedas the explanandum event, and to the statement that such an event

25


has occurred as the explanandum statement. Those factsbothparticular and generalthat are invoked to provide theexplanation are known as the explanans. If we want to referspecifically to statements that express such facts, we may speakof the explanans statements. The explanans and explanandumtaken together constitute the explanation.29

What the quotation appears to say is that the explanation relation perse relates facts. The events such facts are about are the explanansevent(s) and the explanandum event. The statements which expresssuch facts are the explanans statement(s) and the explanandumstatement. On Salmons view, we explain facts, which are aboutevents, by means of making various statements. One consequence ofthis view is that there must be a significant distinction between factsand statements. In chapter V, I return to these questions, and especiallyto the theme of facts, and the role they might play in a theory ofexplanation.

Theories of explanation

Let me introduce what shall prove to be some useful distinctionsbetween different types of theories of full explanation, although theextent of that usefulness can only be apparent as those distinctionsare applied in subsequent chapters. I stress that these are theories offull explanation; I shall try and add some remarks about partialexplanation as I go along. The distinctions provide allegedly necessaryconditions for explanation, not sufficient conditions. Thus, thesedistinctions do not themselves yield specific theories of explanation,but rather permit us to catalogue specific theories as being of one oranother of the types.30

The distinctions make use of concepts such as: event, causation,determinism, indeterminism, certainty, probability, deductive and non-deductive argument. In a book on explanation, it will be unnecessary tooffer analyses of these concepts. The purpose of introducing them is onlyto show how they relate to explanation. I use them, hopefully in waysuncontroversial to the matters at hand.

I introduce three sets of distinctions by which to categorize theories ofexplanation, (A), (B), and (C). The typology which these three sets of

26


distinctions produce permits us to categorize theories of explanation intwo different ways: epistemologically and metaphysically. The first twosets of distinctions, (A) and (B), are epistemological. Hempel, for example,says that we explain something when we see that it was to be expected;and it is in this sense that the explanation enables us to understand whythe phenomenon occurred.31 An expectation is a belief. Must our beliefabout the occurrence of the explained phenomenon be certain, or might itonly be likely? Under (A), I distinguish between theories of explanationwhich offer different answers to these questions.

Theories also differ about the form an explanation may take. I discussthese distinctions under (B). Must an explanation be an argument? Idistinguish between argument theories of explanation (which answer thepreceding question in the affirmative) and non-argument theories (whichanswer it in the negative). Argument and non-argument theories givesomewhat different answers to the epistemological question of the certaintyor probability of ones belief about the explanandum phenomenon.Argument theories can use the ideas of deductive and non-deductivearguments, as a way of giving substance to the ideas of certainty andepistemic probability; non-argument theories do not have this manoeuvreavailable to them.

The third set of distinctions which I use to classify theories ofexplanation is metaphysical, and I discuss this under (C). The relevantmetaphysical distinctions involve, among other things, the ideas ofcausation, determinism, indeterminism, and nondeterminism. That is,different theories of explanation presuppose different things about thenature and extent of causation. A theme that runs throughout this book isthe way in which an epistemic concept like explanation requires orpresupposes a metaphysical backing. I try to show how those differingmetaphysical commitments partially motivate different epistemic viewsabout explanation.

Probability is a highly ambiguous term, and although there are manykinds of probability, and various further distinctions one can draw withinthe two broad kinds of probability I distinguish, I want simply to separateepistemic or inductive probability from physical or objective or descriptiveprobability.32 There are many competing accounts of each (e.g. frequencyand propensity theories are competing accounts of physical probability;logical and Bayesian theories are competing accounts of epistemicprobability). Epistemic probability is concerned in some way with supportor degree of rational belief; physical probability is meant to be a matter ofobjective fact about the world. Obviously the two concepts of probability

27


are related, although distinct. (Another term that can have both an epistemicand a metaphysical sense is the concept of what is necessary. Necessarycan either be construed as certain, or as objectively necessary.)

(A) I begin, now, drawing the epistemological distinctions betweendifferent theories of explanation. There are certainty, high epistemicprobability, and low epistemic probability models of explanation. (Theseare three rival accounts.) On a certainty model of explanation, an explananscan explain an explanandum only if the explanandum is certain, giventhe information contained in the explanans. This is what we might callrelative or conditional rather than absolute certainty; something may becertain, given something else, without being certain or indubitable per se.This is one of the ways in which one might interpret von Wrights remark:what makes an explanation explanatory is that it tells us why [an event]E had to be (occur), why E was necessary once the basis is there and thelaws are accepted.33 Necessary here might be construed as an epistemicidea; it is certain that E would occur, given knowledge of the basis andthe laws.

On the other hand, one might only require of an explanation that theexplanandum be (epistemically) probable, given the explanansinformation. we we might try to salvage what we can by demandingthat an explanation that does not necessitate its explanandum must makeit highly probable.34 Salmon, in this passage, is suggesting that anexplanans need only to make its explanandum epistemically probable(we need not discuss just yet whether highly so or not), but need notmake it certain. An epistemic probability model says that there can bemore kinds of full explanations than the certainty model allows. The formerallows that there can be full explanations which meet the certainty model,and others beside. So they are rival accounts.

An epistemic probability model comes in a stronger and a weaker form.The strong model is a high epistemic probability model. It requires thatin a full explanation the explanandum is at least highly likely, given theexplanans information (or, the explanans highly supports theexplanandum). Given the information in the explanans, we have goodreason to believe that the explanandum is true, but perhaps not conclusivereason. It is true that the strong model has a certain vagueness about it,but it is not clear whether vagueness here is a strength or a weakness.What is highly likely? Any cut-off we select will appear arbitrary andunmotivated. But we might argue that this captures accurately thevagueness of explanation itself. The higher the probability of theexplanandum, given the explanans, the more clearly we have an

28


explanation. We have no clear intuitions, the strong modellist might say,about at precisely what point we cease having even a poor explanationand have instead no explanation at all. Moreover, this is entirely consistentwith the non-pragmatic view that there is a distinction between poorexplanations and no explanations.

Is explanation genuinely ambiguous according to the epistemicprobability model? This depends on the way in which the certainty-conferring and probability-conferring models of explanation are set out.The high epistemic probability model need not hold that there is a radicaldifference between the two kinds of explanation, since certainty is thelimiting case of high probability. On the other hand, Hempel often speaksof these as two different types or kinds of explanation. That on its own isno more evidence for ambiguity than the fact that there are vertebrate andinvertebrate kinds of animals is evidence for the ambiguity of animal.However, Hempel says that there are different models for explanation,and given his views on models and analysis, this ought to mean thatexplanation for him is ambiguous and stands for no single concept.Thus, he says: we have to acknowledge that they [explanationsconforming to the I-S model] constitute explanations of a distinct logicalcharacter, reflecting, as we might say, a different sense of the wordbecause (Hempel 1965:393).

The weaker version of an epistemic probability model does not evenrequire that, in a full explanation, the explanans information provide good,although not necessarily conclusive, reason to believe that theexplanandum is true. On this weaker version, the explanans informationmay only give some, albeit small, reason to believe that the explanandumis true; in one sense of expectation, the explanandum phenomenon wasnot to be expected at all. As Peter Railton says, the explanation does notexplainwhy the decay could be expected to take place. And a goodthing, too;there is no could be expected to about the decay to explainit is not only a chance event, but a very improbable one.35

Wesley Salmon has also argued that the explanandum might have alow probability, given the explanans. The quote from him four paragraphsback continues: even this demand [for high probability] isexcessivewe must accept explanations in which the explanandum eventends up with a low posterior weight (Salmon et al. 1971:64).

There are two sorts of arguments for a low epistemic probability modelof explanation. The first is simply the presentation of cases of explanationwhich appear to support such a theory. The most convincing examplesare indeterminisitic ones, since they ground the low epistemic probability

29


of the explanandum statement on the objective low conditional probabilityof the corresponding event. Both Salmon and van Fraassen use thisexample:

a uranium nucleus may have a probability as low as 1038 ofdecaying by spontaneously ejecting an alpha-particle at aparticular moment. When decay does occur, we explain it in termsof the tunnel effect, which assigns a low probability to thatevent.

(Salmon et al. 1971:152)

The thought here is that in both cases, the decays occurring and thedecays not occurring, precisely the same information is relevant tothe outcome. It seems arbitrary to allow that the information hasexplanatory force in the case of one outcome, but to deny that theinformation has any explanatory force in the case of the other.

This question should be considered by anyone who is inclined to accepta high epistemic probability model and deny a low epistemic probabilitymodel: why should exactly the same information, which intuitively seemsequally relevant to both events, explain one but not the other? Of course,the convinced high epistemic probability modellist can always reply:because the information makes what is to be explained highly probableor likely in one case but not in the other. What is wrong with the reply isthat it seems as arbitrary and unmotivated as the original doctrine. So theconclusion would seem to be: if explanations meeting the high epistemicprobability model are acceptable, then we should sometimes be in aposition to explain an explanandum on the basis of an explanans on whichthe explanandum is only improbable or unlikely.

A second argument that Salmon uses for the low epistemic probabilitymodel derives from a famous argument due to Kyburg.36 Kyburgsargument concerned the class of reasonably accepted statements, anidealized body of scientific knowledge. The question he raises is this: isthe class of reasonably accepted statements closed under conjunction?Closure of the set under conjunction would amount to this: If S is a bodyof reasonably accepted statements, then the conjunction of any finitenumber of members of S belongs to S. Suppose p and q are members ofthe set of statements which are reasonably accepted by me. Closure underconjunction means that, for any p and q, if p and q are reasonably accepted,then (p&q) is reasonably accepted.

30


Now, a statement need not be certain (have a probability of 1) in orderto gain admittance to S. Suppose that we decide to admit to the body ofour reasonably held beliefs only those beliefs which are either certain orhighly probable, say, with a probability of at least 0.85. We admit p and q(which we assume throughout are statistically independent), each of whichhas a probability of 0.9, and therefore qualifies for admission. Because ofthe basic multiplicative rule of the probability calculus, the belief (p&q)will have a probability of 0.81, below the bottom limit of acceptability.For whatever lower limit of acceptability that we set, some conjunctionof what is accepted will itself be unacceptable. This seems in contradictionto the intuitively plausible closure principle, but Kyburg himself, in hisarticle, counsels abandoning the principle despite its original appeal: Itis difficult to give an argument against the conjunction principle, partlybecause it is so obvious to me that it is false, and partly because it is soobvious to certain other people that it is true (Kyburg 1970:77).

One implication of the conjunction principle that Kyburg thinks isfalse is that one has a right to believe the conjunction of all the statementsone has a right to believe. Even if one has good reasons for believingeach and every statement that one believes, one may still have a generalargument for believing that some (but of course one would not knowwhich) of the things one believes are false. If such a general argumentwere sound, then one would not have the right to believe the conjunctionof all the statements that one has a right to believe.

Salmon has used this same basic argument in several of his writings,37

but applied it to explanation rather than reasonable acceptance, to link thefates of the high and low probability models. Suppose S is now taken to bethe body of explained statements (a statement is explained iff there is someexplanans that explains it). Lets pretend we are high epistemic probabilitymodellists, and say that a statement gains admittance to this set S only ifthere is some information on which the statement has a probability of atleast 0.85. Again, suppose that p and q each have a probability of 0.9. If weaccept the following conjunctive closure principle for explanation,

If S is the body of statements which have an explanation, then theconjunction of any finite number of members of S belongs to S,

then we can argue that (p&q) has an explanation which confers on ita probability below the required level. Whatever lower probabilitylimit we set for explanation, an application of this argument (Salmon1971:801) will force us to admit as an explained statement some

31


statement whose probability, given the explanans, is lower than theintended lowest limit. Hence, a low epistemic probability model ofexplanation must be true. Unlike Kyburg, Salmon holds fast to theconjunction principle, and accepts the consequence that there is nolowest limit to the epistemic probability an explanans must confer onthe explanandum in order for the former to explain the latter.

Salmon, as far as I know, has never given any argument for holding onto the conjunctive closure principle for explanation, which is odd giventhe fact that Kyburg seeks to resolve his own puzzle by denying the parallelconjunctive closure principle for reasonable acceptance. Colin Howsonhas suggested rejecting the conjunctive closure principle for explanation,in order to hold on to a high epistemic probability model without beingthereby saddled with a low epistemic probability model.38 He points outthat there is no general support for such a closure principle. In view ofthe havoc conjunctive closure rules would bring in an example such asthe set of reasonably accepted statements, Howson counsels arguing caseby case for their use, and not assuming universally, as Salmon seems todo, that conjunctive closure rules are reasonable. However, even if we donot accept Salmons second argument for a low epistemic probabilitymodel of explanation, adapted from Kyburg, the sorts of examples he andvan Fraassen cite still constitute some evidence in favour of such a theory.

There is even a strong and a weak version of a low epistemic probabilitymodel of explanation. A strong low epistemic probability model willrequire that the explanans raise the probability of the explanandum fromsome prior probability, even though the resulting probability may still below. The weak low epistemic probability model allows an explanation tofurther lower the explanandums probability from some already low priorprobability. (There is an analogous variant of th

Explaining Explanation - PhilPapers · Explaining Explanation ... original and challenging...

Documents

Transcript of Explaining Explanation - PhilPapers · Explaining Explanation ... original and challenging...