Schopenhauer and free will

1. Introduction

Schopenhauer remarked that the only freedom of the will we can coherently have is freedom to do what we will, not to will what we will.

More recent philosophers have not all respected the boundary that Schopenhauer drew. Some of them have defined the freedom of the will that they defend in terms that push in the direction of being able to will what one wills.

In this post we shall start by touching on a few well-established positions (section 2). Then we shall look at the option of choices being random within a constrained range (section 3). Finally we shall explore one feature of human beings that may promise to place them close to willing what they will. This is the ability to choose to make changes to one's own character (sections 4 and 5).

Our concern is with ways to look at the human experience of making choices and acting on them. We do not seek to study what happens in the way that psychologists would study us. Our approach is philosophical, rather than scientific.

We do not claim originality for the various theories that we shall use. They have mostly been around for quite some time. But we shall offer a way to put some options to work against the background of Schopenhauer's remark.

2. Theories of free will

2.1 Choices as the focus of theories

We take freedom to be freedom in choice of action. Two questions that are regarded as absolutely central by some philosophers, and unimportant by others, are these:

(a) Could the subject have made a different choice?

(b) Was a choice truly the subject's choice?

The focus on choices leaves some things out of account. On the side after a choice, we shall take no interest in whether a subject is physically restrained or otherwise incapable of carrying out a chosen action. On the side before the choice our focus will not be on the origins of subjects' desires, except when we consider choosing to change one's character and therefore the desires that may carry weight on future occasions of choice. Apart from that, we shall be happy to take desires as given. We shall not be concerned with whether it is physically determined that a given subject shall have any particular desire. We shall however be very concerned with the mechanism by which desires and perhaps other things give rise to choices.

2.2 Schopenhauer

In chapter 1 of his Prize Essay on the Freedom of the Will, Schopenhauer starts by pointing out that the concept of freedom from physical restraint makes perfect sense. Such freedom means that one can do what one wills. He then points out that freedom to will what one willed would demand a will that chose in one way rather than another without grounds for its choices. The rest of the essay is largely an attack on the view that we have or even could have such a freedom.

In chapter 2 he examines the role of motives, arguing that the way to provide a ground for the will's choices would be for the will to choose on the basis of the strongest motive, not to be free to decide whether to be guided by motives. He also refuses a half-way position in which motives would incline but not determine the will. And he points out that self-consciousness only reveals our choosing as we will, in response to our motives, not any trace of willing as we will. (There is here a dependence on Schopenhauer's more general doctrine of the Will, in which a person is his or her will.)

Moreover, while we can wish for several contradictory things, we can only will one of them. The possibility of wishing does not mean that we could in fact have willed differently without any reason (such as a stronger motive) determining us to do so. And in that case we could not have willed the option we in fact chose.

In chapter 3 Schopenhauer sets out the necessity of the effects of causes, up to and including motives as causes, even when they are the motives of human beings who can rationally consider the past and the future as well as the present when deciding how to act. He adds that an effect, a human action, will not follow without a cause, a motive. And he goes on to explore the notion of a person's character. It is the combination of character and external circumstances that will determine the choices made. The fact that the choices made cannot be reliably predicted by the subject does not mean that other choices could have been made. 

In chapter 3 Schopenhauer also sets out his view that a person's character is unalterable. Apparent changes merely result from being at different stages in life. And a change in conduct results not from a change in character but from increasing knowledge of the consequences of different choices. Moreover, the idea of a free will unbound by anything at all, a will that could will what it willed, would not only be incompatible with having a definite character. It would prevent such a character from being developed over a life.

Chapter 4 is a survey of views expressed in theology, philosophy and literature in the past.

In chapter 5 Schopenhauer rounds off his essay with an attribution to each of us of moral responsibility for our actions, despite their being determined, precisely because each person's actions are determined by his or her character. Here he comes close to basing moral responsibility on the guidance control that is now favoured by many compatibilists.

2.3 Compatibilists

Compatibilists, who seek a freedom that is entirely compatible with a deterministic physical world and the supervenience of the mental on the physical, have on the whole respected Schopenhauer's boundary. Guidance control, one's own brain and character having significant roles in the causal chains that lead from circumstances to actions, is enough for many compatibilists. It is a formulation of freedom to do what one wills which places emphasis on the will having substantive content and playing a significant role.

2.4 Incompatibilist libertarians

Incompatibilist libertarians maintain that a deterministic universe would be incompatible with the free will they seek, but that we have that free will anyway. If we ask what the implications of such a freedom would be, we find that we at least edge toward the freedom to will what one wills that Schopenhauer wanted to deny us.

A great deal hinges on the mechanism by which it is alleged that we make free choices. What mechanism could there be which would give freedom without degenerating into randomness?

There is not usually any suggestion that choices are random and inexplicable. To that extent, Schopenhauer's warning is heeded. But the deterministic physical world cannot be allowed to have total control. Nor can it be the sole guiding factor, leaving a residue of mere randomness, for that would leave us with a disrespectful vision of the subject who chooses.  And given that the link from willing to acting is within the physical world and therefore deterministic or subject merely to vagaries of the world by virtue of which results happen not to be what was intended, there would have to be a guiding factor before the willing of a particular action. To that extent the subject would need a method to will what he or she willed.

The deterministic physical world may be allowed to set some limits, even at the level of the neurons that embody decisions, but there is also a role for some rational influence on choices that does not supervene on the physical - and must not supervene on it if the incompatibilist libertarian position is to survive.

One candidate for such a rational influence is the subject's character. Another is the reasons for possible choices of which the subject is consciously or subconsciously aware. (We shall call these the available reasons, which may not be the full set of reasons that would be relevant to a decision.)

The avoidance of supervenience on the physical may be achieved by a denial that such things do supervene on the physical, perhaps on the basis that characteristics and reasons are inherently mental things. Then their causal efficacy may be achieved through some physical force of agency that characteristics and reasons invoke in favour of a particular choice. Alternatively characteristics and available reasons may supervene on structures in the physical brain, so that they are as determined as anything else, but some force of agency that is not part of a closed physical order of causes and effects may select and put into effect options chosen from a range that is already narrowed down by characteristics and available reasons. These may not be the only possibilities, but they indicate the kind of thing that would be needed.

A physical force of agency that was invoked by something not supervenient on the physical or a force that was itself not part of a closed physical order of causes and effects would be somewhat mysterious. But the idea of a force that operates within limits laid down by characteristics and reasons, whether or not those characteristics and reasons correspond to structures in the physical brain, does give us a clue to an approach that would avoid the need for a mysterious force. This approach is constrained randomness, and we shall now turn to it.

3. Constrained randomness

3.1 The basic idea

Suppose that a subject who faces a decision has a reasonably set character and a range of available reasons to prefer options. And suppose that the reasons cannot all be added up on a single scale to yield a clear conclusion as to which is the best option. (Such a computation would indeed often be impossible.) Then there would seem to be scope to make a free choice between a limited range of options. How would it be if that choice were random?

3.2 Kinds of randomness

We must first think about what we have in mind by way of randomness. It might be genuine physical randomness, such as is seen under some interpretations of quantum mechanics, triggering or not triggering neurons. Or it might be the result of some deterministic physical process, such that the selection of one option rather than another from within the range set by character and available reasons could not be explained in terms of the relative weights of characteristics of the subject or the weights of available reasons because there was no even remotely comprehensive way reliably to link the relevant physical process to any such explanation. (The writings of Donald Davidson on anomalous monism and of Christian List provide a point of comparison here. See Davidson, "Mental Events", as a starting point, although there is much subsequent discussion; List, Why Free Will is Real.)

Genuine randomness and non-explanatory determinism would be equally good for our purposes, because our concern will be with how things are when a subject's choices are not to be explained entirely in terms that look like respectable ways to explain choices - broadly, in terms of characteristics and reasons.

3.3 Assessing constrained randomness

How would constrained randomness look? Would it keep the worry of a lack of explanations of choices at bay? And would it give rise to a different worry, by constituting an attack on the unity and the reflexive transparency of the self?

3.3.1 Randomness and explanation

Constrained randomness would mean that people generally acted in character, and were not often liable to make strange choices that would lead observers to regard them as out of control. To that extent, it would fit our observations of how people act. It would also be a good fit with the observation that there is a degree of unpredictability in people's choices, however well one knows their characters and the ranges of reasons available to them.  And the restrictions imposed by characteristics and available reasons would accommodate the point Schopenhauer made that we can think of all sorts of implausible choices (which could happen through a deterministic exercise of imagination), tell ourselves that we could make them, but in fact find it difficult to the point of practical impossibility to do so (Schopenhauer, Prize Essay on the Freedom of the Will, chapter 3).

How would constrained randomness seem from the inside? One might anticipate a sense of lack of control when the randomness took effect, which would certainly be disturbing and would undermine one's self-conception as a free subject steering one's own life. But actually a sense of lack of control should not be expected, even though there might be justification for attributing lack of control.

Justification would require the detection of randomness. It could be detected in the following way. If we think about choices we have made, we may give explanations by reference to characteristics and available reasons. An ability to give such explanations allows us to see ourselves as captains of our lives. But what we may not notice is that we could have given equally good explanations had we made other choices. If we do notice that, we can expect to notice a residue of unexplained choice of one option over another.

Despite this, one would still not expect a sense of lack of control. The sensation one had as a choice was made would be the same, whether the choice was made by virtue of something random or by virtue of a mysterious force of agency. So the randomness would be hidden from the subject at the time of decision, even though it could be appreciated later.

3.3.2 The unity and transparency of the self

While constrained randomness might fit the data on what people do, and while it might also be consonant with the impression of choice that one has from the inside, there is a concern. Choice through such a mechanism might seem to undermine the unity and the reflexive transparency to oneself of the I that willed.

It is natural to think that there should be a central locus of control which sees all, or at least sees everything that makes a substantive contribution to choices (so that its own contents, perhaps not transparent to itself save in the form of an inactive copy that may not reflect current operations, are merely formal and do not have substantive content).

If such a central locus of control were to see a source of randomness, that would preserve the transparency of everything outside itself to the extent of recognising contributing entities. But on the other hand the source of randomness would be something outside the central locus's control. And its internal workings would have to be opaque in the sense that they could not be seen well enough for its outputs to be predicted reliably in specific cases, because that would eliminate randomness and reduce us to the determinism which provokes incompatibilist fears and drives those who have such fears but who want a strong form of free will to posit mysterious forces of agency.

(There might however be enough transparency of the source of randomness to allow statistical predictions about the relative frequencies with which different outputs would arise. That would not impugn the indeterminism achieved on individual occasions. To that extent the objection that agent-level compliance with the implications of micro-level statistical laws would be a wild coincidence, an objection set out in Pereboom, Free Will, Agency, and Meaning in Life, pages 66-69, may be one that can be ignored.)

Moreover, in order to fit the picture of constrained randomness, it would have to be impossible for the central locus to disconnect the source of randomness so that it had no effect on choices. If that happened, we would be back to a purely deterministic mechanism which would leave nothing that might satisfy the yearnings of those who feared incompatibilism. Nor would it make sense to allow the central locus a substantive as opposed to a merely formal contribution to choices, on pain of either losing transparency or falling back into the picture of willing what one willed that Schopenhauer found unsatisfactory.

We could of course give up on the unity and the transparency to oneself of the I that wills. We might say that imperfect unity and transparency were good enough. We could even say that imperfection here was a good fit with our first-person observation of ourselves. Sometimes we and our choices are puzzles to ourselves, and it is through living, making choices, and acting on those choices that we engage in an unending process of self-discovery.

4. Choosing to change one's character

4.1 The scope to change

There is another way in which a subject's presumed central locus of control might be given a role. This would be to consider deliberate changes of character. We would here exploit a scope for flexibility in character that Schopenhauer denied (Prize Essay on the Freedom of the Will, chapter 3).

What we consider here would not satisfy the incompatibilist libertarian as he or she focused on individual decisions, but it might still allow us to go beyond the mere guidance control of the compatibilists and safely edge a bit toward a subject's being able to will what he or she willed. And we would do so without any need for randomness, whether constrained or unconstrained. 

A human being is capable of standing outside himself or herself, contemplating his or her own character, and considering whether changes would be appropriate. One cannot stand entirely outside oneself, because the standpoint is within oneself. There is no Archimedean point on which to stand. A subject engaged in changing his or her character is in the position of repairing a raft at sea. The raft can be repaired piecemeal, but one must always stand on a part that it not currently under repair. (It is not clear whether there is any part of a personality that by virtue of its centrality can never be changed. In that respect things may be different from how they would be on a raft, in that the part of a raft where one was currently standing could always be repaired later.)

Inability to stand outside oneself means that unity and reflexive transparency of the subject cannot be perfect. But a degree of detachment is possible. Then it is possible to make a plan to change one's character, not by giving direct orders to oneself (which are likely to be ineffective) but by adopting habits that will in due course effect the desired changes. Success in this is never assured, but it is not out of reach.

4.2 Willing what one wills?

To the extent that one does succeed and that one's choices on individual occasions are constrained by one's character, there is a form of willing what one wills. The willing does not relate to any immediate choice, but to future choices. It is a willing that in certain circumstances one should have a will to do certain things and not others. At the present time, one only wills that one's character should change in certain ways, and freedom at that point consists in one's being able actually to promote those changes, to do what one currently wills. We would not have a case of freely willing to will to change one's character.

The modest nature of the claimed victory over a supposed inability to will what one wills, a victory that relies on separating in time various elements in an individual's psychic history, means that it would not matter if the process of deciding to make certain changes to one's character and then acting to make them was deterministic, rendered inevitable in every detail by supervenience on the physical universe. And it would also not matter if the process were supervenient on a physical universe that incorporated a mixture of determinism and randomness. It would suffice for the choice to make certain changes to one's personality to be one that was internally driven, along the lines of the guidance control beloved of compatibilists.

4.3 Two-layer structures

To what would this victory amount? Would it provide any comfort to someone whose natural inclination was to be a libertarian but who felt unable to sustain that position because of its seeming incompatibility with the nature of the world as viewed scientifically?

One point of comparison is the structure of first-order and higher-order desires that has been set out by Harry Frankfurt and that has been seen as providing comfort on questions of free will (Frankfurt, "Freedom of the Will and the Concept of a Person"). A subject can have first-order desires to do different and incompatible things, for example to eat delicious food and to stick to a diet. But a subject can also have a higher-order desire, that the desire to stick to the diet should prevail. The effect on conduct of having a higher-order desire that was strong enough to ensure that the preferred first-order desire would prevail could be seen as comparable to the effect on conduct of having a certain characteristic. And in Frankfurt's view, acting on first-order desires that are in conformity with higher-order desires is a sign that the actions really are one's own, and are in that sense free. The subject wills certain choices (at the first-order level) and wants (at the higher-order level) to will those choices. This is not the willing what one wills to which Schopenhauer objected, because the desires of the two orders are different and the higher-order desire has substantive content.

Comparability of effect should not be assumed to imply comparability of underlying source of the effects. There is a difference between a conscious higher-order desire, cited at the time when first-order desires are in conflict, and a standing disposition, perhaps unconscious, to choose to satisfy certain first-order desires rather than others.

When however we restrict our comparison to conscious higher-order desires at the moment of choice and characteristics that were at some time in the past deliberately developed, they do seem to come together as examples of a limited form of willing what one wills.

They come together because the form of willing is in both cases willing that there be an efficacious rule concerning how to decide on specific actions, a rule that is separable from those decisions on specific actions. A characteristic, whether deliberately developed or not, gives a rule that can be applied over and over. Likewise a higher-order desire is a rule that can be applied repeatedly. Unlike a characteristic, it might not be so well established that it would be applied reliably. It might come to consciousness on only a few of the occasions on which it was relevant, and might sometimes arise but be too weak to have the effect it promoted. But it would naturally be thought of as a rule to be applied repeatedly. And any thought that a higher-order desire was only meant for occasions when the first-order desire to be opposed was not especially strong would be odd.

This is however only a limited form of willing what one wills because we still find no reason to see the end results as entirely optional. It is perfectly possible, indeed highly likely if the mental supervenes on the physical, that the only way things could have turned out differently, either inside a subject's head or in his or her conduct, would have been by virtue of random processes that were in no sense under the subject's direction.

We may compare choices of characteristics and conscious higher-order desires with Robert Nozick's approach, which could be seen as a programme for the establishment of higher-order desires on the fly. Nozick speaks of a decision as setting a precedent under which it is itself subsumed (Nozick, Philosophical Explanations, chapter 4, section 1). He sets out how we assign weights to the various considerations that might bear on a decision. The assignment of weights on one  occasion will influence assignments on future occasions by setting a precedent, but not an inviolable one. Nozick also argues that the assignment of weights on a given occasion can set a precedent under which that assignment is itself subsumed (pages 300-301). He also addresses the risk of seeing our choices as random and uncontrolled by arguing that a choice can be explicable without its falling under a covering law, and in making that argument he uses our own experience of choosing (pages 301-306).

4.4 The advantages of two-layer structures

How much comfort does a two-layer structure give, whether the upper layer is chosen characteristics or higher-order desires?

It does not suffice to allow that things could have turned out differently, save by the unflattering subjection of subjects to randomness. But it does look significantly different from a single-layer picture in which a subject, faced with a given situation, automatically follows the tramlines of the world into a particular choice. It is a modest form of willing what one wills, and one that it is unsurprising Schopenhauer did not develop given his conviction that we have settled and invariable characters.

What is perhaps most comforting about a two-layer structure is that it makes explicit that we are subjects with directed and consistent lives which are not random, while we still have the option to change direction. At this point the idea of a rule to be applied repeatedly becomes important. We can create our future selves, at least to a limited extent, and the choices made in that process of creation are validly our own even if they were inevitable. Moreover, when it comes to choosing characteristics, the chosen characteristics can eventually seep into our bones, so that they do indeed become aspects of who we are.

It is here important to reflect on the process of deliberately moulding character. One consciously makes certain choices. This is an instance of a higher-order desire to take on some new characteristic which will in due course direct choices that may be influenced by first-order desires. Eventually the habit of choosing in what the subject considers to be the right way becomes automatic. In this respect the process of choosing characteristics has the edge over higher-order desires that operate by coming to consciousness on specific occasions of choice. Those higher-order desires do not seep into the bones of the subject, and there is a marked risk that they will not arise or not be strong enough when they are needed.

A further elaboration is possible, whether or not one accepts the supervenience of the mental on a physical world that may be entirely deterministic or perhaps deterministic with a bit of randomness. Within the range of characteristics that a subject might aim to develop through the practice of appropriate habits, there may be higher-order and lower-order ones. Examples of higher-order characteristics would be being thoughtful rather than impulsive, and not being indulgent in relation to pleasures in general. Examples of  lower-order characteristics would be not making assumptions about people newly encountered, and refusing foods that would be bad for one's health.

Higher-order characteristics would be of wider application than lower-order ones. But they would be of less certain application to specific choices. This would be both because their lack of much concrete content would make it hard to compute precise recommendations, and because a subject who did not relish the choices they indicated would have scope to argue internally that other choices more or less complied with their imperative content. Indeed, higher-order characteristics might largely take effect through their influence on the selection of lower-order characteristics, which would have enough concrete content to be of reasonably certain application to specific choices.

This point about influences of some characteristics on the selection of others brings out an advantage of having interactions between different characteristics in the process of their being chosen or otherwise coming to accrue to a subject. Such interactions can enhance the coherence of a set of characteristics, with some possible characteristics failing to accrue because of their conflict with other characteristics. And such interactions would not be limited to downward influences through hierarchies of characteristics. They might include influences upward, or influences back and forth across a network that might not have a clear hierarchy. Nor should we ignore a broader picture of influences, as set out in Michael Bratman's discussion of the formulation of coherent plans to be implemented over time (Bratman, "Planning Agency, Autonomous Agency").

5. Stability and flexibility of character

5.1 The importance of stability

While we have made use of the ability deliberately to change one's character, it would be undesirable if people's characters did not have considerable inertia, resisting change. Fortunately, they do have such inertia.

If a subject's character has inertia and hence stability, we can speak of the subject's personality. If the character fluctuated widely, either over the full range of possibilities for someone of that age, sex and culture or over some narrower but still reasonably broad range, we would have to say that the subject had no personality in particular. And that would be an attack on the very notion of a human subject. It would also make social interaction difficult, because people like that would be a good deal more unpredictable than people in general actually are.

Another aspect is that a degree of settled personality is required for whatever freedom of the will we have to be something worth having. A subject exercises his or her will to carry out his or her projects, in accordance with his or her own ideas of what is valuable. If the subject had no stable character, the subject could not say "I did this" with any more than an etiolated sense of "I". For a subject's utterance of "I did it my way" to carry weight and be a source of satisfaction, the way has to be the subject's in the sense that it reflects how the subject actually is. And if the subject is unstable from moment to moment, it is not possible to say how the subject is in a sense that would allow him or her to look back on something done and affirm that it reflected how he or she really was at any time other than the moment of action. Even potential great instability which had not been actualised would cast doubt on a claim that the action reflected how the subject really was, because truly having one's characteristics requires having them securely.

Stability is also needed for the compatibilist view that is based on guidance control to be at all appealing. If satisfaction comes from being able to claim that a subject's choices and actions reflect substantial causal roles for the subject's own brain and the psyche that supervenes on it, the psyche had better be stable for the claim of substantial roles to have much significance.

5.2 Layers and networks

Layers or networks of interacting characteristics may be a source of stability. An isolated characteristic is at constant risk of losing its grip on choices and conduct, prompted by some immediate temptation. But if different characteristics, vulnerable to different temptations, interact, any one characteristic may be kept at its task because a tendency for it to lose its grip would be corrected by the fact that this would be inconsistent with other characteristics to which it was related. This might be so either within a non-hierarchical network or, in a specifically top-down way, within a hierarchy of  higher-order and lower-order characteristics. Indeed the higher-order characteristics, having only limited concrete content, should be relatively invulnerable to temptations to relax their grip, although on the other hand their grip at the level of specific choices and conduct would be limited by the very fact that they were short of concrete content.

5.3 The balance between stability and flexibility

We can reflect on the balance between the twin desiderata of stability of character to give ownership of our choices and freedom to change character. Here we stop thinking about how to conceptualise what we actually have, or what it is that we actually have, and consider what we might like.

We can imagine a horizontal axis of fixity of personality, going from great changeability on the left to great fixity on the right. A vertical axis can measure desirability. One curve, representing the benefit of flexibility, can run downward from left to right. Another, representing the benefit of stability, can run upward from left to right. The curve representing overall benefit can be some function of values indicated by those two curves, perhaps addition or perhaps something more complicated. Then in designing a world that would be ideal, we could identify the point at which the curve representing benefit achieved a maximum.

(What constituted benefit would be a separate question. It might be contentment, or degree of respect for oneself or for human subjects as agents, or something else.)

This picture of a graph is of course somewhat fanciful. It is not at all clear how the relevant variables could be quantified or the function defined. And even if those obstacles were overcome, there might be no unique point of maximum benefit. But the picture may bring out how conflicting desiderata can be related to one another.

6. Conclusion

We have not found a coherent notion of willing what one wills that Schopenhauer would consider admissible. Nor have we come up with anything that would entirely satisfy the incompatibilist libertarians, because constrained randomness is tainted by randomness, something that does not look compatible with human dignity, and because what appeared in hierarchies of desires or characteristics could still be fully determined by the merely physical. What would still be lacking would be a will that was neither determined by the physical world nor undetermined, but determined by the subject.

What we can do is move a little way in the direction of willing what one wills, giving more freedom than either Schopenhauer or compatibilists would allow, without exposing ourselves to Schopenhauer's complaint of incoherence. But central to this move is a rejection of Schopenhauer's belief in the unalterability of human character. If we see character as changeable, and in particular as open to deliberate moulding by the subject, we can exploit the passage of time to let what is willed at one time have influence not only over what is done at a later time, but also over what is willed at a later time.

References

Bratman, Michael E. "Planning Agency, Autonomous Agency". Chapter 1 of James Stacey Taylor (ed.), Personal Autonomy: New Essays on Personal Autonomy and its Role in Contemporary Moral Philosophy. Cambridge, Cambridge University Press, 2005.

Davidson, Donald. "Mental Events". Essay 11 of Davidson, Essays on Actions and Events, second edition. Oxford, Clarendon Press, 2001.

Frankfurt, Harry G. "Freedom of the Will and the Concept of a Person". Chapter 16 of Gary Watson (ed.), Free Will, second edition. Oxford, Oxford University Press, 2003.

List, Christian. Why Free Will is Real. Cambridge, MA, Harvard University Press, 2019.

Nozick, Robert. Philosophical Explanations. Cambridge, MA, The Belknap Press, 1981.

Pereboom, Derk. Free Will, Agency, and Meaning in Life. Oxford, Oxford University Press, 2014.

Schopenhauer, Arthur. Prize Essay on the Freedom of the Will, edited by Günter Zöller, translated by Eric F. J. Payne. Cambridge, Cambridge University Press, 1999.