Teleosemantics cartoon
Back to Home Under construction Mark Rowlands - University College/Cork 

1. The Teleofunctional Account: An Outline

By `teleosemantics' is meant a teleofunctional account of what determines the semantic content of mental representations.1 And any teleofunctional account will employ, as a pivotal concept, what Millikan calls proper function. The proper function of some mechanism, trait, or process is what it is supposed to do, what it has been designed to do, what it ought to do. More precisely, consider the following simplified version of Millikan's already simplified version of her definition of proper function given in Language, Thought and Other Biological Categories.2

An item X has proper function F only if (i) X is a reproduction of some prior item that, because of the possession of certain reproduced properties, actually performed F in the past, and X exists because of this performance; or (ii) X is the product of a device that had the performance of F as a proper function and normally performs F by way of producing an item like X.

This definition, simplified though it is, takes some unpacking. Firstly, the concept of a proper function is a normative concept. The proper function of an item is defined in terms of what an item should do, not what it normally does or is disposed to do. The concept of proper function, being normative, cannot be defined in causal or dispositional terms. What something does, or is disposed to do, is not always what it is supposed to do. This is for three reasons. Firstly, any mechanism, trait, or process will do many things, not all of which are part of its proper function. A heart pumps blood; it also makes a thumping noise and produces wiggly lines on an electrocardiogram. But only the first of these is its proper function since only pumping blood is something performed by hearts in the past that explains the existence of hearts in the present. Secondly, a mechanism, trait, or process can have a proper function even if it never, or hardly ever, performs it. To use a flagship example of Millikan's, the proper function of the tail of a sperm cell is to propel the cell to the ovum. The vast majority of sperm cell tails, however, do not accomplish this task. Third, a mechanism, trait, or process may have a proper function and yet not be able to perform it properly. A person's heart may be malformed and, thus, not be able to pump blood properly. Nevertheless, pumping blood is its proper function because ancestors of the person whose heart it is had hearts which pumped blood and this (in part) explains why they survived and proliferated and, thus, why the person in question possesses a heart (although not why that heart is malformed). The concept of proper function is, thus, a normative concept. The proper function of an item is its Normal function where, following Millikan, the capitalized `N' indicates that this is a normative sense of normal and not a causal or dispositional sense.

What underlies the normativity of the concept of proper function is that the concept is essentially historical in character. The proper function of an item is determined not by the present characteristics or dispositions of that item but by its history. In particular, the possession of a proper function F by an item depends on that item existing because it possesses certain characteristics that have been selected for because of the role they play in performing F. This is the import of (i). That (i) be satisfied is a necessary condition of an item possessing what Millikan calls a direct proper function. Such possession is essentially a matter of history. There are no first-generation direct proper functions.

There is, however, an important distinction to be observed between direct and derived proper functions. The idea of a derived proper function is captured by condition (ii). Consider the distinction between the particular pattern of pigmentation distributed over the skin of a chameleon and the mechanism that produces this distribution. Only the mechanism itself possesses a direct proper function, roughly the function of distributing pigmentation in such a way that the chameleon will match its immediate environment. However, the state of the chameleon's skin - the particular distribution of pigmentation - also possesses a proper function; a function that derives from the mechanism that produces it. The derived proper function of the `Pollockian' arrangement of pigmentation in the chameleon's skin is to match the chameleon to the Jackson Pollock No. 4 upon which the poor chameleon has been placed. And, unlike direct proper functions, there can be first-generation derived proper functions. The Pollockian arrangement of the Chameleon's pigment has the (derived) proper function it has even if no chameleon has ever been placed on a Pollock No. 4 before, consequently even if no chameleon has ever produced this particular pattern before.

The proper or Normal function, direct or derived, of many evolved items is relational in character. Often this is because the characteristic has evolved to enable the organism to cope with its environment: to locate food, evade predators, protect itself against heat and cold, and so on. Thus, proper functions are often defined relative to some environmental object or feature: the function of the chameleon's skin is to make the chameleon the same colour as its immediate environment; the function of the lion's curved claws is to enable it to catch and hold onto prey; the function of the bee's dance is to indicate the distance and direction of nectar. In each case, the function of the characteristic is specified in terms of the relation it bears to an environmental item. And the reason for this is that the very reason the characteristic in question exists is that it has evolved to meet certain environmental pressures.

The core idea of the teleofunctional account of representation is that the mechanisms responsible for mental representation are evolutionary products also. As such, they will have (direct) relational proper functions. The idea, then, is that the representational capacities of a given cognitive mechanism are specified in terms of the environmental objects or features that are incorporated into that mechanism's (direct) relational proper function. That is, if a cognitive mechanism M has evolved in order to detect an environmental feature E, then this is what makes an appropriate state S of M about E; this is what makes state S mean that E, this is what gives the state S the content that E. In this way, the representational content of a cognitive state S derives from the direct relational proper function of the mechanism M that produces S, and S itself has the derived proper function of representing E.

2. Strengths of the Teleofunctional Account

The teleofunctional account is one way of attempting to naturalize semantics: very roughly, to show how the semantic properties of mental representations can arise out of non-semantic properties. One of the most significant strengths of the teleofunctional account is its ability to resolve a problem that plagued other attempts at naturalizing semantics, attempts embodied in both causal and purely informational accounts of representation. This problem is that of misrepresentation.

Consider a mental representation of a horse. Adopting common practice, I shall refer to this by way of the capitalized HORSE to show that we are talking about the representation and not the horse itself. The representation HORSE, it seems, means `horse'. This is what makes it the representation it is, and not the representation of something else. However, it also seems possible, indeed likely, that the representation HORSE can be caused by things that are not horses. Donkeys in the distance and cows on a dark night might, in certain circumstances, be equally efficacious in causing a tokening of the HORSE representation. Now, according to an informational account, representation is to be explained in terms of nomic dependence. However, if the representation HORSE can be tokened in the absence of horses, then HORSE does not seem nomically dependent on horses in the relevant sense. Rather, what HORSE does seem nomically dependent upon is not the property of being a horse but the disjunctive property of being a horse or a donkey-in-the-distance or a cow-on-a-dark-night. Thus, if information is a matter of nomic dependence, and if representation is a matter of information, then we seem forced to say that what HORSE represents is not the property of being a horse but the above disjunctive property. But, fairly clearly, when we think, for example, that Arkle was a horse, and thus token the representation HORSE, we do not think that Arkle was either a horse or a donkey-in-the- distance or a cow-on-a-dark-night. We think that Arkle was a horse. And if I am inclined to have a flutter on the 2.45 at Uttoxeter, still less do I think of myself as betting on something that is either a horse or a donkey-in-the-distance or a cow-on-a-dark-night. The moral of the story is that there seems to be a divergence between the meaning of a representation and the information carried by that representation. The information is disjunctive in a way that the meaning is not. This problem of misrepresentation is, thus, often referred to as the disjunction problem.

One strength of the teleofunctional account is that it yields an elegant solution to the problem of misrepresentation. HORSE represents the property of being a horse and not the above disjunctive property because the direct proper function of the mechanism that produces HORSE is to produce this representation in the presence of horses and not donkeys or cows. That is, the HORSE representation, we can suppose, is a state of a mechanism M. And this mechanism has the direct proper function of producing HORSE in the presence of HORSES. That is what (presumably among other things) the mechanism has been selected for. It does not have the proper function of producing HORSE in the presence of donkeys or cows, whether in the distance or on a dark night. And, on a teleofunctional account, the content of a state S of mechanism M derives from the direct proper function of M. Thus HORSE is about horses and not about donkeys, cows, or disjunctions of the three. Providing a solution to the disjunction problem requires, in effect, detaching the content of a representation from the property with which it is maximally correlated. And this is precisely what the teleofunctional account allows us to do. Representation, on this view, derives from proper function. And proper function, being normative, cannot be defined in causal or dispositional terms. The fact that HORSE is tokened not only in the presence of horses but also in that of donkeys and cows, and is maximally correlated with a disjunction of the three, is, therefore, irrelevant. What determines the representational content of HORSE is not what environmental item in fact does causally produce it, but what should causally produce it. And this is determined by the direct proper function of its producing mechanism.  

3. Unresolved Problems?

The problem that has perhaps most exercised both opponents and proponents of teleofunctional accounts is often referred to as the problem of indeterminacy. Some have argued that a teleofunctional account of representation entails the indeterminacy of mental content. It does so because, it is claimed, the biological function of a mechanism, trait, or process is itself indeterminate.

Consider an example that by now has assumed the status of a classic. Frogs catch flies by way of a rapid strike with their tongue. Thus, it is plausible to suppose, mediating between the environmental presence of a fly and the motor response of a tongue strike is some sort of neural mechanism that registers the fly's presence in the vicinity and causes the strike of the frog's tongue. In more detail, we might suppose the presence of the fly causes the relevant mechanism to go into state S, and its being in state S causes, by way of various motor intermediaries, the tongue to strike. According to the teleofunctional account, the content of state S should be, roughly, `fly!', or `fly, there!', and it derives this content from the fact that the proper or Normal function of its underlying mechanism is to detect the presence of flies. The state is, thus about flies; it means that there are flies in the vicinity.

There is, however, an alternative construal of the function of the mechanism. On this construal, what the mechanism in question has been selected to respond to are little ambient black things (To avoid becoming entangled in a completely different issue, let me make it clear that the little ambient black things are environmental entities and not dots on a retinal image). The proper function of the mechanism, on this construal, is to mediate between little ambient black things and tokenings of a state that causes the frog's tongue to strike. This state will then beabout little ambient black things and will, therefore, mean that there are little ambient black things in the vicinity.

The proper function of the mechanism is different in each case since, in the latter case but not the former, the frog's mechanism is functioning properly or Normally when the frog strikes at a little ambient black thing that is not, in fact, a fly (but, say, a lead pellet or `BB'). And the content underwritten by the function is different in each case since not all little ambient black things are flies.

The problem of indeterminacy, it is argued, arises because there seems to be no fact of the matter that could determine which of these interpretations is the correct one. Evolutionary theory, it has been argued, does not provide any means of adjudicating between these interpretations. That is, evolutionary theory, by itself, is neutral between the claim that the neural mechanism Normally mediates fly strikes and the claim that the mechanism Normally mediates little ambient black thing strikes that are situated in an environment where little ambient black things are Normally flies. Therefore, whether the mechanism is about, and hence means, that there are flies in the vicinity or that there are little ambient black things in the vicinity is also indeterminate. In this way, the teleofunctional account of representation is thought to entail the indeterminacy of mental content. And so, to use Fodor's colourful expression, it is most unlikely that Darwin is going to pull Brentano's chestnuts out of the fire.3

The most influential recent attempts to resolve the problem of indeterminacy can be divided into two sorts, according to whether they endorse what are sometimes known as a stimulus-based or a benefit-based account of representation. In his more recent writings. Dretske endorses a stimulus based account of representation.4 According to Dretske, when an indicator C indicates both F and G, and its indication of G is via its indication of F, then C acquires the function of indicating F. And this is so even if it is G that is most relevant to explaining the indicational role of C. Thus, to use an example made famous by Dretske himself, the magnetosome possessed by certain marine bacteria represents the direction of geomagnetic north and not the direction of oxygen-free water. It represents this even though the purpose of the magnetosome - the reason for which the magnetosome has been selected - is to direct the bacteria away from the oxygen-rich surface water that would be lethal for it. Thus, the magnetosome represents geomagnetic north and not the direction of oxygen free water even though the only relevance - or benefit - of geomagnetic north to the bacterium is that by heading for geomagnetic north it thereby moves into oxygen-free water (in the southern hemisphere the magnetosomes are reversed). Geomagnetic north is the stimulus for the magnetosome, and representation, for Dretske, follows or tracks stimulus not benefit.

Millikan, on the other hand, endorses a benefit-based account of representation. To this end, she has, in her more recent writings, been emphasizing the role played by the consumers, rather than the producers, of representations in fixing content (and this puts her position somewhat at odds with the outline of the teleofunctional account given in the opening section).5 In her later works, Millikan argues that while the mechanism that produces a representation may determine that representation's status as a representation, it is the consumer of the representation, the mechanism (typically) that interprets and uses the representation that determines its content. Thus, in the case of Dretske's marine bacteria, what the magnetosome represents, for Millikan, is only what its consumers require that it correspond to in order to fulfil their functions. And what these mechanisms require, MIllikan argues, is only that the pull on the magnetosome be in the direction of oxygen-free water. Thus, the magnetosome represents the direction of oxygen-free water, not the direction of geomagnetic north.

I think it is fair to say that both stimulus- and benefit-based approaches face unresolved difficulties. However, perhaps most fundamentally, it is simply unclear what could force the issue one way or another. While defences of each position are often ingenious, it is difficult to see what argument could rationally compel on to adopt one approach over the other. In a recent paper, I have argued, in effect, that there may, in fact, be no need to adjudicate between approaches.6 This is because acceptance of both a stimulus- and a benefit-based approach does not, in fact, lead to the indeterminacy of mental content. Central to my account is the idea that possession, by a single mechanism, of two non-equivalent proper functions does not lead to indeterminacy of content as long as the content underwritten by each function does not attach to the same thing. And the possession by a single mechanism of two distinct proper functions does not entail that the contents underwritten by each function both attach to the mechanism. I distinguish between what I call organismic and algorithmic proper functions of a mechanism. In the case of the frog, the algorithmic proper function of the mechanism might be to detect the presence of little ambient black things, while the organismic proper function is to enable the organism - the frog - to detect the presence of (following Gibson) an affordance of the environment: eatability. (On this account, detection of flies is not a proper function of the mechanism but a consequence of the mechanism performing its proper function in Normal conditions). The algorithmic proper function underwrites attribution of the content `Little ambient black thing, there!' to the mechanism (or to a state of the mechanism). But the organismic proper function, since its function is not to detect eatability as such but, rather, to enable the organism to detect the presence of eatability, underwrites attribution of the content `Eatability, there!' not to the mechanism but to the organism. Thus, since the content in each case attaches to different objects, we do not have a genuine case of indeterminacy but, rather, a case of two perfectly determinate contents being attributed to distinct things.


1. This definition is taken from Millikan, `Compare and contrast Dretske, Fodor, and Millikan on teleosemantics' in her White Queen Psychology and Other Essays for Alice, Cambridge, Mass., MIT Press, 123.

2. Millikan's original definition of the notion of proper function is, in reality, a series of definitions, spread out over the opening chapter of Language, Thought, and Other Biological CategoriesPhilosophy of Science, 56, 2, 1989, 288-302. Reprinted in White Queen Psychology, 13-29.

3. Jerry Fodor, `A Theory of Content I' in A Theory of Content and Other Essay, Cambridge, Mass., MIT Press, (1990), 70.

4. Fred Dretske, `Reply to reviewers', Philosophy and Phenomenological Research, 1, 4, 1990, 819-39.

5. See, for example, Millikan, `Compare, and contrast Dretske, Fodor, and Millikan on teleosemantics'.

6. Mark Rowlands, `Teleological Semantics', in Mind, Vol. 106, no. 422, April 1997, 279-303.


Under construction                                                                                                 Top of page