The informational approach to communication
Here are three examples of the informational definition of communication: ‘behavior that enables the sharing of information between interacting individuals’ (Smith, 1977, p. 2); ‘the transfer of information via signals sent in a channel between a sender and a receiver’ (Hailman, 1977, p. 52); and ‘the behaviors by which one member of a species conveys information to another member of the species’ (Kimura, 1993, p. 3). There is, for some, an intuitive sense in which any adaptationist account of signalling and communication is ‘incomplete’ (Stegmann, 2005, p. 1016) as the content of the signal appears to be ‘missing’ (p. 1015) and the definition ‘has nothing explicit to say about the [signal]’s information itself’ (p. 1017). Hence if our intuitions are any guide then perhaps it is desirable to invoke information, even if we do not need to. Indeed, information does appear to offer something that the adaptationist definition does not: specification of a given signal’s content at a very fine-grained level (for a wide range of examples and extensive discussion see Bradbury & Vehrencamp, 1998; chapter 13). Yet the logical conclusion of adaptationist accounts of communication is to eschew the notion of information altogether and instead focus only on behaviours and adaptation (Dawkins & Krebs, 1978). This does not imply that information is not part of communication; only that it need not (and, as we shall see, should not) be definitional. This section explains why that is so.
The argument, in short, is that in order to specify exactly what information is transferred we first have to specify the signal’s function, which brings us immediately back to the adaptationist definition. To begin, we must first define information. Appeals to intuition are insufficient here if our goal is to derive a precise definition of communication, so we cannot simply invoke some ‘everyday’ sense. More formally, information is considered to be equivalent to a reduction in uncertainty (Shannon & Weaver, 1949). But uncertainty of what? Such a definition is incomplete without a statement about what is relevant. A simple thought experiment illustrates the problem. What uncertainty is reduced if I receive an email from a friend with instructions about where to meet for dinner this evening? Most obviously, there is a reduction in the uncertainty of where I should expect to find my friend this evening, but there is also a reduction in uncertainty about many other things: the correct functioning of my computer, the correct functioning of my email program, the correct functioning of the satellites that carried the message in its digital form, the continued non-existence of a meteor in the location previously occupied by the satellite that carried the message, and so on.
[This is analogous to a famous problem in the philosophy of language, that of referential indeterminacy (Quine, 1960). The thought experiment runs as follows: an anthropologist sees a member of a tribe, whose language he does not know, shout the word ‘gavagai!’ just as a rabbit runs past. What does ‘gavagai’ mean? Although we would intuitively say ‘rabbit’, we cannot know this deductively. We instead have to make an inductive leap, and hence our conclusion about the meaning of gavagai is not logically determined. It could, after all, mean ‘fluffy thing’, ‘dinner’, ‘undetached rabbit parts’ or any one of an actually infinite number of possibilities.]
Hence although a signal most certainly conveys information, exactly what information it conveys is determined at least in part by the receiver: if I wish to know about the correct functioning of my email program (it may have recently played up on me, say) then my friend’s email is informative in that regard; if, on the other hand, I have no concerns about my email program then the email is not informative (in that regard). This is all, of course, in addition to the information about dinner that forms the body of the email, which we assume to be relevant. But why do we assume that the body of the email is relevant? Only because it would be unusual indeed for my friend to email me about where to meet if we had no previous plans to meet at all, or if my friend knew I was busy elsewhere, or if, even, my friend was busy and could not meet me. In short, we assume that the signal is in some way designed to achieve a particular purpose. Other aspects of the signal (like, for example, its appearance in my email program in an apparently normal manner) may carry other sorts of information; specifically, they may carry information that was not part of the signaller’s design of the signal (i.e. about the normal functioning of my email program). These sorts of inferences are the human equivalent of a cue.
Therefore in order to argue that communication entails the provision of information we must also specify what is relevant about the signal, otherwise we are left with a definition so broad that it includes cues: traits or behaviours that are informative but are not designed to be so. How do we define what it means to be relevant? Relevant signals are those that are of use to the receiver; irrelevant signals those that are not. And of course, if the receiver has nothing to gain from the signal then we should expect the receiver to adapt to ignore the signalling behaviour. Correspondingly, the signaller will cease to perform the behaviour. Hence relevant signals are those that are adaptive for the signaller to perform and for the receiver to respond to. If the communication is not in the interests of one or the other organism then any signalling or responding behaviour will be maladaptive: even if there are no direct costs to being ignored (as a signaller) or drawing false conclusions (as a receiver) then the behaviour will still be maladaptive rather than neutral because of opportunity costs and wasted energy (Maynard Smith & Harper, 1995).
We have now arrived back at the non-informational, adaptationist definition of communication: the signal must be designed to have an effect on the receiver, and the receiver’s response should also be designed to utilize the signal appropriately. In other words, when we flesh out a definition of communication based upon the transfer of information then we see that it collapses to the non-informational adaptationist definitions that we already know are sufficient to define communication. Or to put it another way, the adaptationist definition necessarily incorporates the informational definition. If we insist that information is transferred in communication – and we need not, as the adaptationist account makes clear – then the best we can say is that it is an emergent feature of communication, and ‘not an abstraction that can be discussed in the absence of some specific context’ (Hauser, 1996, p. 6). Consequently an informational account that does not discuss adaptation necessarily brings with it a degree of ambiguity, as we cannot even specify what the informational content of a signal is until we have specified what the signal is (adapted) for.
As an example, consider fluctuating asymmetry (FA), the deviations from perfect symmetry exhibited by almost symmetrical structures. It has been suggested that FA is a reliable proxy measure of homeostasis (Mather, 1953), and hence that organisms that exhibit low FA should be more attractive than those with high FA scores (Møller, 1993). Is symmetry a signal? In the absence of any behaviour that brings attention to symmetry, the intuitive answer must be no. But on an information-based view the answer will often be yes: FA carries information, that information is of value to the receiver (the potential mate), and conveying that information would be beneficial for at least some senders; specifically those with low FA scores. In contrast, FA does not qualify as a signal on an adaptationist definition unless there is a specific display that brings attention to it. (In this case we may even wish to argue that the display is the signal, rather than the FA itself, though whichever we prefer the argument is not changed.) In the absence of a display, FA should instead be seen as a cue (Maynard Smith & Harper, 2003). This example illustrates the general problem with information-only definitions: they do not specify how or why a signal is relevant, and hence they have no consistent means by which to distinguish between signals, cues and coercion.
Other alternatives to the adaptationist account
One historically popular view is that signals should be conspicuous, exaggerated or stylized. This idea arises from early ethological work on how signals first emerge (Tinbergen, 1952; Huxley, 1966), which emphasized the fact that signals are likely to be derived from other behaviours and that once they have obtained their new function are likely to increase its salience in order to have the communicative effect desired. [The roots of this idea can be found in Darwin’s (1889) Principle of Antithesis and are also observable in the landmark Krebs & Dawkins (1984) paper published nearly 100 years later.] However, there are plenty of subtle examples, observable only with great care or detailed video analysis, that will fall foul of this criterion. Indeed, economy of effort is often noted to be a common feature of animal signals, an observation that comes into direct conflict with the idea that signals will evolve to be highly salient (Dawkins, 1995). Relative salience is therefore an inappropriate feature with which to define signals and hence communication. Intention is another suggestion occasionally put forward as diagnostic of signalling (Batteau, 1968) but is undesirable for the simple reason that we have no reliable way in which to measure an organism’s intentions. Moreover, there must be considerable doubt on whether it is even coherent to talk about the intentions of non-human species.
Other authors have suggested refinements to the adaptationist approach, despite its internal logic. Such proposals are likely to lead to a degree of confusion. Indeed, the stated aim of Maynard Smith & Harper (2003, p. 1) book was ‘not to report any new facts [but rather]… to bring order out of chaos’, and their discussion of the definition of an animal signal suggests that they see these proposed alternatives as a route to ‘semantic confusion’ (p. 6). They criticize two specific suggestions. The first is that the notion of a cue be limited to behavioural traits, a move which would allow for a distinction between cues and signs (Galef & Giraldeau, 2001). While there are interesting differences between physical and behavioural traits, a term is nevertheless necessary for those features that affect behaviour but are not functionally designed for this purpose. That role is presently performed by cue, and is widely understood as such. Moreover, this distinction is, as we have seen, utterly crucial to a rigorous definition of communication; without it we fail to represent the nature of communicative dynamics and are thus left, inevitably, with ambiguous examples like FA. A distinction between the physical and the behavioural would be of value, but redefinition of an established term like cue seems the wrong way in which to achieve this, particularly in a field that already has a large degree of semantic inconsistency (Maynard Smith & Harper, 2003).
The same problem bedevils the suggestion that the key distinction between a signal and a cue is not whether or not it has evolved for the given effect in question but is instead the combination of two other features: first, that cues are always ‘on’ while signals are switched between ‘off’ and ‘on’; and second, that once a cue has been produced it costs nothing more to express it while the same is not true of signals (Hauser, 1996). This plays to our intuitive idea that signals are projected into the world in some way, rather than being consistently observable features. Yet it is unsatisfactory not only because of the semantic confusion that it is likely to induce but also because there are several examples that do not fit at all easily into this new framework. For example, under this definition fixed badges of status are cues while coverable badges are signals. Yet some supposedly fixed badges can be more or less prominently displayed, according to context (Maynard Smith & Harper, 2003). Another example would be aposematic colouration. In some species, for example monarch butterflies, this is permanently ‘on’, while in others, for example octopuses, it is switched between ‘off’ and ‘on’. Yet we would surely wish to use the same term for both examples. Neither is there an escape route offered by insistence that the overt display of such traits be considered the signal, for no such display takes place in the case of monarch butterflies. In sum, the distinction should be recognized but it is not definitional.
Indeed, similar criticisms will apply to most and perhaps all refinements of the adaptationist approach. This is because that approach contains within it the logical form of the communicative act, and hence any refinements will necessarily be deviations from that form. Once that occurs, ambiguous examples are likely to arise.