20

There are paradigms in which there is a consensus that representations are necessary constructs for explaining behaviour. One clear example is filial imprinting (e.g., Hollis et al., 1991; Johnson, 1992; Bateson and Horn, 1994). Here there is compelling evidence that chicks store information about the characteristics of the maternal hen, and the neurophysiological basis of this process is now understood in some detail (Horn, 1985; 1990; Honey et al., 1995). The considerable time-lag that may elapse between exposure to an imprinting object and successful performance on a discrimination task makes the inference that behaviour is dependent on a stored representation unavoidable. This work suggests another possible approach for evaluating whether or not functionally referential signals are also representational. Experiments could be designed to assess whether exposure to referential signals significantly affects subsequent behaviour measured hours or days later. For example, if birds were exposed to aerial alarm calls in one context, perhaps arranged to be visually distinctive, would they then be persistently more likely to engage in behaviour that functions to detect aerial predators (e.g., scanning upward) than in other environments of a similar type? This strategy for detecting representations relies upon tests to determine whether changes in behaviour caused by signals are based on memories stored for longer periods than the seconds or minutes that normally elapse in playback experiments. Intriguingly, there is anecdotal evidence to suggest that playback of snake alarms affected the behaviour of vervet monkeys passing through the same area some hours later (Cheney and Seyfarth, 1990) but systematic experimental tests would be required to verify this.

XI

THE PROBLEM OF DECEPTIVE SIGNALLING

There is probably no single issue that has generated more philosophical interest and debate than the possibility that animals are capable of behaving deceptively (Griffin, 1981, 1984, 1992; Ristau, 1983; Mitchell and Thompson, 1986; Whiten and Byrne, 1988; Byrne and Whiten, 1990, 1991). This is perhaps because successful deception requires a degree of cognitive complexity and flexibility in behaviour that is qualitatively distinct from that envisaged both in traditional ethological accounts and in behaviourist analyses. As such, strong evidence for deception in the vernacular sense narrows considerably the gap between the cognitive properties of non-human animals and those that we impute to other humans. Like the issue of language, the problem of deception bears directly on the degree of continuity between humans and non-human animals and on the question of human uniqueness.

I will focus especially on deceptive signalling in the specific sense of transmitting false information about external events. I shall argue that analyses of referential signals make a critical contribution to understanding whether this phenomenon occurs in the natural behaviour of animals.

It is important to be rigorous in separating the cognitive use of 'deception' from a purely functional one (Mitchell, 1986). The Batesian mimicry of the viceroy butterfly provides a classic example of functional deception, but this hardly encourages us to speculate about its mental state. It is not, however, always straightforward to deduce which of these senses is intended in published accounts of deceptive behaviour (see below).

There are several types of evidence for deception in animals. The first relies on the systematic collection and analysis of unique social interactions. These anecdotes are then assembled and examined to determine whether there are consistent trends (e.g., Whiten and Byrne, 1988). The second involves studies of inter-specific communication, focussing upon signals that are designed to affect the behaviour of potential predators (e.g., Ristau, 1983, 1991). And the third, which I will concentrate on here, involves intra-specific communication and the selective production of signals that are normally evoked by the approach of predators or the discovery of food. It is logical to separate these latter two data sets because there are likely to be important