11

argue that the term 'leopard alarm' is misleading, because it connotes a greater degree of referential specificity than can be supported by the empirical data. Such criticism would, however, be unfair, because it is clear that the term 'leopard' was intended as a shorthand description for a category of eliciting stimuli of which large spotted cats are, if not the prototypical exemplar, at least the most common (Seyfarth et al., 1980b). Nevertheless, there is the real danger that the terms selected to describe eliciting conditions may reveal more about human linguistic conventions and about the categorization behaviour of the investigators than about the information content of the signals being studied. Such labels, when applied prematurely, may also affect the choice of future experiments (Snowdon, 1990). If a class of signals is labelled 'food calls' then we are perhaps less likely to conduct studies involving qualitatively different stimuli, even though they might reveal behaviour that would fundamentally alter our interpretation, such as calling during affiliative social interactions in the absence of food. Labels are rarely neutral; they reflect our intuition about the properties of the system being studied and they have the potential to constrain the direction of our research.

There are several possible solutions to the labelling problem. The first is suggested by the study of animal signals that do not have the property of functional reference. There is consequently no temptation to apply linguistic terms. The songs of birds or the calls of anuran amphibians are identified with labels that are either completely arbitrary or perhaps reflect some aspect of acoustic structure. This has, of course, been the traditional practice in studies of primate vocal behaviour, so that we read of 'rough grunts' in chimpanzees (Marler, 1976), 'threat-alarm-barks' in vervets (Struhsaker, 1967) and 'isolation peeps' in squirrel monkeys (Snowdon et al., 1985). It is difficult, however, to employ structural terms when faced with a vocal repertoire containing sounds for which we cannot readily coin onomatopoeic descriptions. The chicken's vocal repertoire includes a number of pulsatile sounds that do not lend themselves to economical prose description, although the eliciting conditions (food, social affiliation, and approach of predators) do. One way to avoid this problem would be to use completely arbitrary terms that were neutral with regard to issues of call meaning - so that we might speak of pulsatile call1, pulsatile call2, and pulsatile call3. The obvious cost of such an approach is that our prose would become opaque and probably quite incomprehensible to the non-specialist. The demands of terminological precision are thus to some degree incompatible with those of eloquence.

The most straightforward solution might be to inform the reader at the outset about the way in which terms are to be used. We will then be able to discriminate between: (i) 'leopard' as a shorthand description for a category of calls with particular acoustic characteristics, elicited by, among other things, leopards, and (ii) 'leopard', a signal evoking mental representations of spotted cats in the minds of receivers, and sharing with some words the property of denoting a precisely-defined category of external objects. Clearly these two usages make very different claims about the cognitive underpinnings of the signal being studied, and they will be evaluated according to correspondingly different logical and empirical criteria.

VII.UNDERSTANDING VARIATION IN SIGNAL STRUCTURE: THE IMPORTANCE OF LEVELS

I have so far been discussing analyses that are concerned with variation at the level of call type. These involve identification of the eliciting event and of the factors mediating transitions from one signal class to another. Much less is known about the factors responsible for variation in signal structure within call type (Marler et al., 1992), although efforts have been made to ensure that such variation is represented in the exemplars chosen for use in playback experiments (e.g., Seyfarth et al., 1980b; Evans et al., 1993a). I suggest that the distinction between within-type and between-type variation is an important one, because systems may well prove to be referential at one level but not at the other.