12

Consider the properties of chicken food calls. These signals appear to be evoked by food items and by stimuli that reliably predict the availability of food. They are clearly not simply dependent upon social or sexual interaction (Evans and Marler, 1994). Analyses at the level of call type thus suggest that food calls predict events in the external environment (i.e., that they are functionally referential). There is also variation in the temporal characteristics of call bouts. The rate at which males produce food calls reflects their preference for a food stimulus, so that it is highest for live invertebrates and lower for less preferred items, such as their regular ration (Marler et al., 1986a). This relationship was initially considered to be evidence for highly- specific referential signalling about object characteristics (Marler et al., 1986a), but subsequent analyses have demonstrated that the call rate also co-varies with the speed at which males perform an operant task to obtain food. Males food called and key-pecked vigorously at the beginning of test sessions, but both of these responses decayed with repeated food deliveries (Evans and Marler, 1994). This correlation between instrumental performance and food calling suggests a more parsimonious explanation for the earlier results, which is that temporal variation may simply reflect the male's motivational state with regard to food. Analyses of food-associated calls in cotton-top tamarins suggest that this system may have similar properties. While 97% of tamarin calls were produced in the presence of food (Snowdon, 1993b), the rate of calling was dependent on idiosyncratic individual preferences for particular food types, and therefore most likely reflected a motivational response (Elowson et al., 1991).

Determining whether food-associated calls are functionally referential proves to be more complicated than we might at first suppose. I suggest that the answer obtained will be, in part, dependent upon the level of analysis selected. It is possible for transitions between call types to encode stimulus class while within-category variation in signal structure remains principally dependent upon motivational state. There is thus the very real danger that results from studies of fine-grained variation will appear to contradict those obtained from analyses of differences between call types when, in fact, these two data sets are not comparable because they concern quite different phenomena.

The challenge for future studies of fine-grained variation in referential signals will be to determine whether there are structural parameters that vary with stimulus characteristics, yet are independent of those that serve as vehicles for affect (Marler et al., 1992). Addressing this issue will require the systematic exploration both of the consequences of manipulating stimulus attributes and of variation in the internal state of the sender. Such experiments would begin to redress the relative neglect of motivational or 'conotative' factors in studies of referential signalling (Owings, 1994).

VIII. DESIGN OF REFERENTIAL SIGNALS

Analyses of the information content of referential signals have revealed that some systems have a surprising degree of specificity. This finding does much to explain both the calling behaviour of senders and the responses of receivers. Such work does not yet, however, allow us to make predictions about signal design. Clearly, we would expect calls that encode information about very different classes of events to have contrasting structure, and this is often true. But analyses of meaning provide little else in the way of insights into the physical form of referential signals. Why, for example, are vervet leopard alarms made up of a rapidly-repeated series of pulses with a relatively low fundamental frequency, while snake alarms are delivered at a lower rate and have a much higher dominant frequency, sometimes exceeding 16 kHz (Seyfarth et al., 1980b)? No-one has yet attempted a comprehensive model of the relationship between stimulus category and signal structure along the lines of Morton's (1977) motivational-structural rules, and given how little we still know about the details of referential signalling, such an effort would almost certainly be premature.