5

difficult to distinguish such signals from those that had external referents (category [c]). I suggest, however, that it is not necessary to treat behavioural referents and external referents as mutually exclusive interpretations. Vocalizations such as alarm calls make available many types of information about the sender; a partial list might include caller size, individual identity and affective state. It seems probable that such signals allow conspecific receivers some success in predicting the subsequent behaviour of the sender. This is perhaps especially likely to be true in species that have stable social groups with individually-distinctive vocalizations.

There are a number of other complications. For example, it is particularly difficult to determine whether signals should be thought of as denotative (i.e., as labels for stimulus categories) or imperative (i.e., as instructions describing appropriate responses) (Cheney and Seyfarth, 1990, 1992; Baron-Cohen, 1992; Marler et al., 1992). Attempts to explore the 'meaning' of animal signals also tempt us to address difficult philosophical issues, such as the level of intentionality required to explain the observed behaviour (e.g., Dennett, 1983; Cheney and Seyfarth, 1990) and whether animals are aware of their own knowledge or that of their companions (Allen, 1992; Armstrong, 1992; Schull and Smith, 1992; Snowdon, 1992). Problems of this kind are not unique to the study of animal communication; they are also characteristic of work on the behaviour of preverbal human infants (Marler et al., 1992).

I shall focus instead on questions that are clearly accessible to experimental investigation. Systematic studies of animal signal systems can only establish that our subjects behave as if  their vocalizations encode information about events in the external environment. The term 'functional reference' has been coined to describe this property. It acknowledges the constraints inherent in analyses of animal signals, including the difficult distinctions described above (Marler et al., 1992), and is intended to be neutral about philosophical issues that are not addressed directly by empirical evidence.

IV.

RECOGNIZING FUNCTIONAL REFERENCE

The discovery of referential signals in the natural behaviour of nonhuman primates and birds invites comparative and developmental studies. An essential prerequisite for such a program is the development of agreed criteria for recognizing the property of functional reference. Recent theoretical papers have suggested that this should involve consideration both of the caller's behaviour and of the effects of the signal on companions (Marler et al., 1992; Macedonia and Evans, 1993). Studies of signal production and perception thus assume equal importance.

The key considerations with regard to production are that referential signals should be structurally discrete and that they should have a degree of stimulus-specificity. Eliciting stimuli should belong to a coherent category, although the absolute size of this category, and hence the degree of referential specificity, could vary considerably. Variation of this sort is also characteristic of human speech, which provides paradigmatic examples of referential signalling. We are able to denote individuals and also to discuss much larger groups that are delineated by characteristics such as age or occupation. Despite differences in the number of possible eliciting stimuli, the terms 'Mary' and 'university professor' are both unambiguously referential. The key point is thus not the absolute level of specificity, but rather the relationship between event class and signal type. We would not expect the same class of referential signal to be produced in response to stimuli that are clearly drawn from qualitatively distinct categories.

The importance of this distinction is illustrated by work on California ground squirrels. These sciurid rodents have a complex series of alarm calls which form a continuum from broad-band 'chatters' to tonal 'whistles' (Owings and Virginia, 1978). Whistles are usually produced in response to raptors, whereas chatter calls are evoked by terrestrial carnivores. However, there