Agonistic screams and the classification of dominance relationships: are monkeys fuzzy logicians?

Scream vocalizations of group-living rhesus macaques provoked by higher-ranking aggressors were examined in two contexts: encounters in which the rank difference between opponents was either large or small. Such vocalizations are important in eliciting support from the caller's allies in the group (usually matrilineal kin). Five acoustically distinct recruitment screams encode specific information about features of the agonistic context, for example, relative rank of the opponent and the severity of the attack. Against higher-ranking opponents, noisy screams are most likely to be given during encounters that involve contact aggression, and tonal or undulated screams are most likely for non-contact aggression. We predicted that scream bouts directed to small rank difference opponents would be more likely to comprise calls from scream classes other than the expected type, and that scream bouts from large rank difference encounters would be more in accord with the expected class. For large rank difference bouts, on average 74.2% of calls were of the expected class; for small rank difference bouts, a significantly smaller per cent (53.8%) of calls was accurately predicted. In the large rank difference encounters, 65.6% (55/84) of bouts were composed entirely of expected scream types, but only 47.6% (39/82) were uniformly correct for small rank difference encounters. Large rank difference interactions also yielded significantly fewer bouts where none of the screams produced by the victim of attack belonged to the predicted class. We suggest that a multi-valued or fuzzy logic system, that is one with more than two truth values, might be a more realistic way to conceptualize the categorization of certain referents of monkey vocalizations. (C) 1998 The Association for the Study of Animal Behaviour.