Biological recognition is often modeled as involving discrimination of continuously-distributed (and continuously-perceived) traits according to decision thresholds. However, traits such as animal signals can be categorically distributed. Here, we test how such categorical distributions may influence fundamental trade-offs in signal recognition, using a brood parasite–host system involving identity recognition. The African cuckoo finch Anomalospiza imberbis parasitizes several host species, each of which has evolved inter-individual variation in egg appearance (“egg signatures”) that facilitates recognition and rejection of mimetic cuckoo finch eggs. We demonstrate that egg signature traits in one host species, the zitting cisticola Cisticola juncidis , are categorically distributed. Field experiments reveal that zitting cisticolas make fewer Type II errors (accepting parasitic eggs) and Type I errors (rejecting their own eggs) than hosts exhibiting continuous variation. This challenges the long-standing expectation from classification models, statistics, and signal detection theory that there must be a strict trade-off between these two error types. Individual-based simulations clarify mechanisms by which categorical variation can generate low error rates, especially when combined with “category-based rejection,” whereby hosts only reject eggs of different categories to their own. Our findings show that the categorical distribution and category-based perception of trait variation can shape error trade-offs and coevolutionary dynamics, which should inform studies on other mimicry or self/non-self recognition systems, including immune recognition. They also highlight the importance of quantifying trait distributions and how they are perceived, when understanding coevolution between deceivers and those they deceive.
Dixit et al. (Fri,) studied this question.