The ontogenetic ritualization of bonobo gestures

Abstract

Great apes communicate with gestures in flexible ways. Based on several lines of evidence, Tomasello and colleagues have posited that many of these gestures are learned via ontogenetic ritualization—a process of mutual anticipation in which particular social behaviors come to function as intentional communicative signals. Recently, Byrne and colleagues have argued that all great ape gestures are basically innate. In the current study, for the first time, we attempted to observe the process of ontogenetic ritualization as it unfolds over time. We focused on one communicative function between bonobo mothers and infants: initiation of “carries” for joint travel. We observed 1,173 carries in ten mother–infant dyads. These were initiated by nine different gesture types, with mothers and infants using many different gestures in ways that reflected their different roles in the carry interaction. There was also a fair amount of variability among the different dyads, including one idiosyncratic gesture used by one infant. This gestural variation could not be attributed to sampling effects alone. These findings suggest that ontogenetic ritualization plays an important role in the origin of at least some great ape gestures.

Appendix

A DICE-coefficient (D_C) measures the similarity between the repertoires of two individuals: D_C = (2C _xy )/(R _x  + R _y ), where C _xy is the number of initiating behaviors common to two individuals (x and y), R _x is the number of initiating behaviors in the repertoire of individual x, and R _y is the number of initiating behaviors in the repertoire of individual y. A DICE-coefficient ranges from 0 to 1, where a value of 0 means that two individuals have no initiating actions in common and a value of 1 means that they have identical repertoires.

We tested whether the level of similarity in the action repertoires of dyads in the same class (mother–mother and infant–infant) differed from the level of similarity in the action repertoires of dyads in different classes (mother–infant). To do this, we first averaged the DICE-coefficients for each of the three groupings of dyads and used as a test-statistic the sum of the squared deviations of these averages from the mean of the averages. To test whether this test-statistic was significantly larger than chance expectation, we used a permutation procedure (using 1,000 permutations) applying the same randomization technique as in a Mantel test (Sokal and Rohlf 1995). The P value was determined as the proportion of permutations revealing a test-statistic at least as large as that of the original data. This test was run using a function written for R (R Development Core Team 2011) by Roger Mundry.

We ran a Generalized Linear Mixed Model (GLMM) in order to determine whether there were any trends in the frequencies in which mothers and infants initiated carries (Baayen 2008). Into this model, we included age (z-transformed to a mean of zero and a standard deviation of one), initiator (levels mother or infant), and their interaction as fixed effects. We included the particular day and the identity of the mother and infant as random effects. To account for varying observation times per day (ranging from less than 1 h to more than 5 h), we included it (log-transformed) as an offset term in the model. The model was fit assuming a Poisson error structure and with a log link function. Overdispersion was no obvious issue (dispersion parameter: 1.04; c ² = 668.5, df = 642, P = 0.227). We established the significance of the full as compared to the null model (comprising only the random effects and the offset term) by using a likelihood ratio test (Dobson 2002). The model was fit in R using the function lmer of the R package lme4 (Bates et al. 2011).

We ran a second GLMM in order to determine whether there was a correlation between an infant’s gestural repertoire size and an infant’s “motivation” to initiate carries, which we defined as the proportion of infant-initiated carries relative to all agent-initiated carries in a dyad. As presented in the results, the infants’ tendencies to initiate carries increased with their age and hence needed to be controlled. Thus, we ran the GLMM with the infants’ age as a predictor and their frequency of initiation as a response. To further control for observation effort and the total number of initiations per mother–infant dyad and day, we included these two variables (log-transformed) as effort terms into the model. For such effort terms, no coefficient was estimated (but was just set to one) because their effect is trivial. We also included the identity of each infant as a random effect to avoid confounding the effect of age with differences between infants.

We used the derived coefficient for age as well as the respective total number of initiations and the observation effort per infant and day to determine the expected number of initiations per infant and day. We then used the average differences between its actually observed and expected numbers of initiations, averaged across the period until its final repertoire was reached. Also, we residualized the final repertoire size as it weakly related to the total observation time per infant (Spearman’s rho = 0.51). Hence, we first estimated the relationship between total observation time and final repertoire size assuming the relationship to be: final repertoire = c₁ × (1 − exp(c₂ × observation time)). We then took the difference between observed and expected final repertoire as a measure of repertoire size. The GLMM was fitted in R using the function lmer of the R package lme4 with Poisson error structure and log link function. The relationship between total observation time and final repertoire size was estimated using the R-function nls. For the correlation between residualized initiation rate and residualized final repertoire size, we used Spearman’s correlation coefficient.