Triadic awareness, or knowledge of the relationships between others, is essential to navigating many complex social interactions. While some animals maintain relationships with former group members post-dispersal, recognizing cross-group relationships between others may be more cognitively challenging than simply recognizing relationships between members of a single group because there is typically much less opportunity to observe interactions between individuals that do not live together. We presented acorn woodpeckers (Melanerpes formicivorus), a highly social species, with playback stimuli consisting of a simulated chorus between two different individuals, a behavior that only occurs naturally between social affiliates. Subjects were expected to respond less rapidly if they perceived the callers as having an affiliative relationship. Females responded more rapidly to a pair of callers that never co-occurred in the same social group, and responded less rapidly to callers that were members of the same social group at the time of the experiment and to callers that last lived in the same group before the subject had hatched. This suggests that female acorn woodpeckers can infer the existence of relationships between conspecifics that live in separate groups by observing them interact after the conspecifics in question no longer live in the same group as each other. This study provides experimental evidence that nonhuman animals may recognize relationships between third parties that no longer live together and emphasizes the potential importance of social knowledge about distant social affiliates.

Material and Methods

(a) Study site and population monitoring

We collected all data at Hastings Natural History Reservation in central coastal California, where the acorn woodpecker population has been the subject of a long-term study since 1971 and >95% of the individuals are color-banded (MacRoberts and MacRoberts 1976; Koenig 1981a). Each year, approximately 50 social groups are monitored, and a census is conducted of each group approximately every 8–10 weeks. Subjects for this experiment were 26 wild adult acorn woodpeckers, including 9 breeder females, 5 helper females, 7 breeder males, and 5 helper males from 18 social groups. Experimental trials were conducted from 19 Jul to 27 Nov 2017 and from 6 May to 8 Jul 2018.

(b) Experimental design

We conducted a playback experiment with a violation-of-expectation paradigm following a previously published protocol (Pardo et al. 2018). Woodpeckers were presented with playback stimuli consisting of waka calls recorded from two different individuals, overlapped artificially to simulate two birds calling simultaneously. Waka calls are individually specific, affiliative vocalizations that are frequently given in an overlapping chorus between two or more members of the same group, but rarely given between individuals with no affiliative relationship (MacRoberts and MacRoberts 1976; Yao 2008). If the two overlapping callers in a playback stimulus had no affiliative relationship that the subjects recognized, the playback stimulus was expected to violate the expectations of the subjects and the subjects were expected to respond more strongly by reacting more rapidly and approaching the speaker more closely. Conversely, if the two overlapping callers in a playback stimulus had an affiliative relationship that the subjects recognized, the subjects were expected to respond less strongly (Pardo et al. 2018).

We presented subjects with playbacks from the following 5 treatment categories and attempted to present each subject with all 5 treatments, although this was not always possible (Supplementary Table S1). In all cases, the callers used in each playback stimulus were unrelated to and had never lived in the same group as the subject. The order of presentation was balanced using an incomplete Latin square design (Supplementary Table S1), and playbacks to the same group were spaced apart by 2–48 days (median = 4 d) to avoid habituation. Playbacks to groups within 250 m of one another were also spaced apart by at least 2 days.

(T1) Related callers–currently live together: two related callers that lived in the same social group at the time of the experiment.

(T2) Related callers–last together after subject fledged: two related callers that formerly lived in the same social group but ceased to do so because of dispersal or death of one individual 1.0­–5.8 years (mean 2.1 y) prior to the experiment, but after the subject had fledged. Thus, the subject would have had the opportunity to observe the callers living together in the past during extraterritorial forays to the callers’ group (Barve et al. 2020).

(T3) Related callers–last together before subject hatched: two related callers that formerly lived in the same group, dispersed into separate groups before the subject hatched in the nest and 2.2–6.4 years (mean 4.5 y) prior to the experiment, and were both confirmed to be alive after the subject had fledged. Thus, the subject would not have had the opportunity to observe the callers living together but could have observed them visiting one another post-dispersal.

            (T4) Related callers–never lived together: two genetically related callers that never lived in the same group and were both confirmed to be alive after the subject fledged.

            (T5) Unrelated callers–never lived together (control): two unrelated callers that never lived in the same group and were both confirmed to be alive after the subject fledged.

            Because of the difficulty of obtaining playback-quality recordings from a canopy-dwelling species with unpredictable calling patterns, for some playback stimuli we were forced to use the calls of individuals that had died or disappeared from the study area prior to the experiment. Preliminary analyses indicated that subjects responded more quickly to playback stimuli containing the call of a dead or missing individual (Supplementary Table S2; Figure S1). Therefore, we excluded all trials that used the call of a dead or missing individual from the analyses presented here (see Supplementary Materials for analyses including both live and dead callers). Most (19 of 24) playback stimuli for treatment T2 (related callers–last together after subject fledged) contained the call of a dead or missing individual, so this treatment was excluded from analysis. After excluding the aforementioned trials, there remained 24 playbacks of related callers–currently live together, 11 playbacks of related callers–last together before subject hatched, 13 playbacks of related callers–never lived together, and 13 playbacks of unrelated callers–never lived together, with a total sample size of 25 subjects (8 breeder females, 5 helper females, 7 breeder males, and 5 helper males) from 17 different groups (Supplementary Table S1).

Playback stimuli contained only recordings from callers of the same sex as the subject to increase the likelihood that the subjects would respond (Hannon et al. 1985). For the three treatment categories consisting of a pair of related callers, relatedness between the two callers in each playback stimulus was calculated using a pedigree, which was constructed from parentage assignments based on microsatellite markers (J. Haydock, unpublished data). Mean relatedness (r_(T1)=0.36, r_(T3)=0.38, r_(T4)=0.38) did not differ significantly across these three treatment categories (ANOVA, F_2,45=0.10, P=0.91).

To increase the likelihood that subjects had the opportunity to become familiar with all the callers, we presented subjects with callers from territories as near as possible to the subject’s territory. The mean ± standard deviation (SD) distance between the territory centroids of the subjects and the callers was 435 ± 233 m and did not differ significantly across treatment categories (ANOVA, F_3,57=0.43, P=0.73). As acorn woodpeckers make daily forays to the territories of other groups with a mean foray distance of 500–600 m (Barve et al. 2020), subjects were likely familiar with all or most of the callers with which they were presented.

Whenever possible, we constructed each playback stimulus from a unique pair of call exemplars to minimize pseudoreplication (Supplementary Table S1). All of the playback stimuli for the treatments “related callers–last together before subject hatched” and “unrelated callers–never lived together” consisted of a unique pair of call exemplars that was only used once, although we sometimes used multiple different exemplars of the same callers. However, only 14 of 24 stimuli for “related callers–currently live together” and 8 of 13 stimuli for “related callers–never lived together” consisted of a unique pair of call exemplars.

The speaker was placed in a tree 40–50 m away from the focal bird, and playback volume was standardized at 100.1 ± 1.3 dB re 20 µPa at 1 m, which is near the upper end of the range of natural waka calls (Pardo et al. 2018). Playback stimuli consisted of 1 min of background noise with a fade-in, followed by the two overlapping calls, then 30 sec of background noise, then the same two overlapping calls again, and a final 10 sec of background noise with a fade-out. The trial was aborted if the focal bird flew away before the first set of overlapping calls began. As waka calls consist of a variable number of repeated notes, we could not standardize the duration of the playback stimuli without heavily modifying the calls. The duration of the overlapping call chorus within each playback stimulus ranged from 3.7 to 7.0 sec and did not differ significantly among treatments (ANOVA, F_3,57=0.21, P=0.89).

(c) Measuring response to playback

We videotaped the subject during each trial and measured the following 6 response variables within a 3-min period beginning with the start of the playback: latency to the first “reaction” (vocalizing, flying to higher vantage point, or flying toward speaker), latency to the first “positive flight” (flying up to higher vantage point or toward speaker), latency to the first approach to the speaker, latency to the closest approach to the speaker, distance of the first approach to the speaker, and distance of the closest approach to the speaker. For latency variables, if the behavior of interest did not occur within the allotted 3 min, the latency was assigned the maximum possible value of 180 sec and marked as “censored”. We were blind to the experimental condition in each trial until all the videos had been scored. Before the trials, we measured distances between the speaker and various nearby landmarks using a transect tape and used these measurements to estimate approach distances to the nearest 5 m during the playbacks.

We used only latency to react, latency to positive flight, and distance of first approach in the analysis, as the other response variables were highly correlated (Pearson’s r > 0.80) with at least one of these variables. We used both latency to react and latency to positive flight because it was unclear whether reaction strength was best measured by flight behaviors and vocalizations or by flight behaviors alone. We could not measure latency to vocalize as a stand-alone variable because there were many trials in which the subject flew toward the speaker and out of sight before the first vocalization was heard, and thus it was impossible to know whether the vocalization was produced by the subject or another individual.

(d) Statistical analyses

            All analyses were conducted in R 3.6.3 (R Core Team 2020), and the significance level was set to 0.05 for all tests. We used mixed-effects Cox proportional hazards regression, which accounts for censored observations, in the package coxme (Therneau 2019) to analyze latency to react and latency to positive flight , and a linear mixed model in the packages lme4 (Bates et al. 2015) and lmerTest (Kuznetsova et al. 2017) to analyze distance of first approach.

The residuals for distance of first approach were not normally distributed (Shapiro-Wilk test, W=0.95, P=0.02), so we rank transformed distance of first approach before running the linear model (Shapiro-Wilk test on rank-transformed model: W=0.97, P=0.11). Each model contained treatment, sex, treatment*sex, order of presentation, and days since previous playback as fixed effects, and individual ID as a random effect. Days since previous playback represented the number of days since the last playback to the same group and was coded as 0 for the first playback to a given group. We compared all treatments to the control (unrelated callers–never lived together) separately for females and males in the package “emmeans” (Lenth 2019) using Dunnett’s method to adjust for multiple comparisons within a given model.

The second spreadsheet explains each variable in the dataset. Some variables have some missing values.