Numerous studies have shown that territorial animals exhibit less aggression in response to neighbours than to strangers, a phenomenon known as dear enemy effect. The influence of acoustic features, such as song type sharing and repertoire sizes, in neighbour recognition has been widely documented in male songbirds. However, few studies have focused on duetting species, and particularly on those where pairs have pair-specific duet codes (consistent associations of their individual phrase types). Given that each pair in the population can have a unique repertoire of duet types, duet codes have been hypothesized to enhance discrimination. In this context, we tested for evidence of neighbour recognition and duet code discrimination in two closely related species of neotropical wrens, the riverside wren, Cantorchilus semibadius, and the canebrake wren, C. modestus zeledoni. Although both species have moderately large repertoires, riverside wrens have higher levels of phrase type and duet type sharing across the population. We compared the approach and vocal responses of focal pairs to three playback treatments: neighbours’ correct duet type, neighbours’ incorrect duet type, and a strangers’ duet type. We found that riverside wrens displayed a stronger response to the strangers’ playback than to both neighbours’ playbacks, whereas no differences among treatments were found in canebrake wrens. Given that both species exhibited similar levels of aggression during neighbour playbacks, regardless of whether the correct duet code was used, our findings suggest duet codes do not facilitate neighbour recognition. We conclude that the function of duet codes in these species might be more closely related to intra-pair communication. Finally, we suggest that the level of dear enemy effect a species exhibits depends on ecological factors that influence the perceived level of threat of territory intruders.

We conducted the experiment with riverside wrens at Osa Conservation’s Piro field station in southern Costa Rica (8°24'N, 83°20'W). We have studied this population since 2013 and we have colour banded almost 130 individuals at the study site. We conducted the experiment with canebrake wrens at La Selva Biological Station in northern Costa Rica (10º26'N, 83º59'W). We have studied this population since 2012 and we have colour banded almost 60 individuals at the study site.

We presented focal pairs of each species with three playback treatments: a correct duet type from a neighbouring pair, an incorrect duet type from the same neighbouring pair, and a duet type from a stranger pair. The neighbours’ correct duet type and the strangers’ duet type were used to test if birds display the ‘dear enemy’ effect and respond more aggressively to strangers’ than to neighbours’ vocalizations. The neighbours’ correct and incorrect duet type were used to test if birds pay attention to the duet code used by their neighbours. The stimuli consisted of 10 bouts of duets with one minute of silence after the first 5 bouts. Each bout consisted of 1 male introductory phrase and 7 phrase types of each sex (I(FM)₇) and was separated from the next bout by 10 seconds. 

The trials were conducted with 12 pairs of riverside wrens (20 out of the 24 individuals were previously captured and banded) from the 15^th of April to the 4^th of May 2016; and with 13 pairs of canebrake wrens (21 out of the 26 individuals were previously captured and banded) from the 10^th to the 28^th of May 2016 (Fig. 2). All trials took place between 0530-0800 h to minimize effects of time of day on the behavioural responses to playback. 

We investigated focal pairs’ responses by measuring approach and vocal variables designed to assess both individual and collective behaviours (Logue and Krupp 2016). The individual behaviours considered were: (1) female mean distance to the closest speaker, (2) male mean distance to the closest speaker, (3) female closest distance to any speaker, (4) male closest distance to any speaker, (3) female song initiation, (4) male song initiation, (5) female responsiveness, and (6) male responsiveness. The collective behaviours considered were: (7) duet rate (duets/min) and (8) duet length (number of FM repetitions per duet). Initiation was measured as the number of instances an individual started singing (solos + initiated duets). Responsiveness was defined as the propensity of each sex to answer its partner’s song. We assessed responsiveness by calculating the proportion of the partner’s songs that were joined to form duets (e.g. female responsiveness = male led duets/(male solo songs + male led duets)). 

For both approach and vocal traits, mixed-effects models were adopted, in which territory and trial order were set as random effects (in which the effect of order was nested within that of territory). Distance-related traits, as well as responsiveness, were modelled using Gaussian distributions, whereas duet rate, duet length, and initiation were assumed to follow negative binomial distributions. The latter probability distribution was chosen for count data because it models overdispersion, which was found to characterize the data. In the models for duet rate, an offset for the duration of the trial was included, such that the estimated rate is in duets per minute rather than in duets per trial duration. We modelled linear mixed models with R package lme4 (Bates et al. 2015) and negative binomial mixed models using R package glmmADMB (Skaug et al. 2013).