Social learning is an essential component of normal song development for oscine passerines (Beecher and Brenowitz, 2005). In many species, birds that are not exposed to tutor song during the sensitive phases of song ontogeny develop atypical vocalizations, exhibiting unusual note structures, decreased stereotypies, abnormal song length, and other temporal abnormalities (Price, 1979; Marler and Sherman, 1985; Chaiken et al., 1993; Feher et al., 2009; Kagawa et al., 2014). Similarly, birds that are tutored by heterospecifics may incorporate song features of the tutor species in their vocalization, rather than solely producing species-typical songs (Johannessen et al., 2006; Eriksen et al., 2009; Mann et al., 2020). In this way, cultural inheritance of vocal behavior can lead to vocalizations that are shaped by a bird’s social environment (e.g., Greig et al., 2013).

However, a number of other factors also influence song development in juvenile birds. Genetic factors can guide song learning and development. For example, in canaries (Serinus canaria), genetic differences show complex interactions with learning and song production, influencing the proportion of low- and high-pitched syllables (Wright et al., 2004; Mundinger, 2010; Mundinger and Lahti, 2014). Genetic factors can also interact with the environment. In Bengalese finches, juveniles produced more accurate imitations of an experimental tutor’s song if songs were played back at a tempo that was similar to their genetic father’s song (Mets and Brainard, 2018, 2019), and the influence of genetic background and environment differed between passively and socially tutored birds (Mets and Brainard, 2018). Other heritable traits, e.g., morphology and neural anatomy, can influence song production. Morphological characteristics, such as body size (Kirschel et al., 2009; Kagawa and Soma, 2013; Derryberry et al., 2018; García and Tubaro, 2018), beak morphology (Podos et al., 2004; Kirschel et al., 2009; Derryberry et al., 2018; García and Tubaro, 2018), and syrinx morphology (Elemans et al., 2015; Christensen et al., 2017), are often correlated with song characteristics. The developmental environment, mediated by social interactions and resource availability, also plays a key role in song development. As song production may incur neural costs during development (Gil and Gahr, 2002), early developmental stress, such as sibling competition (Soma et al., 2006) or nutritional stress (Nowicki et al., 1998), may influence adult songs. Social interactions may guide song production through fraternal inhibition (Tchernichovski and Nottebohm, 1998) and social reinforcement from parents (Carouso-Peck et al., 2020). Finally, both laying order and maternal androgens may contribute to song development (Soma et al., 2009).

Bird song is thought to advertise the relative quality of the singer and to that effect plays an important role in sexual selection (Gil and Gahr, 2002). Songs are multi-faceted signals with learned and unlearned features. Consequently, different aspects of song can reveal different information about the singer’s quality. The aspects of quality these traits convey depend partly on their pattern of inheritance. Learned or environmentally influenced traits may reveal information about developmental environment or learning ability (Nowicki et al., 2002a,b; Boogert et al., 2008; Zann and Cash, 2008), whereas genetically inherited traits may signal “good” genes, which will be inherited regardless of tutor (Hasselquist et al., 1996; Forstmeier et al., 2009). Both types of traits may also inform potential mates of direct benefits, such as adaption to the local environment (Podos and Warren, 2007; Snowberg and Benkman, 2007; Badyaev et al., 2008; Branch and Pravosudov, 2015) or ability to provision offspring (Buchanan and Catchpole, 2000; Halupka and Borowiec, 2006; Bartsch et al., 2015). Of the various song features, significant attention has been paid to three categories: song structure, acoustic characteristics of notes, and song performance measures, which demonstrate complex inheritance patterns and provide a wide range of information about singer’s quality (Table 1).

Although song inheritance patterns are well-understood for a handful of model species, whether these patterns replicate more broadly across species, and particularly in rapidly diverging lineages and species of conservation concern, is not known. We studied patterns of song inheritance in the Java sparrow (Lonchura oryzivora), an estrildid that is endangered in its native range but common in captivity and invasive in some locations (BirdLife International, 2018). We examined an existing song library (Kagawa and Soma, 2013; Ota and Soma, 2014; unpublished data) with songs from multiple generations of father-son pairs for which the genetic pedigree was known. Some birds in our dataset were reared by their genetic fathers, and others were cross-fostered by social fathers with songs that were also included in the dataset. We mined our data for patterns of social learning, genetic heritability, and environmental effects on the development of song structure, note spectral and temporal characteristics, and vocal performance measures, and we quantified evidence for the patterns that we found. Based on previous findings in this and other species, we expect that (1) song structure will be socially inherited, (2) spectral and temporal characteristics of notes will be socially learned but may also have some non-learned components, and (3) vocal performance, measured as vocal consistency, will be genetically inherited.

For the whole song analyses, we found strong evidence that features of song structure are socially learned; son’s songs were similar to those of their social father (Table 2). For manually assigned note types, there was a strong positive relationship between the social father’s and son’s songs for all measures with a large associated effect size (all p < 0.001, Table 2; Figure 6), i.e., for the structural features measured, sons produced songs with features closely resembling those of their social father. Song structure at the level of computer-assigned note subtypes was also learned (repertoire size, p < 0.0001; Shannon diversity, p < 0.0001; first order differential entropy, p = 0.0014; song linearity, p = 0.0027; Figure 6), but the effect sizes were generally smaller than those reported for manually-assigned note types. There was no strong evidence that age of the bird at time of recording influenced any structural measure, although positive relationships were found for the number of notes (p = 0.068), the second order differential entropy (p = 0.056) when considering manually assigned note types, and the first order differential entropy (p = 0.024) when considering note subtypes. There was no evidence that genes or the rearing environment influenced song structure (Table 2). In this analysis, we treated the one note type that was not reliably clustered into subtypes as if it were a single subtype, but our results are qualitatively unchanged if we conduct the same analysis using the original computationally assigned subtypes (Supplementary Information 1).

For the individual note analyses, we found strong evidence that acoustic characteristics of note types are learned. For all measures considered, there was a strong positive relationship between the notes produced by social fathers and those of sons (all p < 0.001), with large effect sizes (Table 3; Figure 6); within categories, sons produced notes that were similar to those of their social father. There was no evidence for a relationship between age at time of recording and any of the acoustic characteristics considered (all p > 0.1). There was evidence for an effect of clutch for time median (p = 0.014), time IQR (p = 0.0070), and mean dominant frequency (p = 0.041), indicating that birds from the same clutch were more similar than expected by chance alone (Table 3). We found no evidence for a genetic effect on any acoustic characteristic (Table 3).

We found no clear evidence for social learning of vocal consistency (Table 4; Figure 6), but we found a trend suggesting that social fathers with more consistent note durations had sons with more consistent note durations (p = 0.094). Older birds produced note subtypes with more consistent durations than younger birds (p = 0.0022). However, there was a trend in the opposite direction for spectral cross correlation – older birds appeared to produce less consistent note subtypes (p = 0.066). We found no evidence for an effect of age for any other consistency measure (Table 4). There was evidence that the random effects influenced consistency measures in all cases except the variance of frequency change (Table 4). For the variance of mean dominant frequency (p = 0.0033), the dynamic time warping distance (p = 0.0025), and spectral cross correlation (p = 0.015), birds from the same clutch were more similar than we would expect by chance alone. For the variance of duration, there was a random effect of either clutch or relatedness (p = 0.0056). However, because birds from the same clutch were always genetic brothers in our data, natal clutch and genetic relatedness are correlated, and we cannot determine which of these explains the effect. Visual inspection of the scatterplots (Figure 6) revealed three potentially influential points, which were all birds from a single clutch. Results remain qualitatively unchanged if we conduct the same analyses excluding these individuals (Supplementary Information 2).

We examined the roles of cultural and genetic inheritance in shaping song phenotypes. There was strong evidence for the social learning of song structure and the acoustic characteristics of notes. Sons’ song structure and note characteristics were similar to those of their social fathers. There was no effect of genetics, and with few exceptions, no effect of age. Here, and throughout, the effect of age may be limited as analyses were confined to the songs of adult birds, where age-related song changes may be slow to manifest or small in magnitude (James and Sakata, 2014). For some features, there was an effect of the clutch in which the bird was reared, with individuals from the same clutch more similar than expected by chance alone, indicating an influence of the developmental environment. For vocal consistency, we found no strong evidence of social learning or genetic heritability. However, vocal consistency was influenced by the age of the bird and by the clutch in which the bird was reared, again indicating an influence of the developmental environment. Further empirical work will be required to confirm these patterns in Java sparrows and other species as our analyses were largely exploratory.

Sons resembled their social father in all measures of song structure, with no effect of genetic relatedness in any case, suggesting that these traits are culturally inherited within populations. Note repertoire size was similar in sons and their social fathers, as is the case for many other species (Grant and Grant, 1996; Takahasi and Okanoya, 2010; Soma, 2011; Labra and Lampe, 2018). We also found evidence for cultural inheritance of song complexity, as measured by linearity (Scharff and Nottebohm, 1991), and of higher order note sequencing, as measured by differential entropy. Similar sequence learning has been reported recently in Bengalese finch, and birds were more likely to learn note transitions commonly used by their social fathers (James et al., 2020), although we did not examine this in our dataset.

Regression coefficients for the social learning of fathers’ song structure were large, suggesting that there is limited regression toward the population mean in each generation. Thus, sons produced faithful copies of their social fathers’ songs, and differences among song lineages could persist for many generations. When considering note subtypes (i.e., those that were computationally assigned based on clustering of acoustic characteristics), the magnitude of the regression coefficients was smaller. This suggests that regression toward the population mean (as indicated by coefficients closer to zero) is greater when considering note subtypes, and that innovation may be more likely to involve changes among note subtypes than among note types.

Within note types, sons sang notes with similar acoustic characteristics to those of their social fathers. This may mean that birds learn how to produce notes of a particular type from their fathers, i.e., they learn the acoustic characteristic values of their social father’s note types, or it may mean that they learn which distinct note subtypes to produce from their fathers, which would be reflected in similar mean and variance of note type acoustic characteristics in social father-son pairs.

We found no evidence of genetic inheritance of acoustic characteristics; there was no effect of relatedness on any characteristic measured. This is in contrast to findings from a number of other species [e.g., zebra finch (Forstmeier et al., 2009) and Bengalese finch (Kagawa et al., 2014; Mets and Brainard, 2018, 2019)] where genetic differences underpin some differences in acoustic characteristics. It is possible that levels of genetic variation within the present laboratory population were not large enough to assess the genetic heritability of acoustic characteristics. Reduced genetic variability in laboratory compared to wild populations has been reported in other species (Forstmeier et al., 2007). In Bengalese finches and white-rumped munia, strain-specific differences in acoustic characteristics of notes are apparent (Kagawa et al., 2014). However, the two strains have high levels of disparity in morphology (Soma, 2005) and presumably genetics. A potential caveat of our analysis method (using a relatedness matrix) in determining heritability is that relatedness of founders in our population is unknown. As such, individuals may have been more closely related than suggested by our relatedness matrix.

In contrast to song structure and note acoustic characteristics, we found no strong evidence for the cultural inheritance of vocal consistency, which is a common measure of song performance (but see Supplementary Information 2). The vocal consistency of the social father did not predict his sons’ vocal consistency. However, contrary to our predictions, we also found no relationship between genetic relatedness and measures of vocal consistency; birds that were related did not have similar levels of vocal consistency. Although evidence was limited, some interesting patterns are apparent. In all models, effect of social father’s phenotype was positive, which is consistent with social learning. However, the effect sizes are small in comparison to those for song structure and the acoustic characteristics of notes, so, even if such effects exist, we expect that they will be small. There was some evidence that vocal consistency was related to age at recording; older birds had more consistent note duration across vocalizations than younger birds, although this was not the case for other measures of consistency. Increased vocal consistency with age has been reported in a number of studies across a broad range of bird species, for example (Kao and Brainard, 2006; Botero et al., 2009; de Kort et al., 2009; Rivera-Gutierrez et al., 2010). Improved vocal consistency with age may reflect greater opportunity to practice motor patterns involved in vocalizations (Sakata and Vehrencamp, 2012; Botero and de Kort, 2013). Thus, if age indicates good genes because the bird has survived or good parental care because the bird has experience, then vocal consistency may be an honest signal of quality in Java sparrows. However, there was no effect of age on the consistency of mean dominant frequency or frequency change for note subtypes within songs, and no effect of age on the dynamic time warping distance among notes of the same subtype. Similarly, James and Sakata (2014) also found no significant age-dependent changes in the mean or variability in a range of frequency-based syllable features in Bengalese finches, although this also included syllable duration. When considering spectral cross correlation within note subtypes, we report a trend in the opposite direction; older birds tended to sing less consistently than younger birds. Variance of note duration and distance by spectral cross correlation are negatively correlated, so it would be somewhat surprising for either of these patterns to appear by chance alone in the presence of the other. It is, therefore, not clear how vocal consistency relates to age in male Java sparrows. Our findings suggest the need for further examination of vocal consistency in this species, and for including age in analyses when considering similar questions in other species. There is evidence that differences in consistency are salient (de Kort et al., 2009) and can influence female preference (Woolley and Doupe, 2008), male reproductive success (Byers, 2007; Cramer et al., 2011), and social status (Botero et al., 2009) in some species. The consistency measures we studied carry information about a singer’s age and rearing environment (discussed below). However, determining whether these measures are salient to Java sparrows will require further empirical work.

For a range of measurements across the features examined, we found evidence for an effect of clutch. This reflects an impact of the social father or developmental environment independent of the social father’s song. This was especially relevant to vocal consistency, where the majority of measures indicated an important role for the rearing environment. Due to the nature of our dataset, it was not possible to disentangle a number of possible effects, as many factors might contribute to the clutch variable. One possibility is that clutch effects result from differences in the early developmental environment between nests (Nowicki et al., 2002b; Holveck et al., 2008). In the Bengalese finch, birds from larger, male-biased broods had lower song complexity than those from smaller broods (Soma et al., 2006). In this case song features, such as consistency, may be honest indicators of male quality. Vocal development incurs neural costs during early development (Gil and Gahr, 2002) when birds are likely to be exposed to stressors. High quality songs may, therefore, indicate lower stress levels during development (Nowicki et al., 1998; Nowicki and Searcy, 2004). As well as increasing developmental stress, large, male-biased broods may also result in fraternal inhibition of song learning. In zebra finches, birds with more male siblings had shorter motif durations and reduced note numbers compared to their tutors (Tchernichovski and Nottebohm, 1998). Social reinforcement of learning from parent birds may also play a role in clutch-specific differences in learning accuracy, as parents are likely to show variation in the levels of reinforcement provided. In zebra finches, social feedback from both the father and mother was correlated with song learning, with birds that received appropriate input producing more faithful copies of fathers’ songs (Carouso-Peck et al., 2020).

The strong influence of vocal learning, particularly of song structure and the acoustic characteristics of notes, has implications for the evolution and maintenance of song in the Java sparrow. Sons do not precisely copy their social fathers’ songs; there are differences, particularly when considering note subtypes, which may relate to improvisation or copying errors during learning. However, we report large effects of the social father’s phenotype, suggesting that learning fidelity is high for these traits. In this case, novel variants that arise during the learning process may be preserved and accumulate over generations, contributing to population divergence in song and the formation of vocal dialects (Baker and Cunningham, 1985; Catchpole and Slater, 2008). Differences in songs among populations, if coupled with female preference for local song types, can result in pre-mating reproductive isolation and, in some cases, speciation (Kirkpatrick, 2000; Verzijden et al., 2012). The potential role of song in speciation may be of particular interest in Estrildids, as the family has recently undergone a period of rapid speciation (Olsson and Alström, 2020). In a closely related species, the Bengalese finch, female preference for song complexity, coupled with release from selection pressures, has been highlighted as a driver for increasingly complex songs in captivity compared to ancestral wild populations (Okanoya, 2012; Suzuki et al., 2014). While the Java sparrow and Bengalese finch share similar life histories and domestication history, little is known about female preference for song features in Java sparrows. It is, therefore, not possible to predict how female mate choice could impact song evolution in the Java sparrow.

The Java sparrow has been widely bred in captivity as part of ex situ conservation efforts (BirdLife International, 2018) and as a popular species in aviculture (Restall, 1996). Genetic and behavioral change is frequently reported in captive breeding programs and can accumulate over relatively short time periods, spanning few generations (Håkansson and Jensen, 2005; Frankham, 2008; Suzuki et al., 2013), and these differences may extend to vocal behavior (Tanimoto et al., 2017). The potential for song evolution and cultural divergence in the Java sparrow is, therefore, likely to be of interest to conservation practitioners, as vocal changes may influence the success of conservation programs and in particular reintroduction programs (Lewis et al., 2020; Crates et al., 2021).

Overall, we find that cultural processes play a large role in the song inheritance of Java sparrow, influencing song structure and complexity, as well as acoustic characteristics of notes, in line with findings in other species. Social inheritance of these features has the potential to influence the formation and maintenance of population specific differences, with implications for evolution and conservation in Java sparrows. However, we found no clear evidence for the inheritance of a performance-related factor, vocal consistency, which was instead related to age at recording and clutch-specific differences. While we did not find a relationship with genetic relatedness, our findings support the hypothesis that vocal consistency is an honest signal of quality, revealing information about the age and developmental environment of the signaler.

The original contributions presented in the study are publicly available. This data can be found here: FigShare; https://doi.org/10.48420/14555247.

The animal study was reviewed and approved by The University of Manchester Animal Welfare and Ethics Review Board.

RL, MS, SK, and TG were responsible for the conception and design of the study and contributed to the extraction, preparation, and analysis of data. RL and MS collected and collated the data. The original draft was written by RL. RL and TG contributed to writing sections of the manuscript. All authors were involved in reviewing and editing the manuscript and agreed on the final version.