Ministry of Education Key Laboratory for Biodiversity and Ecological Engineering, College of Life Sciences, Beijing Normal University, Beijing 100875, China
Funds:
the National Natural Science Foundation of China32170516
the National Natural Science Foundation of China31872243
Functionally referential signals are a complex form of communication that conveys information about the external environment. Such signals have been found in a range of mammal and bird species and have helped us understand the complexities of animal communication. Corvids are well known for their extraordinary cognitive abilities, but relatively little attention has been paid to their vocal function. Here, we investigated the functionally referential signals of a cooperatively breeding corvid species, Azure-winged Magpie (Cyanopica cyanus). Through field observations, we suggest that Azure-winged Magpie uses referential alarm calls to distinguish two types of threats: ‘rasp’ calls for terrestrial threats and ‘chatter’ calls for aerial threats. A playback experiment revealed that Azure-winged Magpies responded to the two call types with qualitatively different behaviors. They sought cover by flying into the bushes in response to the ‘chatter’ calls, and flew to or stayed at higher positions in response to ‘rasp’ calls, displaying a shorter response time to ‘chatter’ calls. Significant differences in acoustic structure were found between the two types of calls. Given the extensive cognitive abilities of corvids and the fact that referential signals were once thought to be unique to primates, these findings are important for expanding our understanding of social communication and language evolution.
Song of passerine birds is one of the best studied examples of animal culture (Riebel et al., 2015; Whiten, 2019). As it is transmitted by learning from conspecifics (e.g., Catchpole and Slater, 2008), it gradually changes by cultural evolution, resulting in differentiation between the populations in time and space. The latter is reflected in birdsong geographic variation, which, although studied for decades, still remains a popular research topic (e.g., Podos and Warren, 2007; Singh and Price, 2015; Nelson, 2017; Ku-Peralta et al., 2020).
Dialects are a specific case of geographic variation of learned vocalization, where songs have regionally specific distinct characteristics, and boundaries between different dialect areas are rather sharp even in cases where there are no obvious physical or environmental barriers (e.g., Bjerke and Bjerke, 1981; Kroodsma, 2004). Dialects are fascinating cultural phenomena per se; however, studying them also may shed light on many general biological processes such as migration, gene flow or emergence of reproductive barriers (e.g., Petrinovich et al., 1981; Tracy et al., 2009; Wilkins et al., 2013).
Although dialects have been observed in many songbirds (e.g., Catchpole and Slater, 2008), three species, all from the superfamily Emberizoidea, serve as traditional ‘dialect models’ studied on large spatial scales as well as over time. The North American White-crowned Sparrow (Zonotrichia leucophrys; Passerellidae) was one of the first species in which the dialects were examined objectively by spectrogram analyses in the early 1960s (Marler and Tamura, 1962). Since then, many studies focused on this model species (e.g., Baker et al., 1981; MacDougall-Shackleton et al., 2001; Derryberry, 2009). Dialects of its South American congener, the Rufous-collared Sparrow (Zonotrichia capensis), have been also studied over several decades (e.g., Nottebohm, 1969; Tubaro et al., 1993; Bistel et al., 2022). The third model species, the Yellowhammer (Emberiza citrinella; Emberizidae), is widespread in the Old World, and its song variation in Europe has captivated the interest of naturalists for more than one and half century (e.g., Oppel, 1869; reviewed in Petrusková et al., 2015).
The Yellowhammer is a common, though in many regions declining, species of European agricultural landscapes (BirdLife, 2017). Across most of the continent, the species is year-round resident, and its easy-to-recognize simple song can be heard—and recorded—every year for several months, from late winter until mid-summer (in Central European conditions from late February to July; Cramp and Perrins, 1994). This makes the species a good candidate not only for large-scale studies on geographic song variation but specifically for involvement of the general public in dialect research within citizen science projects.
The Yellowhammer song (Fig. 1) is composed of two parts—it begins with a repetitive sequence of short syllables, which is highly variable among individuals (e.g., Hansen, 1999; Caro et al., 2009). In a complete song, this initial phrase is followed by terminal elements, whose different arrangement, frequency and modulation characteristics define the species’ dialects (Hansen, 1985). In common Yellowhammer dialects, a complete song is terminated by two elements (usually long and whistle-like; Fig. 1) but three-element dialects also exist, as do rare exceptions of birds consistently singing songs terminated by one element only (see e.g., Hansen, 1985; Diblíková et al., 2019; Frauendorf, 2005; Pipek et al., 2018). However, males frequently drop one or more terminal elements and sing incomplete songs.
Figure
1.
A spectrogram of a typical completed Yellowhammer song, with highlighted initial and final (dialect) phrases. The song, recorded in Rožmberk (48.66793 N, 14.32394 E) on 17 July 2013, represents the BE dialect. Frequency and temporal measures used to characterize the terminal element are indicated (f1: end frequency of the penultimate element B, f2 and f3: start and end frequency of the terminal element E, df: frequency offset between B and E (f2 – f1), bw: bandwidth of the terminal element E (f3 – f2), t2: duration of the terminal element E). Crosses indicate 10 equidistant points along the terminal element where frequency measures were taken.
The existence of regional differences in Yellowhammer song have been mentioned in literature already in the 19th and early 20th centuries (Oppel, 1869; Röse, 1869; Heinroth and Heinroth, 1924; Salomonsen, 1935), but it was likely noticed by keen observers much earlier. Several studies between the 1960s and 1980s characterized in more detail the variation in this species' song in some European countries (e.g., Kaiser, 1965; Møller, 1982; Hansen, 1985). Until the mid-1980s, there was no common nomenclature for Yellowhammer dialects and different authors used their own classification (reviewed in Petrusková et al., 2015). Hansen (1985) introduced a system based on labelling the distinct terminal elements by letters B, C, D, E and X. A dialect was then defined by a letter combination indicating the arrangement of these elements at the end of a complete song. Hansen's dialect nomenclature was adopted in most papers on Yellowhammer dialects published since then (see Petrusková et al., 2015).
Hansen (1985) not only unified the nomenclature of Yellowhammer dialects, but his study was also among the first ones on songbird species, to which the general public, approached by mass media, contributed by recordings, thus pioneering the approach today labelled as citizen science (Vohland et al., 2021). In recent years, the Yellowhammer was a target species in several interlinked citizen science projects. Its dialects were studied with the help of the public in Switzerland (Ambühl et al., 2017), Great Britain and New Zealand (Pipek et al., 2018), and especially Czechia (Diblíková et al., 2019). The project Dialects of Czech Yellowhammers (DCY) paved the way for the other two studies (for which data were collected within the Yellowhammer Dialects project; www.yellowhammers.net). DCY assembled over 4000 recordings of Yellowhammer songs across Czechia since 2011 (most of them contributed by volunteers), and the project resulted in the species’ dialects mapped in unprecedented detail at the whole-country level (Diblíková et al., 2019).
The DCY results as well as the compilation of data from various European countries (Petrusková et al., 2015; www.yellowhammers.net) show that Yellowhammer dialects are distributed in an interesting mosaic-like fashion, with geographic replicates of boundaries of the same dialect types observed in various parts of the continent. In the areas of contact of different dialects, birds alternating between both song types (‘mixed singers’; e.g., Frauendorf, 2005; Diblíková et al., 2019) or three-element song variants (e.g., Hansen, 1985; Frauendorf, 2005; Pipek et al., 2018) have been documented.
However, birds assigned to certain dialects did not form geographically well-delineated clusters and rather seemed scattered. Occasional presence of unexpected song type within another dialect region is not unusual, and may reflect for example dispersal of birds originating from elsewhere. However, such inconsistent patterns of spatial distribution also may be due to unsuitable criteria for dialect delineation. The first studies following Hansen's pivotal study from Denmark (Glaubrecht, 1989; Frauendorf, 1994) already pointed out that some dialects defined by Hansen (1985) may not be generally applicable, as the frequency and length of individual elements vary considerably (Frauendorf, 1994). Inconsistencies in some dialect assignments based on arbitrarily defined quantitative boundaries (frequency or temporal measures) may lead to non-adjacent or scattered distributions of birds assigned to different dialect categories.
The DCY study (Diblíková et al., 2019) specifically indicated two groups of previously recognized dialects that at least in our study area (Czechia) tend to show highly fragmented distribution. Long and short variants of the X element (Xl and Xs) were originally considered to define separate dialects (XlB and XsB) by Hansen (1985). Wonke and Wallschläger (2009) supported the distinctness of Xl and Xs elements in a quantitative analysis of recordings from Eastern Germany, but pointed out that the Xs variant lacked the exclusive geographic separation expected from a distinct dialect. The same was reported for Yellowhammer populations in Czechia (Diblíková et al., 2019), Switzerland (Ambühl et al., 2017) or western Belgium and eastern France (Caro et al., 2009). In all those regions, occasional birds singing short variants of the X element, or alternating between the long and short one, were observed scattered among those singing the dominant XlB variant. Due to these observations, a study on British and New Zealand Yellowhammers (Pipek et al., 2018) refrained from Xl and Xs differentiation when assigning song to dialect types.
Another problematic group of song endings that were originally recognized by Hansen (1985) are BE, BlBh and BD (Fig. 2), i.e., those with lower-pitched penultimate element B, followed by a higher-pitched terminal element of either markedly increasing frequency (E), or with a terminal element of roughly constant or weakly decreasing frequency (Bh or D, depending on its pitch). Hansen (1985) considered song endings with two B elements (BhBl, BlBh, BBe), which may differ in their relative frequency (h = higher, l = lower, e = equal), as ‘subdialects’ of one dialect BB. However, studies from various regions (e.g., Conrads, 1984; Frauendorf, 2005; Diblíková et al., 2019) clearly indicated that BlBh and BhBl variants (i.e., those, in which the terminal element is of either higher or lower frequency than the preceding one) are common in some regions and tend to have clearly distinct distributions, so they deserve recognition as separate dialects.
Figure
2.
Examples of complete songs originally assigned to BlBh, BD or BE dialects, indicating the variation of terminal element shape, slope and offset. The original classification is indicated on the left, the reclassification based on criteria redefined in this study on the right. More examples of the terminal element variation are provided in Supplementary material (Appendix A1). Origin of the recordings (all obtained via the Dialects of Czech Yellowhammers project in 2013): A: Kolín (50.09148 N, 15.24853 E), recorded on 08 July; B: Jaroměř (50.34435 N, 15.94887 E), 14 May; C: Černá Voda (50.29131 N, 17.14847 E), 29 May; D: Lomnice (50.2192 N, 12.64917 E), 20 July; E: Koštěnice (50.00442 N, 15.88917 E), 17 July; F: Jeseník (50.21714 N, 17.18778 E), 11 June; G: Mikulov (48.81244 N, 16.62953 E), 30 March; H: Liberec (50.66018 N, 15.08290 E), 09 July.
There is nevertheless a substantial variation in the slope, modulation, and frequency offset of a terminal element assigned by various authors to Bh, D and E, and transitional shapes can be recorded (Fig. 2), as already reported by, e.g., Frauendorf (1994) from Saxony (Germany). In his subsequent study from that region, Frauendorf (2005) differentiated multiple additional categories within this apparent Bh–D–E continuum. In the DCY project, we applied two coarser criteria to differentiate presumably distinct BlBh, BD and BE dialects. First, we evaluated the bandwidth of the terminal element (bw in Fig. 1), and the upward-sloping elements with the bandwidth over 250 Hz were assigned to E; this choice was based on the bimodal bandwidth distribution in the Czech dataset (Diblíková et al., 2019). Second, we measured the difference between the final peak frequency of the first terminal element and the initial peak frequency of the second one (i.e., the frequency offset; df in Fig. 1). The threshold of 1500 Hz, as applied by E. Frauendorf (pers. comm.) in his studies from Germany, was then used to differentiate between lower-pitched BlBh and high-pitched BD.
The resulting fragmented spatial pattern, however, was also inconsistent with a hypothesis that such delineated element variants define spatially distinct dialect regions. In particular, birds assigned to dialect BD, which is structurally related to either BlBh or BE (Fig. 2), were usually scattered among birds singing one of those dialects. We thus concluded that a broader definition of dialects, if resulting in less fragmented areas, might be biologically more relevant.
In this follow-up study (which includes additional recordings gathered from the DCY project during the past years), we quantified the variation of the terminal elements in songs assigned to BE, BlBh and BD dialects. Specifically, we evaluated whether these song categories can be considered distinct or whether they form a continuous gradient that should not be further subdivided. We hypothesized that despite their substantial variation, we will be able to differentiate at least two distinct clusters of structurally similar terminal elements, considering the bimodal distribution of their slopes in Czech recordings (Diblíková et al., 2019). Then, we evaluated whether the spatial distribution of reclassified dialects, with criteria based on cluster analysis of quantitatively characterized elements, results in more consistent patterns in our study regions.
2.
Methods
2.1
Song analyses
We used data from the Dialects of Czech Yellowhammers project (www.strnadi.cz), including some recordings obtained after the project results have been published in Diblíková et al. (2019). Altogether, we considered recordings containing finished songs of 2831 Yellowhammer males from 2790 unique sites, which were recorded in Czechia between 2011 and 2022. Out of these, we screened recordings containing songs assigned to BlBh, BD and BE dialects (i.e., with a terminal element that has ascending, constant or weakly descending frequency, and higher pitch than the preceding one). Then, we selected a representative subset of recordings that seemed, based on visual inspection of spectrograms, of sufficient quality to allow detailed frequency measurements of elements in the final phrase. Altogether, we quantitatively processed 430 recordings (each of a different male), 56 of which had been assigned to BD, 67 to BlBh, 306 to BE, and one was a mixed singer of BE and BhBl dialects.
Our aim was to assess comprehensively the variation within the focal dialects, focusing on among-individual differences. When switching between multiple types of initial phrases, some males may also change the frequency characteristics of the terminal elements. Such variation, however, is lower or comparable to differences between males singing the same dialect in the same population (T. Petrusková, pers. obs.). From each of the selected recordings, converted to WAV 44.1 kHz mono format, we thus included in the analysis only one complete song representative of a given male.
The spectrograms, on which the measurements were conducted, were created in Avisoft SASLab Pro v. 5 (Specht, 2007) with the following settings: FFT length 512, Hamming Window, Temporal Overlap 87.5%, resulting in a 112 Hz bandwidth with 86 Hz frequency and 1.45 ms temporal resolution. For even more precise assessment of the element pitch, a higher FFT length would have been desirable. However, the settings above were sufficient to capture the overall patterns within the dataset, and in most cases provided values allowing consistent and reproducible assignment of the measured elements to clusters (see Results).
The following characteristics of the final phrase were quantified in each analysed song: frequency difference between the end of the first and the beginning of the second terminal element (frequency offset), and duration, initial frequency and modulation of the second element. The modulation was characterized by temporal and frequency characteristics of ten points equally distributed throughout the element duration (Fig. 1, Appendix A1). We aimed to measure a peak frequency within the element's trace at the specified time point. For some recordings of medium quality, however, the frequency measurements may have been manually corrected to adequately follow the trace.
To obtain values independent of the absolute frequency and duration of the second element, its initial frequency (f2 in Fig. 1) was subtracted from the respective frequency measurements, and its duration (t2) was standardized to 1. The resulting frequency values were then fitted against time by a second-degree orthogonal polynomial. The three polynomial coefficients (c1, c2, c3) that captured the bandwidth and modulation of the second element, together with its absolute duration (t2; Fig. 1) and offset from the preceding element (df = f2 – f1; Fig. 1), were used as input variables for subsequent multivariate analyses. The terminal element bandwidth and offset were also used to check the assignment of the analysed songs to the dialects according to criteria in Diblíková et al. (2019).
Using the quantitative measurements from the 430 measured songs, we first run a principal component analysis (PCA) using rda function from ‘vegan’ package (Oksanen et al., 2022) to visualize patterns of variation in the final phrase. Second, to better identify potential distinct clusters in the dataset, we have applied uniform manifold approximation and projection (UMAP; McInnes and Healy, 2018) using ‘umap’ package (Konopka, 2022). Based on the clusters identified by UMAP, we defined simple criteria to assign songs to the redefined dialects, which could be easily applied to field recordings. Then, we reclassified available recordings with the focal dialects from Czechia (those measured in this study, as well as all unmeasured ones originally assigned to BD) according to these new criteria.
We compared the distribution of the original (according to the criteria in Diblíková et al., 2019) as well as the newly reclassified dialect assignment in maps (projected in the JSTK Krowak East–West coordinate system). In the visualization of reclassified dialects, we also combined the original XlB and XsB song variants to a pooled XB dialect. We plotted Voronoi polygons around data points to better characterize the area of the different dialect regions. The maps were post-processed in Inkscape 1.2.2.
2.3
Data accessibility
Data underlying the present study (the origin of quantitatively analysed Yellowhammer songs, frequency and time measurements of terminal elements, input and output parameters of multivariate analyses, and original and reclassified dialect assignment) are provided in Appendix Table S1. The analysed songs in the WAV format, information on all recordings included in the maps, and the R code used to calculate statistics and generate outputs, are available in the Zenodo repository (https://doi.org/10.5281/zenodo.8092674).
3.
Results
Neither the principal component analysis (Fig. 3A) nor the uniform manifold approximation and projection (Fig. 3B) based on terminal element characteristics of 430 recordings assigned to BlBh, BD and BE dialects supported the separation of the analysed final phrases into more than two categories. While the majority of terminal elements of BE and BlBh songs were clearly separated, those initially assigned to BD were distributed to both clusters recognizable in UMAP and PCA plots. The clusters were primarily differentiated by the three coefficients of the polynomial, which characterize the modulation of the terminal element, and its frequency offset (Fig. 3A). The same approach, applied to a subset comprising only songs originally assigned to BD and BlBh, did not reveal any further clustering tendency (results not shown).
Figure
3.
The principal component analysis (PCA; A) and the uniform manifold approximation and projection (UMAP; B) based on temporal and frequency characteristics of the terminal elements of studied Yellowhammer dialects, and the projection of this dataset against the terminal element frequency offset and bandwidth (C). The original dialect assignment (BlBh, BD and BE) is indicated by symbol colours, the UMAP clusters (mostly corresponding to new reclassified dialects) by symbol shapes. The three regions in C delineated by black lines indicate the original criteria for dialect assignment used in Diblíková et al. (2019), the red broken line delineates the ‘new dialects’ according to reclassification criteria (see Results). Light grey background rectangles highlight regions in the plot where reclassified boundaries deviate from a simple merging of original BlBh and BD dialects, and darker grey rectangle is a region of rarely observed transitional (unassigned) song types. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)
One UMAP cluster comprised all songs assigned to BlBh and most of those assigned to BD (especially with the slightly downward sloping elements), while the other cluster comprised all but two songs assigned to BE, and five high-pitched BD (Fig. 3C). The criteria that would assign almost all analysed songs to one of the clusters according to these two frequency parameters are summarized in Table 1, and indicated as the red dashed border line in Fig. 3C. These characterize our redefined dialects (further referred to ‘new BlBh’ and ‘new BE’).
Table
1.
Criteria used to reassign dialects based on frequency bandwidth and offset of the last element (as visualized in Fig. 3C).
Interestingly, a visual assessment of the spectrogram of the one ‘mismatched’ song in Fig. 3C, for which the assignment to UMAP cluster 2 differed from the dialect classification according to the simplified frequency criteria (‘new BE’), indicated that the overall shape of its terminal element (ID 298003 in Appendix A1) matches the reclassification better, as the polynomial coefficients were apparently affected by unusually deviating frequency measurements.
After reclassification of original BlBh and BD recordings (those included in the analyses above as well as the remaining ones that were not suitable for measurements) to ‘new BE’ a ‘new BlBh’ categories according to these criteria, we observed substantially more contiguous dialect regions (Fig. 4B) than when the criteria from Diblíková et al. (2019) were used (Fig. 4A). Most songs originally assigned to BD merged with adjacent BE or BlBh regions. A noteworthy is the observation that the five cases where high-pitched BD elements were reassigned to ‘new BE’ (despite a bandwidth not exceeding 250 Hz; see example in Fig. 2D) were within or adjacent to the originally defined BE regions. This supports the idea that those songs indeed form part of the BE continuum.
Figure
4.
Distribution of Yellowhammer song variants in Czechia mapped within the DCY project, (A) as assigned to dialects according to original criteria from Diblíková et al. (2019), and (B) as reclassified in this study. In the latter map, XlB and XsB song variants were merged to XB, and recordings originally assigned to BD mostly reclassified to newly defined BlBh and BE dialects (except of a few ‘transitional songs’). Eight isolated Yellowhammer individuals that sung aberrant song variants (not considered rare dialects) were excluded. The dialect distribution is indicated by Voronoi polygons, black dots mark origin of recordings that were quantitatively analysed in this study, white dots mark recordings reclassified by the new criteria after visual inspection. Positions of all individual recordings and their dialect assignment, including the aberrant songs, are shown in Appendix Fig. S1.
Elements with a relatively high frequency offset (>2000 Hz) and almost constant frequency (peak frequency bandwidth between −100 and +100 Hz) were considered ‘transitional’ according to our criteria. Such completed songs (only rarely observed in our dataset; in total five measured songs with the bandwidth between 0 and 90 Hz, and two more assessed visually) would remain unassigned to a dialect. Even in those cases, variation in song delivery of a particular singing male might indicate to which dialect to assign it, as other songs sung by the same individual may show a tendency of decreasing or increasing pitch (A. Petrusek, pers. obs.). These seven unassigned males singing the ‘transitional songs’ were observed scattered in various places of the country, adjacent to BE of BlBh areas (Fig. 4B). This is comparable, both in number and distribution, to eight males that sung various unusually shaped aberrant elements deviating from common dialects, which were even more unpredictably scattered (Appendix Fig. S1).
4.
Discussion
Our quantitative analysis of frequency and temporal characteristics of terminal elements of songs assigned to three arbitrarily delineated Yellowhammer dialects (BlBh, BD and BE) confirmed that such categorization is not substantiated, at least not in our study region and as defined in Diblíková et al. (2019). The analyses and visualizations nevertheless confirmed our assumption that the dataset can be divided into biologically meaningful groups. These were not more than two, largely—but not fully— corresponding to merged BlBh with BD, and BE dialects, with only a narrow transition zone of elements that could not be unambiguously categorized (those of very high pitch, and almost constant frequency). A further subdivision of these groups was not supported, as there seemed to be a continuous variation of the terminal element characteristics within each of the two clusters identified in our analyses (‘new BlBh’ and ‘new BE’ dialects).
As expected, when the spatial distribution of original dialect assignment was compared with that based on reclassification of songs to two above-mentioned broader groups, the latter approach resulted in a more orderly pattern. Regions where songs originally assigned to BD were scattered among BlBh or BE got largely unified (Fig. 4), and occasional BD birds isolated within otherwise contiguous BE or BlBh areas were mostly reassigned, based on their frequency characteristics, to the surrounding dialect. Sharp borders between newly defined dialect areas and adjacent dialects were not always obtained, but that could not be expected. What may seem like an abrupt change from one dialect to another at a country level is, when studied locally in detail, a boundary zone where birds singing different song types are to some extent spatially interspersed (see, e.g., Frauendorf, 2005; Diblíková et al., 2019).
Not surprisingly, a few recorded birds singing ‘transitional’ songs remained unassigned to dialect. However, these were comparable in number with isolated birds singing other unusual terminal elements, not matching any commonly recognized dialects. Among over 2800 individuals from the DCY dataset, for which completed songs were recorded, such unassigned birds represent less than 0.6%. Occasionally, such aberrant songs may be learned by other birds and give rise to local rare dialects (as were those reported in the DCY project; Diblíková et al., 2019) but in general they are unlikely to persist and spread, considering the quite conservative nature of Yellowhammer final phrases across the continent.
Frauendorf (1994; 2005) approached the problem with the presence of transitional types of the terminal phrases along Bh–D–E gradient in Saxony by defining additional song type categories with finer arbitrarily defined frequency thresholds. Our data, however, suggest that neither the spectral and temporal characteristics of the song elements, nor the patterns of spatial distribution of birds singing such songs, warrant recognition of transitional terminal phrases as distinct dialects. We therefore suggest that instead of detailed splitting to multiple dialect categories, which likely rather reflect variation within the same song type, lumping these to broader categories is more relevant when studying dialect distribution and boundaries.
Unlike other model species for dialect research (e.g., Nottebohm, 1969; MacDougall-Shackleton and MacDougall-Shackleton, 2001; Nelson et al., 2004), the Yellowhammer is very specific by the mosaic distribution of the same dialect variants across its European range (Petrusková et al., 2015; www.yellowhammers.net). However, it is conceivable that in different regions within the species’ range, song variants assignable to the same dialects differ in quantitative characteristics or extent of their variation, due to local cultural evolution, founder effects, or constrained distribution. Transposing the results from one area to another may then lead to substantial biases.
Specifically, Hansen's recognition of BD as a song variant distinct from BlBh might have reflected the restricted distribution of the high-pitched D element in his dataset: in Denmark, it was recorded only on a single small island, Læsø (Hansen, 1985). If present in a larger contiguous area, it is not unlikely that broader frequency variation of that element would be observed. However, Hansen (1985) considered the dialect BD distinct from BlBh, and accordingly assigned to BD a few matching recordings available to him from other countries (Hungary and former Yugoslavia). His categorization was so influential that authors of further studies (e.g., Frauendorf, 1994; 2005; Ambühl et al., 2017; Pipek et al., 2018; Diblíková et al., 2019) expected to encounter songs assignable to a specific BD category, and set the criteria accordingly.
However, transitional forms between BD and BlBh, and/or BD and BE, were observed in detailed studies in Central Europe (Frauendorf, 2005; Diblíková et al., 2019), and our results point out that reconsideration of the dialect classification was warranted. Although it is apparently not as erratic as the delineation of the former BD dialect, a thorough analysis of the boundary between BlBh and BC dialects might be also worth a study in the region(s) where these are in contact and seem to merge into each other, such as Saxony in Germany (Frauendorf, 2005).
It would be also highly desirable to apply a comparable methodological approach as in our study in some other region(s) where such song variants have been reported from, to test whether patterns of variation are comparable to the Czech dataset or show different clustering tendency. The latter result would suggest that at least sometimes, song variants assigned to the same dialect types in fact follow different paths of song cultural evolution. For such a purpose, however, a sufficient number of recordings would have to be available, capturing variation within the dialect areas as well as across their borders.
Several previous studies used various quantitative approaches for Yellowhammer dialect delineation (e.g., Hansen, 1985; Frauendorf, 2003; 2005; Wonke and Wallschläger, 2009). Although their methods differed, as did the number of measured parameters, all evaluated individual temporal and frequency measures separately. The approach applied in our study, which takes into account the element modulation, seems convenient for capturing variation of relatively simple elements, such as those that differ between the Yellowhammer dialects. For quantitative analyses of variation of more complex songs or their structures, other methodological approaches are nevertheless necessary—for example, the comparison by the dynamic time warp analysis (e.g., Lachlan et al., 2013; Soha et al., 2016; Oñate-Casado et al., 2023) or other suitable (dis)similarity measures. In our specific case, however, not only characterizing the studied element by several equidistant points was methodologically simple, but it also allowed analyses of recordings from our dataset that could be hardly processed otherwise due to background noise, overlapping sounds, or relatively weak signal.
The varying quality of our source recordings is a natural consequence of their origin from a citizen science project. They were obtained by a wide range of recording equipment, from high-quality sound recorders to cell phones or cameras, and often in suboptimal field conditions. Despite this, a large part of the dataset was of a sufficient quality not only to assess the local dialect bus also for this study of quantitative analysis of the elements. Although the quality of birdsong recordings is likely to vary between recordings by experts and citizen scientists (Jäckel et al., 2021), the latter group may provide large-scale data that would be otherwise difficult or impossible to obtain.
At present, most studies on bird geographic variation that claim to use material contributed by citizen scientists (e.g., Searfoss et al., 2020; Otter et al., 2020) benefit from recordings deposited by the public to various databases (such as the Cornell Lab of Ornithology Macaulay Library of Natural Sounds, and xeno-canto). Dedicated projects mobilising the public to obtain targeted birdsong records to study a specific research question are much scarcer (but see, e.g., Jäckel et al., 2022) but may be expected to increase in number in the near future.
Our citizen science projects dedicated to the Yellowhammer dialects (Ambühl et al., 2017; Pipek et al., 2018; Diblíková et al., 2019) allowed assembling large-scale data across several countries, and provided material not only to study the geographical distribution of dialects but also allowed this quantitative analysis focusing on a meaningful dialect delineation. As songbird dialects provide opportunities for focused studies of processes underlying the emergence of spatial patterns in vocalization (e.g., Harbison et al., 1999; Bell et al., 2003; García et al., 2015), not only their sufficiently detailed mapping but also their appropriate characterization is essential.
Recordings of citizen scientists in public databases contributed substantially to a recent continental-wide study of dialect shift in White-throated Sparrows (Zonotrichia albicolis) (Otter et al., 2020). Dialects of two songbirds, including that species and the Yellowhammer, were also used as case studies demonstrating the usefulness of data collected by the public via a free bird sound identification application for smartphones, BirdNET (Wood et al., 2022). In case of the Yellowhammer, however, Wood et al. (2022) only distinguished between two broad groups of song types that begin with either X or B elements, previously considered to be geographically separated (e.g., Kaiser, 1965; Cramp and Perrins, 1994) but in fact lacking such clear distributional pattern (Petrusková et al., 2015). More detailed information on Yellowhammer dialect distribution and variation (which could substantially extend the whole-Europe coverage) thus remain to be mined from that dataset.
In the near future, big data assembled by applications such as BirdNET (Wood et al., 2022) will certainly become troves of useful information on birdsong geographic variation, especially when machine learning techniques are applied to process them. To maximise their usefulness for further dialect research, however, careful evaluation of dialect boundaries, which are then recognized by the classification algorithms, is essential.
Authors’ contributions
LD, PP, AP and TP conceived the research. LD selected recordings for the analyses and processed most of them. PP and SV designed analytical approaches, with contribution of LD and AP. PP wrote R scripts and performed statistical analyses. LD, PP, AP and TP interpreted the results. PP and AP created most figures. The present version of the manuscript was mostly drafted by AP, with substantial contribution from all authors. All authors read and approved the final manuscript.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We thank all contributors of recordings to the projects Dialects of Czech Yellowhammers (www.strnadi.cz) and Yellowhammers Dialects (www.yellowhammers.net), all colleagues involved in project set up and outreach, and the Czech Society for Ornithology for its long-term support of citizen science. This work was also funded by the Charles University Grant Agency (project number 312213). We are grateful to Jana Černochová and Hana Kyliánková for their involvement in the song measurements.
Berthet, M., Mesbahi, G., Cäsar, C., Zuberbühler, K., 2022. Impact of predator model presentation paradigms on titi monkey alarm sequences. Behav. Ecol. Sociobiol. 76, 143.
Blumstein, D.T., Armitage, K.B., 1997. Alarm calling in yellow-bellied marmots: I. The meaning of situationally variable alarm calls. Anim. Behav. 53, 143–171.
Bolt, L.M., 2020. Affiliative contact calls during group travel: chirp and wail vocalization use in the male ring-tailed lemur (Lemur catta). Folia Primatol. 91, 575–594.
Bolt, L.M., Sauther, M.L., Cuozzo, F.P., Jacky, I.A.Y., 2015. Antipredator vocalization usage in the male ring-tailed lemur (Lemur catta). Folia Primatol. 86, 124–133.
Bond, A.B., Kamil, A.C., Balda, R.P., 2003. Social complexity and transitive inference in corvids. Anim. Behav. 65, 479–487.
Bouchard, A., Zuberbühler, K., 2022. An intentional cohesion call in male chimpanzees of Budongo Forest. Anim. Cogn. 25, 853–866.
Bradbury, J.W., Vehrencamp, S.L., 2011. Principles of Animal Communication, second ed. Sinauer, Sunderland, MA.
Brown, C.H., 1982. Ventroloquial and locatable vocalizations in birds. Z. Tierpsychol. 59, 338–350.
Brown, E.D., 1985. The role of song and vocal imitation among common crows (Corvus brachyrhynchos). Z. Tierpsychol. 68, 115–136.
Butler, N.E., Watson, S.J., Peters, R.A., 2022. Life history characteristics and alarm calling in Australian arid-zone birds. Emu 122, 155–166.
Clayton, N.S., Emery, N.J., 2015. Avian models for human cognitive neuroscience: a proposal. Neuron 86, 1330–1342.
Cunningham, S., Magrath, R.D., 2017. Functionally referential alarm calls in noisy miners communicate about predator behaviour. Anim. Behav. 129, 171–179.
Deshpande, A., van de Waal, E., Zuberbühler, K., 2023. Context-dependent alarm responses in wild vervet monkeys. Anim. Cogn. 26, 1199–1208.
Dutour, M., Kasper, J., Ridley, A.R., 2021. Transfer of information between a highly social species and heterospecific community members. Behav. Ecol. Sociobiol. 75, 137.
Emery, N.J., Clayton, N.S., 2004. The mentality of crows: convergent evolution of intelligence in corvids and apes. Science 306, 1903–1907.
Evans, C.S, Evans, L., Marler, P., 1993. On the meaning of alarm calls: functional reference in an avian vocal system. Anim. Behav. 46, 23–28.
Fedurek, P., Slocombe, K., 2011. Primate vocal communication: a useful tool for understanding human speech and language evolution? Hum. Biol. 83, 153–173.
Fernández, G.J., Dutour, M., Carro, M.E., 2023. Information transfer during mobbing: call rate is more important than the number of callers in a southern temperate passerine. Behav. Ecol. Sociobiol. 77, 78.
Freeberg, T.M., 2006. Social complexity can drive vocal complexity: group size influences vocal information in Carolina chickadees. Psychol. Sci. 17, 557–561.
Freeberg, T.M., Dunbar, R.I.M., Ord, T.J., 2012. Social complexity as a proximate and ultimate factor in communicative complexity. Phil. Trans. R. Soc. B 367, 1785–1801.
Gao, L.F., Zhang, H.Y., Zhang, W., Zhang, X.D., Zhu, Z.Q., Du, B., 2021. Fitness consequences of divorce in the azure-winged magpie depends on the breeding experience of a new mate. Curr. Zool. 67, 17–25.
Greene, E., Meagher, T., 1998. Red squirrels, Tamiasciurus hudsonicus, produce predator-class specific alarm calls. Anim. Behav. 55, 511–518.
Grieves, L.A., Logue, D.M., Quinn, J.S., 2014. Joint-nesting smooth-billed anis, Crotophaga ani, use a functionally referential alarm call system. Anim. Behav. 89, 215–221.
Hailman, J.P., Ficken, M.S., Ficken, R.W., 1985. The ‘chick-a-dee’ calls of Parus atricapillus: a recombinant system of animal communication compared with written English. Semiotica 56, 191224.
Igic, B., Magrath, R.D., 2014. A songbird mimics different heterospecific alarm calls in response to different types of threat. Behav. Ecol. 25, 538–548.
Jurisevic, M.A., Sanderson, K.J., 1998. Acoustic discrimination of passerine anti-predator signals by Australian raptors. Aust. J. Zool. 46, 369–379.
Krause, J., Ruxton, G.D., 2002. Living in Groups. Oxford University Press, Oxford.
LaPergola, J.B., Savagian, A.G., Smith, M.G., Bennett, B.L., Strong, M.J., Riehl, C., 2023. Referential signaling in a communally breeding bird. P. Natl. Acad. Sci. U. S. A. 120, e2222008120.
Leighton, G.M., 2017. Cooperative breeding influences the number and type of vocalizations in avian lineages. Proc. R. Soc. A B 284, 20171508.
Macedonia, J.M., Evans, C.S., 1993. Essay on contemporary issues in ethology: variation among mammalian alarm call systems and the problem of meaning in animal signals. Ethology 93, 177–197.
R Core Team, 2023. R: a Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. .
Ren, Q., Luo, S., Du, X., Chen, G., Song, S., Du, B., 2016. Helper effects in the azure-winged magpie Cyanopica cyana in relation to highly-clumped nesting pattern and high frequency of conspecific nest-raiding. J. Avian Biol. 47, 449–456.
Rosa, G.L.M., Ellis, J., Bonaccorso, E., dos Anjos, L., 2016. Friend or foe? Social system influences the allocation of signals across functional categories in the repertoires of New World jays. Behaviour 153, 467–524.
Slocombe, K.E., Zuberbühler, K., 2005. Functionally referential communication in a chimpanzee. Curr. Biol. 15, 1779–1784.
Smith, C.L., 2017. Referential signalling in birds: The past, present and future. Anim. Behav. 124, 315–323.
Struhsaker, T.T., 1967. Auditory communication among vervet monkeys. In: Altmann, S. A. (Ed.), Social Communication Among Primates. University of Chicago Press, Chicago, pp. 281–324. Cercopithecus aethiops.
Suzuki, T.N., 2011. Parental alarm calls warn nestlings about different predatory threats. Curr. Biol. 21, 15–16.
Tanimoto, A.M., Hart, P.J., Pack, A.A., Switzer, R., 2017. Vocal repertoire and signal characteristics of ’Alalā, the Hawaiian crow, Corvus hawaiiensis. Wilson J. Ornithol. 129, 25–35.
von Frisch, K., 1974. Decoding the language of the bee. Science 185, 663–668.
Wang, X., Yu, F., Si, H., Li, Z., 2023. Vocal repertoire of azure-winged magpies, Cyanopica cyanus: a context-associated communication system. Avian Res. 14, 100089.
Wood, S.R., Sanderson, K.J., Evans, C.S., 2000. Perception of terrestrial and aerial alarm calls by honeyeaters and falcons. Aust. J. Zool. 48, 127–134.
Zuberbühler, K., 2003. Referential signaling in non-human primates: cognitive precursors and limitations for the evolution of language. Adv. Study Behav. 33, 265–307.
DownLoad:
Email Alerts
RSS Feeds