Introduction

Males of passerine bird species usually combine songs into song sequences that may carry information about their behavior and the context in which they are singing. In particular, males of many passerine species vary certain aspects of their singing behavior when engaged in territorial interactions (Todt and Naguib 2000). Song sequences can be characterized by how the vocal units (e.g., song types) are combined, and by timing aspects such as song rate. Several song parameters and singing behaviors, such as song rate, song type switching rate, song matching, song overlapping, and the usage of specific song or call types, have been proposed to be aggressive signals (Todt and Naguib 2000; Catchpole and Slater 2008; Searcy and Beecher 2009). Song complexity also plays a role in territory acquisition and defense. For example, New Zealand Tui Prosthemadera novaeseelandiae males reacted more strongly to playback of complex songs that had the highest number of syllables and highest syllable diversity (Hill et al. 2017). Similarly, stronger responses to songs with more phrases were found in Chaffinches Fringilla coelebs (Leitão et al. 2006) and Wood Warblers Phylloscopus sibilatrix (Goretskaia 2013). In addition to this, song type diversity, i.e., the number of song types in a song sequence, depended on the context in some species (Kroodsma and Verner 1978; Searcy and Yasukawa 1990; Searcy and Beecher 2009; Vehrencamp et al. 2007). The conclusion from the majority of studies was that males increase their song diversity and complexity when confronted with a female or a competitor.

In many songbirds, the transitions between different song/syllable types in the course of singing are non-random (Dobson and Lemon 1978; Lemon et al. 1993; Todt and Hultsch 1998; Gil and Slater 2000; Bhattacharya et al. 2007; Ivanitskii et al. 2012, 2017a, b; Sasahara et al. 2012; Okanoya 2013). Although some authors presumed that non-random vocal streams are of significance to signaling (Ruxton and Schaefer 2011), in most cases we do not know what these signals comprise or if conspecifics could use and encode information in them. It was proposed, however, that the ordering of songs may convey biologically relevant information about individual fitness (Leitner et al. 2006). Indeed, in Common Nightingales Luscinia megarhynchos the transitional patterns of song types were found to play a role in singing interaction (Weiss et al. 2014). However, data are still scarce.

In this study, we used playback experiments to investigate whether singing plays a role in male–male interactions in Claudia’s Leaf-warbler (Phylloscopus claudiae). Along with traditional metrics (i.e., song type diversity, song rate, between-song pause length, song length and song type switching rate), we applied an information theory approach in order to analyze the role of song type transitional patterns in male–male interactions.

Methods

Claudia’s Leaf-warbler is a small passerine bird endemic to China. It breeds abundantly in the mountains of Central and North China mostly above 1000 m. This species is comprises the Phylloscopus reguloides superspecies, which has been given monotypic species status based on genetic and bioacoustic data (Päckert et al. 2009; Martens et al. 2011).

Data were collected in April–June 2016 in Hupingshan National Nature Reserve, Hunan Province, South Central China (30°02′N, 110°31′E). Claudia’s Leaf-warbler is one of the commonest leaf-warbler species in the reserve, breeding at altitudes of 1500–2000 m above sea level.

Preparation of playback stimuli

The songs used to prepare playback stimuli were recorded in Hupingshan National Nature Reserve in 2014. We recorded males during spontaneous singing using a Marantz PMD-660 digital recorder with a Sennheiser ME66-K6 microphone. We selected three high-quality recordings from different males and used these for the preparation of playback stimuli. These recordings had a total duration of 45.4 min, and contained 296 songs. For each recording, we firstly determined the number of different song types. In total, we identified 11 song types for these three males. The recordings were then analyzed as described below (see “Analyzing song bouts”). To construct playback stimuli, we used the median values of the parameters examined. The playback stimuli thus roughly correspond to the averages determined for spontaneous song: repertoire size = four song types, between-song pause length = 7 s, song type switching rate = 0.66, first-order relative entropy (RE1) = 0.90. We selected four out of 11 song types that appeared to us the most distinct and used them to prepare playback stimuli. To summarize, in each experiment, we used the same playback stimuli that corresponded to the ‘average’ Claudia’s Leaf-warbler’s song. Playback stimuli comprised a repertoire of 50 songs and lasted for 7.4 min.

Playback experiments

Experiments were conducted in the morning, between 0700 and 1200 hours. Firstly, we selected a male that sang spontaneously when no other males were nearby. A loudspeaker (XqBeats PF305) was then placed on a branch at a height of 1.5–2 m from the ground, within 10–20 m of the focal male. Two observers (A. O. and Y. K.) were ca. 10–20 m away from the loudspeaker during each experimental trial. We did not standardize the amplitude of the playback stimuli. However, we used the same amplitude in all experiments as we set the maximum volume level of the loudspeaker. In our opinion, the resulting amplitude appeared quite similar to the natural amplitude of Claudia’s Leaf-warbler songs.

The experimental trial consisted of two subsequent stages without interruption: recording of spontaneous singing before the start of the playback, and recording of singing during the playback. During the experiment we always noted the songs of a male by simply recording “it is a male.” In total, the experiments involved 14 different males. In each experiment, we recorded 6.1 ± 2.6 min (range 2.5–11.1 min, n = 14) of spontaneous singing and 7.4 min of singing during the playback.

In all trials, males responded to the playback by approaching the loudspeaker on the tree. Males typically came close to the loudspeaker, sang actively, flew back and forth, and performed wing flicking. All these behaviors occur during natural, aggressive interactions between males (personal observation). We scored the number of flyovers (where males flew for more than 1 m within 10 m of the loudspeaker) as an index of aggressive response.

Songs were recorded using a Marantz PMD-660 digital recorder with a Sennheiser ME66-K6 microphone. For sound visualization and analysis, we used Syrinx PC version 2.6 (John Burt, www.syrinxpc.com) with fast Fourier transform size = 256, and a Hanning window type.

Only two (14.3%) of the tested males had been captured, 1 and 34 days before the experiment, and banded with color rings. All males had established territories prior to the experiment.

The singing activity of Claudia’s Leaf-warbler males had at least two clear peaks in 2014 and 2016. Males sang actively from late April to early/middle May, just after their arrival at the breeding grounds. Later in May, singing was rare. The second peak was observed from late May/early June and continued until at least mid-June, when our fieldwork ended in both years. The June peak may have been caused by the beginning of the second breeding cycle, but we do not have supporting data for this. In our sample, eight males were tested from 27 April to 10 May, and another six males were tested between 4 and 9 June. We did not find any differences between the males recorded in April–May and in June, hence we analyzed all phonograms jointly.

Analyzing song bouts

In the course of this study, we defined song bout as a song sequence recorded during either spontaneous singing or in response to playback. All bouts were processed separately, but not blindly, for each male. Males normally tended to return to their “normal” singing soon after the playback, but we did not analyze a male’s song immediately after the treatment.

The following parameters were calculated for each song bout: (1) median song length, (2) median between-song pause length, (3) song rate, (4) song type diversity, (5) song type switching rate, (6) RE1, (7) sequence linearity index (SLIN); (8) sequence consistency index (SCONS) and (9) stereotype index (S).

We successively measured the durations of songs, and pauses between them, directly from the sonograms. These measurements gave us three variables: median song length (seconds), median pause length (seconds), and song rate (number of songs recorded/record length; number/seconds).

Further, we determined the song type diversity, i.e., the number of different song types in a bout. We defined song type as a unique sequence of elements arranged in a certain order (Catchpole and Slater 2008). The Claudia’s Leaf-warbler songs were simple and discrete, and there was little, if any, sharing of elements between song types. Thus, the delimitations of song types were quite clear and robust in this species (Fig. 1S). We then calculated song type switching rate, i.e., the number of transitions to different song types/total number of transitions.

At the next stage of analysis each song type was labeled with a number. The numbers corresponding to the labeling were inputted into Past 3 software (Hammer et al. 2001). Using the transitional probability matrix generated in Past 3 software, we calculated RE1 (Briefer et al. 2010; Palmero et al. 2014; Opaev 2016). RE1 describes how much the observed song sequence differs from a hypothetical “accidental” sequence (in the latter, transition frequencies between all vocalization types are equal, i.e., all song types considered equally likely). Therefore, RE1 could be used as a measure of freedom in choosing the next song type in a bout. RE1 varies from 0 to 1 and reaches its maximum value when all song types have the same probability of being chosen. RE1 was calculated as follows: RE1 = E1/E0. The Shannon-Weaver formula was used to calculate zero-order entropy (E0) and first-order entropy (E1):

$$ E_{0} = - \sum {\frac{1}{\text{K}}} { \log }_{2} \left( {\frac{1}{\text{K}}} \right), $$

where K is the observed repertoire size (i.e., song type diversity);

$$ E_{1} = - \mathop \sum \nolimits P_{i} \log_{2 } P_{i} , $$

where Pi is the observed probability of each transitional type.

Equations SLIN, SCONS and S were modified from Scharff and Nottebohm (1991). These metrics have been used by some authors in the analysis of song sequencing in birds with small repertoires (Woolley and Rubel 1997; Okanoya 2013; Ivanitskii et el. 2017a, b). SLIN, SCONS and S were calculated using the transitional frequency matrix generated in Past 3 software.

SLIN was calculated as the number of song types in a bout divided by the number of all transition types. In a completely linear sequence, each song type has only one transition type, i.e., it is followed by only one other song type, and thus SLIN equals 1. Otherwise, SLIN is zero.

SCONS = ∑ typical transition types/∑ total transition types. The typical transition type for a song type is the one most frequently observed. Thus, SCONS does not address how songs are ordered but how often a particular sequence is followed. SCONS expresses the frequency with which the main, or typical, sequence appears. If a given sequence always occurs, SCONS amounts to 1. Otherwise SCONS is zero.

$$ {\text{Stereotype}}\;{\text{index}}\,(S) = \left[ {S_{\text{LIN}} + S_{\text{CONS}} } \right]/ 2. $$

Statistical analysis

The analysis was performed in R 3.3.2 (R Core Team 2016). All the acoustic parameters and variables analyzed had a normal distribution, except for song type diversity, and thus parametric tests were performed (with the exception of the Spearman rank order correlation).

We used generalized least square regression (GLS) in the nlme package in R (Pinheiro et al. 2017) to test for differences between spontaneous singing and singing in response to playback for each parameter. GLS offers a simpler and more direct approach to the analysis of correlated data than its main competitors (e.g., ANOVA and generalized linear mixed models) (Pekár and Brabec 2016). In the GLS, experimental treatment (spontaneous vs. playback) was used as a fixed effect and male identity (ID) (14 levels) was used as a random effect. The total number of songs in a bout, bout length and song type switching rate can potentially affect song type diversity and, by influencing the number of different transitions that occurs, RE1, SLIN, SCONS and S. Therefore, while analyzing song type diversity, RE1, SLIN, SCONS and S we firstly fitted a model for the only random effect (male ID). Secondly, we incorporated the total number of songs in a bout, bout length, song type switching rate and (except for song type diversity) song type diversity in the GLS by using exponential models (corExp in the nlme package in R). The Bonferroni correction procedure was used to adjust the significance level because two-group comparisons after multiple comparisons increase the probability of type I errors. We set the α-levels to 0.05/9 = 0.006, but present the exact p-values in Table 1.

Table 1 Values of variables of spontaneous singing before the playback and singing during the playback
Full size table

We visualized transitions between song types in a bout using the markovchain package in R (Spedicato et al. 2017).

Results

Spontaneous singing

Each Claudia’s Leaf-warbler’s song lasted for about 2 s. The median song durations individually varied from 1.7 to 2.4 s (n = 14 males). The median durations of pauses between successive songs in different males ranged from 2.1 to 12.3 s (n = 14) (Table 1). The songs were often preceded by a short introductory element and had a repetitive structure of from two to four successive syllables consisting of from two to four elements each (Fig. 1). Each male possessed a repertoire of stereotypical song types, or song phrases. The number of song types produced during spontaneous singing individually varied from two to 12 (median 3, n = 14 males) (Table 1).

Fig. 1
figure 1

Catalogue of song types of male no. 12 Phylloscopus claudiae. Song types produced during spontaneous singing (1, 2) are shown in the top panel, and those produced under the playback treatment (1–10) in the middle and bottom panels

Full size image

Responses to the playback

All males clearly responded to playback stimuli, as they approached the loudspeaker and flew around it. The number of flyovers during 7.4 min of the playback trial varied from three to 60 (median 33, n = 14).

We found that males responded to playback with significantly more song types (increased song type diversity), shorter between-song pauses, increased song rate, and decreased RE1 relative to the control period of spontaneous singing immediately before the playback. The effect of treatment (spontaneous vs. playback) was still significant after accounting for the total number of songs in a bout, bout length, song type switching rate and song type diversity in the data. Median song length, song type switching rate, SLIN, SCONS and S were not significantly different between spontaneous singing and that elicited by playback (Table 1). The lack of effect of the treatment on SLIN, SCONS and S was surprising given that all these parameters significantly correlated with RE1 (Spearman rank order correlations: R = − 0.48, p = 0.009, n = 28; R = − 0.51, p = 0.006, n = 28 and R = − 0.55, p = 0.002, n = 28, respectively).

In response to the playback, males usually added some new song types but also continued to sing all or the majority of song types produced during spontaneous singing. In fact, eight out of 14 males not only continued to sing “spontaneous” song types when responding to the playbacks (see Fig. 1 for an example) but also added some new ones.

The song type transitions during singing in response to the playback turned out to be more predictable than those during spontaneous singing. The patterns occurred concurrently with a decrease of entropy but not with the expected increase of SLIN, SCONS and S. The decrease of the number of transition types per song type possibly explained the decrease of entropy (e.g., males 8 and 14: Fig. 2).

Fig. 2
figure 2

All transition types between song types in three Claudia’s Leaf-warbler males during spontaneous singing and singing elicited by playback. Transition probabilities are shown; numbers in circles show the different song types of a male. Song type diversity values, first-order relative entropy values (RE1) and stereotype indexes (S) are given

Full size image

Discussion

This study showed that Claudia’s Leaf-warbler males may use several different mechanisms for aggressive signaling in the context of playback-simulated territorial intrusion. Compared to singing prior to playback, song type diversity increased, RE1 decreased (i.e., the non-randomness of singing increased), and song rate increased in response to playback.

Song rate (as well as between-song pause length) has been suggested to be an aggressive signal in several songbird species (Todt and Naguib 2000; Searcy and Beecher 2009), including Phylloscopus (Scordato 2017; Szymkowiak and Kuczyński 2017). It is a flexible trait that can be easily modified by singing males during vocal interactions with competitors to signal short-term variations in the signaler’s condition or motivation (Searcy et al. 2000; Baker et al. 2012; Szymkowiak and Kuczyński 2017). Modification of song rate could be of special importance in species with simple vocalizations. However, songbirds with complex songs and large repertoires also modify their song rate during vocal interactions [e.g., Common Nightingale (Hultsch and Todt 1982)]. It is thus not surprising that Claudia’s Leaf-warbler, a species with a moderate repertoire size, also modified its song rate in response to playback stimuli.

Claudia’s Leaf-warbler males also increased song type diversity to signal aggression. The role of song type diversity in communication is known in many species of birds, and was first demonstrated by Krebs (1976, 1977). For example, male Sedge Wrens Cistothorus platensis tended to use a greater variety of song types when other males were singing nearby (Kroodsma and Verner 1978). Male Chestnut-sided Warblers Setophaga pensylvanica were much more likely to use their rare song types during playback-simulated territorial intrusions than during spontaneous singing (Byers 2017). Some species use their song repertoire in intersexual encounters. For example, a male Red-winged Blackbird Agelaius phoeniceus sang more song types when a female was present (Searcy and Yasukawa 1990). To summarize, males of many species, including Claudia’s Leaf-warbler, increase song type diversity when they detect environmental changes such as the appearance of a female or a competitor.

The singing of Claudia’s Leaf-warblers appeared to be more strictly determined and linear during playback trial, than just before playback, as witnessed by the decrease of entropy. But the result was not strong because SLIN, SCONS and S, though significantly correlated with RE1, did not differ between spontaneous singing and singing in response to playback. We thought that this was because SLIN, SCONS and S were around 0.5 when song type diversity = 2–3 and all transition types occurred. At the same time, the entropy as well as the actual freedom of choice was close to 1 because all transition types had nearly the same probability of occurrence (e.g., spontaneous singing of male 15; Fig. 2).

We hypothesized that determined song could be a strategy to maximize song diversity. For example, when song type diversity in 10 or 20 successive songs was higher, the male sang in a more linear style (i.e., the transitions between different song types were more determined and predictable). Song type diversity, in turn, might play a role in male-male communication.

An important consideration for playback studies is pseudo-replication, i.e., the use of inferential statistics to test for treatment effects for data where replicates are not statistically independent (McGregor et al. 1992; Kroodsma 1989). Practically all playback studies involve some pseudo-replication, possibly because only one location is studied, or the number of playback songs is limited, etc. In this study, we used the only playback stimuli that lead to songs from three males. This limits our ability to generalize from the results. However, in this study, the intention was to learn if conspecific singing affects the vocal behavior of a resident male, no matter what song types from which males were used in the playback. Moreover, some authors think that if the stimuli result from a manipulation (e.g., playback vs. control) then there is no need to replicate stimuli (McGregor 2000; see also Schank and Koehnle 2009).

To conclude, this study revealed that song type diversity, song rate, and non-randomness of a song sequence could play roles in male-male competition. In particular, our study demonstrated the role of non-random vocal streams in songbird communication. However, as the results were not conclusive, further studies are needed.