Turning the tables: a tiny bird uses alarm calls and mimicry to deceive its nest predator

Abstract

Animals often eavesdrop on other species’ alarm calls to gain information about danger, but this can allow for deception by callers. Such deception often uses ‘aerial’ alarm calls, which normally warn of airborne predators and prompt immediate fleeing. The calls are deceptive if they are given when no flying predator is present and the caller benefits from the victim’s response, typically by gaining food dropped when the listener flees. We studied deceptive alarm calling by brown thornbills, Acanthiza pusilla, defending offspring against predatory pied currawongs, Strepera graculina. Thornbills give their own and mimetic aerial alarm calls when defending nestlings against currawongs, who are fooled into scanning for danger or flying away. We tested whether deception works by exploiting the predator’s response to aerial alarm calls, and what role mimicry plays. Currawongs were more likely to flee, and delayed feeding longer, after playback of purely aerial compared with purely mobbing alarm choruses. They responded the same regardless of what type of mimetic alarm followed the thornbill’s aerial alarm. We conclude that vocal deception is effective because it exploits currawong response according to call meaning, while mimicry likely creates an illusion of a multi-species alarm chorus.

1. Introduction

Animals often eavesdrop on calls of other species, with potential costs or benefits for the caller and receiver that set the stage for signal evolution [1–4]. At one extreme, callers can suffer lethal consequences from eavesdroppers, such as when predators hunt prey by listening for courtship calls of adults [5] or begging calls of offspring [3]. In response, victims may evolve calls that are harder to hear or locate [6–9] or vary call usage according to current risk [10,11]. At the other extreme, callers can benefit from eavesdropping through the formation of mixed-species groups that increase foraging efficiency or the ability to detect and deter predators [12]. The result can be calls that are designed to be easy to overhear [8,13], and even used in deliberate communication with other species [14]. In contrast to these situations in which eavesdroppers benefit, we consider an unusual case in which callers ‘turn the tables’, and use vocal mimicry to deceive eavesdropping predators.

Vocal deception commonly uses ‘flee’ alarm calls, because the costs of responding to false calls are usually low but the cost of ignoring true calls could be death [15,16]. Flee alarm calls signal immediate danger from fast-moving predators, and prompt rapid responses such as fleeing to cover; they contrast with ‘mobbing’ alarm calls, given to stationary or slow-moving predators not posing an immediate threat, that prompt approach, monitoring and harassment of the predator [17,18]. Most commonly, deceptive flee alarms are used to steal food, by scaring victims from their prey [19,20]. The calls given in this context meet the definition of functional deception [21] because the receivers hear the flee alarm calls and respond as if a fast-moving predator is present, despite there being no such predator, and the callers benefit from the receivers’ response. Deception therefore works because of the receivers’ response according to the calls’ normal referent, a fast-moving predator.

Vocal mimicry commonly entails acoustic imitation of another species’ sounds [22], and may enhance alarm-call deception in two ways. First, deceptive calls work only when they are relatively uncommon, or they will be ignored, so deceptive alarm callers must balance deceptive and honest calling [19,23,24]. However, vocal mimicry eases this constraint on deception. For example, fork-tailed drongos, Dicrurus adsimilis, mimic the alarm calls of heterospecifics, so if victims cease to respond to one call type, they can switch to another, and thereby maintain the effectiveness of deception [19,25]. Second, vocal mimicry can simulate the presence of multiple species—and therefore also individuals—giving alarm calls, creating a more realistic cue of the presence of a dangerous and generalist predator [26–31].

We studied deceptive alarm calling by parents protecting their vulnerable offspring. Brown thornbills, Acanthiza pusilla, are tiny birds that have their own alarm calls and also mimic the alarm calls of other species [32]. They usually mimic flee and mobbing alarm calls in contexts appropriate to their referent. Specifically, they give choruses of their own and mimetic mobbing alarm calls to predators that are not in flight—including snakes, mammals and birds on foot—but choruses of their own and mimetic ‘aerial’ flee alarm calls when warning of airborne birds, such as Accipiter hawks or omnivores such as pied currawongs, Strepera graculina [33]. However, thornbills switch from mobbing calls when a predator is on the ground near their nest to including aerial alarm calls if a predator attacks their nest and provokes nestlings to give distress calls [33,34]. These choruses are false alarms because there is no predator in flight; the only predator is on foot. Playback experiments show that these aerial alarms fool attacking currawongs into scanning for danger and sometimes flying away, potentially giving the young a chance to flee the nest and hide in nearby cover [34]. The context of the use of these aerial alarm calls, and the response by the predator, together imply deception. Furthermore, currawongs respond more strongly to a mixture of species-specific and mimetic aerial alarm calls than to purely species-specific alarm calls, showing the inclusion of mimicry prompts a greater response, probably because it creates the illusion of multiple species calling [21,33,34]. To our knowledge, this remains the only experimental demonstration of using alarm calls to fool predators.

Here, we test a key assumption in the evidence for deception by thornbills and further assess the likely mechanism for effectiveness. The evidence for deception rests on the thornbill’s use of aerial alarm calls when no airborne predator is present, and the currawong’s fearful response because they respond according to the normal referent of these heterospecific calls [21,33,34]. Here, we test the idea that the currawong’s response depends on the alarm calls being aerial compared to mobbing calls. We predicted that attacking currawongs would be deterred more by the playback of an aerial alarm chorus than a mobbing alarm chorus, in each case including both species-specific and mimetic calls. We also assessed if the type of mimetic alarm call itself affects currawong response, given a species-specific aerial alarm call, or whether any mimicry is sufficient to create the illusion of a greater number of callers and therefore a greater response [34]. If mimetic type is the primary benefit, we predicted that currawongs would respond more strongly to alarm choruses including aerial mimicry compared to mobbing mimicry. A mimetic aerial alarm call would reinforce the thornbill’s own message of airborne danger, whereas a mimetic mobbing call may imply that a flying predator has landed and potentially poses a reduced threat. By contrast, currawong response may be unaffected by mimetic call type, given that either type of alarm call would create the illusion of multiple callers and the presence of a nearby predatory bird.

2. Methods

(a). Study species and site

The brown thornbill is a small (ca 6−7 g) passerine that breeds in pairs, constructing domed nests in dense vegetation on or near the ground, and laying eggs from July to November [35]. The pied currawong is a large, passerine omnivore (ca 300 g) that commonly eats fruits and seeds in winter, but shifts to a protein-rich diet, including nestlings of other bird species, when feeding its own young from October to December [36–38].

The study was conducted from August to November 2017 in and near the Australian National Botanic Gardens (35°16′34″ S, 149°06′33″ E) in Canberra, where both brown thornbills and pied currawongs are common. Thornbills and other small passerines give alarm calls in response to resident predators including currawongs, collared sparrowhawks (Accipiter cirrocephalus), brown goshawks (A. fasciatus), laughing kookaburras (Dacelo novaeguineae), grey butcherbirds (Cracticus torquatus), southern boobook owls (Ninox boobook) and eastern brown snakes (Pseudonaja textilis) ([32,39,40]; pers. obs.). Flying Accipiter hawks, in particular, provoke a chorus of aerial alarm calls from small passerines, and goshawks can also take large prey including currawongs [41]. About 50% of thornbill nests are depredated, primarily by birds, and particularly by currawongs [39].

(b). Playback experiment

We carried out a playback experiment on 20 focal currawongs from September to November 2017 to test the response of currawongs to thornbill vocalizations while feeding at artificial thornbill nests, following general methods in Igic et al. [34]. Each bird received four playback treatments, in balanced random order, that replicated alarm choruses of species-specific and mimetic alarm calls given during nest defence. Each playback file was unique and consisted of a non-mimetic followed by a mimetic alarm, a common natural pattern [33] (figure 1). The four treatments were (i) control: thornbill song then song from a different individual thornbill; (ii) mobbing–mobbing: thornbill mobbing alarm then mimicry of heterospecific mobbing alarm; (iii) aerial–mobbing: thornbill aerial alarm then mimicry of a heterospecific mobbing alarm; and (iv) aerial–aerial: thornbill aerial alarm then mimicry of a heterospecific aerial alarm (electronic supplementary material, audio S1−S4). Mimicry was of alarm calls of New Holland honeyeaters, Philodonyris novaehollandiae, the species most frequently mimicked during nest defence [34]. The order of playbacks was determined by random selection of 20 of the possible 24 orders of four playbacks. Playbacks were about 2 s long, with each component lasting about 1 s, separated by 0.1 s of background noise. See further details in the electronic supplementary material.

Figure 1.

Spectrograms of exemplars of playback treatments. (a) Control: thornbill song then a different thornbill song. (b) Mobbing–mobbing: thornbill mobbing alarm then mimicry of honeyeater mobbing alarm. (c) Aerial–mobbing: thornbill aerial alarm then mimicry of honeyeater mobbing alarm. (d) Aerial–aerial: thornbill aerial alarm then mimicry of honeyeater aerial alarm. Mimicry was of New Holland honeyeaters. Spectrograms from Raven Pro 1.5, with Blackman window function and 512 sample window size, using electronic supplementary material, audio S1–S4.

Playbacks were carried out when a focal bird was handling chicken meat—simulating nestlings—at an artificial nest, and responses were videoed to enable blind scoring. See details in the electronic supplementary material. Two cameras were placed 5 m from the feeder at a 90° angle (front- and side-on) to enable a clear view regardless of the currawong’s orientation and recorded for 5 min from the start of playback. The time at the start of the playback was noted, and responses were scored after the sound was muted and the videos were randomized and renamed by another person. We used two measures of response, following Igic et al. [34]. First, the immediate response was classified as: (i) no response, the bird continued to feed; (ii) scan, the bird raised its head and looked around; and (iii) flee, the bird flew off within 2 s of playback initiation (electronic supplementary material, video S1). Second, we measured the delay to resume feeding as the period from the start of the playback until the bird resumed eating from the feeder. In three of the 80 playbacks, the bird fled and did not return, so we recorded the time as 5 min. In two other cases, the bird did not flee nor resume eating at the feeder, so we scored the period the bird scanned before feeding elsewhere: one fed on the ground; the other resumed manipulating food in its bill after pausing while scanning.

(c). Statistical analysis

We used R 3.4.3 [42] to carry out linear models to examine the effect of playback treatment on currawong response. To examine immediate response, we used two dichotomous analyses: (i) any response versus no response; and (ii) fleeing versus not fleeing. As there was complete separation of the data in some treatments (e.g. 100% response), we used bias-reduced binomial-response general linear models (‘brglm’ function in the package ‘brglm’; [43,44]). These models included Bird ID and playback order as fixed terms. To examine the duration of response, we used a linear mixed model (‘lme’ in package ‘nlme’), with playback order and sex as fixed terms and Bird ID as the random term. We used Wald tests for the statistical significance of model terms and Tukey’s post hoc test with p-value adjustment for pairwise comparisons. The duration data were first rank-transformed to meet model assumptions, due to non-normally distributed variation and several outliers, with a higher rank meaning a greater duration. The raw data are in the electronic supplementary material.

3. Results

(a). Alarm calls versus controls

Currawongs responded more strongly to alarm than control playbacks. Currawongs almost always responded, at least with scanning, to playbacks of alarm calls, but rarely responded to control song (figure 2; table 1; brglm pairwise comparisons versus control: aerial–aerial, z = 3.43, p < 0.001; aerial–mobbing, z = 3.46, p < 0.001; mobbing–mobbing, z = 3.00, p < 0.01). Similarly, birds delayed feeding for longer to alarm compared to control playbacks (figure 3; table 2; aerial–aerial, t₅₆ = 8.26, p < 0.001; aerial–mobbing, t₅₆ = 7.93, p < 0.001; mobbing–mobbing, t₅₆ = 3.45, p < 0.01). Playback order did not affect responses (tables 1 and 2).

Figure 2.

Immediate responses of currawongs to playback of control thornbill songs, and mobbing–mobbing, aerial–mobbing and aerial–aerial alarm calls. Alarm playbacks included a species-specific thornbill alarm followed by thornbill mimicry of a New Holland honeyeater alarm. n = 20 focal birds, each receiving all treatments.

Table 1.

Likelihood of currawongs displaying any immediate response (scan, flee) versus no response to playback, using a biased-reduced binomial general linear model (brglm). Log-odds ratios > 0 indicate currawongs are more likely to immediately respond to the first-named treatment compared to the second. Bolded p-values indicate statistical significance.

model term	log-odds ratio (s.e.)	z-value	p
pairwise among alarm-call treatments
aerial–aerial : mobbing–mobbing	1.276 (1.098)	1.162	0.245
aerial–aerial : aerial–mobbing	0.0946 (1.261)	0.075	0.940
aerial–mobbing : mobbing–mobbing	1.181 (1.060)	1.115	0.265
pairwise alarm calls versus control
mobbing–mobbing : control	2.552 (0.851)	2.997	<0.01
aerial–aerial : control	3.828 (1.115)	3.432	<0.001
aerial–mobbing : control	3.734 (1.079)	3.459	<0.001
other
PB order (1–4)	0.0606 (0.325)	0.186	0.852

Figure 3.

Boxplots of the rank duration of the delay to resume feeding by currawongs after playback, showing median, interquartile range and outliers. Larger ranks mean longer durations. Data were rank-transformed for analysis to normalize residuals. Playbacks included control thornbill songs, and mobbing–mobbing, aerial–mobbing and aerial–aerial alarm calls. Alarm playbacks consisted of a species-specific thornbill alarm followed by thornbill mimicry of a New Holland honeyeater alarm. n = 20 focal birds, each receiving all treatments.

Table 2.

Duration of delay to resume feeding after initiation of playback. Results from a linear mixed model comparing the ranked duration of delay time. T ratios > 0 indicate that the first-named treatment has a higher rank, and thus a longer delay, than the second-named treatment. Post hoc comparison p-values were Tukey adjusted. Bolded p-values indicate statistical significance.

term	F-value	df	p
intercept	488.548	1, 56	< 0.0001
PB order (1-4)	0.027	1, 56	0.870
sex	1.677	1, 18	0.212
treatment (4 levels)	31.040	3, 56	< 0.001

post hoc comparison	mean rank difference (s.e.)	t-ratio	df	adjusted p
mobbing–mobbing : control	17.275 (5.01)	3.451	56	< 0.01
aerial–aerial : control	41.338 (5.01)	8.257	56	< 0.001
aerial–mobbing : control	39.687 (5.01)	7.928	56	< 0.001
aerial–aerial : mobbing–mobbing	24.063 (5.01)	4.807	56	< 0.001
aerial–aerial : aerial–mobbing	1.651 (5.01)	0.330	56	0.988
aerial–mobbing : mobbing–mobbing	22.412 (5.01)	4.477	56	< 0.001

(b). Effectiveness of deception by alarm call type

Currawongs responded more strongly to aerial alarm choruses than to mobbing alarm choruses, showing that aerial calls are effective in deception. First, 40% (8/20) of currawongs fled after aerial–aerial alarm playback, but none did so after mobbing–mobbing alarm calls (figure 2; table 3; brglm pairwise comparison of aerial–aerial versus mobbing–mobbing: z = 2.62, p < 0.01). All but two birds scanned to the mobbing alarm playbacks, and none fled. Second, currawongs delayed feeding longer after aerial–aerial than mobbing–mobbing alarm playbacks (figure 3; table 2; aerial–aerial versus mobbing–mobbing: comparison of rank duration, t₅₆ = 4.81, p < 0.001). The median delay duration after aerial–aerial alarm playbacks was 8.3 s (IQ range 6.2−47.0), compared with 4.6 s (IQ range 3.9−5.8) after mobbing–mobbing playbacks. Playback order did not affect responses (tables 2 and 3).

Table 3.

Likelihood of currawongs fleeing to playback, using a biased-reduced binomial general linear model (brglm). Log-odds ratios > 0 indicate currawongs are more likely to flee to the first-named treatment compared to the second. Bolded p-values indicate statistical significance.

model term	log-odds ratio (s.e.)	z-value	p
pairwise among alarm-call treatments
mobbing–mobbing : control	0.134 (1.575)	0.085	0.932
aerial–aerial : control	3.626 (1.373)	2.640	< 0.01
aerial–mobbing : control	3.486 (1.374)	2.537	0.011
pairwise alarm calls versus control
aerial–aerial : mobbing–mobbing	3.491 (1.330)	2.624	< 0.01
aerial–aerial : aerial–mobbing	0.140 (0.750)	0.186	0.852
aerial–mobbing : mobbing–mobbing	3.352 (1.330)	2.521	0.012
other
PB order (1–4)	−0.473(0.354)	−1.334	0.182

(c). Type of mimicry

Currawongs reacted equally strongly to aerial–aerial and aerial–mobbing alarms, showing that the type of mimicry did not affect response when the initial species-specific call was an aerial alarm. Currawongs were equally likely to flee to these playbacks (figure 2; table 3; pairwise comparison: flee: z = 0.19, p = 0.85), with all birds either scanning or fleeing. Similarly, there was no difference in the delay to resume feeding (figure 3; table 2; pairwise comparison of ranked data: t₅₆ = 0.33, p = 0.99). The median distraction lasted 8.3 s for aerial–aerial and 8.8 s for aerial–mobbing playbacks. Again, playback order did not affect responses (tables 2 and 3).

4. Discussion

Our results confirm that thornbill alarm calling at the nest is deceptive, and suggest that the addition of mimetic alarm calls enhances deception by simulating more calling species rather than because of the type of alarm call mimicked. Currawongs were more likely to flee, and delayed feeding for longer, after purely aerial alarm choruses than purely mobbing choruses. The currawong’s response is therefore enhanced by thornbills using contextually misleading aerial alarm calls compared to contextually expected mobbing calls, confirming earlier evidence for deception. Currawongs reacted similarly to playbacks of thornbill aerial alarms combined with either mobbing or aerial alarm mimicry. This means that the type of alarm mimicry that follows species-specific aerial alarms does not affect deception, despite the finding that the inclusion of alarm mimicry itself does enhance deception [34]. This implies that mimicry is effective by creating the illusion of multiple calling species or individuals, rather than because a mimetic call is unusually effective or because it amplifies the message of the thornbill’s own aerial alarm calls.

(a). Deception of a nest predator

As predicted if thornbill calling is deceptive, currawongs were more likely to flee and delayed feeding for longer in response to aerial than mobbing alarm choruses. In natural situations, thornbills switch from giving only mobbing calls to including species-specific and mimetic aerial calls when their nestlings are disturbed and give distress calls [33,34]. The thornbill’s misleading switch to including aerial alarm calls, despite the lack of an airborne threat, is probably effective because it signals an immediate threat from an airborne predator, such as a goshawk, which is also a predator of currawongs [34,41]. Thornbills are therefore exploiting the currawong’s eavesdropping on heterospecific alarm calls and response according to call meaning, that would normally provide reliable information on danger. To our knowledge, deceptive use of alarm calls to fool nest predators has not yet been demonstrated in other bird species, although female superb lyrebirds (Menura novaehollandiae) produce mimetic vocalizations, including alarm and predator calls, when their nest is under threat, suggesting a role in nest defence [45]. Some other species of birds mimic sounds that resemble dangerous species such as snakes or sparrowhawks, which is potentially Batesian mimicry [46–50]. Overall, our results support the conclusion that thornbill calls are deceptive, by confirming the assumption that currawongs discriminate between aerial and mobbing alarm calls.

(b). Mimetic model selection

Currawongs did not respond more strongly to aerial–aerial choruses than aerial–mobbing choruses, implying that following a species-specific aerial call, the type of alarm call mimicked has no additional effect. Currawongs were equally likely to flee and delayed feeding for a similar period. This result is consistent with natural patterns of calling, given that thornbills typically do not stop giving mobbing alarms when they switch to including aerial alarms in response to nestling distress calls [34]. Nonetheless, although the type of mimetic alarm calls did not matter, our previous experiment found that currawongs delayed feeding for about 70% longer when playbacks of aerial alarm choruses included mimicry, not just species-specific alarm calls [34]. Taken together, these studies show that it is the inclusion of alarm mimicry, rather than the specific type of call mimicked, that produces the stronger reaction in an eavesdropping currawong.

We suggest that the inclusion of vocal mimicry in addition to species-specific alarm calls is effective because it creates a strong illusion of multiple individuals calling, which is normally a reliable indication of danger. Dangerous predators often prompt choruses of alarm calls, so a greater number of callers is a more reliable cue of danger than a single caller [26,28,51]. Multiple individuals giving aerial alarm calls would reinforce the message of an airborne predatory bird, while a chorus including an aerial and then mobbing call implies that a predatory bird is nearby, even if currently landed. Consistent with indicating a greater threat, playback of multiple alarm calls can lead to stronger anti-predator responses by both birds and mammals [26,29,31,52,53]. Assessing the number of callers of the same species could be relatively difficult and rely on subtle differences in voice between individuals or spatial separation of callers, as simulated by stereo speakers in most experimental studies. By contrast, including mimicry of heterospecifics could be effective at creating an illusion of multiple callers, even when calling from the same location, because each species has a distinct call. This is consistent with our previous study, which found that the inclusion of mimicry enhanced deception even though, when broadcast alone, the mimicked honeyeater aerial alarm did not have a greater effect on currawong response than the thornbill’s own aerial alarm call [34]. Another benefit of mimicry is that alarm calling by multiple species might be a more reliable indication of danger than multiple calls from a single species, and in addition, indicate a generalist predator [30], and so more likely a threat to currawongs.

Another potential benefit of using mimetic alarm calls during nest defence is that it may slow a predator’s ability to learn about and recognize deception. In general, the effectiveness of defensive mimicry declines as it becomes more common [54–56] so the ability to mimic multiple models could counteract the predator’s ability to recognize deception [25]. In support of this idea, fork-tailed drongos switch the species of alarm call mimicked once their kleptoparasitic victims reduce their response to one type of alarm call [19].

In conclusion, our results confirm that thornbills deceptively exploit the currawong’s ability to eavesdrop on thornbill alarm calls and respond according to alarm-call types. We suggest that deceptive alarm calling at nests may be more common than appreciated, given that alarm calling is also used to communicate with conspecifics when predators are nearby [57], so researchers may assume that calls are used exclusively to communicate with conspecifics. Furthermore, although less effective than choruses including mimicry, purely conspecific thornbill aerial alarm choruses also deterred currawongs [34], so even non-mimetic species might use aerial alarms deceptively in this context, to add to other deceptive behaviours near nests and young [58]. Deceptive calling might therefore be widespread and not limited to mimetic species. Among mimics, there is the opportunity for research on model selection, testing the ‘multiple callers’ and ‘multiple species’ hypotheses, and the role of predator behaviour in driving mimicry.

Ethics

The study was approved by the Australian National University Ethics Committee (A2015/67). The experiment used artificial nests, and playbacks and did not endanger subjects.

Data accessibility

All data are provided in the electronic supplementary material, in addition to the R code used to statistically analyse the results. The data file has self-explanatory column names and coding, without abbreviations, and all methods are in the main document or text supplement.

Supplementary material is available online [59].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

L.A.: conceptualization, data curation, formal analysis, investigation, methodology, validation, writing—original draft, writing—review and editing; B.I.: conceptualization, methodology, supervision, writing—review and editing; R.M.: conceptualization, funding acquisition, methodology, project administration, resources, supervision, visualization, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

The work was funded by the Research School of Biology at the Australian National University and an Australian Research Council Discovery Grant, DP150102632.