Ethics statement

Our stimuli collection from human participants and all experimental procedures were approved by the University of Roehampton Life Sciences Ethics Committee (Ref. LSC 19/ 280), with the latter being in line with ASAB guidelines for the use of animals in research (ASAB/ABS, 2020). The feasibility and adequacy of protocols we used to address our research questions were additionally approved by a Life Sciences Research Student Review Board prior to onset of testing (protocols not externally registered). All tests were non-invasive and lasted a maximum of seven minutes per subject for each trial. No animals were euthanised and all were released into a large outdoor paddock to join the rest of the herd following testing. We collected signed consent forms from human participants who provided stimuli for the current experiment.

Study site & sample population

We conducted experiments between 8th September–19th October 2020 and 19th May–16th July 2021 at Buttercups Sanctuary for Goats (http://www.buttercups.org.uk/) in Kent, UK (51°13′15.7″N 0°33′05.1″E). During these periods, the sanctuary was open to visitors who were freely able to approach and interact with the goats, including opportunities to feed them. Goats had daily access to a large outdoor area (approximately 3.5–4 acres) and were kept individually or in small groups within a large stable complex at night (mean pen size = 3.5 m²). Throughout the day, animals had ad libitum access to hay, grass and water, and were supplemented with commercial concentrate according to age and condition. Subjects came from a variety of backgrounds having been removed from their owners due to mistreatment or when owners could no longer care for them. We selected a target sample size based on a power analysis of our results from the first year of testing (2020; using the response, time spent looking after the start of each voice playback) and on sample sizes of similar studies (e.g., Adachi, Kuwahata & Fujita, 2007; Proops, McComb & Reby, 2009). Our final sample comprised 26 adult rescue goats (17 castrated males and nine intact females) that were well habituated to human proximity, were of various breeds and ages, and had resided at the sanctuary for over eight months (for detailed subject information, see Table S1).

Stimuli collection & preparation

We selected goats that had been described by a particular caretaker, volunteer or frequent visitor (from whom stimuli were collected for the current experiment) as being particularly responsive to them. We assumed the reported preferential attention given to a particular person (hereafter known as a preferred person) would indicate goats were more familiar with their individual characteristics. We always tested goats with stimuli from one preferred person and one gender-matched, non-preferred caretaker to avoid potential reliance on gender as a cue and with the aim of ensuring subjects were familiar with all human stimuli presented.

In 2020, we collected photographs and voice samples from two male and three female caretakers. Photographs and voice samples for each person were collected in a single session at a two-meter distance, outdoors (as stimuli would normally be experienced by goats) and at the same location and time of day. In 2021, we collected stimuli from five additional volunteers and frequent visitors to the sanctuary and one familiar staff member not involved with carrying out husbandry procedures (five women and one man) who had developed bonds with individual goats. For these, we took photographs and voice samples based on availability, meaning similarity in time of day could not be maintained, and location was sometimes shifted based on prevailing light conditions (otherwise conditions were as described above). Goats tested in 2020 were subject to photographic and vocal stimuli from a pair of caretakers, but in 2021, one of the two people goats experienced stimuli from (the preferred person) was not responsible for carrying out husbandry procedures (e.g., hoof trimming), which goats likely perceived as aversive. Caretakers were also involved in feeding the goats, occasionally providing favoured food items, with the volunteers and frequent visitors regularly provisioning preferred food more specifically to particular goats. This could have affected the nuance of the relationship with their preferred person for goats tested in 2020 compared to those tested in 2021 (the year a goat was tested was controlled for during statistical analysis).

To collect visual stimuli, we asked each person to maintain a neutral expression and face the camera (Panasonic Lumix DMC-FZ45) before several frontal photographs of their head and shoulders were taken against a white background. Photographs were later edited so only the head and neck were visible, processed to improve clarity and brightness, before being blown up to A3 landscape size (slightly larger than natural head-size).

While collecting vocal recordings, we asked each person to say the phrase: “Hey, look over here”, several times in a manner they would normally use to address the goats. Speech was intonated, and we avoided using goats’ names and other potentially salient words (e.g., food-related vocabulary) to test whether potential recognition can be generalised based on vocal features, rather than being restricted to specific familiar words or phrases (Kriengwatana, Escudero & Ten Cate, 2015). Voice samples were recorded using a Sennheiser MKH 416 P48 directional microphone and a Marantz PMD-661 digital recorder (sampling rate: 48 kHz, with an amplitude resolution of 16 bits in WAV format). We selected the clearest voice sample (mean sample length ± SD = 1.28 s ± 0.35) with the lowest background noise and shifted the mean amplitude to 70dB to ensure consistency between samples and compiled these into a playback sequence using Praat v.6.1 (Boersma & Weenink, 2019). Playback sequences comprised five seconds of silence before the first voice playback, followed by 10 s of silence, a repeat of that playback, and 30 s of silence. The 10 s following initiation of each of the two playbacks in a sequence will be hereafter referred to as the ‘response period.’

Experimental enclosure

The experimental enclosure was constructed out of opaque metal agricultural fencing and barred metal hurdles in a large outdoor paddock that goats had ready access to throughout the day. The enclosure was divided into five sections (Figs. 1; S1). Goats entered through the preparation pen where they were equipped with a heart rate monitor before experiments. Trials took place in the experimental arena. A corner of the arena (the holding pen) was sectioned off with hurdles and used both for training (see ‘Training Phase’) and to prevent subjects from having visual access to experimenters during trials. We erected a semi-opaque green barrier around the fencing’s edge to prevent subjects from being able to see inside the preparation pen and outside the arena to reduce distractions and unintentional cuing from experimenters, visitors and other goats. Two Sony CX240E video cameras (frame rate = 25 FPS) were positioned at the front and back of the arena. The camera at the front was hidden under camouflage netting to prevent subjects attributing it as the sound source. We placed a target bucket filled with compressed hay next to the hurdle barrier, directly in front of the stimuli array. Goats were encouraged to approach and investigate this bucket as from this position they would have a clear view of the photograph, unimpeded by the barrier.

During experiments, we affixed the appropriate photograph to a predetermined position at the front of the arena (top edge of image was 93.5 cm off the ground). This position ensured that the speaker (a Bose Soundlink Mini Bluetooth Speaker II), which was placed directly behind the image was at approximately mouth level (70 cm off the ground). The speaker used to broadcast playbacks in the current experiment has been confirmed to accurately reproduce the frequency range of the human voice (Ben-Aderet et al., 2017). We separated off the front of the arena using hurdles, imposing a distance between the subject and stimuli set of a minimum of 2.4 m. Goats have good visual acuity, being able to discriminate small details (3.4 cm symbols) at distances of 1.5–2 m (Blakeman & Friend, 1986).

Habituation phase

We habituated subjects to staying in the experimental arena alone for extended periods over three, four-minute sessions that took place prior to testing, and occurred between a minimum of three hours and a maximum of eight days apart (mean ± SD = 2 days 6 hrs ± 1 day 21 hrs). The variation in interval between successive habituation sessions was necessary as goats were not always willing to be led to the experimental enclosure or were displaced by more dominant individuals en route. Before starting each habituation session, we equipped goats with a Zephyr™ BioHarness 3.0, which in conjunction with AcqKnowledge v. 4.4.2 software (BIOPAC System Inc.) transmitted live cardiac data to a laptop (HP ProBook 650 G4) via Bluetooth during experimental trials. If subjects exhibited signs of aversion or aggression while this device was being fitted, we stopped and restricted future measurements taken during such trials to behaviour only (cardiac data collected from 15 out of 26 goats; for further details see Table S1).

For the first habituation session, an experimenter sat still with their head facing downwards at the front of the arena, where the photograph would be placed during experimental trials. As initially goats were not completely isolated, we aimed for this to reduce neophobia and encourage subjects to investigate their surroundings. For the following two sessions, subjects were kept alone in the arena. We occasionally repeated sessions when previous ones were halted due to adverse conditions such as rain, or when a long period of time had elapsed since the goat’s third habituation session (approximately seven days). For all three habituation sessions, we ensured animals had access to water and small food rewards (dry pasta), the latter of which were placed in and around the target bucket to encourage them to investigate and spend more time at the front of the arena. The habituation phase preceded the test phase by a minimum of three hours and 15 min and a maximum of eight days (median = 1 day; interquartile range, IQR = 1 day 4 hrs 38 mins).

Training phase

The training phase took place directly before each experimental trial and aimed to incentivise goats to approach and investigate the target bucket during the test phase. Goats entered the enclosure through the preparation pen where we fitted them with the BioHarness, which, subjects had become accustomed to wearing during habituation sessions. To attain a clearer ECG trace, we clipped a patch of fur around the left shoulder blade at least one day prior to testing, over which the BioHarness module would be positioned. ECG gel was applied liberally to sensors on the BioHarness belt before it was placed around the subject’s thorax to improve conductance to the skin. Once a clear ECG trace had been secured, we led goats into the holding pen to begin training.

For training trial 1, Experimenter 1 and the subject were positioned inside the holding pen. Experimenter 2 held a piece of pasta (or cracker if goats were not sufficiently motivated by pasta) near the subject’s head and once they had noticed it, walked slowly backwards towards the bucket while actively maintaining their attention. They then slowly and overtly placed two pieces of pasta inside, before crouching by the bucket and looking at it. Experimenter 1 then released the subject from the holding pen and stood aside, allowing the goat to pass and retrieve the pasta. The subject consumed both pieces of pasta before being led back into the holding pen. Training trial 2 proceeded similarly to the first, but instead of crouching down by the bucket, Experimenter 2 released the goat from the holding pen and stood aside to allow them to retrieve the reward. For training trials 3 and 4, Experimenter 2 gained the subject’s attention from the bucket, before baiting it and releasing them from the holding pen. Goats passed a trial if they successfully reached the bucket within 30 s after release and needed to pass all four trials consecutively to proceed to the test phase. If a subject failed to pass a trial, the training process was repeated from the beginning and if they were not sufficiently motivated to participate in training, they were released, and the experimental trial was carried out on a different occasion.

Test phase

Goats experienced four experimental trials each (3–14 days apart; mean = 7 days). Two trials were congruent (face and voice were from the same person) and two incongruent (from different people), with subjects experiencing all combinations of photograph and voice playbacks from a preferred person and a caretaker in a random order (randomisation carried out using sample function in R: R Core Team, 2020).

Once training had been completed, we led subjects back into the preparation pen and following checks to ensure the ECG signal had been maintained, they were distracted while the photograph was positioned. The goat was then led back into the arena and the holding pen shut behind them. Experimenters 1 and 2 quietly hid behind the visual barrier in the preparation pen and monitored subject movement in a nearby video camera’s LCD monitor (Fig. 1). We expected goats to approach and inspect the bucket (no pasta or water available during test phase) and when they were positioned close to and facing the front, we initiated the playback sequence (mean maximum amplitude ± SD = 76 dB ± 2 measured 1m away under field conditions using a CEM™ DT-8851 sound-level meter). Goats listened to a single sequence of two voice playbacks, with behavioural and cardiac responses being measured for 10 s following the initiation of each playback. If a subject moved away within the five seconds preceding the first playback, where possible the playback sequence was aborted, and reinitiated when they were in a good position. If goats failed to be in a suitable position for six minutes following trial initiation (n = 11 trials), the trial was suspended, and they were released. We placed an event marker in the ECG trace to indicate occurrence of each playback for later analysis of cardiac measures. Goats that had heard both playbacks were released 30 s after the final playback (median duration in arena = 1 min 20.95 s; IQR = 64.12 s). Of the 33 subjects tested, we excluded one from further testing due to health concerns, one for repeatedly failing to be in a suitable position for stimuli presentation, and a further subject and two trials from another due to a lack of training motivation.

Video coding

A single observer coded behavioural data using BORIS v. 7.8.2 (Friard & Gamba, 2016), with goat latency to look towards the photograph and looking duration measured for 10 s following the initiation of each voice playback defined to the nearest frame (0.04 s). Goat looking behaviours were coded from when the goat’s head was turned in the direction of the stimuli array, until they turned away. Behavioural measurements were extracted from footage captured from the front and back of the arena, before being compared and combined. Measurements taken from the front were often clearer and without a blind spot, so these took precedence when quantifying behaviour. However, this was not always possible, with technical issues in the front camera for three trials and the back camera for two preventing experiments from being successfully captured by both, so in these cases we coded behaviours from the footage available. For one of these trials, the subject went into the camera’s blind spot, so only behaviours that could be coded with certainty were recorded, with others defined as missing. In one further trial, the photograph fell down during testing, so it was also excluded from analysis.

Although we analysed behaviour coded by a single observer, a second observer, who was blind to the experimental condition, independently extracted behavioural measurements from 20% of trials. Agreement between observers proved to be high for goat duration of looking (concordance correlation coefficient: 0.917, 95% CI [0.849–0.956]) and latency to look (0.902, 95% CI [0.816–0.949]; DescTools package: Signorell, 2017). However, Bland-Altman plots indicated differences in latency measurements for observations where goats took longer to look.

Exclusion criteria

Prior to analysis, we applied three exclusion criteria to determine which trials should be included in our final data set. Firstly, if goats failed to look during both response periods in the playback sequence, the trial was excluded as subjects were interpreted as not being sufficiently attentive to human cues to notice incongruencies between them (n = 21 trials: n = 6 congruent and n = 15 incongruent). Secondly, if goats were already looking towards the stimuli array before a playback had been initiated, these trials were excluded as goats were considered unlikely to be responding to differences in congruency between visual and vocal cues (n = 23 observations: n = 16 congruent and n = 7 incongruent; n = 4 trials: n = 2 congruent and n = 2 incongruent). Finally, once the first and second exclusion criteria had been applied, we also excluded goats that did not look for at least one congruent and one incongruent trial as these prohibited within-subjects comparison (n = 3 subjects excluded). Ultimately, we analysed 85 trials from 26 subjects (n = 45 congruent and n = 40 incongruent trials; see Table S1 for more information), eight experiencing stimuli from men, and 18 from women.

Data analysis

Latency to look following playbacks

The time taken for goats to look towards the photograph following the presentation of a voice playback was not significantly affected by congruency between these human cues (Table 1; Fig. 2A). Playback number (1 or 2), the year a goat was tested (2020 or 2021), their sex, the gender of the human stimuli presented, and whether the voice or the photograph came from a preferred person also did not affect their latency to look.

Looking duration following playbacks

Congruency between the human face and voice presented did not significantly affect how long goats looked towards the stimuli array following presentation of voice playbacks. Neither did playback number, the year a goat was tested, their sex, the gender of the human cues presented or whether the face and voice shown belonged to a preferred person (Table 2; Fig. 2B).

Heart rate & heart rate variability

Changes in goat heart rate (Fig. 3A) and HRV from baseline values (Fig. 3B) were observed over the playback sequence (congruency x playback number interaction significant: Table 3), with heart rate decreasing between playbacks 1 and 2 when the human face and voice presented were incongruent (β ± s.e. = 7.198 ± 1.641, t-ratio = 4.385, p = 0.0006: Table S5). The results were broadly similar between the full models presented (Table 3) and more basic models containing only the effect of the congruency and playback number interaction, and the random effect of trial number nested within the goat being tested (for results of basic models, see Tables S2–S4). No significant changes in heart rate were observed when goats were shown congruent cues, nor were there any significant pairwise changes in HRV between playbacks 1 and 2 based on congruency (Tables S5; S6). However, when there was a longer measurement period over which to calculate goat HRV, we found that these values were significantly lower than when shorter periods were available (Table 3).

We applied the same exclusion criteria for our cardiac data set as for our behavioural one. Specifically, trials were excluded when goats did not look for both response periods of the playback sequence, and goats when they did not look for one congruent and one incongruent trial. Additionally, trials were excluded when goats were looking before a playback had been initiated. We considered this latter criterion to more likely effect behavioural over cardiac measures. Therefore, to evaluate the robustness of the effect congruency and playback number on goat cardiac responses, we added observations where goats had already been looking before the onset of a playback back into our cardiac data set. However, when these observations were reinstated, congruency, in combination with playback number no longer affected goat heart rate (β ± s.e. = −3.426 ± 2.214, t = −1.547, p = 0.129) or HRV (β ± s.e. = 5.220 ± 4.333, t = 1.205, p = 0.233).