Theoretical Background

Why study music perception in other animals? Cross-species studies of music cognition provide a way to study the evolutionary roots of our musical capacities (McDermott & Hauser, 2005; Fitch, 2006). Music cognition involves many distinct capacities, ranging from "low-level" capacities not specific to music, such as the ability to perceive the pitch of a complex harmonic sound, to "high-level" capacities which appear unique to music, such as the processing of tonal-harmonic relations based on learned structural norms (Peretz & Coltheart, 2003; Koelsch, 2011). It is unlikely that all of these capacities arose at the same time in evolution. Instead, different capacities are likely to have different evolutionary histories. Cross-species studies can help illuminate these histories using the methods of comparative evolutionary biology (for further discussion, see Patel & Demorest, 2013).

For example, consider the capacity to sense, and move in synchrony with, a perceived regular temporal pulse (e.g., via foot tapping, head bobbing, or dance). Musical beat perception and synchronization (BPS) is a fundamental aspect of music cognition that develops without special training in young children, and is observed in every human culture (Brown & Jordania, 2011; Nettl, 2000). Did the capacity for BPS arise uniquely in the human lineage as part of an evolutionary adaptation for music-making (Bispham, 2006; Honing, 2012), or does it reflect ancient abilities shared by many species? Darwin (1871) seems to have favored the latter view, stating that "The perception, if not the enjoyment, of musical cadences and of rhythm is probably common to all animals, and no doubt depends on the common physiological nature of their nervous systems" (p. 1207 in the 2006 edition, E.O. Wilson, Ed.). Darwin's intuition is reasonable: it seems plausible that the capacity to sense a beat in music and synchronize to it could be widespread among animals. After all, the auditory systems of humans and other mammals (especially primates) have many structural and functional parallels (Rauschecker & Scott, 2009), and all vertebrates produce voluntary rhythmic movements (e.g., as part of locomotion). Therefore, it is reasonable to expect that many animals could learn to coordinate rhythmic movements with a perceived auditory beat. Furthermore, nonhuman animals (henceforth, animals), such as rhesus monkeys, can learn complex sensorimotor tasks in the laboratory (Georgopoulos, Taira, & Lukashin, 1993). Thus one might expect that many animals could learn to synchronize movements with an auditory beat, especially since rhythm is increasingly thought to be a fundamental organizing principle of brain function (Buzsáki, 2006).

Another conceptual possibility is that the capacity for BPS is neither uniquely human nor widespread among animals, but restricted to a few species. This view was suggested by Patel (2006), who proposed that the capacity for BPS arose as a fortuitous byproduct of the brain circuitry for complex vocal learning. According to this view, the capacity for BPS did not arise as a result of an evolutionary adaptation for music-making, but a secondary use (or "exaptation", Gould & Vrba, 1982) of brain circuits that evolved for other reasons. This "vocal learning and rhythmic synchronization hypothesis" was motivated by the fact that vocal mimicry, like BPS, involves tight links between the auditory and motor systems, and the fact that the neural substrates for vocal learning and BPS appear to overlap in the brain (see Patel, 2006 for details). Crucially, vocal learning is a rare capacity found only in a few groups of mammals (including humans, dolphins, elephants, and seals) and birds (songbirds, parrots, and hummingbirds), and is associated with neuroanatomical specializations including modified auditory-motor cortical connections (Jarvis, 2007). Thus the vocal learning hypothesis makes a testable prediction: only vocal learners are capable of BPS.

The discovery of BPS in parrots has supported this hypothesis (Hasegawa, Okanoya, Hasegawa, & Seki 2011; Patel, Iversen, Bregman, & Schulz 2009a; Schachner, Brady, Pepperberg & M. Hauser, 2009) as has the finding that rhesus monkeys (who are vocal non-learners) could not learn to synchronize their taps to a metronome despite over a year of concerted training (Zarco, Merchant, Prado, & Mendez, 2009). However, many other vocal non-learner species remain to be tested. As noted above, horses are of particular interest due to the many anecdotal accounts of BPS in domesticated horses. If horses show the key features of BPS, this would challenge the vocal learning hypothesis and raise new questions about the evolutionary substrates of this ability.

Key features of musical beat perception and synchronization

Synchronization to a musical beat in humans has several important features that distinguish it from other examples of rhythmic entrainment in nature, such as the synchronous chorusing of certain insects and frogs (Patel et al., 2009b). First, BPS involves the extraction of a beat from a complex acoustic stimulus (i.e., music vs. from simple pulse trains). Second, BPS involves substantial flexibility in movement tempo: humans can easily synchronize their movements to a musical beat across a fairly wide range of tempi. Third, BPS is cross-modal, with rhythmic sound driving movement that is not necessarily aimed at sound production. Thus a convincing demonstration of BPS in animals requires demonstrating these three features. In addition, an important feature of BPS is phase matching: people spontaneously align their taps and other rhythmic movements with a beat. That is, people tap slightly before the beat or on the beat. Stated more formally, the average temporal asynchrony between taps and beats is slightly negative or around zero (Patel, Iversen, Chen, & Repp, 2005; Rankin, Large, & Fink, 2009). This indicates accurate temporal prediction of beats. Hence if an animal is trained to tap with a beat, and taps at the correct tempo after each beat (e.g., with a delay of a few hundred ms), this behavior could be largely reactive rather than anticipatory. (This sort of behavior was observed in the study of Rhesus monkeys mentioned above.)

General considerations

In testing animals for synchronization to a musical beat, several important methodological issues arise (see Patel et al., 2009b for a detailed discussion of 11 such issues). A primary issue is eliminating possible visual, auditory, or tactile rhythmic cues from humans, which could lead to a "Clever Hans" effect. This concern particularly applies to horses, which are capable of sensing human cues in many modalities (e.g., Proops et al., 2010; Saslow, 2002). In particular, the horse's sensitivity to tactile cues means that an essential requirement for studies of BPS is that the horse not be ridden. In addition, any humans in close proximity to a horse during testing should be "deaf" to the music that the horse is hearing so that they cannot give inadvertent cues.

Another primary issue is that the animal be tested for tempo flexibility, i.e., for its ability to synchronize to music at different tempi. The danger with using a single tempo is that a horse could appear to synchronize to music if the music tempo happened to match its natural trotting tempo, even without any auditory-motor coupling. In prior research on BPS in another species (a cockatoo), testing for tempo flexibility was accomplished by choosing a single song with a clear beat and creating versions at several different tempi using software that allows changing tempo without changing pitch (Patel et al., 2009a). In that prior work, 11 different versions of a song were created ranging from 20% slower to 20% faster than the original. The use of different tempi allows one to test for tempo flexibility, a key feature of BPS, and is also crucial for the statistical permutation tests used to check for true synchronization to a musical beat, as discussed further below. For certain animals, the number of tempi used might need to fewer, depending on the particular motor abilities of the animal. For example, with horses one should ensure that the horse is capable of trotting as slow as the slowest tempo and as fast as the fastest tempo. In general, we suggest that no fewer than five distinct tempi be tested (e.g., -15%, -7.5%, 0, +7.5%, and +15% relative to the original musical piece). According to one source (Gallo, 2007), the average trotting tempo for horses is 152 beats per minute [BPM], based on the timing of footfalls of the front legs. That is, if one measures the number of steps per minute as the horse trots, counting each front leg footfall as a step, the average rate is 152 steps per minute, though this number depends on the particular horse. Author DL feels that the comfortable range of trotting for horses likely spans about 140 — 200 BPM. Thus if using a tempo range of +/-15% relative to the original musical piece, a base tempo of about 165 BPM (e.g. "Hound Dog" by Elvis Presley) would ensure that the slowest and fastest tempi fall within this range.

A final issue is the use of objective methods to measure synchronization to a musical beat. Simply examining videos and making perceptual judgments of synchrony is not a valid way to study BPS in other species, as humans have a tendency to perceive synchrony between rhythmic auditory and visual stimuli even when it is not objectively present (Arrighi, Alais, & Burr, 2006). Thus objective measures, such as frame-by-frame coding of the animal's movement from video (with the sound track turned off) are needed. Some portion (ideally all) of the data should be coded independently by at least two researchers in order to measure inter-rater reliability.

The circular trotting to music (CTM) method

The circular trotting to music (CTM) method is a way of testing whether horses spontaneously synchronize their trotting to a musical beat. This method involves a horse trainer and an equipment controller working with one horse in an indoor enclosure (to minimize background noise). The trainer stands at the center of the enclosure and the horse trots in circles while music is played over loudspeakers in the arena. To prevent the trainer from hearing this music, the trainer wears sound-isolating insert earphones and closed-ear noise-protecting earmuffs (e.g., of the type used when operating loud power equipment). The earphones are connected to a portable digital audio device, and throughout the experiment the trainer listens to music without any beat, such as meditation music, at a sound level sufficiently high to mask out the music heard by the horse. The equipment controller triggers the start of experimental trials by starting the music and video camera, at which point s/he gives a visual cue to the trainer that an experimental trial has begun. Further information on the roles of the trainer and equipment controller are given below. Trials from the current study can be viewed online (see Appendix 2), and the reader is encouraged to watch these videos in conjunction with reading the information below. Our study focused on a young adult male horse named "True" (Breed: Morgan), who was approximately 6 years old. Testing took place in an indoor enclosure measuring approximately 75' x 125', with a dusty sand floor (Horse Quarters in Davis, CA). The study took place in March of 2010.

ROLE OF THE TRAINER

The trainer interacts with the horse during the experiment. Specifically, once the music for a given trial has started, the trainer gives the horse a "go" signal to start trotting in circles around the trainer. The horse continues trotting until it elects to stop or the music ends (in which case the trainer gives a "stop" signal). The trainer can give the horse a food reward at the end of trials if needed.

In our study, the trainer and horse were connected by a long slack rope, a 22-ft "lead line" (pronounced: "leed" line) made of yachting braid made by Parelli Natural Horsemanship, used in conjunction with a ¼" yacht braid rope halter. The trainer held one end of this rope with one hand, and with the other hand held a thin rigid pole connected to a flexible whip (in Parelli terminology, a "carrot stick and string", Figure 2). Using body language, the trainer gave the "go" signal to the horse using the rope and/or whip. The horse was allowed to circle the trainer either clockwise or counterclockwise, depending on its preference, which could vary from trial to trial. As the horse circled (typically at a distance of about 10' from the trainer), the trainer turned in place to maintain eye contact with the horse. The trainer continued to hold the rope while the horse circled, but no tension was applied to the rope, which often dragged lightly on the ground, meaning that the horse set its own pace. The whip was never applied once the horse started trotting, and thus was not a source of rhythmic cues. On five trials in the current study the horse trotted around the trainer "at liberty", meaning that no rope connected the horse and trainer (the trainer still turned in place to face the horse as it trotted). Also, on three trials the horse started trotting before the music started playing. In future work, the use of a lead line (vs. liberty) and the onset of horse trotting relative to the music (before or after music onset) should be made consistent.

Fig 2. Screenshot from video of an experimental trial, showing the horse and trainer. Please see the online videos for full experimental trials.

Data analysis method

The experimental video was segmented into individual trials based on the start and end of the music stimulus on each trial. For each trial, the video and audio tracks were separated so that an audio file and a silent video file were obtained for further analysis, using AviSynth (avisynth.org).

STATISTICAL SYNCHRONIZATION ANALYSIS: IDENTIFYING SYNCHRONIZED BOUTS

Following the methods of Patel et al. (2009), the statistical test of BPS is divided into two steps. The first step is to examine each trial for bouts of synchrony (i.e. time windows when footfalls and musical beats are aligned). This process was conducted on the current study, and is described in this subsection. The second step (not possible in the current study, due to lack of trials at different tempi) is a statistical permutation test that provides the true test of BPS. This permutation test is further described in the next subsection.

To look for synchronized bouts within a trial, the timing of each footfall was converted into a relative phase (RP) value within the time interval bounded by the preceding and succeeding musical beats (Figure 3). For example, if a musical beat occurred at 1.0 and 2.0 seconds and a footfall at 1.25 seconds, the footfall's RP would be coded as 90 degrees (¼ of the beat interval after the beat); whereas a footfall at 1.50 seconds would be labeled as 180 degrees (½ of the beat interval), and a footfall of 1.75 seconds would be coded as -90 degrees (¼ of the beat interval before the beat) Thus a sequence of N footfall times in a given trial would be converted to a sequence of N RP values.

Fig 3. Illustration of relative timing of footfalls and musical beats. Four footfalls (vertical grey lines) and four musical beats (black dots) are depicted (FL = Front left foot, FR = front right foot). The first two footfalls have positive relative phase (40 and 20 degrees) since they occur after the nearest musical beat. The third footfall has zero relative phase since it aligned in time with the musical beat. The fourth footfall has negative relative phase (-20 degrees) since it occurs before the nearest musical beat.

Once the RP sequence for a given trial was computed, we searched for bouts of synchrony by performing a windowed analysis on each sequence. Specifically, the first eight RP values were examined with a phase-sensitive Rayleigh test (Fisher, 1993) to test if footfalls and beats were synchronized. Then, the window was shifted by 4 RP values and this test was repeated (i.e., window 1 was RP values 1-8, window 2 was RP values 5-12). This process of window-shifting and statistical testing was repeated until the end of the RP sequence was reached. A synchronized bout was defined as at least two consecutive windows (i.e., a minimum of 12 total RP values) with p<0.05.

How does the phase-sensitive Rayleigh test work? Briefly, this test uses a sequence of RP vectors to compute a mean vector, which can be represented on the unit circle (Figure 4). After computing this mean vector, the length of this resultant vector projected onto zero phase (the horizontal axis) provides a measure of synchrony. More intuitively, the test examines whether the RP vectors tend to cluster around zero phase, as would be expected if footfalls and beats were in synchrony and phase-matched.

Fig 4. Illustration of 8 relative phase values depicted as light blue vectors on the unit circle. The numbers around the circle are degrees (0 represents perfect alignment of footfall and musical beat; negative numbers indicate footfalls preceding the nearest musical beat; positive numbers indicate footfalls following the nearest beat). Note that two vectors have a very similar value (near 3 degrees) and hence overlap in this image. The dark red arrow is the mean vector of these eight phase vectors, and the black arrow is a projection of the mean phase vector onto the zero phase axis.

Results from the current data

We collected data from 16 trials and analyzed 15 of them (one trial was excluded since we were unable to get an accurate result from the beat-tracking algorithm). Trials varied in length from 21.5 - 62.7 seconds (mean=41.3 seconds; trial onset was defined as the point when True started trotting to the music, and offset as the point when he ceased trotting, which always occurred before the music ended). The number of footfalls per trial ranged from 57-156 (mean=101 footfalls per trial, total number of footfalls across trials = 1537). The two musical stimuli used for this experiment were very similar in tempo (both had tempi between 155.05 and 156.81 BPM when measured using the beat tracker).

After converting the footfall and musical beat timing data to sequences of relative phase values (as described above), we found that nearly all trials (14/15) had at least one synchronized bout (range: 1-4 bouts, mean: 1.79 bouts per trial). The number of footfalls that were part of synchronized bouts ranged from 12-72, with a mean of 35.42 synchronized footfalls per trial in the 14 trials with bouts (on these trials, an average of 33.9% of the footfalls were part of synchronized bouts). The total number of synchronized footfalls across all trials was 496. To test whether this amount of synchrony could occur by chance, and to test for tempo flexibility (a key feature of BPS), one would need data from trials at several different tempi, as discussed in the previous section.

Figure 5 shows relative phase data from one trial with a single bout of 40 synchronized footfalls. The synchronized bout is indicated with a black box. One notable aspect of this figure is that when True's tempo is slightly faster or slower than the music tempo, the relative phase slowly drifts from -180 to +180 degrees ("phase wrapping"), but during the bout stabilizes near zero, indicating footfalls at the same tempo as the music and synchronized with the musical beat.

In summary, True exhibits synchronized bouts, but given the limitations of this dataset it is not possible to say whether or not this is a coincidence. Testing genuine synchronization to a musical beat in horses awaits future research that includes more experimental trials and a range of musical tempi.

Fig 5. Time series of relative phase values from one trial of the current study. The 40 values within the black box are part of a synchronized bout. Relative phase (in degrees) is plotted on the vertical axis with zero indicating perfect alignment of footfalls and musical beats. Negative values indicate footfalls that occur before the nearest beat, positive values indicate footfalls that occur after the nearest beat, and values of +/-180 indicate footfalls that are midway between beats (i.e., in antiphase with the beat).