Romain Adrien Lefévre:
|
Biphonation, the occurrence of two independent fundamental frequencies termed 'f0' and 'g0' in animals' vocalisations, is believed to enhance social communication. In diverse species such as amphibians, cetaceans, terrestrial mammals, or birds, biphonic calls have been proposed to captivate listeners' attention, prevent habituation, provide orientation cues, and foster group cohesion. In addition, biphonic calls act as indicators of the signaller's motivational and emotional states, as supported by variations in acoustic parameters associated with changes in the autonomic and somatic nervous systems.
This PhD project explored the biomechanisms behind the production of horses' whinnies, the function of their composing independent frequencies and the potential for combinatoriality in their structure. The outcomes of this thesis sought to contribute to a better understanding of how animals can simultaneously transmit multiple and separate pieces of information, particularly emotional content. To this end, we proposed to address the following questions: 1) What mechanisms are selectively involved in producing f0 and g0 frequencies in the horse whinnies? 2) Are these vocalisations structured in a compositional manner? 3) What roles do these distinct components play in fulfilling social communication purposes? 4) Finally, a separate project investigated if positive and negative emotional information is consistently expressed across horses and other ungulates.
The structures involved in producing f0 and g0 frequencies in horses whinnies were investigated in Chapter 1, where we benefitted from a unique combination of in vivo and ex vivo methods, including computed tomography (CT) scans of excised larynges, endoscopic examinations of the live-functioning larynx, helium experiments through the larynx in isolation, and acoustic analysis of vocalisations of horses diagnosed with recurrent laryngeal neuropathy. The results suggested that g0 production is generated by an aerodynamic whistle mechanism and not through tissue vibration. We propose that the glottal opening, altered by the action of intrinsic laryngeal muscles and cartilages, creates a narrow, slit-like opening leading to rapid airflow increases that results in a whistle-like sound during the horse whinny. In addition, in subjects affected with atrophy of the recurrent laryngeal nerve, we found that the ability to narrow the glottic cleft is also reduced, causing airway obstruction resulting in fragmented f0. At the same time, the g0 frequency seems unaltered, suggesting different production sources.
The existence of compositional expressions in the horse whinnies was investigated in Chapter 2. In this experiment, we examined subjects' behavioural and physiological responses to natural nickers (which contain only f0 frequency), natural squeals (which contain only g0 frequency), natural whinnies (which contain both g0 and f0 frequencies), and artificially constructed whinnies made of a nicker followed by a squeal or inversely (reversed). Results showed that nickers perception aligned with its suggested communicative functions to promote group cohesion and attract others' attention. For their part, squeals may not have been perceived as threatening because they were played from a distance, which could have been irrelevant for these calls mainly produced at close range during agonistic physical interactions. More importantly, our findings reported that horses did not significantly differentiate between natural and artificial whinnies, suggesting that whinnies in the horse acoustic repertoire could have a composite meaning that would be derived from their individual nicker and squeal components. However, the absence of marked differences between responses to artificial compared with reversed artificial whinnies does not support the hypothesis of a compositional structure, as whinnies' reaction seems independent of the squeal and nicker components ordering. Instead, horses' whinnies could follow a holistic combinatorial rule, similar to the structural flexibility observed in human language, where the meaning is extracted regardless of the order of the composing elements.
The functional role of the two fundamental frequencies, f0 and g0, in the horse's biphonic calls was further explored in Chapter 3. In this study, we conducted playbacks of artificially created whinnies with combined and isolated f0 and g0 frequencies extracted from natural whinnies to investigate horses' behavioural and physiological responses. Contrary to expectations, no significant differences were observed in the horses' responses to natural and artificial calls. This observation indicates that horses may have an adaptive auditory processing mechanism robust enough to discern the crucial information content of a whinny, an ability that could be beneficial in noisy environments to preserve group cohesion. Additionally, results showed that subjects responded similarly to the f0 and g0 components of the whinnies when presented in isolation and to the artificial whinnies made of their combination. These findings suggest redundancy of the transmitted information and contradict previous discoveries that attributed distinct communicative functions to these independent components.
Finally, in a separate project (Chapter 4), we investigated the extent to which the emotional content of the vocalisations of horses and other domestic and wild ungulates highlighted in previous studies is universal and can hence be identified by a machine learning algorithm. In this experiment, we trained a machine learning algorithm (XGBoost) to identify common rules governing the expression of positive and negative emotions in animals, possibly uncovering universal acoustic indicators of emotional valence expression. The findings indicate that an algorithm can be trained to reach 91.15% accuracy, and that negative calls have lower amplitude and frequency modulation but higher peak frequency, harmonicity and harmonics energy than positive calls.
This thesis explored the unique phenomenon of biphonation in horses, which suggests a greater sophistication in this species' vocal communication than previously thought. First, it provides insights into the biomechanics involved in biphonic calls, which revealed unique production mechanisms of the higher frequency in this species. It could also lead to the development of non-invasive diagnostic methods for assessing respiratory diseases by monitoring alterations in the horses' vocal signals. Then, it suggests that horses demonstrate a robust communication system well adapted to overcome distortions in the transmitted signal due to noisy environments while raising questions about a potential combinatorial structure akin to some aspects of human language processing. Ultimately, by exploring machinelearning methods to assess animal emotional valence non-invasively, this thesis hints at supporting the development of non-invasive tools to better assess animal welfare based on vocal signal modulations.