Voici les éléments 1 - 10 sur 205
  • Publication
    Accès libre
    From Forest to Zoo: Domain Adaptation in Animal Behavior Recognition for Great Apes with ChimpBehave
    This paper addresses the significant challenge of recognizing behaviors in non-human primates, specifically focusing on chimpanzees. Automated behavior recognition is crucial for both conservation efforts and the advancement of behavioral research. However, it is significantly hindered by the labor-intensive process of manual video annotation. Despite the availability of large-scale animal behavior datasets, the effective application of machine learning models across varied environmental settings poses a critical challenge, primarily due to the variability in data collection contexts and the specificity of annotations. In this paper, we introduce ChimpBehave, a novel dataset featuring over 2 hours of video (approximately 193,000 video frames) of zoo-housed chimpanzees, meticulously annotated with bounding boxes and behavior labels for action recognition. ChimpBehave uniquely aligns its behavior classes with existing datasets, allowing for the study of domain adaptation and cross-dataset generalization methods between different visual settings. Furthermore, we benchmark our dataset using a state-of- theart CNN-based action recognition model, providing the first baseline results for both within and cross-dataset settings. The dataset, models, and code can be accessed at: https://github.com/MitchFuchs/ChimpBehave
  • Publication
    Accès libre
    ASBAR: an Animal Skeleton-Based Action Recognition framework. Recognizing great ape behaviors in the wild using pose estimation with domain adaptation
    To date, the investigation and classification of animal behaviors have mostly relied on direct human observations or video recordings with posthoc analysis, which can be labor-intensive, time-consuming, and prone to human bias. Recent advances in machine learning for computer vision tasks, such as pose estimation and action recognition, thus have the potential to significantly improve and deepen our understanding of animal behavior. However, despite the increased availability of open-source toolboxes and large-scale datasets for animal pose estimation, their practical relevance for behavior recognition remains under-explored. In this paper, we propose an innovative framework, To date, the investigation and classification of animal behaviors have mostly relied on direct human observations or video recordings with posthoc analysis, which can be labor-intensive, time-consuming, and prone to human bias. Recent advances in machine learning for computer vision tasks, such as pose estimation and action recognition, thus have the potential to significantly improve and deepen our understanding of animal behavior. However, despite the increased availability of open-source toolboxes and large-scale datasets for animal pose estimation, their practical relevance for behavior recognition remains under-explored. In this paper, we propose an innovative framework, ASBAR, for Animal Skeleton-Based Action Recognition, which fully integrates animal pose estimation and behavior recognition. We demonstrate the use of this framework in a particularly challenging task: the classification of great ape natural behaviors in the wild. First, we built a robust pose estimator model leveraging OpenMonkeyChallenge, one of the largest available open-source primate pose datasets, through a benchmark analysis on several CNN models from DeepLabCut, integrated into our framework. Second, we extracted the great ape’s skeletal motion from the PanAf dataset, a large collection of in-the-wild videos of gorillas and chimpanzees annotated for natural behaviors, which we used to train and evaluate PoseConv3D from MMaction2, a second deep learning model fully integrated into our framework. We hereby classify behaviors into nine distinct categories and achieve a Top 1 accuracy of 74.98%, comparable to previous studies using video-based methods, while reducing the model’s input size by a factor of around 20. Additionally, we provide an open-source terminal-based GUI that integrates our full pipeline and release a set of 5,440 keypoint annotations to facilitate the replication of our results on other species and/or behaviors. All models, code, and data can be accessed at: https://github.com/MitchFuchs/asbar.
  • Publication
    Restriction temporaire
    The evolutionary origins of event cognition in hominids
    (Neuchâtel : Université de Neuchâtel, 2024) ;
    How did syntax evolve in humans? Is there a direct transition from the simple forms observed in animal communication to the complex sentence structures seen in human languages? This thesis tested a new theory positing that human syntax is rooted in prelinguistic event cognition, an evolutionarily ancient propensity to decompose events into their causal components. Humans spontaneously attribute agents (doers of actions) and patients (undergoers) roles while viewing their interactions, i.e., “who does what to whom”, with a particular focus on agents. The current literature suggests that humans recognise agents faster than patients and dedicate more attention to them. Similarly in language, agents are typically placed before patients, are marked by simpler forms, and play a privileged role in language processing. But, how do animals understand the causal structure of events, how do they keep track of event roles, and in doing so, do they also experience a preference for the agent? This thesis aims to shed light on the evolution of event decomposition, as a putative precursor to human language and syntax, by comparing hominid species. I first explored whether great apes, like humans, spontaneously encoded event roles. Using a switch cost paradigm (i.e., a reaction time experiment), I showed that switching the role (i.e., agent or patient) of the assigned target significantly increased the processing time of the participants. I then concluded that the propensity to spontaneously encode social events in terms of agents and patients is shared amongst hominids, and that it is impossible or very difficult to override this deeply anchored cognitive propensity. Secondly, I examined whether great apes also showed indications of agent preference. Using a non-linguistic task based on video clips that mimicked decision making during language processing, I found that all hominids had a robust preference for animate agents, especially when acting on objects. This preference transcended simple perceptual biases, again suggesting a shared propensity. Building on these results I then tested for potential modulation of the agent preference by differences in cooperation levels. Results demonstrated that in both chimpanzees and humans, agent preference increased along the perceived cooperativity of the agent. Such results suggested that agent preference is affected by social factors, and similarly so in animals and humans. Finally, I focused on the computational abilities of hominids’ event perception. After watching video clips, participants selected the components of the events in serial order: the II agent, patient and action. Surprisingly, and in contrast to standard linguistic theory, all species, including humans, preferentially selected the action first, followed by the agent. The second preferred order was agent-action. I concluded that identifying the action was crucial for constructing the event structure, after which the agent preference became prominent. Overall, these findings provide evidence that the ability to decompose events in terms of agent, patient and action evolved long before modern humans and, by extension, language. It implies that syntax-ready brains already existed for millions of years, dating back to at least the last common ancestor of hominids. What still remains unclear, however, is why the ability to externalise these structures with communication then evolved in humans and, correspondingly, why great apes, despite shared perceptual abilities, are prevented from sharing them with others.
  • Publication
    Accès libre
    Call combinations and compositional processing in wild chimpanzees
    (2023)
    Maël Leroux
    ;
    Anne M. Schel
    ;
    Claudia Wilke
    ;
    Bosco Chandia
    ;
    ;
    Katie E. Slocombe
    ;
    Simon W. Townsend
    Abstract : Through syntax, i.e., the combination of words into larger phrases, language can express a limitless number of messages. Data in great apes, our closest-living relatives, are central to the reconstruction of syntax’s phylogenetic origins, yet are currently lacking. Here, we provide evidence for syntactic-like structuring in chimpanzee communication. Chimpanzees produce “alarm-huus” when surprised and “waa-barks” when potentially recruiting conspecifics during aggression or hunting. Anecdotal data suggested chimpanzees combine these calls specifically when encountering snakes. Using snake presentations, we confirm call combinations are produced when individuals encounter snakes and find that more individuals join the caller after hearing the combination. To test the meaning-bearing nature of the call combination, we use playbacks of artificially-constructed call combinations and both independent calls. Chimpanzees react most strongly to call combinations, showing longer looking responses, compared with both independent calls. We propose the “alarm-huu + waa-bark” represents a compositional syntactic-like structure, where the meaning of the call combination is derived from the meaning of its parts. Our work suggests that compositional structures may not have evolved de novo in the human lineage, but that the cognitive building-blocks facilitating syntax may have been present in our last common ancestor with chimpanzees.
  • Publication
    Accès libre
    Social learning in wild sooty mangabey vocal communication
    Élucider l'origine du langage est considéré comme l'un des problèmes les plus difficiles de la science. Il existe un soutien considérable à l'idée que les composants de base de la cognition humaine, y compris la faculté de langage, ne sont pas apparus de novo, mais ont des racines profondes dans la lignée des primates. Bien que l'on en sache beaucoup sur la façon dont les humains acquièrent le langage, et même sur l’acquisition du chant des oiseaux, une question restant largement ouverte est de savoir comment les primates non humains acquièrent leur compétence communicative. La recherche comparative s’avère être d’une grande aide pour palier à ce manque de connaissances sur la compréhension de la profondeur de ces racines évolutives. Ici, j'ai mené une série d'expériences sur le terrain avec des mangabeys fuligineux en liberté, une espèce de primate présente dans le parc national de Taï, en Côte d'Ivoire, pour éclairer la façon dont les primates non humains acquièrent leur faculté de communication. Tout d'abord, j'ai combiné des observations comportementales directes avec des expériences de lecture pour étudier le développement de la reconnaissance des cris d'alarme con- et hétérospécifiques chez les mangabeys fuligineux. J'ai constaté que la compétence communicative était acquise au stade juvénile, avec un apprentissage de la compréhension des appels d'alarme précédant l'utilisation vocale appropriée et sans différence claire dans l'apprentissage des signaux con- et hétérospécifiques. J'ai également constaté que, durant les premières étapes de la vie, la référence sociale, une forme proactive d'apprentissage social, était essentielle à l'acquisition d'un comportement d'appel d'alarme approprié. Dans une deuxième expérience, j'ai fourni l'un des rares tests systématiques, directs et empiriques dans des conditions naturelles du modèle de développement influent de Seyfarth et Cheney : l'idée que les primates apprennent à communiquer par processus d'élagage cognitif. En utilisant des modèles de vipères dangereuses et de serpents colubridés non venimeux, j'ai montré que les jeunes juvéniles percevaient tous les serpents comme dangereux, étaient fréquemment engagés dans des références sociales et lançaient des appels d'alarme sans discernement. Par opposition, les adultes répondaient plus rapidement aux vipères que les colubridés et ne faisaient jamais d'appel d'alarme à ces derniers, contrairement aux juvéniles. Ces résultats démontrent empiriquement que l'apprentissage de la prédation de serpents chez les mangabeys fuligineux commence par une surgénéralisation avant d’être suivie d'un raffinement ultérieur par apprentissage social. Dans une troisième expérience, j'ai évalué la capacité des mangabeys fuligineux à apprendre socialement des connaissances sur les prédateurs à partir de cris d'alarme. Par l’utilisation de présentations individuelles, j'ai exposé des sujets à un nouveau modèle de prédateur chimérique avec des caractéristiques visuelles de serpent et de léopard en conjonction avec des lectures d'alarmes de léopard et de serpent de congénères. J'ai trouvé que les schémas de réponse des sujets au modèle prédateur correspondaient à la signification des appels d'alarme qu'ils avaient entendus lors de la première rencontre. Cela suggère qu'ils classaient le modèle comme appartenant à la catégorie serpent ou léopard. Près de deux ans plus tard, j'ai exposé un tiers des mêmes individus au même modèle de prédateur pour une deuxième évaluation et j'ai constaté que les sujets continuaient à montrer la même réponse anti-prédateur, démontrant une mémoire sémantique à long terme à partir d'une seule expérience. En conclusion, les primates non humains apprennent à communiquer par un processus d'élagage cognitif qui affine le comportement anti-prédateur et d'appel d'alarme. L'apprentissage vocal des primates est régi par un développement cognitif sous-jacent, qui s'écarte des réponses innées et trop généralisatrices aux classes animales de base, suivi d'un raffinement ultérieur entraîné par l'apprentissage social. Ce processus d'apprentissage social commence au cours des premiers stades de vie et s'affine à mesure que les animaux grandissent, et façonne également l'interprétation des appels d'alarme de leur propre et des autres espèces. Dans l'ensemble, ce travail a fait avancer un problème scientifique important, à savoir la façon dont les primates acquièrent la capacité d'apprendre socialement à communiquer, en observant et en interagissant avec les autres, une composante essentielle de la faculté de langage humain. ABSTRACT Elucidating the origin of language is considered one of the hardest problems in science. There is considerable support for the idea that the basic components of human cognition, including the language faculty, did not emerge de novo, but have deep roots in the primate lineage. Although much is known about how humans acquire language, and even how birds acquire song, a largely open question is how non-human primates acquire their communicative competence. Comparative research can help us this gap of knowledge undermines our understanding of how deep these evolutionary roots are. Here, I conducted a set of field experiments with free-ranging sooty mangabeys, a primate species occurring in Taï National Park, Côte d'Ivoire, to shed light on how non-human primates acquire their communicative capacities. First, I combined direct behavioural observations with playback experiments to study the development of con- and heterospecific alarm calls recognition in sooty mangabeys. I found that communicative competence was acquired during the juvenile stage, with alarm call comprehension learning preceding appropriate vocal usage and with no clear difference in learning of con- and heterospecific signals. I also found that, during early stages of life, social referencing, a proactive form of social learning, was key in the acquisition of competent alarm call behaviour. In a second experiment, I provided one of the few systematic, direct, empirical tests under natural conditions of Seyfarth & Cheney’s influential developmental model –the idea that primates learn to communicate by undergoing a cognitive pruning process. Using models of dangerous vipers and not-venomous colubrid snakes, I showed that young juveniles perceived all snakes as dangerous, frequently engaged in social referencing and indiscriminately alarm called, whereas adults responded faster to vipers than colubrids but never alarm called to the latter, unlike juveniles. These results empirically demonstrate that snake predator learning in sooty mangabeys starts with over-generalisation followed by subsequent refinement via social learning. In a third experiment, I assessed the capacity of sooty mangabeys to socially learn predator knowledge from alarm calls. During individual presentations, I exposed subjects to a novel, chimeric predator model with both snake- and leopard-like visual features in conjunction with playbacks of conspecifics’ leopard and snake alarms. I found that the subjects’ response patterns to the predator model corresponded to the meaning of the alarm calls they heard during the first encounter, suggesting that they classed the model as either belonging to the snake or leopard category. Nearly 2 years later, I exposed a third of the same individuals to the same predator model for a second assessment and found that subjects continued to show the same anti-predator response, demonstrating long-term semantic memory from one single experience. In conclusion, non-human primates learn to communicate by a cognitive pruning process that fine-tunes anti-predator and alarm call behaviour. Primate vocal learning is governed by an underlying cognitive development, which departs from innate, over-generalising responses to basic animal classes, followed by subsequent refinement driven by social learning. This social learning process starts during their early stages of life and is refined as the animals mature, and equally shapes the interpretation of alarm calls from their own and other species. Overall, this work made progress with an important scientific problem, that is, how primates acquire the ability to socially learn to communicate, by observing and interacting with others, a core component of the human language faculty.
  • Publication
    Accès libre
    Juvenile vervet monkeys rely on others when responding to danger
    AbstractPrimate alarm calls are mainly hardwired but individuals need to adapt their calling behaviours according to the situation. Such learning necessitates recognising locally relevant dangers and may take place via their own experience or by observing others. To investigate monkeys alarm calling behaviour, we carried out a field experiment in which we exposed juvenile vervet monkeys to unfamiliar raptor models in the presence of audiences that differed in experience and reliability. We used audience age as a proxy for experience and relatedness as a proxy for reliability, while quantifying audience reactions to the models. We found a negative correlation between alarm call production and callers’ age. Adults never alarm called, compared to juveniles. We found no overall effect of audience composition and size, with juveniles calling more when with siblings than mothers or unrelated individuals. Finally, concerning audience reactions to the models, we observed juveniles remained silent with vigilant mothers and only alarm called with ignoring mothers, whereas we observed the opposite for siblings: juveniles remained silent with ignoring siblings and called with vigilant siblings. Despite the small sample size, juvenile vervet monkeys, confronted with unfamiliar and potentially dangerous raptors, seem to rely on others to decide whether to alarm call, demonstrating that the choice of the model may play an important key role in the ontogeny of primate alarm call behaviour.
  • Publication
    Accès libre
    On potential cooperation in predator-prey interactions in fishes
    (Neuchâtel, 2023) ; ; ;
    Barbara König
    L'interaction prédateur-proie fournit le contexte de certains des cas de coopération les mieux étudiés. Certaines espèces de prédateurs peuvent chasser ensemble et coordonner leurs mouvements dans des rôles actifs et divers pour augmenter les taux de capture de leurs proies ; c'est ce qu'on appelle la chasse coopérative. Les poissons-lions sont des piscivores communs dans l'Indo-Pacifique et envahissants dans les Caraïbes. Étant donné que les poissons-lions chassent seuls par nature, une étude d'un ancien chercheur qui a démontré un recrutement actif, une coordination et une frappe alternée (peut-être réciproque) chez le poisson-lion nain Dendrochirus zebra a suscité beaucoup d'intérêt. On a vu que le poisson-lion zèbre utilise un motif d'évasement des nageoires qui implique une ondulation de la nageoire caudale et des évasements successifs des deux nageoires pectorales pour indiquer le début de la chasse coopérative. Les résultats suggèrent que la capacité de chasser en coopération peut avoir contribué au succès d'une espèce sœur de poisson-lion, Pterois miles et P. volitans, à envahir les Caraïbes. Ici, j'ai étudié Pterois miles - l'une des espèces envahissantes - dans son aire de répartition naturelle en mer Rouge. Sur le terrain, je n'ai trouvé aucun signe de chasse coordonnée. J'ai complété les observations de terrain par une expérience en laboratoire, dans laquelle j'ai exposé des individus à un éventuel partenaire de chasse et à des proies inaccessibles dans un logement transparent. J'ai observé le schéma d'évasement des nageoires, mais il est essentiel de noter que le partenaire n'était pas la cible de ce signal d'évasement des nageoires. Les découvertes sur le terrain selon lesquelles cette espèce de la mer Rouge ne dépend pas de la chasse coopérative pour attraper du poisson sont également appuyées par le fait que les deux poissons-lions ne se sont pas rassemblés dans les zones de proies. J'ai en outre étayé ces résultats en examinant les mouvements coordonnés et l'alternance des frappes pendant la chasse. J'ai exposé des sujets d'appariements de P. miles à des proies inaccessibles dans trois logements clairs. En présence de proies, les deux poissons-lions ne se sont pas rassemblés dans la même maison de proies dans l'espace ou dans le temps. Dans une deuxième expérience, j'ai mis des morceaux de nourriture sur un bâton "d'arbre à nourrir" pour tester l'alternance réciproque des frappes. J'ai généralement vu moins d'alternances que prévu par hasard, et j'ai découvert que les alternances peuvent être augmentées en mettant des contraintes sur la monopolisation individuelle de la nourriture. En conclusion, le modèle de mouvement d'évasement des nageoires observé chez l'espèce sœur de la mer Rouge, P. miles, qui était auparavant considéré comme un signal, était maintenant interprété comme un mode de nage. De plus, les paires de milles P. dans la mer Rouge n'alternaient pas réciproquement leurs frappes. J'ai interprété le résultat de l'étude précédente de D. zebra qu'une certaine alternance pourrait être générée si les proies devenaient alternativement disponibles à deux coins dans un espace confiné, chaque poisson-lion préférant monopoliser un coin chacun. Finalement, en raison de certains défis rencontrés dans l'étude originale, qui m'ont empêché de mener des recherches plus empiriques, j'ai complété les chapitres empiriques de la thèse de doctorat avec une revue de recherche sur les perspectives de l'inspection des prédateurs chez les poissons. Le sujet a été choisi parce que l'inspection des prédateurs est un autre exemple classique de coopération. Cet examen a examiné les objectifs et les fonctions possibles des poissons proies inspectant les prédateurs potentiels et la manière dont les proies les approchent en fonction des objectifs d'inspection. Nous avons discuté de l'évaluation des motivations des proies et des indicateurs qui sont passés de la présence de prédateurs et des indicateurs d'état qui pourraient montrer pourquoi une attaque a été faite. Nous avons également discuté des types de jeux impliqués dans le comportement d'inspection des prédateurs. L'examen a identifié plusieurs lacunes importantes dans nos connaissances qui empêchent actuellement une évaluation appropriée des jeux de coopération qui pourraient s'appliquer.
    Abstract Predator – prey interaction provide the context for some of the best-studied cases of cooperation. Some predator species can hunt together and coordinate their moves within active and diverse roles to increase capture rates of their prey; this is known as cooperative hunting. Lionfish are common piscivores in the Indo-Pacific and invasive in the Caribbean. Since lionfishes hunt alone by nature, a study by a former researcher that demonstrated active recruitment, coordination, and alternated (perhaps reciprocal) striking in the dwarf lionfish Dendrochirus zebra has attracted much interest. Zebra lionfish have been seen to use a fin-flaring pattern that involves undulation of the caudal fin and successive flares of both pectoral fins to indicate the start of cooperative hunting. The findings suggested that the ability to hunt cooperatively may have contributed to the success of a sister lionfish species, Pterois miles and P. volitans, in invading the Caribbean. Here, I investigated Pterois miles - one of the invasive species - in its natural range in the Red Sea. In the field, I found no signs of coordinated hunting. I supplemented field observations with a laboratory experiment, in which I exposed individuals to a possible hunting partner and inaccessible prey in a transparent housing. I observed the fin-flaring pattern, but it is vital to note that the partner was not the target of this fin-flaring signal. Also supporting the field findings that this species in the Red Sea does not rely on cooperative hunting to catch fish is the result that the two lionfish did not congregate at the prey patches. I further supported these findings by examining coordinated movement and strike alternation during hunting. I exposed subjects of P. miles pairings to inaccessible prey in three clear housings. In the presence of prey, the two lionfish did not congregate at the same prey house in space or time. In a second experiment, I put food pieces on a "feeding tree" stick to test for reciprocal alternation of strikes. I generally saw fewer alternations than expected by chance, and found that alternations can be increased by putting constaints on individual monopolization of food. In conclusion, the fin flaring movement pattern seen in the Red Sea sister species P. miles that were previously thought to be a signal was now interpreted as a swimming mode. Additionally, pairs of P. miles in the Red Sea did not reciprocally alternate their strikes. I interpreted the result of the previous study of D. zebra that some alternation might be generated if prey items become alternately available at two corners in a confined space, with each lionfish preferring to monopolize one corner each. Eventually, due to some challenges encountered in the original study, which prevented me from conducting more empirical research, I complimented the empirical chapters of the PhD thesis with a research review on the perspectives of predator inspection in fishes. The topic was chosen because predator inspection is another classic example of cooperation. This review looked at the possible goals and functions of prey fishes inspecting potential predators and how prey approach them based on the inspection goals. We discussed the assessment of prey motivations and indicators that moved from predator presence and state indicators that could show why an attack was made. Also, we discussed what types of games are involved in predator inspection behaviour. The review identified several important gaps in our knowledge that currently prevent a proper assessment of what cooperation games might apply. In conclusion, the thesis demonstrates that it is an important scientific task to revisit apparently well-established examples and to challenge previous interpretations.
  • Publication
    Restriction temporaire
    The ontogeny of pant hoot vocalisations and social awareness in wild chimpanzees
    (Neuchâtel : Université de Neuchâtel, 2023) ;
    While some scholars have regarded primate vocal communication as innate, inflexible, and insensitive to the context, recent advances suggest instead that vocal behaviours can be flexible, insofar as they are affected by individual and situational factors, notably the social context. However, whether the same is true for the acquisition of communicative capacities remains largely unknown, particularly in great apes. In my thesis, I address this by examining the ontogeny of vocal behaviours in the pant hoots of immature chimpanzees (Pan troglodytes schweinfurthii) of the Budongo Forest, Uganda. Furthermore, I investigate audience effects on pant hoot sequences used during displays to determine the extent to which these vocal structures are flexibly modulated depending on the social environment. Pant hoots are a multi-phase vocal sequence typically used to maintain contact and coordinate movements between individuals and groups over long distances. The question of how this complex and flexible vocal signal develops is key for a better understanding of how chimpanzees navigate dynamic social interactions in fission-fusion societies from both an ontogenetic and a comparative perspective. Results from my thesis show that chimpanzees produced rudimentary pant hoot sequences since birth, suggesting that vocal repertoires are largely innate. However, these sequences presented some structural and acoustic differences when compared to those of older individuals, suggesting they also undergo ontogenetic changes. In addition, the vocal usage and responses to pant hoots in immature chimpanzees was enhanced by greater vocal and social exposure to key group members, such as the mother and adult males, when compared to the development of less gregarious immature individuals. Finally, social context modulated the use of pant hoot phases during vocal displays, likely enhancing the communicative capacities of a species with limited vocal production learning and relatively small vocal repertoire. Taken together, findings from my thesis suggest that the ontogeny of complex chimpanzee vocalisations is socially mediated and that chimpanzee vocal communication is flexibly adjusted depending on the social environment.