S’adapter aux autres : le rôle de la prédiction dans la communication humaine

PREDI-COM est un projet pluridisciplinaire en sciences humaines porté par le laboratoire SCALab de l’Université de Lille soutenu par la Maison Européenne des Sciences de l'Homme et de la Société et l’Université de Lille.

L’objectif scientifique de ce projet est de caractériser à quoi renvoie la notion de prédiction à travers différents champs disciplinaires en sciences humaines et sciences exactes (philosophie, linguistique, psychologie et neurosciences). En particulier, cette notion sera discutée dans le contexte de la communication humaine marquée par une variabilité forte d’indices fournis par les individus (sonores, linguistiques, non-verbaux, et sociaux) et le besoin de s’adapter entre les individus pour se comprendre. Dans ce but, deux laboratoires en sciences humaines de l’Université de Lille, l’un spécialisé en psychologie cognitive (Laboratoire Sciences Cognitives et Sciences Affectives, SCALab, UMR 9193) et l’autre spécialisé en linguistique et en philosophie (Laboratoire Savoirs, Textes, Langage, STL, UMR 8163) vont organiser trois demi-journées en anglais autour de la prédiction dans la communication humaine.

Première demi-journée

• Vendredi 14 février 2020

La première demi-journée aura pour objet de présenter la conception actuelle de la notion de prédiction en neurosciences et de la discuter entre philosophes, linguistes, psychologues et neuroscientifiques. Deux représentants les plus éminents en neurosciences présenteront cette approche (le Dr. Wolfram Schultz, de l’Université de Cambridge, et le Dr. Karl Friston, de l’University College London).

Wolfram Schultz, Department of Physiology, Development & Neuroscience, University of Cambridge, UK
Predictions and reward

Rewards induce learning (positive reinforcement), approach behaviour, economic decisions and positive emotions (pleasure, desire). We investigate basic neuronal reward signals during learning and decision-making, using behavioural and neurophysiological methods.

The phasic dopamine reward prediction error signal is composed of two components, resembling two-component responses in main sensory neurons. The early dopamine response component detects events indiscriminately and is influenced by physical impact, novelty, reward generalisation and reward context. The second component codes reward value. Although the first component detects punishers by their physical impact, none of the activations reflects the aversive nature of punishers. Dopamine activity shows also slower, heterogeneous and inconsistent changes related to widely ranging behavioural events not easily conceptualised by formal behavioural theories. A much slower, basically tonic dopamine neurotransmission is compromised in Parkinson’s disease and often subjected to pharmacological tests.

Our recent experimental economics studies reveal that monkeys are risk seeking with small rewards and risk neutral and then risk avoiders with larger rewards. Specifically structured behavioural tests allow us to estimate formal economic utility functions from certainty equivalents assessed during choices under risk (Von Neumann-Morgenstern utility). The animals' choices are meaningful in satisfying first, second and third order stochastic dominance, which defines rational choices governed by value, symmetric (variance) risk and skewness risk, respectively. Utility itself is coded in dopamine neurones as utility prediction error that incorporates risk into subjective value. Consistent with this neuronal signal, the dopamine response satisfies first- and second-order stochastic dominance. These data unite concepts from animal learning theory and economic decision theory at the level of single reward neurons.

Karl Friston, Institute of Neurology, University College London, UK
Active inference & deep prediction

This presentation considers deep temporal models in the brain. It builds on previous formulations of active inference to simulate behaviour and electrophysiological responses under deep (hierarchical) generative models of discrete state transitions. The structured temporal aspect of these models means that evidence is accumulated over time, enabling inferences about narratives and policies. We illustrate this behaviour in terms of Bayesian belief updating – and associated neuronal processes – to reproduce epistemic foraging for reward. These simulations reproduce these sort of delay period activity and local field potentials seen empirically; including evidence accumulation, place cell activity and transfer of dopamine responses. These simulations are presented as an example of how to use basic (first) principles to constrain our understanding of functional architectures in the brain.
https://webtv.univ-lille.fr/video/10789/karl-friston-active-inference-amp-deep-prediction ?
 

Seconde demi-journée

• Mardi 6 octobre 2020

La seconde demi-journée aura pour objet de présenter la notion de prédiction dans la communication humaine et ses liens avec le phénomène de convergence entre individus. Ces notions seront discutées entre philosophes, linguistes, et psychologues. Deux experts internationaux en psychologie et en linguistique présenteront ces notions (le Dr. Martin Pickering, de l’Université de l’Édimbourg, et le Dr. Antje Schweitzer, de l’Université de Stuttgart).

Martin Pickering, Université d'Edimbourg, UK
Prediction in Language

What do comprehenders predict and how do they use those predictions?  I first present a series of “visual world” experiments that suggest that comprehenders initially predict in an automatic, associative manner and then shift to predicting what the speaker or character under discussion is likely to say.  I argue that comprehenders predict by simulating the speaker, adjusting for differences between self and other, and then engaging aspects of the system that they use to produce language. 

Antje Schweitzer, Université de Stuttgart, DE
Phonetic convergence in conversational speech

In this talk, I will discuss results of phonetic convergence studies on spontaneous speech. In contrast to experiments in more controlled settings, studies that investigate less artificial scenarios seem to find much more subtle and more variable convergence effects, and they often depend on extra-linguistic aspects such as social factors. Moreover, there is also no established method of quantifying convergence in such scenarios, making comparison across studies difficult.

Troisième demi-journée

• Jeudi 26 novembre

La troisième demi-journée aura pour objet de présenter la notion de prédiction d'indices temporels et ses liens avec les comportements rythmiques dans la communication animale et humaine. Ces notions seront discutées entre philosophes, linguistes, et psychologues.

Deux experts internationaux en biologie évolutive et en neurosciences présenteront ces notions (le Dr. Andrea Ravignani, Max Planck Institute for Psycholinguistics et le Dr. Benjamin Morillon, Institut de Neurosciences des Systèmes).

Andrea Ravignani, Max Planck Institute for Psycholinguistics, NE
Predicting, reacting or entraining? Rhythm and timing from an evolutionary, comparative perspective

Capacities for rhythm, beat perception and synchronization are key in human music, speech and action. Just in the last decade, many thoughts that humans were quite unique in their rhythmic capacities. Since 2009, however, this theoretical landscape started changing: Research in a parrot (Snowball the cockatoo) and other animals provided evidence that other species can perform predictive and flexible rhythmic synchronization. In this talk, I will take a broad comparative approach, showing the previously underestimated richness of animal rhythms. In particular I will focus on my work in non-human primates and seals. These data suggest that rhythmic abilities in other mammals are more developed than previously surmised. I will discuss how evidence for rhythm in other species can inform the evolution of rhythmic capacities in our own.

Benjamin Morillon, Institut de Neurosciences des Systèmes, FR
Motor contribution to auditory temporal predictions

Temporal predictions are fundamental instruments for facilitating sensory selection, allowing humans to exploit regularities in the world. Recent evidence indicates that the motor system instantiates predictive timing mechanisms, helping to synchronize temporal fluctuations of attention with the timing of events in a task-relevant stream, thus facilitating sensory selection. Accordingly, in the auditory domain auditory-motor interactions are observed during perception of speech and music, two temporally structured sensory streams. I will present a behavioral and neurophysiological account for this theory and will detail the parameters governing the emergence of this auditory-motor coupling, through a set of behavioral and magnetoencephalography (MEG) experiments.

I will first review the prominence of delta (~2 Hz) oscillatory rhythms in the motor cortex and show that they constraint the interaction between motor and auditory systems. At this rate –and this rate only– overt rhythmic movements sharpen the temporal selection of auditory stimuli, thereby improving performance. I will next show that the implication of the motor system during auditory perception depends also on the temporal predictability of the sensory stream. Behaviorally, the feeling of groove induced by a melody –i.e. the wanting to move during passive listening– strongly depends on its temporal predictability. MEG results reveal that auditory and motor regions have a distinctive sensitivity to auditory temporal dynamics, with motor areas being more flexible in their ability to track temporal information. Together, these findings are compatible with active sensing theories, emphasizing the prominent role of motor areas in sensory processing, which are notably implicated in the analysis of contextual temporal information.