Table ronde ILCB : Interpreting machine learning in hearing, communication and language sciences

ILCB Table ronde

Interpreting machine learning in hearing, communication and language sciences: why, how, and the current challenges

Programme :

12h/12h30 – Etienne Thoret (Post-doc ILCB, PRISM, LIS) – Deciphering the acoustical bases of hearing by interpreting biomimetic deep-neural-networks (20 min + 10 min)
12h30/13h – Philippe Blache (LPL) – Is language processing incremental? A comparison between Transformer and RNN-based language models and their ability to model human language processing.  (20 min + 10 min)
13h/13h30 –  Ronan Sicre (LIS) – Visual interpretability of deep neural networks: a brief overview.  (20 min + 10 min)
13h30/13h45 Adrià Torrens (University of Ostrava) Building a grammar for gradient linguistic evaluative expressions: Do Machine learning, neuronal networks, and deep learning help? (10 min + 5min)
13h45/14h30 – Discussion (45 minutes)

Plus d'informations : Events | ILCB


ILCB Lunchtalk : Virginia B. Penhune

What we learn and when we learn it: the interaction of maturation and experience in music and language

Virginia B Penhune
Department of Psychology, Concordia University
Laboratory for Motor Control and Neural Plasticity

The impact of training or experience is not the same at all points in development. Children who learn to play a musical instrument or speak a second language early in life are often more proficient as adults. In the domain of music, a wealth of anecdotal evidence suggests that early training is important for musical skill, however, there has been little evidence directly demonstrating the impact of the age of start. To address this question, work in my laboratory has compared behavior and brain structure in early- (<7) and late-trained ( >7) adult and child musicians, showing differences in behavior and brain structure. More recently, we have compared early- and late-trained musicians with simultaneous and sequential bilinguals, showing differential effects of age-of-start in the arcuate fasciculus. I will discuss these findings in the context of our understanding of the interaction between normative development and specific experience, and describe a model of gene-environment interactions that integrates the contribution of age of start.

Website :

ILCB Lunchtalk : Abdellah Fourtassi

Using Data Science to Study Children’s Cognitive Development

Abdellah Fourtassi

Following the seminal work of Piaget, the traditional approach in cognitive development has focused on studying the structure of children’s knowledge in controlled situations (e.g., laboratory experiments). While this approach allows for precise inference about how children behave in certain tasks, it cannot provide an understanding of the social context within which knowledge emerges. In fact, it has been known, at least since Vygotsky, that children acquire new skills and concepts with the help of more competent members of society who scaffold the children’s learning, allowing them to attain skills that are just beyond their current abilities. In fact, much of the children’s abstract knowledge about the world, it has been argued, is mediated through discussions with their parents/caregivers.

In this talk, I explain how new advances in Data Science, especially in Natural Language Processing (NLP), allow us to 1) account for what and how information is presented to children by their parents through language, and 2) make precise predictions about the way this information can be used by children in controlled designs. Thus, NLP can create a fruitful synergy between controlled and naturalistic research methods. More generally, I argue that a complete theory of cognitive development requires interdisciplinary research across computer science and psychology.

Website :

ILCB Lunchtalk : Robert Zatorre

Musicians at the cocktail party: Neural correlates of bottom-up and top down mechanisms

Robert Zatorre
Montreal Neurological Institute
McGill University

Segregating sound mixtures makes demands on multiple cognitive and neural mechanisms that musical training may enhance or exploit. In a series of studies we have documented the music-related enhancement behaviorally in the context of speech in noise, and also in a selective attention context with competing speech streams. Using functional MRI, we observed that musicians’ enhanced speech-in-noise perception was associated with better decoding of speech in auditory areas at high signal-to-noise ratios (SNR), whereas under low SNR conditions the enhancement was associated with decoding in frontal and motor cortical regions. We interpret this finding as indicating a shift from bottom-up to top-down mechanisms depending on the quality of the input, with musicians being better able to deploy either mechanism as a function of the conditions. We then used MEG to look at the neural representation of competing speech streams via decoding of the neural signature (amplitude envelope) of attended vs unattended items. The behavioral advantage associated with musical training was related to enhanced ability to represent both streams in auditory cortex, consistent with their capacity to follow multiple sound streams in music. These cognitive neuroscience approaches help us to develop better models to explain why musicians are good at cocktail parties (apart from their reputed drinking abilities).

Website :

ILCB Lunchtalk : Suzanne Dikker

ILCB Lunchtalk

Suzanne Dikker
New York University, Department of Psychology & Center for Neural Science

• 12.00 Suzanne Dikker, PhD
• 13.00 Lunch

Confirm attendance (mandatory) by sending an email to

Brains in Harmony: the role of brain-to-brain synchrony in naturalistic social interactions

Neuroscience research has produced tremendous insight into how the human brain supports dynamic social interactions. Still, laboratory-generated findings do not always straightforwardly generalize to real-world environments. To fill this gap, I collaborate with scientists, artists, and educators to take neuroscience out of the laboratory, into schools, museums, and underserved neighborhoods. We consistently find a relationship between brain-to-brain synchrony and successful social interaction. For example, empathy, joint action, and social motivation predicts synchrony in dyadic interactions, and synchrony among high schoolers is related to classroom social dynamics and student engagement. Taken together, our multidisciplinary approach may provide a potential new avenue to investigate social interactions outside of the laboratory.

ILCB Lunchtalk : Véronique Izard et Antje S. Meyer

ILCB Lunch Talk*, April 26, 2019
Station Marine d’Endoume – 13007 Marseille

Véronique Izard : Integrative Neuroscience and Cognition Center, CNRS & Université Paris Descartes
Antje S. Meyer : Max Planck Institute for Psycholinguistics, Nijmegen The Netherlands

11h Véronique Izard : In search for the cognitive foundations of Euclidean geometry
12h Antje S. Meyer : Towards processing theories of conversation
• 13h Lunch
• Confirm attendance (mandatory) by sending an email to

In search for the cognitive foundations of Euclidean geometry
Euclidean geometry has been historically regarded as the most « natural » geometry. Taking inspiration from the flourishing field of numerical cognition, in the past years I have been looking for the cognitive foundations of geometry: Do children, infants, and people without formal education in geometry have access to intuitive concepts that bear some of the content of Euclidean concepts? Results have been mixed. In particular, we found that angle, a central tenant of Euclidean geometry, is not intuitive for children. These results call into question the status of Euclidean geometry as a natural geometry.

Towards processing theories of conversation
Most experimental research into spoken language has focused either on speaking or on listening. However, these processes should also be studied together, not only because they naturally co-occur in conversation and likely affect each other, but also because an integrated research approach can lead to novel insights into the architecture of the cognitive system supporting language use. I will provide an overview of a research program on speaking and listening in dyadic contexts. The starting point is the model of turn-taking in conversation proposed by Levinson and Torreira (2015). Though based exclusively on observational data the model makes strong processing predictions. A key claim is that speakers begin to plan their utterances as early as possible during their interlocutor’s turn, in order to be prepared to respond quickly. Experimental evidence showed that speakers indeed begin to plan their utterances before the end of the preceding turn but, contrary to the prediction, not necessarily as early as possible. Rather than following a fixed rule (« plan as early as possible ») they appear to be quite flexible in their utterance planning. Current work aims at uncovering the factors that limit this flexibility. It appears that, in addition to social and pragmatic factors that define the speaker’s processing goals, capacity limitations arising in different components of the cognitive system play an important role. I will end by discussing how speakers might achieve smooth turn-taking without intensive linguistic dual-tasking.

Lien vers le site Web de l’ILCB

ILCB Lunchtalk : Understanding publication practices, across disciplines to improve the impact of your inter-disciplinary research

ILCB Lunchtalk

Jeudi 28 Mars de 11h à 15h30 en salle des voûtes (Saint-Charles)
Organisé par Xavier Alario et Elin Runnqvist, avec des interventions de Christopher Chambers (Cardiff University, Editor of the journal Cortex), Didier Torny (Mines ParisTech – Paris), et Marie Farge (Ecole Normale Supérieure, Paris).

• 11h00 – 11h10: Motivations for this seminar: Alario, Blache, Runnqvist
• 11h10 – 11h50: Didier Torny (20 min. talk, 20 min. debate)
• 11h50 – 12h30: Chris Chambers (20 min. talk, 20 min. debate)
• 12h30 – 13h30: Lunch on site
• 13h30 – 14h10: Marie Farge (20 min. talk, 20 min. debate)
• 14h10 – 14h50: Round table (20 min. talk, 20 min. debate)
• 14h50 – end: Coffee, discussions, etc.

• Confirm attendance (mandatory) by sending an email to

Understanding publication practices, (models and time-courses) across disciplines to improve the impact of your inter-disciplinary research.
An inter-disciplinary discussion at ILCB

In our modern science practices, it would not be surprising to hear that the three most important assets for a scientist are… publication, publication, publication!

But what exactly is a publication? The answer to this question could be very different across disciplines, and many of its significant aspects are evolving.

Across disciplines, the increment of knowledge is conceived and packaged in diverse formats (working papers, proceedings, monographies, articles,
etc.) which are attributed vastly diverse value.

Across disciplines, the relationship between the author, the reviewers and the publishers can be vastly different. The business models of scientific publication has experienced major innovations in recent times, and presumably more changes lie ahead.

Across disciplines, the course of conception, dissemination, and archiving of a scientific contribution are practiced quite differently.
The public discussion of findings can precede or follow publication. In some disciplines, the predictions preceding an empirical study can now be archived in advance as pre-registrations, which can be referred to later to clarify whether the authors observed exactly what they predicted or predicted exactly what they observed, thus strengthening the impact of the contributions.

The inter-disciplinary « Institut Convergences ILCB » (Institute for Language, Communication, and the Brain) organizes a scientific discussion about current publication practices across disciplines. The issues above and related topics will be discussed by 4 specialist speakers.

This scientific discussion will have two goals. First, to create common knowledge, across practitioners of different disciplines, of what their collaborators in other domains consider a scientific contribution.
Second, to reflect upon and hopefully improve the publication strategies of researchers.

ILCB Lunch-talk : Dezso Nemeth

ILCB Lunch-talk par Dezso Nemeth
(Hungarian Academy of Sciences and Eotvos Lorand University ; titulaire de la Chaire ILCB/IMéRA « Langage et cerveau »)

1er mars 2019, 12h
Salle de conférences B011, LPL
5 avenue Pasteur, Aix-en-Provence

Plus d’infos