Hans Op de Beeck

Animal bias refers to the human tendency to prioritise the neural processing of animate over inanimate stimuli, a phenomenon observed across multiple cognitive domains, such as attention, perception, memory, and language. Lookalikes (also referred to as zoomorphic objects), inanimate objects that visually resemble animals, tend to be interpreted as animate regardless of their inanimate identity. We examined whether this animal bias can be modified through learning to understand how experience and plasticity shape visual object representations. We tested whether feedback-based categorisation training in which lookalikes had to be categorised as inanimate could reduce the animal bias. Thirty healthy participants completed three weekly sessions of animacy categorisation training, followed by an EEG recording session during which they performed an animacy task. Repeated-measures ANOVA on behavioural data revealed that categorisation efficiency, measured as the inverse efficiency score, improved both within and across sessions, but there was no evidence that this improvement was specific to lookalikes. Time-resolved multivariate pattern analysis and representational similarity analysis on EEG data obtained after this training showed a descriptively smaller animal bias relative to the no-training dataset from \citet{Chen2023}, but this reduction was not statistically significant in any analysis. Moreover, a significant animal bias was still present in participants who underwent training. Thus, neural reorganisation related to the animal bias is limited even after several training sessions to overcome this bias.

What transforms a novice into an expert? Decades of research show that expertise relies on domain-specific knowledge, but a neural account of this transformation has remained fragmentary: we lack an understanding of what information expert representations encode, how they are structured for efficient use, and where in the brain they reside. Using chess as a model system, we combine neuroimaging with multivariate pattern analysis to reveal three principles of the expert brain. Expertise drives a shift in representational content, from surface visual features to high-level, relational information. It is accompanied by a structural change to low-dimensional, optimised representation: codes become more compact and better organised, yet retain the detail needed for precise evaluation. Finally, the representational load shifts from sensory-specific cortices to domain-general frontoparietal networks. The expert brain packs more into less, concentrating richer knowledge into fewer, better-organised representations that support the rapid, flexible decisions of mastery.

The human visual system, and in particular the Visual Word Form Area (VWFA), adapts to process letters and words, even when the stimuli do not share canonical script features, like Braille. Here we set-up to compare the organization of typical orthographic and peculiar visual scripts such as Braille in computational models. In a first experiment, we looked at how Braille letters are represented in an illiterate Convolutional Neural Network (AlexNet) and compared them to Latin alphabet and to Line Braille, a custom line-based script. We observed a predisposition of the network, pre-trained to perform object recognition, for line-based scripts. This finding suggests an initial advantage of line junctions over Braille in processing scripts likely based on typical visual computations applied to the visual world. In a second experiment, we trained two benchmark neural network architectures (AlexNet, CORnet Z) to classify words in the Latin script (literacy acquisition) and then in the Braille script (expertise acquisition). We modelled the processing of reading visual Braille and explored the networks representations at different layers. We observed clustering of features based on the visual properties of the scripts and not by the network’s expertise. Unlike human participants, the representations of linguistic categories do not converge to a model of the linguistic (orthographic, phonological, semantic) properties. Overall, the lack of alignment between the visual processing of the trained computational models and neural data recorded in expert humans suggests that the fundamental processing of reading cannot be fully explained by simple feed-forward visual processing of the script, but likely relies on additional mechanisms including interactive relations between the visual and linguistic systems.

Most written systems share basic shape features with natural objects such as line junctions, a commonality thought to be at the basis of fluent reading. Studies that compared reading acquisition for different scripts used line-based scripts and investigated them in terms of complexity, novelty, or associations of stimuli from different categories. Here, we directly compared visual Braille, a script that only includes patterns of dots and no line junctions, to Line Braille, a novel, custom-made script based on line junctions drawn between Braille dots. For four consecutive days, two groups of participants (each N = 40) underwent online training of either one of the scripts, during which they first mapped new letters onto the Latin alphabet and then trained on full words. Each day, participants were tested by asking them to transcribe a set of stimuli (words and pseudowords) from the novel to the Latin script. Across sessions, we found no significant differences between scripts in the overall transcription accuracy nor in the time required to transcribe words. We only found a small delay in the learning trajectory of the Braille group represented through an interaction between group and session in overall accuracy, and slightly higher sensitivity for stimulus length in the Braille group. Overall, these results show that line junctions only provide a small and temporary benefit when learning a new script, contrasting with the idea that line junctions is a core visual features for learning orthography.

One of the distinctive features of the human visual system is the presence in occipito-temporal cortex (OTC) of regions that show preferential activation to specific categories of visual objects. To understand how this selectivity relates to categorisation behaviour, studies have employed a distance-to-bound approach (DTB), where multivariate brain activity is used to estimate a decision boundary, from which behavioural performance can be predicted. Using this approach, correlations have been found between activity in OTC, and behavioural performance when carrying out certain categorisation tasks. However, it remains unclear what determines where in OTC this correlations can be found, and with which categorisation tasks they can be found. Here, we bridged this gap by relating category-selective regions of OTC, to behavioural performance while participants categorised images as belonging or not to their preferred categories. We adopted a more basic approach and considered simple, univariate activity, rather than relying on decoding to build our DTB. Our results show that activation in regions selective to faces (FFA & OFA), bodies (EBA), and scenes (PPA), is sufficient to predict behavioural performance while categorising images as being faces, bodies, or scenes, respectively. These results are largely consistent across reaction time and motor movements, and generalise to animacy classification. Overall, our data adds to evidence that category-selective regions in OTC can serve to guide categorisation behaviour, and underlines the validity of the DTB approach to address this relationship.

High-level visual cortex contains category-selective areas embedded within larger-scale topographic maps like animacy and real-world size. Here, we propose action as a key organizing factor shaping visual cortex topography and assess the ability of topographic deep artificial neural networks (DANNs) in capturing this organization. Using fMRI, we examined responses to images of body-parts and objects with different degrees of action properties. In left lateral occipitotemporal cortex, we identified a topographically-organized action gradient, with overlapping activations for bodies, hands, tools, and manipulable objects along a dorsal-posterior to ventral-anterior axis, culminating at the intersection of body parts and objects exhibiting higher action properties. Multivariate analyses confirmed action as a crucial organizing principle, while shape and animacy dominated ventral occipitotemporal cortex and DANNs, which exhibited no action-based organization. Our proposed action dimension serves as a further organizing principle of object categories, advancing understanding of visual cortex organization and its divergence from DANN-based models.

How does sensory experience shape the development of the visual brain? To answer this eluding question, we examine brain responses to visual categories in a rare group of cataract-reversal individuals who experienced a short transient period of early blindness. Encoding of low-level visual properties is impaired in the early visual cortex (EVC) of cataract-reversal participants, whereas categorical responses in downstream ventral occipito-temporal cortex (VOTC) are preserved. In controls, degrading visual input to mimic the visual deficits of cataracts produces cascading disruptions extending from EVC to VOTC, unlike in the cataract group. A deep neural network trained on altered visual input reproduces this dissociation, supporting the brain findings. These results demonstrate that while EVC is permanently affected by early deprivation, categorical coding in VOTC shows resilience, highlighting different sensitive periods for specific brain regions and computations.

What transforms a novice into an expert? Decades of research show that expertise relies on domain-specific knowledge, but a neural account of this transformation has remained fragmentary: we lack an understanding of what information expert representations encode, how they are structured for efficient use, and where in the brain they reside. Using chess as a model system, we combine neuroimaging with multivariate pattern analysis to reveal three principles of the expert brain. Expertise drives a shift in representational content, from surface visual features to high-level, relational information. It is accompanied by a structural change to low-dimensional, optimised representation: codes become more compact and better organised, yet retain the detail needed for precise evaluation. Finally, the representational load shifts from sensory-specific cortices to domain-general frontoparietal networks. The expert brain packs more into less, concentrating richer knowledge into fewer, better-organised representations that support the rapid, flexible decisions of mastery.