General features of the model

We implemented a neurobiologically constrained model replicating cortical areas of fronto-temporo-occipital lobes and their connectivity to shed light on the mechanism underlying semantic processing grounded in action and perception. We created a neural architecture with 15,000 representative neurons for simulating activity in twelve cortical areas in the left language-dominant hemisphere (see Figure ​Figure1A).1A). These “areas” represented three levels of processing—primary, secondary, and higher-association cortex—in four modality-systems: (motor) frontal superior-lateral hand-motor, (articulatory) inferior face-motor, (auditory) superior-temporal and (visual) inferior-temporo-occipital system. Two of these, the auditory and articulatory systems (areas highlighted in blue and red, Figure ​Figure1A)1A) are in perisylvian language cortex and appear most relevant for language processing (Zatorre et al., 1996; Pulvermüller, 1999; Fadiga et al., 2002; Pulvermüller and Fadiga, 2010). The motor and visual system (yellow and green highlighted areas) are outside the perisylvian language cortex (called “extrasylvian” in the present work) and involved in processing visual object processing (Ungerleider and Haxby, 1994), and the execution of manual actions (Deiber et al., 1991; Lu et al., 1994; Dum and Strick, 2002, 2005).

(A) Structure and connectivity of 12 frontal, temporal and occipital cortical areas relevant for learning the meaning of words related to actions. Perisylvian cortex comprises an inferior-frontal articulatory-phonological system (red colors), including primary motor cortex (M1_i), premotor (PM_i) and inferior-prefrontal (PF_i), and a superior-temporal acoustic-phonological system (areas in blue), including auditory parabelt (PB), auditory belt (AB) and primary auditory cortex (A1). Extrasylvian areas comprise a lateral dorsal hand-motor system (yellow to brown), including lateral prefrontal (PF_L), premotor (PM_L) and primary motor cortex (M1_L), and a visual “what” stream of object processing (green), including anterior-temporal (AT), temporo-occipital (TO), and early visual areas (V1). When learning words in the context of perceived objects or to actions, both peri- and extrasylvian systems are involved. Numbers indicate Brodmann Areas (BAs) and the arrows (black, purple, and blue) represent long distance cortico-cortical connections as documented by neuroanatomical studies. (B) Schematic global area and connectivity structure of the implemented model. The colors indicate correspondence between cortical and model areas. (C) Micro-connectivity structure of one of the 7,500 single excitatory neural elements modeled (labeled “e”). Within-area excitatory links (in gray) to and from cell e are limited to a local (19 × 19) neighborhood of neural elements (light-gray area). Lateral inhibition between e and neighboring excitatory elements is realized as follows: the underlying cell i inhibits e in proportion to the total excitatory input it receives from the 5 × 5 neighborhood (dark-purple shaded area); by means of analogous connections (not depicted), e inhibits all of its neighbors. Adapted from (Garagnani and Pulvermüller, 2013).

The model replicates a range of important anatomical and physiological features of the human brain (e.g., Garagnani et al., 2008, 2017; Tomasello et al., 2017). As follow a summary of the six neurobiological principles incorporated in the neural network model:

Neurophysiological dynamics of spiking pyramidal cells including temporal summation of inputs, threshold-based spiking, nonlinear transformation of membrane potentials into neuronal outputs, and adaptation (Connors et al., 1982; Matthews, 2001);

Synaptic modification by way of Hebbian-type learning, including the two biological mechanisms of long-term potentiation (LTP) and long-term depression (LTD) (Artola and Singer, 1993);

Area-specific global regulation mechanisms and local lateral inhibition (global and local inhibition) (Braitenberg, 1978; Yuille and Geiger, 2003);

Within-area connectivity: a sparse, random and initially weak connectivity was implemented locally, along with a neighborhood bias toward close-by links (Kaas, 1997; Braitenberg and Schüz, 1998);

Between-area connectivity based on neurophysiological principles and motivated by neuroanatomical evidence; and

Uncorrelated white noise was constant present in all neurons during all stages of learning and retrieval with additional noise added to the stimulus patterns to mimic uncorrelated input conditions (Rolls and Deco, 2010).

Note that the connectivity structure implemented in the network reflects existing anatomical pathways between corresponding cortical areas of the cortex revealed by neuroanatomical studies using diffusion tensor and diffusion-weighted imaging (DTI/DWI) in humans and non-human primates (Table ​(Table2)2) (Rilling et al., 2011; Thiebaut de Schotten et al., 2012). A detailed description of the single-neuron properties, synaptic plasticity rule, and single-area model structure is provided next, followed by details of the network anatomy and connectivity structure.

Connectivity structure of the modeled cortical areas.