Poster A62, Saturday, March 25, 5:00 – 7:00 pm, Pacific Concourse
A neurobiologically inspired computational model of sensorimotor grounding of abstract semantics
Malte R Schomers1,2, Friedemann Pulvermüller1,2; 1Brain Language Laboratory, Freie Universität Berlin, Germany, 2Berlin School of Mind and Brain, Humboldt-Universität zu Berlin, Germany
Explaining abstract semantics in terms of grounding in sensorimotor systems poses a particular challenge (Dove 2016, Psych.Bull.Rev). Concrete concepts can be grounded through high overlap of perceptuo-motor semantic features (e.g., the visual features of ‘eye’ or the motor features of ‘to grasp’ are highly similar across instances). But how can grounding occur for abstract concepts that have highly variable features across all instances in which the concept is used, exhibiting a family-resemblance relationship? We investigate this question using a neurobiologically well-informed computational model of the human brain areas involved in phonological and semantic word representations. We extended an existing model that demonstrated grounding of concrete action vs. object concepts (Garagnani & Pulvermüller 2016 Eur.J.Neurosci.; Tomasello et al., 2016, Neuropsychologia) to abstract semantics: The model was trained to associate phonological word form representations (neuronal patterns in acoustic-articulatory language cortex) with highly variable sets of sensorimotor semantic features (patterns in visual and motor cortices) exhibiting a family-resemblance relationship. Compared to implementations of ‘concrete’ semantics relying on conceptual-semantic feature overlap, the neuronal representations of ‘abstract’ words (1) shifted more strongly towards the perisylvian brain areas, (2) were less ‘cohesive’ (less likely to be active together), and (3) exhibited lower average firing rates during ‘concept retrieval’. Crucially, these results show that abstract meaning can be grounded in action and perception, while indicating a relatively greater reliance of abstract semantics on perisylvian cortical areas as well as processing advantages of concrete over abstract words. The results are discussed in light of the current embodied-cognition debate.
Topic Area: LANGUAGE: Semantic