Abstract

One of the core human abilities is that of interpreting symbols. Prompted with a perceptual stimulus devoid of any intrinsic meaning, such as a written word, our brain can access a complex multidimensional representation, called semantic representation, which corresponds to its meaning. Notwithstanding decades of neuropsychological and neuroimaging work on the cognitive and neural substrate of semantic representations, many questions are left unanswered. The research in this dissertation attempts to unravel one of them: are the neural substrates of different components of concrete word meaning dissociated? In the first part, I review the different theoretical positions and empirical findings on the cognitive and neural correlates of semantic representations. I highlight how recent methodological advances, namely the introduction of multivariate methods for the analysis of distributed patterns of brain activity, broaden the set of hypotheses that can be empirically tested. In particular, they allow the exploration of the representational geometries of different brain areas, which is instrumental to the understanding of where and when the various dimensions of the semantic space are activated in the brain. Crucially, I propose an operational distinction between motor-perceptual dimensions (i.e., those attributes of the objects referred to by the words that are perceived through the senses) and conceptual ones (i.e., the information that is built via a complex integration of multiple perceptual features). In the second part, I present the results of the studies I conducted in order to investigate the automaticity of retrieval, topographical organization, and temporal dynamics of motor-perceptual and conceptual dimensions of word meaning. First, I show how the representational spaces retrieved with different behavioral and corpora-based methods (i.e., Semantic Distance Judgment, Semantic Feature Listing, WordNet) appear to be highly correlated and overall consistent within and across subjects. Second, I present the results of four priming experiments suggesting that perceptual dimensions of word meaning (such as implied real world size and sound) are recovered in an automatic but task-dependent way during reading. Third, thanks to a functional magnetic resonance imaging experiment, I show a representational shift along the ventral visual path: from perceptual features, preferentially encoded in primary visual areas, to conceptual ones, preferentially encoded in mid and anterior temporal areas. This result indicates that complementary dimensions of the semantic space are encoded in a distributed yet partially dissociated way across the cortex. Fourth, by means of a study conducted with magnetoencephalography, I present evidence of an early (around 200 ms after stimulus onset) simultaneous access to both motor-perceptual and conceptual dimensions of the semantic space thanks to different aspects of the signal: inter-trial phase coherence appears to be key for the encoding of perceptual while spectral power changes appear to support encoding of conceptual dimensions. These observations suggest that the neural substrates of different components of symbol meaning can be dissociated in terms of localization and of the feature of the signal encoding them, while sharing a similar temporal evolution.