The Representation of Objects in the Brain, and Its Link with Semantic Memory and Language: a Conceptual Theory with the Support of a Neurocomputational Model

Recognition of objects, their representation and retrieval in memory and the link of this representation with words is a hard cognitive problem, which can be summarized with the term “lexico-semantic memory”. Several recent cognitive theories suggest that the semantic representation of objects is a distributed process, which engages different brain areas in the sensory and motor regions. A further common hypothesis is that each region is organized by conceptual features, that are highly correlated, and neurally contiguous. These theories may be useful to explain the results of clinical tests on patients with lesions of the brain, who exhibit deficits in recognizing objects from words or in evoking words from objects, or to explain the use of appropriate words in bilingual subjects. The study of the cognitive aspects of lexico-semantic memory representation may benefit from the use of mathematical models and computer simulations. Aim of this chapter is to describe a theoretical model of the lexico-semantic system, which can be used by cognitive neuroscientists to summarize conceptual theories into a rigorous quantitative framework, to test the ability of these theories to reproduce real pieces of behavior in healthy and pathological subjects, and to suggest new hypotheses for subsequent testing. The chapter is structured as follows: first the basic assumptions on cognitive aspects of the lexico-semantic memory model are clearly presented; the same aspects are subsequently illustrated via the results of computer simulations using abstract object representations as input to the model. Equations are then reported in an Appendix for readers interested to mathematical issues. The model is based on the following main assumptions: i) an object is represented as a collection of features, topologically ordered according to a similarity principle in different brain areas; ii) the features belonging to the same object are linked together via a Hebbian process during a phase in which objects are presented individually; iii) features are described via neural oscillators in the gamma band. As a consequence, different object representations can be maintained simultaneously in memory, via synchronization of the corresponding features (binding and segmentation problem); iv) words are represented in a lexical area devoted to recognition of words from phonemes; v) words in the lexical area and the features representing objects are linked together via a Hebbian mechanism during a learning phase in which a word is presented together with the corresponding object; vi) the same object representation can be associated to two alternative words (for instance to represent bilinguism). In this case, the two words are connected via inhibitory synapses, to implement a competition among them. vii) the choice of words is further selected by an external inhibitory control system, which suppresses words which do not correspond to the present objective (for instance to choose between alternative languages). Several exempla of model possibilities are presented, with the use of abstract words. These exempla comprehend: the possibility to retrieve objects and words even in case of incomplete or corrupted information on object features; the possibility to establish a semantic link between words with superimposed features; the process of learning a second language (L2) with the support of a language previously known (L1) to represent neurocognitive aspects of bilinguism