A feature-based model of semantic memory: the importance of being chaotic.

Semantic memory representations have often be modeled in terms of a collection of semantic features. Although feature-based mod- els show a great explanatory power with respect to cognitive and neu- ropsychological phenomena, they appear to be underspecified if inter- preted from a neuro-computational perspective. Here we investigate the retrieval dynamics in a feature-based semantic memory model, in which the features are represented by neurons of the Hindmarsh-Rose type in the chaotic regime. We study the state of synchronization among fea- tures coding for the same or different representations and compare the correlation patterns obtained by analyzing the whole neural signal and a manipulated signal in which the sub-threshold component is ruled out. In all cases we find stronger correlations among features belonging to the same representations. We apply a formal method in order to represent the state of synchronization of features which are simultaneously coding for different representations. In this case, the synchronization and de- synchronization pattern that allows for a shared feature to participate in multiple memory representations appears to be better defined when the whole signal is considered. We interpret the simulation results as sugges- tive of a role for chaotic dynamics in allowing for flexible composition of elementary meaningful units in memory representations.