The processing of verbs and nouns in neural networks Insights from synthetic brain imaging.pdf

The Processing of Verbs and Nouns in Neural Networks: Insights from Synthetic Brain Imaging Angelo Cangelosi Centre for Neural and Adaptive Systems and Plymouth Institute of Neuroscience University of Plymouth (UK) Domenico Parisi Institute of Cognitive Sciences and Technologies National Research Council (Italy) 1. Language processing in natural and artificial neural networks Artificial neural networks have been frequently used to build models of language processing abilities in adults and children. They have been employed to study the acquisition of lexicon and meaning, the processing of morphology and syntax, reading and speech production (cf. Christiansen et al., 1999). However, much connectionist work on language tends to study language in isolation from other cognitive abilities and from the sensory-motor interactions of the organism with the environment. This is an obstacle to considering the important issue of the “grounding” of symbols on sensory-motor experience through which linguistic symbols acquire their meaning. Furthermore, in most connectionist models the issue of the neural plausibility and significance of the network architecture and functioning is not addressed, and this makes it impossible to compare the simulation results with such neural data as neuroimaging data. Computational models have also been successfully employed for investigating the evolution of language through simulation (Cangelosi Parisi, 2002; Kirby 2002). These models use various approaches: artificial neural networks (e.g. Batali, 1994; Cangelosi Harnad, 2000), rule-based systems (Kirby, 2001), and robotics (Steels Kaplan, 1999). Neural networks have proven particularly useful because they can focus on the influence of both cognitive and neural mechanisms on language development and evolution. For example, evolutionary neural networks, or networks viewed in an Artificial Life perspective (Cangelosi Harnad, 2000; Parisi 1997), are used to