A dynamic approach to semantic content modeling

In the last decade, an increased scientific awareness manifested towards the importance of dealing with the dynamics of phenomena for a deeper understanding in any domain of science. Within cognitive linguistics, the dynamic approach has been argued by several authors to be a promising alternative to the symbolic paradigm based on logics and algebraic algorithms. The dynamic perspective on linguistic phenomena is also supported by recent researches in neural sciences. The results suggest that event-related brain potentials reflect a lexical-semantic integration which can be interpreted in terms of dynamical system theory. Other experiments have shown the presence of a separate nonverbal mechanism that is accessed by pictorial information, and may be later accessed by image mediated words. The paper starts from the premise that semantic structures can be identified in natural language at the neural level and investigates the possibility of implementing such a structure using self-organizing maps. Each linguistic component is modeled by an attractor implemented by a self-organizing neural map. More complex linguistic constructs are formed by a superposition of attractors controlled by chaotic sources, starting from the elemental level of phonemes. At the sentence level, the constituent words combined together convey a unitary meaning in the form of a resultant self-organizing map. The experimental results are relevant for this kind of dynamic approach and encourage further developments.