THE USE OF AGENT-BASED MODELS IN COGNITIVE LINGUISTICS: AN APPROACH TO CHOMSKY’S LINGUISTICS THROUGH THE CLARION MODEL
Authors: Miriam Bait & Raffaella Folgieri & Oscar Scarpello
Number of views: 457
In this paper we propose the use of Agent-Based Models (ABM) (Gilbert 2008) to study the development of historical natural languages starting from a universal grammar according to Chomsky’s "Theory of the principles and parameters" (Chomsky 1995) . The CLARION architecture, designed by Ron Sun (Sun 2002) integrates implicit and explicit knowledge, cognitive and meta-cognitive levels, with the motivational aspect, i.e. accepting the cardinal principles of the embodied mind (Clark 1997) and recognizing the basic role of direct men- environment interaction in cognitive mechanisms. Ron Sun develops these points in a theory of mind and in a thorough discussion of learning problems. The goal of an artificial neural network (ANN), based on a CLARION architecture, is to verify theoretical assumptions through simulation, bringing together the dichotomy between implicit (subsymbolic) and explicit (symbolic) knowledge through a learning mechanism realized by the extraction of explicit rules by subsymbolic knowledge, based on interaction with the world. In the real world, cognitive operations are mostly performed unconsciously. Moreover, learning is carried out through attempts, in dynamic circumstances. The methodology allows to observe the development of cognitive structures of individual agents through ABM and contribute to studying the emergence of unplanned and unexpected routines or mechanisms. The use of neural models as learning tools implies that the simulations are realistic, considering the relationship between intentional behaviour, learning, desires, individual structures and social structures. The simulation, thus, enables a study the mind from an evolutionary perspective (that of satisfying a particular need in a physical and sociocultural world), understanding how individual structures and social institutions and environment could change each other. Through ANN-based models one can build realistic 'intelligent agents', i.e. with a 'mind', minimizing the programming of rules of behaviour and letting the interaction with the environment produce efficient behaviour.