Search

Categories and word meanings tend to be interpreted as single entities that are activated in the mind when an individual categorizes the world or uses language. We argue that this view is mistaken by using models based on neural networks and adaptive behavior simulations. Connectionist simulations of populations of organisms that live and reproduce in an environment show that categories and word meanings are open and changing collections of activation patterns in the network's internal units rather than single activation patterns.

Internal representations of the environment are often invoked to explain performance in tasks in which an organism must make a detour around an obstacle to reach a target and the organism can lose sight of the target along the path to the target. By simulating a detour task in evolving populations of robots (Khepera) we show that neural networks with memory units perform better than networks without memory units in this task. However, the content of the memory units need not be interpreted as an internal representation of the position of target. The memory units send a time-varying internally generated input to the network's hidden units that allows the network to generate the appropriate behavior even when there is no external input. Networks without memory units do not have this internal input and this explains their inferior performance.