Live Chat
1-888-369-5603 44-20-7183-0318 61-28-014-8214

Connectionism term paper

[...]
"Attributing a cause to an event is an indispensable mental capacity that enables humans to identify the factors in their environment responsible for their hardship or well-being, to predict similar events in the future, and to increase their control over their occurrences. In fact, not only humans but also animals use the capacity to detect cause-effect relationships in order to prosper and safeguard their everyday adaptation and long-term survival. Given this long evolutionary history from simple invertebrates like the mollusk (Hawkins, 1989) on, it seems reasonable to assume that much, if not all, causal learning is governed by very elementary and simple cognitive processes, operating in animals as well as humans. The causal learning process involves the development of mental associations or connections between potential causes and the effect, and that this learning process can be profitably analyzed from a connectionist perspective. The associative approach to causal learning grew from research on animal conditioning (Rescorla & Wagner, 1972), and has gained increasing support in current cognitive research on human causality and categorization (for reviews, see Allen, 1993; Shanks, 1993, 1995) and in connectionist or adaptive network models of human memory and thinking (Gluck & Bower, 1988a; McClelland & Rumelhart, 1988). The various theoretical proposals put forward in these diverse areas of research seem to converge to a few fundamental principles. For instance, the popular associative model of animal conditioning proposed by Rescorla and Wagner in 1972 is, in fact, formally equivalent to a specific class of two-layer connectionist models based on the delta learning algorithm."
[...]

 

[...]
"The original Cohort model (Marslen-Wilson & Welsh, 1978) represents the first attempt at providing systematic answers to the three previous questions. In doing so, this model assumes two successive stages of processing. During the first stage, all words that exactly match the onset (i.e., the initial one or two segments) of the target word are activated. In this way, a set of competitors or the word initial cohort of the target is generated. No other words are allowed to enter the cohort and to compete for recognition. After the initial activation phase is a stage of deactivation during which the cohort members that mismatch later arriving sensory input are eliminated from the cohort via bottom-up inhibition. The number of cohort members decreases as more stimulus information becomes available. Thus, in response to the question about the lexical competitors of a target word, the cohort model asserts that the competitors at any given time are those words that match and are aligned with the target word. Neither words that are aligned but that mismatch the target (even in only a single distinctive phonological feature) nor did words that match the target but are not initially align count as competitors. The answer to the question about the competitors' influence on the target is also simple. The mere presence of competitors in the cohort prevents the target from being recognized, but neither the number (cohort size) nor the nature (e.g., competitor frequency) of the cohort members affects the time course of word recognition in this model. Only the final cohort member(s)-the word(s) matching the target the longest--determine the target's recognition point."
[...]

 

[...]
"Searle's original arguments treated the field of artificial intelligence as a single endeavor, and the arguments were put forward as applicable to any computational system. But as we have seen, AI is far from unified, and it is vital to draw careful distinctions between different approaches before issuing blanket statements. The "Syntax is not sufficient for semantics" argument, superficially very appealing, may well apply strongly to traditional, symbolic AI. If one's intuitions about AI are derived entirely from the consideration of traditional models, one might take the argument as a condemnation of the entire field. But this would be to ignore the fact that other approaches are possible."
[...]