CGSC 2001 Lecture Notes - Lecture 5: Generative Grammar, Universal Grammar, Long-Term Memory
Document Summary
Decision making: search trees have many branches, heuristics! Explanation: explain a fact from what is known (rules, hypothesizing based on what is known = abduction. Learning: rule based view of learning: acquisition, modification, deletion, and application. Inductive generalization: rules learned from instances: chunking (soar) and composition (act-r): learning rules from other rules, specialization: learn subcases of rules. Incremental learning: associate numerical weights with rules, augment when rule is used, favour selection of rules with higher weights. Language: generative linguistics: complex grammar that consists of rules! (unconscious knowledge) Ex tenses and rules, exceptions, etc: chomsky: innate universal grammar (space of possible rules, particular languages learned abductively vs deductively. Human like reasoning: rate of learning slows down as more is learned, problem space is less structured at beginning of a task but chunks can be built rapidly. As chunks build, speed of performance increases. As higher level chunks build they become less useful, because the situations they apply to are rare.