THE STRUCTURE AND ORGANIZATION OF A SEMANTIC NET FOR COMPREHENSION AND INFERENCE

摘要: We have developed a network representation for propositional knowledge that we believe to be capable of encoding any proposition expressible in natural language. The can regarded as computer-oriented logic with associative access paths from concepts propositions. Its syntax is closely modeled on predicate calculus but includes constructs expressing some kinds vague and uncertain knowledge. allows the efficient use caselike semantic constraints arguments purpose language comprehension: these are simply implications predicates concerned. Our approach comprehension based nonprimitive representations. argue primitive representations simple propositions often extremely complex, offer no real advantages. demonstrated ideas mini-implementation mapping certain declarative sentences into representation. implementation emphasizes proper handling iterated adjectival modifiers, especially comparative modifiers. More recently, worked problem rapid facts relevant query. solution involves back-link structures propositions, called “topic skeletons,” which conform general topic hierarchies memory. For example, “Clyde grey” classified under “coloring” Clyde, subsumed “appearance” topic, turn “external quality” finally “physical Clyde. form query (or an assertion) used determine what memory should accessed starting points, associated skeletons followed order information. feasibility building such hierarchies, inserting information them automatically, accessing inserted second experimental implementation. hierarchic organization appears providing order-of-magnitude improvements question-answering efficiency, only doubling storage costs.