Fri Mar 2, 2007
5101 Tolman Hall, 11 AM–1 PM
|Institute of Cognitive and Brain Sciences
Lotfi Zadeh (UC Berkeley)
“A Natural-Language-Based Computational Theory of Perceptions (CTP)”
In conventional modes of computation, the objects of computation are values of variables. In computation with information described in natural language, or NL-Computation for short, the objects of computation are not values of variables but states of information about the values of variables, with the added assumption that information is described in natural language. A simple example: X is a real-valued random variable. What I know about X is: (a) usually X is much larger than approximately a; and (b) usually X is much smaller than approximately b, with a less than b. What is the expected value of X? Another simple example: (a) overeating causes obesity; and (b) obesity causes high blood pressure. I overeat. What is the probability that I will develop high blood pressure? The importance of NL-Computation derives from the fact that much of human knowledge, and particularly world knowledge, is expressed in natural language.
Natural languages are intrinsically imprecise. A prerequisite to computation is precisiation, that is, translation of the given information into what is referred to as precisiation language. A key idea in the approach which is described is that of representing information as a so-called generalized constraint. This idea serves to construct an expressive precisiation language called the Generalized Constraint Language (GCL), whose elements are generalized constraints. Propositions and predicates drawn from a natural language are precisiated via translation into GCL. Through precisiation, computation with information described in natural language is reduced to computation with generalized constraints. Computation in GCL is carried out through application of rules which govern combination, qualification, propagation, and counterpropagation of generalized constraints.