next up previous contents
Next: Metalevels Up: Methods as functions Previous: Methods as functions

Introspection and Transparent Functions

This section can be skipped by the casual reader.

In this section we discuss the rationale of Noos methods being also feature terms. In tex2html_wrap_inline1861 -calculus, functions have the property of also being values--that is to say functions-as-values can be manipulated by functions--passed as parameters, returned by functions, etc. In languages like Lisp or Scheme closures--a function plus the set of bindings referenced by that function--are first class objects that can be manipulated by other functions. Closures can be returned by functions, passed as parameters--but they are opaque. A closure can be understood as a value that packages a function and a representation of stategif. However, there is no way any external program is able to inspect any variable binding inside a closure. A closure can be passed around and applied (evaluated)--but nothing else. We say closures are opaque because of this property. For instance, if we consider a closure where first parameter of a function in the closure, say x, is bound to 3, there is no any external function can obtain the value of x. What is needed to access x is to have a function defined in such that ``captures'' x and returns the value of x. Even so, in order to update that x variable, there has to be another function defined that ``captures'' x and performs an assignment on x. In fact, with these basic constructs on closure handling is easy to provide object-oriented extensions of Lisp and Scheme.

On the other hand, Noos methods are also feature terms, and they are not only first-class objects (values) but they are inspectable to external methods. Since method parameters are passed by names (feature names), we can have path references to those features. We can say thus that Noos methods are transparent. For instance, if we define Wet-Causal-Expl-Home as a refinement of wet-causal-expl of Example 2.9 as below, we can have references to the features of Wet-Causal-Expl-Home like those to any other feature term.


Notice that in this example the features place and effect of the method can be referenced

Why we need methods to be transparent? The reason is to be able to perform introspective tasks, i. e. to perform forms of inference that need to inspect and reason about other methods and how they have been used to solve particular tasks. An example of using introspection is a form of compilative learning--also called explanation-based learning (EBL). A learning that performs EBL builds a more efficient and compact method examining the proof structure of the methods that were used to solve a particular problem. Such a method requires the introspective capability of accessing and manipulating the terms and formulae in the proof structure. Consequently, integrating some forms of learning techniques--a main goal in Noos language design--requires some way of introspection that in Noos is achieved by the methods themselves being feature terms. In fact, also other reflective capabilities are needed, like the method and task constructs. The constructs method and task allow to access the method pertaining to a specific feature. For instance, in order to access the place feature of the wet-causal-expl method used in the ground-status task of Example 2.10 we can use these constructs to build up a reflective reference as follows:

     (>> place of (method of (task ground-status of (>> house of John))

These and other reflective capabilities are explained in Chapter 4.

next up previous contents
Next: Metalevels Up: Methods as functions Previous: Methods as functions

Enric Plaza
Thu Jan 23 11:36:28 MET 1997