MAGMA-LISP: a machine language for artificial intelligence

The paper describes MAGMA-Lisp, an extended Lisp system proposed as an implementation tool for A.I. languages exploiting nondeterministic techniques. The main idea informing MAGMA-Lisp is that a tree structure of conceptually indipendent computation environments (context tree) is the supporting structure of any nondeterministic system. MAGMA-Lisp proposes this structure in a quite virginal form, so that the user can state his own techniques to prune, select and explore the available alternatives. In this sense, MAGMA-Lisp is to be viewed as a "machine language". This ideas of MAGMA-Lisp are contrasted with the systems that most influenced its design. The technique used in context implementation is described, showing how a very flexible context mechanism can be realized with a tolerable loss of efficiency. In particular, in spite of the complexity of the system, garbage collection does not result much more time consuming than in standard Lisp systems.