ok@quintus (06/16/88)
This is a forwarded message.
In the future, I suggest that instead of sending things to me for
forwarding, people should send mail to the Prolog Digest at
Prolog@Polya.Stanford.EDU
FORWARDED MESSAGE:
Does anyone know of any KRL written in Prolog besides APES? Has anyone any
comments on attempts to mimic KEE or ART type systems in Prolog?
Jerry Harper: Merrion Gates Software, Rosemount Court, Booterstown Avenue,
: Co Dublin, IRELAND
jharper@euroies.uucpfinin@antares.PRC.Unisys.COM (Tim Finin) (06/17/88)
In article <121@quintus.UUCP> jharper@euroies.UUCP (Jerry Harper) writes: > Does anyone know of any KRL written in Prolog besides APES? Has anyone any > comments on attempts to mimic KEE or ART type systems in Prolog? > > Jerry Harper: Merrion Gates Software, Rosemount Court, Booterstown Avenue, > : Co Dublin, IRELAND > jharper@euroies.uucp We here at the Unisys Paoli Research Center have been using KNET, a Knowledge Represention system implemented in Prolog, on a number of AI projects since about 1982. Attached is a brief description of KNET. More details can be found in an article which will appear in a forthcoming issue of the Journal of Logic Programming. Tim ___________________________________________________________________________ KNET is a frame-based knowledge representation system developed at the Unisys Paoli Research Center to support the acquisition and maintenance of large, real-world knowledge bases. The KNET system is currently being used in a rule-based diagnosis and maintenance system (KSTAMP), in a model-driven configuration expert system (BEACON) and in our natural language processing system (PUNDIT). The KNET representation language is similar to KL-ONE in that it is based on a formal semantic model which defines the meaning of objects in its knowledge bases. Objects in the knowledge base are called concepts, each of which has a unique name. A KNET knowledge structure consists of a number of concepts organized into two distinct data structures, called hierarchies. Each hierarchy has its own structure, but the two are related in a complex manner. The first hierarchy is the specialization hierarchy. Concept B is said to specialize concept A if B is a kind of A. There is a single designated root concept, and all concepts participate in the specialization hierarchy. It is essentially the IS-A hierarchy used as a basis of conventional semantic networks. It is used so that components (see below) may be described at a single concept and inherited by all the children of that concept. The specialization hierarchy is more general than the conventional IS-A network in that it is possible for a concept to specialize more than one other concept, thus inheriting properties from all its parents. This feature is termed multiple inheritance. The second hierarchy is the aggregation hierarchy. In this hierarchy, the children of a concept are its components, and taken as an aggregate they completely define that concept. In some cases these children are not components in a literal sense, but are properties, as for example the wall voltage required by a device may be a property of that device. A link in this hierarchy consists of an object called a role which names the component, together with a connection to the owner of the role and another connection to the type of the role (which is another concept). The final constituent of a KNET structure is the constraint. A constraint is a piece of executable code attached to the network. The code is available for use by a program using KNET (an application); whether, when, and how it is used is up to the application. A constraint is housed at some particular concept, and refers to one or more roles. Exactly one of the roles referred to by a constraint is designated as the trigger of the constraint; the remaining roles are the targets of the constraint. The purpose of the trigger is to provide a means for indicating within the KNET structure when an application program should use a constraint. The meaning of a constraint is always defined relative to the context in which the constraint occurs. This means that references to roles made from within an inherited constraint always refer to the local context rather than to the context in which the constraint was originally defined. Finally, it is important to maintain consistency in knowledge networks. The definition of consistency varies for differing kinds of knowledge representation, and depends on the semantic model implemented by the knowledge representation. For KNET, a fundamental consistency requirement is the subsumption requirement, defined as follows: If concept A has a role R with type B, and if concept A2 specializes A, then the type of the inherited role R2 must be B or a specialization of (a specialization of ...) B. If this requirement is not met, a subsumption violation is said to occur. The program which builds KNET structures, the Browser/Editor, automatically checks for and disallows subsumption violations and several other types of inconsistencies. KNET has been implemented in a standard subset of Prolog which has allowed it to be ported to several different Prolog systems on a variety of machines. Versions of KNET run on VAXes, Suns, Lisp machines and PCs. An interactive browser/editor system for KNET knowledge bases can use a simple ASCII terminal interface, enabling its use on each of these machines. A more sophisticated graphical browser/editor is available for use on Sun workstations. Tim Finin finin@prc.unisys.com Paoli Research Center ..!{psuvax1,sdcrdcf,cbmvax,bpa}!burdvax!finin Unisys Corporation 215-648-7446 (o) PO Box 517, Paoli PA 19301 215-386-1749 (h)
shardy@teknowledge-vaxc.ARPA (Steve Hardy) (06/17/88)
Jerry Harper asks about knwoledge representation languages written in Prolog (besides APES.) Teknowledge developed a prolog-based expert system shell called M.1. It was released in June 1984 and has sold close to 4,000 copies. M.1 is unusual in that it is a complete logic programming language as well as being an easy-to-use expert system shell. For example: /* --- simple EMYCIN-like heuristic rule --- */ if main-component-of-meal = beef then best-color-of-wine = red cf 75. /* --- list processing --- */ infix <>. /* for "append" */ [] <> LIST = LIST. if LIST1 <> LIST2 = LIST3 then [ITEM|LIST1] <> LIST2 = [ITEM|LIST3]. /* --- objects and inheritance --- */ issquare(obj-22). size(obj-22) = 5. if isrectangle(X) and height(X) = H and width(X) = W and H * W = R then area(X) = R. if issquare(X) then isrectangle(X). if issquare(X) and size(X) = S then height(X) = S. if issquare(X) and size(X) = S then width(X) = S. After releasing four versions of M.1 in Prolog, Teknowledge recoded the system in C. This led to the system being approximately five times faster and able to handle knowledge systems five times larger (up to 2000 rules on a 640K IBM-PC.) It was easier to work out the design of M.1 with Prolog than it would have been with C. Steve Hardy, Teknowledge Inc., (415) 424-0500 DISCLAIMER: My opinioins are not necessarily those of Teknowledge.