“ An Assessment of Tools for Building Large KB Systems , ”

As a primary architect of CRL-Ops and a former KnowledgeCraft class instructor, I had to deal with the general public’s misconceptions about forward versus backward chaining systems. Mr. Mettrey’s article, in my opinion, is the type which generates the confusion that forward chaining rule systems cannot “backwards chain.” This nonsensical view was held by the vast majority of our customers in the KC class. The section on Rule-Based inference implies that backward chaining is done only by Prolog in KC with its statement “by contrast, KnowledgeCraft implements backward chaining by supporting a version of Prolog.” Any forward chaining rules system can efficiently implement constrained backward chaining by simply using a goal structure to search for the required knowledge. Additionally, to the best of my knowledge, which I will gladly admit is at least a year out of date, Art does not really “implement[s] backward chaining by allowing a single backward-chaining goal pattern as one of the conditions in a forward-chaining rule.” More realistically, Art allows default backward chaining to occur when no forward chaining matches are available by automatically creating a goal to search for a match which would instantiate some partially matched rule. The section on debugging claimed that “Art and Kee provide graphic tracing of rule activity during inferencing” by failed to mention CRLOps’s facility for this. Finally, the statements in the “Requirements for Delivery Enviments” regarding parallelism in production systems are particularly misleading. The results quoted are early work of Gupta and Forgy. Gupta’s thesis (CMU CS Technical Report 861221, which, I’m sure, must significantly predate eh submission of this article, clearly shows that it is not true that “Production systems in particular appear to lend them selves to parallel execution.” A large amount of work is still required to demonstrate if productions will lend themselves to parallel implementations. Only one highly optimized general purpose parallel production system compiler exists [“Results of Parallel Implementation of 0ps.S on the Encore Multiprocessor,’ Gupta, Forgy, Kalp, Newell, and Tambe Technical report CMU-CS-87=146) and this achieved only reasonable speed ups. The amount of parallelism in large production system programs so far demonstrated by this implementation is limited to an optimistic 13 f this is in a large but highly match intensive program. What is the review procedure? Was this article passed around in Inference, IntelliCorp, and Carnegie Group? Was it reviewed by experts in the knowledge engineering / production systems research areas!