Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!watmath!clyde!burl!mgnetp!ihnp4!houxm!vax135!cornell!uw-beaver!tektronix!hplabs!sri-unix!SHEBS@UTAH-20.ARPA From: SHEBS@UTAH-20.ARPA Newsgroups: net.ai Subject: Re: Taxonomies Message-ID: <13052@sri-arpa.UUCP> Date: Wed, 29-Aug-84 11:31:47 EDT Article-I.D.: sri-arpa.13052 Posted: Wed Aug 29 11:31:47 1984 Date-Received: Tue, 4-Sep-84 07:52:45 EDT Lines: 28 From: Stan Shebs Some of the most recent KR systems attempt to provide meta-taxonomies; I know of RLL/Eurisko, MRS, and AGE, all Stanford products. Am not sure what LOOPS provides in the way of knowledge about representation schemes (although one could build something to recommend whether a given piece of information should be a logical assertion, an object, an instance variable of an object, Lisp code, etc). Meta-taxonomies are HARD. The ability to create a taxonomy of some body of knowledge implies that one has both a deep and broad understanding of that body. The creation of a meta-taxonomy implies that there is a similar level of understanding for many issues in knowledge representation, which is definitely *not* the case. We're still lacking adequate theories of multiple inheritance, nor have we plumbed the depths of strange logical systems. Looking at library science is an interesting idea; while I imagine that many of the classification schemes are informal (probably relying on human judgement), librarians have been classifying massive databases (books) for a long time. Moving farther afield, taxonomies in other AI areas are lacking. I asked a while back about taxonomies for rule systems, and found that there was about one paper, by Davis and King in a ca. 1976 MI. This, however, was an informal taxonomy, and not particularly susceptible to mechanization. Am still waiting for a tree that puts OPS5, Emycin, and Prolog on different leaves... stan