The Semantic Web - Social Semantics: The Search for Meaning on the Web - page 53

results through programs like SHRDLU (Winograd 1972), since the 'semantics'
were ad-hoc and task-dependent, procedural semantics could not be used outside
the limited domain in which they were created. Furthermore, there became a
series of intense debates on whether these programs often purported to do what
they wanted even within their domain, as Dreyfus critiqued that it was ridiculous
that just because a program was labelled 'understand' that it did actually in any
way understand (1979). Interestingly enough, the debate between declarative and
procedural semantics is, under the right formal conditions, a red herring since the
Curry-Howard Isomorphism states that given the right programming language, there
is a tight coupling between logical proofs and programs so that the simplification of
proofs can be equivalent to steps of computation (Wadler 2001).
Within AI, research began into other forms of declarative knowledge repre-
sentation languages besides first-order logic that were supposed to be in greater
concordance with human intelligence and that could serve as more stable substrates
for procedural knowledge-based systems. Most prominent among these alterna-
tives were semantic networks , “a graphic notation for representing knowledge in
patterns of interconnected nodes and arcs”(Sowa 1987). Semantic networks are
as old as classical logic, dating back to Porphyry's explanation of Aristotelian
categories (Sowa 1987), although their first self-described usage was as a common
knowledge-representation system for machine-translation systems by Masterman
(1961). Motivated by a correspondence with natural language, semantic networks
were used by many systems in natural language processing, such as the work of
Wilks in resolving ambiguities using preference semantics and the work of Schank
using conceptual dependency graphs to discover identical sentences regardless of
their syntactic form (Schank 1972; Wilks 1975). Soon semantic networks were
being used to represent everything from human memory to first-order logic itself
(Quillian 1968; Sowa 1976). The approach of semantic networks was given some
credence by the fact that often when attempting to make diagrams of 'knowledge,'
humans often start by drawing circles connected by lines, with each component
labelled with some human-readable description. A semantic network about 'The
architect of the Eiffel Tower was Gustave Eiffel' is given in Fig. 3.1 . Note that it
refers declaratively to things in the world, but uses 'natural-language-like' labels on
its nodes and edges.
When researchers attempted to communicate or combine their knowledge rep-
resentation schemes, no-one really knew what the natural language description
'meant' except the author, even when semantic networks were used as a formal
language. The 'link' in semantic networks was interpreted in at least three different
ways (Woods 1975) and no widespread agreement existed on the most common sort-
of link, the IS-A link, which could represent both subclassing, instantiation, close
similarity, and more. This led to an assault on semantic networks by champions of
first-order logic like Hayes, who believed that by providing a formal semantics that
defined 'meaning', first-order logic at least allowed knowledge representations to be
transportable across domains, and that many alternative knowledge representations
could be re-expressed in first order-logic (Hayes 1977). In response, the field
of knowledge representation bifurcated into separate disciplines. Many of the