Re: histrionics

[b-collins@home.com (Bob Collins)]

In article <7rb5lm$dhu$1@nnrp1.deja.com>, Ted Dennison
<dennison@telepath.com> wrote:

> In article <7r9rkj$g75$1@nnrp1.deja.com>,
>   Robert Dewar <robert_dewar@my-deja.com> wrote:
> > US experts in expert systems were unconvinced, and considered
> > LISP a better choice. That opinion has not significantly
> > changed, although you probably find more expert systems written
> > in "conventional" languages like C or Ada these days than
> > before.
> 
> While taking my AI class last summer, it occurred to me that inference
> engines share a number of similarities with compilers. There are
> differences too of course. But it might be possible to use an
> OpenToken-style approach to create a general Ada inference engine and
> rule-based-system whose rules could be programmed entirely in Ada.

Now getting far afield, but no further than our hero (RD)
in discussions about Prolog, both if-then expert system
rules and parser rewrite rules are variants of Post
Production Systems. One of my doctoral students wrote a
thesis that used LR parsing as the inference engine for
an expert system. He combined an earlier idea about using
Finite State Automata to provide procedural control for
expert systems, and his resulting grammars served as both
procedural control and as the if-then rules. Of course, the
grammars were ambiguous, so as to allow conflict resolution
heuristics.

He showed that his method outdid the Rete algorithm (in the
lab). The AI journal he submitted the work to did not publish
his results (I think) because they did not understand LR
parsing and the PL journal he submitted was not interested in
the AI implications.

To bring the subject back to Ada, he did get a conference
invitation to one of the Ada and AI conferences at George
Mason University. The parser, the AI hooks, the parser
generators, the semantics were all written in Ada, just as
suitable as LISP or Prolog for big AI work. (Big = thousands
of productions).

-- 
Bob Collins     <mailto:b-collins@home.com>