MSG:  *MSG   1487  
Date: 12/17/82 18:47:30 
From: LRC.Slocum at UTEXAS-20
Re:   Cognitive Systems

Date: 17 Dec 1982 1745-CST
From: Jonathan Slocum <LRC.Slocum at UTEXAS-20>
Subject: Cognitive Systems
To: bboard at MIT-AI


Two days ago I was in Houston, at the Open House for the new Symbolics,
Inc., regional office.  Cognitive Systems [Roger Schank's company] was
represented by Steven Shwartz and Ann Drinan, and they gave a "live" demo
presentation of their new English-to-database interface, supposedly
already delivered to a local customer.  This system promises understanding
of UNRESTRICTED English queries [emphasis mine] COMPARABLE TO A HUMAN'S
(within the subject area covered by the database).  To quote from their
slide, "DISADVANTAGES: None".  To open their talk, they went over the
failure of Machine Translation in the 50's and early 60's, and how
enormous amounts of world knowledge was necessary to correct the glaring
deficiencies of those [20-year-old!]  approaches.  In their demonstration,
they were following a fixed, prepared script (if you'll pardon the pun).

Then it happened: when they said something about spelling correction, and
how wonderfully better their program does it than anyone else's in the
world, someone in the audience (a Houston bank officer, or something such)
asked them to misspell "scale" as "scle" in the pending command.  I sensed
a distinct reluctance to try this (a [prepared] example in a subsequent
query was promised), but they did it anyway.  The resulting input was

	Use a scle of 1 inch to 2000 feet.

The response was

	There are two states where Bibb county is located: Alabama and
	Georgia.  Pick the state you mean.

Wow!  But the topper was the following: in the subsequent prepared query,
there was a real ambiguity: the query mentioned Bibb County without
specifying a state.  The system responded with EXACTLY the same request
for clarification -- which, in this new context, was precisely the right
thing to do.  Hmmmm....