Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!oitnews.harvard.edu!news.cs.umb.edu!CTCnet!news.math.psu.edu!feeder.chicago.cic.net!news.cic.net!wolverine.hq.cic.net!newsxfer.itd.umich.edu!newsxfer3.itd.umich.edu!su-news-hub1.bbnplanet.com!cpk-news-hub1.bbnplanet.com!news.bbnplanet.com!cam-news-hub1.bbnplanet.com!howland.erols.net!ix.netcom.com!whitten
From: whitten@netcom.com (David Whitten)
Subject: Re: Does CYC really understand natural language? How?
Message-ID: <whittenE4s97F.BCD@netcom.com>
Organization: Netcom On-Line Services
X-Newsreader: TIN [version 1.2 PL2]
References: <5cmjb9$8jp@atlantis.utmb.edu>
Date: Wed, 29 Jan 1997 18:40:27 GMT
Lines: 85
Sender: whitten@netcom5.netcom.com

Rong Shen (rshen@marlin.utmb.edu) wrote:
: Hi:

: 	I was just re-reading Lenat's article in the Sept. 1995 
: Scientific American. On page 82, he wrote that "Similarly, CYC could 
: parse the request 'Show me happy people' and deliver a picture whose 
: caption reads 'A man watching his daughter learning to walk.'" 

: 	It sounds like the CYC could understand the request as we humans 
: do and could generate an appropriate response. I checked the web site at 
: www.cyc.com (CYC-NL System), and what I found was a simple description of 
: how the 
: lexicon, the syntactic parser, and the semantic interpreter worked. But 
: what happened between the semantic interpreter and the final response to 
: the request was vague.
: 	Can anyone tell me a little bit more about how CYC turns a 
: semantically correct sentence (a request) into a response? 
: 	Thanks for any hints or pointers.
: 	Rong --- rshen@marlin.utmb.edu
: 	 



Many of the details are proprietary, but at least some of the concepts
have been publicly discussed.  Cyc's Natural Language system seems to
take the sentence and then (using a lot of the knowledge it has about 
individual words and the concepts represented by those words) it will 
produce a query in a formal syntax called CycL.
Thus you example of "show me happy people", it will generate 
a query of 'show' me  the pictures which depict:

a) people recieving something they like 
b) people engaged in behaviour which they enjoy
c) people who are displaying happy emotions
d) people who are doing something which most people consider to be enjoyable.
e) people not doing something which they wouldn't enjoy.

Basically, each picture must be classified by a human to say what it
is depicting.  Cyc then uses the information it has about what is Common
Knowledge among Twentieth Century Americans to determine what kind of 
conclusions to draw from it.
A picture might be described as showing a smiling boy holding a lollipop.

The Knowledge Bank may have several rules or facts which apply to this 
description of the picture.

One is that humans smile to show the interior emotion of happiness.
One is that happy people have the interior emotion of happiness.
One is that people who have good things are happier than people who don't.
One is that a boy considers sweet things to be good things. (even if his
mother may not)
One is that a lollipop is a sweet thing.
One is that if a person is holding something, then that person has that
thing. 
One is that a boy is a person.(actually probably derived from the following
two rules/facts)
One that all humans are persons.
One is that all boys are humans.

Of course, these facts are just a few of the facts that could be relevant 
to a system like Cyc's.

When you add time as an element (like a picture of a smiling boy with his
hand extended toward a hand holding a lollipop)
then you get a more complex set of rules and facts that involve
expectations, scripts of behaviour, etc.

Regardless of what sets of rules and facts needed, they are all part of
the information used by a Knowledge Bank to draw new conclusions (or 
in your example) to retrieve specific pictures from a separate storehouse
from the Knowledge Bank. 

A simple database could record the same description as index terms for
our example picture but it would only be able to find the picture if
you asked for pictures involving boys or involving lollipops or involving
smiling.

A Knowledge Bank is able to work from your request for happy people 
back through reasons people might be happy or forward from descriptions
that indicate happy emotions.

Hope this helps,

David (whitten@netcom.com) (713) 791-1414 ext 6116

