[schubert@cs.rochester.edu: definitions and analytic truth]

dam@ai.mit.edu (David McAllester)
From: dam@ai.mit.edu (David McAllester)
Date: Tue, 7 Jan 92 22:51:07 EST
Message-id: <9201080351.AA01379@raisin-scone>
To: schubert@cs.rochester.edu
Cc: interlingua@isi.edu, sowa@watson.ibm.com
Subject: [schubert@cs.rochester.edu: definitions and analytic truth]

    I'm wondering if JMc and DAM think that with the fix JMc proposes, the
    definition facility as outlined by RF now conforms with their precepts 
    -- or do the constructs still allow us to "sneak in new assertions"? 
    In particular, if we use defprimrelation to define a type hierachy 
    (which the syntax certainly permits) are we "sneaking in new assertions"?

I am convinced that any one of the proposed changes to the way constants are
defined fixes the problem (definitions become conservative extensions).