Re: Time ontology in PDDL

From: George Ferguson (ferguson@cs.rochester.edu)
Date: 10/22/02

  • Next message: pat hayes: "Re: Time ontology in PDDL"
    I'll let Jerry have first shot at most of these questions, but a few of them 
    seem like my problem (at first glance).
    
    I also have a few comments of my own at the bottom.
    
    Drew McDermott wrote:
    >>Axioms 4.2-1 et seq. assume <=> is associative and commutative, which it 
    >isn't (unless this is a KIFism I'm unfamiliar with).
    
    This may be a problem with the parser I used to read Jerry's logical notation 
    and turn it into KIF. I had some problems getting the precedence of operators 
    like "<=>" to come out right, but I thought I had fixed it. If 4.2-1 et seq. are 
    the axioms that I think they are (the ones related the three forms of clock/cal 
    notations), then they may have come out wrong. Yes, I see that they are. I'll 
    have to take a look. Maybe I'll just break them into two separate cases in the 
    text for ease of parsing (or, at this point, I could do it by hand...).
    
    >>Why do we need both *second* and *sec*?  (etc.)  What does
    >   *DayOfWeek* denote exactly?  I realize that they are introduced in
    >   connection with time intervals that lie on "official" boundaries,
    >   such as next Friday, as opposed to some random 24-hour period.  I
    >   just don't see how the extra temporal units contribute.  If we start
    >   with an "era" (such as CE(..)), and carve intervals out by first
    >   getting the n'th year, then the m'th day of that year, and so forth,
    >   we can use the original units to keep track of how long each
    >   interval is.  That's what happens in my version, which lacks all
    >   temporal units except *second*, *minute*, *hour*, *day*, *week*,
    >   *month*, and *year*.  I could make it look more like the original if
    >   I understood the original.
    
    I had the same question regarding *second* and *sec* just yesterday as I was 
    working on some markup derived from the Ontology.
    
    >>I disagree that we need both clock-int and cal-int.  Subintervals should 
    > begin counting from 1 in all cases.  So what if 8:00 names the first 
    > minute of the ninth hour of the day?
    
    I also ran into this. When one considers typing the arguments of these 
    predicates (as I was doing), Jerry's notation of "integer(n)" doesn't really cut 
    it. I was trying to decide if he meant "natural" (0, 1, 2, ...) or "ordinal" (1, 
    2, 3, ...).
    
    In the case of the "time-of" predicate, it depends on whether we intend to pass 
    in "1" for the first minute, which makes sense "theoretically", or "0", which 
    makes sense in the sense of "that's what the unix time function would give as 
    for that slot".
    
    I also have a message from Gerard Donlon (gdonlon@bbn.com):
    ---cut---
    The following axiom:
    
    <axiom id="2.4-1">
           (A t1,t2)[instant(t1) & instant(t2)
                        --> [before(t1,t2) v t1 = t2 v before(t1,t2)]]
    </axiom>
    
       The second "before" either needs to swap arguments to before(t2,t1) or change 
    to after(t1,t2).
    ---cut---
    He looks right to me. The arguments should be swapped.
    
    I'll start preparing a revision of the document to accomodate this fix and 
    whatever comes out of discussion of Drew's points.
    
    George
    


    This archive was generated by hypermail 2.1.4 : 10/22/02 EDT