COMMENT ⊗   VALID 00006 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Random ideas:
C00009 00003	PRESENT PRIMITIVES:
C00011 00004	LATEST PROGRAM APPROACH:
C00012 00005	NEXT:?
C00014 00006	MULTIPLE REPRESENTATION: for calendar
C00018 ENDMK
C⊗;
Random ideas:

Room problem = intersections in time and space.
	Like an appointment scheduler.
	If times are fairly definite, could run through both comparing
	locations to determine best time and place for a meeting.
	Must identify relevant time & distance scales.

Need better distinctions between generalities and specifics, or maybe just
	a lot more data.
	Chuck doesn't have anything fancy (see pp 59-66).
	Want to say "It's like x except for:"
	1) A is replaced by B.
	2) A is inserted (deleted) between B and C.
	3) A (everything) gets additional property (PROP VAL).
Want general algorithm for searching up ISA and PART_OF trees to locate
	inheritable information.
	English names (descriptions) are inheritable.  Must find inherited
	name and then add on adjectives derived from local information.
	Then, could get ordering of LOUIS_15 and LOUIS_16 from sequence number
		stored separately from general name, "LOUIS".

Need better English recognition than PARRY's.  Could bypass the problem
	by inputting parenthesized conceptual representation.

Relating two time sequences:
	First, estimate absolute (up to a cyclic point) times on each side.  
	Beware of cyclic tricks where the answer is "Repeatedly" or "Never".
	Look for BEFORE & AFTER relations between elements on opposite sides.
	Search should spread out in parallel from all known points.
	If DURATIONs and a single point of contact are known, can step down
	through both sides and match up subsequent points.
	When CONCURRENT points are found, can recurse to a finer time(space)
	scale.
	Some things just aren't related (normally). eg:
	Nixon resigned today is in the context of US-HISTORY, and
	I ate lunch is in the context of MY-RECENT-PAST.

Identify implicit time relations (enabling, precondition, resultative).

Facts are often approximations with pointers to better approximations.
	e.g.
	YEAR ≡   4  SEASON	SEASON ≡  3  MONTH	DAY    ≡ 24 HOUR
	YEAR ≡  12  MONTH	MONTH  =  4+ WEEK	HOUR   ≡ 60 MINUTE
	YEAR =  52+ WEEK	MONTH  = 30+ DAY	MINUTE ≡ 60 SECOND
	YEAR = 365+ DAY		WEEK   ≡  7  DAY
Know how much out of phase 2 periodic things are.
Every sequence of things is named and/or numbered.

Finer and finer grain makes (STOP (START (STOP X))) meaningful.
The START of something measured in YEARs might be measured in WEEKs.

See for time references:
	Bertram C. Bruce, "...Temporal References...", Artificial Intelligence,
	III, 1972.

(SEQUENCE (event1 ... eventN) event) brings up some problems.
One is the implicit BEFORE relations among the members of a
SEQUENCE.  Another is the wrap-around nature of calendars in general while
the specific instances of months or days don't wrap-around.
PRESENT PRIMITIVES:

event =
	atom
	(BEFORE event event)
	(CAUSE event event)
	(CONCURRENT event event)
	(DURATION event time)
	(INVOLVE event event)
	(ISA event event)
	(KING atom atom)
	(START event)
	(STOP event)
	(TIME event time)
	(time event time)	for use in cheating
	(WAR atom atom)
	(PART event event)
	(SEQUENCE (event1 ... eventN) event)
time =
	(YEAR number number)

Definitional time relations:

(CONCURRENT X Y) ↔ (AND (BEFORE (START X) (STOP Y)) (BEFORE (START Y) (STOP X)))
(BEFORE (START X) X)
(BEFORE X (STOP X))
(BEFORE X Y) ↔ (BEFORE (STOP X) (START Y))
(AND (BEFORE X Y) (BEFORE Y Z)) → (BEFORE X Z)
(AND (TIME X (YEAR TX1 TX2)) (DURATION X (YEAR DX1 TX2-TX1+1))) ↔
	(AND (TIME (START X) (YEAR TX1 TX2-DX1)) (TIME (STOP X) (YEAR TX1+DX1 TX2)))
(TIME (YEAR X Y) (YEAR X Y))

Implicit time relations:

(CAUSE X Y) → (BEFORE X Y)
(INVOLVE X Y) → (CONCURRENT X Y)
(DURATION (KING ? US) (YEAR 4 8))
(DURATION (KING ? ?) (YEAR 10 40))
(DURATION (WAR ? ?) (YEAR 5 10))
(WAR X Y) → (BEFORE (POSS X MONEY) (WAR X Y))
(TAX X Y) →
	(AND (BEFORE (POSS Y MONEY) (TAX X Y)) (BEFORE (TAX X Y) (POSS X MONEY)))
LATEST PROGRAM APPROACH:

Who was the king of France during the constitutional convention?
(CONCURRENT (KING ? FRANCE) CONSTITUTION)

(BEFORE (START LOUIS_16) (STOP US_REVOLUTION))
(BEFORE (STOP US_REVOLUTION) (STOP CONSTITUTION))

(BEFORE (START CONSTITUTION) (START FRENCH_REVOLUTION))
(BEFORE (START FRENCH_REVOLUTION) (STOP LOUIS_16))
NEXT:?

Stay with historical data base (BOOK).
	Need typical questions to answer.
- - - - - - -
What did Eisenhower do during the Nixon 72 campaign.
	Not much, since he was dead.
Could I teach an afternoon class at Stanford, and be at an evening party in Boston?
What about vice versa?
Who was president during the the Israeli Yom Kippur War?
	Well Nixon was before hand he did x.
	And he was president after since he was demoted.
	Continuity is usual, so probably Nixon.
When was the Louisiana Purchase made?
	Well Washington and Adams didn't do it.
	And it was before 1812 because it belonged to us in that war so it must
	have been done between.  Maybe even search through those presidents.
- - - - - - -
	Many more implicit time relations
		(POSS ? MONEY) figures in many.
		Brings up problem of extraneous information.
		Same thing at different times brings up use of ISA.

Could switch data base back to people's daily activities (CALEND).
	Finer and finer time scales.
	Locations as well as times.
MULTIPLE REPRESENTATION: for calendar

I can see uses for the following facts:
	(TIME APRIL-75 (YEAR 1975 1975))
	(TIME APRIL-75 (MONTH 4 4))
But, since a span of time one unit long often has a name:
	(TIME APRIL-75 APRIL)
Also, most things are instances of others.
	(ISA APRIL-75 APRIL)
And events take place during (as part of) more global events.
	(PART APRIL-75 (YEAR 1975 1975))
Others:
	(SEQUENCE JAN FEB ... NOV DEC)
	(ISA JAN MONTH)
	(PART MONTH YEAR)
	(DURATION MONTH (WEEK 4 4))
?	(SEQ_NUM WED 4 DAY WEEK)

However the  potential number  of connections  between the  following
concepts is staggering:
	MAY-1-1975, MAY-1, (DAY 1 1), THURSDAY, DAY,
	MAY-1975, MAY, (MONTH 5 5), MONTH,
	(YEAR 1975 1975), YEAR

Besides this  problem, I don't  like this propositional  notation for
calendar  information.  I think  of calendars in terms  of nested and
sequence-numbered arrays.  A large  number of ISAs, PARTs, and  TIMEs
obscures the  "important" ideas  more than it  clarifies them.   This
leads  to  the  idea of  multiple  representations.    The "SEQUENCE"
primitive which  I think I  mentioned last  time was  a step in  this
direction.   Even though it looked  like other primitives,  it had an
unpredictable  number   of   positions  which   made   retrieval   of
partially-specified ideas much harder.

 SUBJECT: Dealing with Time
 
After thinking about  it for a while,  I have come to  the conclusion
that redundant  data is the most appropriate way  to go.  However, in
order to  minimize the  search, there  needs to  be more  distinction
between your various uses of time; that is, the fact that April-75 is
in 1975, and is an april which is a fourth month can be distinguished
by a kind  of typing mechanism.   Data should be both  redundant, and
carefully distinguished to make  clear what relations are being used.
Also, connections need only be established at one level, in the sense
that may-1-75 ought  to have some substructure which  indicates that
it is in a  month(may) with  day 1 and  year 75,  and that  all other
properties can  follow from the  structure, not  just the  relational
links.   If  you want  we can  talk about  this more.   We  could  get
together  Friday if you want.  Let  me know.  Or are you  sufficiently
near the end of your time here to not want to do more?