00100	              NOTES FOR SEMANTICS POSITION PAPER
00200	
00300		1. Semantics is the study of the function
00400		denotation(<expression>,<situation>)
00500	for various sets %E% of expressions and %S% of situations.
00600	
00700		2.  Two  expressions  have the same sense or meaning if
00800	they have the same denotation for all situations in %S.
00900	
01000		3. The notions of sense and denotation date from Frege,
01100	but  mathematical  logic  has so far found little use for sense
01200	and has relied entirely on denotation.  We propose  to  do  the
01300	same as much as possible.
01400	
01500		4. The greatest success of formal semantics has been in
01600	studying predicate and function logics. One of the high  points
01700	of this was Tarski's formal definition of truth.  Many theorems
01800	relate the syntax and semantics of logic and  various  theories
01900	expressed in logic such as arithmetic.
02000	
02100		5.  Formal  semantic methods were more recently applied
02200	by Kripke and others to modal logic with good results.
02300	
02400		6. Application of these ideas to define  the  semantics
02500	of  programming  languages  leads  to formal definitions of the
02600	correctness of compilers which probably can't  be  accomplished
02700	by purely syntactic methods.
02800	
02900		7.  This  seminar  is  concerned  with the semantics of
03000	natural language because  of  its  importance  to  the  various
03100	fields represented here.
03200		a.  Linguistics.   The  purely   syntactic   study   of
03300	language,  i.e.  of  the set of strings that are its sentences,
03400	has turned out to be very shallow.  In particular, it does  not
03500	provide enough information to study translation.
03600		b. Psychology. People use  language  to  express  their
03700	thoughts.   The  meaning  of  the  expressions  used  ought  to
03800	correspond to the thoughts expressed.
03900		c.  Artificial  intelligence.   For a machine to behave
04000	intelligently in a wide collection of  situations  it  must  be
04100	able  to  express internally what it knows about the particular
04200	situation it is in.  Thus we must be able to formalize much  of
04300	the expressive capability of natural language although we don't
04400	have to accept any aspects of particular natural languages that
04500	we  regard  as mere irregularities.  Our tendency is to use the
04600	formalism of predicate logic to express the content of English.
04700	
04800		8. All this leads us to ask that  semantics  treat  the
04900	following kinds of questions.
05000		a. In a particular situation do the phrases "the  noble
05100	dog" and "Beauregard" denote the same object.
05200		b. Do the phrases "Pegasus" and "the  winged  horse  of
05300	Greek mythology" always denote the same object if any.
05400	
05500		11. To put a point in a ddiλλifferent way,
05600	a major issue iswλ whether "deep structure" is a fact
05700	about human beings or about the situations that call
05800	forth speech.  If it is a mere psychological fact,
05900	then we should be able to invent other structures
06000	that represent the facts just as conveniently.  My
06100	own belief is that humans could use conveniently
06200	any structure that represented the facts and did not
06300	put a larger burden on short term memory than present
06400	languages.
     

00100		c.   From  what sentences and in what situations can we
00200	deduce "John knows Pete's  telephone  number"  and  what  other
00300	sentences  must  we have to be able to deduce from it "John can
00400	telephone Pete".
00500		d.  How can we abstract from the content of a situation
00600	so that we don't have to describe the whole world in  order  to
00700	determine  whether  one  sentence follows from two others or to
00800	determine the  referent  of  a  pronoun  in  a  sentence  of  a
00900	paragraph.
01000		e.  What consequences can be drawn from the sequence of
01100	sentences  of  a  narrative as in a newspaper story or a novel.
01200	What is the difference between what can be inferred in the case
01300	of literature and reportage.
01400	
01500		9.   I am reasonably confident that semantics should be
01600	able to deal with the above questions. In fact, I hope to  give
01700	some tentative answers to some of them in this seminar and some
01800	ideas are already  presented  in  (McCarthy  and  Hayes  1969).
01900	However,   I   am   not  at  all  confident  that  they  are  a
02000	representative collection of questions  and  particularly  fear
02100	that  I  have  included  some advanced questions while omitting
02200	more elementary ones.      I  hope  to  learn  from  the  other
02300	presentations  to  devise  a  more representative collection of
02400	questions.
02500	
02600		10.  I think that the most  important  questions  about
02700	semantics  do  not  depend on human psychology or on the way in
02800	which language has developed historically.    This  is  because
02900	they   concern  the  units  of  information  available  to  and
03000	transmittable by any automaton in the same position as a  human
03100	and  with  the same possibilities for obtaining information and
03200	the same information requirements for action.  Thus  a  machine
03300	or  a  Martian would have to ask a question equivalent to "What
03400	is John's telephone number?" in order to be able  to  telephone
03500	John.     Examples wherein the question is culturally dependent
03600	can be found, but I believe the basic sentences of any  natural
03700	language  correspond to units of information that are objective
03800	requirements of the communication situation. I  would  like  to
03900	see  if  I can maintain this position in argument.   It has the
04000	consequence  that   modern   syntax   is   not   so   important
04100	philosophically as is claimed, for example, in (Katz 1965).  It
04200	also means semantics should  be  investigated  by  asking  what
04300	information  transfer  is required by common situations as well
04400	as by purely linguistic methods.
04500