D.

Lewis General semantics

I. Introduction
Assuming that all possible languages may be given a categorially based transformational grammar,
we may answer two uestions!
"#$% what are meanings
"#&% how does semantic composition occur
'estrictions on semantics!
$. semantics is truth-conditional semantics "not translational semantics% (for a discussion, see
")arston &**&! +, ff%-! .emantics with no treatment of truth conditions is not semantics/
translational semantics 0ust postpones the problem of given a truth1conditional semantics
&. semantics is not psychological or sociological! that a population spea2s a language may be
factored into two moments! "a% the abstract description of a correlation between strings and
meanings/ "b% the description of the psychological or sociological facts in virtue of which a
population uses one semantics rather than another
II. Categorially based grammars
"a% Base component!
A categorial grammar es a conte3t1free phrase structure grammar with!
$. a small number of basic categories "., 4, )%
&. infinitely many derived categories! for any categories c, c
1
c
n
, we have the derived
category "c/c
1
c
n
%
5. context-free phrase structure rules of the form!
c 6 "c/c
1
c
n
% 7 c
1
7 7 c
n
where 878 indicates )94)A:
;. a finite lexicon
A categorial grammar thus defined is a notational variant of a conte3t1free grammar. 9nly the
output of the base component is relevant for semantic interpretation "< the transformational
component does not affect meaning%.
"b% Transformational component!
:he base component overgenerates li2e cra=y. A suitable transformational component may filter
agrammatical sentences due to! "a% strange word order/ "b% inappropriate iteration of operators/ "c%
allow for noun phrase ob0ects.
>f meaning is determined by the base component, then it is easy to provide answers for uestions
"#$% and "#&% above.
III. Intensions for basic categories
An answer to "#$% has two moments! "a% determine the function of meanings/ "b% posit something
that accomplishes that function.
One function of meaning is to determine, together with facts about the world, e3tension "relative
to relevant parameters%. >n particular!
$. sentence meaning determines the conditions under which a sentence is true or false "cf.
>nde3, conte3t, and content for a different ta2e on sentential meaning%
&. name meaning determines a thing
5. common noun meaning determines a set
Carnapian intensions "possibly partial functions from indices to appropriate extensions% may
fulfill this function of meaning.
An index is an n1tuple of elements "coordinates% which enter into the determination of e3tensions.
An inde3 has several coordinates!
1. possible world coordinate
&. contextual coordinates! time, place, spea2er, audience, indicated ob0ects, previous discourse
3. assignment coordinate
?owever, intensions are not all there is to meaning "since all tautologies differ in meaning and share
intension%.
(Digression! :ars2ian ontological hierarchies and the impossibility of a universal semantic
metalanguage-
IV. Intensions for derived categories
:here are some )@) "< ad0ectives% that do not have e3tensions v.g., alleged. >t is best to thin2 of
meanings of derived categories as non-Carnapian compositional intensions! functions from
intensions to intensions!
an appropriate intension for a derived category "c/c
1
c
n
% is a function from Ac
1
1
intensions c
n
1intensionsB to c1intensions
Semantic projection is functional application (#&-!
(("c/c
1
c
n
% 7 c
1
7 7 c
n
-- <

!
1

n
"
C3amples!
)@) (ad0ectives-! function from common noun intensions to common noun intensions
.@4 (DE-! function from name intensions to sentence intensions
".@4%@".@4% (adverbs-! function from DE1intensions to DE1intensions
(.impler intentions might be possible for some categories, but only if functional application as the
only form of semantic pro0ection is dropped.-
(#election restrictions may be accounted for by allowing partial compositional intensions! 8green81
intension may be undefined for the 8idea8 argument. Alternatively, we may allow for the assignment
of null intensions for the problematic cases.-
V. eanings
Differences in intension give only coarse differences in meaning!
compound e3pressions may have same intension and different meaning
co1intensionality is a sufficient condition of synonymy only for le3ical constituents
eanings may be identified with semantically interpreted phrase mar$ers minus the terminal
nodes! these constructions give, for any constituent, "a% its syntactic category, "b% its intension, and
"c% the categories and intensions of its constituents.
A few ualifications!
$. interpreted phrase structures do not cut too finely! in ordinary tal2, we say that . and it is
not the case that not . have the same meaning/ however, the ordinary concept of meaning
may signify anything from intensions to interpreted phrase structures "cf. >nde3, )onte3t,
and )ontent for essentially the same point about saying%
&. interpreted phrase structures are theoretical constructs! if they are not meaning themselves,
they represent meanings in one1to1one correspondence
Ehrase structure rules encode conditions of well1formedness of meanings! a meaning is well formed
if it is compound following phrase structure rules. A meaning is a tree obeying certain constraints!
A.,
.
<
DE&
"
4
%B
A4,
4
B A.@4,
DE&
<
Adv
"
DE$
%B
A.@4,
DE$
B A".@4%@".@4%,
Adv
B
9nce we define truth at an index, we are able to define further semantic notions!
a sentence # is analytic iff it is true at every inde3
a sentence # is analytic on a given occasion iff it is true at every inde3 differing from the
inde3 of the conte3t in at most world an assignment
a sentence # is an eternal trut! on an occasion iff it is true at every inde3 differing from
the inde3 of the conte3t in at most time and assignment
a sentence # is true on an occasion iff it is true under every assignment
Fe also are able to define notions of truth stronger than truth at every inde3 by means of the
concepts of semantic variant of !a meaning" m and of s-fixed semantic variant of m! s-true,
logically true, mathematically true, s-conse%uence, and others. (?ere, logical and mathematical
truth are defined without considering alternative interpretations. >t seems to be a case of Gol=ano1
conseuence.-
VI. "rammars reconstructed
:he base component of a categorially based transformational grammar may be specified by giving a
lexicon! each le3ical item is a triple Ae, c, B, where e is an e3pression, c is a category, and is an
intension. (> omit the other way of constructing the base.-
:he transformational component imposes finitely many constraints on finite seuences of phrase
mar2ers (i.e., it bloc2s certain derivations as ill1formed-.
Fe have the following definitions!
A seuence Ap
1
H p
n
B of phrase mar2ers is a transformational derivation of p
n
from p
1
in &
iff it satisfies the constraints imposed by :
An e3pression e has a meaning m iff there is a transformational derivation Ap
1
H p
n
B in :
such that e is the terminal string of p
n
and p
1
represents m "relative to the le3icon L%
e is a meaningful expression iff e has a meaning m, for some m
p
n
is a surface structure of e
p
1
is a base structure of e
p
'
p
n-1
are intermediate #deep$ structures of e
CO%V&%TIO%! let a base structure be any phrase mar2er that represents a meaning relative to L
"regardless of whether it is the base structure of an e3pression or not%.
Given a representation relation "between e3pressions and meanings%, semantic relations defined for
meanings may be defined for e3pressions.
:he transformational component imposes t'o constraints on se(uences of p!rase mar)ers! "a% a
local constraint, that two phrase mar2ers in a seuence stand in one "of finitely many%
transformational relation/ "b% global constraints, holding between non1ad0acent phrase mar2ers in a
seuence.
(.election restrictions do not occur upon le3ical insertion, but abnormal sentences are filtered out
before reaching the transformational component.-
VII. Treatment of (uantification and noun p!rases
#uantifier e3pressions are of type .@".@4%. Determiners appearing in uantifiers are of type ".@
".@4%%@). :hen, compositional intensions for uantifier e3pressions are functions from .@41
intensions to .1intensions.
?owever, occurrences of uantifiers in ob0ect position raise problems!
$. Fe may try doubling the ob0ect ta2ers to allow for uantifiers in ob0ect position (this is
not actually 0ust doubling! it is type1shifting-. ?owever, this way of interpreting
(uantifiers in situ can not account for the e3istence of two readings in uantificational
constructions li2e!
"$% Cvery woman loves a man
>t can not account for the reading in which a man has wide scope.
&. Another solution is variable binding! we may posit a binder, of type ".@4%@., different
from the uantifier itself, and have only uantifiers of type .@".@4%. :hus, we have a
structure li2e!
.
.@".@4% .@4
something ".@4%@. .
$ .@4 4
grunts 3$
?owever, the e3tension of variable binding to all cases of uantifier interaction puts too
much weight on transformations (actually, Lewis8 treatment of binding is the standard
treatment of EIE/ the problem is that in the $JK*s, nobody thought of uantifier movement,
or of the idea of covert movement-.
5. 9ne may try to assimilate names to (uantifiers! for every name of category 4, we define
a pseudo1name of category .".@4%/ then, we call 4E to the category .".@4%. :his
eliminates names as sub0ects/ to eliminate them as ob0ects, we must define pseudo1verbs "as
in pseudo1transitives% in order to allow them to ta2e 4Es as ob0ects "rather than 4s%.
9nce names are eliminated, we can have uantifiers as ob0ects, and avoid variable binding.
.ome sort of variable binding is still needed, but variables are now pseudo1variables with a
lifted type.
;. 9ne last possibility is to assimilate (uantifiers to names! we posit that the e3tensions of
4Es are characters "individual or generic, according to whether the 4E is a proper name or
a uantifier e3pression%. .till, some variable binding is reuired
VIII. Treatment of non-declaratives
(ethod of sentence radicals! one way of dealing with non1declaratives is to analy=e every sentence
into "a% a sentence radical and "b% a mood operator, as in!
.C4:C4)C
L99D .
declarative .@4 4
imperative
interrogative
be late you
.entence radicals ".% have truth values as e3tensions. :he semantics for moods consists of rules of
language use. (Again, this view was standard "and still is% in the $JM*s.-
(ethod of paraphrased performatives! instead, Lewis proposes that non1declaratives be treated as
paraphrases of performative sentences!
Ge late A > command you to be late
Are you lateN A > as2 you whether you are late
>ssues raised!
$. whether performatives are to be counted as declaratives or not "Lewis feels inclined towards
a positive answer%
&. this method calls for an assignment of truth value to non1declarative sentences, and the truth
value assigned corresponds to the truth value of the paraphrased performative, which might
seem counter1intuitive
5. performatives might not have truth value "Austin%
;. ordinary declaratives should not be treated as paraphrased performatives "otherwise, the
Carth is flat would be a paraphrase of > declare that the Carth is flat and that would assign
the wrong truth conditions%
*++&%,I-. I%,IC&S &-+*%,&,
Lodification of inde3 coordinates!
$. :o accommodate more than one designatum! the indicated ob0ect coordinate is to be thought
of as an infinite seuence of sets of indicated ob0ects
&. Erominent ob0ects coordinate! in order to account for incomplete definite descriptions
5. Delineation coordinate! in order to account for vagueness
+ostscripts to /"eneral Semantics0
*. Index and Context
>t is difficult to specify in advance all the relevant coordinates for the determination of e3tensions.
.o, instead of indices to determine e3tensions, let context do that 0ob! the features of conte3ts are
given mostly implicitly/ a conte3t is to be though of as a triple of world, time and spea2er!
c < Aw, t, aB
A conte3t is characteri=ed by the fact that the spea2er is present at the time of the conte3t at the
world of the conte3t. :his relatedness ensures that we can retrieve the features of conte3t relevant
for the determination of e3tensions.
?owever, conte3ts cannot replace indices! the evaluation of sentences containing certain operators
demand the variation of parameters in a fashion that destroys the relatedness characteristic of
conte3ts. Fe must be able to shift certain parameters while 2eeping others fi3ed.
Fe may define an inde3 as an n1tuple containing as coordinates! "a% the conte3t and "b% all the
parameters that may be shifted. An original index is one in which the shiftable parameters are
determined by the conte3t of the inde3/ a s!ifted index is one in which at least one parameter is
shifted away.
:ruth at a conte3t is truth at the original inde3/ truth at an inde3 is an au3iliary notion used to define
truth at a conte3t whenever we have shifting operators in the language.
B. Variables and binding
(H-
C. Infinitives versus clauses
.urface infinitives may be preserved! they need not be transformed into underlying sentential
clauses. :his is in line with Attitudes de dicto and de se, where Lewis defends the idea that the
ob0ects of attitudes are properties rather than propositions.