Sentences, speech acts, and thoughts are alike in that they have propositional
content. Thus, `La neige est blanche' means that snow is white;
in uttering `Over my dead body', Betty was letting you know that the
probability of her going out with you wasn't very high; and one of your
mental states is a belief that Palermo is south of Rome. Because
sentences, speech acts, and thoughts all have propositional content, one
can't sensibly limit one's semantic interests to the philosophy of language,
and my focus today will be on issues that cut across both the philosophy
of language and the philosophy of mind.
As we all know, philosophical investigations can't be neatly contained within the traditional curriculum headings of philosophy of language, epistemology, metaphysics, and so on, and one of the things that makes the theory of content exciting is how quickly it leads to some hairy issues in the area of metaphysics we call ontology. For consideration of sentences that ascribe propositional content has led Frege and many others to suppose that the truth of these sentences requires the existence of abstract entities called `propositions', and many other philosophers have found propositions so mysterious or otherwise repugnant that they've gone to great lengths to try to avoid being committed to their existence. Quine once got so lathered about propositions that he called them creatures of darkness (if Quine were more widely read there would no doubt be a cult rock band called The Propositions). Still, there is a case for the existence of propositions, and we may put it in the following way. It's a two-part case, and it's first part is a case for the claim that that-clauses, such as `that Fido is a dog' in `Ralph believes that Fido is a dog', are referential singular terms, where to say that an expression t in a sentence S(t) is a referential singular term is to say, at least to a first approximation, that t stands for an object x such that the sentence S(t) is true just in case the predicate S( ) is true of x. So, for example, we say that `Fido' in `Fido is a dog' is a referential singular term because there is a certain dog, Fido, such that the sentence is true just in case the predicate `is a dog' is true of Fido, and we say that `the woman over there' in an utterance of `Lester loves the woman over there' is a referential singular term because that utterance is true just in case the predicate `Lester loves _' is true of the indicated woman. Now, the reason for taking that-clauses to be referential singular terms is quite simply that it's apparently the best way to account for the evident validity of such inferences as the following:
Lester believes that eating liver increases sexual potency, and so does
So, there is something that they both believeto wit, that eating liver
increases sexual potency.
Lester believes that eating liver increases sexual potency.
That eating liver increases sexual potency is Carlotta's theory.
So, Lester believes Carlotta's theory.
Lester believes everything that Carlotta says.
Carlotta says that eating liver increases sexual potency.
So, Lester believes that eating liver increases sexual potency.
We can readily account for the validity of these inferences if we suppose
that the that-clause contained in them, `that eating liver increases sexual
potency', is a referential singular term, and it's not clear that we can
account for it otherwise.1
In connection with this, notice how in the premise `That eating liver increases
sexual potency is Carlotta's theory' the that-clause occupies the grammatical
subject position, and how we can substitute the singular term `Carlotta's
theory' for it in the sentence `Lester believes that eating liver increases
sexual potency'. Despite this evidence, there do exist various attempts
to account for the validity, or seeming validity, of these inferences without
the assumption that that-clauses are referential singular terms. There's
even my own labored attempt in my book Remnants of Meaning to use
non-objectual quantification in aid of the denial project. My present view,
however, is that all these attempts fail; that-clauses really do typically
function as referential singular terms.2
At any rate, that they do so function will be a working assumption of the
rest of this talk.
Very well, that-clauses refer. The next question, thenturning to the second part of the two-part case for propositionsis: To what do they refer? What are the referents of that-clauses and, thereby, the things we believe and assert? That is easy to answer: the referent of `that eating liver increases sexual potency' is that eating liver increases sexual potency; that eating liver increases sexual potency is precisely the referent of the that-clause singular term. To be sure, to be sure, I can hear you muttering, but what manner of thing is this thing, that eating liver increases sexual potency, which is the referent of the that-clause singular term? Happily, there are a number of things we can say in response to this question right off the tops of our heads. First, that eating liver increases sexual potency is abstract, or immaterial. It doesn't occupy space and has no physical properties at all. Second, it's mind and language independent in at least two senses: it exists in possible worlds in which there are neither thinkers nor speakers, and although it can be expressed by a sentence of every language, it itself belongs to no language; that eating liver increases sexual potency isn't Japanese, Italian, or English. Third, it has a truth condition: that eating liver increases sexual potency is true iff eating liver increases sexual potency. Fourth, it has its truth condition essentially; it's a necessary truth that that eating liver increases sexual potency is true iff eating liver increases sexual potency. This is in contrast to the sentence `Eating liver increases sexual potency', which, while also true iff eating liver increases sexual potency, has its truth condition only contingently on our actual linguistic practices. Had our use of language been different, it might have had a different truth condition or none at all. Fifth, and finally, that eating liver increases sexual potency has its truth condition absolutely, without relativization to anything. The contrast is again with the sentence `Eating liver increases sexual potency', which has its truth condition only in English or among us, and may have a different truth condition in some other language or among some other population of speakers. But that eating liver increases sexual potency has its truth condition everywhere and everywhen. In short, the referents of that-clauses, and therewith the contents of our speech acts and propositional attitudes, are what philosophers call propositions: abstract, mind- and language-independent entities that have truth conditions, and have their truth conditions both essentially and absolutely.
As you may have heard, there are those who agree that that-clauses refer but who deny that they refer to propositions. These guys attempt to make do with linguistic surrogates, and I have time only to tell you that in my opinion they fail. At any rate, another working hypothesis of this talk will be that propositional attitudes really are relations to propositions, to those propositions to which that-clauses refer. If we're stuck with propositions, then we should try to demystify their existence, and I'll have something to say about this later.
Our working hypothesis is that propositions, being the referents of that-clauses,
are the things we mean and believe. This still leaves plenty of room for
propositionalists to disagree among themselves about what else is true of
propositions. It is arguable that the most plausible further account of
propositions remains Frege's. At all events, the Fregean position is still
a dominant position, and it enjoys its most recent important development
in Christopher Peacocke's A Study of Concepts.3
The Fregean position may be characterized in the following way.
Pretend that the Superman fiction is fact and consider these two sentences:
Lois Lane believes that Superman eats groundhogs.
Lois believes that Clark Kent eats woodchucks.
The Fregean, quite sensibly, holds two things initially. First, she holds
that these sentences may differ in truth-value notwithstanding that Superman
= Clark Kent and the property of being a groundhog = the property of being
a woodchuck. Second, she holds that these sentences have the form they appear
to have: `believes' occurs in them as standing for a two-place relation
holding between believers and the propositions they believe, while the two
singular terms in each sentence stand for alleged terms of that relation.4 In other words, the
Fregean would claim that these two sentences enjoy the following form-revealing
B(Lois, the proposition that Superman eats groundhogs)
B(Lois, the proposition that Clark Kent eats woodchucks)
These two sensible initial assumptions commit the Fregean to a simple
and plausible account of how it is that our two belief sentences can differ
in truth-valueviz., the two that-clauses refer to distinct propositions.
The proposition that Superman eats groundhogs is not the same proposition
as the proposition that Clark Kent eats woodchucks.
The Fregean next explains why these two propositions are different: they have different constituents. Propositions, for the Fregean, are structured entities whose basic constituents are what we may call propositional building blocks (although, I warn you, I'm about to call them three other things as well!). Our two propositions are different because they're built from different building blocks. These propositional building blocks are the references words have in that-clauses. Let me explain. Some singular terms are semantically simple in that their references aren't determined by the references or semantic values of any of their parts. Proper names, such as `Fido', are like that. Other singular terms are semantically complex in that their references are determined by their syntax and the references of their constituent expressions. `The capital of Italy' owes its reference to its syntax and the fact that `Italy' refers to Italy and `the capital of' refers to that function which maps countries onto their capitals. Evidently, that-clauses are semantically complex singular terms, given that they are singular terms. According to the Fregean, propositional building blocks are the references words have in that-clauses (which references, we can already deduce, won't be references words have outside of that-clauses).
We still haven't asked what sorts of things the Fregean takes propositional building blocksthe references words have in that-clausesto be, but before turning to that important question, let's notice three further things we can say about Fregean propositional building blocks. The first is a merely verbal point, but verbal points can be interesting. The word `concept' is used in philosophy as a term of art, although it's unfortunately used as more than one term of art. But a dominant use, highly congenial to the Fregean, is that concepts are constituents of the contents of thoughts. On this common way of speaking, propositional building blocks, whatever they turn out to be, are concepts. The second further thing I wanted to say about Fregean propositional building blocks is that, subject to a certain qualification, they're also word meanings. We get this result in the following simple way. Consider the true sentence
`Superman eats groundhogs' means that Superman eats groundhogs.
Now, it's a platitude that the meaning of a word is its contribution
to the meanings of the sentences in which it occurs, and, as the displayed
truth illustrates, the contribution that the words in `Superman eats groundhogs'
contribute to the meaning of the sentence are precisely the references those
words have in the displayed that-clause. The qualification to which I alluded
has to do with the fact that we can't regard every indicative sentence as
meaning a complete proposition (e.g., the sentence type `She's now there'
expresses no complete), and even if we could, the Fregean would be constrained
to say that words can make different contributions to the meaning of an
utterance in different contexts of utterance. But it won't hurt us to ignore
this complication for now and to appreciate what for the Fregean is the
near truth that propositional building blocks are both concepts and meanings.
The third thing I wanted to say at this point about Fregean propositional building blocks is that, for the Fregean, the truth-value of a proposition is determined in a certain way by "semantic values" of its constituent concepts. We've already noticed that for the Fregean propositional building blocks can't be the objects and properties our beliefs are about. If they were, then the proposition that Superman eats groundhogs would be identical to the proposition that Clark Kent eats woodchucks. Still, propositional building blocks must bear some important relation to those ordinary objects and properties, otherwise there would be nothing to make our beliefs about those things. The Fregean suggests an accommodation of this constraint via her use of the metaphor of a mode of presentation. Before proceeding, though, let's pause to keep from getting overwhelmed by a surfeit of labels. According to the Fregean, the propositions we believe and assert are structured entities, and we're already calling their constituents propositional building blocks, concepts, and meanings. The Fregean also calls them modes of presentation. Her point is that while propositional building blocks aren't the objects and properties our beliefs are about, they are modes of presentation of those things. Modes of presentation are propositional building blocks, and the things of which they are modes of presentation are their "semantic values." Use of the metaphor of a mode of presentation affords a neat way of spelling out what the Fregean means in saying that the truth-value of a proposition is determined by, or is a function of, semantic values of its constituent concepts. At a certain level of analysis, the Fregean can represent all propositions as being of the form
<<m1,..., mn>, mn>
where <m1,..., mn> is an n-ary sequence of
modes of presentation of things of any kind and mn is a mode of presentation
of an n-ary relation (one-place relations are properties). Then the
sense in which, for the Fregean, the truth-value of a proposition is determined
by semantic values of its constituent concepts is given by the following
definition of truth and falsity for propositions (`
_ !v' means "there is a unique
v such that"):
<<m1,..., mn>, mn>
iff _ !x1,...,xn
_ ! n(m1,...,
mn are modes of presentation of x1,..., xn respectively
& mn is a mode of presentation of n & <x1,...,
xn> instantiates n);
false iff _ !x1,...,xn
_ ! n(m1,...,
mn are modes of presentation of x1,..., xn respectively
& mn is a mode of presentation of n & <x1,...,
xn> doesn't instantiate n);
neither true nor false iff ~ _ !x1,...,xn _
! n(m1,..., mn are
modes of presentation of x1,..., xn respectively & mn
is a mode of presentation of n)
Thus, the proposition that Fido is a dog may be represented as
where mf is a mode of presentation of Fido and md is a
mode of presentation of doghood (and where we allow ourselves to drop the
brackets for unit sequences). This proposition is therefore true just in
case Fido instantiates doghood, i.e., just in case Fido is a dog. The proposition
that Fido loves Gina may be represented as
<<mf, mg>, ml>
where mf, mg, and ml are modes of presentation respectively
of Fido, Gina, and the love relation, and the proposition is therefore true
just in case <Fido, Gina> instantiates the love relation; i.e., just
in case Fido loves Gina. And the complex proposition that roses are red
and violates are blue may be represented as
<<mr, mv>, mconj>
where mr, mv, and mconj are modes of presentation
respectively of the proposition that roses are red, the proposition that
violets are blue, and the conjunction relation, and the proposition is therefore
true just in case <the proposition that roses are red, the proposition
that violets are blue> instantiate the conjunction relation; i.e., just
in case roses are red and violets are blue.
This brings us to the $64 question: What are Fregean modes of
presentation? What sorts of things satisfy the foregoing characterization
of modes of presentation? As so far characterized, our understanding of
the notion of a mode of presentation is simply that they are whatever things
play such-and-such theoretical role, if indeed there are things that play
that role. What I'm now asking is what things, if any, play that
role. To ask this question, we know, is the same as to ask what, for the
Fregean, are concepts or meanings.
One strategy for answering this question would be to try to give an account of modes of presentation which satisfies what I've elsewhere called the intrinsic-description constraint.5 According to this constraint, if a thing is a mode of presentationif, that is, it plays the mode-of-presentation rolethen it must be intrinsically identifiable in a way that does not describe it as a mode of presentation or as a possible mode of presentation. If a thing is a mode of presentation, then it must be intrinsically identifiable as some other kind of thing. Denying this constraint is apt to seem tantamount to introducing the notion of a gene as whatever plays such-and-such role in the transmission of inheritable characteristics and then insisting that the things which play that role enjoy no more intrinsic characterization than `things that play such-and-such role in the transmission of inheritable characteristics'. To deny the intrinsic-description constraint is to insist that propositional building blocks enjoy no more intrinsic characterization than that they are propositional building blocks. So the intrinsic-description constraint is hardly unmotivated. At the same time, I'm pretty confident that the Fregean won't be able to satisfy the constraint. As I've argued elsewhere,6 when you go through the list of candidates for modes of presentation that satisfy the constraint, you can find pretty good reasons for striking each candidate off the list. I don't have time to review the case for this now, but let me briefly mention one example. Many have read Frege, rightly or wrongly, as suggesting that modes of presentation are uniqueness properties of the form the property of uniquely having the property that is, that property a thing has when it has the property and nothing else has . Whether or not Frege actually held this, it's clear that Bertrand Russell once did. But the view is hopeless. Since properties need modes of presentation, you couldn't think of something under a mode of presentation without having a distinct mode of presentation for that mode of presentation, thereby setting off a self-refuting regress.
If the Fregean theory is to have a chance, it must deny the intrinsic-description constraint. This is recognized by the Fregean Christopher Peacocke, who in his A Study of Concepts offers an account of conceptsi.e., modes of presentationwhich doesn't satisfy the constraint, and then explicitly argues that "the intrinsic-description constraint is quite generally false of abstract objects, and its falsity for concepts or modes of presentation is a special case of this general falsity."7 I'm now inclined to agree with Peacocke that there is a reasonable way of denying the intrinsic-description constraint, although this way isn't exactly Peacocke's and although I believe that the theory of modes of presentation Peacocke offers is less than fully correct. But there isn't time to discuss Peacocke's interesting views on these matters, and in the time remaining I'm going to lay out, as simply and as baldly as I can, what I think must be said if one's to have modes of presentation that don't satisfy the intrinsic-description constraint. The Fregean, I'll propose, needs to identify modes of presentation, her propositional constituents, with what I'm soon to call pleonastic concepts. Let me explain.
It's my view that certain kinds of objects are in a sense language-created "linguistic posits," hypostatizations of certain linguistic practices, even though, in another sense, they enjoy a mind- and language-independent existence.8 Propositions are linguistic posits in this sense. They are mind and language independent in the two senses already mentioned: propositions would have existed no matter what linguistic or conceptual practices we employed, and one and the same proposition can be grasped and expressed by speakers off different languages. They are linguistic posits, hypostatizations of the linguistic practices that introduce propositions into our ontology, in a sense that includes the following claims.9
1. Linguistic posits enter our language via
what may be called something-from-nothing transformations. These
are trivial transformations that take one from a sentence in which no singular
term refers to the linguistic posit to a sentence that does contain such
a singular term. Thus, from
Fido is a dog,
whose only singular term is `Fido', we can
infer its pleonastic equivalent
That Fido is a dog is true,
or, more colloquially,
It's true that Fido is a dog,
which contains the singular term `that Fido
is a dog' whose referent is the proposition that Fido is a dog. It's because
of our ability to move back and forth between any sentence `S' and
its pleonastic equivalent `That S is true' that we have the well-known
truth schema for propositions:
(The proposition) that S is true iff
(The existence of something-from-nothing transformations
doesn't imply that all references to linguistic posits can be paraphrased
away. There's no paraphrasing away the that-clause in `Ralph believes that
Fido is a dog'. But the use of that-clauses in these constructions is parasitic
on its uses in the hypostatizing something-from-nothing transformations.)
have knowledge of linguistic posits, one merely needs to be party to the
linguistic practices by which they are introduced, and there is no other
way of gaining knowledge of them. Imagine a possible world, ß , exactly like the actual world, , except that in ß we don't have linguistic practices that license the
formation of that-clauses; we in that world lack the practice that allows
us to transform the sentence `S' into the singular term `that S'.
It follows from our actual practices governing that-clauses, our practices
in , that propositions exist in ß , but, lacking the concept
of a proposition, we in ß would lack all knowledge
of them. What would it take to bring us in ß up to epistemological
snuff with us in ? It's simple: what we'd need to do, and all that we'd
need to do, is adopt the proposition-introducing language games we actually
play. We certainly couldn't become aware of the existence of things that
are not linguistic posits in this way. You couldn't, for example,
become aware of trees simply by introducing talk of trees. You would
first have to discover trees and then introduce talk of trees.
An important corollary of this point is that no substantial relationcertainly
no causal relationmust obtain between us and propositions in order for
us to refer to them and to have knowledge about them; it's enough that
we indulge in a certain linguistic practice, a certain language game.
3. There is nothing more to the nature of linguistic posits
than is determined by the hypostatizing linguistic practices by which linguistic
posits are introduced. What we can learn about them is what our linguistic
practices license us to learn about them. The essences of things that enjoy
the highest degree of independence from our linguistic and conceptual practicestrees
and electrons, for examplecan be discovered by a posteriori, scientific
investigation, but the essence of linguistic posits can't be discovered
in any such way. Whatever belongs to their essence can be read off the
something-from-nothing linguistic practices that posit them in our ontology.
As Mark Johnston aptly puts it, linguistic posits have no "hidden
and substantial nature for a theory to uncover."10
4. A corollary of the nature-determination point, but one worth mentioning
separately, concerns the individuation of linguistic posits.11
Linguistic posits needn't have nontrivial criteria of individuation, and
this is true of propositions. At the same time, we are able to make confident
judgments of nonidentity, for we have criteria for ascribing properties
to propositions. Just think of all the differences between the proposition
that Harry Truman had toes and the proposition that the Pope is Catholic.
In a sense, what "makes" the proposition that Superman eats groundhogs
distinct from the proposition that Clark Kent eats woodchucks is that someone
can believe one of them without believing the other. I'm not trying to
say something trivial. If Al kissed Betty but did not kiss Carla, then
Betty _ Carla, and
if Lois believes that Superman eats groundhogs but doesn't believe that
Clark Kent eats woodchucks, then the proposition that Superman eats groundhogs
_ the proposition
that Clark Kent eats woodchucks. However, the important difference between
the two kinds of cases is as follows. Suppose you were wondering whether
Betty and Carla were the same person. You could hardly hope to determine
that they were not by determining that Al kissed Betty but not Carla. Imagine
saying, "Well, I saw that Al kissed Betty and that he didn't kiss
Carla, so I knew Betty wasn't Carla." The absurdity, of course, is
that you couldn't rationally conclude in this case that Al kissed Betty
but not Carla unless you already knew that Betty was not Carla. But you
can determine that Lois believes that Superman eats groundhogs and
that she doesn't believe that Clark Kent eats woodchucks without any prior
opinion, as it were, about the identity or difference of the two propositions.
It's because the criteria for the truth of these belief statements are
independent in this way that we can first determine the different
truth-values of `Lois believes that Superman eats groundhogs' and `Lois
believes that Clark Kent eats woodchucks' and then conclude that
the proposition that Superman eats groundhogs _
the proposition that Clark Kent eats woodchucks.
In short, our criteria for determining the truth of belief statements doesn't
require a prior individuation of the propositions involved; rather, our
individuation of propositions is culled entirely from prior criteria for
assessing belief statements.
What I now want tentatively to propose is that we might construe ourselves
as having linguistic practices that afford us a conception of concepts as
linguistic posits in pretty much the way propositions are. Pleonastic
propositions is what I like to call propositions conceived as linguistic
posits, so the view of concepts I'm proposing may be called a conception
of pleonastic concepts. The primary linguistic practice I have in
mind is the one that licenses inferences like the following:
Giorgio believes that Satan lurks everywhere.
So, Giorgio believes something involving his concept of Satan.
To be sure, we don't seem to have very rich practices involving the word
`concept', but the foregoing doesn't seem too strained, and is a version
of a something-from-nothing introductory practice in that `Giorgio believes
that Satan lurks everywhere' contains no singular term that explicitly refers
to the concept of Satan.12
Given that we have such a practice, concepts will be linguistic posits in
just the way propositions are. Among other things, this means that our only
basis for knowing that the concept of X _ the concept of Y is that one can believe that
X without believing that Y . Criteria for ascribing beliefs
come first, and from them we cull our ways of individuating concepts. Better
yet, criteria for ascribing beliefs come first, and from them we cull our
ways of individuating propositions, and from them we cull our ways of individuating
There's more that we can say about this deflationary conception of concepts that we find riding piggyback on our deflationary conception of propositions, although time constraints prevent me from giving a complete elaboration.
a. Pleonastic concepts, like the pleonastic
propositions from which they derive, don't satisfy the intrinsic-description
constraint. The only intrinsic characterization of the concept of Satan
is `the concept of Satan'.
b. As we'll presently observe, the context
`the concept of ' may be construed as admitting of two readings, but the
one that's been in play so far is intentional in two senses. First, the
concept of X may exist even though X doesn't exist. Giorgio's
concept of Satan exists even though, presumably, Satan doesn't exist. Second,
it may be that the concept of X _ the concept of Y even though X = Y.
For example, even though Superman = Clark Kent, Lois's concept of Superman
_ her concept
of Clark Kent.
c. We noticed that for the Fregean propositional building blocks must
determine semantic values that can be used to determine the truth-values
of the propositions those building blocks help to build. A major task for
the Fregean is to give a theory of this determination, a theory that tells
us how the constituents of propositions determine their semantic values.
When the Fregean calls these building blocks concepts, the task
is to say what it is for a particular object or property to fall under
a concept. Our pleonastic conception of concepts affords us an easy way
of characterizing the determination relation. We've noticed that since
the position of `X' in `the concept of X' is intentional
(at least as so far characterized), we can't say that a particular object
or property falls under the concept of X just in case that object
or property = X. However, we can say that every instance of the
If X exists, then X falls under the concept of X
is analytic in just the way that every instance of the schema
The proposition that S is true iff S
is analytic. These schemas are trivial consequences of our ways of introducing
talk of concepts and propositions respectively. Since we can speak of a
thing as falling under the concept of X, we may easily introduce
a nonintentional way of understanding `a concept of y', the lower-case
letter being our way of marking the difference. We can say that, for any
y, the concept of X is a concept of y just in case
y falls under the concept of X.
d. The way pleonastic concepts are obtained from that-clauses suggests
we can construe them as the references words have in that-clauses, for
each word in a that-clause determines a concept involved in the proposition
to which the whole that-clause refers.
In this way we arrive at a singularly deflationary version of Frege's
theory of propositions which, since I'm already speaking of pleonastic propositions
and concepts, we might as well call pleonastic Fregeanism. It's the
view that the propositions we believe and assert are pleonastic propositions
composed of pleonastic concepts. It's the view that Fregean modes of presentation
are pleonastic concepts. One reason that this is not full-blown Fregeanism
is that for Frege what explains the fact that the proposition that
Superman eats groundhogs _ the
proposition that Clark Kent eats woodchucks is that they have different
constituents. But on our notion of pleonastic concepts, the fact that these
two propositions have different constituents is entirely derivative on their
being different propositions, which in turn, I've suggested, is entirely
derivative on the conceptually prior fact that `Lois believes that Superman
eats groundhogs' and `Lois believes that Clark Kent eats woodchucks' may
differ in truth value.
I now want to conclude by bringing pleonastic Fregeanism to bear on the
vexing question of compositional semantics. For simplicity, let's ignore
indexicality, ambiguity, vagueness, and grammatical moods other than the
indicative. Then we may say that:
A compositional meaning theory for a Language L is a finitely
statable theory of L that associates compositional rules with the
basic syntactic structures of L and assigns meanings to the primitive
words of L in such a way as to generate for each sentence of L
a truth of the form
means p in L.
Thus, a compositional meaning theory for French stated in English would
entail the statement
`La neige est blanche' means in French that snow is white.
If we assume, with Frege, that sentence meanings are structured propositions,
then the meanings assigned to the words of L will be propositional
building blocks. Now, it's very widely held that each natural language has
a compositional meaning theory. It's held that each natural language has
such a compositional semantics because the hypothesis that it does is needed
to explain certain things. Among the things theorists have claimed we need
a compositional meaning theory to explain are:
(i) Our ability to understand utterances of novel sentences, sentences
we've never heard before. It's argued that what explains this is that although
the sentence was novel, its words and structure weren't. We could figure
out the meaning of the novel sentence because we already knew the meanings
of its parts and knew a rule for putting those meanings together to get
the meaning of the sentence. And this story, it's further held, makes sense
only on the assumption that languages have compositional meaning theories.
(ii) The productivity of language: the fact that each of infinitely
many sentences has its own unique meaning.
(iii) The systematicity of language: the fact that each word makes a
uniform contribution to the meanings of the infinitely many sentences in
which it occurs.
(iv) The productivity of thought: our ability in principle to entertain
any one of an infinity of propositions.
(v) The systematicity of thought: the fact that the ability to entertain
any one thought carries with it the ability to entertain numerous permutations
of that thought. For example, someone who can entertain the thought that
John loves Mary can also entertain the thought that Mary loves John.
For the past ten years, I have been arguing that languages neither have
nor need compositional semantics. I now think I may have been only half
right. If we avail ourselves of pleonastic concepts as word meanings, then
we can allow that languages have compositional meaning theoriespleonastic
compositional meaning theories. So I may have been wrong to think that languages
don't have compositional meaning theories. But pleonastic compositional
meaning theories won't explain any of the things theorists have thought
they needed compositional semantics in order to explain. This is because
of the way pleonastic concepts are abstractions from the already determined
pleonastic propositions and not genuine building blocks of them. As I've
the issue about whether languages have compositional meaning theories boils
to down the issue of whether propositions are compositionally determined.
Propositions are compositionally determined if there's a finitely definable
function from sequences of propositional building blocks onto the propositions
they build. If there are pleonastic concepts, then propositions are compositionally
determined. But only in a very Pickwickian sense, because there's no identifying
the building blocks until you already have the propositions they build.
Pleonastic concepts are an epiphenomenon of that-clauses, and they contribute
nothing to the mechanisms whereby that-clauses determine propositions. Propositional
building blocks would explain what needs to be explained only if they really
were an essential part of the mechanism that explained the business of that-clauses.
Well, if pleonastic compositional semantics is the best compositional semantics we can have, and if it doesn't explain language understanding and the different versions of productivity and systematicity, then what does explain those things? Our understanding of natural languages may be quite easy to explain without a compositional semantics: as I argued in Remnants of Meaning, we don't need a compositional semantics to explain our understanding of Mentalese, our internal system of mental representation, and natural language understanding can be understood wholly in terms of certain "translation" functions that map spoken utterances onto meaning-equivalent Mentalese sentences but do so wholly on the basis of the syntactic features of the sentences on which they operate.14 We would still need to explain the productivity and systematicity of Mentalese, and therewith, directly or indirectly, the productivity and systematicity of thought and natural languages, but most of that can be done via what I've elsewhere called compositional supervenience theories,15 theories that are compositional but don't imply compositional semantics. And so it goes, one thing leading to another, and where will it ever end? Not here, not today.