You are on page 1of 16

JOHN ETCHEMENDY

T H E D O C T R I N E OF L O G I C AS F O R M

I. T H E D O C T R I N E

There is a prima facie implausibility attaching to any theory of logical


form. To earn the name, a theory of logical form must assign distinct
forms or structures to sentences displaying, from all appearances, the
same grammatical form, simply on the basis of a difference in the logical
properties of the sentences.
So, for example, several observations indicate that the sentences
'Benjamin is healthy' and 'Someone is healthy' differ in their logical
properties. We note that the followingargument is valid:

Benjamin is healthy.
Benjamin is wealthy.
So, Benjamin is healthy and wealthy.
while observing that the next is not:

Someone is healthy.
Someone is wealthy.
So, someone is he'althy and wealthy.

Or we note that by conjoining the sentences 'Benjamin is healthy' and


'Benjamin is not healthy' we arrive at a contradiction, though the
conjunction of 'Someone is healthy' and 'Someone is not healthy' is
actually true, not logically false at all. Clearly, our two sentences display
a marked difference in their logical properties.
Now our pre-theoretic grammatical intuitions, indeed perhaps even
our post-the0retic intuitions, might urge us to consider both of these
sentences to be of subject-predicate form. But of cot/rse a theory of
logical form must beg to differ. It is standard to claim that 'Benjamin is
healthy' does display subject-predicate form, that is, logical form, while
'Someone is healthy' does not. But we might equally well assign subject-
predicate form to the latter and something else to the former, at least so
far as the present evidence is concerned. What is crucial to a theory of
logical form is just that sentences with different logical properties be
assigned different logical structures.
I will call the view that underlies this demand the doctrine of logic as

Linguistics and Philosophy 6 (1983) 319-334. 0165--0157/83/0063-0319501.60


Copyright 0 1983 by D. Reidel Publishing Co., Dordrecht, Holland, and Bostonl U.S.A.
320 JOHN ETCHEMENDY

form. That doctrine is this: Two sentences cannot differ logically if they
do not also differ formally or structurally. Should this structural
difference not show up in the grammatical form assigned to the sen-
tences, we are free to claim just one of two things: Either logical form is
independent of grammatical form, or our grammatical theory is some-
how wrong; it has not come up with the proper underlying grammatical
structures.
In this paper I will explain two reasons the doctrine of logic as form
has seemed so plausible to philosophers. Both stem from the philosophy
of logic, from the branch of philosophy concerned with, among other
things, the general nature of the logical properties and relations that hold
among sentences of a language. Both involve theories about what that
general nature is, about what distinguishes logical truths from run-of-the-
mill truths, about what distinguishes logically valid arguments from your
average string of sentences. Should either of these theories turn out
correct, the doctrine of logic as form can scarcely be denied: Sentences
with precisely the same structural properties could not possibly differ in
their logical properties. But neither view is correct, or so I will argue,
and hence in the end the doctrine of logic as form gleans no support
from the philosophy of logic.

II. PRELIMINARIES

Before turning to the two theories mentioned, let me make a few points
that I take to be obvious, points nevertheless good to have out in the
open. The first is that the doctrine of logic as form gains no particular
support from our ability to construct languages in which it holds. In an
artificially regimented first-order language, two sentences with identical
structural properties always display similar logical properties. But this
fact does not imply that the same must hold true in English. After all, in
our first-order language the logical properties go hand in hand with
surface grammatical structure, and we know that this is not the case in
English.
A related point concerns translation. In any adequate translation of
English sentences into a regimented first-order language, two English
sentences with different logical properties will go over into sentences
that differ formally. But of course if they did not, the translators would
simply have made a mistake: An adequate translation must, among other
things, preserve the logical properties of the translated sentences. And
when logic and structure coincide in the receiving language, we cannot
help but assign structurally different translations to logically different
THE DOCTRINE OF LOGIC AS FORM 321

sentences of English. But there is no reason to believe, at least not yet,


that such translation reveals any formal or structural properties of the
original sentences. Translation into French doesn't; why should trans-
lation into Fregean?
Finally, I want to point out that the formal tradition in logic, a
tradition that dates back to Aristotle, is neutral on the doctrine of logic
as form. Fluent speakers of a language have fairly strong and fairly
consistent intuitions about the logical properties and relations that hold
among sentences in their language. T h e y also have fairly strong and
fairly consistent intuitions about certain structural properties of sen-
tences, specifically about their grammatical structure. Viewed naively,
formal logic just involves a tremendously productive interplay between
these two sorts of intuition. On the one hand, the logician is primarily
interested in the logical properties of sentences and arguments, and
hence of course in the former intuitions. But it turns out that the latter
intuitions occasionally enable the logician to treat an infinite set of
sentences or an infinite set of arguments in a single blow. This occasion
arises when all the grammatically appropriate instances of a sentence or
argument form have the same logical properties.
Thus, Aristotle took the plunge into formal logic by offering us
'Barbara':

All As are Bs.


All Bs are Cs.
So, all As are Cs.

Here it seems clear - bringing to bear of course both sorts of intuition -


that any grammatically appropriate instance of the argument form will
be valid, will have a conclusion that is entailed by the premises. And
since 'Barbara' has an infinite number of instances, the study of logic
has made a great stride forward, if only as a taxonomic discipline.
The formal technique pioneered by Aristotle has proven to be an
extremely valuable tool, but it does not thereby provide evidence for the
doctrine of logic as form. For again, considered naively, there remain as
many sentences and arguments that do not submit to formal treatment
as sentences and arguments that do. Thus our logical intuitions tell us
that 'Tom is a brother' entails 'Tom is a sibling', that the latter is a
consequence of the former. But our grammatical intuitions preclude
the formal treatment of this argument, since many other arguments with
the same intuitive structure are obviously invalid. Just so, our logical
intuitions indicate that the sentences 'Benjamin is healthy' and 'Ben-
jamin is wealthy' entail 'Benjamin is healthy and wealthy'. But again
322 JOHN ETCHEMENDY

there seem to be arguments with precisely the same structure, the same
intuitive grammatical structure, which nonetheless fail to be valid. As we
saw earlier, replacing 'Benjamin' with 'Someone' produces one such
argument.
Now my point here is simple. The partial success of the formal
technique when it employs simple grammatical descriptions (e.g. Aris-
totle's sentence and argument forms) cannot itself imply the doctrine of
logic as form. It cannot imply that under some other structural descrip-
tion, the logical properties of sentences will always be distinguishable
formally. To put it even more simply, the fact that all instances of
certain argument forms are intuitively valid does not imply that all
intuitively valid arguments are, under any description, instances of
argument forms of this sort. The formal tradition in logic is neutral when
it comes to the doctrine of logic as form.

III. THE TEMPLATE ACCOUNT

We need some independent motivation for adopting the doctrine of logic


as form, for believing that sentences cannot differ logically without
differing formally or structurally. For, otherwise, our logical intuitions
can offer no grounds for thinking that 'Someone is healthy' possesses
structural properties not shared by 'Benjamin is healthy', no grounds for
revising our simple grammatical intuitions about these sentences.
Now viewed in the naive fashion I have suggested, formal logic does
not offer an explanatory account of the logical properties and relations it
studies. That is, it hazards no theory about what makes a sentence
logically true or about what makes an argument logically valid. It simply
takes our logical intuitions as primitive, attempting no explanation of
what underlies them. Nothing is said about exactly what the properties
are that we come to recognize when we judge a sentence to be logically
true or an argument to be logically valid. Let me turn now to two
accounts that go somewhat further, that say a bit more about what
underlies the intuitive logical properties. Either of these accounts, if
correct, could provide the independent motivation we are looking for,
motivation for adopting the doctrine of logic as form.
I will call the first theory the template theory of logic. The heyday of
the template theory was the late twenties and early thirties, and its
clearest spokesman was Carnap. But Carnap was by no means its only
proponent, nor was he the first or the last. The view is implicit in much
of Russell's work, arguably in Frege's, and certainly in Hilbert's and
Gentzen's.
THE DOCTRINE OF LOGIC AS FORM 323

During these years, Carnap promoted the following picture of lan-


guage: A language (or language system, as Carnap called it) consists of a
vocabulary or lexicon and three sets of rules. Two sets of rules fall
under the general heading of syntax; these are the rules of grammatical
syntax, on the one hand, and the rules of logical syntax, on the other.
The former, the rules of grammatical syntax, determine which sequences
of words constitute a sentence; the latter, the rules of logical syntax,
determine all the logical properties and relations that hold within the
language. Both are syntactic, Carnap explains, because they are
specified "without reference to sense or meaning", but concern only "the
kinds of words and the order in which they follow one a n o t h e r " J
Carnap says very little about the remaining rules, those that are not
syntactic, but presumably this third class of rules determines everything
else about the language. Carnap points out, in particular, that among
these are the rules that determine the meanings of our sentences, and
hence in what circumstances they are true or false. Thus the third class
includes, but may not be limited to, the semantic rules of the language.
How do the rules of logical syntax determine the logical properties of
sentences in the language? In exactly the same way that grammatical
rules determine which expressions are sentences: b y direct, syntactic
fiat. If we ask why a given sequence of English words is a sentence, say
'Benjamin is healthy', the answer is that it consists of a particular sort of
sequence - a singular noun phrase, followed by the word 'is' followed by
an adjective - and that a rule of English syntax proclaims any such
sequence to be a sentence. It is a sentence for no other reason than that
it fits a particular template and this template has the caption "sentence".
Just as we have " s e n t e n c e " templates, so too we have "logical truth"
templates and "logical c o n s e q u e n c e " templates. Carnap illustrates with
an instance of 'Barbara':

All eagles are birds.


All birds are beasts.
So, all eagles are beasts.

This argument is valid, Carnap says, because it displays a particular,


purely syntactic structure - it starts with the word 'all', followed by a
substantive, followed by the word 'are', etc. - and because there is a rule
of English syntax that proclaims all such sentence trios to be valid
arguments. The argument is valid for no other reason than that it fits a
given template and this template is captioned "logical consequence".
Now, it is important to realize that these rules are meant to be
genuinely syntactic: We adopt them, we learn them, and we apply them
324 JOHN ETCHEMENDY

without regard to the meanings of any expressions in the language. This


is not to say they have nothing whatsoever to do with the meanings of
items in the lexicon. Our adoption of 'Barbara' as one of our logical
templates will naturally place certain constraints on the meaning of 'all'.
But the same might conceivably be said about the rules of grammatical
syntax: Maybe the meaning of 'is' is somehow constrained by the
grammatical rules in which it is singled out.
Perhaps the best way to see how the template theorist views this
relation is by asking the following question: Wouldn't 'Barbara' cease to
be a valid argument form, if we changed the meaning of the word 'all',
say, if we let 'all' mean some? The template theorist must view this
question as oddly confused. The answer of course is "yes"; instances of
'Barbara' would not be valid if 'all' meant some. But this is because we
can only change the meaning of 'all' by altering our logical templates, by
abondoning the rules of logical syntax governing our language. To the
template theorist the question is akin to asking: Wouldn't my muscles
flex if I raised my arm? Yes indeed; but simply because I can only raise
my arm by flexing my muscles. In the genuine, physiological scheme of
things the flexing of muscles comes first; the flexing makes my arm go
up. Just so, in the genuine, linguistic scheme of things the syntactic
templates come first: They make the arguments valid, and in turn make
'all' mean all.
So goes the template theory of logic. An argument comes to be
logically valid, a sentence to be logically true, simply because it displays
a certain syntactic structure, a structure that fits a distinguished tem-
plate. And clearly if this is indeed how sentences and arguments acquire
their logical properties, the doctrine of logic as form can scarcely be
denied. For if there are no structural differences between two sentences,
they will of course fit all the same templates. And by that fact alone they
will share all their logical properties. Conversely, if we perceive an
intuitive logical distinction between two superficially similar sentences,
then we must be applying our templates to some less apparent, underly-
ing structure.
I will not spend much time arguing against the template theory. It has
pretty much gone by the board, at least among philosophers of language
and philosophically minded linguists. There are still pockets of support
among philosophers of logic, in particular, among inheritors of the
logicist and formalist traditions. But I suspect that in rejecting Carnap's
early view of language, we will encounter little resistance.
Still, a few words should be said about how the template view came to
be abandoned. First of all, there was a direct attack by Quine: 'Truth by
THE DOCTRINE OF LOGIC AS FORM 325

Convention', the earliest of the Quine classics, is aimed at precisely this


account of logic. And understood as such, the argument seems unim-
peachable. But had it not been for two other developments, both of
which preceded Quine's article, I doubt that many template theorists
would have been converted.
The other developments were G6del's completeness theorems and the
emergence of viable semantic theories for simple languages. The effect
of the first was strangely backhanded. We have all heard how the
formalist program in mathematics crumbled with the trumpeting of the
incompleteness results. Well, in his earlier paper G/Sdel inflicted similar
damage on the template account of logic, though it took some time for
the effect to be felt. The problem was not really the completeness results
themselves, but the question to which they were addressed. More
accurately, it was the phrasing of that question: Here is a formal system
of derivation, the logical templates of the language. We are now going to
find out whether the system of derivation yields all the genuine 'logical
truths, the genuinely valid arguments of the language.
Notice that on the template theory, this makes absolutely no sense.
The templates are what produce the logic of the language; the system of
derivation embodied in them is by definition complete. Now, I have
some qualms about how G6del and others have marketed these results:
Even the heading "completeness" rather than "co-extensiveness" begs
many significant philosophical questions. But what is important here is
that G6del's question seemed to everyone to make perfectly good sense.
And its being a sensible question was itself a reductio ad absurdum of
the template account of logic.
The final development, the emergence of semantics, came oddly
enough to be spearheaded by Carnap, with much help of course from
Tarski. Inspired by Tarski, Carnap the semanticist began to take a
serious look at the third class of rules, in particular, the rules that
determine in what circumstances a sentence is true or false. And when
these semantic rules are taken seriously, the rules of logical syntax begin
to look very queer indeed. If according to our semantics there are
possible circumstances in which a given sentence comes out false, then
what business have the other rules got proclaiming its logical truth? Is it
logically true in spite of the fact that it might very well be false? If on
the other hand the semantic rules show a sentence to be true in any
circumstance, say, regardless of the truth values of its components, then
what need have we for rules of logical syntax, rules that can at best
reiterate that fact? To borrow a figure, the rules of logical syntax may
spin on, but they've been disengaged from the linguistic mechanism.
326 JOHN ETCHEMENDY

The template theory of logic may no longer move us, but it would be
a mistake to underestimate its historical importance. It certainly con-
vinced many people of the truth, indeed the analytic truth, of the
doctrine of logic as form. And in so doing, the template view was largely
responsible for the emergence of the expression "logical form", for its
first appearance in the philosophical vocabulary.

IV. THE SUBSTITUTIONAL/INTERPRETATIONAL ACCOUNT

The template theory isn't the only motivation for believing the doctrine
of logic as form. At this point let us turn to the second of the two
accounts of logic that I mentioned earlier. I will call this account the
substitutional~interpretational theory of logic, or the S/I theory for short.
The account first appears in Bolzano's Wissensschaftslehre; it is implicit
in Padoa's early work on consistency and definability, and is cham-
pioned, in somewhat different forms, by Quine and Tarski. With slight
modifications Tarski's version now appears in virtually every text on
mathematical logic. Quine's version appears, naturally enough, in
Quine's logic texts, as well as countless philosophy books and articles by
various authors.
So much for its pedigree; let me present a simple, general version of
the theory. The guiding idea is that we can define logical truth in terms
of ordinary, run-of-the-mill truth, and logical validity in terms of
ordinary, run-of-the-mill truth preservation. What I mean by "truth
preservation" is just the property any argument has when one of its
premises happens to be false, or else when its conclusion happens to be
true. So this is a truth preserving argument:
Reagan is president.
So, Etchemendy is in New Jersey.
It is truth preserving simply because the conclusion is presently 2 true.
But last summer the argument was not truth preserving: Then, the
premise was true and the conclusion false. On the other hand, three
years ago the argument did preserve truth; not because I was in New
Jersey (I was no0, but because Carter was president.
Obviously there is quite a gap between run-of-the-mill truth and
logical truth, and between run-of-the-mill truth preservation and logical
validity. The S/I account attempts to close the gap with some syntactic
manipulations. Specifically, we will take logical truth to be a property
that a sentence has just in case each and every member of an associated
class of syntactically similar sentences is true. Similarly, an argument
THE DOCTRINE OF LOGIC AS FORM 327

will be dubbed logically valid, if each member of an associated class of


syntactically similar arguments preserves truth. Of course we have to
ensure that the associated classes contain the original sentences or
arguments, as the case may be. Otherwise, we might tap false sentences
as logical truths, and proclaim arguments logically valid that do not
preserve truth, arguments with true premises and a false conclusion.
So here goes: First we choose a subset f of the lexicon of our
language. Thus f is just a set of words or atomic expressions of the
language. I will call f the set of fixed terms, for reasons that will soon be
apparent. Now, given an arbitrary sentence S, here is how we specify
the associated class of sentences. A sentence will be a member of the
class associated with S, just in case it can be gotten from S by a series of
syntactically permissible substitutions, none of which disturb any fixed
term occurring in S. A substitution is syntactically permissible if it is
uniform throughout the sentence and if the expression pulled out is of
the same grammatical category as the one we put in its place. For an
arbitrary argument A, the associated class is defined similarly, though
now the substitutions must be uniform throughout the whole argument,
not just within the individual sentences.
L e t us look at some examples. Consider the sentence 'Snow is white
or snow is not white'. The first point to notice is that no matter what our
choice of i, no matter what fixed terms we settle on, the sentence will be
in its own associated class. This, simply because it is itself the result of
the null series of substitutions. Now if f contains the entire lexicon, or
even if it contains just all the words in our present sentence, then no
other sentence will be in that class. Any nontrivial replacement would
then involve pulling out a fixed term.
But as f gets smaller and smaller, the class associated with our
sentence gets, in a non-Cantorian sense, bigger and bigger. First suppose
f does not contain the expression 'snow'. Then a whole raft of sentences
gets thrown in, sentences like 'Grass is white or grass is not white' and
'Porridge is white or porridge is not white', and so forth. If in addition
the word 'white' is omitted from f, then 'Snow is green or snow is not
green' and 'Snow is tasty o, snow is not tasty' b e c o m e members of the
associated class. At this point even the double substitution 'Porridge is
tasty or porridge is not tasty' is permitted. Finally, if we omit either the
word 'or' or the word 'not' from f, we introduce sentences like 'Snow is
white and snow is not white' and maybe 'Snow is white or snow is
necessarily white'.
As I suggested earlier, we take a sentence to be logically true if its
associated class contains only true sentences, and an argument to be
328 JOHN ETCHEMENDY

valid if the associated class contains only truth preserving arguments.


Now obviously, the logical truths and the logically valid arguments will
fluctuate according to our selection of f. So for example 'Snow is white
or snow is not white' is logically true when we hold fixed 'or' and 'not',
but it is no longer logically true when we exclude 'or' from f. For then
the associated class contains the false sentence 'Snow is white and snow
is not white'.
Bolzano and others have seen this as quite an advantage. After all,
different terms have different "logics". Consider for example the fol-
lowing argument:

Benjamin knew Tom to be a policeman.


So, Benjamin believed Tom to be a policeman.

If we hold fixed the expressions 'knew' and 'believed', this argument


comes out logically valid: All the substitution instances have either a
false premise or a true conclusion. But if we allow 'knew' to be replaced,
we get substitution instances like the following:

Benjamin wanted Tom to be a policeman.


So, Benjamin believed Tom to be a policeman.

Clearly, one of these instances will have a true premise and a false
conclusion. Perhaps not this one, since Benjamin may be prone to
wishful thinking, but one very much like it. Thus, we might consider this
a valid argument of epistemic logic, the logic of knowledge and belief,
but not a valid argument of mere doxastic logic, the logic of belief alone.
I will have more to say about the selection of f in a moment. But at
this point let me pause for a brief remark about Tarski. The account I've
just sketched is more faithful to Bolzano and Quine than to Tarski,
though Tarski's basic idea is the same. Tarski was worried about the
dependence of the strictly substitutional account, the account I have just
described, on the expressive resources of the language. As it stands, a
sentence in our language can be logically true with respect to a given
selection of f, yet n o t be logically true in a simple expansion of the
language. For when we add new words to the lexicon, we may very well
add new members to the substitution class associated with our sentence.
And one of these might well turn out false.
I will not go into the details here, but the upshot of Tarski's
modification is just that the logical properties are made to persist
through certain simple expansions of the language. Briefly, we can think
of the bodily replacement of the word 'white' by the word 'green' as a
way of finding out what would happen to our sentence, to its truth value,
THE DOCTRINE OF LOGIC AS FORM 329

if the predicate 'is white' meant is green, that is, if the extension of 'is
white' encompassed all and only the green things in the world. Well, now
suppose our language had no predicate expession, simple or complex,
whose extension comprised just my little finger and a certain cumquat in
Brazil° Tarski's idea is here to skip the actual demand for an expression
like 'is green', an expression we can substitute for 'is white' but which
extends to my little finger, to the cumquat, and no further. We can just
pretend 'is white' itself fills the bill, and on that interpretation my finger
will be "white", the cumquat will be "white", but snow will not. In this
way our determination of the logical properties cannot be affected by
introducing a new word, say 'fingquat', that actually has precisely the
right meaning.
Now, back to the problem of selecting the fixed terms. It has often
been remarked that with various choices of f, the S/I account yields an
unintuitive crop of logical truths and logically valid arguments. This is
most obvious if we fix all the words in the language. For then a n y true
sentence will qualify as logically true, and a n y truth preserving
argument will qualify as logically valid. On that selection of f, the
associated classes have shrunk to their smallest size, to singletons
including just the sentence or argument itself. On the other hand, if f
contains no words at all, if no terms are held fixed, then in general n o
sentence will be logically true, no argument logically valid: It would take
real stamina to survive the resulting orgy of substitution. At points in
between we sometimes get plausible results, sometimes not.
This problem is a nagging one, but it needn't be taken too seriously. It
is generally thought that the account should be supplemented with a
specification of the distinctly "logical" terms of the language - basically
the ones singled out in Carnap's rules of logical syntax - and that these
should be the only ones held fixed, the only ones included in f. For
example, we will certainly include 'or', 'not', and 'some'; perhaps be
daring and add 'believes' or 'necessarily'; and maybe even terms like
'very' and 'quickly'. Of course the more expressions we include, the
harder it will be to say what they have in common. But for now I will
just say that the logical terms are the ones for which the substitutional
test w o r k s , that is, the ones whose inclusion in f produces an intuitively
plausible harvest of logical truths and logically valid arguments.
Naturally this can't be the final word. We will eventually need some
explanation of why our shared logical intuitions seem keyed to this
motley collection of terms, while those s a m e intuitions rebel against the
s a m e test when other choices of f are made, when f contains expressions
like 'snow' and 'white'.
330 JOHN ETCHEMENDY

In any event, suppose that with an appropriate supplement about


selecting f, the S/I account is right. That is, suppose the account actually
captures those properties underlying our pre-theoretic logical intuitions,
those properties that lead us to judge some sentences logically true and
others not, to view some arguments as valid and others as not. Then it
seems again we have sustained the doctrine of logic as form. For to be
logically true is just to display a certain skeletal structure, a structure
articulated by the peculiarly logical terms and which, fleshed out, never
fails to yield a true sentence. If our intuitions recognize a logical truth
and yet the skeletal structure is not apparent, then it must of course be
present at some deeper level.
Now I believe that the present account of the logical properties is
incorrect; incorrect, that is, in the sense just described. It bears no clear
relation to our intuitive notions of logical validity or logical truth, no
relation to what the logician initially set out to study. And because of
this, the S/I account can offer no support for the doctrine of logic as
form, for the belief that intuitions about the logical properties of a sen-
tence are at base intuitions about hidden formal or structural properties.
There are three arguments to the contrary, arguments that the account
is indeed, as Tarski puts it, "close in essentials" to our common
conception of the logical properties. All three are, I believe, seriously
misguided. But they have each been influential in gaining acceptance for
the account.
The first argument appears in two related forms. In the first, it appeals
to our intuitions about the modal properties of logically valid arguments,
and in the second, to intuitions about their epistemological properties.
Both Bolzano and Tarski explicitly offer the modal version as
justification for the S/I definition of validity? For various reasons, Quine
never actually gives the argument as explicit justification for the ac-
count, but he comes about as close to offering the epistemological
version as is otherwise possible. 4 I will call the argument Bolzano's
fallacy, in deference to historical precedence.
Let me quote from Tarski. After explaining the definition of logical
validity, or actually of the logical consequence relation, he provides the
following intuitive motivation for the account:
It s e e m s to me that e v e r y o n e who u n d e r s t a n d s the content of the above definition m u s t
admit that it agrees quite well with c o m m o n usage. This b e c o m e s still clearer from its
various c o n s e q u e n c e s . In particular, it can be proved, on the basis of this definition, that
every consequence of true sentences must be trueJ

Using my terminology, Tarski's point is this: On the basis of the S/I


definition, we can actually prove that it is impossible for a valid
THE DOCTRINE OF LOGIC AS FORM 331

argument to have true premises and a false conclusion. And of course


this sits nicely with our intuitive understanding of validity: A valid
argument is one in which the premises could not possibly be true while
the conclusion is false.
To see the fallacy here, we need only consider the proof of the
impossibility. We have suggested a precise criterion for validity: An
argument is valid with respect to some choice of f just in case every
member of its associated class preserves truth. Now suppose we have a
valid argument with true premises. We know, as a quite general fact,
that every argument is a member of its own associated class. Con-
sequently, if our argument has a false conclusion, then there is at least
one member of the associated class that does not preserve truth,
namely, the original argument itself.
The fallacy involves a misplaced modality. If it weren't fallacious, it
would indeed be a surprising result. For recall that any truth preserving
argument is valid on s o m e choice of f, and no peculiarities about f were
mentioned in the proof. We have proven, for arbitrary A and f, that the
following are jointly impossible:
(i) A is logically valid with respect to f.
(ii) A has true premises.
(iii) A has a false conclusion.

Now to fix ideas, imagine that A is the following truth preserving


argument:
(A) Reagan is president.
So, Etchemendy is in New Jersey.
If f contains all the component expressions in A, then this argument
satisfies our criterion for validity. In other words, clause (i) is true. But
as we can see, the truth of (i) and the joint impossibility of (i), (ii) and
(iii) do not guarantee the joint impossibility of (ii) and (iii). A n d that is
w h a t we really wanted. A valid argument is one in which it is impossible
for the premises to be true while the conclusion is false. Bolzano's
fallacy only demonstrates that arguments cannot simultaneously
preserve truth, have true premises, and have a false conclusion. But that
is obviously true of any argument, intuitively valid or not; preserving
truth just m e a n s not having a true premise and a false conclusion. From
there the modal property attributed to the allegedly valid argument is
constructed out of whole cloth. 6
I will just skim over the epistemological version of the fallacy, since
by now it should be quite easy to see through. We begin by noting that it
332 JOHN ETCHEMENDY

is irrational to believe the premises of a valid argument and simul-


taneously to deny the conclusion. That is the intuitive property we are
after. Next we note that it would indeed be irrational to believe that an
argument meets the suggested criterion for validity, yet also that it has
true premises and a false conclusion. For then we would have to believe
both that it preserves truth and that it does not preserve truth. But of
course this fact is entirely irrelevant to the intuitive property we were
originally aiming for. This fact holds of any argument, intuitively valid
or not.
Let me emphasize two things about the above discussion. First of all,
my appeal to the fact that argument A meets the criterion of validity on
some selections of f was simply intended to emphasize the fallacy. The
argument is equally fallacious on any choice of fixed terms, even when
we restrict f to the "logical" terms, most narrowly construed. Second, I
do not deny that on some selections of f, with certain syntactically
simple languages, the account produces arguments with the right modal
and epistemological properties. But that is only because, on occasion,
the definition does manage to ferret out a class of genuinely valid
arguments. It is not because the S/I account captures any modal or
epistemological properties underlying our pre-theoretic intuitions.
Now I said that there were three reasons for imagining the S/I account
to be "close in essentials" to our intuitive conception of the logical
properties. The second appeals to those who believe logical truth to be a
type of analytic truth. Suppose we think, not implausibly, that a logical
truth is a sentence which is true solely in virtue of the meanings of its
component logical terms. On this view, we include the traditional "logi-
cal" expressions in f in order to ensure that our results are sensitive to
their meanings, while we exclude the traditional "nonlogicar' expres-
sions to cancel out undue influence from their meanings.
If only things were so simple. But of course the S/I account is
completely incapable of distinguishing between matters of meaning and
matters of fact. And consequently, it is incapable of testing for analy-
ticity, for sentences whose truth depends only on meaning, whether that
dependence extends to a sentence's entire complement of atomic
expressions or merely to some "logical" subset. To see this we need
only consider what happens when we extend f to include the entire
lexicon. According to the present suggestion our test should then yield
the class of all analytic truths, sentences true in virtue of the meanings
of both the logical and the nonlogical terms. But of course our test then
welcomes with open arms sentences like 'Snow is white' and 'Etche-
mendy is in New Jersey'. To be sure, it now includes among the "logical
T H E D O C T R I N E OF L O G I C AS F O R M 333

truths" all analytic truths, but only because they happen to be true. The
account cannot reliably distinguish analytic truths from truths influenced
by mere matters of fact.
Let me explain that I am here expressing a partial agreement with
Quine. Quine realizes that the S/I account has nothing to do with
analyticity, that it is not sensitive to the sorts of distinctions needed to
make sense of this notion. But I am not thereby expressing agreement
with his other views about analyticity. There is no need to go into that
now; for present purposes it suffices that the S]I account cannot guaran-
tee a harvest of analytic truths. It may sometimes produce a plausible
collection of such truths, but this is no more surprising than the
production, on occasion, of genuine logical truths and genuinely valid
arguments.
The final reason offered in favor of the S/I account, in support of its
agreement with our intuitive understanding of the logical properties, is
simple. I quote from Tarski. The account shows the logical consequence
relation "to be uniquely determined by the form of the sentences
between which it holds". Tarski is right: The account does agree with
that aspect of the "common conception", the common conception held
by the template theorists to which he was addressing his article. But of
course this just brings us full circle. We will find no support here for the
doctrine of logic as form.

V. CONCLUSION

Let me briefly review the bidding, and offer a few words in conclusion.
The doctrine of logic as form requires support from somewhere other
than the traditional conduct of formal logic. Historically it has derived
much of that support from general views in the philosophy of logic,
views about the nature of the logical properties themselves. The tem-
plate account sees these properties arising from the conventional adop-
tion of syntactical rules. When we learn a language, we in effect
internalize a formal system of derivation; our intuitions about the logical
properties are then seen as intuitions about formal derivability. The
substitutional-interpretational theory replaces this purely syntactic view
of ~ogic with a quasi-semantic account. It is semantic in that it presup-
poses the notion of truth. But it retains some of the crucial syntactic
characteristics of the earlier view. It defines the logical properties in
terms of certain syntactic features of a sentence plus the single semantic
notion of truth, the simple run-of-the-mill variety.
I have argued that there is no reason to think that either of these
334 JOHN ETCHEMENDY

theories is a'correct account of the logical properties, of the intuitive


logical properties we recognize among sentences of our native language.
Consequently, I see no reason, no reason stemming from the philosophy
of logic, to adopt the doctrine of logic as form. And in turn I find no
reason for seeing 'Someone is healthy' as structurally or formally
different from 'Benjamin is healthy'. Obviously, they contain different
lexical items. And just as obviously, the semantic characteristics of
these expressions function very differently; otherwise the sentences
would not differ logically.
Now these semantic differences might be reflected in the structural
properties of some underlying level of representation, perhaps the level
we subject to semantic evaluation rules. But then again it might not.
Montague in effect showed us how a semantic theory might account for
the differences in the logical properties of these sentences without those
differences being reflected in any underlying structural divergence. At
least this is one way of looking at his treatment of quantifiers. But I am
not particularly endorsing Montague's approach. I simply fail to see why
formal or structural properties need be thought central to a semantic
account of the logical properties of a language. So far I have seen little
reason to think that form has much to do with logic at all.

Princeton University

NOTES

L Philosophy and Logical Syntax, Kegan Paul, London (1935), p. 39.


2 At time of writing.
3 See Bolzano, Theory of Science, Reidel, Dordrecht (1973), p. 206; and Tarski, 'On the
Concept of Logical Consequence', in Logic, Semantics, Metamathematics, Oxford Uni-
versity Press, London (1956), p. 417.
4 I am thinking, in particular, of the first two paragraphs on page 4 of Methods of Logic,
Holt, Rinehart & Winston, New York (1972).
Logic, Semantics, Metamathematics, p. 417, my emphasis.
6 It might be argued that Tarski never meant to make the fallacious inference from the
truth of (i) and the joint impossibility of (i), (ii) and (iii) to the joint impossibility of (ii) and
(iii). But if not, it is difficult to see why he thought the observation had any bearing on the
agreement of his definition with "common usage". In any event, the strictly historical
question is secondary to our present concerns.

You might also like