Mill maintains that we are always, though not necessarily consciously,
applying his canons in daily life and in the courts of law. Thus, to illustrate
the second canon he says, ‘When a man is shot through the heart, it is by
this method we know that it was the gunshot which killed him: for he was
in the fullness of life immediately before, all circumstances being the same,
except the wound.’
Mill’s methods of agreement and disagreement are a sophistication of
Bacon’s tables of presence and absence.1 Like Bacon’s, Mill’s methods seem
to assume the constancy of general laws. Mill says explicitly, ‘The propos-
ition that the course of Nature is uniform, is the fundamental principle, or
general axiom, of Induction.’ But where does this general axiom come
from? As a thoroughgoing empiricist, Mill treats it as being itself a gener-
alization from experience: it would be rash, he says, to assume that the law
of causation applied on distant stars. But if this very general principle is the
basis of induction, it is difficult to see how it can itself be established by
induction. But then Mill was prepared to affirm that not only the funda-
mental laws of physics, but those of arithmetic and logic, including the
very principle of non-contradiction itself, were nothing more than very
well-confirmed generalizations from experience.2
Frege’s Refoundation of Logic
On these matters Frege occupied the opposite pole from Mill. While for
Mill propositions of every kind were known a posteriori, for Frege arith-
metic no less than logic was not only a priori but also analytic. In order to
establish this, Frege had to investigate and systematize logic to a degree that
neither Mill nor any of his predecessors had achieved. He organized logic in
a wholly new way, and became in effect the second founder of the
discipline first established by Aristotle.
One way to define logic is to say that it is the discipline that sorts out
good inferences from bad. In the centuries preceding Frege the most
important part of logic had been the study of the validity and invalidity
of a particular form of inference, namely the syllogism. Elaborate rules had
been drawn up to distinguish between valid inferences such as
1 See vol. III, p. 31. 2 See Ch. 6 below.
LOGIC
100
All Germans are Europeans.
Some Germans are blonde.
Therefore, Some Europeans are blonde.
and invalid inferences such as
All cows are mammals.
Some mammals are quadrupeds.
Therefore, All cows are quadrupeds.
Though both these inferences have true conclusions, only the first is valid,
that is to say, only the first is an inference of a form that will never lead
from true premisses to a false conclusion.
Syllogistic, in fact, covers only a small proportion of the forms of valid
reasoning. In Anthony Trollope’s The Prime Minister the Duchess of Omnium
is anxious to place a favourite of hers as Member of Parliament for
the borough of Silverbridge, which has traditionally been in the gift of the
Dukes of Omnium. He tells us that she ‘had a little syllogism in her head as
to the Duke ruling the borough, the Duke’s wife ruling the Duke, and
therefore the Duke’s wife ruling the borough’. The Duchess’s reasoning is
perfectly valid, but it is not a syllogism, and cannot be formulated as one.
This is because her reasoning depends on the fact that ‘rules’ is a transitive
relation (if A rules B and B rules C, then A does indeed rule C), while
syllogistic is a system designed to deal only with subject–predicate
sentences, and not rich enough to cope with relational statements.
A further weakness of syllogistic was that it could not cope with
inferences in which words like ‘all’ or ‘some’ occurred not in the subject
place but somewhere in the grammatical predicate. The rules would not
determine the validity of inferences that contained premisses such as ‘All
politicians tell some lies’ or ‘Nobody can speak every language’ in cases
where the inference turned on the word ‘some’ in the first sentence or the
word ‘every’ in the second.
Frege devised a system to overcome these difficulties, which he
expounded first in his Begriffsschrift. The first step was to replace the gram-
matical notions of subject and predicate with new logical notions, which Frege
called ‘argument’ and ‘function’. In the sentence ‘Wellington defeated
Napoleon’ grammarians would say (or used to say) that ‘Wellington’ was
the subject and ‘defeated Napoleon’ the predicate. Frege’s introduction of
LOGIC
101
Trollope’s Lady Glencora Palliser ruled not just one but two Dukes of Omnium. Here,
in Millais’ illustration to Phineas Finn, she establishes her dominion over the elder Duke
by presenting him with a grandson.
102
LOGIC
the notions of argument and function offers a more flexible method of analysing
the sentence.
This is how it works. Suppose that we take our sentence ‘Wellington
defeated Napoleon’ and put into it, in place of the name ‘Napoleon’, the
name ‘Nelson’. Clearly this alters the content of the sentence, and indeed it
turns it from a true sentence into a false sentence. We can think of the
sentence as in this way consisting of a constant component, ‘Wellington
defeated . . . ’, and a replaceable element, ‘Napoleon’. Frege calls the first,
fixed component a function, and the second component the argument of
the function. The sentence ‘Wellington defeated Napoleon’ is, as Frege
would put it, the value of the function ‘Wellington defeated . . . ’ for the
argument ‘Napoleon’ and the sentence ‘Wellington defeated Nelson’ is the
value of the same function for the argument ‘Nelson’.
We could also analyse the sentence in a different way. ‘Wellington
defeated Napoleon’ is also the value of the function ‘ . . . defeated Napoleon’
for the argument ‘Wellington’. We can go further, and say that the
sentence is the value of the function ‘ . . . defeated . . . ’ for the arguments
‘Wellington’ and ‘Napoleon’ (taken in that order). In Frege’s terminology,
‘Wellington defeated . . . ’ and ‘ . . . defeated Napoleon’ are functions of a
single argument; ‘ . . . defeated . . . ’ is a function of two arguments.3
It will be seen that in comparison with the subject–predicate distinction
the function–argument dichotomy provides a much more flexible method
of bringing out logically relevant similarities between sentences. Subject–
predicate analysis is sufficient to mark the similarity between ‘Caesar
conquered Gaul’ and ‘Caesar defeated Pompey’, but it is blind to the
similarity between ‘Caesar conquered Gaul’ and ‘Pompey avoided Gaul’.
This becomes a matter of logical importance when we deal with sentences
such as those occurring in syllogisms that contain not proper names like
‘Caesar’ and ‘Gaul’, but quantified expressions such as ‘all Romans’ or
‘some province’.
Having introduced these notions of function and argument, Frege’s next
step is to introduce a new notation to express the kind of generality expressed
by a word like ‘all’ no matter where it occurs in a sentence. If ‘Socrates is
3 As I have explained them above, following Begriffsschrift, functions and arguments and their
values are all bits of language: names and sentences, with or without gaps. In his later writings
Frege applied the notions more often not to linguistic items, but to the items that language is
used to express and talk about. I will discuss this in the chapter on metaphysics (Ch. 7).
LOGIC
103
mortal’ is a true sentence, we can say that the function ‘ . . . is mortal’ holds
true for the argument ‘Socrates’. To express generality we need a symbol to
indicate that a certain function holds true no matter what its argument is.
Adapting the notation that Frege introduced, logicians write
(x)(x is mortal)
to signify that no matter what name is attached as an argument to the
function ‘ . . . is mortal’, the function holds true. The notation can be read
as ‘For all x, x is mortal’ and it is equivalent to the statement that
everything whatever is mortal.
This notation for generality can be applied in all the different ways in
which sentences can be analysed into function and argument. Thus
‘(x)(God is greater than x)’ is equivalent to ‘God is greater than everything’.
It can be combined with a sign for negation (‘�’) to produce notations
equivalent to sentences containing ‘no’ and ‘none’. Thus ‘(x)� (x is
immortal)’ ¼ ‘For all x, it is not the case that x is immortal’ ¼ ‘Nothing
is immortal’. To render a sentence containing expressions like ‘some’ Frege
exploited the equivalence, long accepted by logicians, between (for
example) ‘Some Romans were cowards’ and ‘Not all Romans were not
cowards’. Thus ‘Some things are mortal’ ¼ ‘It is not the case that nothing
is mortal’ ¼ ‘� (x)�(x is mortal)’. For convenience his followers used, for
‘some’, a sign ‘(Ex)’ as equivalent to ‘� (x)�’. Frege’s notation, and its
abbreviation, can be used to make statements about the existence of things
of different kinds. ‘(Ex)(x is a horse)’, for instance, is tantamount to ‘There
are horses’ (provided, as Frege notes, that this sentence is understood as
covering also the case where there is only one horse).
Frege believed that objects of all kinds were nameable—numbers, for
instance, were named by numerals—and the argument places in his logical
notation can be filled with the name of anything whatever. Consequently
‘(x)(x is mortal)’ means not just that everyone is mortal, but that every-
thing whatever is mortal. So understood, it is a false proposition, because,
for instance, the number ten is not mortal.
It is rare, in fact, for us to want to make statements of such unrestricted
generality. It is much more common for us to want to say that everything
of a certain kind has a certain property, or that everything that has a certain
given property also has a certain other property. ‘All men are mortal’ or
‘What goes up must come down’ are examples of typical universal
LOGIC
104
sentences of ordinary language. In order to express such sentences in
Frege’s system one must graft his predicate calculus (the theory of quan-
tifiers such as ‘some’ and ‘all’) on to a propositional calculus (the theory of
connectives between sentences, such as ‘if ’ and ‘and’).
In Frege’s system of propositional logic the most important element
is a sign for conditionality, roughly corresponding to ‘if ’ in ordinary
language. The Stoic logician Philo, in ancient times, had defined ‘If p
then q’ by saying that it was a proposition that was false in the case in
which p was true and q false, and true in the three other possible cases.4
Frege defined his sign for conditionality (which we may render ‘!’) in a
similar manner. He warned that it did not altogether correspond to
‘if . . . then’ in ordinary language. If we take ‘p ! q’ as equivalent to ‘If p
then q’ then propositions such as ‘If the sun is shining, 3� 7 ¼ 21’ and ‘If
perpetual motion is possible, then pigs can fly’ turn out true—simply
because the consequent of the first proposition is true, and the antecedent
of the second proposition is false. ‘If ’ behaves differently in ordinary
language; the use of it that is closest to ‘!’ is in sentences such as ‘If
those curtains match that sofa, then I’m a Dutchman’. Frege’s sign can be
looked on as a stripped-down version of the word ‘if ’, designed to capture
just that aspect of its meaning that is necessary for the formulation of
rigorous proofs containing it.
In Frege’s terminology, ‘ . . .! . . . ’ is a function that takes sentences as
its arguments: its values, too, are sentences. Whether the sentences that are
its values (sentences of the form ‘p! q’) are true or false will depend only
on whether the sentences that are its arguments (‘p’ and ‘q’) are true or
false. We may call functions of this kind ‘truth-functions’. The conditional
is not the only truth-function in Frege’s system. So too is negation,
represented by the sign ‘�’, since a negated sentence is true just in case
the sentence negated is false, and vice versa.
With the aid of these two symbols Frege built up a complete system of
propositional logic, deriving all the truths of that logic from a limited set of
primitive truths or axioms, such as ‘(q ! p)! ( �p ! �q)’ and
‘��p! p’. Connectives other than ‘if ’, such as ‘and’ and ‘or’, are defined
in terms of conditionality and negation. Thus, ‘�q ! p’ rules out the case
in which p is false and �q is true: it means that p and q are not both false,
4 See vol. I, p. 138.
LOGIC
105
and therefore is equivalent to ‘p or q’ (in modern symbols, ‘p V q’). ‘p and q’
(‘p & q’), on the other hand, is rendered by Frege as ‘� ðq! �pÞ’. As Frege
realized, a different system would be possible in which conjunction was
primitive, and conditionality was defined in terms of conjunction and
negation. But in logic, he maintained, deduction is more important than
conjunction, and that is why ‘if ’ and not ‘and’ is taken as primitive.
Earlier logicians had drawn up a number of rules of inference, rules for
passing from one proposition to another. One of the best known was called
modus ponens: ‘From ‘‘p’’ and ‘‘If p then q’’ infer ‘‘q’’ ’. In his system Frege
claims to prove all the laws of logic using this as a single rule of inference.
The other rules are either axioms of his system or theorems proved from
them. Thus the rule traditionally called contraposition, which allows the
inference from
If John is snoring, John is asleep
to
If John is not asleep, John is not snoring,
is justified by the first of the axioms quoted above.
When we put together Frege’s propositional calculus and his predicate
calculus we can symbolize the universal sentences of ordinary language,
making use of both the sign of generality and the sign of conditionality.
The expression
(x)(Fx ! Gx)
can be read
For all x, if Fx then Gx,
which means that whatever x may be, if ‘Fx’ is true then ‘Gx’ is true.
If we substitute ‘is a man’ for ‘F ’ and ‘is mortal’ for ‘G ’ then we obtain
‘For all x, if x is a man, x is mortal’, which is what Frege offers as the
translation of ‘All men are mortal’. The contradictory of this, ‘Some men
are not mortal’, comes out as ‘�(x)(x is a man ! x is mortal)’ and its
contrary, ‘No man is mortal’, comes out as ‘(x)(x is a man!� x is
mortal)’. By the use of these translations, Frege is able to prove as part of
his system theorems corresponding to the entire corpus of Aristotelian
syllogistic.
LOGIC
106
Administrator
线条
Administrator
线条
Frege’s logical calculus is not just more systematic than Aristotle’s; it is
also more comprehensive. His symbolism is able, for instance, to mark the
difference between
Every boy loves some girl ¼ (x)(x is a boy! Ey(y is a girl & x loves y) )
and the apparently similar (but much less plausible) passive version of the
sentence
Some girl is loved by every boy ¼ (Ey(y is a girl & (x)(x is a boy ! x
loves y) ).
Aristotelian logicians in earlier ages had sought in vain to find a simple and
conspicuous way of bringing out such differences of meaning in ambiguous
sentences of ordinary language. A final subtlety of Frege’s system must be
mentioned. The sentence ‘Socrates is mortal’, as we have seen, can be
analysed as having ‘Socrates’ for argument, and ‘ . . . is mortal’ as function.
But the function ‘ . . . is mortal’ can itself be regarded as an argument of a
different function, a function operating at a higher level. This is what
happens when we complete the function ‘ . . . is mortal’ not with a deter-
minate argument, but with a quantifier, as in ‘(x)(x is mortal)’. The
quantifier ‘(x)(x . . . )’ can then be regarded as a second-level function of
the first-level function ‘ . . . is mortal’. The initial function, Frege always
emphasizes, is incomplete; but it may be completed in two ways, either by
having an argument inserted in its argument place, or by itself becoming
the argument of a second-level function. This is what happens when the
ellipsis in ‘ . . . is mortal’ is filled with a quantifier such as ‘Everything’.
Induction and Abduction in Peirce
A number of Frege’s innovations in logic occurred, quite independently, to
C. S. Peirce; but Peirce was never able to incorporate his results into a
rigorous system, much less to publish them in a definitive form. Peirce’s
importance in the history of logic derives rather from his investigations
into the structure of scientific inquiry. Deductive logic assists us in
organizing our knowledge; but the kind of reasoning that extends our
knowledge (‘ampliative inference’ as Peirce calls it) is of three kinds:
induction, hypothesis, and analogy. All of these inferences, Peirce claimed,
LOGIC
107
depend essentially on sampling. Any account, therefore, of non-deductive
inference must be related to the mathematical theory of probability
(EWP 177).
Scientists frame hypotheses, make predictions on the bases of these
hypotheses, and then make observations with a view to confirming or
refuting their hypotheses. These three stages of inquiry are called by
Peirce abduction, deduction, and induction. In the abductive phase the
inquirer selects a theory for consideration. In the deductive phase
he formulates a method to test it. In the inductive phase he evaluates
the results of the test.
How does a scientist decide which hypotheses are worth inductive testing? Indef-
initely many different theories might explain the phenomena he wishes to
investigate. If he is not to waste his time, his energy, and his research funding,
the scientist needs some guidance about which theories to explore. This guidance
is given by the rules of the logic of abduction. The theory must, if true, be
genuinely explanatory; it must be empirically testable; it should be simple and
natural and cohere with existing knowledge, though not necessarily with our
subjective opinions about antecedent likelihood. (P 7.220–1)
Rules of abduction, however, do not by themselves explain the success of
scientists in their choice of hypotheses. We have to believe that in their
investigation of nature they are assisted by nature herself.
Science presupposes that we have a capacity for ‘guessing’ right. We shall do better
to abandon the whole attempt to learn the truth . . . unless we can trust to the
human mind’s having such a power of guessing right that before very many
Modern symbolic logic no longer uses the actual symbol system of its founder Frege,
which was difficult to print. The illustration shows the pattern, in his notation, for
deriving results such as ‘‘If this ostrich is a bird and cannot fly, it follows that some
birds cannot fly’’.
LOGIC
108
hypotheses shall have been tried, intelligent guessing may be expected to lead us to
the one which will support all tests. (P 6.530)
This trust has to be presupposed at the outset, even though it may rest on
no evidence. But in fact the history of science shows such trust to be well
founded: ‘it has seldom been necessary to try more than two or three
hypotheses made by clear genius before the right one was found’ (P 7.220)
Once the theory has been chosen, abduction is succeeded by deduction.
Consequences are derived from the hypothesis, experimental predictions
that is, which will come out true if the hypothesis is correct. In deduction,
Peirce maintained, the mind is under the dominion of habit: a general idea
will suggest a particular case. It is by verifying or falsifying the predictions of
the particular instantiations that the scientist will confirm, or as the case
may be refute, the hypothesis under test.
It is induction that is the all-important element in the testing, and
induction is essentially a matter of sampling.
Suppose a ship arrives in Liverpool laden with wheat in bulk. Suppose that by some
machinery the whole cargo be stirred up with great thoroughness. Suppose that
twenty-seven thimblefuls be taken equally from the forward, midships, and aft
parts, from the starboard, center and larboard parts, and from the top, half depth
and lower parts of her hold, and