T H E L OGI C MANU AL

for Introduction to Logic

����/����

Volker Halbach

Oxford
�th August ����

�is text is to be used by candidates

in their �rst year in ����/����. �e
set text for Literae Humaniores stu-
dents sitting Moderations in ���� is
Hodges’s Logic.
 Content

� Sets, Relations, and Arguments �

�.� Sets �
�.� Binary Relations �
�.� Functions ��
�.� Non-Binary Relations ��
�.� Arguments, Validity and Contradiction ��
�.� Syntax, Semantics and Pragmatics ��

� Syntax and Semantics of Propositional Logic ��

�.� Quotation ��
�.� �e Syntax of the Language of Propositional Logic ��
�.� Rules for Dropping Brackets ��
�.� �e Semantics of Propositional Logic ��

� Formalisation in Propositional Logic ��

�.� Truth-Functionality ��
�.� Logical Form ��
�.� From Logical Form to Formal Language ��
�.� Ambiguity ��
�.� �e Standard Connectives ��
�.� Natural Language and Propositional Logic ��

� �e Syntax of Predicate Logic ��

�.� Predicates and Quanti�cation ��
�.� �e Sentences of L� ��
�.� Free and Bound Occurrences of Variables ��

© Volker Halbach ����/����

�.� Notational Conventions ��
�.� Formalisation ��

� �e Semantics of Predicate Logic ��

�.� Structures ��
�.� Truth ��
�.� Validity, Logical Truths, and Contradictions ���
�.� Counterexamples ���

� Natural Deduction ���

�.� Propositional Logic ���
�.� Predicate Logic ���

� Formalisation in Predicate Logic ���

�.� Adequacy ���
�.� Ambiguity ���
�.� Extensionality ���
�.� Predicate Logic and Arguments in English ���

� Identity and De�nite Descriptions ���

�.� Qualitative and Numerical Identity ���
�.� �e Syntax of L= ���
�.� Semantics ���
�.� Proof Rules for Identity ���
�.� Uses of identity ���
�.� Identity as a Logical Constant ���

Natural Deduction Rules ���

�������

�e Logic Manual is a relatively brief introduction to logic. I have tried to

focus on the core topics and have neglected some issues that are covered
in more comprehensive books such as Forbes (����), Guttenplan (����),
Hodges (����), Smith (����), and Tennant (����). In particular, I have
tried not to include material that is inessential to Preliminary Examina-
tions and Moderations in Oxford. For various topics, I could not resist
adding footnotes o�ering extra information to the curious reader.
Logic is usually taught in one term. Consequently, I have divided the
text into eight chapters:
�. Sets, Relations, and Arguments
�. Syntax and Semantics of Propositional Logic
�. Formalisation in Propositional Logic
�. �e Syntax of Predicate Logic
�. �e Semantics of Predicate Logic
�. Natural Deduction
�. Formalisation in Predicate Logic
�. Identity and De�nite Descriptions
If the reader wishes to read selectively, chapters �–� constitute a self-con-
tained part, to which Section �.� (Natural Deduction for propositional
logic) can be added; and chapters �–� yield an introduction to predicate
logic without identity.
I have set the core de�nitions, explanations, and results in italics like
this. �is might be useful for revision and for �nding important passages
more quickly.
In some cases, the end of an example or a proof is marked by a square
�. �e word ‘i�’ is short for ‘if and only if’.
I have written an Exercises Booklet that can be used in conjunction
with this Logic Manual. It is available from WebLearn. �ere also some
additional teaching materials may be found such as further examples of
proofs in the system of Natural Deduction.
I am indebted to colleagues for discussions and comments on previous
versions of the text. In particular, I would like to thank Stephen Blamey,
Paolo Crivelli, Geo�rey Ferrari, Lindsay Judson, Ofra Magidor, David
McCarty, Peter Millican, Alexander Paseau, Annamaria Schiaparelli, Se-
bastian Sequoiah-Grayson, Mark �akkar, Gabriel Uzquiano, and David
Wiggins. I am especially grateful to Jane Friedman and Christopher von
Bülow for their help in preparing the �nal version of the Manual.
 � Sets, Relations, and Arguments

�.� ����

Set theory is employed in many disciplines. As such, some acquaintance

with the most basic notions of set theory will be useful not only in logic,
but also in other areas that rely on formal methods. Set theory is a vast
area of mathematical research and of signi�cant philosophical interest.
For the purposes of this book, the reader only needs to know a fragment
of the fundamentals of set theory.�
A set is a collection of objects. �ese objects may be concrete objects
such as persons, cars and planets or mathematical objects such as numbers
or other sets.
Sets are identical if and only if they have the same members. �erefore,
the set of all animals with kidneys and the set of all animals with a heart
are identical, because exactly those animals that have kidneys also have
a heart and vice versa.� In contrast, the property of having a heart is
usually distinguished from the property of having kidneys, although both
properties apply to the same objects.
�at a is an element of the set M can be expressed symbolically by

� �ere are various mathematical introductions to set theory such as Devlin (����),
Moschovakis (����) or the more elementary Halmos (����). In contrast to rigorous
expositions of set theory, I will not proceed axiomatically here.

� I have added this footnote because there are regularly protests with respect to this
example. For this example, only complete and healthy animals are being considered. I
have been told that planarians (a type of �atworms) are an exception to the heart–kidney
rule, so, for the sake of the example, I should exclude them as well.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments �

writing ‘a ∈ M’. If a is an element of M, one also says that a is in M or

that M contains a.
�ere is exactly one set that contains no elements, namely, the empty
set �. Obviously, there is only one empty set, because all sets containing
no elements contain the same elements, namely none.
�ere are various ways to denote sets.
One can write down names of the elements, or other designations of
the elements, and enclose this list in curly brackets.
�e set {London, Munich}, for instance, has exactly two cities as its
elements. �e set {Munich, London} has the same elements. �erefore,
the sets are identical, that is:

{London, Munich} = {Munich, London}.

�us, if a set is speci�ed by including names for the elements in curly

brackets, the order of the names between the brackets does not matter.
�e set {the capital of England, Munich} is again the same set be-
cause ‘the capital of England’ is just another way of designating London.
{London, Munich, the capital of England} is still the same set: adding
another name for London, namely, ‘the capital of England’, does not add
a further element to {London, Munich}.
�is method of designating sets has its limitations: sometimes one
lacks names for the elements. �e method will also fail for sets with
in�nitely many or even just impractically many elements.
Above I have designated a set by the phrase ‘the set of all animals
with a heart’. One can also use the following semi-formal expression to
designate this set:

{ x ∶ x is an animal with a heart }

�is is read as ‘the set of all animals with a heart’. Similarly, { x ∶ x is a

natural number bigger than � } is the set of natural numbers bigger than �,
and { x ∶ x is blue all over or x is red all over } is the set of all objects that

© Volker Halbach ����/����

 � Sets, Relations, and Arguments �

are blue all over and all objects that are red all over.�

�.� ������ ���������

�e expression ‘is a tiger’ applies to some objects, but not to others. �ere
is a set of all objects to which it applies, namely the set { x ∶ x is a tiger }
containing all tigers and no other objects. �e expression ‘is a bigger city
than’, in contrast, does not apply to single objects; rather it relates two
objects. It applies to London and Munich (in this order), for instance,
because London is a bigger city than Munich. One can also say that the
expression ‘is a bigger city than’ applies to pairs of objects. �e set of all
pairs to which the expression ‘is a bigger city than’ applies is called ‘the
binary relation of being a bigger city than’ or simply ‘the relation of being
a bigger city than’.� �is relation contains all pairs with objects d and e
such that d is a bigger city than e.�
However, these pairs cannot be understood simply as the sets {d, e},
such that d is a bigger city than e, because elements of a set are not ordered
by the set: as pointed out above, the set {London, Munich} is the same
set as {Munich, London}. So a set with two elements does not have a
�rst or second element. Since London is bigger than Munich, but not
vice versa, only the pair with London as �rst component and Munich as

� �e assumption that any description of this kind actually describes a set is problematic.
�e so-called Russell paradox imposes some limitations on what sets one can postulate.
See Exercise �.�.

� By the quali�cation ‘binary’ one distinguishes relations applying to pairs from relations
applying to triples and strings of more objects. I will return to non-binary relations in
Section �.�.

� O�en philosophers do not identify relations with sets of pairs. On their terminology
relations need to be distinguished from sets of ordered pairs in the same way properties
need to be distinguished from sets (see footnote �). In set theory, however, it is common
to refer to sets of ordered pairs as binary relations and I shall follow this usage here.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments �

second component should be in the relation of being a bigger city than,

but not the pair with Munich as �rst component and London as second
component.
�erefore, so-called ordered-pairs are used in set theory. �ey are
di�erent from sets with two elements. Ordered pairs, in contrast to sets
with two elements, have a �rst and a second component (and no further
components). �e ordered pair �London, Munich� has London as its �rst
component and Munich as its second. �Munich, London� is a di�erent
ordered pair, because the two ordered pairs di�er in both their �rst and
second components.� More formally, an ordered pair �d, e� is identical
with � f , g� if and only if d = f and e = g. �e ordered pair �the largest
city in Bavaria, the largest city in the UK� is the same ordered pair as
�Munich, London�, because they coincide in their �rst and in their second
component. An ordered pair can have the same object as �rst and second
component: �London, London�, for instance, has London as its �rst and
second component. �Munich, London� and �London, London� are two
di�erent ordered pairs, because they di�er in their �rst components. Since
I will not be dealing with other pairs, I will o�en drop the quali�cation
‘ordered’ from ‘ordered pair’.
���������� �.�. A set is a binary relation if and only if it contains only
ordered pairs.
According to the de�nition, a set is a binary relation if it does not
contain anything that is not an ordered pair. Since the empty set � does
not contain anything, it does not contain anything that is not an ordered
pair. �erefore, the empty set is a binary relation.
�e binary relation of being a bigger city than, that is, the relation
that is satis�ed by objects d and e if and only if d is a bigger city than e is
the following set:

� Using a nice trick, one can dispense with ordered pairs by de�ning the ordered pair
�d, e� as {{d}, {d, e}}. �e trick will not be used here.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments �

��London, Munich�, � London, Oxford�, �Munich, Oxford�,

�Paris, Munich�, . . . �
In the following de�nition I will classify binary relations. Later, I shall
illustrate the de�nitions by examples. Here, and in the following, I shall
use ‘i� ’ as an abbreviation for ‘if and only if ’.
���������� �.�. A binary relation R is
(i) re�exive on a set S i� for all d in S the pair �d, d� is an element of R;
(ii) symmetric i� for all d, e: if �d, e� ∈ R then �e, d� ∈ R;
(iii) asymmetric i� for no d, e: �d, e� ∈ R and �e, d� ∈ R;
(iv) antisymmetric i� for no two distinct d, e: �d, e� ∈ R and �e, d� ∈ R;
(v) transitive i� for all d, e, f : if �d, e� ∈ R and �e, f � ∈ R, then also
�d, f � ∈ R;
(vi) an equivalence relation on S i� R is re�exive on S, symmetric and
transitive.
In the following I shall occasionally drop the quali�cation ‘binary’.
As long as they are not too complicated, relations and their properties
– such as re�exivity and symmetry – can be visualised by diagrams. For
every component of an ordered pair in the relation, one writes exactly
one name (or other designation) in the diagram. �e ordered pairs in the
relation are then represented by arrows. For instance, the relation
{�France, Italy�, �Italy, Austria�, �France, France�,
�Italy, Italy�, �Austria, Austria�}
has the following diagram:
⌃ ⌃
France
II Austria
u:
II uu
II uu
II uu
I$ uu
Italy
V

�e arrow from ‘France’ to ‘Italy’ corresponds to the pair �France,

Italy�, and the arrow from ‘Italy’ to ‘Austria’ corresponds to the pair �Italy,

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

Austria�. �e three loops in the diagram correspond to the three pairs

�France, France�, �Italy, Italy�, �Austria, Austria�.
Since ‘France’, ‘Italy’ and ‘Austria’ all have such a loop attached to
them, the relation is re�exive on the set {France, Italy, Austria}. �e
relation is not re�exive on the set {France, Italy, Austria, Spain}, because
the pair �Spain, Spain� is not in the relation.
�e relation is not transitive. For transitivity it is required that if there
is an arrow from a point d to a point e and one from e to f in the diagram,
then there must be a ‘shortcut’, that is, a (direct) arrow from d to f . In the
diagram above there is an arrow from ‘France’ to ‘Italy’ and an arrow from
‘Italy’ to ‘Austria’, but there is no arrow from ‘France’ to ‘Austria’. Hence
the relation is not transitive. If the additional pair �France, Austria� were
added to the relation, then a transitive relation would be obtained.
If a relation is symmetric, then there are no ‘one-way’ arrows. �at
is, if there is an arrow from d to e, then there must be an arrow back to
d from e. �e relation above is not symmetric. For instance, the pair
�France, Italy� is in the relation, but not the pair �Italy, France�. �at is,
in the diagram there is an arrow from ‘France’ to ‘Italy’ but no arrow back
from ‘Italy’ to ‘France’.
�e relation is also not asymmetric. If a relation is asymmetric and
�d, e� is in the relation, then �e, d� cannot be in the relation. �e pair
�France, France� is in the relation, but the pair with its elements reversed,
that is, �France, France� (which happens to be the same ordered pair
again), is in the relation as well, thereby violating the condition for asym-
metry.
Generally, in the diagram of an asymmetric relation there are only
‘one-way’ arrows: there is never an arrow from an object d to an object e
and then an arrow back from e to d. �is implies that in the diagram of
an asymmetric relation there cannot be any loops, because if there is an
arrow from d to d, there is also, trivially, an arrow ‘back’ from d to d: the
very same arrow.
�e relation in the diagram on page � is antisymmetric: in an anti-
symmetric relation there must not be two di�erent objects with arrows

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

in both directions between them. �us, antisymmetry is the same as

asymmetry except that in an antisymmetric relation elements may have
loops attached to them. In the above diagram there are objects with loops,
but no two di�erent objects with arrows in both directions between them.
�erefore, the relation is antisymmetric.
I turn to another example, a relation with the following diagram:
⌃ ⌃ ⌃
Mars^ Pluto 5 Venus

v
Mercury
V

�is relation is re�exive on the set {Mars, Pluto, Venus, Mercury}; it is

also symmetric. It fails to be transitive since direct arrows are missing,
for instance, from Mars to Venus. �e relation is not asymmetric or
antisymmetric since there are pairs of objects – such as Mars and Mercury
– that have arrows going back and forth between them .
Both relations discussed so far are not equivalence relations, as they
are not transitive.
�e relation � has some peculiar properties: its diagram is empty. It is
re�exive on the empty set �, but on no other set. It is symmetric, as there
is no arrow for which there is not any arrow in the opposite direction.
But it is also asymmetric and antisymmetric because there is no arrow
for which there is an arrow in the opposite direction. � is also transitive.
Consequently, � is an equivalence relation on �.
�e relation with the diagram below is not re�exive on the set with
the two elements Ponte Vecchio and Ei�el Tower, because there is no loop
attached to ‘Ei�el Tower’.
. ⌃
Ei�el Tower m Ponte Vecchio
�e relation is symmetric, but not asymmetric or antisymmetric. It is
also not transitive: there is an arrow from ‘Ei�el Tower’ to ‘Ponte Vecchio’

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

and an arrow back from ‘Ponte Vecchio’ to ‘Ei�el Tower’, but there is no
shortcut from ‘Ei�el Tower’ directly to ‘Ei�el Tower’, that is, there is no
loop attached to ‘Ei�el Tower’.
Now I turn to a relation that cannot easily be described by a diagram
or by listing the pairs in the relation, namely to the relation that obtains
between persons d and e if and only if d is at least as tall as e, that is, the
relation that contains exactly those pairs �d, e� such that d is at least as
tall as e. �is relation is re�exive on the set of all persons because every
person is at least as tall as themselves. �e relation is not symmetric: I am
taller than my brother, so I am at least as tall as he is, but he is not at least
as tall as I am. �us the pair �Volker Halbach, Volker Halbach’s brother�
is an element of the relation, while �Volker Halbach’s brother, Volker
Halbach� is not an element of the relation. �e relation is transitive: if d
is at least as tall as e and e is at least as tall as f , then surely d is at least
as tall as f . Since the relation is not symmetric it is not an equivalence
relation.
�e relation of loving contains exactly those ordered pairs �d, e� such
that d loves e. �is relation is presumably not re�exive on the set of all
persons: some people do not love themselves. Much grief is caused by the
fact that this relation is not symmetric, and the fortunate cases of mutual
love show that the relation is also not asymmetric or antisymmetric. It
clearly fails to be transitive: there are many cases in which d loves e and
e loves f , but in many cases d does not love his or her rival f .
�e relation of not having the same hair colour is the set containing
exactly those pairs �d, e� such that d does not have the same hair colour
as e. �is relation is surely not re�exive on the set of all persons, but
it is symmetric: if d’s hair colour is di�erent from e’s hair colour, then
surely e’s hair colour is di�erent from d’s hair colour. �e relation fails to
be transitive: my hair colour is di�erent from my brother’s hair colour
and his hair colour is di�erent from mine. If the relation were transitive,
then I would have a hair colour that di�ers from my own hair colour.
More formally, the pairs �Volker Halbach, Volker Halbach’s brother� and
�Volker Halbach’s brother, Volker Halbach� are in the relation, while

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

�Volker Halbach, Volker Halbach� is not. �is example illustrates again

that in the de�nition of transitivity it is not presupposed that d must be
di�erent from f .
�e relation of being born on the same day is re�exive on the set of
all persons; it is also symmetric and transitive. �us it is an equivalence
relation on the set of all persons.
I will now turn to another very important kind of relation. It is so
important that it deserves a section of its own.

�.� ���������

���������� �.�. A binary relation R is a function i� for all d, e, f : if

�d, e� ∈ R and �d, f � ∈ R then e = f .
�us a relation is a function if for every d there is at most one e such
that �d, e� is in the relation.
In the diagram of a function there is at most one arrow leaving from
any point in the diagram. In order to illustrate this, I will consider the
function with the following ordered pairs as its elements: �France, Paris�,
�Italy, Rome�, �England, London�, and �the United Kingdom, London�.
�e function has the following diagram:
France / Paris

Italy / Rome

England i4 London
/
iii
iiii
iiiiiii
ii
the United Kingdom
In this diagram, there are arrows from ‘France’, ‘Italy’, ‘England’, and
‘the United Kingdom’. �e set containing France, Italy, England and the

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

United Kingdom is called ‘the domain’ of the function. �e names of the

three cities receive arrows; the set of these three cities is called ‘the range’
of the function.
���������� �.�.
(i) �e domain of a function R is the set { d: there is an e such that
�d, e� ∈ R }.
(ii) �e range of a function R is the set {e: there is a d such that �d, e� ∈
R }.
(iii) R is a function into the set M if and only if all elements of the range
of the function are in M.
�e elements of the domain serve as ‘inputs’ or ‘arguments’ of the
function; the elements of the range are ‘outputs’ or ‘values’.
In the above example the set containing France, Italy, England and the
United Kingdom is the domain of the function, while the set with Paris,
Rome and London as its elements is the range. According to (iii) of the
above de�nition, the function is a function into the set of all European
cities, for instance.
���������� �.�. If d is in the domain of a function R one writes R(d) for
the unique object e such that �d, e� is in R.
�e relation containing all pairs �d, e� such that d has e as a biological
mother is a function: if d has e as biological mother and d has f as
biological mother, then e and f must be identical. Its domain is the set
of all people and animals, its range the set of all female animals with
o�spring.
In contrast, the relation containing all pairs �d, e� such that d is the
biological mother of e is not a function: my brother and I have the same
biological mother, yet we are not identical.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

�.� ���-������ ���������

�e relations I have considered so far are binary; they contain only ordered
pairs. Expressions such as ‘d loves e’ express binary relations; the expres-
sion ‘d loves e’ expresses the relation that contains exactly those ordered
pairs �d, e� such that d loves e. In contrast, the expression ‘d prefers e
over f ’ expresses a ternary (�-place) relation rather than a binary one. In
order to deal with ternary relations, ordered triples (or ‘triples’ for short)
are used. Triples are very much like ordered pairs.
A triple �d, e, f � is identical with a triple �g, h, i� if and only if they
agree in the �rst, second and third component, respectively, that is, if and
only if d = g, e = h and f = i.�
Ternary relations are sets containing only triples.
Besides ordered pairs and triples there are also quadruples and so on.
�is can be generalised to even higher ‘arities’ n: an n-tuple �d� , d� , . . . , d n �
has n components. An n-tuple �d� , d� , . . . , d n � and an n-tuple �e� , e� , . . . , e n �
are identical if and only if d� = e� and d� = e� and so on up to d n = e n .
Now n-tuples allow one to deal with n-place relations:
An n-place relation is a set containing only n-tuples. An n-place rela-
tion is called a relation of arity n.
For instance, there is the relation that contains exactly those �-tuples
�d, e, f , g, h� such that d killed e with f in g with the help of h. �is is
a �-ary relation, which, for instance, contains among others the �-tuple
�Brutus, Caesar, Brutus’ knife, Rome, Cassius�.
I also allow �-tuples as a special case. I stipulate that �d� is simply
d itself. �us a �-place or unary relation is just some set.

� As has been remarked in footnote � above, one can de�ne ordered pairs as certain sets.
Similarly one can de�ne the triple �d, e, f � using ordered pairs as ��d, e�, f �. So in the
end only sets are needed.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

�.� ���������, �������� ��� �������������

In logic usually sentences are taken as the objects that can be true or false.
Of course not every sentence of English can be true or false: a command
or a question is neither true nor false.
Sentences that are true or false are called declarative sentences. In what
follows I will focus exclusively on declarative sentences. I will o�en drop
the restriction ‘declarative’, because I will be concerned exclusively with
declarative sentences.
Whether a sentence is true or not may depend on who is uttering the
sentence, who is addressed, where it is said and various other factors. �e
sentence ‘I am Volker Halbach’ is true when I say it, but the same sentence
is false when uttered by you, the reader. ‘It is raining’ might be true in
Oxford but false in Los Angeles at the same time. So the truth of the
sentence depends partly on the context, that is, on the speaker, the place,
the addressee and so on. Dealing with contexts is tricky and logicians
have developed theories about how the context relates to the truth of a
sentence. I will try to use examples where the context of utterance does
not really matter, but for some examples the context will matter. Even in
those cases, what I am going to say will be correct as long as the context
does not shi� during the discussion of an example. �is will guarantee
that a true sentence cannot become false from one line to the other.
We o�en draw conclusions from certain sentences, and a sentence
is o�en said to follow from or to be a consequence of certain sentences.
Words like ‘therefore’, ‘so’, or ‘hence’, or phrases such as ‘it follows that’
o�en mark a sentence that is supposed to follow from one or more sen-
tences. �e sentences from which one concludes a sentence are called
‘premisses’, the sentence, which is claimed to be supported by the pre-
misses is called ‘conclusion’. Together premisses and conclusion form an
argument.
���������� �.�. An argument consists of a set of declarative sentences (the
premisses) and a declarative sentence (the conclusion) marked as the con-

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

cluded sentence.
�ere is no restriction on how the conclusion is marked as such.
Expressions like ‘therefore’ or ‘so’ may be used for marking the conclusion.
O�en the conclusion is found at the end of an argument. �e conclusion,
however, may also be stated at the beginning of an argument and the
premisses, preceded by phrases such as ‘this follows from’ or ‘for’, follow
the conclusion.
In an argument there is always exactly one conclusion, but there may
be arbitrarily many premisses; there may be even only one premiss or no
premiss at all.
�e following is an argument with the single premiss ‘Zeno is a tor-
toise’ and the conclusion ‘Zeno is toothless’.

Zeno is a tortoise. �erefore Zeno is toothless.

A biologist will probably accept that the conclusion follows from the pre-
miss ‘Zeno is a tortoise’, as he will know that tortoises do not have teeth.
�at the conclusion follows from the premiss depends on a certain biolog-
ical fact. �is assumption can be made explicit by adding the premiss that
tortoises are toothless. �is will make the argument convincing not only
to biologists but also to people with no biological knowledge at all. �e
biologist, if prompted for a more explicit version of the argument, would
probably restate the argument with the additional premiss on which he
may have implicity relied all along:

Zeno is a tortoise. All tortoises are toothless. �erefore Zeno

is toothless.

Now no special knowledge of the subject matter is required to see that

the conclusion follows from the premisses. �e conclusion follows from
the two premisses purely formally or logically: the conclusion is a conse-
quence of the premisses independently of any subject-speci�c assump-
tions. It does not matter who Zeno is, what tortoises are, what being
toothless is, or which objects the argument is thought to be about.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

In this argument the conclusion follows from the premisses indepen-

dently of what the premisses and conclusion are about. Whatever they
are taken to be about, in whatever way the subject-speci�c terms are
(re-)interpreted, the conclusion will be true if the premisses are. Argu-
ments of this kind are called ‘logically valid’ or ‘formally valid’. �us
in a logically valid argument the conclusion follows from the premisses
independently of the subject matter.
���������������� �.� (������� ��������). An argument is logically
valid if and only if there is no interpretation under which the premisses are
all true and the conclusion is false.�
In particular, if all terms are interpreted in the standard way, then,
according to Characterisation �.�, the conclusion is true if the premisses
are true. �us the conclusion of a logically valid argument is true if the
premisses are true.
�e notion of an interpretation employed in Characterisation �.�
needs some clari�cation: An interpretation will assign meanings to the
subject-speci�c terms such as ‘Zeno’, ‘tortoise’, and ‘iridium’. It will also
determine which objects the argument is taken to be about. �e logical
terms, that is, the subject-independent terms, such as ‘all’ are not subject
to any (re-)interpretation. �ese logical terms belong to the form of the
argument and they are not a�ected by interpretations.
In later chapters I shall provide an exact de�nition of interpretations
or ‘structures’, as I shall call them in the case of formal languages. �ese
formal accounts of logical validity can also be seen as attempts to elucidate
the notion of logical validity in natural languages such as English at least
for those parts of English that can be translated into the formal languages.
According to the above characterisation of logical validity, the mean-
ings of the subject-speci�c terms do not matter for the logical validity
of the argument. �us, one can replace these terms by other terms and

� A precise and informative de�nition of the logical validity of an argument is not so easy
to give. Sainsbury (����, chapter �) provides an critical introductory discussion.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

thereby obtain a logically valid argument again. �e following argument

has been obtained from the argument about Zeno by such a substitution
of nonlogical, that is, subject-speci�c terms:

Iridium is a metal. All metals are chemical elements. �ere-

fore iridium is a chemical element.

Both the argument about Zeno and the argument about iridium have the
same pattern; they share the same form. �e conclusion follows from the
premisses solely in virtue of the form of the argument. �is is the reason
for calling such arguments ‘formally valid’.
�e notion of logical or formal validity is occasionally contrasted with
other, less strict notions of validity, under which more arguments come
out as valid. Some arguments in which the truth of the premisses does
guarantee the truth of the conclusion are not formally valid. Here is an
example:

Hagen is a bachelor. �erefore Hagen is not married.

In this argument the conclusion is bound to be true if the premiss is true,

but it is not logically or formally valid, that is, valid in virtue of its form.
‘Hagen is not married’ follows from ‘Hagen is a bachelor’ in virtue of the
meaning of the word ‘bachelor’, which is subject-speci�c.
Also, arguments in which the premisses do not guarantee the truth of
the conclusion are o�en called valid. Here is an example:

All emeralds observed so far have been green. �erefore all

emeralds are green.

�e premiss may support the conclusion in some sense, but it does not
guarantee the truth of the conclusion. Such arguments as the argument
above are said to be inductively valid. In logically valid arguments, in
contrast, the truth of the premisses guarantees the truth of the conclusion.
Logically valid arguments, are also called ‘deductively valid’.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

In this book I will focus on logical validity and not consider other, less
stringent kinds of validity. �erefore, I shall o�en drop the speci�cation
‘logical’ or ‘formal’: validity will always be understood as logical validity.
�ere are good reasons to focus on logically valid arguments. Philoso-
phers o�en suppress premisses in arguments because they think that
these premisses are too obvious to state. However, one philosopher’s
obvious premiss can be another philosopher’s very contentious premiss.
Trying to make an argument logically valid forces one to make all hidden
assumptions explicit. �is may unearth premisses that are not obvious
and uncontroversial at all. Also, there is usually not a unique way to
add premisses to render an argument logically valid, and it may remain
controversial which premisses were implicitly assumed by the original
author, or whether he relied on any implicit premisses at all. At any rate,
if an argument is formally valid, then the validity does not rely on any
potentially controversial subject-speci�c assumptions: all the assump-
tions needed to establish the conclusion will be explicitly laid out for
inspection.
�is is not to say that logical validity is always obvious: all required
premisses may have been made explicit, but it might not be obvious that
the conclusion follows from the premisses, that is, one might not be able
to see easily that the argument is logically valid. Characterisation �.� of
logical validity does not demand an obvious connection between the
premisses and the conclusion that is easy to grasp. Almost all of the
examples of logically valid arguments considered in this book are toy
examples where it will be fairly obvious that they are logically valid, but
showing that an argument is logically valid can be extremely di�cult.
Mathematicians, for instance, are mainly concerned with establishing
that certain sentences (theorems) follow from certain premisses (axioms),
that is, with showing that certain arguments are logically valid. Of course
one can try to break up valid arguments into chains of short and obvious
steps. In Chapter � this task is taken up and a formal notion of proof is
developed.
A valid argument need not have a true conclusion. In the following

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

example the non-logical terms of the logically valid argument about Zeno
(or iridium) have been replaced in such a way as to make the conclusion
false:

Water is a metal. All metals are chemical elements. �erefore

water is a chemical element.

Although the conclusion ‘Water is a chemical element’ is false, the argu-

ment is logically valid: the conclusion still follows from the premisses.
In a logically valid argument the conclusion may be false as long as at
least one premiss is false. In this case ‘Water is a metal’ is false. �erefore,
one cannot refute the validity of an argument by merely pointing out a
false conclusion. If the conclusion of an argument is false, then either at
least one of the premisses is false or the argument is not logically valid
(or both).
So far I have used only one argument form (argument pattern) in my
examples. Here is an argument of a di�erent pattern:

Either CO� -emissions are cut or there will be more �oods. It

is not the case that CO� -emissions are cut. �erefore there
will be more �oods.

�e argument is logically valid according to Characterisation �.� of logi-

cally valid arguments since the validity of the argument does not depend
on the subject-speci�c terms such as ‘CO� -emissions’ and ‘�oods’. �e
validity of the argument depends on the logical terms ‘either . . . or . . . ’
and ‘it is not the case that . . . ’.
In the argument about Zeno I could replace various terms, but not
complete sentences. In the present example one can replace entire sen-
tences. In this case the argument will still be valid a�er replacing the
sentences ‘CO� -emissions are cut’ and ‘�ere will be more �oods’ with
some other sentences. �e pattern of the valid argument is a pattern of
whole sentences. Valid arguments of this kind are said to be proposi-
tionally valid. �us an argument is propositionally valid if and only if

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

there is no (re-)interpretation of the sentences in the argument such that

all premisses are true and yet the conclusion is false. �ese patterns of
propositionally valid arguments are studied in sentential or propositional
logic. Propositional validity will be treated in Chapters � and �.
�e argument about Zeno can be adequately analysed in predicate
logic only, and not in propositional logic. Predicate logic is based on
propositional logic; from the technical point of view it is a re�nement of
propositional logic. �us I shall start with propositional logic and then
move on to predicate logic.
�e notion of consistency is closely related to the notion of validity.
���������������� �.� (�����������). A set of sentences is consistent if
and only if there is a least one interpretation under which all sentences of
the set are true.
�e negation of a sentence is obtained by writing ‘It is not the case
that’ in front of the sentence (in English there are various stylistically
more elegant ways to express negation). A sentence is false if and only if
its negation is true.
For a valid argument there is no interpretation under which the pre-
misses are all true and the conclusion is false. �us, for a valid argument
there is no interpretation under which the premisses are all true and the
negation of the conclusion is also true. �us, if an argument is valid, the
set obtained by adding the negation of the conclusion to the premisses is
inconsistent; and if the set obtained by adding the negation of the conclu-
sion to the premisses is inconsistent, then there is no interpretation under
which all sentences of that set are true, and, consequently, there is no
interpretation under which all the premisses are true and the conclusion
is false. Hence one can de�ne validity in terms of consistency:

An argument is valid if and only if the set obtained by adding

the negation of the conclusion to the premisses is inconsis-
tent.

I have not imposed any restrictions on the number of premisses in

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

an argument. In particular, there may be no premisses at all. Arguments

with no premisses may still be logically valid. �e following argument
does not have any premisses but only a conclusion:

All metaphysicians are metaphysicians.

�e sentence is true, and it is true for any interpretation of ‘metaphysician’,

which is the only non-logical, subject-speci�c term in the sentence. �ere-
fore, there is no interpretation under which all premisses are true (there
is none) and the conclusion is false. �erefore, the argument is logically
valid. �e conclusion of a logically valid argument with no premisses is
also called ‘logically true’ or ‘logically valid’.
���������������� �.� (������� �����). A sentence is logically true if
and only if it is true under any interpretation.
�ere are also sentences that cannot be made true by any interpreta-
tion. �ese sentences are called ‘logically false’. �ey are called ‘contra-
dictions’.
���������������� �.�� (�������������). A sentence is a contradiction
if and only if it is false under any interpretation.
If a sentence A follows logically from a sentence B and B follows
logically from A, that is, if the argument with A as its only premiss and
B as conclusion, and the argument with B as premiss and A as conclusion,
are logically valid, then the sentences A and B are logically equivalent.
According to Characterisation �.�, the argument with A as premiss and B
as conclusion and the argument with B as premiss and A as conclusion
are both logically valid if and only if A and B are true under the same
interpretations:
���������������� �.�� (������� �����������). Sentences are logically
equivalent if and only if they are true under exactly the same interpreta-
tions.

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

�.� ������, ��������� ��� ����������

In the following chapters I will examine formal languages. �ese lan-

guages are in many respects much less complicated than natural languages
such as English or German. �ey are intended to mirror certain prop-
erties of natural languages. Some philosophers conceive of these formal
languages as models for natural languages.
Usually, in analysing either natural or formal languages one distin-
guishes three aspects of a language: syntax, semantics and pragmatics.�
In order to use a language competently, one must master all three aspects
of it.
Syntax is concerned with the expressions of a language bare of their
meanings. In the syntax of a language it is speci�ed what the words or
sentences of the language are. In general, the grammar of a language
belongs to the syntax of that language, and o�en syntax is identi�ed with
grammar. In order to use the language competently, one must know the
grammar of the language. In particular, one must know how to form
sentences in the language.
Semanticssemantics may be described as the study of the meanings
of the expressions of a language. Clearly, to use a language one must not
only know what the words and the sentences of the language are; one
must also know what they mean.
�e expression ‘Im Mondschein hockt auf den Gräbern eine wild
gespenstische Gestalt’ is a well-formed German declarative sentence. In
a syntactic analysis of that sentence one may remark that ‘hockt’ is a verb
in present tense, and so on. All this is merely syntactic information; it
does not tell one anything about the meaning of that sentence. In order
to understand the sentence, you need information about meaning. For
instance, it is a semantic fact of German that ‘im Mondschein’ means ‘in
the moonlight’.

� �e trichotomy was introduced by Morris (����).

© Volker Halbach ����/����

 � Sets, Relations, and Arguments ��

�e third component, pragmatics, will not be studied here. Pragmatics

is, roughly speaking, the study of language in use. Assume John calls
Mary and asks her whether she wants to come along to the cinema. She
replies, ‘I am ill.’ Obviously, John should not expect Mary to come along,
but the sentence ‘I am ill’ does not mean the same thing as ‘I don’t want
to come along to the cinema’; the former sentence only says something
about Mary’s health. But uttered by Mary in this particular situation, the
sentence ‘I am ill’, spoken by Mary, conveys the information that she will
not join John. �us, John needs pragmatics in order to understand that
Mary is not going to come along. Pure semantics would not tell him.

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic

In this section I shall introduce the language of propositional logic. All

other formal languages that will be discussed in this manual are based on
the language of propositional logic.
Before introducing the syntax of this language I will brie�y outline a
method for talking e�ciently about the expressions of a language and for
describing the syntax of a language. �e method is by no means speci�c
to the language of propositional logic.

�.� ���������

By enclosing an expression in quotation marks one can talk about that

expression. Using quotation marks one can say, for instance, that ‘A’ is
the �rst letter of the alphabet and that ‘Gli enigmi sono tre’ is an Italian
sentence. �e quotation of an expression is that very expression enclosed
in quotation marks.�
Quotation marks allow one to designate single expressions. Describ-
ing the syntax of a language usually makes it necessary to talk about a
large or in�nite number of expressions. For instance, one would like to
be able to state that one can construct new sentences in English by com-
bining sentences using ‘and’ (ignoring capitalisation and punctuation).

� Cappelen and LePore (Spring ����) provide an overview of the intricacies of quotation
and of proposed theories. A classical text on quotation is Quine (����).

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

Logicians would express that general claim by saying the following:

(���) If � and ψ are English sentences then ‘� and ψ’ is an

English sentence.

‘It is raining’ is an English sentence. If one takes ‘It is raining’ as both

� and ψ in the above rule, then the rule says that ‘It is raining and it is
raining’ is also an English sentence (again we ignore the absence of the
full stop and the missing capitalisation). One can then use ‘It is raining
and it is raining’ as � again and ‘It is raining’ as ψ to conclude from the
rule that also ‘It is raining and it is raining and it is raining’ is an English
sentence. In this way one can construct longer and longer sentences and
there is no limit to the iterations.
I think that (���) is fairly straightforward and should be easy to
understand. �ere is, however, something puzzling about it as well: the
part of (���) claiming that ‘� and ψ ’ is an English sentence is decidedly
not about the expression in quotation marks. �e letters � and ψ are
Greek letters, and the expression with ‘�’ as �rst symbol, followed by a
space, followed by ‘and’ and another space, followed by ‘ψ ’, is de�nitely
not an English sentence. Only once ‘�’ and ‘ψ ’ are replaced by English
sentences does ‘� and ψ ’ become an English sentence.
�e Greek letters used in this way are metavariables or metalinguistic
variables.
�us, the above rule may also be expressed in the following way:
An English sentence followed by ‘and’ (in spaces) and another
or the same English sentence is also an English sentence.
�is way of rephrasing (���) does not rely on quotation marks but on
talking about expressions following one another. �is method is perhaps
safer than using (���) with its quotation marks and metavariables, but it
is also more cumbersome when applied to intricate grammatical rules.
�us, I will present de�nitions in the style of (���) rather than talking
about expressions following one another.

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

Logicians hardly ever use the expressions of formal languages in

the way they use the expressions of their mother tongue, but they o�en
talk and write about the expressions of these formal languages. Since
the expressions of the formal languages they are concerned with di�er
from expressions of English, logicians usually drop the quotation marks.
Instead of saying
‘(P → (Q ∧R))’ is a sentence of the language of propositional
logic
they say,
(P → (Q ∧ R)) is a sentence of the language of propositional
logic.
I will follow this convention and usually drop quotation marks around
the expressions of formal languages in this manual. �is also applies to
expressions containing metavariables.

�.� ��� ������ �� ��� �������� �� ������������� �����

Now I can describe the syntax of the language L� of propositional logic.

���������� �.� (�������� �������). P, Q, R, P� , Q� , R� , P� , Q� , R� , P� ,
Q� , R� , and so on are sentence letters.
Using metavariables I will de�ne the notion of a sentence of the lan-
guage L� of propositional logic.
���������� �.� (�������� �� L� ).
(i) All sentence letters are sentences of L� .
(ii) If � and ψ are sentences of L� , then ¬�, (� ∧ ψ), (� ∨ ψ), (� → ψ)
and (� ↔ ψ) are sentences of L� .
(iii) Nothing else is a sentence of L� .
Given what I have said about metavariables, (ii) implies that ‘(� ∧ ψ)’
becomes a sentence of the language of propositional logic when the Greek

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

letters ‘�’ and ‘ψ ’ have been replaced by sentences of the language of

propositional logic. �e Greek letters ‘�’ and ‘ψ ’ themselves are not
expressions of the language L� .
As I explained on page ��, I could have formulated part (ii) of De�ni-
tion �.� without using the metavariables ‘�’ and ‘ψ’ by expressing (ii) in
the following way:
�e negation symbol followed by a sentence of L� is again a
sentence of L� . �e symbol ‘(’ followed by a sentence of L� ,
followed by the symbol ‘∨’ (or ‘∧’, ‘→’, ‘↔’), followed by a
sentence (not necessarily distinct from the �rst one), followed
by the symbol ‘)’, is a sentence of L� .
I hope that (ii) is not only shorter but also easier to grasp.
�e last clause in De�nition �.� says that only expressions one can
obtain by using clauses (i) and (ii) are sentences. Very o�en this last clause
is omitted and the clauses (i) and (ii) are implicitly taken to be the only
means of arriving at sentences. At various points in this manual I will
provide de�nitions that are similar to De�nition �.�. In those cases I will
drop the analogues of clause (iii) for the sake of simplicity.
Logicians also say ‘the negation of �’ rather than ‘¬�’. In this termi-
nology, ¬� is the negation of �, and similarly (� ∧ ψ) is the conjunction of
� and ψ, and � ∨ ψ is the disjunction of � and ψ. �e sentence ¬(P → Q),
for instance, is the negation of (P → Q).
������� �.�. �e following expressions are sentences of the language L� :
((P → P) ∧ Q��� ),
¬(R ∨ (P ∨ (P� ∨ ¬Q� ))),
((¬P ∧ Q� ) → P).
In the next example I show how to prove that the last sentence above
is indeed a sentence of L� .
������� �.�. By De�nition �.�(i), P is a sentence of L� . �us, by (ii),
¬P is also a sentence of L� . By (i) again, Q� is a sentence of L� . By (ii) and

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

by what has been said so far, (¬P ∧ Q� ) is a sentence, and by (ii) again and
by what has been said so far, ((¬P ∧ Q� ) → P) is also a sentence of L� .
�e symbols ¬, ∧, ∨, →, ↔ are called ‘connectives’. �ey roughly cor-
respond to the English expressions ‘not’, ‘and’, ‘or’, ‘if . . . , then . . . ’ and ‘if
and only if’, respectively.
name in English symbol alternative
used here symbols
conjunction and ∧ ., &
disjunction or ∨ +, v
negation it is not the ¬ –, ∼
case that
arrow if . . . then → ⊃, ⇒
(material
implication,
conditional)
double arrow, if and only if ↔ ≡, ⇔
(biconditional
material
equivalence)
�e names in brackets and the symbols in the rightmost column are used
by some other authors; they will not be used here.
�e expressions in the ‘in English’ column indicate how the connec-
tives are commonly read, rather than their precise meanings.

�.� ����� ��� �������� ��������

A sentence with many brackets can be confusing. For convenience I shall

employ certain rules for dropping brackets. �ese rules are not revisions
of the de�nition of a sentence and they do not form part of the o�cial
syntax of the language L� of propositional logic. �ese rules are mere
conventions that allow one to write down abbreviations of sentences
instead of the sentences themselves in their o�cial form.

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

Most logicians employ at least some of these rules. For instance, hardly
anyone writes (P ∧ Q) instead of P ∧ Q. �is, then, is the �rst rule:
���������� ���������� �. �e outer brackets may be omitted from a
sentence that is not part of another sentence.
For instance, one may write P → (Q ∨ P) instead of (P → (Q ∨ P)).
However, this convention does not allow one to drop any brackets from
¬(P → (Q ∨ P)), because the sentence (P → (Q ∨ P)) is written here as
a part of the sentence ¬(P → (Q ∨ P)).
Here a warning is in order: �e syntactic de�nitions in Section �.�
apply to L� -sentences but not to their abbreviations. For instance, I have
de�ned the negation of the sentence � as ¬�. Now one might think that
¬P ∧ Q is the negation of P ∧ Q. �is is not the case, however. P ∧ Q is
short for (P ∧ Q) according to Convention �; and the negation of (P ∧ Q)
is ¬(P ∧ Q) and not ¬P ∧ Q, which is short for (¬P ∧ Q).
���������� ���������� �. �e inner set of brackets may be omitted
from a sentence of the form ((�∧ψ)∧ χ). An analogous convention applies
to ∨.
One may abbreviate ((P ∧ Q� ) ∧ P� ) as (P ∧ Q� ) ∧ P� by Conven-
tion �, and then one may also drop the remaining pair of brackets by
Convention �; so P ∧ Q� ∧ P� is short for ((P ∧ Q� ) ∧ P� ).
((� ∧ ψ) ∧ χ) may be part of a larger sentence. So, using Conventions
� and �, one can abbreviate the sentence

((((P� ∧ P� ) ∧ Q) ∧ R) → ((P� ∨ ¬P� ) ∨ Q))

with the following expression:

(P� ∧ P� ∧ Q ∧ R) → (P� ∨ ¬P� ∨ Q).

In arithmetic one may write � × � + � instead of (� × �) + �, because

the symbol × for multiplication binds more strongly than the symbol +
for addition. Analogously, there are conventions for ordering logical

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

operations: ∧ and ∨ bind more strongly than → or ↔. And just as in the

case of arithmetic, this yields further conventions for dropping brackets.
For instance, the sentence P ↔ Q ∧ R is in abbreviated form. In order
to decide whether the sentence is an abbreviation of ((P ↔ Q) ∧ R) or
of (P ↔ (Q ∧ R)), imagine that the connectives ↔ and ∧ ‘compete’ for
the sentence letter Q. Since ∧ binds more strongly than ↔, ∧ gains the
upper hand and (P ↔ (Q ∧ R)) is the correct reading.
In more abstract terms, this rule for dropping brackets can be ex-
pressed as follows:
���������� ���������� �. Assume �, ψ, and χ are sentences of L� , ● is
either ∧ or ∨, and ○ is either → or ↔. �en, if (� ○ (ψ ● χ)) or ((� ●ψ) ○ χ)
occurs as part of the sentence that is to be abbreviated, the inner set of
brackets may be omitted.
A�er dropping the outer brackets from ((P ∧ Q) → R) according
to Convention �, one may shorten the sentence further to P ∧ Q → R
since ∧ binds more strongly than →. Similarly ¬(P → ((Q ∧ P� ) ∨ R))
may be abbreviated as ¬(P → (Q ∧ P� ) ∨ R). In (¬(P ∨ Q) ↔ Q)
only the outer brackets can be omitted in virtue of Convention �, so
that ¬(P ∨ Q) ↔ Q is obtained as an abbreviation. One cannot use
Convention � to obtain ¬P ∨ Q ↔ Q, because the original sentence does
not contain ((P ∨ Q) ↔ Q) as a part. Intuitively, in ¬P ∨ Q ↔ Q the
negation symbol ¬ refers only to P and not to the sentence (P ∨ Q) as it
does in the original sentence.
�ere is no deeper reason for the choice of these three conventions.
It is never incorrect to use the unabbreviated sentences of L� with all
their brackets rather than their abbreviations. In situations where the
application of the bracketing conventions can give rise to confusions,
it is better not to use them. Also it is perfectly legitimate to apply the
rules selectively. For instance, one may apply only Convention �, but not
Convention �, and write (P ∧ Q) → R for ((P ∧ Q) → R). I will apply
these conventions extensively.

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

�.� ��� ��������� �� ������������� �����

In Section �.� I gave a characterisation of the logical validity of arguments

in English. In this section I will de�ne validity for arguments in the
language L� of propositional logic. �e informal Characterisation �.�
of validity for English arguments will be adapted to the language L� of
propositional logic and thereby be transformed into a precise de�nition.
In order to de�ne logical validity for the language L� , the notion of
an interpretation for the language L� needs to be made precise. First, I
need to say which expressions can be interpreted in di�erent ways and
which are always interpreted in the same way. �e connectives are logical
symbols, the brackets merely auxiliary symbols; logical and auxiliary
symbols cannot be re-interpreted (insofar as one can speak of auxiliary
symbols’ being interpreted at all). �e sentence letters are the only non-
logical symbols of L� ; they can be interpreted in di�erent ways.
�e interpretations of the sentence letters will be provided by so-called
L� -structures. �ese L� -structures need only provide enough information
to determine whether a sentence is true or false. Now, under which
conditions is the sentence P ∧ Q true? If the connective ∧ functions
like ‘and’ in English, then both P and Q must be true for P ∧ Q to be
true; otherwise P ∧ Q will be false. Similarly, since ¬ works like ‘not’, the
sentence ¬R is true if and only if R is false. As ∨ corresponds to ‘or’, the
sentence P ∨ Q is true if and only if P or Q is true (or both are true).
�e arrow → corresponds roughly to the English ‘if . . . then’, but the
latter has a rather complicated semantics. �e L� -sentence P → Q is false
if and only if P is true and Q is false; otherwise it is true. �e phrase ‘if . . .
then’, which corresponds to the arrow, does not always behave like this.
How the arrow → is related to ‘if . . . then’ will be discussed in Section �.�.
Generally, in L� , the truth or falsity of a sentence of L� depends only
on the truth or falsity of the sentence letters occurring in the sentence;
any further information is not relevant. �erefore, L� -structures need
only determine the truth and falsity of all sentence letters.
Instead of saying that a sentence is true, logicians say that the sentence

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

has the truth-value True. �is sounds like a philosophically profound

move since new objects are required: truth-values. However, truth-values
are hardly mysterious objects. Some more mathematically minded logi-
cians use the number � for the truth-value True and � for the truth-value
False. In the end it matters only that True and False are distinct. It is
possible, although not very customary and technically less convenient,
to develop the semantics of L� without truth-values by saying that a sen-
tence is true (or false) instead of saying that it has the truth-value True (or
False). I shall use the letter ‘T’ as a name for the truth-value True and ‘F’
for the truth-value False.
�us an L� -structure provides interpretations of all sentence letters
by assigning to every sentence letter exactly one truth-value, T or F.
���������� �.� (L� -���������). An L� -structure is an assignment of ex-
actly one truth-value (T or F) to every sentence letter of the language L� .�
One may think of an L� -structure as an in�nite list that provides a
value T or F for every sentence letter. �e beginning of such a list could
look like this:
P Q R P� Q� R� P� Q� R�
�
T F F F T F T T F
Starting from the truth-values assigned to the sentence letters by an
L� -structure A, one can work out the truth-values for sentences con-
taining connectives in the following way. �e shortest sentences are the
sentence letters; their respective truth-values are �xed directly by the
L� -structure A. For instance, P could be assigned the truth-value T and
R could be assigned the same truth-value. In this case P ∧ R would receive
the truth-value T, too. If P� is given the truth-value F by A, the sentence
P� ∨ (P ∧ R) gets the truth-value T, because P ∧ R is true; and ψ’s being
true is su�cient to make a sentence � ∨ ψ true.

� In more mathematical terms, an L� -structure is a function into the set {T, F} with the
set of all sentence letters of L� as its domain.

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

�us, the truth-values of the shortest sentences, that is, of the sentence
letters, are �xed by the L� -structure A, and then the truth-values for longer
sentences are determined successively by the truth-values of the sentences
they are made up from.
I will write ���A for the truth-value of � that is obtained on the basis
of A; it is determined by A in the following way:
���������� �.� (����� �� �� L� -���������). Let A be some L� -structure.
�en � . . . �A assigns to every sentence of L� either T or F in the following
way:�
(i) If � is a sentence letter, ���A is the truth-value assigned to � by the
L� -structure A.
(ii) �¬��A = T if and only if ���A = F.
(iii) �� ∧ ψ�A = T if and only if ���A = T and �ψ�A = T.
(iv) �� ∨ ψ�A = T if and only if ���A = T or �ψ�A = T.
(v) �� → ψ�A = T if and only if ���A = F or �ψ�A = T.
(vi) �� ↔ ψ�A = T if and only if ���A = �ψ�A .
Instead of writing ���A = T, I will sometimes write that � is true in A
or that T is the truth-value of � in A.
�e de�nition of � . . . �A does not say explicitly when a sentence has the
truth-value F in A. Nonetheless, extra clauses for falsity are not required,
since a sentence has the truth-value F (in A) if and only if it does not have
the truth-value T. In particular, a sentence letter has the truth-value F if
and only if it is not true in A. Similarly, for negation the following falsity
clause follows from De�nition �.�:

�¬��A = F if and only if ���A = T.

De�nition �.� also implies the following claim for conjunction (and simi-
larly for the other connectives):

� More formally, � . . . �A is a function with the set of all L� -sentences as its domain into
the set {T, F}. It properly extends the L� -structure A, that is, it contains all the ordered
pairs that the function A contains and more besides them.

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

�� ∧ ψ�A = F if and only if ���A = F or �ψ�A = F.

�us, De�nition �.� also says whether a sentence is false in a given struc-
ture). For example, consider the sentence (¬(P → Q) → (P ∧ Q)) and a
structure B that assigns T to the sentence letter P and F to the sentence
letter Q. I want to determine the truth-value �¬(P → Q) → (P ∧ Q)�B .
Using the various clauses of De�nition �.�, one can calculate that its truth-
value by calculating the truth-values of the sentences that were used in
building it up according to the syntactic rules (De�nition �.�) for forming
L� -sentences. Here is how:
�. �P�B = T by assumption and De�nition �.�(i)
�. �Q�B = F by assumption and De�nition �.�(i)
�. �P → Q�B = F by �, �, and De�nition �.�(v)
�. �¬(P → Q)�B = T by � and De�nition �.�(ii)
�. �P ∧ Q�B = F by �, �, and De�nition �.�(iii)
�. �¬(P → Q) → (P ∧ Q)�B = F by �, �, and De�nition �.�(v)
�erefore, ¬(P → Q) → (P ∧ Q) is not true in B.
�e clauses (ii)–(vi) of De�nition �.� can be neatly expressed by truth
tables. According to (ii), for instance, a sentence ¬� has truth-value T if
and only if � has truth-value F. �us if ���A = F, we have �¬��A = T; and
if ���A = T, we have �¬��A = F. �is is expressed in the following table:

� ¬�
T F
F T

�e clauses (iii)–(vi) correspond to the following tables respectively:

� ψ (� ∧ ψ) � ψ (� ∨ ψ)
T T T T T T
T F F T F T
F T F F T T
F F F F F F

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

� ψ (� → ψ) � ψ (� ↔ ψ)
T T T T T T
T F F T F F
F T T F T F
F F T F F T
Truth tables are also useful for calculating the truth-values of sen-
tences with more than one connective. I will use the same example as
above to show how this can be done. �e �rst step is to write below each
sentence letter the truth-value assigned to it by the L� -structure A:

P Q ¬ (P → Q) → (P ∧ Q)
T F T F T F

�e next step is to calculate the truth-values of sentences directly built up

from sentence letters according to the truth tables (in this case the tables
for → and ∧ are needed):
P Q ¬ (P → Q) → (P ∧ Q)
T F T F F T F F

�en one can go on to determine the truth-values for more and more
complex sentences:

P Q ¬ (P → Q) → (P ∧ Q)
T F T T F F T F F

Finally, one will obtain the truth-value for the entire sentence (here high-
lighted in using boldface):

P Q ¬ (P → Q) → (P ∧ Q)
T F T T F F F T F F

One can also use a (multi-line) truth table to work out the truth-values
of a given sentence for all L� -structures.
In a truth table one can also work out the truth-value of the sentence
¬(P → Q) → (P ∧ Q) in any given L� -structure. I employ again the

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

sentence ¬(P → Q) → (P ∧ Q) as an example; it contains two sentence

letters. In a given structure, P can be true or false, and Q can be true or
false. �us there are four possibilities: in any given L� -structure, either
both P and Q are true, or P is true and Q is false, or P is false and Q is true,
or both sentence letters are false. �ese four possibilities are captured in
the two le�most columns of the truth table below. Now one can calculate
the truth-values of the sentence for all four possibilities, and, thereby, for
all L� -structures:

P Q ¬ (P → Q) → (P ∧ Q)
T T F T T T T T T T
T F T T F F F T F F
F T F F T T T F F T
F F F F T F T F F F

Again, the column with the truth-value of the entire sentence is in boldface
letters. I will call this column the main column.
Clearly, if there are only T’s in the main column of a sentence the
sentence is true in all L� -structures; if there are only F’s in the main
column the sentence is false in all L� -structures. �us one can use truth
tables to determine whether a sentence is always true, or whether it is
always false, or whether it is true in some structures and false in others.
�e notion of an L� -structure corresponds to that of an interpretation
of an English sentence in Section �.�. In that section I also used the
notion of an English sentence being true under an interpretation; this
corresponds to the notion of an L� -sentence being true in a structure. �e
de�nitions of logical validity in English (Characterisation �.�), of logical
truth in English (Characterisation �.�), and so on, can be adapted to the
language L� of propositional logic. �e de�nitions are the same for L�
as for English, except that the informal notion of an interpretation from
Section �.� is replaced by the technical notion of an L� -structure.
���������� �.�.
(i) A sentence � of L� is logically true if and only if � is true in all L� -

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

structures.
(ii) A sentence � of L� is a contradiction if and only if � is not true in
any L� -structure.
(iii) A sentence � and a sentence ψ are logically equivalent if � and ψ are
true in exactly the same L� -structures.
Logically true sentences are also called ‘tautologies’.
Logical truths, contradictions and logically equivalent sentences of L�
can also be characterised in terms of truth tables. In what follows, truth
tables are always understood as full truth tables with lines for all possible
combinations of truth-values of all the sentence letters in the sentence.
������� �.�.
(i) A sentence of L� is logically true (or a tautology) if and only if there
are only T’s in the main column of its truth table.
(ii) A sentence is a contradiction if and only if there are only F’s in the
main column of its truth table.
(iii) A sentence � and a sentence ψ are logically equivalent if they agree
on the truth-values in their main columns.
One of the main purposes of developing semantics for L� was to
de�ne the notion of a valid argument in L� that would be analogous to
Characterisation �.� of validity for arguments in English.
���������� �.�. Let à be a set of sentences of L� and � a sentence of L� .
�e argument with all sentences in Γ as premisses and � as conclusion is
valid if and only if there is no L� -structure in which all sentences in Γ are
true and � is false.
�e phrase ‘�e argument with all sentences in Γ as premisses and
� as conclusion is valid’ will be abbreviated by ‘Γ � �’; this is also o�en
read as ‘� follows from Γ’ or as ‘Γ (logically) implies �’. �us Γ � � if and
only if the following holds for all L� -structures A:

If �ψ�A = T for all ψ ∈ Γ, then ���A = T.

�us an L� -argument is not valid i� there is a structure that makes all

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

premisses true and the conclusion false:

���������� �.��. An L� -structure is a counterexample to the argument
with Γ as the set of premisses and � as conclusion if and only if �ψ�A = T
for all ψ ∈ Γ and ���A = F.
�erefore, an argument in L� is valid if and only if it does not have a coun-
terexample.
Following the pattern of the De�nition �.� of consistency for sets of
sentences in English I will de�ne consistency for sets of L� -sentences:
���������� �.�� (�������� �����������). A set à of L� -sentences is
semantically consistent if and only if there is an L� -structure A such that
���A = T for all sentences � of Γ. Semantic inconsistency is just the opposite
of semantic consistency: a set Γ of L� -sentences is semantically inconsistent
if and only if Γ is not consistent.�
A�er De�nition �.� I argued that an argument is valid if and only if
the set obtained by adding the negation of the conclusion to the premisses
is inconsistent. �e argument carries over to L� :
������� �.��. If � and all elements of à are L� -sentences, then the fol-
lowing obtains:

Γ � � if and only if the set containing all sentences in Γ and ¬�

is semantically inconsistent.

�us, for an argument with, say, two premisses � and ψ and a con-
clusion χ, this means that �, ψ � χ if and only if the set {�, ψ, ¬χ} is
semantically inconsistent. �e proof of the theorem is le� to the reader.
Logicians usually allow in�nite sets of premisses, but such in�nite
sets of premisses will not play an important role here. One can actually
prove that if a sentence � of L� follows from a set Γ of sentences, then

� �ere is an alternative way of de�ning the consistency of sets of L� -sentences, which is

known as ‘syntactic consistency’. Although the de�nition looks very di�erent, both no-
tions of consistency coincide. Syntactic consistency will be introduced in De�nition �.�.

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

� already follows from a �nite set of sentences in Γ. �is result is known

as the Compactness �eorem of propositional logic.
�e set Γ of premisses may also be empty. A sentence follows from the
empty set of premisses if and only if it is a tautology (that is, i� it is logically
true). �is is fairly obvious; it is also a special case of �eorem �.�� below.
If Γ has only �nitely many elements, one can use truth tables to check
whether Γ � �. I will show how to answer the question whether Γ � � by
means of an example.
������� �.��. {P → ¬Q, Q} � ¬P.
Claims like the one above may be abbreviated by dropping the curly
brackets around the premisses.Generally, ‘ψ� , . . . , ψ n � �’, where ψ� , . . . ,
ψ n and � are L� -sentences, is short for ‘{ψ� , . . . , ψ n } � �’. So the claim of
Example �.�� may be written P → ¬Q, Q � ¬P.
First I draw a truth table for the premisses and the conclusion in the
following way:
P Q P→¬Q Q ¬P
T T T F F T T FT
T F TTT F F FT (�.�)
F T F T F T T TF
F F F TT F F TF
Now I will check whether there is any line in which the entries in the
main columns of the premisses all have T’s, while the conclusion has an F.
In the �rst line of truth-values the �rst premiss receives an F, the second
a T, and the conclusion an F. �e second and fourth lines also have F’s
for one premiss. �e third line has T’s for both premisses, but also a T for
the conclusion. �us, there is no line where all premisses receive T’s and
the conclusion an F. �erefore, the argument is valid, that is, ¬P follows
from {P → ¬Q, Q} or, formally, P → ¬Q, Q � ¬P.
For �nite sets Γ of premisses, one can reduce the problem of checking
whether Γ � � to the problem of checking whether a single sentence is
logically true. To do this one combines all of the premisses, that is, all
sentences in Γ, using ∧, and then one puts the resulting conjunction in

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

front of an arrow followed by the conclusion. �e resulting sentence is

logically true if and only if the argument is valid. �is can be expressed
more succinctly as follows:
������� �.��. ψ� , . . . , ψ n � � if and only if ψ� ∧. . .∧ψ n → � is a tautology
(that is, i� ψ� ∧ . . . ∧ ψ n → � is logically true).
I do not give a proof of this theorem here.
I will apply �eorem �.�� to the example above: First, the two pre-
misses are combined into (P → ¬Q) ∧ Q. It is necessary to reintroduce
the brackets around P → ¬Q because otherwise the result would be an
abbreviation for P → (¬Q ∧ Q) as ∧ binds more strongly than →. Next,
the arrow is put between this conjunction and the conclusion. �is yields
((P → ¬Q) ∧ Q) → ¬P. �e brackets around the conjunction of the two
premisses are not really necessary since ∧ binds more strongly than →,
but they might make the sentence easier to read. �e truth table for this
long sentence looks like this:

P Q ((P → ¬ Q) ∧ Q) → ¬ P
T T T F FT FT T F T
T F T T TF FF T F T (�.�)
F T F T FT TT TT F
F F F T TF FF TT F

�us, the sentence ((P → ¬Q) ∧ Q) → ¬P is valid, that is, it is a tautology.

By �eorem �.��, it follows that P → ¬Q, Q � ¬P. Of course, we know
this already from truth table (�.�).
Drawing truth tables for arguments or sentences with many sentence
letters is cumbersome: every new sentence letter doubles the number of
lines of the truth table, because for any already given line two possibilities
must be considered: the new sentence letter can have truth-value T or F.
�us, a sentence or argument with � sentence letter requires only � lines,
one with � di�erent sentence letters requires �, one with � requires � lines,
and so on. Generally, the truth table for a sentence or argument with n dif-
ferent sentence letters will have �n lines of truth-values. In Exercise �.�

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

there will be an argument in L� with � sentence letters. �us, writing

down the corresponding truth table would require �� = ��� lines.
To show that an L� -sentence is a tautology, one does not need to draw
a complete truth table. One only needs to show that there cannot be a
line in the truth table that yields an F in the main column. In order to
refute the existence of such a line for Example �.��, the best strategy is to
start with the assumption that the value in the main column is F:

P Q ((P → ¬ Q) ∧ Q) → ¬ P
F

A sentence of the form � → ψ has the truth-value F only if � has truth-

value T and ψ has truth-value F. �us, I must have:

P Q ((P → ¬ Q) ∧ Q) → ¬ P
T F F

I can continue to calculate truth-values ‘backwards’:

P Q ((P → ¬ Q) ∧ Q) → ¬ P
T TT F FT

I write the calculated truth-values also under the �rst two occurrences of
P and Q:
P Q ((P → ¬ Q) ∧ Q) → ¬ P
T T ? T TT F FT
But now there is no way to continue. �e slot marked with a question
mark cannot be �lled with a truth-value: there should be an F under the
negation symbol ¬, as Q has truth-value T, but there should also be a T,
because (P → ¬Q) and P have T’s. It follows that there cannot be a line
with an F in the main column. �erefore, in the full truth table with all
the lines, all truth-values in the main column are T’s. �is proves again
that ((P → ¬Q) ∧ Q) → ¬P is a tautology.

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

Since it is not easy (for an examination marker, for instance) to recon-

struct how the truth-values have been calculated, it is useful to record the
order in which the values were obtained:
P Q ((P → ¬ Q) ∧ Q) → ¬ P
T� T� ? T� T� T� F F� T�

Of course I could have written down the truth-values in a di�erent order.

For instance, a�er arriving at F� , I could have calculated the value in the
last column and only then have turned to the part preceding →.
Now I will use the same method to show that (P → ¬Q) → ¬P is not
a tautology. As before, an F is written in the main column:

P Q (P → ¬ Q) → ¬ P
F

�e following table results from the �rst backwards-calculation:

P Q (P → ¬ Q) → ¬ P
T F F

�us, P must receive the truth-value T:

P Q (P → ¬ Q) → ¬ P
T F FT

�us one can also write a T under the �rst occurrence of P:

P Q (P → ¬ Q) → ¬ P
TT F FT

Since P → ¬Q has the truth-value T and P also has the truth-value T, the
sentence ¬Q receives a T, and Q, accordingly, an F. Hence, the line can
be completed as follows (I will also insert the obligatory indices):

P Q (P → ¬ Q) → ¬ P
T F T� T� T� F� F F� T�

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

At any rate, when one has arrived at a ‘possible’ line one should cal-
culate the truth-values from bottom to top (starting from the truth-values
that have been obtained for the sentence letters) to ensure that one has not
missed a column that cannot given a unique truth-value. Only once this
�nal check has been carried out, one knows that the line obtained is a
possible line in a truth table.
�e above backwards-calculations shows that the sentence (P →
¬Q) → ¬P has truth-value F if P has the truth-value T and Q has the
truth-value F.
Technically speaking, if A(P) = T and A(Q) = F for a structure A,
then (P → ¬Q) → ¬P is false in A. So, by De�nition �.�(i), (P → ¬Q) →
¬P is not logically valid, that is, it is not a tautology.
Sometimes this method of calculating truth-values backwards re-
quires more than one line. �is is the case in the following example:
P Q (P ∨ ¬ Q) ↔ (Q → P)
F
If a sentence � ↔ ψ is false, there are two possibilities: � could have
truth-value T and ψ truth-value F, or, � could have F and ψcould have
truth-value T. As such, one has to take these two possibilities into account:
P Q (P ∨ ¬ Q) ↔ (Q → P)
� T F F
� F F T
I have underlined the truth-values that cannot be uniquely determined,
and so more than one possibility (line) needs to be checked.
�e rest is routine. �e indices in the table below indicate the order
in which I arrived at the truth-values. �e order in which the values
are calculated does not really matter, but the indexing makes it easier to
follow the reasoning.
P Q (P ∨ ¬ Q) ↔ (Q → P)
� F� T� ? T� F T� F� F�
� F� F� F� T� F T� T� ?

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

Neither of the two lines can be completed. �is shows that (P ∨ ¬Q) ↔
(Q → P) is a tautology.
Of course it can happen that more lines are required and that di�erent
cases under consideration have to be split up into further subcases.
�e method of backwards-calculation can also be applied in order to
check whether an argument is valid or not. To show that an argument
is not valid, one has to �nd a line (that is, a structure) where all of the
premisses are true and the conclusion is false. If there is no such line,
the argument is valid. Here is an example of how to use the method to
determine whether an argument is valid. I have picked an example that
forces me to consider several di�erent cases. So, I want to determine
whether
P → Q, ¬(P� → Q) ∨ (P ∧ P� ) � (P ↔ Q) ∧ P� .
I will start by writing the two premisses and the conclusion in one table.
I have to check whether there can be a line in the table where the two
premisses come out as true while the conclusion is false. As such, I should
start by writing T’s in the main columns of the premisses and an F in the
main column of the conclusion:
P Q P� P → Q ¬(P� → Q) ∨ (P ∧ P� ) (P ↔ Q) ∧ P�
T T F
Now I have to make a case distinction: the �rst sentence could be true
because P is false or because Q is true. Similarly, in the case of the other
sentences, there is no unique way to continue. Given that I can make a
case distinction with respect to any of the three sentences, it is not clear
how to proceed. But some ways of proceeding can make the calculations
quicker and less complicated. It is useful to avoid as much as possible
picking a sentence that will require a new case distinction in the next
step.Ultimately though, so long as all possible cases are systematically
checked, the order in which one proceeds will not a�ect the end result.
At this stage in the calculation, a case distinction cannot be avoided,
and so I will pick the last sentence: (P ↔ Q) ∧ P� can be false either

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

because P ↔ Q is false or because P� is false. I will try to complete the

line for the latter case and leave the former for later:
P Q P� P → Q ¬ (P� → Q) ∨ (P ∧ P� ) (P ↔ Q) ∧ P�
� T T F� F
� T F� F� T� T ? F� F F�
�e second line of the table cannot be completed: the second premiss
must be true, but it follows from my assumption that both ¬(P� → Q)
and P ∧ P� must be false. �is means that the rules tell me to write an F
in the slot marked by ?, because P� is false, but they also tell me to write
a T for ?, because the entire sentence ¬(P� → Q) ∨ (P ∧ P� ) is true and
¬(P� → Q) is already false.
Line � is more complicated, because P ↔ Q can be false for two
reasons: �rst, P can be true while Q is false or, second, P can be false while
Q is true. �us, I need to distinguish the subcases marked �.� and �.�:
P Q P� P → Q ¬ (P� → Q) ∨ (P ∧ P� ) (P ↔ Q) ∧ P�
�.� T� T ? T T� T� F� F� F
� T F� F� T� T ? F� F F�
�.� T T F� F� T� F
Since P is true and P → Q is true, Q has to be true as well. But according
to the assumption, Q is false. �erefore, line �.� cannot be completed.
Only case �.� remains:
P Q P� P → Q ¬ (P� → Q) ∨ (P ∧ P� ) (P ↔ Q) ∧ P�
�.� T� T ? T T� T� F� F� F
� T F� F� T� T ? F� F F�
�.� T F� T� T� T F� ? F� F� T� F
Since the second premiss is true and ¬(P� → Q) is false, P ∧ P� ought to
be true. But this cannot be the case, because P is false. So line �.� cannot
be completed either. Since this exhausts the possible ways of refuting the
argument, the following claim has been established:
P → Q, ¬(P� → Q) ∨ (P ∧ P� ) � (P ↔ Q) ∧ P� .

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

Generally the entries in a table where one calculates truth-values ‘back-

wards’ should be indexed by numbers that show the order in which the
values were obtained. �e assumptions in the di�erent cases should be
indicated by underlining the respective truth-values. Subcases should be
marked as such (in the way I have done this above). A proof without in-
dices for the truth-values will be considered to be incomplete. �ese
conventions merely serve the purpose of making the calculations easier
to reconstruct: otherwise the calculations of truth-values can be very
di�cult to follow.
A�er having developed semantics for the language L� , I will now
look at alternatives. One might wonder why the connectives ¬, ∧, ∨, →
and ↔ have been chosen. �ese connectives are used in many logic texts,
and they more or less correspond to expressions in English. However,
there are also other English phrases for connecting sentences. �e phrase
‘neither . . . nor . . . ’ is an example. �e sentence

Neither Jones nor Brown is in Barcelona

will be true if and only if Jones is not in Barcelona and Brown is not
in Barcelona. In L� there is no connective that directly corresponds to
‘neither . . . nor . . . ’. If one added a connective for the phrase ‘neither . . .
nor . . . ’, it would have the following truth table:

� ψ �↓ψ
T T F
T F F
F T F
F F T

However, the connective ↓ for ‘neither . . . nor . . . ’ is not really needed

because one can generate the same truth table using the old connectives
of L� . ‘Neither . . . nor . . . ’ can be re-expressed in English as ‘It is not the
case that . . . , and it is not the case that . . . ’. In L� one can also de�ne ↓ in

© Volker Halbach ����/����

 � Syntax and Semantics of Propositional Logic ��

terms of ¬ and ∧ in the following way:

� ψ ¬�∧¬ψ
T T F FF
T F F FT
F T T FF
F F T TT

Alternatively, one can re-express or de�ne � ↓ψ as ¬(�∨ψ). �us, adding ↓

to the language L� would not increase the expressive power of L� . �e
connective ↓ would only allow one to abbreviate some sentences. �ere
are many more truth tables for which L� does not have connectives. So
far I have looked only at binary connectives (connectives conjoining two
sentences) such as ∧, ∨, →, ↔ and ↓, but there are also unary connectives
(connectives taking one sentence) other than ¬; and there are ternary
connectives (connectives taking three sentences), and so on. Can all these
connectives be expressed with the connectives of L� , that is, with ¬, ∧, ∨,
→, and ↔? �e answer is ‘yes’: all truth tables can be produced with the
old connectives of L� . �is fact is called the truth-functional completeness
of L� . In fact, ¬ and ∧ together without any further connectives are
su�cient for expressing all other connectives. And even ↓ on its own
would do the trick. At any rate, adding more connectives to L� than those
used here is not really necessary and would not increase the expressive
power of L� . I will not prove these results here.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic

In the previous chapter I focused on the formal language L� . �e con-

nectives ∧, ∨, →, and ↔ of L� can be used to combine sentences of L� to
form new compound sentences; the connective ¬ can be added in front
of a sentence of L� to build a new sentence.
English sentences can also be combined and modi�ed with many
di�erent connectives to form new sentences. For instance, one can write
the word ‘and’ between two English sentences to obtain a new sentence.
‘Or’, ‘because’, ‘although’, ‘but’, ‘while’, ‘if’ and many others can be used
in the same way. An expression that connects sentences can also be
more complex: the expression ‘due to the fact that’ between two English
sentences yields a new English sentence. Other expressions, such as ‘if . . . ,
then’, connect sentences even though they are not written between two
sentences.
Other expressions do not combine sentences, but rather modify a
sentence, as is the case with ‘not’, ‘as is well known’, ‘John strongly be-
lieves that’, and ‘regrettably’. ‘Not’ is special insofar as it o�en cannot be
simply inserted into a sentence, but rather requires the introduction of
the auxiliary verb ‘to do’: the introduction of ‘not’ into ‘Alan went to
London’ yields ‘Alan did not go to London’. In this respect ‘not’ is more
complicated than an adverb such as ‘regrettably’ or the connective ¬ of
L� .
In the previous chapter I de�ned the notion of a connective; now I
will apply it to English as well: Expressions that can be used to combine or
modify English sentences are connectives. �is de�nition is far from being
precise, but an exact de�nition of the notion of a connective of English
is not easy to give because sometimes the connectives are not simply

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

plugged into or written between sentences. Occasionally the sentences

themselves have to be modi�ed, for instance, by introducing auxiliary
verbs, as the above examples of ‘not’ shows.

�.� �����-�������������

�e connectives of L� , that is, ¬, ∧, ∨, → and ↔, correspond to connectives

in English. �e semantics of the connectives of L� is very simple; it is
encompassed in their truth tables. In contrast, many connectives of
English function in a much more intricate way.
As an example I will consider the connective ‘because’. Imagine that I
drop my laptop computer on the street. It’s had it: the screen is broken.
So my laptop computer does not work. �e sentence

My computer does not work because I dropped my computer

is also true: the laptop would still be functional if I had not dropped it.
Moreover, it is true that the computer does not work and it is true that I
dropped it. �us, ‘because’ connects the two true sentences ‘My computer
does not work’ and ‘I dropped my computer’ together forming a new true
sentence. In this respect it seems similar to ‘and’.
In other cases, however, one can use ‘because’ to connect two true
English sentences A and B and end up with a false sentence. A�er picking
up my broken laptop, I consider the following sentence:

My laptop computer does not work because it is not plugged

in.

In the situation I just described, it is true that my computer does not work,
and it is true that it is not plugged in as I am standing in the street with
my broken laptop. Nevertheless the sentence that my laptop computer
does not work because it is not plugged in is false: it would work if I had
not dropped it. Even if it were now plugged in, it would not work. It does
not work because I dropped it, not because it is not plugged in. So this

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

is a case in which using ‘because’ to connect two true sentences yields a

false sentence.
Nevertheless, the truth of ‘A because B’ is not completely independent
of the truth and falsity of the English sentences A and B. If A or B (or
both) are false, then ‘A because B’ is also false. �ese dependencies can
be summarised in the following truth table for the English connective
‘because’, where A and B are declarative sentences of English:

A B A because B
T T ?
T F F
F T F
F F F

�e question mark indicates that in this case the truth-value of ‘A be-

cause B’ depends not only on the truth-values of the direct subsentences,
that is, on the truth-values of the sentences A and B that ‘because’ connects.
�is means that when ‘because’ is used to connect two true sentences,
sometimes the resulting sentence is true and sometimes the resulting
sentence is false; so the truth-value of the compound sentence is not de-
termined by the truth-values of the sentences connected by ‘because’. In
this respect ‘because’ di�ers from ‘and’. If the truth-value of the compound
sentence is determined by the truth-value of the connected sentences,
as is the case with ‘and’, then the connective is called ‘truth-functional’.
Connectives like ‘because’ are not truth-functional.
�e following is a general, less than precise characterisation of truth-
functionality :
���������������� �.� (�����-�������������). A connective is truth-
functional if and only if the truth-value of the compound sentence cannot
be changed by replacing a direct subsentence with another sentence having
the same truth-value.
For instance, ‘because’ is not truth-functional: replacing the true
sentence ‘I dropped my computer’ with the equally true sentence ‘the

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

computer is not plugged in’ does change the truth-value of the compound
sentence

My computer does not work because I dropped my computer

from True to False.

�us, the de�nition of truth-functionality of an English connective
can be paraphrased in terms of truth tables: a connective is truth-functional
if and only if its truth table does not contain any question marks.
‘If . . . then’ is usually translated as the arrow →. Some of its occur-
rences, however, are de�nitely not truth-functional. A sentence � → ψ
is true if � is false or ψ is true. In the following sentence, ‘if . . . then’
functions di�erently:

If Giovanni hadn’t gone to England, he would not have caught

a cold in Cambridge.

Assume that Giovanni really did go to England, but did not catch a cold
in Cambridge. In this case one may hesitate to assign a truth-value to
the sentence: some people would say that the sentence is neither true
nor false; others would say that it is false. At any rate, in that case the
sentence is not true. But if the whole sentence is not true, then this is a
case in which the �rst subsentence following ‘if’ is false, but the whole
sentence is also false. But according to the truth table for → a sentence
with a false antecedent is true. �is means that the arrow → cannot be
used to formalise the sentence correctly.
‘If’-sentences describing what would have happened under circum-
stances that are not actual are called ‘subjunctives’ or ‘counterfactuals’.
In these sentences ‘if’ does not function like the arrow → and cannot
be translated as the arrow. �e proper treatment of counterfactuals is
beyond the scope of this book.�
Indicative conditionals such as

� Lewis (����) is a classic text on counterfactuals.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

If Jones gets to the airport an hour late, his plane will wait
for him

are o�en formalised using the arrow →, but it is somewhat questionable

whether → really is appropriate.
Assume, for instance, that Jones arrives at the airport early and he
easily catches the plane. Suppose also that Jones is not a VIP and so the
airline would not have waited for him. Using the arrow for ‘if’ one can
try to translate this sentence as P → Q with the following dictionary:

P: Jones gets to the airport an hour late,

Q: Jones’s plane will wait for Jones.

If �P�A = F, that is, if P is false in the structure A, then P → Q is true

in A, that is, �P → Q�A = T by De�nition �.� or by the truth table of → on
page ��. According to the assumptions, ‘Jones gets to the airport an hour
late’ is actually false. �us, if the formalisation is correct, the displayed
English ‘if’-sentence should be true. But it is highly questionable whether
it is true: one may hold that ‘If Jones gets to the airport an hour late, his
plane will wait for him’ is simply false, even if Jones gets there on time.
�ere is an extensive literature on the treatment of ‘if’-sentences. �e
treatment of ‘if’-sentences, including counterfactuals, has interesting
philosophical implications. I shall not go further into the details of this
discussion here. �e above example should be su�cient to show that
formalising ‘if’ by the arrow is problematic even in the case of indicative
conditionals. For most purposes, however, the arrow is considered to
be close enough to the ‘if . . . then . . . ’ of English, with the exception of
counterfactuals.
�e de�nition of truth-functionality also applies to unary connectives:
a unary connective is truth-functional if and only if the truth-value of the
sentence with the connective cannot be changed by replacing the direct sub-
sentence with a sentence with the same truth-value.
‘It is necessarily the case that A’ or ‘It is necessary that’ is a unary
connective that is not truth-functional. If A is a false English sentence,

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

then ‘It is necessary that A’ is false, but if A is true, ‘It is necessary that A’
may be either true or false:
It is necessary that all trees are trees.
�is sentence is true: ‘all trees are trees’ is logically true and thus necessary.
But, if the true sentence ‘All trees are trees’ is replaced by the true sentence
‘Volker has ten coins in his pocket’ then the resulting sentence
It is necessary that Volker has ten coins in his pocket
is not true, because I could easily had only nine coins in my pocket.
Generally if A is only accidentally true, ‘It is necessary that A’ will be false.
�us the corresponding truth table looks like this:
A it is necessary that A
T ?
F F
Some other connectives – like‘Bill believes that . . . ’ – have nothing but
question marks in their truth tables. In contrast, ‘Bill knows that . . . ’ has
the same truth table as ‘it is necessary that’.�

�.� ������� ����

In this section and the next I will show how to translate English sentences
into L� sentences. �ese translations are carried out in two steps: First
the sentence is brought into a standardised form, which is called the
‘(propositional) logical form’.� In the second step the English expressions

� A more comprehensive account of truth-functionality is given by Sainsbury (����,

Chapter �).

� In this chapter I will usually drop the speci�cation ‘propositional’ from ‘propositional
logical form’ since I will not deal with any other kind of logical form for now. �ere is
also a more complex predicate logical form of an English sentence. �e predicate logical
form will be studied in Chapter �.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

are replaced by symbols. Obtaining the logical form is the non-trivial

part; the step from the logical form to a sentence of the language L� of
propositional logic is simple and purely mechanical.
Here I shall sketch a practical procedure for bringing an English
sentence into its logical form. �e procedure is to be applied to the
sentence, reapplied to its sentences from which the main sentence is built
up, and then repeated until the subsentences cannot be further analysed
by the means of propositional logic.
I have broken down this procedure into �ve steps. In practice they are
all carried out at the same time. �e �rst step is the di�cult one: in this
step it is checked whether the sentence can be broken down into a truth-
functional connective and one or more subsentences, that is, whether the
sentence is built up from one or more sentences with a truth-functional
connective.
For instance, the sentence

�e car doesn’t start because the battery is �at or there is no

petrol in the tank.

is not built up from other sentences with a truth-functional connective:

it is built up from the sentence ‘�e car doesn’t start’ and the sentence
‘�e battery is �at or there is no petrol in the tank.’ with the connec-
tive ‘because’, which is not truth-functional. �e connective ‘or’ is truth-
functional, but it only connects ‘�e battery is �at’ and ‘�ere is no
petrol in the tank.’; so only the subsentence ‘�e battery is �at or there
is no petrol in the tank.’ is built up from other sentences with the truth-
functional connective ‘or’, but not the entire sentence.
To put it in a di�erent way, one identi�es the ‘topmost’ or ‘main’ con-
nective and checks whether it is truth-functional. It is permissible to
reformulate the sentence slightly to put into a form such that it is built
up with a truth functional connective. �e truth-functional connectives
should be taken from a �xed list of truth-functional connectives; this
restriction will enable one to formalise the connectives as the �ves con-

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

nectives of the formal language L� , respectively. Finally, one reapplies the

procedure to the sentence(s) from which the main sentence is built up.
I give now the �ve steps of the procedure and then show how it works
by means of some examples:
�. Check if the sentence can be reformulated in a natural way as a sen-
tence built up from one or more sentences with a truth-functional
connective. If this is not possible, then the sentence should be put in
brackets and not analysed any further.
�. If the sentence can be reformulated in a natural way as a sentence
built up from one or more sentences with a truth-functional connec-
tive, do so.
�. If that truth-functional connective is not one of the standard con-
nectives in Table �.�, reformulate the sentence using the standard
connectives.
�. Enclose the whole sentence in brackets, unless it is a negated sen-
tence, that is, a sentence starting with ‘it is not the case that’.
�. Apply the procedure, starting back at �., to the next subsentence(s)
(that is, to the sentence(s) without the standard connective of step �).

name standard connective some other formulations

conjunction and but, although
, [a comma
between sentences]
disjunction or, unless
negation it is not the case that not, none, never
arrow if . . . then given that, . . .
double arrow if and only if exactly if,
precisely if

Table �.�: standard connectives

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

�ere is no need to memorise this description of the procedure; the

point is to learn how to apply it. �us instead of describing the procedure
in more detail, which would be fairly di�cult, I’ll illustrate the procedure
with several examples:
������� �.�. Rob and Tim will laugh, if the tutor can’t pronounce Siob-
han’s name.
�e sentence is built from ‘Rob and Tim will laugh’ and ‘�e tutor
can’t pronounce Siobhan’s name’ with the connective ‘. . . , if . . . ’. As such,
there is no need for the sort of reformulation called for in �. But ‘. . . ,
if. . . ’ is not a standard connective. So, in accordance with step � I will
reformulate the sentence with the standard connective ‘if . . . then . . . ’:
If the tutor can’t pronounce Siobhan’s name, then Rob and
Tim will laugh.
In step � the entire sentence is enclosed by brackets:
(If the tutor can’t pronounce Siobhan’s name, then Rob and
Tim will laugh)
Step � sends me back again to step �. �e two subsentences to which step �
is applied are:
�e tutor can’t pronounce Siobhan’s name,
Rob and Tim will laugh.
�e �rst sentence contains a negation and so I will reformulate it with the
standard connective ‘it is not the case that . . . ’:
It is not the case that the tutor can pronounce Siobhan’s name
According to step �, the sentence does not need to be put in brackets
since it starts with ‘it is not the case that’.
I still have to apply the procedure to the second sentence. ‘Rob and
Tim will laugh’ is not a sentence built up using a truth-functional connec-
tive, but it can be reformulated in accordance with step � as a sentence
with a truth-functional connective in the following way:

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

Rob will laugh and Tim will laugh,

since ‘. . . and . . . ’ is already a standard connective, there is no need for the

sort of reformulation described in step �. A�er applying step � I obtain
the following expression:

(Rob will laugh and Tim will laugh).

�us the whole sentence now reads as follows:

(If it is not the case that the tutor can pronounce Siobhan’s
name, then (Rob will laugh and Tim will laugh)).

Now I have to start again with step �. �e sentence ‘the tutor can pro-
nounce Siobhan’s name’ cannot be reformulated in a natural way as a
sentence built up with a truth-functional connective. �us it is put in
brackets according to step �:

(If it is not the case that (the tutor can pronounce Siobhan’s
name), then (Rob will laugh and Tim will laugh)).

Next, neither ‘Rob will laugh’ nor ‘Tim will laugh’ can be reformulated as
a sentence built with a truth-functional connective, so they are each put
into brackets as required by step �:

(If it is not the case that (the tutor can pronounce Siobhan’s
name), then ((Rob will laugh) and (Tim will laugh))).

Now this is the (propositional) logical form of the sentence.

������� �.�. Unless the ignition is turned on and there is petrol in the
tank, the engine will not start and I’ll not be able to arrive in time.
I can skip steps � and � because the sentence is already built up with
a truth-functional connective, though not by a standard one. Next, I
replace ‘unless’ by the standard connective ‘or’ in accordance with step �;
then I apply step � and enclose the entire sentence in brackets.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

(�e ignition is turned on and there is petrol in the tank, or

the engine will not start and I’ll not be able to arrive in time).

Since ‘and’ is already a standard connective, step � is applied twice more.

((�e ignition is turned on and there is petrol in the tank),

or (the engine will not start and I’ll not be able to arrive in
time)).

Next, I turn to the part of the sentence a�er ‘or’. �ere are two sentences
containing ‘not’. According to step �, they are to be reformulated with the
corresponding standard connective.

((�e ignition is turned on and there is petrol in the tank),

or (it is not the case that the engine will start and it is not the
case that I’ll be able to arrive in time)).

Now step � is applied four times:

(((�e ignition is turned on) and (there is petrol in the tank)),

or (it is not the case that (the engine will start) and it is not
the case that (I’ll be able to arrive in time))).

�e process terminates here since the remaining sentences not containing

brackets, that is, ‘�e ignition is turned on’ and so on cannot be further
analysed.

�.� ���� ������� ���� �� ������ ��������

Once the logical form of a sentence has been determined, the translation
into the language L� of propositional logic is simple.
In order to translate the logical form of an English sentence into L�
apply the following procedure:
�. Replace standard connectives by their respective symbols in accor-
dance with the following list:

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

standard connective symbol

and ∧
or ∨
it is not the case that ¬
if . . . then . . . →
if and only if ↔
�. Replace every English sentence by a sentence letter and delete the
brackets surrounding the sentence letter.� Use di�erent sentence let-
ters for distinct sentences and the same sentence letter for multiple
occurrences of the same sentence.
�. Give a list (the ‘dictionary’) of all sentence letters in the resulting L� -
sentence together with the respective sentences they have replaced.
I shall carry out this procedure on Example �.�. �e logical form of
that sentence is
(If it is not the case that (the tutor can pronounce Siobhan’s
name), then ((Rob will laugh) and (Tim will laugh))).
To translate this into L� I �rst replace all standard connectives by the
respective symbols, as required by step �:
(¬ (the tutor can pronounce Siobhan’s name) → ((Rob will
laugh) ∧ (Tim will laugh))).
According to step � the sentences are to be replaced by sentence letters:
(¬P → (Q ∧ R)).
I complete the formalisation by adding the dictionary required in step �:
P: �e tutor can pronounce Siobhan’s name.
Q: Rob will laugh.
R: Tim will laugh.

� English sentences usually do not contain any L� connectives or brackets. �us, one will
replace with sentence letters exactly those English sentences that could not be further
analysed in accordance with step � of the procedure on page ��.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

�is was the logical form of the English sentence from the second
example:

(((�e ignition is turned on) and (there is petrol in the tank)),

or (it is not the case that (the engine will start) and it is not
the case that (I’ll be able to arrive in time))).

Its formalisation is the sentence ((P ∧ Q) ∨ (¬R ∧ ¬P� )), or, using the
bracketing conventions, (P ∧ Q) ∨ (¬R ∧ ¬P� ). �e dictionary is obvious:
P: �e ignition is turned on.
Q: �ere is petrol in the tank.
R: �e engine will start.
P� : I’ll be able to arrive in time.
In both examples, I used the sentence letter P to formalise the �rst
sentence, and Q to formalise the next and so on. �is is not obligatory. It
would have been equally correct (but awkward) to employ the sentence
letter R��� instead of P.

�.� ���������

Determining the logical form of an English sentence can be tricky. Some-

times there is no unique solution.

Brown is in Barcelona and Jones owns a Ford or Smith owns

a Ford.

�is sentence is ambiguous: ‘and’ could have been used to connect the two
claims ‘Brown is in Barcelona’ and ‘Jones owns a Ford or Smith owns a
Ford’. It could equally well be used to express that there are the following
two possibilities: �rst, Brown is in Barcelona and Jones owns a Ford;
second, Smith owns a Ford.
Corresponding to these two possible readings there are at least two
possible formalisations of this sentence:

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

(i) P ∧ (Q ∨ R)
(ii) (P ∧ Q) ∨ R
�e dictionary is as follows:
P: Brown is in Barcelona
Q: Jones owns a Ford
R: Smith owns a Ford
�e formalisations (i) and (ii) correspond to the two readings of the
original English sentence. In a given situation it may be clear which
reading is intended, and thus which formalisation is preferable. Without
further hints, however, one cannot decide between (i) and (ii).
Ambiguities like the one above are called ‘scope ambiguities’. Roughly
speaking, the scope of an occurrence of a connective in a sentence is
that part of the sentence to which the connective applies. In (i) the
connective ∧ applies to the entire sentence, as it connects P and (Q ∨ R),
while in (ii) the scope of ∧ is only (P ∧ Q).
���������� �.� (����� �� � ����������). �e scope of an occurrence of
a connective in a sentence � is (the occurrence of) the smallest subsentence
of � that contains this occurrence of the connective.
By ‘subsentence of �’ I mean any sentence that is part of �.
In the sentence
((P → (P ∨ Q)) → (¬P ∧ Q))
��� � � � � � � � � � � � � � � � � ��� � � � � � � � � � � � � � � � � � �
the scope of the second occurrence of the arrow → is the entire sentence;
the scope of the �rst occurrence of → is the underbraced part of the
sentence.
�e de�nition of the scope of an occurrence of a connective refers
to the sentence, not to any of its abbreviations. So, the scope of the �rst
occurrence of the arrow in
(P → P ∨ Q) → (¬P ∧ Q)
��� � � � � � � � � � � � � ��� � � � � � � � � � � � � � �

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

is still the underbraced part, because P → P is not a part (subsentence) of

the real sentence.
�e problem of scope ambiguity highlights a general di�erence be-
tween natural languages such as English and formal languages like L� :
while sentences of English are o�en ambiguous in their structure, sen-
tences of L� are never structurally ambiguous. �us, there is no chance
of translating the ambiguous English sentence into an equally ambiguous
sentence of L� . �is will mean that one might have to choose between
di�erent possible formalisations of an English sentence.
�ere are also two possibilities of formalising the following sentence:

Brown is in Barcelona and Jones owns a Ford and Smith owns

a Ford.

�e sentence may be formalised as P ∧ Q ∧ R, which is short for the

sentence ((P ∧ Q) ∧ R), or, alternatively as the sentence P ∧ (Q ∧ R).
�ese are two di�erent sentences of the language L� . But they are logically
equivalent: the two L� -sentences have the same truth table. �us, it is
does not matter for the purpose of checking validities of arguments etc
which formalisation is used.

�.� ��� �������� �����������

�e syntax of the connectives of L� is very simple: ¬ is written in front

of a sentence, and the sentence that results from writing ¬ in front of a
sentence is the negation of that original sentence �e other connectives
are written between sentences, and the entire string of expressions is
surrounded by brackets. �e syntax of the corresponding expressions in
English is far more complicated.
�e grammar of ‘not’ is a bit complicated: usually one cannot simply
insert ‘not’ into a sentence to obtain the negation of the sentence. O�en
the sentence is reformulated with the auxiliary verb ‘to do’. ‘Bill does not
write an essay’ is the negation of ‘Bill writes an essay’. But there are more

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

ways to express negation: one could also say ‘Bill writes no essay’; in this
case the negation is expressed by ‘no’.
‘And’ and its counterpart ∧ seem less problematic. In some cases ‘and’
does not connect complete sentences:

Liz and Anne are mountaineers.

Here ‘and’ combines two proper names. �e sentence, however, can be

seen as an abbreviation of a sentence in which ‘and’ does connect two
sentences:

Liz is a mountaineer and Anne is a mountaineer.

�is sentence can then be formalised with the help of ∧. But the trick
does not always work. �e sentence

Liz and Anne are friends

can hardly be rephrased as

Liz is a friend and Anne is a friend.

Some English sentences can be reformulated in a way that introduces

‘and’:

Liz is an Australian mountaineer

can be rephrased as

Liz is Australian and Liz is a mountaineer.

However, the sentence

Liz is an avid mountaineer

does not mean the same as

Liz is avid and that Liz is a mountaineer.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

Similarly,

Kentaro is a slim sumo-wrestler

cannot be rewritten as

Kentaro is slim and Kentaro is a sumo-wrestler.

For a slim sumo-wrestler might not be slim at all, but only slim for a
sumo-wrestler.
�e connective ‘but’ and similar words are o�en translated as ∧, al-
though ‘but’ o�en indicates a contrast between the two sentences that are
combined.
‘Or’ is fairly straightforward. ‘Unless’ is in many uses very similar to
‘or’ and may then be translated by ∨. O�en ‘either . . . or . . . ’ is assumed
to be exclusive, that is, ‘Either A or B’ is taken to be false, if A and B are
both true. As a rule of thumb this is correct, but sometimes ‘either . . . or
. . . ’ may be equivalent to the simple ‘or’, and in some cases, with some
emphasis, the simple ‘or’ may be exclusive.

�.� ������� �������� ��� ������������� �����

In the previous sections I have shown how to translate English sentences

into sentences of the language L� of propositional logic. �e concepts of
Section �.� can now be applied to the sentences that have been obtained
as translations.
�e sentence of L� that is obtained by translating an English sentence
into the language of propositional logic is the formalisation of that sentence.
I will o�en speak of ‘the formalisation’ of an English sentence as if
there were always exactly one (best) formalisation. Of course one always
has the choice of di�erent sentence letters, and when one is translating
a phrase like the exclusive ‘either . . . or . . . ’, for which there is no direct
equivalent in L� , one has a choice between di�erent ways of rendering
this phrase.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

Although these di�erent possibilities show already that there cannot

be only a single best formalisation, the di�erences between these for-
malisations do not really matter, because these di�erences do not a�ect
the properties of L� -sentences I am interested in. In particular, one can
replace sentence letters in a tautology, that is, in a logically true sentence
of L� and one will obtain a tautology again as long the same letter is in-
serted for all occurrences of a given sentence letter. �us it does not matter
with respect to the property of being logically true which sentence letters
are used in a translation. Also, how exactly a connective such as ‘either
. . . or . . . ’ is translated does not matter for the property of being logical
truth, being valid etc.: whether ‘Either it rains or it snows’, for instance, is
formalised as (P ∨ Q) ∧ ¬(P ∧ Q), or as P ↔ ¬Q, or as ¬P ↔ Q �with
the obvious dictionary) does not matter for the validity of an argument
in which these formalisations are used, as all these formalisations are
logically equivalent.
In some cases, however, there are equally correct formalisations of
a sentence that di�er in their relevant properties: one of them may be
logically true while the other is not, for instance. Ambiguous sentences
may have more than one formalisation in L� . In such cases one should
be more precise and talk about the formalisation of a sentence under a
certain reading of that sentence. In what follows I will be less precise and
mostly ignore problems of ambiguity.
�e notions of De�nition �.� will now be applied to formalisations of
English sentences, and the English sentences will be categorised accord-
ingly.
���������� �.�.
(i) An English sentence is a tautology if and only if its formalisation in
propositional logic is logically true (that is, i� it is a tautology).
(ii) An English sentence is a propositional contradiction if and only if its
formalisation in propositional logic is a contradiction.
(iii) A set of English sentences is propositionally consistent if the set of all
their formalisations in propositional logic is semantically consistent.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

Instead of saying that a sentence is a tautology one can also describe it

as propositionally valid or propositionally true.
�e following sentence is a tautology:

Unless Alfred is an eminent logician, it is not the case that

both Kurt and Alfred are eminent logicians.

‘Unless A, B’ can be rephrased as ‘if not A, then B’ or simply as ‘A or B’. �e

other steps in the translation are routine, and the following L� -sentence
is obtained as a translation:

P ∨ ¬(P ∧ Q).

�e dictionary is as follows:

P: Alfred is an eminent logician.

Q: Kurt is an eminent logician.

�e truth table shows that P ∨ ¬(P ∧ Q) is a tautology by �eorem �.� (i):

P Q P ∨¬ (P ∧ Q)
T T T TF TT T
T F T TT TF F
F T F TT FF T
F F F TT FF F

�erefore, by De�nition �.� (i), the sentence ‘Unless Alfred is an eminent

logician, it is not the case that both Kurt and Alfred are eminent logicians’
is a tautology.
In the formal language L� of propositional logic, the logically true
sentences are exactly the tautologies. In contrast, in English there are
logically true sentences that are not tautologies. �e sentence ‘All logi-
cians are logicians’ is logically true but it is not a tautology, because the
formalisation in propositional logic is a single sentence letter. A single
sentence letter never is logically true, that is, it never is a tautology.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

Similarly, an English sentence can be contradiction without being a

propositional contradiction: �e sentence ‘�ere is an oak that is not an
oak.’ is an example.
A set of sentences may be propositionally consistent without being
consistent. �e inconsistent set containing the three sentences ‘All birds
can �y’, ‘Tweety is a bird’, ‘Tweety can’t �y’, for instance, is propositionally
consistent.
I turn now to the formalisation of entire arguments. �e formalisation
of an argument in English is that argument in L� that has as its premisses
all the formalisations of the premisses of the English argument, and has
as its conclusion the formalisation of the English conclusion.
���������� �.�. An argument in English is propositionally valid if and
only if its formalisation in L� is valid.
Every propositionally valid argument is also valid, but not every valid
argument is propositionally valid (for an example see the argument about
Zeno on page�.�).
De�nition �.� is a more formal elaboration of the notion of proposi-
tional validity mentioned on page ��. Using the methods developed one
can check arguments in English for their propositional validity.
I will consider some examples.

Jones owns a Ford or Brown is in Barcelona. If Brown is

in Barcelona, Smith is in Barcelona too. But Smith isn’t in
Barcelona. �erefore, Jones owns a Ford.

�e premisses of this argument are translated as the following sentences:

P∨Q
Q→R
¬R

�e conclusion is then formalised as the sentence P. �e sentence letters

stand for the following English sentences:

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

P: Jones owns a Ford.

Q: Brown is in Barcelona.
R: Smith is in Barcelona.
By De�nition �.� the English argument is propositionally valid if and
only if P ∨ Q, Q → R, ¬R � P. �e claim that P ∨ Q, Q → R, ¬R � P can
be established, according to �eorem �.��, by showing that the sentence

(P ∨ Q) ∧ (Q → R) ∧ ¬R → P

is a tautology:

P Q R ((P ∨ Q) ∧ (Q → R)) ∧ ¬ R →P
T T T TTT T T TT FF T TT
T T F TTT F T F F FT F TT
T F T TTF T F TT FF T TT
T F F TTF T F T F TT F TT
F T T FTT T T TT FF T T F
F T F FTT F T F F FT F T F
F F T FFF F F TT FF T T F
F F F FFF F F TF FT F T F

�us the English argument is propositionally valid.

�e following argument might look puzzling and too optimistic with
respect to my �nances:

Jones has ten coins in his pocket; and it is not the case
that Jones has ten coins in his pocket. �erefore there are
� ��� ��� in my bank account.

�e premiss can be formalised as P ∧ ¬P, and the conclusion as Q, with

the obvious dictionary:

P: Jones has ten coins in his pocket.

Q: �ere are � ��� ��� in my bank account.

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

By the truth-table method one can easily establish P ∧ ¬P � Q (see

Exercise �.�). �erefore the argument is propositionally valid.
�e argument looks so puzzling because there is no ‘connection’ be-
tween the premiss and the conclusion: the premiss does not seem to
say anything about my bank account. But Characterisation �.� of a valid
argument does not say that the premisses of a valid argument need to be
relevant in this way to the conclusion. In this case, the premiss of the
argument is a (propositional) contradiction; thus there is no interpreta-
tion that would make the premiss true (and the conclusion false); and,
therefore, the argument is propositionally valid. �is principle, that any
argument with a contradiction as premiss is called ‘ex falso quodlibet’,
which is Latin for ‘From something false everything (follows)’.
Blocking this kind of argument might seem desirable, but it is not a
simple task since abandoning the ex falso quodlibet principle will require
abandoning rules that are o�en applied in reasoning. Only such simple
rules are applied in the following argumentation, which starts from ‘Jones
has ten coins in his pocket’ and ‘it is not the case that Jones has ten coins
in his pocket’, and arrives at the conclusion that there is � ��� ��� in my
bank account.
Jones has ten coins in his pocket. So Jones has ten coins in
his pocket or there is � ��� ��� in my bank account. But
Jones does not have ten coins in his pocket. �erefore there
is � ��� ��� in my bank account.
�e reasoning may sound odd, but it is hard to tell where things go wrong.
If a claim ‘A’ is true, surely the weaker claim ‘A or B’ must be true as
well. And from an alternative ‘A or B’ and the negation ‘not-A’ or the
�rst alternative, one usually concludes ‘B’. If these steps may be used to
establish he validity of an argument, then the above argument is valid
Blocking the ex falso quodlibet principle would involve the rejection of
one of those steps.
Usually when one hits upon a contradiction, one does not carry on
reasoning but rather starts to doubt the premisses. �e ex falso quodlibet

© Volker Halbach ����/����

 � Formalisation in Propositional Logic ��

principle shows that it is pointless to reason on the basis contradictory

premisses, because from such premisses everything follows.

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic

Many arguments in English are valid without being propositionally valid.

�at is, these arguments are valid but when they are translated into the
language L� of propositional logic the resulting argument in L� is not
valid. An example of a valid argument that is not propositionally valid is
the example from page ��:
Zeno is a tortoise. All tortoises are toothless. �erefore Zeno
is toothless.
�e argument is not propositionally valid because each of the premisses
and the conclusion have to be translated into di�erent sentence letters.
�is means that in the language of propositional logic the argument will
look like this: P, Q � R. It is certainly not the case that P, Q � R. �e
English argument, however, served as an example of a valid argument
in Section �.�. In order to capture the validity of arguments like this one
about Zeno, a formal language more powerful and more sophisticated
than the language L� of propositional logic is required.

�.� ���������� ��� ��������������

In this section I shall motivate and introduce the basic elements of the
syntax of the language L� of predicate logic; the precise de�nition of a
sentence of L� will be given in Section �.�.
For its analysis in predicate logic, a simple sentence like ‘Tom loves
Mary’ must be broken down into its constituents: the sentence contains
two designators, ‘Tom’ and ‘Mary’, that is, two expressions intended to
denote a single object. �e expression ‘loves’ is is a ‘predicate expression’

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

or ‘predicate’, for short: it connects the two designators and expresses that
a certain relation obtains between Tom and Mary.� �e predicate ‘loves’
can take two designators. I indicate the slots where the singular terms
can be put by dots: Replacing the two strings of dots in the predicate
expression ‘. . . loves . . . ’ by designators, respectively, yields a declarative
English sentence.
Here are further examples of other sentences built from predicate
expressions and designators:
�e lecture is boring .
��� � � � � � � � � � � � � � � � � � � � � �� ��� � � � � � �� � � � � � � �
designator predicate

Leon sees the Ei�el Tower .

� � ��� � � � � � � � � � � � � � � � � � ��� � � � � � � � � � � � � � � � � � � ��
designator predicate designator

�e tallest student in Oxford gives his friend the CD .

��� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ��� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ��� � � � � � � ��� � � � � � � � � ��� � � �� � � � �
designator predicate designator designator

�e tallest student in Oxford gives the CD to his friend .

��� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ��� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � ��� � � �� � � � � � ��� � � � � � � ��� � � � � � � � �
designator �rst part designator second designator
of predicate part of
predicate

�e engineer loosens the nut with the wrench .

� � � ��� � � � � ��� � � ��� � � � � ��� � � � � � � � � � � � � � � � � � � � � � �
��� � � � � � � � � � � � � ��� � � � � � � � � � � � � � �� ���rst
designator part designator second designator
of predicate part of
predicate

�e predicates can be simple and consist in just one word, as is ‘. . .

sees . . . ’ in the second sentence and ‘. . . gives . . . . . . ’ in the third; or they
can be formed from two or more words, as is ‘. . . is boring’ in the �rst
sentence, ‘. . . gives . . . to . . . ’ in the fourth, and ‘. . . loosens . . . with . . . ’
in the last.

� �e terminology in logic di�ers here from traditional grammar, where ‘loves Mary’
would be the predicate.

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

In predicate logic predicate expressions are translated into predicate let-

ters. �ese predicates have an upper index that corresponds to the number
of designators the corresponding English predicate expression can take.
For instance, ‘. . . is boring’, which can take one designator, is translated
into a predicate letter with upper index �: P � , for instance, is such a predi-
cate letter. �e predicate ‘. . . sees . . . ’ can be translated as Q � since it can
take two designators, and ‘. . . gives . . . . . . ’ can be translated as R � because
it can take three designators. �e upper index (�, �, � here) is called the
predicate letter’s ‘arity-index’. A predicate letter with upper index n is
called an ‘n-place predicate letter’. �-place predicate letters are also called
‘unary’, �-place ‘binary’ and �-place ‘ternary’. So, the predicate expression
‘. . . loosens . . . with . . . ’, for instance, is translated into a ternary predicate
letter.
It is not hard to �nd English sentences that require �- or �-place
predicate letters for their formalisation. �us, I will include predicate
letters with arity-indices for any n in the language L� of predicate logic
to make sure that there is always a su�cient stock of predicate letters
available for the various English predicates.
I will also include �-place predicate letters in the language L� . �ey
are useful for formalising English sentences like ‘It is raining’. It just seems
to be a quirk of the English language that the pronoun ‘it’ in this sentence
is required; ‘it’ does not serve the same purpose as the designators in
the sentences I have considered so far; other languages such as Italian
have dispensed with the pronoun in the corresponding sentence. �us, I
will formalise the sentence ‘It is raining’ and similar sentences as �-place
predicate letters. �is provides the reason for including �-place predicate
letters in the language L� .
Now that I have dealt with predicates, I will turn to the formalisation
of designators. In the easiest cases the designators are proper names like
‘Tom’, ‘the Ei�el Tower’, or ‘the United Kingdom of Great Britain and
Northern Ireland’; other types of designators will be discussed later. Cor-
responding to the proper names of English, the language L� of predicate
logic features constants, namely a, b, c, a� , b� , c� , and so on.

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

�e example sentence ‘Tom loves Mary’ from the beginning of the

section can now be translated into the language L� as

P � ab.

In such sentences, the predicate letter is always put a the beginning of

the sentence. �e predicate letter and the two constants are translated as
follows:
P�: . . . loves . . .
a: Tom
b: Mary
�e order of the constants in P � ab is crucial: P � ba is a formalisation of
‘Mary loves Tom’, which says something di�erent from what ‘Tom loves
Mary’ says.
In the dictionary, the entry for ‘loves’ contains two strings of dots.
�ey stand for the places that are taken in the sentence by the designators
‘Tom’ and ‘Mary’. In the corresponding formal sentence P � ab the �rst
place of the binary predicate letter P � is taken by a, and the second place
is taken by b. Now the �rst string of dots in ‘. . . loves . . . ’ corresponds
to the �rst place of P � and the second string of dots corresponds to the
second place of P � . In order to emphasise this correlation, one can attach
subscripts to the dots:
P�: . . . � loves . . . �

�is is tantamount to the translation for P � given above.

If there are subscripts in the dictionary, the string of dots marked � al-
ways corresponds to the �rst place of the predicate letter, the string marked �
corresponds to the second place of the predicate letter, and so on. �e num-
ber of strings of dots must always correspond to the arity-index of the pred-
icate letter.
�erefore, if I had used
P�: . . . � loves . . . �

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

as the entry for P � , while keeping the entries for a and b unchanged, the
proper translation of ‘Tom loves Mary’would be P � ba.
As my next example I consider a sentence with a slightly more com-
plicated predicate: �e sentence

Ebenezer is a scrooge

can be translated into L� as the sentence R � c, with the following transla-

tion of the predicate letter and the constant:
R� : . . . is a scrooge
c: Ebenezer
�e noun ‘scrooge’ is not translated separately but forms part of the
predicate ‘. . . is a scrooge’, which is translated as the predicate letter R � .
Typically, phrases of the form ‘is a . . . ’ are translated as predicate letters.
One way to see why ‘. . . is a scrooge’ can be translated as a simple predicate
letter only, is to observe that instead of ‘is a scrooge’ one could say ‘is
stingy’, which obviously can be translated as a unary predicate letter.
Generally, in the language L� an n-place predicate letter followed by
n constants yields a sentence. Hence, P � ab, that is, a binary predicate
letter followed by two constants, and R � c, that is, a unary predicate letter
followed by one constant, are sentences of L� . �-place predicate letters
form a sentence without any further symbol: each �-place predicate letter
is already a sentence. �us, they behave in the same way as the sentence
letters of the language L� of propositional logic. In fact, I will identify
the �-place predicate letters with sentence letters; sentence letters are,
therefore, merely a certain sort of predicate letters.
One can build sentences of the language L� using connectives in
the same way as in the language L� of propositional logic. For instance,
(P � ab ∧ R � c) is the translation of the following sentence of L� :

Tom loves Mary and Ebenezer is a scrooge.

�e dictionary is the same as above:

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

P�: . . . loves . . .
a: Tom
b: Mary
R� : . . . is a scrooge
c: Ebenezer
�e techniques for translation into propositional logic carry over to
predicate logic (see page ��): ‘Liz is an Australian mountaineer’ can be
rephrased as ‘Liz is Australian and Liz is a mountaineer’. �e caveats
explained there also apply to predicate logic.
With the techniques developed so far, certain occurrences of personal
pronouns can be readily translated.
Caesar came, he saw, he won.
�is can be paraphrased as the following sentence:
Caesar came and Caesar saw and Caesar won.
�e pronouns (or rather their occurrences) in this example are known
as ‘lazy’ pronouns.� Using ‘he’ here saves one the e�ort of repeating the
name ‘Caesar’. Lazy pronouns can easily be eliminated by repeating the
name (or whatever they refer back to), and thus their formalisations do
not pose any special problems.
�ere are other uses of pronouns that cannot easily be dispensed with.
If a politician speaks the truth, he won’t be elected.
In this sentence the pronoun ‘he’ cannot be replaced by ‘a politician’. �e
sentence
If a politician speaks the truth, a politician won’t be elected
has a di�erent meaning; it says that some politician will not be elected if
some politician (not necessarily the same one) speaks the truth. In fact,
the original sentence is equivalent to

� �is terminology comes from Geach (����).

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

All politicians speaking the truth are not elected.

In the original sentence ‘If a politician speaks the truth, he won’t be
elected’ the pronoun is used to express a generalisation. Uses of pronouns
for the purpose of generalisation (and some similar purposes) are called
‘quanti�cational’ uses. Quanti�cation can be expressed in many di�erent
ways in English. Sometimes pronouns are used, and sometimes quanti�-
cation can be expressed without pronouns, as is the case with the sentence
‘All politicians speaking the truth are not elected’.
At any rate, the translation of quanti�cational uses of pronouns re-
quires additional resources in L� beyond the ones I have mentioned so
far. Personal pronouns in English come in di�erent genders (‘he’, ‘she’,
‘it’), in di�erent cases (‘he’, ‘him’, and so on). �ese di�erent forms help
to disambiguate sentences. Here is a somewhat tricky case:
If a visitor wants to borrow a book from the library, she is
required to complete the form for it, which must then be
submitted to a librarian, who can grant her permission to
check it out, if it looks satisfactory to him.
In order to make the reference of the various occurrences of the personal
pronouns clearer, one can attach indices to them:
If a visitor� wants to borrow a book� from the library, she� is
required to complete the form� for it� , which� must then be
submitted to a librarian� , who can grant her� permission to
check it� out, if it� looks satisfactory to him� .
It is natural to assume that the �rst and second occurrences of ‘it’ refer
back to ‘book’, while the third occurrence refers back to ‘the form’. �is
assumption is made explicit by using subscripts. Natural languages o�er
other resources for disambiguation, but indexing is a straightforward
method. Since the help of gender etc. is no longer required, when the ref-
erence is made clear by indexing, one can dispense with English pronouns
and replace them with what logicians call variables:

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

If a visitor x� wants to borrow a book x� from the library, x� is

required to complete the form x� for x� , which� must then
be submitted to a librarian x� , who can grant x� permission
to check x� out, if x� looks satisfactory to x� .
In the sentence variables refer back to such phrases as ‘a visitor’ and ‘some
book’. One could introduce expressions corresponding to these phrases
into the formal language. Logicians have found a method to simplify
the language further and to manage with just one additional expression:
�e purpose of expressions such as ‘a visitor’ is to restrict the focus to
visitors; the above sentence makes a general claim about visitors. But one
can replace this with a general claim about all things (whether they are
persons, animals or inanimate objects). Instead of saying
Every visitor is a classicist,
one can say
If something is a visitor, then it is a classicist
or
For everything� : if it� is a visitor, then it� is a classicist.
By substituting variables one obtains:
For all x� , if x� is a visitor, then x� is a classicist.
�e last formulation is basically the analysis of typical quanti�ed state-
ments. ‘For all’ is translated as the symbol ∀ (a rotated ‘A’ reminding one
of ‘all’).
‘Is a visitor’ is a predicate that is translated as P � , while ‘is a classicist’
is translated as Q � . So ‘x� is a visitor’ is translated as P � x� , and ‘x� is a
classicist’ becomes Q � x� . ‘If . . . , then . . . ’ becomes the arrow, so ‘if x� is a
visitor, then x� is a classicist’ becomes (P � x� → Q � x� ). �us the sentence
‘Every visitor is a classicist’ translates into the following L� -sentence:
∀x� (P � x� → Q � x� ).

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

In addition to ∀ logicians also use the symbol ∃. In the following

sentence an existence claim is made:

At least one visitor is a classicist.

If this is rewritten with variables, then the following expression is ob-

tained:

For at least one x� , x� is a visitor and x� is a classicist.

�is is then formalised as ∃x� (P � x� ∧ Q � x� ).

Later, formalisation in predicate logic will be discussed in more detail,
but I have now motivated all elements of the syntax of predicate logic that
will be introduced in the formal de�nition of a sentence of L� .

�.� ��� ��������� �� L�

In this section the syntax of L� is introduced in a formally precise way.

First I turn to predicate letters.
�e language L� of predicate logic contains �-place predicate letters.
For simplicity, they do not have an arity-index, that is, an upper index;
they are the sentence letters P, Q, R, P� , Q� , and so on. L� also contains
predicate letters with arbitrary arity-index �, �, �, and so on. Having only
one binary, that is, �-place, predicate letter will not su�ce. In order to
formalise a sentence containing the predicate expressions ‘. . . hates . . . ’
and ‘. . . loves . . . ’, one will need two distinct binary predicate letters; and
of course one might also need a third and even more binary predicate
letters. In order to make sure that there is always a su�cient stock of
binary predicate letters, I include in�nitely many binary predicate letters
in L� , namely the expressions P � , Q � , R � , P�� , Q�� , R�� , P�� , and so on. �is
applies not only to binary predicate letters but also to predicate letters
with other arity-indices. �us, the general de�nition of predicate letters
looks like this:

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

���������� �.� (��������� �������). All expressions of the form Pnk , Q nk ,

or R nk are sentence letters, where k and n are either missing (no symbol) or
a numeral ‘�’, ‘�’, ‘�’, . . .
So the letter P with or without numerals ‘�’, ‘�’, and so on as upper
and/or lower indices is a predicate letter, and similarly for Q and R. �e
sentence letters P, Q, R, P� , Q� , . . . are also predicate letters, according to
this de�nition. Furthermore, P � , Q � , R � , P�� , Q�� , R�� , P�� , Q�� , R�� , . . . , P�� , Q�� ,
R�� , P�� , Q�� , R�� , and so on, are predicate letters. �is de�nition ensures
that L� contains in�nitely many n-place predicate letters for any n. Using
only predicate letters with P, but not Q or R, would su�ce, but having a
choice between letters enables me to generate more readable formulae.
�-place predicate letters (sentence letters) have arity �; �-place predi-
cate letters have arity �, and so on:
���������� �.�. �e value of the upper index of a predicate letter is called
its arity. If a predicate letter does not have an upper index its arity is �.
�e predicate letter P�� , for example, has arity �.
�e language L� contains constants, which will be used to translate
English proper names and some similar expressions.
���������� �.� (���������). a, b, c, a� , b� , c� , a� , b� , c� , a� , . . . are con-
stants.
Moreover, L� contains in�nitely many variables.
���������� �.� (���������). x, y, z, x� , y� , z� , x� , . . . are variables.
Now the notion of an atomic L� -formula can be de�ned:
���������� �.� (������ �������� �� L� ). If Z is a predicate letter of
arity n and each of t� , . . . , t n is a variable or a constant, then Zt� . . . t n is
an atomic formula of L� .
In this de�nition, the upper case letter Z serves as a metavariable for
predicate letters, that is, for P, R��
�
, Q � , and the like. According to this
�
de�nition, Q x, P c y, P� x�� c� y, and R � xx are examples of atomic formu-
� �

lae. De�nition �.� allows for the case in which n = �. �is means that all

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

sentence letters, that is, P, Q, R, P� , and so on, are also atomic formulae.
���������� �.�. A quanti�er is an expression ∀v or ∃v where v is a vari-
able.
�us, ∀x��� and ∃z are quanti�ers.�
���������� �.� (�������� �� L� ).
(i) All atomic formulae of L� are formulae of L� .
(ii) If � and ψ are formulae of L� , then ¬�, (� ∧ ψ), (� ∨ ψ), (� → ψ)
and (� ↔ ψ) are formulae of L� .
(iii) If v is a variable and � is a formula, then ∀v � and ∃v � are formulae
of L� .
Examples of formulae of the language L� of predicate logic are:

∀x (P � xa → Q � x),
∀z�� ¬∃y� ∃z�� (P � x y → ∃x� (R�� z�� c� xz�� ∧ Q)),
(∃x P � x ↔ ¬∃y ∃y Q � y y),
∀x ∃z R � az.

�ere is no point in trying to understand these formulae; the point

here is that they all classify as L� -formulae. In order to show that a given
expression is a formula of L� , one can build up the formula step by step
according to the rules laid down in De�nition �.�. As an example I will
show that the last formula, (∃xP � x ↔ ¬∃y∃yQ � y y), is a formula of L� :
�. P � is a predicate letter by De�nition �.� with arity � (De�nition �.�),
and x is a variable by De�nition �.�.
�. �erefore, by De�nition �.�, P � x is an atomic formula.
�. ∃x P � x is thus a formula of L� by De�nition �.�(iii).
�. Similarly, Q � y y is an atomic formula (I will not go through the
tedious reasoning of �. and �. again).

� �ere are alternative symbols for ∀ and ∃, which will not be used here: �v, Πv and (v)
are sometimes used instead of ∀v, and � v and Σv instead of ∃v.

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

�. ∃y Q � y y is a formula of L� by �.�(iii).
�. ∃y ∃y Q � y y is a formula of L� by �.�(iii).
�. ¬∃y ∃y Q � y y is a formula of L� by �.�(ii).
�. (∃x P � x ↔ ¬∃y ∃y Q � y y) is a formula of L� by �.�(ii). �is follows
from the previous item and �.
In cases like this, one will be able to see without a long proof whether an
expression is a formula, and so it will not be necessary to go through all of
these steps. �e above proof of the claim that (∃x P � x ↔ ¬∃y ∃y Q � y y)
is a formula only shows how exactly the de�nition of L� -formulae works.

�.� ���� ��� ����� ����������� �� ���������

In the formula ∀x (P � x → Q � x) the last two occurrences of x refer back

to or depend on the quanti�er ∀x. In the formula P � x → Q � x, by contrast,
there is no quanti�er to which they can refer back; they occur freely, as
logicians say. In the next de�nition this notion of a free occurrence of a
variable is made precise.
���������� �.�.
(i) All occurrences of variables in atomic formulae are free.
(ii) �e occurrences of a variable that are free in � and ψ are also free in
¬�, � ∧ ψ, � ∨ ψ, � → ψ and � ↔ ψ.
(iii) In a formula ∀v � no occurrence of the variable v is free; all occur-
rences of variables other than v that are free in � are also free in
∀v �.
An occurrence of a variable is bound in a formula if and only if it is not
free.
Less formally speaking, occurrences of variables are free as long as
they are not ‘caught’ by a quanti�er. For instance, in the atomic formulae
R � xx or P � x all occurrences of x are free according to clause (i) of the
de�nition, and so, according to clause (ii), all occurrences of x are free in
(R� xx ∨¬P � x) and in ¬(P � x ↔ R � aa). Similarly, all occurrences of x� are
free in (P � x� → Q � x� ), but, according to clause (iii) of De�nition �.�, all

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

occurrences of x� are bound in ∀x� (P � x� → Q � x� ). In ∀y (P � x� → Q � x� )

all occurrences of the variable x� are free because y is a variable di�erent
from x� .
In ¬(Q � z ∧ ∃z R � zz) the �rst occurrence of the variable z is free, while
the remaining occurrences are bound. In ∀x (R � x y → R � xa) ↔ R � ax all
but the last occurrence of x are bound.
���������� �.�. A variable occurs freely in a formula if and only if there
is at least one free occurrence of the variable in the formula.
As pointed out on page ��, ∀x� (P � x� → Q � x� ) is the formalisation of
the sentence ‘Every visitor is a classicist’ with the following logical form:

For everything� : if it� is a visitor, then it� is a classicist.

�e sentence can be true or false depending on the circumstances. �e

formula (P � x� → Q � x� ) corresponds to

(F) if it� is a visitor, then it� is a classicist.

On its own (F) is not a sentence that is true or false: �ere is no quantifying
phrase like ‘for everything� ’ the pronoun ‘it� ’ can refer back to; also, ‘it� ’ is
not a lazy pronoun referring back to a certain designator. �us, (F) does
not have a truth-value. One can only assign a truth-value to (F), if one
makes an arbitrary choice and takes ‘it� ’ to stand for a particular thing.
But without such an arbitrary choice (F) cannot be assigned a truth-value.
�e L� -formulae behave similarly: only formulae without free occur-
rences of variables are sentences; and only sentences will be assigned
truth-values by L� -structures, which will be introduced in the follow-
ing chapter. Also, sentences but not formulae with free occurrences of
variables will be used as premisses and conclusions in arguments.
���������� �.�� (�������� �� L� ). A formula of L� is a sentence of L�
if and only if no variable occurs freely in the formula.
Again informally speaking, in a sentence of L� all occurrences of
variables are ‘caught’ by some quanti�er. �e following are examples of

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

L� -sentences:

∀x (P � x → (Q��
�
xa ∨ ∃x R� xax))
(P � ab ∧ ∃y(P � by ∧ ∀x ¬P � x y))

�.� ���������� �����������

In Section �.� I introduced some conventions for dropping brackets from

sentence of L� . �ese rules did not form part of the o�cial syntax of L� ;
they merely allow one to abbreviate sentences.
In this section I will specify some rules for abbreviating formulae
of L� . Again, they do not form part of the o�cial syntax. Applying the
rules does not yield L� -sentences but rather only abbreviations of L� -
sentences. Like the rules for dropping brackets in L� , the conventions
do not have to be applied: one can always write down the full formula
instead of the abbreviated form.
�e Bracketing Conventions �–� apply also to formulae of L� . �e
quanti�ers have to be taken into account: In the sentence ∃x(P � x ∧ Q � x)
the brackets are not outer brackets, so they cannot be dropped. �e
expression ∃x P � x ∧ Q � x is an abbreviation of the formula (∃x P � x ∧ Q � x),
which is not a sentence, because the second occurrence of x is free.
As a further example I will consider the following L� -sentence:

∀x ((P � x ∧ R�� xa) → ∃y� ((R�� x y� ∧ Q � x) ∧ P � y� )) (�.�)

�is sentence may be abbreviated in the following ways:

∀x (P � x ∧ R�� xa → ∃y� ((R�� x y� ∧ Q � x) ∧ P � y� ))

∀x ((P � x ∧ R�� xa) → ∃y� (R�� x y� ∧ Q � x ∧ P � y� ))
∀x (P � x ∧ R�� xa → ∃y� (R�� x y� ∧ Q � x ∧ P � y� ))

In the �rst line, Bracketing Convention � is applied, in the second Brack-

eting Convention �, and in the third both conventions are applied. �ere
are no further ways of saving brackets.

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

Very o�en the upper index of predicate letters, that is, their arity-
index, is also omitted. �is is due to the fact that there is only one way to
add these upper indices to the predicate letters of an expression that is
supposed to abbreviate a formula. �erefore, sentence (�.�) also has

∀x ((Px ∧ R� xa) → ∃y� ((R� x y� ∧ Qx) ∧ P y� ))

as an abbreviation. So, combined with the rules for dropping brackets,

the most economical form of (�.�) is the following abbreviation:

∀x (Px ∧ R� xa → ∃y� (R� x y� ∧ Qx ∧ P y� ))

�us, when the arity-index is missing, this does not necessarily mean
that the predicate letter is a sentence letter: it could be an abbreviation of
another predicate letter from which the arity-index has been omitted.
Abbreviations of formulae that have been obtained by omitting arity-
indices can be misleading: one might think that ∀x ∀y (Px ↔ Px y)
abbreviates an L� -sentence that contains the same predicate letter twice.
Inserting the missing indices, however, shows that the sentence contains
two di�erent predicate letters, P � and P � :

∀x ∀y (P � x ↔ P � x y)

�erefore, there is only one occurrence of P � in the formula and only

one occurrence of P � . �e abbreviation ∀x ∀y (Px ↔ Px y) is correct
according to the above conventions, but in such cases it may be helpful
to retain the arity-indices.

�.� �������������

�e basic strategy for obtaining the logical form of an English sentence in

predicate logic is the same as in propositional logic (cf Sections �.� and �.�):
a given sentence is analysed from top to bottom. �at is, one starts with the
entire sentence and works one’s way deeper and deeper into the sentence.

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

In contrast to propositional logic, one does not have to stop at quanti�ed

sentences; one can analyse them in the way outlined in Section �.�. In
particular, the logical form of universal sentences is obtained in the way
sketched on page ��.
Rather than going over the general rules again, I will show how the
method works by way of some examples. I have already dealt with simple
sentences like ‘Tom loves Mary’ on page ��, so here I will focus on complex
sentences.
�e following sentence is an example of a universally quanti�ed sen-
tence, that is, a sentence making a claim about all objects of a certain
sort:
All frogs are amphibians.
First, I will determine its logical form. It is clearly a universal claim that
is to be parsed as outlined on page ��:
For all x (if x is a frog, then x is an amphibian).
�e expression in brackets contains the standard connective ‘if . . . , then
. . . ’, so it can be further parsed as follows:
For all x (if (x is a frog), then (x is an amphibian)).
Both ‘x is a frog’ and ‘x is an amphibian’ are enclosed in brackets; they
contain no connectives and are not quanti�ed. ‘Is a frog’ and ‘is an
amphibian’ are then formalised by two distinct predicate letters P � and Q � ,
respectively. So ‘(x is a frog)’ becomes Px (omitting the arity-index);
and ‘x is an amphibian’ becomes Qx. �e expression ‘for all’ becomes
the universal quanti�er ∀, and ‘if . . . , then . . . ’, the arrow →. So, the
formalisation is
∀x (Px → Qx),
with the following dictionary:
P: . . . is a frog
Q: . . . is an amphibian

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

Generally, universal claims can be formalised this way. Completely

unrestricted universal claims are rare, but philosophers occasionally do
make claims like the following, that are meant to be completely general:
Everything is material.
�is sentence can be formalised as ∀x Rx, where R stands for ‘is material’.
Existential claims are usually formalised by the existential quanti-
�er; restrictions to a certain kind of objects is expressed by conjunction.
�erefore, ‘�ere are poisonous frogs’ has the following logical form:
(R) �ere is at least one x ((x is a frog) and (x is poisonous))
�e formalisation is
∃x(Px ∧ Q� x)
with the following dictionary:
P: . . . is a frog,
Q� : . . . is poisonous.
�e English phrase ‘No . . . is . . . .’ can be taken to be a negated existen-
tial quanti�cation. �e sentence
No frog is poisonous.
can be rephrased as
It is not the case that there are poisonous frogs.
‘It is not the case that’ is a standard connective and it is formalised as ¬. I
have already shown how to go about formalising ‘�ere are poisonous
frogs’. So the sentence ‘No frog is poisonous’ is formalised as the following
sentence, with the same dictionary as above:
¬∃x (Px ∧ Q� x) (�.�)
Alternatively, one could have rephrased the original sentence ‘No frog is
poisonous’ as the following sentence:

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

All frogs are non-poisonous.

�is formalises into ∀x (Px → ¬Q� x). �is sentence and the alternative
formalisation (�.�) are logically equivalent under the semantics I will
expound in Chapter �. Both formalisations are equally sound.
�e formalisation of the following sentence requires two quanti�ers:

Every student has a computer.

�is is clearly a universal claim; so in the �rst step one obtains:

For all x (if x is a student, then x has a computer).

�is is not yet the full logical form of the sentence: ‘x has a computer’
contains an existential claim and can be further analysed as

there is at least one y (x has y and y is a computer).

Now ‘x has y’ and ‘y is a computer’ cannot be further analysed and so

they are put in brackets:

there is at least one y ((x has y) and (y is a computer)).

�us the full logical form of ‘Every student has a computer’ is

For all x �if (x is a student), then there is at least one y

((x has y) and (y is a computer))�.

�e formalisation is now straightforward:

∀x (Px → ∃y(Rx y ∧ Q y))

�e dictionary is speci�ed in the following way:

P: . . . is a student
Q: . . . is a computer
R: . . . has . . .

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

Generally, the dictionary must provide translations of all sentence letters,

predicate letters and constants occurring in the formalisation. However,
the dictionary must not contain translations for the variables. Variables in
sentences never refer to particular objects; they are only used for making
universal or existence claims.
Here is a somewhat more complicated example:
If it’s raining, then Bill reads a book or a newspaper.

�If (it’s raining), then there is at least one x ((Bill reads x)

and ((x is a book) or (x is a newspaper))�.

�e proper name ‘Bill’ is translated as a constant; the sentence ‘It’s raining’

is translated as a sentence letter:

P → ∃x (P � ax ∧ (Qx ∨ Rx)).

I have dropped the outer brackets according to Bracketing Convention �.

In the dictionary I have restored all arity-indices. In particular, one must
avoid any confusions between the sentence letter (�-place predicate letter)
P and the �-place predicate letter P � .
P: it’s raining,
Q �: . . . is a book,
R� : . . . is a newpaper,
P�: . . . reads . . .
Ternary predicate letters are needed for formalising sentences such as
the following:
�ere is a country between Spain and France.
�e logical form of this sentence is
�ere is at least one x �(x is a country) and (x is between
Spain and France)�.

© Volker Halbach ����/����

 � �e Syntax of Predicate Logic ��

By formalising this, one obtains the following sentence of predicate logic:

∃x (Px ∧ Qxbc).

�e dictionary is as follows:

P: . . . is a country
Q: . . . is between . . . and . . .
b: Spain
c: France
Using the techniques outlined so far, one can formalise fairly com-
plicated sentences. �ere are, however, some problem cases. Before
discussing more intricate problems of formalisation in Chapter �, I shall
introduce the semantics of predicate logic. Without having discussed
the semantics of L� �rst, it would be di�cult to judge the soundness of
translations between English and the language L� .

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic

Discussions in metaphysics and in other areas in philosophy have been

spurred by investigations into semantics. Whereas the semantics of the
language L� of propositional logic is somewhat crude and philosophically
not very exciting, the semantics of the language L� of predicate logic
touches upon questions that are at the core of old debates in metaphysics.
In this chapter I shall con�ne myself to the technical core of the se-
mantics of L� . �e philosophical issues will resurface in later discussions
about translating English sentences into sentences of L� . �e technical
account that I am going to present can be traced back to Tarski (����) and
subsequent work by Tarski, although I will deviate from Tarski’s original
approach in many details and in my notation. Tarski’s de�nition of truth
had a profound in�uence on many areas not only in philosophy, but also
in mathematical logic, linguistics and computer science.
In English, phrases such as ‘Paris’ or ‘Julius Caesar’, which are usually
formalised as constants, and predicate expressions such as ‘is tired’ or
‘loves’, have �xed meanings. In the language L� of predicate logic, the
constants and predicate letters will not be assigned �xed meanings. �is
is not because it is not possible to assign �xed meanings to them, but
rather because the validity of arguments or the property of being logically
true do not depend on the particular meanings of constants and predicate
letters. A sentence of the language L� will be de�ned to be logically
true, for instance, if and only if it is true under any interpretation of the
constants and predicate letters. �us any particular interpretations of
constants and predicate letters do not matter for logical truth. A similar
remark applies to validity: an argument in L� will be de�ned to be valid
if and only if there is no interpretation under which the premisses are all

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ��

true and the conclusion is false. �us, again, the validity of an argument
does not depend on any speci�c interpretation we could assign to the
constants and predicate letters.
As in the case of propositional logic, the notion of an interpretation
from Characterisation �.� will be made precise by the notion of a structure:
structures provide interpretations for the non-logical, subject-speci�c
vocabulary, that is, for predicate letters and constants. �e interpretation
that is assigned to a symbols by a structure is called the ‘semantic value’
or the ‘extension’ of the symbol in the structure.

�.� ����������

�e semantics of the language L� will be given in the few de�nitions

in italics in this chapter. �e bulk of the chapter is only an attempt to
motivate and elucidate these de�nitions.
I start by looking back at the semantics of the language L� of proposi-
tional logic: whether a sentence of the language L� is true depends on the
truth or falsity of the sentence letters in that sentence. �e truth-values of
all the sentence letters are given by an L� -structure. �en the truth tables
of the connectives allow one to calculate the truth-values of sentences
formed with connectives.
Structures for predicate logic are more complicated: L� -structures
need to determine more than merely the truth-values of sentence letters
because the language L� contains also other symbols, namely predicate
letters and constants. L� -structures assign semantic values to these sym-
bols as well. Sentence letters will receive truth-values as their semantic
values in the same way as in propositional logic, but predicate letters will
be assigned semantic values of a di�erent kind.
Whether a sentence of L� is true or false does not only depend on the
semantic values of the constants and the sentence and predicate letters, but
also over which objects the quanti�ers are taken to range. �is situation
is similar to English: the truth-value of the English sentence ‘All glasses

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ��

are empty’ depends in part on whether the expression ‘all glasses’ is taken
to range only over the glasses on a particular table or in a particular room
or over all glasses in the world. �e sentence is usually uttered when one
is talking about particular glasses.
�us, one of the things that an L� -structure does is specify a domain
of discourse, which is just some non-empty set of objects. �ere are
no restrictions on the domain of discourse except that it must not be
empty.� If A is an L� -structure, I will write DA for the structure’s domain
of discourse.
In the language L� , constants play a role comparable to proper names
in English, and in English proper names refer to objects: the English
proper name ‘Rome’ refers to (or ‘denotes’) the capital of Italy, ‘Volker
Halbach’ refers to Volker Halbach, and so on. �us, an L� -structure
assigns elements of the domain of discourse to the constants as their
semantic values.
Sentence letters are treated as in propositional logic: they receive
truth-values, that is, either T or F, as semantic values in an L� -structure.
Hence, an L� -structure contains also an L� -structure.
Unary (�-place) predicate letters correspond to English expressions
such as ‘is green’, ‘walks’, or ‘is a philosopher’. Unary predicate letters
have sets as their semantic values. �e predicate letter P � , for instance,
can have as its semantic value the set of all green objects (or the set of
all walking objects, or the set of all philosophers, or the empty set). On
page ��, sets were conceived of as unary relations; so predicate letters
have unary relations as semantic values, and an L� -structure must assign
unary relations to unary predicate letters.

� Empty domains are not allowed in the traditional accounts of semantics for predicate
logic. Admitting the empty domain would make the semantics for L� more clumsy, but it
is perfectly possible to admit them. From a philosophical point of view it would probably
be more satisfying to admit the empty domain, but I want to avoid the additional
technical complications, and I shall therefore follow the traditional account. �e e�ects
of the exclusion of the empty domain will be explained below by means of examples.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ��

Binary predicate letters correspond to expressions such as ‘loves’ or

‘is bigger than’. Binary predicate letters are interpreted by binary relations,
that is, by sets of ordered pairs. �e predicate letter P � , for instance, can
have the relation of loving, that is, the set of all ordered pairs �d, e� such
that d loves e, as its semantic value. �us, an L� -structure must assign
binary relations to binary predicate letters as their semantic values.
Analogously, �-place predicate letters are interpreted by �-place rela-
tions, that is, sets of triples, and generally predicate letters with arity n
are assigned n-ary relations (see Section �.�).
In the following list I summarise which objects are assigned to expres-
sions of L� by an L� -structure as their semantic values or ‘extensions’.

L� -expression semantic value

constant object
sentence letter truth-value
unary predicate letter set, unary relation
binary predicate letter binary relation
(= set of ordered pairs)
predicate letter of arity � �-place relation
(= set of triples)
⋮ ⋮
In sum, an L� -structure speci�es a non-empty set as domain of dis-
course, it assigns elements of the domain to constants, it assigns a truth-
value to every sentence letter, and it assigns an n-ary relation to every
predicate letter.
�e de�nition of an L� -structure can be spelled out more precisely
in technical terms. I mention this de�nition only for the sake of those
readers who want the full story. I shall not make use of this de�nition in
what follows.
���������� �.� (L� -���������). An L� -structure is an ordered pair �D, I�
where D is some non-empty set and I is a function from the set of all con-

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ��

stants, sentence letters and predicate letters such that the value of every con-
stant is an element of D, the value of every sentence letter is a truth-value
T or F, and the value of every n-ary predicate letter is an n-ary relation.
One might wonder why variables are not mentioned in the de�nition
of an L� -structure. But just as ‘he’ does not stand for a particular object
in the general claim ‘If a reader is perplexed, he stops reading’, a bound
variable does not stand for a particular object in a sentence of L� . For this
reason, L� -structures do not assign semantic values to variables.
For technical reasons, however, it is convenient to have semantic
values not only for sentences but also for formulae with occurrences of
free variables. Formulae with occurrences of free variables will also be
assigned truth-values as semantic values. Whether a formula like P � x
with an occurrence of a free variable will receive the truth-value T or F
depends on what the variables stand for in the same way ‘He stops reading’
is true or false for some persons but not for others. More than one variable
may occur freely in a formula of L� : whether the formula R � x y ∧ R � z
receives the truth-value T or F depends on what the variables x, y, and
z stand for. In addition to L� -structures, I therefore introduce a list that
assigns an object to every variable of L� . �is list a�ects only the truth or
falsity of formulae with occurrences of free variables, but it does not a�ect
the truth or falsity of sentences (that is, formulae with no free variables).
A variable assignment over an L� -structure A assigns to each variable
an element of the domain DA of A.� Occasionally I will drop the speci�-
cation ‘over the L� -structure A’, when it is clear from the context which
L� -structure is meant.
One may think of a variable assignment as a table with two lines
that has all variables as entries in the �rst line line, and elements of the
domain of discourse as entries in the other line. For instance, there is a
variable assignment α over an L� -structure with the set of all European

� More formally, one can take a variable assignment over D to be a function from the set
of all variables into D.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ��

cities as domain, that assigns Rome to x, y� , and z� , Paris to y, Berlin to

z, London to x� , Oslo to x� . �e assignment α assigns elements to all
further variables, but of course I cannot specify an in�nite list here. �e
beginning of the variable assignment α may be visualised as follows:

x y z x� y� z� x�
�
Rome Paris Berlin London Rome Rome Oslo
An L� -structure A and a variable assignment over A together assign
semantic values to every variable, constant, sentence letter, and predicate
letter. I will write �e�Aα
for the semantic value of the expression e in the
L� -structure A under the variable assignment α over A. �us, for any L� -
structure A and any variable assignment α over A the semantic values of
the respective L� -expressions are as follows:
(i) For any constant t, �t�Aα
is the object in the domain DA of A assigned
to t by A.
(ii) For any variable v, �v�Aα
is the object in DA assigned to the variable v
by the variable assignment α.
(iii) For any sentence letter Φ, �Φ�A α
is the truth-value (either T or F)
assigned to Φ by A.
(iv) For any unary predicate letter Φ, �Φ�A α
is the unary relation, that is,
the set, assigned to Φ by A.
(v) For any binary predicate letter Φ, �Φ�A α
is the binary relation, that
is, the set of ordered pairs, assigned to Φ by A.
(vi) For any �-ary predicate letter Φ, �Φ�A α
is the �-ary relation, that is,
the set of ordered triples, assigned to Φ by A.
And so on for predicate letters of higher arity.�

� �erefore, if A is the ordered pair �D, I�, then for all constants, and sentence and
predicate letters Φ, �Φ�A
α
= I(Φ). �is is what is expressed by (i) and (iii)–(vi).

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ��

�.� �����

�e function � . . . �A α
gives semantic values for all variables, constants, sen-
tence letters, and predicate letters. In this section, � . . . �A α
will be extended
to cover complex formulae as well, that is, formulae that are not mere
sentence letters.
�e de�nition in which truth values are assigned to formulae with
connectives and quanti�ers will be inductive. �at is, �rst I shall de-
�ne � . . . �A
α
for atomic formulae (De�nition �.�), and then I shall de�ne
the semantic values of formulae containing connectives and quanti�ers.
Any formula of L� is either true or false in an L� -structure A under a
variable assignment α over the L� -structure A. �us, for any formula �
either ���A α
= T or ���A
α
= F obtains (but not both). ���A α
= T is o�en read
as ‘α satis�es � in A’. �is use of the term ‘satis�es’ is motivated by its
�

use in �gures of speech like ‘the property of being red is satis�ed by the
apple’ or ‘the equation x � = y is satis�ed by � and �’; the only di�erence
is that α is not a single object but rather an entire sequence of objects
providing semantic values for all variables.
�e truth-value of atomic formulae, that is, of sentences such as P � b or
R xc, is de�ned in the following way: P � b, for instance, is true if and only
�

if the object assigned to b is in the extension (semantic value) of P � , that

is, if �b�A
α
is an element of the set �P � �A α
. Similarly, R � xc is true if and only
if the ordered pair ��x�A α
, �c�A
α
�, that is, the ordered pair with the value of x
as its �rst component and the extension of c as its second component,
is in the extension of R � , that is, in the relation �R � �A . �erefore, the
variable assignment α impinges on the truth-values of formulae with
free occurrences of variables, such as R � xc, because �x�A α
is given by the
variable assignment α.
In the sentence P � a the unary predicate letter P � receives a unary
relation, that is, some set, as its extension (semantic value). I shall assume

� Many authors prefer to write A � �[α] or something similar to express that α satis�es
the formula � in A.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

that �P � �A
α
is the set {Rome, London, Paris} and �a�A α
is Rome. On that
assumption, P a is true because Rome is in the set {Rome, London, Paris},
�

that is, �a�Aα

∈ �P � �Aα
. �e case of unary predicate letters is covered by
the following clause because �a�A α
is the same as ��a�A α
� according to the
assumption on page �� where it has been stipulated that d and �d� are the
same for any object d.
(i) �Φt� . . . t n �A
α
= T if and only if ��t� �A
α
, . . . , �t n �A
α
� ∈ �Φ�Aα
, where Φ is an
n-ary predicate letter (n must be � or higher), and each of t� , . . . , t n
is either a variable or a constant.
�us, this clause determines whether a variable assignment satis�es a
formula like P � a, P � x, R � x y, or Q � xc y in a structure.
If a formula is built up by means of connectives from other sentences,
then truth-values can be assigned to this formula in the style of the De-
�nition �.� of truth in an L� -structure: for instance, if the L� -formulae
� and ψ both have semantic value T, then the formula � ∧ ψ should also
have truth-value T; otherwise it should have F as its extension (semantic
value). �us, a variable assignment α satis�es the formula � ∧ ψ in an
L� -structure, if and only if α satis�es � and ψ in that structure. Similarly
a variable assignment α satis�es a formula ¬� in an L� -structure, if and
only if α does not satisfy � itself in the structure. �is can be expressed
more formally by the following two de�nitional clauses:
(ii) �¬��Aα
= T if and only if ���A α
= F.
(iii) �� ∧ ψ�A = T if and only if ���A
α α
= T and �ψ�A α
= T.
�e clauses (iv)–(vi) for the remaining three connectives ∨, →, and ↔ are
similar and will be listed below.
It remains to de�ne the semantic value, that is, the truth-value, of
quanti�ed formulae from the semantic values of shorter formulae. �at
is, I want to �nd clauses analogous to (i) and (ii) for quanti�ers. As an
example I consider the following L� -sentence:
∃x Rx y
When should a variable assignment α satisfy this formula in a structure A?

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

Assume, for instance, that the variable assignment α looks like this:
x y z x� y� z� x�
�
Rome Paris Berlin London Rome Rome Oslo
In this case �y�Aα
is Paris.
Assuming that R is translated as ‘. . . is smaller than . . . ’, the formula
∃x Rx y corresponds to the English phrase ‘�ere is something smaller
than it’. �e pronoun ‘it’ plays the role of the variable y that occurs freely
in ∃x Rx y. Now ‘�ere is something smaller than it’ is satis�ed by Rome
(which is assigned to y by α) if there is something smaller than Rome,
that is, if there is something (for ‘it� ’) satisfying ‘it� is smaller than it� ’
when ‘it� ’ is taken to stand for Rome.
One can express this more precisely and perspicuously in terms of
variable assignments for the formal language L� : the variable assignment
α satis�es ∃x Rx y if and only if there is a variable assignment β satisfying
Rx y that di�ers from α at most in what is assigned to x. �ere is such
a variable assignment β, assuming that �R�A α
is a relation containing the
pair �Oslo, Rome�:
x y z x� y� z� x�
�
Oslo Paris Berlin London Rome Rome Oslo
�is variable assignment di�ers from α only in the entry for x. Since
there is such a variable assignment, α satis�es the formulae ∃xRx y (in a
structure A where �Oslo, Rome� is an element of the extension �R�A α
of
R).
Of course the variable assignment β must not di�er in the entry for
y, as the question is whether the variable assignment α satis�es ∃xRx y,
that is, whether ∃xRx y is true when y is taken to stand for Rome.
Generally, a variable assignment α satis�es a formula ∃x� if and only
if there is a variable assignment β satisfying � that di�ers from α only
in the entry for x. � may have free occurrences of other variables than
y; for this reason β must agree with α on all variables with the possible
exception of x.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

So I de�ne for all variables v and formulae �:

(viii) �∃v ��A = T if and only if ���A = T for at least one variable assign-
α β

ment β over A di�ering from α in v at most.

By saying that α di�ers from β in v at most, I mean that �u�A = �u�A
α β

for all variables u with the possible exception of v. Hence, in terms of

tables, a variable assignment α and a variable assignment β di�er in a
given variable v at most, if the they agree in all columns with the possible
exception of the column for the variable v. �e two tables above are an
example of two variable assignments di�ering in x.
Universal quanti�ers can be treated in a similar way. When should
one say, for instance, that a variable assignment α satis�es the formula
∀y (Rx y ∧ Ryz) in a structure A? �at is, when should the following
obtain?
�∀y (Rx y ∧ Ryz)�A α
=T
∀y expresses generality. α will satisfy ∀y (Rx y ∧ Ryz) in A if and only
if everything in the domain of A will make Rx y ∧ Ryz true if it is taken
to stand for y (with the values of x and z unchanged from α). �us, α
satis�es ∀y (Rx y ∧ Ryz) in A if and only if every β that di�ers from α
only in y satis�es Rx y ∧ Ryz in A.
�is can be generalised to all variables v and L� -formulae �: a variable
assignment α satis�es a formula ∀v� in a structure A if and only if every
variable assignment β that di�ers from α at most in v satis�es �. Of course,
only variable assignments over A are considered: the variable assignments
can only assign objects from the domain of A to the variables.
�e general clause can now be stated as follows:
(vii) �∀v ��A = T if and only if ���A = T for all variable assignments β
α β

over A di�ering from α in v at most.

I will now collect the di�erent clauses into a de�nition of satisfaction.
Given an L� -structure, this de�nition determines for any variable assign-
ment α and any L� -formula whether α satis�es � in A, that is, whether
���A
α
= T or ���A
α
= F.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

���������� �.� (������������). Assume A is an L� -structure, α is a vari-

able assignment over A, � and ψ are formulae of L� , and v is a variable.
For a formula � either ���A α
= T or ���A α
= F obtains. Formulae other than
sentence letters then receive the following semantic values:
(i) �Φt� . . . t n �A
α
= T if and only if ��t� �A
α
, . . . , �t n �A
α
� ∈ �Φ�A
α
, where Φ is a
n-ary predicate letter (n must be � or higher), and each of t� , . . . , t n
is either a variable or a constant.
(ii) �¬��Aα
= T if and only if ���A α
= F.
(iii) �� ∧ ψ�A = T if and only if ���A
α α
= T and �ψ�A α
= T.
(iv) �� ∨ ψ�A = T if and only if ���A = T or �ψ�A = T.
α α α

(v) �� → ψ�A α
= T if and only if ���Aα
= F or �ψ�A α
= T.
(vi) �� ↔ ψ�A = T if and only if ���A = �ψ�A .
α α α

(vii) �∀v ��A = T if and only if ���A = T for all variable assignments β
α β

over A di�ering from α in v at most.

(viii) �∃v ��A = T if and only if ���A = T for at least one variable assign-
α β

ment β over A di�ering from α in v at most.

In general, what α assigns to variables not occurring freely in a formula
� does not impinge on whether α satis�es � in A. So, if ���A α
= T, and if
β is a variable assignment di�ering from α only in variables that do not
occur freely in �, then also ���A = T. In particular, the variable v does
β

not occur freely in a formula of the form ∃v ψ. �us, α satis�es ∃v ψ in

A independently of what α assigns to the variable v. A similar remark
applies to formulae with a universal quanti�er.
If � is a sentence, that is, if no variable occurs freely in � then ���A
α

does not depend on the variable assignment α at all at all. Hence, if

� is a sentence, then ���A is the same truth value as ���A for all variable
α β

assignments β over A.
In order to simplify the notation one may drop the index for the
variable assignment and write ���A if � is a sentence. In general, one can
drop the variable assignment when the semantic value is the same for
all variable-assignments. �is is the case for constants, sentence letters,
predicate letters, and sentences.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

Truth in an L� -structure is now de�ned in terms of satisfaction:

���������� �.� (�����). A sentence � is true in an L� -structure A if and
only if ���A
α
= T for all variable assignments α over A.
As pointed out above, the truth-value of a sentence in a structure is the
same for all variable assignments. �erefore, if a sentence is satis�ed (in a
structure) by some variable assignment, it will be satis�ed by all variable
assignments. Consequently, a sentence � is true in an L� -structure A if
and only if ���Aα
= T for at least one variable assignment α over A.
�e De�nition �.� of truth has generated and continues to generate
extensive discussion. �e views on its philosophical value di�er wildly.
At any rate, De�nition �.� has been a big success as a tool in philosophy,
mathematics, computer science and linguistics. �e extent to which this
de�nition can also be adapted to natural languages such as English is also
a matter of some controversy.
As an example I will consider a speci�c L� -structure, which I call E.
Its domain of discourse is the set of all European cities. It assigns the
set {Florence, Stockholm, Barcelona} to Q � , the relation of being smaller
than to R � , Florence to a, and London to b. �is information can be
displayed in the following way:
�Q � �E = {Florence, Stockholm, Barcelona},
�R� �E = � �d, e�∶ d is smaller than e �,
�a�E = Florence,
�b�E = London.
�us, �R � �E is the set {�Florence, London�, � Florence, Birmingham�,. . . }.
I have dropped the index for the variable assignment and written �Q � �E
rather than �Q � �Eα since variable assignments do not a�ect the semantic
values of predicate letters and constants.�
������� �.�. �e sentence R � ab is true in E.

� What is assigned to other constants, sentence and predicate letters is irrelevant for the

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

Proof. Since Florence is smaller than London, the pair �Florence, London�
is an element of the relation of being smaller than, and I can reason as
follows (the comments to the right explain what justi�es the proof step
on the le�):

�Florence, London� ∈ � �d, e�∶ d is smaller than e �

��a�E , �b�E � ∈ �R � �E de�nition of E
�R � ab�E = T De�nition �.�(i)

�is shows that R � ab is true in E.

������� �.�. �e sentence ∀x (Q � x → R � xb) is true in E.
Proof. Let α be an arbitrary variable assignment. I distinguish two cases.
First case: �x�Eα is in �Q � �E , that is, �x�Eα is either Florence, Stockholm, or
Barcelona. As all three cities are smaller than London, and �b�E is London,
one has the following:

��x�Eα , �b�E � ∈ �R� �E

�R� xb�Eα = T De�nition �.�(i)
�Q � x → R � xb�Eα = T De�nition �.�(v)

Second case: �x�Eα is not in �Q � �E . In this case one can proceed as

follows:

�x�Eα is not in �Q � �E
�Q � x�Eα = F De�nition �.�(i)
�Q � x → R � xb�Eα = T De�nition �.�(v)

following. For the sake of de�niteness, I could stipulate that E assigns the empty set
as extension to all predicate letters other than Q � and R � , T to all sentence letters, and
Rome to all constants other than a and b.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

�erefore, in both cases, that is, for every variable assignment α over E,
the following obtains:
�Q � x → R � xb�Eα = T.
Consequently, according to De�nition �.�(vii), �∀x (Q � x → R � xb)�E = T
β

for every variable assignment β. Hence, by De�nition �.�, the sentence

∀x (Q � x → R � xb) is true in E.
�e �nal example sentence contains two quanti�ers.
������� �.�. �e sentence ∀x ∃y (R� x y ∨ R � yx) is true in E.
Proof. Let α be an arbitrary variable assignment over E.
First case: �x�Eα is not London (the largest city in Europe). �en change
the entry for y into London (if it is not already London) and call the
resulting variable assignment β. By de�nition, β di�ers from α in y at
most. Since every European city except London itself is smaller than
London, one has the following:
��x�Eβ , �y�Eβ � ∈ �R� �E
�R � x y�E = T
β
De�nition �.�(i)
�R � x y ∨ R � yx�E = T
β
De�nition �.�(iv)
�∃y (R � x y ∨ R � yx)�Eα = T De�nition �.�(viii)
�e last line holds because β di�ers from α at most in y.
Second case: �x�Eα is London. Change the entry for y in α into Florence
(or any other European city smaller than London), and call this variable
assignment β; it di�ers from α only in y. �e �rst of the following lines
holds because �y�E is Florence, which is smaller than �x�Eα , that is, London:
β

��y�Eβ , �x�Eβ � ∈ �R � �E
�R� yx�E = T
β
De�nition �.�(i)
�R� x y ∨ R � yx�E = T
β
De�nition �.�(iv)
�∃y (R � x y ∨ R � yx)�Eα = T De�nition �.�(viii)

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

�e last line holds, because α di�ers from β in y at most.

�erefore, I have proved that �∃y(R� x y ∨ R � yx)�Eα = T for any variable
assignment α over E. According to De�nition �.�(vii) this implies by that

�∀x∃y (R� x y ∨ R � yx)�E = T,

which shows that the sentence is true in E.

In practice hardly anyone will go through all these steps explicitly.
�e foregoing examples should have illustrated how the semantics for
the language of predicate logic works and how the truth or falsity of all
sentences is determined by an L� -structure.

�.� ��������, ������� ������, ��� ��������������

With the de�nition of truth in hand one can now de�ne such notions as
logical truth, contradiction, the validity of an argument, and so on. �e
following de�nition is analogous to De�nition �.� for propositional logic.
���������� �.�.
(i) A sentence � of L� is logically true if and only if � is true in all L� -
structures.
(ii) A sentence � of L� is a contradiction if and only if � is not true in
any L� -structure.
(iii) A sentence � and a sentence ψ are logically equivalent if both are
true in exactly the same L� -structures.
(iv) A set Γ of L� -sentences is semantically consistent if and only if there
is an L� -structure A in which all sentences in Γ are true. As in propo-
sitional logic, a set of L� -sentences is semantically inconsistent if and
only if it is not semantically consistent.
Also, the de�nition of validity of an argument in L� follows the pattern
set out in the de�nition of validity of an argument in propositional logic,
that is, in De�nition �.�.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

���������� �.�. Let à be a set of sentences of L� and � a sentence of L� .

�e argument with all sentences in Γ as premisses and � as conclusion is
valid if and only if there is no L� -structure in which all sentences in Γ are
true and � is false.
�is just captures the intuitive idea that an argument is valid if and
only if any L� -structure that makes the premisses true also makes the
conclusion true.
�at the argument with all sentences in Γ as premisses and � as con-
clusion is valid, is abbreviated as Γ � �. Instead of Γ � � one can also say
‘� follows from Γ’. �e symbol � is the negation of �; so, it is de�ned as
follows: Γ � � if and only if not Γ � �.
I have not excluded the possibility that there is not premiss at all in
an argument. So, Γ may be the empty set. If Γ is the empty set and Γ � �,
one may also simply write � �. As I have said above, Γ � � means that
� is true in all L� -structures in which all sentences in Γ are true. Now if
there are no sentences in Γ, one has Γ � � if and only if � is true in all
structures. Consequently, � � means that � is true in all L� -structures
tout court; that is, it means that � is logically true.
I introduce a further notational convention, already adopted for propo-
sitional logic: When the sentences in Γ are written out explicitly, the set
brackets around the sentences may be dropped: For instance, one may
write
∀x Qx, ∀x (Qx → Rx) � ∃x Rx
rather than the following:
�∀x Qx, ∀x (Qx → Rx)� � ∃x Rx.

�.� ���������������

How can one show that an argument in is valid? And how can one show
that it is not valid?
For the language L� of propositional logic, these question are usually
easily answered (if there are not too many or too long sentences involved):

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

One can check out whether an argument in L� is valid by means of a truth

table.
Generally, in the language L� , it is much harder to show that an
argument is valid or not. �ere is not a �nite set of possibilities one
can check out in order to �nd out whether an argument is valid or not.
�ere are in�nitely many domains of discourse, and even a single binary
predicate letter can be interpreted by in�nitely many binary relations.
Showing that an L� -argument is valid by proving that the conclusion is
true in all L� -structures in which all the premisses are true is, therefore,
usually a di�cult task. A more e�cient and elegant way of establishing
that an argument in L� is valid will be introduced in the next chapter.
In order to show that an argument is not valid, however, one does
not have to prove something about all L� -structures; one has only to �nd
an L� -structure in which all premisses of the argument are true and its
conclusion is false. Such L� -structures are called counterexamples. In
this section I will explain how to use counterexamples to disprove the
validity of arguments.
An L� -structure A is a counterexample to an argument if and only if
all premisses of the argument are true in A and the conclusion is false in A.
As explained above, a sentence is logically valid if and only if the
argument with no premisses and the sentence as its conclusion is valid.
�us, one can use counterexamples to show that a sentence is not logically
true:
An L� -structure is a counterexample to an L� -sentence if the sentence
is not true in it. An L� -sentence is logically true if and only if there are no
counterexamples to it.
I will prove the following claim by means of a counterexample.
������� �.�. �e sentence Qb → ∀x Qx is not logically true.
In order to �nd a counterexample to this sentence, one could reason
informally as follows: b could satisfy Q, but other objects might not
satisfy Q and thus ∀x Qx would be false. One can turn this into a proof.
First, an L� -structure with a domain of discourse containing at least two

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

objects is required. And then, the object that is the semantic value of b
needs to be in the set that is the semantic value of Q, while one of the
other objects is not in this set. Now this can be turned into a proper proof
of the claim that Qb → ∀x Qx is not logically true:
Proof. Let B be an L� -structure with the set {�, �} as its domain of dis-
course and the following semantic values for Q and b:�

�Q�B = {�},
�b�B = �.

I will now show that Qb → ∀x Qx receives the semantic value F

in this L� -structure. Let α be the variable assignment that assigns � to
every variable, so �x�Bα = �. Now one can reason as follows, using ∉ as an
abbreviation for ‘is not an element of’:

� ∈� {�}
�x�Bα ∈� �Q�B de�nition of α and B
�Qx�Bα = F De�nition �.�(i)
�∀x Qx�Bα = F De�nition �.�(vii)
� ∈ {�}
�b�B ∈ �Q�B de�nition of B
�Qb�Bα = T De�nition �.�(i)
�Qb → ∀x Qx�Bα = F De�nition �.�(v)

By the De�nition �.� of truth, Qb → ∀x Qx is not true in B and thus,

according to De�nition �.�(i), Qb → ∀x Qx is not logically true.

� For the sake of de�niteness I should specify also the value of other constants, sentence
and predicate letters. But as they do not make a di�erence to the truth-values of sentences,
I will not specify them in this and the following examples.

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

In order to show that Qb → ∀x Qx is not logically true, I could have

employed objects other than the numbers � and �. �e two numbers are
convenient because of their short names. �ere is no need to use more
fancy objects. Generally, it is sensible to keep things simple by choosing
small domains of discourse. In some cases, however, it may be necessary
to use large domains; there are even cases where the domain has to be
in�nite.
Next I will turn to an argument.
������� �.��. ∀x ∃y Rx y � ∃y ∀x Rx y.
�e premiss could be the translation of a sentence such as ‘For every-
thing there is something with the same mass’; the conclusion would then
be the translation of ‘�ere is something that has the same mass as any
object’. If there are exactly two things di�ering in mass, then ‘For every-
thing there is something with the same mass’ is true, because every object
agrees with itself in its mass, and the conclusion is false, because the mass
of neither of the two objects matches the mass of the other object. Hence,
one can use an L� -structure with a domain containing exactly two objects.
R needs to have a relation as extension that relates every object to itself
but not to the other object in the domain.
Proof. �e L� -structure C is de�ned as follows:

DC = {the sun, the moon},

�R�C = ��the sun, the sun�, �the moon, the moon��.

First I will show that the premiss is true in the L� -structure C. Let α be
an arbitrary variable assignment over C. �en change the value of x so
that the values of x and y are the same, that is, change the entry for x into
the sun if �y�Cα is the sun and into the moon if �y�Cα is the moon; call the
resulting variable assignment β. �e �rst line in the following proof, then,

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

holds in virtue of the de�nition of �R�C , and because �x�C = �y�C .

β β

��x�Cβ , �y�Cβ � ∈ �R�C

�Rx y�C = T
β
De�nition �.�(i)
�∃y Rx y�C=T
β
De�nition �.�(viii)
�∀x ∃y Rx y�Cα = T De�nition �.�(vii)
�e last line holds because the foregoing reasoning applies to all variable
assignments α. Hence, the premiss is true in C.
It remains to show that the conclusion is false in C. Assume to the
contrary that ∃y ∀x Rx y is true in C. �en, by De�nition �.�(viii), there
is a variable assignment α such that the following holds:
�∀x Rx y�Cα = T.
�us, by De�nition �.�(vii),
�Rx y�C = T
β

for every variable assignment β that di�ers from α at most in x. But this
is not the case since one can choose a variable assignment β such that
�x�C is di�erent from �y�C and so ��x�C , �y�C � is not in �R�C .
β β β β

Since the premiss is true in C and the conclusion is false in C, the

argument is not valid.
������� �.��. ∀x (Px → Qx ∨ Rx), Pa � Ra.
To motivate the counterexample below one can reason as follows: �e
premiss Pa must be true in the counterexample, call it D, so �a�D must
be in �P�D . �e premiss ∀x(Px → Qx ∨ Rx) therefore implies that �a�D is
either in �Q�D or in �R�D . So if the premiss is to be true at least one of the
latter must be the case. As the conclusion Ra must be false, �a�D must
not be in �R�D , and, therefore, �a�D must be an element of �Q�D . So, one
can employ a counterexample with a single object in its domain, where
that object is in the extensions of P and Q while the extension of R is the
empty set:

© Volker Halbach ����/����

 � �e Semantics of Predicate Logic ���

Proof. �e following L� -structure is a counterexample:

DD = {�},
�a�D = �,
�P�D = {�},
�Q�D = {�},
�R�D = �.
�ere is only one variable assignment over D because its domain DD
contains only one object and every variable is assigned the number �.
�erefore, �x�Dα
= � for all variable assignments α. To show that the �rst
premiss is true, one can reason as follows:
� ∈ {�}
�x�D
α
∈ �Q�D de�nitions of α and D
�Qx�Dα
=T De�nition �.�(i)
�Qx ∨ Rx�Dα
=T De�nition �.�(iv)
�Px → Qx ∨ Rx�Dα
=T De�nition �.�(v)
Since this holds for all variable assignments over D, as α is the only such
variable assignment, ∀x (Px → Qx ∨ Rx) is true in D.
�e second premiss is also true in D:
� ∈ {�}
�a�D ∈ �P�D de�nitions of D
�Pa�D = T De�nition �.�(i)
�e conclusion, however, is false in D:
�∉�
�a�D ∉ �R�D de�nitions of D
�Ra�D = F De�nition �.�(i)
�is shows that both premisses are true and that the conclusion is false in
D. �erefore, the argument is not valid.

© Volker Halbach ����/����

 � Natural Deduction

A valid argument need not be obviously valid. One can establish the
validity of such an argument by breaking it into smaller arguments and by
showing that one can pass from the premisses to the conclusion through
a sequence of small and obvious steps. �at is, one proves the conclusion
from the premisses via intermediate conclusions: the original premisses
are used to derive obvious conclusions, which in turn are employed in the
next step as premisses to derive further conclusions, and so on, until the
original conclusion is obtained. Such a sequence of obvious arguments is
called a ‘proof’.
Whether a step is obvious depends on the perspective. However,
one might try to show that there is a �xed list of simple proof rules
that are su�cient for establishing the validity of any valid argument.
�e rules should be formulated in a way that makes it easy to check
whether any given step in a proof conforms to one of these rules. If a set
rules that can be used in proofs is �xed, then there cannot be a serious
disagreement about the admissibility of any given step in a proof, and
there is an objective notion of proof.
For the languages L� and L� of propositional and of predicate logic
one can provide such a list of admissible rules that legitimate steps in a
proof.
It is obvious for which arguments there should be proofs: �rst, there
should be proofs for valid arguments only. Formally speaking, it must not
be possible to pass from the premisses in a set Γ to a sentence �, if it is not
the case that Γ � �. �e rules must be sound in this sense. Second, the
proof rules should be complete in the sense that there should be proofs
for all valid arguments: if Γ � �, then it should be possible to reach �

© Volker Halbach ����/����

 � Natural Deduction ���

from the premisses in Γ by going through proof steps that conform to the
rules for proofs.
In order to show that Γ � �, one can then simply give a proof rather
than argue using L� -structures as in the previous chapter. �is will greatly
facilitate establishing the validity of arguments in predicate logic.
Logicians have devised various proof systems for di�erent purposes:
some systems are easy to implement on computers, others are very easy
to state (but hard to work in), still others facilitate general investigations
into the notion of provability. I will employ a system that enables one to
use proof steps that are not dissimilar to the steps people take in everyday
reasoning. �e the rules I will specify should be intuitively plausible, but
not every intuitively sound step is a permissible rule in the system: the
system has not been designed to be as e�cient as possible. It is devised
to show that any proof can be broken down into simple and elementary
steps of very few types. If the objective were a very e�cient proof system,
more rules would have to be added.
Because the proof rules are fairly close to proof steps used in informal
proofs, systems of the kind described in this chapter are called Natural
Deduction systems. �ey were introduced independently by Jaśkowski
(����) and Gentzen (����). �e system I am going to present is a variation
of Gentzen’s version.
Proofs in Natural Deduction start with an assumption. Any sentence
can be assumed:
assumption rule �e occurrence of a sentence � with no sen-
tence above it is an assumption. An assumption of � is a proof
of �.
It may seem somewhat odd that the solitary occurrence of a sentence
is already a proof, but it is convenient to consider a line with a single
sentence � as a proof of � from the assumption �, because this makes the
following de�nitions more straightforward.
Every proof begins with assumptions. �e further rules for proofs
show how to extend a proof, that is, how to form longer and longer

© Volker Halbach ����/����

 � Natural Deduction ���

proofs by adding further sentences. When stating the rules I will talk
about ‘appending’ sentences to already existing proofs. By this I mean the
following: one appends a sentence � to a proof by drawing a horizontal
line under the proof and then writing � under this line. One appends
a sentence � to two (or three) proofs by writing the proofs side by side,
then drawing a single line under all of these proofs, and then writing
� under that single line.
All the rules enable one to append only a single sentence in a given
step. �us, in every proof there is always a single sentence � at the bottom
(or the ‘root’) of the proof. �e proof is a proof of this sentence �. Proofs
have therefore the shape of (upward-branching) trees.
For each connective and quanti�er there is an introduction rule and
an elimination rule. I shall use abbreviations: for instance, ‘∧Intro’ is
short for ‘∧-introduction rule’.

�.� ������������� �����

For the sake of those who are concentrating just on propositional logic, I
shall only use examples in L� in this section. Nonetheless, the rules apply
equally to predicate logic.
I will start with the rules for conjunction:

∧Intro �e result of appending � ∧ ψ to a proof of � and a

proof of ψ is a proof of � ∧ ψ.

Graphically, this rule allows one to write a proof ending with � and a
proof ending with ψ side by side, to draw a horizontal line below both,
and to write � ∧ ψ under this line. �us an application of the rule will
have the following shape:

⋮ ⋮
� ψ
∧Intro
�∧ψ

© Volker Halbach ����/����

 � Natural Deduction ���

�e assumptions in the proof of � ∧ψ are all the assumptions in the proofs

of � and ψ, respectively, because any assumption in the proof of � or ψ,
that is, any sentence with no other formula above it in the proofs of � or
ψ, will also be an assumption in the proof of � ∧ ψ, that is, it will not have
a sentence above it in the proof of � ∧ ψ.
�e order of the proofs of � and ψ does not matter. �e rule does
not require that the proof of � is written to the le�. So an application of
∧Intro can also look like this:
⋮ ⋮
ψ �
∧Intro
�∧ψ
�e same applies to other rules: in rules where a sentence is appended to
two (or in one case three) proofs, the rules allow one to write down the
proofs in any order.
For ∧ there are two elimination rules:
∧Elim� �e result of appending � to a proof of � ∧ ψ is a proof
of �.
�e other rule allows one to keep ψ:
∧Elim� �e result of appending ψ to a proof of � ∧ψ is a proof
of ψ.
�e two rules can be depicted as follows:
⋮ ⋮
�∧ψ �∧ψ
∧Elim�
� ψ ∧Elim�
�e �rst rule allows one to drop the second part of the conjunction, the
second rule allows one to drop the �rst part.
With these rules in hand I can already construct a proof. First, I will
assume (P ∧ Q) ∧ R. �e rule ∧Elim� allows me to append P ∧ Q to the
proof, and ∧Elim� allows me to append Q to the resulting proof. So I can
obtain a proof of Q under the assumption (P ∧ Q) ∧ R:

© Volker Halbach ����/����

 � Natural Deduction ���

(P ∧ Q) ∧ R
∧Elim�
P∧Q
∧Elim�
Q
�e labels ‘∧Elim�’ and ‘∧Elim�’ do not belong to the proof; they are
mere comments that are intended to help the reader to grasp the proof.
Occasionally, when I think that the labels will facilitate understanding, I
will add them.
Next I shall specify rules for the arrow →. In order to motivate the
introduction rule for the arrow, I will look at how one might establish
and ‘if . . . , then . . . ’ like the following:

(A) If CO� -emissions are not cut, temperatures will rise globally.

To derive the conclusion, one will use additional assumptions about cli-
mate change, the greenhouse e�ect and so on, which I will not specify
here. Using these additional assumptions, one could argue as follows for
(A):

Assume that CO� -emissions are not cut. �en the CO� -level
in the atmosphere . . . [now one uses the additional assump-
tion, probably talking about the greenhouse e�ect, and con-
cludes:] so temperatures will rise globally. �erefore, if CO� -
emissions are not cut, temperatures will rise globally.

One makes the assumption that CO� -emissions are not cut only in order
to show that in that case temperatures will rise globally. �is assumption is
made only for the sake of the argument and once (A) has been concluded,
one is no longer assuming that CO� -emissions are not cut. �e proof of
(A) is based only on the additional assumptions about climate change etc,
but not on the assumption that CO� -emissions are not cut. �us, when
one concludes (A), one does not make the assumption anymore that
CO� -emissions are not cut; one claims only that temperatures will rise if
CO� -emissions are not cut without assuming anything about whether the
emissions are cut or not. Logicians describe this by saying that, when one

© Volker Halbach ����/����

 � Natural Deduction ���

concludes (A), one has ‘discharged’ the assumption that CO� -emissions
are not cut.
Generally, one can argue for a claim of the form ‘If A, then B’ by
proving B from the assumption A.
in the following way: one proves B by assuming A; then one concludes
‘If A, then B’ without assuming A anymore.
In Natural Deduction the rule for introducing the arrow works in the
same way: one assumes a sentence �, derives a sentence ψ from it, and
then the rule allows one to conclude � → ψ and to get rid of or ‘discharge’
the assumption of �.
In formal proofs one indicates that an assumption has been discharged
by enclosing that assumption in square brackets:
In formal proofs, assumptions are discharged by surrounding them with
square brackets. Of course one must only discharge assumptions in accor-
dance with the rules.
�e proof technique used in the above informal proofs is captured by
the introduction rule for →:
→Intro �e result of appending � → ψ to a proof of ψ and
discharging all assumptions of � in the proof of ψ is a proof of
� → ψ.
So one may add � → ψ to a proof with ψ at the root and then enclose all
assumptions of � (that is, all occurrences of � with no line above them)
in the proof of ψ in square brackets.
�e graphical representation looks like this:
[�]
⋮
ψ
→Intro
�→ψ
�is rule does not require that the proof of ψ actually contains an assump-
tion of �. Only if there are any assumptions of � in the proof of ψ, they
must be discharged.

© Volker Halbach ����/����

 � Natural Deduction ���

�e rule for eliminating → is straightforward:

→Elim �e result of appending ψ to a proof of � and a proof
of � → ψ is a proof of ψ.
�is rule is graphically represented as follows:

⋮ ⋮
� �→ψ
→Elim
ψ
�is rule is also called the ‘cut rule’, because the sentence � is ‘cut o�’
from � → ψ.
Before giving some examples, I will introduce a new piece of notation:
���������� �.�. �e formula � is provable from à (where à is a set of L� -
sentences) if and only if there is a proof of � with only sentences in Γ as
non-discharged assumptions. �e phrase ‘� is provable from Γ’ is abbrevi-
ated as Γ � �. If Γ is empty, Γ � � is abbreviated as � �. If Γ contains
exactly the sentences ψ� , . . . , ψ n , one may write ψ� , . . . , ψ n � � instead of
{ψ� , . . . , ψ n } � �.
������� �.�. � P ∧ Q → P.
Proof. I show step by step how to establish this claim. First, P ∧ Q is
assumed:

P∧Q
Applying ∧Elim� yields the following:
P∧Q
∧Elim�
P
�e rule →Intro allows one to add P ∧ Q → P and to discharge P ∧ Q:
[P ∧ Q]
∧Elim�
P →Intro
P∧Q →P

© Volker Halbach ����/����

 � Natural Deduction ���

All assumptions in this proof have been discharged. �us, P ∧ Q → P is

provable from the empty set of premisses, that is, � P ∧ Q → P.
�is is a typical proof of a sentence of the form � → ψ: usually one
assumes �, arrives through some steps at ψ, and then uses →Intro to
derive � → ψ and to discharge any assumptions of �.
�e introduction rules for disjunction are as follows:
∨Intro� �e result of appending a sentence � ∨ ψ to a proof
of � is a proof of � ∨ ψ.
As in the case for the elimination rules for ∧, there is also another intro-
duction rule for ∨:
∨Intro� �e result of appending a sentence � ∨ ψ to a proof
of ψ is a proof of � ∨ ψ.
�e graphical representations are as follows:
⋮ ⋮
� ψ
∨Intro� ∨Intro�
�∨ψ �∨ψ
�e elimination rule for ∨ is somewhat tricky. It corresponds to the
following informal proof strategy:
Assume ‘A or B’ is given. �en one can try to prove C by mak-
ing a case distinction: First, one tries to derive C assuming
A; then one tries to derive C assuming B. If C can be derived
in both cases, then (given ‘A or B’) one may conclude C.
�is type of reasoning – reasoning by drawing the same conclusion from
both parts of a disjunction – is captured in the following rule:
∨Elim �e result of appending χ to a proof of �∨ψ, a proof of χ
and another proof of χ, and of discharging all assumptions
of � in the �rst proof of χ and of discharging all assumptions
of ψ in the second proof of χ, is a proof of χ.

© Volker Halbach ����/����

 � Natural Deduction ���

An application of ∨Elim looks like this:

[�] [ψ]
⋮ ⋮ ⋮
�∨ψ χ χ
χ ∨Elim

In the �rst proof of χ only assumptions of � are discharged; one must

not discharge assumptions of ψ in this proof when applying the rule. Of
course, this remark applies analogously to the second proof of χ. �e
rule may look somewhat awkward because from two proofs of χ one
only obtains another, longer proof of χ. �e point of the rule is that the
assumptions of � and ψ can be discharged, so that χ now follows from
� ∨ ψ without assuming � or ψ.
In an actual proof one would proceed as follows: assume one has a
proof of � ∨ ψ. �en, one will start a new branch by assuming � and
another new branch by assuming ψ. A�er obtaining χ on both branches,
one can discharge all assumptions of � in the �rst proof of χ, and all
assumptions of ψ in the second proof of χ, and append χ to the three
proofs.
I will illustrate the use of ∨Elim with the following example.
������� �.�. P ∨ Q, P → R � R ∨ Q.
Proof.
[P] P→R
→Elim [Q]
R ∨Intro� ∨Intro�
P∨Q R∨Q R∨Q
∨Elim
R∨Q
In the last step – an application of ∨Elim – the assumptions P and Q are
discharged.
�e introduction rule for negation is another rule that allows one
to discharge assumptions. �e underlying strategy is as follows: If one

© Volker Halbach ����/����

 � Natural Deduction ���

can derive a contradiction from an assumption of A, then one may con-

clude: ‘It is not the case that A.’ �is rule is called ‘reductio ad absurdum’
(reduction to an absurdity).

¬Intro �e result of appending a sentence ¬� to a proof of ψ

and a proof of ¬ψ and of discharging all assumptions of � in
both proofs is a proof of ¬�.

Schematically, the rule for ¬-introduction has the following shape:

[�] [�]
⋮ ⋮
ψ ¬ψ
¬Intro
¬�
I will demonstrate the use of the rule with an example:
������� �.�. ¬(P → Q) � ¬Q.
Proof. In order to arrive at a conclusion of the form ¬�, it is o�en useful to
assume � and to try to derive a contradiction. In this case I will assume Q
and try to obtain a contradiction with the only premiss, viz, ¬(P → Q).
[Q]
→Intro
P→Q ¬(P → Q)
¬Intro
¬Q
In the le� branch of the proof I have applied →Intro even though there is
no assumption of P. �is is in accordance with the formulation of →Intro:
nothing in →Intro actually requires that there is actually an assumption
of P; only if there are any, they must be discharged.
�e rule of negation elimination allows one to discharge assumptions
of negated sentences:�

� �is rule allows for nonconstructive indirect proofs. For instance, the formula � could

© Volker Halbach ����/����

 � Natural Deduction ���

¬Elim �e result of appending as sentence � to a proof of ψ

and a proof of ¬ψ and of discharging all assumptions of ¬�
in both proofs is a proof of �.

Graphically an application of the rule looks like this:

[¬�] [¬�]
⋮ ⋮
ψ ¬ψ
¬Elim
�
Here is an example showing how the negation elimination rule can be
used:
������� �.�. ¬P → Q, ¬Q � P.

[¬P] ¬P → Q
→Elim
Q ¬Q
¬Elim
P
Negation elimination can be used to prove the law of excluded middle:
������� �.�. � P ∨ ¬P.

be an existence claim. Negation elimination allows one to conclude the existence of an

object with a certain property from the inconsistency of the assumption that there isn’t
any object with that property. �us one will be able to prove that an object satisfying
a certain property exists, without being able to show directly, of any particular object,
that it has this property. If one is interested in constructive proofs, that is, proofs
that require one to exhibit a particular example of an object that demonstrably has
the property in question in order to prove the relevant existence claim, then the rule
for negation elimination has to be dropped. �e resulting system, which di�ers from
classical Natural Deduction, is called ‘intuitionistic logic’. �e law of the excluded middle,
� ∨ ¬� (Example �.�), and the law of double negation elimination, ¬¬� → �, are some
of the principles that are theorems of classical logic, which is the logic studied here, but
not of intuitionistic logic. Tennant (����) provides an introduction to intuitionistic
logic with a proof system that is similar to the version of Natural Deduction used here.

© Volker Halbach ����/����

 � Natural Deduction ���

�e proof is surprisingly awkward. One cannot prove one of the sen-

tences P or ¬P (without assumptions) and then apply ∨Intro� or ∨Intro�.
Rather one proves P ∨ ¬P indirectly by assuming ¬(P ∨ ¬P), which is
then shown to lead to a contradiction. �is makes it possible to apply the
negation elimination rule.
Proof. First I assume P and apply ∨Intro�:
P ∨Intro�
P ∨ ¬P
In the next step, I assume ¬(P ∨¬P), which is the negation of the sentence
that is to be proved.
[P]
∨Intro�
P ∨ ¬P ¬(P ∨ ¬P)
¬Intro
¬P
Now ∨Intro� can be used:
[P]
∨Intro�
P ∨ ¬P ¬(P ∨ ¬P)
¬Intro
¬P
∨Intro�
P ∨ ¬P
Now ¬(P ∨ ¬P) is assumed again, so that ¬Intro can be applied once
more. �is time ¬Intro is used to discharge the assumption ¬(P ∨ ¬P),
which occurs twice.
[P]
∨Intro�
P ∨ ¬P [¬(P ∨ ¬P)]
¬Intro
¬P
∨Intro�
P ∨ ¬P [¬(P ∨ ¬P)]
¬Elim
P ∨ ¬P

�e rules for the double arrow are as follows:

© Volker Halbach ����/����

 � Natural Deduction ���

↔Intro �e result of appending � ↔ ψ to a proof of � → ψ

and a proof of ψ → � is a proof of � ↔ ψ.

↔Elim� �e result of appending � → ψ to a proof of � ↔ ψ is

a proof of � → ψ.

↔Elim� �e result of appending ψ → � to a proof of � ↔ ψ

is a proof of ψ → �.

�e graphical representations of the rules are as follows:

⋮ ⋮
�→ψ ψ→�
↔Intro
�↔ψ
⋮ ⋮
�↔ψ �↔ψ
↔Elim� ↔Elim�
�→ψ ψ→�
In the following proof I will illustrate the use of the rules for the double
arrow.
������� �.�. � (P → Q) ↔ (¬Q → ¬P).
Proof.

[P → Q] [P] [¬Q → ¬P] [¬Q]

Q [¬Q] ¬P [P]
¬P Q
¬Q → ¬P P→Q
(P → Q) → (¬Q → ¬P) (¬Q → ¬P) → (P → Q)
(P → Q) ↔ (¬Q → ¬P)
For the last line ↔Intro is used.
In cases like the following it is o�en useful to start from the root and to
consider how one might have obtained the sentence that is to be proved.

© Volker Halbach ����/����

 � Natural Deduction ���

������� �.�. ¬(Q ∧ ¬R) � Q → R.

Proof.
[Q] [¬R]
¬(Q ∧ ¬R) Q ∧ ¬R
R
Q→R
�e assumption ¬(Q∧¬R) is not bracketed; it remains as an undischarged
assumption. �us, ¬(Q ∧ ¬R) � Q → R is established.
Finally, here is a proof of a variant of the ‘ex falso quodlibet’-principle:
������� �.�. P, ¬P � Q.
Proof. �e following proof contains a strange application of ¬Elim. �e
formula Q is introduced although ¬Q has never be assumed. ¬Elim
allows one to introduce � and to discharge all assumptions of ¬� even if
there are not any.
P ¬P ¬Elim
Q
�is application of ¬Elim complies with ¬Elim, as it is not a requirement
that there actually be assumptions of ¬�.
�e following result says that the rules for Natural Deduction have
been chosen in such a way that there are proofs for exactly those argu-
ments in L� that are valid.
������� �.�� (�������� ��� ������������� �����). Assume that
� and all elements of Γ are L� -sentences. �en Γ � � if and only if Γ � �.
I will not prove this theorem here.
In particular, if a sentence of L� is a tautology (logically true), then
there is a proof of that sentence without undischarged assumptions.
�e adequacy result says that the rules given here are su�cient for
proving a conclusion from premisses if the argument is valid. Any rule

© Volker Halbach ����/����

 � Natural Deduction ���

that can be added is either not sound, that is, it allows one to prove
sentences that are not logically true, or it is dispensable (but it may provide
a shortcut). I shall return to adequacy in more detail in Section �.�; there
adequacy will be discussed with respect to predicate logic.

�.� ��������� �����

�e proof rules of propositional logic, which were expounded in the

previous section, apply to all sentences including the sentences of L� .
�e following example contains sentences of the language L� that are not
in L� ; but the proof requires only rules from propositional logic, that is,
introduction and elimination rules for connectives.
������� �.��. � ¬(∀x Px ∨ ∃y P y) → ¬∀x Px.
Proof. Since one wants to prove a sentence of the form � → ψ, the �,
which corresponds to ¬(∀x Px ∨ ∃y P y) in the present example, is as-
sumed. From this assumption one tries to arrive at ¬∀x Px. Since this is
a negated sentence, one may hope to get it by using ¬Intro; so ∀x Px is
assumed. �at is almost a contradiction: applying ∨Intro� to ∀x Px gives
∀x Px ∨ ∃y P y, and the contradiction is obtained:
∀x Px ∨Intro�
¬(∀x Px ∨ ∃y P y) ∀x Px ∨ ∃y P y
Now ¬Intro is applied by appending ¬∀x Px to the two proofs and dis-
charging the assumption ∀x Px:
[∀x Px]
∨Intro�
¬(∀x Px ∨ ∃y P y) ∀x Px ∨ ∃y P y
¬Intro
¬∀x Px
One �nishes the proof by applying →Intro:

© Volker Halbach ����/����

 � Natural Deduction ���

[∀x Px]
∨Intro�
[¬(∀x Px ∨ ∃y P y)] ∀x Px ∨ ∃y P y
¬Intro
¬∀x Px →Intro
¬(∀x Px ∨ ∃y P y) → ¬∀x Px
All assumptions are discharged, so the claim is established.
I will now turn to the rules for ∀ and ∃. First, I will explain the rule
for eliminating ∀, which may be motivated by considering the following
argument:
If some person has made more than ten mistakes then that
person won’t pass. �erefore, if Ben has made more than ten
mistakes he won’t pass.
Here one is going from a universal claim to a claim about a speci�c in-
stance. �e rule for eliminating ∀, which is also known as ‘universal
instantiation’-rule, allows one to pass from a universally quanti�ed sen-
tence to a special instance. For instance, the rule licenses the step from the
universally quanti�ed sentence ∀x (Px → Qx) to the instance Pa → Qa.
In order to give a general formulation of ∀Elim I employ the following
de�nition:
���������� �.��. Assume v is a variable, t a constant, and � an L� -formula
with at most v occurring freely. �en �[t�v] is the sentence obtained by re-
placing all free occurrences of v in � by t.
For instance, Px [b� �x] is Pb� ; and ∀y (Px y → ∃x Rx y) [b�x]is the
sentence ∀y(Pby → ∃x Rx y). In the second case only the �rst occurrence
of x has been replaced because the other two are bound occurrences.
�e rule for eliminating the universal quanti�er can now be stated as
follows:
∀Elim �e result of appending �[t�v] to a proof of ∀v � is a
proof of �[t�v].
In this rule it is assumed that t is a constant, v a variable, and ∀v � is a
sentence (so that only v can occur freely in �).
�us an application of this rule has the following form:

© Volker Halbach ����/����

 � Natural Deduction ���

⋮
∀v �
∀Elim
�[t�v]
Here, � is an L� -formula in which only the variable v occurs freely; t is a
constant.
�us the rule allows one to drop the quanti�er ∀v at the beginning of
a sentence and to replace all free occurrences of v in the resulting formula
with the constant t.
�e following example explains why only free occurrences of the
variable are replaced:

∀y (P y ∧ ∃y Q y)
∀Elim
Pb� ∧ ∃y Q y
If one were allowed to replace also the last occurrence of y, which is
bound, one would to get to Pb� ∧ ∃yQb� , which is logically equivalent to
Pb� ∧ Qb� . �is step is clearly not sound: from ∀y(P y ∧ ∃y Q y) it does
not follow that b� is Q. Intuitively speaking, only occurrences of y in
∀y (P y ∧ ∃y Q y) that are ‘caught’ by the universal quanti�er ∀y can be
replaced by the constant; the last occurrence of y belongs to the existential
quanti�er and must be le� alone.
�e introduction rule for the universal quanti�er is more di�cult to
state. In order to argue for the general claim that every traveller’s journey
from London to Munich in ���� took over two hours, one could reason
as follows:

Assume somebody travelled from London to Munich in ����.

Call him John Doe. If he took the train . . . [now each means
of transport is taken into account, and it is argued in every
case that the journey must have taken more than two hours.]
�erefore John Doe’s journey took over two hours. �erefore,
since John Doe is an arbitrary person, every traveller’s jour-
ney from London to Munich in ���� took over two hours.

© Volker Halbach ����/����

 � Natural Deduction ���

�e idea here is that one talks about a nondescript particular instance as

an example using a name. In the story above, this nondescript instance is
a person who is dubbed ‘John Doe’. �e name ‘John Doe’ is not used to
name a speci�c real person. �en one carries out the argumentation for
this instance, and concludes that this argument applies to every instance
because the chosen instance was just used as an example. For the validity
of the argument it is important that the chosen person is arbitrary and
that one did not bring in any speci�c information about that person or
object (if there is such a person or object at all).
�e rule for introducing the universal quanti�er in Natural Deduction
works in a similar way: � ∀x(Px → Px) will be established by taking an
arbitrary constant, say, c (corresponding to the use of a name ‘John Doe’
above), and arguing as follows:
[Pc]
→Intro
Pc → Pc
Since c was chosen to name a nondescript instance as an example, the
claim also holds for every object. Now, one substitutes the variable x for
the constant c in Pc → Pc and pre�xes the universal quanti�er ∀x to the
formula. �en one can add this sentence to the proof, in accordance with
the rule for introducing ∀:
[Pc]
→Intro
Pc → Pc
∀Intro
∀x(Px → Px)
In formulating the general rule ∀Intro, however, one must exercise
caution. One might think that ∀Intro can be formulated in the following
way: given a proof of a sentence �, one replaces a constant in the sentence
by a variable v, writes ∀v in front of the sentence, and then appends the
resulting sentence to the proof. However, such a rule would allow one to
derive false conclusions from true premisses for three reasons.
����� �������. I will illustrate the problem by considering the fol-
lowing attempted proof:

© Volker Halbach ����/����

 � Natural Deduction ���

∀y (P y → ∃x Ryx)
∀Elim
(P) Pa → ∃x Rax incorrect ∀-introduction
∀x (Px → ∃x Rxx)
If (P) were a correct proof one could derive ∀x (Px → ∃xRxx) from
∀y (P y → ∃x Ryx). So (P) cannot be a correct proof: it would allow one
to go from the formalisation of ‘Every student sees something’ to the
formalisation of ‘For every student there is something that sees itself.’ �e
�rst sentence is true while the latter is false if there is at least one student
and nothing can see itself.�
It should be obvious what went wrong. When the constant a was
replaced by x in the sentence Pa → ∃x Rax, the x replacing the last
occurrence of a got ‘caught’ by the quanti�er ∃x:
∀x (Px → ∃x Rx x)
�
actual binding

Of course the ‘intention’ was that the penultimate occurrence of x should

refer back to the universal quanti�er ∀x and not to the existential quanti-
�er:
∀x (Px → ∃x Rx x)
��� � � � � � � � � � � � � � � � � � ��� � � � � � � � � � � � � � � � � � �
intended binding
�e problem can be avoided in the following way: when one adds
a sentence ∀v � to a proof by applying the rule for introducing ∀ then
the preceding sentence must be � with all free occurrences of the vari-
able v replaced by the constant t. In other words, the rule allows one to
continue a proof ending with �[t�v] by adding the sentence ∀v �. So the
rule for introducing ∀ will take the following form (with some additional
restrictions on occurrences of the constant t speci�ed below):
⋮
�[t�v]
∀v �

� �e proof that there is a counterexample to the argument with ∀y (P y → ∃x Ryx) as

premiss and ∀x (Px → ∃xRxx) as conclusion is the content of Exercise �.�(ii).

© Volker Halbach ����/����

 � Natural Deduction ���

In the above attempted proof (P), I passed from Pa → ∃x Rax to

∀x (Px → ∃x Rxx). But this step is not covered by the rule that licenses
only the step from �[t�v] to ∀v �: �e result (Px → ∃x Rxx)[a�x] of
replacing all free occurrences of x by a is the sentence Pa → ∃x Rxx, not
Pa → ∃x Rax. Only going from Pa → ∃x Rxx to ∀x (Px → ∃x Rxx)
would be covered by the rule.
������ �������. Assume one starts a proof by assuming Qb and
then generalises in the following way:
Qb
incorrect ∀-introduction
∀x Qx
�is is not correct. If it were correct, one could apply →Intro in the next
step:
[Qb]
incorrect ∀-introduction
∀x Qx
→Intro
Qb → ∀x Qx
�us, one would have proved that � Qb → ∀x Qx; but Qb → ∀x Qx is
certainly not logically true, as has been shown in Example �.�.
For this reason, one must make sure that one is not using any speci�c
assumptions containing the constant that is used as an ‘arbitary exam-
ple’ over which one can generalise. So when ∀Intro is applied by going
from �[t�v] to ∀v �, the constant t must not occur in any undischarged
assumption. In the discussion of the example of John Doe I said that it is
important that the person chosen as an example (John Doe) is arbitrary.
�e restriction that the constant must not occur in any undischarged as-
sumption is the formal counterpart of the requirement that John Doe is an
‘arbitrary’ person about whom one does not have any speci�c information.
In the earlier, correct proof of ∀x (Px → Px) the constant c occurred
in the assumption Pc, but that assumption had been discharged by the
time ∀Intro was applied. So this proof meets the restriction that c must
not occur in any undischarged premiss when the universal quanti�er is
introduced.

© Volker Halbach ����/����

 � Natural Deduction ���

����� �������. �e following attempted proof meets all conditions

that were imposed on applications of ∀Intro to avoid the �rst and second
problem:

∀y Ry y
Raa
∀x Rax
However, the argument with ∀y Ry y as premiss and ∀x Rax as its con-
clusion is not valid: it would alow one to go from the formalisation of
‘Everthing is self-identical’ to the formalisation of ‘John is identical to
everything.’� �us, some condition must be imposed on the rule ∀Intro
that blocks the step from Raa to ∀x Rax. �is can be done by disallow-
ing one to keep some occurrences of a in the sentence that is added in
accordance with the rule. Only passing from Raa to ∀x Rxx would be
admissible. �us, since an application of ∀Intro is a step from �[t�v]
to ∀v �, the rule ∀Intro can be applied only if � (that is, the formula
following the universal quanti�er) does not contain the constant t, that
is, if no occurrence of t is retained when ∀Intro is applied.
In the following formulation of the rule for introducing the universal
quanti�er all three problems are avoided:
∀Intro Assume that � is a formula with at most v occurring
freely and that � does not contain the constant t. Assume fur-
ther that there is a proof of �[t�v] in which t does not occur
in any undischarged assumption. �en the result of append-
ing ∀v � to that proof is a proof of ∀v �.
�e rule can be represented as follows:
⋮ provided the constant t does not
�[t�v] occur in � or in any undischarged
∀Intro
∀v � assumption in the proof of �[t�v].

� �e proof of ∀y Ry y � ∀x Rax is Exercise �.�(iii).

© Volker Halbach ����/����

 � Natural Deduction ���

In less formal terms, the restrictions on ∀Intro can be summed up in

two points: First, one must make sure that one is not generalising over a
constant that occurs in an undischarged assumption or that is still in the
sentence one tries to obtain. Second, in an application of the rule, one
has to make sure that the variable of the newly introduced quanti�er is
not caught by a quanti�er that is already in the formula.
In the case of the existential quanti�er the introduction rule is the
easy one:

∃Intro �e result of appending ∃v � to a proof of �[t�v] is a

proof of ∃v �.

�[t�v]
∃Intro
∃v �

Of course t is a constant, v a variable, and ∃v � is an L� -sentence.

�e rule for eliminating the existential quanti�er is the most compli-
cated rule. Consider the following argument:

�ere is at least one epistemologist. All epistemologists are

philosophers. �erefore, there is at least one philosopher.

�e two premisses have ∃x Px and ∀x (Px → Qx) as their respective for-

malisations; the formalisation of the conclusion is ∃x Qx. �e dictionary
has the following two entries:

P: . . . is an epistemologist
Q: . . . is a philosopher
�e corresponding argument in L� is valid. �e question is how one
can prove the conclusion from the two premisses. If one could in some
way get Pc from the �rst premiss ∃x Px, the rest of the proof would be
obvious:

© Volker Halbach ����/����

 � Natural Deduction ���

⋮ ∀x (Px → Qx)
∀Elim
Pc Pc → Qc
→Elim
Qc
∃Intro
∃x Qx
�e problem is that Pc does not follow from ∃x Px. �e premiss ∃x Px
just states that there is some epistemologist; it does not give one a speci�c
epistemologist and, in particular, it does not give one a particular name c
for an epistemologist.
�e conclusion ∃x Qx does not say anything speci�c about c. In
the proof I could have used any other constant in place of c. So one
might apply the following proof strategy: one may assume Pc. �is is
tantamount to picking an arbitrary name like ‘John Doe’ and assuming
that John Doe is an epistemologist. Once a sentence not containing c is
proved, one can discharge the assumption Pc using the premiss ∃x Px:
the conclusion does not depend any more on the assumption that c is
one of the P’s.

∀x (Px → Qx)
∀Elim
[Pc] Pc → Qc
→Elim
Qc
∃Intro
∃x Px ∃x Qx
∃x Qx
�e point in the last step of the proof is that the premiss Pc can be dis-
charged, so one has now proved the conclusion from the existence claim
∃x Px rather than from the speci�c instance Pc. Informally speaking, the
conclusion that there is a philosopher does not depend on the assumption
that there is an epistemologist called ‘John Doe’.
When making an assumption such as Pc one must not choose a con-
stant about which one already has speci�c information: the constant
acts as an ‘arbitrary example’ in the same way as in ∀Intro. �e precise
statement of the elimination rule for ∃ is so convoluted because the re-
strictions on the constant must ensure that the constant can play its role
as an arbitrary label.

© Volker Halbach ����/����

 � Natural Deduction ���

∃Elim Assume that � is a formula with at most v occurring

freely and that the constant t does not occur in �. Assume
further that there is a proof of the sentence ψ in which t does
not occur in any undischarged assumption other than �[t�v].
�en the result of appending ψ to a proof of ∃v � and the proof
of ψ and of discharging all assumptions of �[t�v] in the proof
of ψ is a proof of ψ.

As before, �[t�v] is just � with all free (and only free) occurrences of v
replaced with t.
An application of the rule looks like this:

[�[t�v]] provided the constant t does not

⋮ ⋮ occur in ∃v �, or in ψ, or in any
∃v � ψ undischarged assumption other
∃Elim than �[t�v] in the proof of ψ.
ψ
In practice, ∃Elim is applied in the following way: Assume that one
has proved ∃v �. �en one picks a constant that has not been used yet and
that does not occur in any premisses one has, and assumes �[t�v]. Once
one has proved a sentence ψ not containing t, all assumptions of �[t�v]
are discharged and one writes ψ under the proof of ∃v � and the proof
of ψ. Here I have recommended using a constant t that is completely
‘new’ to the proof; this is not really necessary or forced by the rule, but by
using a new variable one can make sure that the conditions on t in the
rule are satis�ed. Also, by using a constant that is not new one does not
gain anything.
�is rule concludes the description of the system of Natural Deduc-
tion. A list of all the rules may be found in Appendix �.�.
I will give some examples of proofs in which the quanti�er rules are
used.
������� �.��. ∀x ¬Px � ¬∃x Px
Proof. First one assumes ∀x ¬Px and applies ∀Elim:

© Volker Halbach ����/����

 � Natural Deduction ���

∀x ¬Px ∀Elim
¬Pa
In order to be able to apply ∃Elim, one assumes Pa and continues with
¬Intro:
[∀x ¬Px]
∀Elim
Pa ¬Pa ¬Intro
¬∀x ¬Px
With the additional assumption ∃x Px one can apply ∃Elim and dis-
charge Pa:
[∀x ¬Px]
∀Elim
[Pa] ¬Pa
¬Intro
∃x Px ¬∀x ¬Px ∃Elim
¬∀x ¬Px
Now ∀x ¬Px is assumed again to produce a contradiction that allows one
to apply ¬Intro:
[∀x ¬Px]
∀Elim
[Pa] ¬Pa
¬Intro
[∃x Px] ¬∀x ¬Px
∃Elim
∀x ¬Px ¬∀x ¬Px ¬Intro
¬∃x Px
∃x Px can be discharged according to ¬Intro. �e only assumption that is
not discharged is the le�most occurrence of ∀x ¬Px, which is the premiss.

�e next example is the converse of the previous one.

������� �.��. ¬∃x Px � ∀x ¬Px.
Proof. ∀x ¬Px, which is to be derived, can be obtained by ∀Intro from ¬Pa.
�is in its turn can be obtained by ¬Intro. So Pa is assumed and ∃Intro
is applied:

© Volker Halbach ����/����

 � Natural Deduction ���

Pa ∃Intro
∃x Px
Now the assumption ¬∃x Px is added and ¬Intro is applied:
[Pa]
∃Intro
¬∃x Px ∃x Px ¬Intro
¬Pa
Because Pa is now discharged, ∀Intro can be applied:
[Pa]
∃Intro
¬∃x Px ∃x Px ¬Intro
¬Pa ∀Intro
∀x ¬Px
All assumptions with exception of the premiss are discharged.
In the following example, ∃Intro is applied in a cunning way.
������� �.��. ∀x Rxx � ∀y ∃z Ryz.
Proof. ∀x Rxx ∀Elim
Raa ∃Intro
∃z Raz ∀Intro
∀y ∃z Ryz
�e application of ∃Intro is legitimate: nothing in the formulation of
the rule forces one to replace all occurrences of a by the variable z. �e
formula Raz is a formula with only one variable occurring freely; thus
�[t�v] is Raz[a�z], which is the formula Raa. ∃v � is then ∃z Raz.
In Example �.�� on page ���, I refuted ∀x ∃y Rx y � ∃y ∀x Rx y by
means of a counterexample. Now I can establish the converse direction
∃y ∀x Rx y � ∀x ∃y Rx y.
������� �.��. ∃y ∀x Rx y � ∀x ∃y Rx y.
Proof.

© Volker Halbach ����/����

 � Natural Deduction ���

[∀x Rxb]
∀Elim
Rab ∃Intro
∃y ∀x Rx y ∃y Ray
∃Elim
∃y Ray
∀Intro
∀x ∃y Rx y
�e assumption ∀x Rxb has been discharged in the penultimate step by
applying ∃Elim.

© Volker Halbach ����/����

 � Formalisation in Predicate Logic

In this chapter I shall bring three strands together: the semantics of

the language L� of predicate logic, the system of Natural Deduction,
and arguments in English. First I will turn to the relation between the
semantics of L� and proofs in Natural Deduction.

�.� ��������

�e rules of Natural Deduction have been chosen so as to guarantee

that, if a sentence is provable (with no undischarged premisses), it is
logically true; and if a sentence is provable from certain premisses, the
corresponding argument is valid. �is way, formal proofs allow one to
establish that certain arguments in L� are valid (and that certain sentences
are logically true). In the following lemma, it is assumed that � and all
elements of Γ are sentences of L� .
����� �.� (���������). If à � �, then à � �.
I will not give a proof of this claim here, but I have tried to present
the rules of Natural Deduction in such a way that their soundness ought
to be plausible, if not obvious.
While the soundness of the rules of Natural Deduction is fairly plau-
sible, it is much harder to see whether one can actually prove � from
premisses in Γ whenever Γ � �, and whether one can prove all logical
truths of L� . In fact, the rules of Natural Deduction are su�cient for this
purpose:
������� �.� (������������). If à � �, then à � �.

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

As before, � and all elements of Γ are L� -sentences here.

�e fact that the proof system of Natural Deduction is sound sound-
ness and complete completeness is o�en expressed by saying that it is
adequate for the semantics introduced in Chapter �. More formally, this
can be expressed in the following way:
������� �.� (��������). Assume that � and all elements of à are L� -
sentences. �en Γ � � if and only if Γ � �.
�us, � and � coincide even though they have been de�ned in com-
pletely di�erent terms. I have de�ned � in semantic terms, that is, in
terms of L� -structures, while � has been de�ned in purely syntactic terms,
that is, in terms of rules for manipulating sentences of L� .
For the special case where Γ is empty, the Adequacy �eorem �.�
means that � is logically true if and only if � is provable (with all assump-
tions discharged). �is can be expressed more formally in the following
way:
� � if and only if � �
In practice, if one wants to show that � follows from sentences in Γ,
that is, if one wants to show that Γ � �, one will usually try to construct a
proof in the system of Natural Deduction because this is in most cases
easier than a proof directly establishing that � is true in all L� -structures
in which all sentences in Γ are also true.
In contrast, if one wants to prove that � does not follow from Γ, that
is, that Γ � � is not the case, or Γ � � for short, one will usually be better
o� constructing a counterexample, that is, an L� -structure in which all
sentences of Γ are true and � is not true. �is is usually easier than
showing that there is no proof of � with sentences from Γ as the only
undischarged assumptions.
In Section �.� I have de�ned further semantic properties of sentences
of L� . �ese properties can be de�ned also in terms of provability. �en
the Adequacy �eorem �.� can be used to show that the syntactic de�ni-
tions (in terms of provability) and the semantic de�nitions (in terms of

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

L� -structures) coincide. Here consistency will serve as an example. First,

consistency is de�ned in terms of proofs:
���������� �.� (��������� �����������). A set à of L� -sentences is
syntactically consistent if and only if there is a sentence � such that Γ � �.
Here Γ � � means that it is not the case that Γ � �. �erefore, a
set Γ of sentences is syntactically consistent if and only if it is not the
case that any sentence whatsoever can be proved from premisses in Γ (as
nondischarged assumptions).
Inconsistency is the opposite of consistency: a set of sentences is
syntactically inconsistent if and only if it is not syntactically consistent.
�us, a set Γ of sentences is inconsistent if and only if from premisses in
Γ every L� -sentence can be proved. �at is, Γ is inconsistent if and only
if Γ � � for every L� -sentence �.
According to De�nition �.�, a set Γ of L� -sentences is semantically
consistent if and only if there is an L� -structure A in which all sentences
in Γ are true. Now the Adequacy �eorem can be used to show that
semantic and syntactic consistency coincide:
������� �.�. A set à of L� -sentences is semantically consistent if and only
if Γ is syntactically consistent.
Proof. Assume that the set Γ is semantically consistent. �en by De�ni-
tion �.�(iv) there is an L� -structure A in which all sentences in Γ are true.
Choose a sentence � which is false in that structure. �en Γ � �, and by
the Soundness Lemma �.�, Γ � �. So Γ is syntactically consistent.
To see the converse, assume that Γ is not semantically consistent.
�en there is no L� -structure A in which all sentences in Γ are true.
Consequently, any sentence � will be true in all L� -structures in which all
sentences of Γ are true (because there are no such structures). So Γ � �
for all sentences � of L� , and by �eorem �.�, Γ � � for all sentences �.
So there is no sentence � that is not provable from premisses in Γ, and,
therefore, Γ is not syntactically consistent.
So far all the premiss sets used in examples contained only a few simple

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

sentences. A premiss set, however, may also be very complicated. For

example, a premiss set could contain all of the sentences a certain person
believes, or the axioms of some theory in philosophy or mathematics
or medicine or any other area. A theory here is nothing more than a
premiss set, that is, a set of sentences. Some theories can be logically very
powerful. For instance, all of mathematics (with some possible minor
exceptions) can be developed in set theory, and the axioms of set theory
can be written down as L� -sentences.
When considering a theory, one would like to know whether it is (syn-
tactically or semantically) consistent. �eorem �.� shows why consistency
is such an important property of a theory: if a theory is inconsistent, any-
thing can be proved from it. �us, in order to show that a set of sentences
is not useless as a theory (because anything can be proved from it), one
has to show that the set is consistent. �is can be achieved by establishing
that there is an L� -structure in which all the sentences of the theory are
true. In that case the set is semantically consistent, and by �eorem �.�
also syntactically consistent. �e tools of logic have o�en been used to
show that theories are consistent.
�ese tools can be used to prove not only that theories are consistent
and that sentences follow logically from certain premisses; they can also
be used to investigate much more general problems about consistency,
validity, and so on. I will sketch some general results of this kind here.
It would be convenient to have a general method for determining
whether an arbitrary argument is valid or whether a given sentence is
logically true. If � is a sentence of the language L� of propositional logic,
there is a method for deciding whether � is logically true or not: one can
use the truth table method. In propositional logic, one can also use the
truth table method to decide whether an argument with �nitely many
premisses is valid or not. �e situation with L� is di�erent: one can use
the system of Natural Deduction to show that a sentence is logically true,
but it hardly can be used to prove that a sentence is not logically true. If
one does not hit upon a proof in a reasonable amount of time, this does
not mean that there is no proof; one could simply have not yet found

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

the proof: �e maximal possible length of a proof cannot be read o�

the sentence. So the system of Natural Deduction does not deliver a
systematic method for deciding whether a sentence is logically true.
Church (����) showed, using Kurt Gödel’s famous Incompleteness
�eorems, that there cannot be a procedure for deciding whether a sen-
tence of L� is logically true. In particular, no computer program can, given
an arbitrary sentence as input, tell one infallibly whether the sentence
is logically true or not, or whether a given sentence can be proved from
sentences in a given theory. �is holds even if there are no restrictions
imposed on the computing time, computer memory, and so on. Church’s
result shows that, given certain sentences as input, the program will keep
on calculating forever (or return an incorrect result). �e precise formu-
lation of this result is beyond the scope of this text, but it explains why
the system of Natural Deduction (or any other system for predicate logic)
cannot be transformed into a method that, like the truth-table method
for L� , takes a sentence of L� as input and then returns in any case a�er
�nitely many steps an answer to the question of whether the sentence is
logically true or not.
Church’s result also shows that there are no simple checks for the
consistency of a set of assumptions: there is no systematic method – a
method that could be implemented on a computer – for deciding whether
a given �nite set of sentences is consistent or not. Of course, one may
try to �nd an L� -structure in which all sentences in the set are true, but
there is again no systematic method that tells one whether there is such
an L� -structure. �us, proving the consistency of sets of sentences of L�
is a highly non-trivial a�air.

�.� ���������

As I said at the beginning of the chapter, I want to bring together three

strands: the semantics of L� , the system of Natural Deduction, and ar-
guments in English. �e relation between the former two has now been

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

clari�ed by �eorem �.�: the semantics of L� and the proof system of

Natural Deduction match up, that is, � and � coincide. Now I will turn
to natural language and how to compare arguments in English to those
in L� .
�e connection between arguments in English and in L� is established
by translating between the two languages. So I will turn now to such
translations.
In Section �.� I provided a sketch of how to go about translating from
English into the language L� . In the �rst step the English sentence is
brought into a regimented form: its logical form. In the second step the
standard connectives are replaced by the respective symbols: ‘there is
at least one x’ is replaced by ∃x, ‘for all x’ is replaced by ∀x, names are
replaced by constants, predicate expressions are replaced by predicate
letters, and so on. Most di�culties occur in the �rst step, while in the
second step English expressions are merely mechanically replaced by
corresponding symbols.
Some di�culties arise from a discrepancy between predicate expres-
sions in English and the predicate letters of L� : in English the number
of designators a predicate takes can vary from sentence to sentence. �e
following English sentences are all well formed:
In his garage the engineer loosens the nut with the wrench.
�e engineer loosens the nut with the wrench.
�e engineer loosens the nut.
In the �rst sentence one would like to formalise the predicate ‘loosens’ as
a �-place predicate letter, in the second as a ternary predicate letter, and
in the last sentence as a binary predicate letter. �e language L� , however,
does not have predicate letters with a variable number of places. If an
argument contains two of the sentences above one would like to formalise
the predicate ‘loosens’ as the same predicate letter in both sentences,
and there are some tricks that can help. For instance, one might try to
reformulate the second sentence as ‘In some place the engineer loosens
the nut with the wrench,’ thereby making it amenable to formalisation by

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

a �-place predicate letter; then ‘loosens’ can be formalised by this �-place

predicate letter in the �rst and second sentence. �e strategy, however,
does not easily generalise, and philosophers and logicians have proposed
other strategies to solve the problem of variable arity. �e discussion of
these proposals, however, goes beyond the scope of this text.
�e problem of variable arity I have just outlined does not arise from
an ambiguity in English; rather it highlights di�erences in the grammars
of English and L� . In Section �.� I have already mentioned certain am-
biguities that can lead to two di�erent formalisations in propositional
logic. In particular, I have discussed ambiguities concerning the scope
of connectives. Of course, a sentence displaying such an ambiguity has
di�erent formalisations also in L� because the connectives work in L� in
the same way as in L� .
I will now browse through some other kinds of ambiguities, begin-
ning with lexical ambiguities. �e word ‘bank’, for instance, is lexically
ambiguous. It can mean the edge of a river or a �nancial institution. One
can analyse lexical ambiguities by formalising ‘is a bank’ as two di�erent
unary predicate letters, where one predicate letter stands for ‘. . . is a bank
(�nancial institution)’ and the other for ‘. . . is a (river) bank’. �us, a sen-
tence containing lexically ambiguous vocabulary may have two di�erent
formalisations that are not logically equivalent.
More interestingly, there are also ambiguities that can be analysed in
predicate logic by using two di�erent sentences that do not only disagree
in their predicate letters of constants but also in their structures. Usually,
the inde�nite article ‘a’ indicates existential quanti�cation, as, for instance,
in the following sentences:
Every student owns a computer.
A house has been damaged by a meteorite.
In some cases, however, the inde�nite article is used to make a general
statement that must be formalised by a universal quanti�er. �e following
sentences are most naturally understood as general claims that are to be
formalised by sentences with a universal quanti�er:

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

A politician will never admit that he has made a mistake.

An electron is negatively charged.

Occasionally, the inde�nite article is ambiguous. O�en one reading will

be more plausible given the context of the sentence, but sometimes it
may be very hard to reach a decision about the appropriate reading. For
instance, it may be di�cult to decide in which way the following sentence
should be understood:

A Labour MP will not agree to this proposal.

�e sentence could be taken to express that at least one Labour MP will

not agree or as expressing that, in general, no Labour MP will agree. Of
course, such ambiguities can be made explicit with formalisations. �e
two readings yield the following two (respective) formalisations:

∃x (Px ∧ ¬Qxa)
∀x (Px → ¬Qxa)

I have used the following dictionary in this formalisation:

P: . . . is a Labour MP
Q: . . . agrees to . . .
a: this proposal
In the next example the inde�nite article expresses existential quan-
ti�cation without ambiguity, but the sentence is ambiguous for another
reason:

A mistake was made by every student.

In a �rst attempt to parse the sentence one could start as follows:

�ere is at least one x (x is a mistake and (every student

made x))

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

Of course one would then go on and analyse ‘every student made x’ as

‘for all y (if (y is a student), then (y made x))’. According to this analysis,
there is at least one mistake that was made by every student; so every
student made the same mistake (and possibly more). On this reading the
logical form would be as follows:

�ere is at least one x ((x is a mistake) and for all y (if (y is a

student), then (y made x)))

Using the dictionary below, this yields the following L� -sentence:

∃x(Qx ∧ ∀y (P y → Ryx)) (�.�)

P: . . . is a student
Q: . . . is a mistake
R: . . . made . . .
�ere is, however, an alternative reading. �e original sentence may be
taken to say what would be more naturally expressed by the following
sentence:

Every student made a mistake,

where it is understood that it could well be the case that each student
made di�erent mistakes, and there is no one mistake that was made by
all students. �is reading results in the following logical form:

For all x (if (x is a student), then there is at least one y ((y is

a mistake) and (x made y)))

�e formalisation is obviously di�erent from (�.�):

∀x (Px → ∃y (Q y ∧ Rx y)) (�.�)

Without additional information, one cannot decide which of the two

readings is the correct one. �e original sentence is ambiguous.

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

�is kind of ambiguity is akin to the scope ambiguities in proposi-

tional logic discussed on page ��. �ere the ‘grouping’ of connectives
was not uniquely determined by the original English sentence, which
resulted in two di�erent formalisations, P ∧ (Q ∨ R) and (P ∧ Q) ∨ R.
In that example it was not clear whether ∨ should be in the scope of ∧
(as in the �rst sentence) or ∧ should be in the scope of ∨. In the present
case, on the �rst reading (�.�), the existential quanti�er comes �rst so that
the universal quanti�er ‘falls under’ the existential one, while the order is
reversed in the second formalisation (�.�). �e de�nition of the scope of
(an occurrence of) a quanti�er is similar to the de�nition of the scope of
a connective on page ��:
���������� �.� (����� �� � ����������). �e scope of an occurrence of
a quanti�er in a sentence � is (the occurrence of ) the smallest L� -sentence
that contains that quanti�er and is part of �.
�us, in (�.�) the entire sentence, including the occurrence of ∀y, is
the scope of (the single occurrence of) the existential quanti�er. �e
scope of the occurrence of the universal quanti�er ∀y is the underbraced
part:
∃x(Qx ∧ ∀y(P y → Ryx))
��� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �
scope of ∀y

So the above ambiguity is another case of scope ambiguity because the

original sentence ‘Every student made a mistake’ leaves it open whether
the universal quanti�er is in the scope of the existential quanti�er or vice
versa.

�.� ��������������

If the constants a and b have the same extension in an L� -structure A,

that is, if a and b denote the same object, then replacing a by b in a true
sentence will yield a true sentence. For instance, if a and b both denote
Rome in some L� -structure A, that is, if �a�A and �b�A are both Rome, then,

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

for example, Pa will be true if and only if Pb is true. �is is easily seen
from De�nition �.� of satisfaction: assume Pa is true, that is, �Pa�A = T
and reason as follows:
�Pa�A = T assumption
�a�A ∈ �P�A De�nition �.�(i)
�b�A ∈ �P�A by assumption �a�A = �b�A
�Pb�A = T De�nition �.�(i)
In the third line I used the assumption that a and b have the same ex-
tension, that is, that �a�A = �b�A . �is shows that if Pa is true in A, so
is Pb.
�e argument generalises to more complex sentences: as long as a
and b have the same extension, they can be replaced in any sentence by
one another without changing the truth-value of that sentence.
Generally, if constants, sentence letters, and predicate letters are re-
placed in an L� -sentence by other constants, sentence letters, and predicate
letters (respectively) that have the same extension in a given L� -structure,
then the truth-value of the sentence in that L� -structure does not change.
I will not prove the general claim that all sentence letters, predicate
letters, and constants with the same extensions respectively can be substi-
tuted ‘salva veritate’ (Latin shorthand for ‘without making a true sentence
false’), but the above example of the sentence Pa should make the gen-
eral claim plausible. Languages in which these substitutions are possible
are called ‘extensional’: in extensional languages a sentence’s truth-value
depends only on the semantical values of the non-logical symbols that is,
on the extensions of the names, on the relations that are the extensions of
predicate expressions, and so on.
In English it is o�en possible to substitute designators denoting the
same object for one another. For instance, the designator ‘Qomolangma’
is just the o�cial Tibetan name for Mount Everest. �us ‘Qomolangma’
and ‘Mount Everest’ denote the same mountain. So if the sentence
Mount Everest is � ��� metres high.

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

is true, then

Qomolangma is � ��� metres high.

must be true. Clearly whether I use ‘Mount Everest’ or ‘Qomolangma’

to denote the highest mountain, the truth-value of the sentence will be
the same. I could even replace ‘Mount Everest’ by the description ‘the
mountain Edmund Hillary climbed on ��th May ����’ without a�ecting
the sentence’s truth-value:

�e mountain Edmund Hillary climbed on ��th May ���� is

� ��� metres high.

So here English behaves very much like L� : I can replace designators

designating the same object by one another without changing the truth-
value of this sentence in the same way I have been able to substitute b
for a in the sentence Pa without changing its truth-value in A (assuming
that �a�A = �b�A ).
English, however, is not extensional. �ere are also sentences such
that substituting designators denoting the same object can change their
truth-value.
Assume that the following sentence is true:

Tom believes that Mount Everest is � ��� metres high.

If Tom does not believe that Qomolangma is Mount Everest if he believes,

for instance, that it is a small mountain in the Alps, then the following
sentence is presumably false:

Tom believes that Qomolangma is ���� metres high.

And the sentence

Tom believes that the mountain Edmund Hillary climbed on

��th May ���� is � ��� metres high

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

might also be false, although ‘Mount Everest’, ‘Qomolangma’, and ‘the

mountain Edmund Hillary climbed on ��th May ����’ all denote the same
object. Tom might have no idea whether Hillary claimed a mountain on
��th May ���� and, if so, which mountain Hillary climbed on that day.
�is example shows that English is not an extensional language.
A similar point can be made about predicate expressions. Assuming
again that the animals with kidneys are exactly the animals with hearts,
the following substitution may transform the true �rst sentence into a
false sentence:�

Tom believes that all snails have hearts.

Tom believes that all snails have kidneys.

Tom might believe that all snails have hearts, but he might not have a
view on whether they also have kidneys. He may even believe that they
lack kidneys.
Another example of the failure of extensionality is the following pair
of sentences:

It is logically true that all animals with hearts have hearts.

It is logically true that all animals with hearts have kidneys.

�e �rst sentence is true, while the second is false: it is not logically true
that all animals with hearts have kidneys.
In the above examples, ‘that’-sentences have been used to produce
counterexamples to the extensionality of English. �e problematic sub-
stitutions were made a�er such phrases as ‘Tom believes that’ or ‘It is
logically true that’. �ere are also cases of simpler sentences, without ‘that’,
where extensionality fails:

Oedipus is looking for his mother.

� �e quali�cations from footnote �, page � footnote � apply.

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

�is sentence may well be true since Oedipus was abandoned as a baby.
Now he is married to and lives with Jocasta who, unbeknownst to him, is
his mother. So the following sentence might be false at the same time:

Oedipus is looking for Jocasta.

She might actually be sitting right next to him. Since ‘Jocasta’ and ‘Oedi-
pus’ mother’ denote the same person, the example provides another case
of the failure of extensionality.
�e failure of extensionality of English imposes certain restrictions
on the formalisations of English sentences in the language of predicate
logic. In the above example, one might be tempted to formalise

Oedipus is looking for Jocasta

as P � ab with the following dictionary:

P�: . . . is looking for . . .

a: Oedipus
b: Jocasta
�is translation is not correct: ‘. . . is looking for . . . ’ does not express
a relation, that is, a set of ordered pairs. If it did, it would express the
set (relation) of all pairs �d, e� such that d is looking for e; and if the
pair �Oedipus, Oedipus’ mother� is in that set, then �Oedipus, Jocasta� is
necessarily also in that set since it is the same ordered pair: ordered pairs
are identical if they agree in the their �rst and second components, and
Oedipus’ mother and Jocasta are the same object. So it would be true that
Oedipus is looking for Jocasta, which is not the case. �us, ‘. . . is looking
for . . . ’ cannot be formalised as a binary predicate letter, because this is
assigned a binary relation as its extension in any L� -structure.
However, one might still formalise ‘. . . is looking for Oedipus’ mother’
by a unary predicate letter. �e English predicate expression ‘. . . is looking
for Oedipus’ mother’ does express a unary relation, that is, a set: the
designator ‘Oedipus’ may be replaced by any other designator for Oedipus

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

without changing the truth-value of the sentence. Generally, whenever

one has a designator t such that ‘t is looking for Oedipus’ mother’ is true
and some designator s refers to the same object as t, the sentence ‘s is
looking for Oedipus’ mother’ is true as well.
By a similar argument one can show that the predicate expression
‘. . . believes that . . . is high’ must not be formalised as a binary predicate
letter. �is can be done by considering the following example again:
Tom believes that Mount Everest is high.
�e formalisation by the sentence R � a� b� with the following dictionary is
not correct:
R� : . . . believes that . . . is high
a� : Tom
b� : Mount Everest
Assume that ‘. . . believes that . . . is high’ is formalised as a binary predicate
letter; then the following problem arises: If B is an L� -structure and the
pair �Tom, Mount Everest� is an element of the semantical value of R � ,
that is, of the relation �R � �B , then �Tom, Qomolangma� is by necessity
also an element of that relation. �is is due to the fact that �Tom, Mount
Everest� and �Tom, Qomolangma� are the same ordered pair with Tom
as �rst component and the highest mountain on earth as the second. But
the sentence
Tom believes that Qomolangma is high
may be false. So ‘. . . believes that . . . is high’ cannot be formalised as a
binary predicate letter.
�e best formalisation with the tools available might be Q � a, where
Q is translated as ‘. . . believes that Mount Everest is high’.
�

Generally one can only use a predicate letter for English predicates if
they express relations.
�is also includes unary predicate letters, which denote unary rela-
tions, that is, sets (see the end of Section �.�).

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

�.� ��������� ����� ��� ��������� �� �������

Many remarks about, and de�nitions of, formalisations carry over from
propositional logic to predicate logic. In particular, De�nition �.� can be
reformulated for predicate logic in the obvious way:
���������� �.�.
(i) An English sentence is logically true in predicate logic if and only if
its formalisation in predicate logic is logically true.
(ii) An English sentence is a contradiction in predicate logic if and only
if its formalisation in predicate logic is a contradiction.
(iii) A set of English sentences is consistent in predicate logic if and only if
the set of their formalisations in predicate logic is semantically con-
sistent.�
Similarly, the de�nition of validity of English arguments in predicate
logic is analogous to De�nition �.�:
���������� �.�. An argument in English is valid in predicate logic if and
only if its formalisation in the language L� of predicate logic is valid.
Of course, an argument in English is valid if it is valid in predicate
logic, that is, if its formalisation in L� is valid. So an argument in English
is valid, if its formalisation in L� is a valid argument. However, on the
view of many logicians, there are valid arguments in English that are not
valid in predicate logic. I will return to the question whether there are
such English arguments later.
As in the case of propositional logic, talking about the formalisation
of a sentence in L� is not unproblematic since there may be more than
one formalisation (see Section �): if a sentence is ambiguous and has two
or more formalisations, the sentence may be logically true in predicate
logic on one reading, but not on another.

� I could have used the notion of syntactic consistency, which is the same as semantic
consistency by �eorem �.�.

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

To illustrate how the methods of predicate logic can be used to analyse

arguments in English, I will now go through some examples, starting
with a simple and famous example:
All men are mortal. Socrates is a man. �erefore, Socrates is
mortal.
Clearly, the argument is valid. In order to establish its validity formally, I
will formalise the premisses as ∀x(Px → Qx) and Pc and the conclusion
as Qc with the following dictionary:
P: . . . is a man
Q: . . . is mortal
c: Socrates
�e resulting argument in L� is valid:
������� �.�. ∀x(Px → Qx), Pc � Qc
Proof. A proof in Natural Deduction looks like this:
∀x(Px → Qx)
∀Elim
Pc Pc → Qc
→Elim
Qc

�e formalisation and the proof show that the English argument is

valid in predicate logic and, thus, formally valid.
Next I will turn to an example that was analysed on page �� in the
chapter on propositional logic:
Unless Alfred is an eminent logician, it is not the case that
both Kurt and Alfred are eminent logicians.
�e sentence has the formalisation
Pa ∨ ¬(Pa ∧ Pb)
with the following dictionary:

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

P: . . . is an eminent logician
a: Alfred
b: Kurt
�e claim � Pa ∨ ¬(Pa ∧ Pb) can be established by the following proof:

[¬(Pa ∧ Pb)]
Pa ∨ ¬(Pa ∧ Pb) [¬(Pa ∨ ¬(Pa ∧ Pb))]
¬Elim
Pa ∧ Pb
∨Intro�
Pa
Pa ∨ ¬(Pa ∧ Pb) [¬(Pa ∨ ¬(Pa ∧ Pb))]
¬Elim
Pa ∨ ¬(Pa ∧ Pb)
Hence, the English sentence is logically true in predicate logic. An inspec-
tion of the proof shows that I have only used rules from propositional
logic: the formalisation of the designator ‘Alfred’ as a constant and of
‘. . . is an eminent logician’ as a predicate letter has not been used at all.
�e parsing of the sentence ‘Alfred is an eminent logician’ into a predicate
expression and a designator is not really needed in order to see that the
sentence is logically true: if I had formalised the sentences ‘Alfred is an
eminent logician’ and ‘Kurt is an eminent logician’ just with two sentence
letters, I would also have obtained a logically true L� -sentence (which is
also a tautology). In fact, on page �� I already showed that the sentence
is logically true in propositional logic, that is, it is a tautology. �us, I did
not need to use the more detailed formalisation in order to establish that
the English sentence is a tautology. �is observation can be generalised:
If a partial formalisation of an English sentence is logically true, then
that English sentence is logically true in predicate logic. Similarly, if a par-
tial formalisation of an English argument is valid, then that English argu-
ment is valid in predicate logic.
By a partial formalisation of a sentence I mean here a translation of
that sentence into the formal language (L� or L� ) that has been obtained
according to the rules for translating, but that has not reached its full
formalisation. �us, in order to show that an argument is valid, one does
not always have to give a full formalisation. Of course it is not wrong to

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

give the full formalisation, but giving merely a partial formalisation will
be less laborious.
I will illustrate this point with the following argument:
Every student has a computer. Wilma doesn’t have a com-
puter. �erefore Wilma isn’t a student.
I have already formalised the �rst premiss on page ��. �is time I will
formalise the expression ‘has a computer’ as a unary predicate letter, that
is, I will not formalise the existential quanti�er contained in ‘. . . has a
computer’ as on page ��.
P: . . . is a student
P� : . . . has a computer
a: Wilma
With this dictionary the formalisation of the �rst premiss is ∀x(Px →
P� x), the formalisation of the second premiss is ¬P� a, and the formalisa-
tion of the conclusion is ¬Pa. �is yields a valid argument in predicate
logic: ∀x(Px → P� x), ¬P� a � ¬Pa (see Exercise �.�(i)). �us, the English
argument is valid in predicate logic. �is has been established without
giving the full formalisation of the �rst premiss as on page ��, that is,
∀x(Px → ∃y(Rx y ∧ Q y)).
�e next argument (or at least a similar one) has played a role in the
development of logic.�
(H) Horses are animals. �erefore every head of a horse is the head of
an animal.

� �e dominating form of logic since antiquity was Aristotle’s syllogistics. But by the ��th
century some shortcomings of syllogistics had become clear. Syllogistics is not incorrect,
and it is not in con�ict with modern logic, but it is weaker than predicate logic. In the
��th century various logicians argued that syllogistics cannot cope with arguments of
certain types. An argument similar to the one above was used by De Morgan (����,
pages ���–���) to demonstrate the insu�ciency of syllogistics because its validity cannot
be shown in syllogistic logic.

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

Here is the logical form of the premiss:

for all x: (if (x is a horse) then (x is an animal))

�us the premiss can be formalised as ∀x(Px → Qx).

�e logical form of the conclusion of (H) is more di�cult to determine.
�e �rst step should be clear, however:

for all x (if x is the head of a horse then x is the head of an

animal)

�e expression ‘x is the head of a horse’ need not be further analysed for

showing the validity of the argument. �e following logical form of the
conclusion will su�ce:

for all x (if there is a y: ((y is a horse) and (x is the head of

y)) then there is a y: ((y is an animal) and (x is the head of
y)))

�us, the conclusion can be formalised as the following L� -sentence:

∀x�∃y(P y ∧ Rx y) → ∃y(Q y ∧ Rx y)�

I have used the following dictionary:

P: . . . is a horse
Q: . . . is an animal
R: . . . is the head of . . .
�e resulting L� -argument is valid:
������� �.��. ∀x(Px → Qx) � ∀x�∃y(P y ∧ Rx y) → ∃y(Q y ∧ Rx y)�

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

Proof.
[Pb ∧ Rab] ∀x(Px → Qx)
Pb Pb → Qb [Pb ∧ Rab]
Qb Rab
Qb ∧ Rab
∃Intro
[∃y(P y ∧ Ray)] ∃y(Q y ∧ Ray)
∃Elim
∃y(Q y ∧ Ray)
→Intro
∃y(P y ∧ Ray) → ∃y(Q y ∧ Ray)
∀Intro
∀x�∃y(P y ∧ Rx y) → ∃y(Q y ∧ Rx y)�

�erefore, argument (H) is valid in predicate logic and, therefore,

logically (formally) valid.
One does need to use a full formalisation in order to show that the
next argument is valid.
�ere is not a single moral person. �erefore all persons are
immoral.
If ‘immoral’ is understood as ‘not moral’, the argument can be formalised
as follows:
¬∃x(Px ∧ Qx) � ∀y(P y → ¬Q y) (�.�)
�e dictionary is obvious:
P: . . . is a person
Q: . . . is moral
Claim (�.�) can be proved as follows:

[Pa ∧ Qa]
∃Intro
¬∃x(Px ∧ Qx) ∃x(Px ∧ Qx) [Pa] [Qa]
¬Intro
¬(Pa ∧ Qa) Pa ∧ Qa
¬Intro
¬Qa
→Intro
Pa → ¬Qa
∀Intro
∀y(P y → ¬Q y)

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

�e arguments considered so far in this section have all been valid. I

will now consider an argument in English that is not valid:
All lottery tickets are winners or losers. �erefore all tickets
are winners.
Using this trivial example I shall explain in some detail how the semantics
of L� can be employed to show that this argument is not valid in predicate
logic.
First I will formalise the argument. �e premiss becomes

∀x(Px → Qx ∨ Rx),

and the conclusion

∀x(Px → Qx).
�e dictionary should be obvious:
P: . . . is a lottery ticket
Q: . . . is a winner
R: . . . is a loser
I want to disprove the validity of the resulting L� -argument, that is, I want
to show the following:

∀x(Px → Qx ∨ Rx) � ∀x(Px → Qx) (�.�)

�is can be achieved by means of a counterexample, that is, by means of

an L� -structure in which the premiss is true and the conclusion is false.
�e L� -structure F constitutes such a counterexample; it has a domain
with only the number � in it:

DF = {�}
�P�F = {�}
�Q�F =�
�R�F = {�}

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

�e semantical values for the other constants, sentence letters, and predi-
cate letters do not matter, and I will not specify them.
Next, I will show that the premiss is true in F. Over the domain DF
there is only a single variable assignment α, because there is only one
object in the domain and for any variable v, α(v) must be �, so �x�Fα = �.
Now one can reason as follows:

�x�Fα ∈ �R�F de�nition of F

�Rx�Fα = T De�nition �.�(i)
�Rx ∨ Qx�Fα = T De�nition �.�(iv)
�Px → Rx ∨ Qx�Fα = T De�nition �.�(v)
�∀x(Px → Rx ∨ Qx)�F = T De�nition �.�(vii)

�e last line holds because there is only one variable assignment over DF .
�us, the premiss is indded true in the L� -structure F.
It remains to show that the conclusion is false in F:

�x�Fα is not in �Q�F de�nition of F

�Qx�Fα = F De�nition �.�(i)
�x�Fα ∈ �P�F de�nition of F
�Px�Fα = T De�nition �.�(i)
�Px → Qx�Fα = F De�nition �.�(v)
�∀x(Px → Qx)�F = F De�nition �.�(vii)

�e last line shows that the conclusion is not true in F, and thus the claim
�.� is established. Consequently, the English argument is not valid in
predicate logic.
Generally, in order to show that an English argument is not valid in
predicate logic, one will formalise the argument and provide a counterex-
ample to the resulting L� -argument.
I have said that one can show that an English argument is valid by
providing a partial formalisation that is a valid L� -argument. In order to

© Volker Halbach ����/����

 � Formalisation in Predicate Logic ���

refute the validity of an English argument, merely partial formalisations

cannot be used: usually an English argument that is valid in predicate
logic will have some L� -formalisation that is not valid. �us, in order to
show that an English argument is not valid in predicate logic, one needs
to consider its full formalisation.

© Volker Halbach ����/����

 � Identity and De�nite Descriptions

In this chapter I will introduce a third formal language. �is new language
L= is the language of predicate logic with identity; it is only a small
re�nement of the language L� of predicate logic: L= is just L� with the
addition of the new symbol = for identity.

�.� ����������� ��� ��������� ��������

Philosophers have distinguished two di�erent notions of identity: quali-

tative identity and numerical identity. In the following example I present
a case of qualitative identity.
�ere is a fountain pen in my teaching room and another
fountain pen in my study at home. �ey are the same model,
the same colour, and both are still in pristine condition. �us,
I have two identical fountain pens.
�ere are two pens, and they are qualitatively identical because they are
in all relevant aspects very similar.
To explain numerical identity, I will expand the example a little bit:
A fountain pen expert sees my pen at home a�er having
seen the pen in my teaching room the day before. He may
wonder whether I have taken the pen home and ask: ‘Is this
the same pen as the pen in your teaching room?’ or ‘Is this
pen identical to the pen I saw yesterday?’
He knows all the ways that the pen at my home and the pen in my teaching
room are similar, and so he is not asking whether they are the same colour

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

or are of the same brand etc; rather he wants to know whether it is the
same pen, that is, whether he has seen two (qualitatively identical) pens
or whether he has seen one and the same pen. So in his question identity
has to be understood numerically. In the numerical sense the pen in my
teaching room is not identical with the pen in my study at home, that is,
there are two pens.
Occasionally it is not clear which kind of identity is at issue in a given
sentence. �e claim

Robin saw the same tree years later in the garden

might be taken to express that Robin saw one and the same tree in the
garden years later, or that he saw a tree of the same kind in the garden
years later.
Qualitative identity may be formalised by a binary predicate letter
of L� . Its treatment in predicate logic with identity does not di�er from
the treatment of most other binary predicates.
Numerical identity, in contrast, is given a special status. In what fol-
lows I shall talk exclusively about numerical identity. Numerical identity
is formalised by a new, special predicate letter.

�.� ��� ������ �� L=

All formulae of L� are also formulae of L= . But L= also includes a new

kind of atomic formula.
���������� �.� (������ �������� �� L= ). All atomic formulae of L� are
atomic formulae of L= . Furthermore, if t� and t� are variables or constants,
then t� = t� is an atomic formula of L= .
Examples of atomic formulae of L= are x = x, x = z�� , a = y, and
a�� = c�� , and atomic formulae of L� such as P�� xxc� . Otherwise, there are
no deviations from the syntax of L� , and one can build complex formulae
using connectives and quanti�ers as in L� in accordance with De�nition
�.� of a formula of L� . Of course, now one must also allow for =. �e new

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

symbol = behaves exactly like a binary predicate letter, with the exception
that = is written between the variables or constants. Writing the identity
symbol like other predicate letters as the �rst symbol in atomic formulae
would look odd, as we are used to writing x = y rather than = x y, but that
is the only reason for writing x = y rather than = x y.
�e formulae of L= are de�ned in the same way as the formulae of L�
in De�nition �.�, with the only exception that the new atomic formulae
can be used.
���������� �.� (�������� �� L= ).
(i) All atomic formulae of L= are formulae of L= .
(ii) If � and ψ are formulae of L= , then ¬�, (� ∧ ψ), (� ∨ ψ), (� → ψ)
and (� ↔ ψ) are formulae of L= .
(iii) If v is a variable and � is a formula of L= then ∀v � and ∃v � are
formulae of L= .
For instance, ¬ x = y and ∀x(Rx y� → y� = x) are formulae of L= .
All other de�nitions from Section �.� carry over as well. Sentences
of L= are de�ned as those formulae in which no variable occurs freely.
Also, the bracketing conventions are the same as for L� .

�.� ���������

�e semantics for L= is just a small variation on the semantics for L� , and

so it is not necessary to introduce a new kind of structure: L� -structures
are used for the semantics of L= .
Only De�nition �.� needs to be amended by adding the following
additional clause to (i)–(viii), where A is an L� -structure, s is a variable
or constant, and t is a variable or constant:
(ix) �s = t�A
α
= T if and only if �s�A
α
= �t�A
α
.
�us, in the semantics for L= , the symbol = is always interpreted as nu-
merical identity. In (ix) the symbol = is used in two di�erent ways: its �rst
occurrence is a symbol of the formal language L= ; the two subsequent

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

occurrences belong to the language we are using to describe L= . In order

to avoid this ambiguity some authors put a dot under the symbol of the
formal language, but this convention has not really caught on. Alterna-
tively, one could avoid the use of = outside the formal language L= by
reformulating (ix) in the following way:
(ix) �e variable assignment α satis�es s = t in A if and only if �s�A
α
and
�t�A are the same object.
α

In the following I will not try to avoid the use of = in these two di�erent
roles: it should be clear for every occurrence of the symbol whether it
is used as a symbol of L= or as a symbol of our everyday mathematical
language.
It follows from clause (ix) that for any L� -structure A, �a = a�A = T
and for any variable assignment α over A, �x = x�A α
= T, because, trivially,
�x�A is the same object as �x�A , and �a�A is the same object as �a�A . Of
α α

course, the same applies to variables other than x and to constants other
than a.
�e de�nitions of validity of arguments, of semantic consistency, of log-
ically true sentences, and so on, carry over from De�nitions �.� and �.�.
�e method of counterexamples can be applied in the same way as it
was for the language L� . As an example I will show that ∃x ∃y ¬ x = y does
not follow in predicate logic with identity from the premiss ∃x Px ∧∃y P y.
������� �.�. ∃x Px ∧ ∃y P y � ∃x ∃y ¬ x = y
�e argument could be a formalisation of the following English argu-
ment:
�ere is a painting and there is a painting. �erefore there
are at least two things.
A more literal translation of the conclusion ∃x ∃y ¬ x = y would be ‘there
is a thing such that there is a thing that is not identical to the �rst’; but
that is just a longwinded way of saying that there are at least two things.
�e premiss just makes the same claim twice, namely that there is a
painting. �e use of the two variables x and y does not imply that there

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

are two di�erent paintings. �is yields the idea for the following proof.
Proof. Let B be an L� -structure with (the painting of) Mona Lisa as the
only element in its domain of discourse, and {the Mona Lisa } as the
extension of P.
DB = { the Mona Lisa }
�P�B = { the Mona Lisa }
First I will show that the premiss is true in this structure.
�ere is exactly one variable assignment α over B: it assigns the Mona
Lisa to all variables, so �x�A
α
is the Mona Lisa.
the Mona Lisa ∈ {the Mona Lisa}
�x�Bα ∈ �P�B
�Px�Bα = T De�nition �.�(i)
�∃x Px�B = T De�nition �.�(viii)
Since α assigns the Mona Lisa to y as well, the same reasoning can be
applied to y:
�y�Bα ∈ �P�B
�P y�Bα = T De�nition �.�(i)
�∃y P y�B = T De�nition �.�(viii)
Taking the last lines together, one can infer the following by De�nition
�.�(iii):
�∃x Px ∧ ∃y P y�B = T
So the premiss is true in B, and it remains to show that the conclusion is
not true in B.
Assume to the contrary that the conclusion is true in B, that is, assume
that �∃x ∃y ¬x = y�B = T. �en, by De�nition �.�(viii), for at least one
variable assignment the following must obtain:
�∃y ¬x = y�Bα = T

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

Applying De�nition �.�(viii) again, one can conclude that there is a vari-
able assignment β over B, di�ering from α in y at most, such that the
following obtains:
�¬ x = y�B = T
β

(In fact, there is only one variable assignment over B and therefore α and
β are the same variable assignment, but I do not make use of this fact
here.)
By De�nition �.�(ii) it follows that

�x = y�B = F.
β
(�.�)

Since there is only one object in the domain of B, namely the Mona Lisa,
�x�B and �y�B are the same object. �us, �x�B = �y�B , which implies the
β β β β

following by the above special supplementary clause (ix) for De�nition �.�:

�x = y�B = T
β

�is contradicts (�.�), which followed from the assumption that the con-
clusion ∃x ∃y ¬x = y is true in B. �us, the conclusion is not true in B,
and it has been shown that ∃x ∃y ¬x = y does not follow from the premiss
∃x Px ∧ ∃y P y.

�.� ����� ����� ��� ��������

In order to obtain a proof system that is sound and complete with respect
to the semantics for L= , the system of Natural Deduction needs to be
expanded so as to include an introduction and an elimination rule for
identity.
�e introduction rule allows one to assume a = a (and similarly for
all other constants) and to discharge a = a immediately:

=Intro Any assumption of the form t = t where t is a constant

can and must be discharged.

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

Hence, a proof with an application of =Intro looks like this:

[t = t]
⋮
To motivate the elimination rule I will look at the following infor-
mal way of reasoning: if one has established that Mount Everest is Qo-
molangma and that Mount Everest is in Asia, then one can conclude that
Qomolangma is in Asia. �e elimination rule for identity is the formal
counterpart of the general principle legitimating this substitution. In this
rule � is a formula of L= with at most one variable v occurring freely.

=Elim If s and t are constants, the result of appending �[t�v]

to a proof of �[s�v] and a proof of s = t or t = s is a proof of
�[t�v].

�e graphical representation of the rule looks as follows:

⋮ ⋮ ⋮ ⋮
�[s�v] s=t �[s�v] t =s
=Elim =Elim
�[t�v] �[t�v]
Strictly speaking, only one of the versions is needed, as from s = t one can
always obtain t = s using only one of the rules, as will be shown in Example
�.�. Having both versions available is, however, more convenient.
I give some examples illustrating the use of these rules.
������� �.�. � ∀x x = x
Proof. �e proof is very short:
[a = a]
∀x x = x
First a = a is assumed and immediately discharged by =Intro. �en ∀Intro
can be applied without violating the restriction on constants.

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

������� �.�. � ∀x ∀y (x = y → y = x)
Proof. [a = a] [a = b]
=Elim
b=a →Intro
a=b → b=a ∀Intro
∀y (a = y → y = a)
∀Intro
∀x∀y (x = y → y = x)
�is proof shows that =Elim does not demand that one replace all the
occurrences of a by b in the step from the �rst to the second line. �is
step is licensed by the rule =Elim, taking � to be the formula x = a.
������� �.�. � ∀x ∀y ∀z (x = y ∧ y = z → x = z)
Proof. [a = b ∧ b = c] [a = b ∧ b = c]
∧Elim� ∧Elim�
a=b b = c =Elim
a=c →Intro
a=b ∧ b=c → a=c ∀Intro
∀z (a = b ∧ b = z → a = z)
∀Intro
∀y∀z (a = y ∧ y = z → a = z)
∀Intro
∀x∀y∀z (x = y ∧ y = z → x = z)

�is proof system, like those of propositional and predicate logic, is

adequate:
������� �.� (��������). Assume that � and all elements of à are L= -
sentences. �en Γ � � if and only if Γ � �.
As in the case of propositional and predicate logic, I will not prove
the Adequacy �eorem here.

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

�.� ���� �� ��������

�e word ‘is’ can play various rôles. It can be used to express predication,
as in ‘Snow is white’ or ‘Jane is a classicist’. In these cases ‘is’ forms part of
the predicate. �e phrase ‘is a classicist’ is formalised as a unary predicate
letter as it does not refer to a speci�c classicist.
In other cases ‘is’ is used to express identity as in ‘Ratzinger is Bene-
dict XVI.’ or ‘St Mary College of Winchester is New College.’ In these
cases, ‘is’ combines two designators and expresses (numerical) identity.
�us, ‘St Mary College of Winchester is New College’ is formalised as
a = b with the obvious dictionary:
a: St Mary College of Winchester
b: New College
�e identity symbol is useful not only for formalising overt identity
statements, as in the above examples. One can also use the identity symbol
to express that there is a certain number of objects of some kind. Assume
that the predicate letter Px has the following entry in the dictionary:
P: . . . is a Wagner opera
�en the claim that there is at least one Wagner opera can be expressed by
existential quanti�cation as ∃x Px. If one wants to say that there are at least
two Wagner operas, however, it does not su�ce to say ∃x∃y (Px ∧ P y) or
∃x Px ∧ ∃y P y, because these two sentences say merely that something
is a Wagner opera and something is a Wagner opera; it does not say that
something is a Wagner opera and something else is a Wagner opera. But
the latter can be expressed using =:

∃x ∃y (Px ∧ P y ∧ ¬ x = y) (�.�)

�is sentence of L= says that there are at least two Wagner operas. Of
course the trick also works with three:

∃x ∃y ∃z (Px ∧ P y ∧ Pz ∧ ¬x = y ∧ ¬x = z ∧ ¬y = z)

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

�is sentence says that there are at least three Wagner operas.
By using identity one can also express that there is at most one Wagner
opera by saying that, if x and y are Wagner operas, then x and y are
identical:
∀x ∀y (Px ∧ P y → x = y) (�.�)
Again this also works for ‘at most two’, ‘at most three’, and so on.
‘�ere are at most two Wagner operas’ can be formalised as

∀x ∀y ∀z (Px ∧ P y ∧ Pz → x = y ∨ y = z ∨ x = z). (�.�)

‘�ere is exactly one Wagner opera’ can now be rephrased as ‘�ere is

at least one Wagner opera and there is at most one Wagner opera’, and I
have already shown how to express the two parts of that claim: ∃x Px says
that there is at least one such opera, and the second part, beginning with
‘at most’, has (�.�) as its formalisation. So ‘�ere is exactly one Wagner
opera’ can be formalised by the following L= -sentence:

∃x Px ∧ ∀x ∀y (Px ∧ P y → x = y) (�.�)

�is can also be expressed by the following logically equivalent formula:

∃x(Px ∧ ∀y(P y → x = y)) (�.�)

�is sentence says that there is a Wagner opera and it is the only one, that
is, any Wagner opera is identical to it. A still more concise version is the
sentence ∃x ∀y(P y ↔ x = y).�
Similarly, one can express in L= that there are exactly two Wagner
operas by combining (�.�) with (�.�):

∃x ∃y (Px ∧ P y ∧ ¬x = y) ∧ ∀x ∀y ∀z (Px ∧ P y ∧ Pz → x = y ∨ y = z ∨ x = z)

By this method one can express, in the language L= , that there are
exactly �� Wagner operas, although this L= -sentence will be painfully long.

� �e equivalence to (�.�) follows from Exercise �.�.

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

One might think that the claim that there are �� Wagner operas involves
also a claim about a mathematical object, namely the number ��. However,
the claim can be formalised without using a predicate letter or constant
for numbers. �erefore, one can dispense with numbers when claiming,
for instance, that there are exactly �� Wagner operas. Some philosophers
have tried to dispense with numbers and other mathematical objects
completely, and the examples of this section show that identity can be
used to express certain claims without reference to numbers, even if these
claims seem to be about numbers at �rst glance.
With these tricks at one’s disposal one can tackle a problem with the
formalisation of designators such as ‘the king of France’, ‘Claudia’s garden’,
‘the tallest tutor of New College who can speak Latin but does not own a
car’, or ‘the car owned by Tim’. Designators of this kind are called ‘de�nite
descriptions’. De�nite descriptions cannot be adequately formalised as
constants. �e following argument is logically valid:

(T) �e car owned by Tim is red. �erefore there is a red car.

Formalising the de�nite description by a constant yields Pa for the pre-

miss and ∃x(Px ∧ Qx) for the conclusion, with the obvious dictionary:

a: the car owned by Tim

P: . . . is red
Q: . . . is a car
Clearly, the argument in L� corresponding to the English argument (T)
is not valid, that is, Pa � ∃x(Px ∧ Qx).� By formalising a de�nite de-
scription as a constant one loses all the information contained in the
de�nite description. As the examples above show, a de�nite description
can contain a lot of information, and condensing ‘the tallest tutor of New
College who can speak Latin but does not own a car’ into a single constant
is bound to be inadequate.

� See Exercise �.�(i).

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

If Tim did not own a car at all, the premiss ‘�e car owned by Tim is
red’ would not be true; if Tim owned two or more cars the premiss would
also be not true, as there would not be any car that is the car owned by
Tim would not exist. �e premiss implies that Tim owns exactly one car.
In fact, the premiss can be rephrased as follows:�
Tim owns exactly one car and it is red.
From the above it is clear how to express the claim that there is exactly
one car owned by Tim. So the sentence can be rephrased in the following
way:
�ere is a car owned by Tim, and it’s his only car (that is,
every car Tim owns is identical to it), and it is red.
Following the pattern of (�.�), the premiss of (T) is formalised as follows:

∃x �(Qx ∧ Rbx) ∧ ∀y(Q y ∧ Rby → x = y) ∧ Px� (�.�)

For this formalisation the dictionary needs to be extended to cover b

and R:
b: Tim
R: . . . owns . . .
When the de�nite description ‘the car owned by Tim’ was formalised
as a constant, the validity of (T) could not be captured by the validity of
its formalisation. �is was the motive for seeking a more re�ned analysis
of the de�nite description. I still have to show that the new, more detailed
analysis actually allows me to show the validity of (T) by establishing
the validity of its translation. �e premiss now is formalised as (�.�) and
the conclusion as before by ∃x(Px ∧ Qx). With this formalisation the
argument is valid in predicate logic with identity:

� �e following is Russell’s (����) theory of de�nite descriptions. For a criticism of Russell’s

theory see (Strawson, ����).

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

������� �.�.
∃x�(Qx ∧ Rbx) ∧ ∀y(Q y ∧ Rby → x = y) ∧ Px� � ∃x(Px ∧ Qx)
�e proof is on the next page.
In some cases it may not be so easy to see that a sentence contains
a de�nite description. Especially identity statements involving de�nite
descriptions may be confusing. Consider the following two sentences:
(i) Jane is a classicist.
(ii) Jane is the classicist.
In the �rst sentence ‘is’ expresses predication. Sentence (ii), however, is
an identity statement: ‘Jane’ is a proper name, while ‘the classicist’ is a
de�nite description. If ‘. . . is a classicist’ is translated as Q� and ‘Jane’ as
c� , sentence (i) becomes Q� c� , while (ii) becomes the following formula,
when formalised in the style of �.�:

∃x(Q� x ∧ ∀y(Q� y → y = x) ∧ c� = x)

�is sentence is logically equivalent to

Q� c� ∧ ∀y(Q� y → y = c� ),

which says that Jane and only Jane is a classicist.�

�e following example of a sentence containing a de�nite description
is due to Russell (����):

�e king of France is bald.

By ‘the king of France’ I mean ‘the present king of France’. �us, this
de�nite description does not refer to any object because France is a re-
public, not a monarchy. Applying the strategy above, one can rephrase
this sentence as follows:

� Someone might could retort that (i) can be analysed as an identity statement as well,
because it says that Jane is identical to some classicist. See Exercise �.� for this analysis.

© Volker Halbach ����/����

© Volker Halbach
[((Qc ∧ Rbc) ∧ ∀y(Q y ∧ Rby → c = y)) ∧ Pc]
(Qc ∧ Rbc) ∧ ∀y(Q y ∧ Rby → c = y)

Pc Qc
[((Qc ∧ Rbc) ∧ ∀y(Q y ∧ Rby → c = y)) ∧ Pc] Qc ∧ Rbc

Pc ∧ Qc
∃Intro
∃x((Qx ∧ Rbx) ∧ ∀y(Q y ∧ Rby → x = y) ∧ Px) ∃x(Px ∧ Qx)
∃Elim
∃x(Px ∧ Qx)
� Identity and De�nite Descriptions

����/����
���
 � Identity and De�nite Descriptions ���

�ere is exactly one king of France, and he is bald.

�is English sentence can be formalised as a sentence expressing that

there is a king of France, he is the only king of France, and he is bald:

∃x(Rxc ∧ ∀y(Ryc → y = x) ∧ Px)

R: . . . is the king of . . .
c: France
P: . . . is bald
Since France is a republic, the sentence ‘�e king of France is bald’ is
false. However, the sentence

�e king of France is not bald (�.�)

is also false under at least one reading: it seems to say that there is exactly
one king of France and that he is not bald, which is also not true since
there is no king of France.
�e following sentence, in contrast, is true:

It is not the case (for whatever reason) that the king of France is bald.
(�.�)
Here the claim that the king of France is bald is rejected: it leaves open
whether there is a king of France who is not bald or whether there is no
king of France at all, or, perhaps, whether there is more than one king, so
that there is nothing that is the king of France.
Sentence (�.�) is most naturally formalised by the following sentence:

∃x(Rxc ∧ ∀y(Ryc → y = x) ∧ ¬Px) (�.��)

�is says that there is a king of France, that he is the only king of France,
and that he is not bald. So only the baldness is denied, not that there is
exactly one king of France.

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

Whereas in �.�� the negation symbol directly precedes Px, in the

formalisation of (�.�) it is at the beginning:

¬∃x(Rxc ∧ ∀y(Ryc → y = x) ∧ Px) (�.��)

�is expresses that it is not the case (for whatever reason) that there is a
king of France, who is the only one, and who is bald.
If these formalisations are correct, then (�.�) and (�.�) have di�erent
meanings. But especially the formalisation of (�.�) is not uncontrover-
sial, and there may be another reading of (�.�) that results in a di�erent
formalisation. �e following claim is a valid argument only if (�.�) is
understood as expressing the same as (�.�):

�e king of France is not bald; for there is no king of France.

�e �rst sentence is the conclusion, the second sentence is the premiss of

the argument. It can be formalised as a valid argument in L� only if (�.�)
is formalised like (�.�) as (�.��). �us, depending on the reading, there
are two formalisations of (�.�) in L= that are not logically equivalent,
namely (�.��) and (�.��). If an English sentence has two non-equivalent
formalisations, it is ambiguous. By comparing (�.��) and (�.��), one can
see that this is a case of a scope ambiguity again. In formalisation (�.��)
the occurrence of ¬ has a ‘narrow’ scope; its scope is only ¬Px. In for-
malisation (�.��) the negation symbol has a ‘wide’ scope: its scope is the
entire sentence. Which formalisation is better has to be decided from
case to case, depending on the context.
�e analysis of de�nite descriptions just sketched allows one to treat
the expression ‘the king of France’ as an expression that does not refer to
an object in any case. �is is an advantage, compared to a formalisation
of ‘the king of France’ as a constant: a constant has exactly one object as
it semantic value in any given L� -structure; a constant refers to an object
in any L� -structure. If a constant is used, the following argument comes
out as valid in L= :

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

It is not the case that the king of France is bald. �erefore

something is not bald.
Let the constant a stand for ‘the king of France’. �en ¬Pa � ∃x ¬Px
can be established by a single application of the rule ∃Intro. However, the
English argument clearly is not valid. �e formalisation is valid, while the
English argument is not, because the semantics of constants in L� and of
de�nite descriptions in English are di�erent: in L� constants always refer
to some object, while in English de�nite descriptions such as ‘the king of
France’ may fail to refer to an object.
If the premiss ‘It is not the case that the king of France is bald’ is
formalised as (�.��)

¬∃x(Rxc ∧ ∀y(Ryc → y = x) ∧ Px),

in accordance with the above proposed analysis of de�nite descriptions,

then its formalisation correctly comes out as not valid. �is example
shows that, at least in some cases, the proposed theory of de�nite de-
scriptions can be used to handle English designators that do not denote
anything.
So far I have formalised proper names as constants. In the light of
what has just been said, one may doubt the universal adequacy of this
formalisation of proper names: a proper name such as ‘Pegasus’ does not
seem to refer to an object (existing now or in the past). One proposal for
dealing with this problem is to formalise proper names in the same way
as de�nite descriptions. But doing so requires a predicate that singles
out Pegasus, such that Pegasus, if he existed, would be the one and only
object satisfying this predicate. Logicians have also played with alternative
semantics, where constants may fail to refer to an object. All of these
proposals are beyond the scope of this text.
�is discussion of de�nite descriptions shows that the identity symbol
of L= is used for more than just formalising overt identity statements
in English. Many more sentences and arguments can be analysed using
identity. So formalisations in L= capture more details than formalisations

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

in L� . Consequently, there are English arguments that can be shown to

be valid in predicate logic with identity but not in predicate logic without
identity. �e argument (T) above is such an argument.
Of course, strictly speaking I still have to explain what I mean by
validity of an argument in predicate logic with identity, but the explanation
should be obvious:
Validity of English arguments (and logical truth etc) in predicate logic
with identity is de�ned analogously to validity of arguments in proposi-
tional and predicate logic in De�nitions �.�, �.�, �.� and �.�.

�.� �������� �� � ������� ��������

Prima facie identity seems to be merely another binary predicate in Eng-

lish. In the semantics of L= , however, the identity symbol is always in-
terpreted as numerical identity, while other binary predicate letters can
be interpreted as arbitrary binary relations. Why does identity receive
this special treatment? Why is there not a predicate logic with special
treatment for other binary predicates such as ‘is smaller than’ or ‘loves’?
One could come up with a special symbol for any of those binary predi-
cates and invent a semantics in which they are always interpreted in the
same way. Why is identity singled out as a logical predicate, while others
are not? I shall illustrate the di�erent treatment of identity and other
relations by considering two examples.
In predicate logic with identity the following argument is valid:

�e morning star is the evening star. �e morning star is a

planet. Hence the evening star is a planet.

In L= the premisses can be formalised as a = b and Pa and the conclusion

as Pb. �e claim a = b, Pa � Pb can easily be established by a proof in
Natural Deduction.
In predicate logic without identity, the best possible formalisation
of the premisses is Rab, Pa and of the conclusion Pb, with an entry in

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

the dictionary for Rx y as ‘. . . is identical to . . . ’. But, by specifying an

L� -structure in which Rab and Pa are true, and Pb is not, Rab, Pa � Pb
can easily be refuted. �us, the above argument is valid in predicate logic
with identity but not in predicate logic without identity. �e reason is
that in L= the interpretation of = is �xed, while in L� the binary predicate
letter R may be interpreted as any arbitrary binary relation.
�e argument can be compared with the following argument:

�e evening star is smaller than Uranus. Uranus is smaller

than Saturn. �erefore the evening star is smaller than Sat-
urn.

�is argument is not valid in predicate without identity or in predicate

logic with identity. If the interpretation of ‘smaller than’ were �xed in the
way the interpretation of identity is �xed in L= , that is, if the predicate
letter for ‘. . . is smaller than . . . ’ were always interpreted as the smaller-
than relation, the argument would come out as valid. Generally, �xing the
interpretation of further predicates, such as ‘is smaller than’, will make
more arguments valid and more sentences logically true, but it is doubtful
whether such notions of validity still capture logical validity and logical
truth.
In Characterisation �.� of valid arguments in Section �.� I stipulated
that an argument is valid if and only if there is no interpretation under
which the premisses are all true and the conclusion is false. In interpreta-
tions, only the vocabulary that is non-logical can be reinterpreted. Logical
vocabulary has been characterised as not subject-speci�c. �e predicate
expression ‘is smaller than’ is arguably subject-speci�c because it can
only be sensibly applied to objects with spatial extension (and perhaps to
numbers); at least it is not so clear whether it can be applied to objects
such as thoughts, properties, laws, or functions. Identity, in contrast, is
not speci�c to any subject. �us, it cannot be reinterpreted. Logicians say
that identity is a logical constant. �is way of distinguishing logical from
non-logical vocabulary is far from being clear and precise, and the dis-

© Volker Halbach ����/����

 � Identity and De�nite Descriptions ���

tinction and the possibility of such a distinction are controversial issues

in the philosophy of logic.
Of course there could still be expressions in English that, while not
contained as logical symbols in L= , are yet logical expressions. In fact,
many logicians believe that the language L= ought to be extended so as to
include additional logical symbols expressing, for instance, ‘it is necessary
that . . . ’ Consequently, they think that there are valid arguments in English
that are not valid in predicate logic with identity. Other philosophers
think that any valid argument of English can be formalised as a valid
argument in L= , so long as certain formalisation tricks are allowed. �ese
controversies are not only crucial for the philosophy of logic and language,
they also impinge on ontology and other core disciplines of philosophy.
At any rate, L� and L= are very powerful languages. Many logicians,
mathematicians, and philosophers believe that L= is su�cient for formal-
ising very comprehensive parts, if not all, of mathematical and scienti�c
discourse. If the language of mathematics and science can indeed be
handled in the language of predicate logic with identity, there is hope that
it can even capture large parts of philosophy.

© Volker Halbach ����/����

 Natural Deduction Rules

������������� �����

⋮ ⋮ ⋮ ⋮
� ψ �∧ψ �∧ψ
∧Intro ∧Elim�
�∧ψ � ψ ∧Elim�

[�]
⋮ ⋮
⋮
� �→ψ
ψ →Elim
→Intro ψ
�→ψ

⋮ ⋮
� ψ
∨Intro� ∨Intro�
�∨ψ �∨ψ

[�] [ψ]
⋮ ⋮ ⋮
�∨ψ χ χ
χ ∨Elim

© Volker Halbach ����/����

 Natural Deduction rules ���

[�] [�] [¬�] [¬�]

⋮ ⋮ ⋮ ⋮
ψ ¬ψ ψ ¬ψ
¬Intro ¬Elim
¬� �

⋮ ⋮
�→ψ ψ→�
↔Intro
�↔ψ
⋮ ⋮
�↔ψ �↔ψ
↔Elim� ↔Elim�
�→ψ ψ→�

��������� �����
⋮ provided the constant t does not
�[t�v] occur in � or in any undischarged
∀Intro
∀v� assumption in the proof of �[t�v].
⋮
∀v�
�[t�v]
∃Intro
∀Elim ∃v�
�[t�v]

[�[t�v]] provided the constant t does not

⋮ ⋮ occur in ∃v �, or in ψ, or in any
∃v� ψ undischarged assumption other
∃Elim than �[t�v] in the proof of ψ.
ψ

��������
⋮ ⋮ ⋮ ⋮
[t = t]
�[s�v] s=t �[s�v] t=s
⋮ =Elim =Elim
�[t�v] �[t�v]

© Volker Halbach ����/����

 Bibliography

Cappelen, Herman and Ernest LePore (Spring ����), Quotation, in

E. N.Zalta, ed., ‘�e Stanford Encyclopedia of Philosophy’.
URL: http://plato.stanford.edu/archives/spr����/entries/quotation/

Church, Alonzo (����), ‘A note on the Entscheidungsproblem’, Journal of

Symbolic Logic �, ��–��.

De Morgan, Augustus (����), Formal Logic, Taylor & Walton, London.

Devlin, Keith (����), Fundamentals of Contemporary Set �eory, second

edn, Springer-Verlag, New York.

Forbes, Graeme (����), Modern Logic, Oxford University Press, Oxford.

Geach, Peter (����), Reference and Generality, Cornell University Press,

Ithaca.

Gentzen, Gerhard (����), ‘Untersuchungen über das natürliche Schließen’,

Mathematische Zeitschri� ��, ���–���, ���–���.

Guttenplan, Samuel (����), �e Languages of Logic, second edn, Blackwell,

Malden.

Halmos, Paul R. (����), Naive Set �eory, D. Van Nostrand Company,

Princeton. Reprinted, Springer-Verlag, New York, NY, ����.

Hodges, Wilfrid (����), Logic: An Introduction to Elementary Logic, sec-

ond edn, Penguin Books, London.

© Volker Halbach ����/����

 Index ���

Jaśkowski, Stanislaw (����), ‘On the rules of supposition in formal

logic’, Nakładem Seminarium Filozo�cznego Wydzialu Matematyczno-
Przyrodniczego Uniwersytetu Warszawskiego .

Lewis, David (����), Counterfactuals, Harvard University Press, Cam-

bridge MA. reissued in ���� by Blackwell, London.

Morris, Charles (����), Foundations of the �eory of Signs, University of

Chicago Press, Chicago.

Moschovakis, Yiannis (����), Notes on Set �eory, Undergraduate texts

in mathematics, second edn, Springer, New York.

Quine, Willard O. V. (����), Mathematical Logic, Norton, New York.

Neuau�age ���� von Harvard University Press (Cambridge, M.A.).

Russell, Bertrand (����), ‘On denoting’, Mind ��, ���–���.

Sainsbury, Mark (����), Logical Forms, second edn, Blackwell, Oxford.

Smith, Peter (����), An Introduction to Formal Logic, Cambridge Univer-

sity Press, Cambridge.

Strawson, Peter F. (����), ‘On referring’, Mind ��, ���–���.

Tarski, Alfred (����), ‘Der Wahrheitsbegri� in den formalisierten

Sprachen’, Studia Philosophica Commentarii Societatis Philosophicae
Polonorum �.

Tennant, Neil (����), Natural Logic, �nd edn, Edinburgh University Press,
Edinburgh.

Troelstra, Arne S. and Helmut Schwichtenberg (����), Basic Proof �eory,

number �� in ‘Cambridge Tracts in �eoretical Computer Science’,
Cambridge University Press, Cambridge.

© Volker Halbach ����/����

 Index

adequacy, ���, ��� subjunctive, ��

for predicate logic, ��� conjunction
for propositional logic, ��� of sentences, ��
ambiguity, ��–��, ��� connectives, ��, ��–��
lexical, ��� in L� , ��
of scope, ��� in L� , ��
antisymmetric, �� in English, ��
argument, ��, ��� main, ��
of a function, �� consistency, ��
arity, �� in predicate logic, ���
arity-index, ��, �� semantic, ���
assignment, ��–��� semantic in L� , ���
assumption, ��� in English, ��
asymmetric, �� semantic, ���, ���
atomic formulae, ��� semantic in L� , ��
of L� , �� syntactic, ���, ���
constants, ��, ��, ��, ��
binary relation, �, �� context, ��
bound occurrence of a variable, contradiction, ��, ��, ��, ���, ���
�� contradictory, ��
Compactness �eorem of propo- counterexamples
sitional logic, �� in L= , ���
completeness, ���, ���, ��� in predicate logic, ���–���
conclusion, �� in propositional logic, ��
conditionals counterfactual
counterfactual, �� conditionals, ��
indicative, �� cut rule, ���

© Volker Halbach ����/����

 Index ���

declarative sentence, �� heart, �

deductive validity, ��
de�nite descriptions, ��� identity
designators, ��, ��, ���, ���, ��� numerical, ���–���
dictionary, �� qualitative, ���–���
disjunction, ��� i�, ��
of sentences, �� inconsistency
domain semantic in L� , ��
of a function, �� syntactic, ���
domains indicative conditionals, ��
of discourse, �� induction, ���
double arrow, ��� inductive validity, ��
double negation elimination, ��� intuitionistic logic, ���

empty set, � kidney, �

equivalence languages
logical, ��, ��, ��, ��� extensional, ���
equivalence relation, �� letters
ex falso quodlibet, ��, ��� propositional, ��
extensional, ��� predicate, ��, ��
extensionality, ��� logical equivalence, ��, ��, ��,
extensions, ��, ��� ���
falsity logical falsity, ��
logical, �� logical form, ���
formal validity, �� propositional logic, ��
formulae logical terms, ��
atomic, ���, ��� logical truth, ��
of L= , ��� logical validity, ��, ��, ��
of L� , �� logically true, ���, ���
free occurrence of a variable, ��
main column, ��, ��
function, ��
main columns, ��
generalisation, �� main connective, ��

© Volker Halbach ����/����

 Index ���

metalinguistic variable, �� root, ���

metavariables, ��–��
satisfaction, ���
Natural Deduction, ��� scope
negation of a quanti�er, ���
of sentences, �� of a connective, ��
scope ambiguity, ��, ���, ���
ordered pair, � semantic inconsistency
pair in L� , ��
ordered, � semantic consistency, ���, ���
partial formalisation, ��� in L� , ��
pragmatics, �� in L� , ���
predicate letters, ��, ��, ��, �� semantic values, ��
predicate logic, �� semantics, ��, ���
premiss, �� sentence letters, ��, ��, ��, ��, ��
proof, ��, ��� sentences
proper names, �� of L� , ��
property, � of L� , ��, ��
propositional consistency, �� of L= , ���
propositional contradiction, �� sentential logic, ��
propositional logic, �� sets, �–�
propositionally valid, ��, �� soundness, ���, ���, ���
provable, ��� structures, ��, ���
in L� , ��
quadruple, �� in L= , ���
quotation, ��–�� in L� , ��–���
subjunctive conditionals, ��
range syllogistics, ���
of a function, �� symmetric, ��
reductio ad absurdum, ��� symmetry, ��
re�exive, �� syntactic consistency, ���, ���
relation, � syntactic inconsistency, ���
binary, �� syntax, ��, ���

© Volker Halbach ����/����

 Index ���

tautology, ��, ��
ternary, ��
ternary relation, ��
transitive, ��
triple, ��
truth
in an L� -structure, ��
in an L� -structure, ���
logical, ��
truth, logical, ���, ���
truth-functional, ��
truth-functional completeness, ��
truth-values, ��, ��, ��

universal quanti�er, ��, ���

valid, ���
valid argument, ��
validity, ��
deductive, ��
inductive, ��
logical, ��, ��
of an argument, ��–��
propositional, ��
value
of a function, ��
values
semantic, ��
variable assignment, ��–���
variables
in L� , ��
metalinguistic, ��

© Volker Halbach ����/����