The Sorites paradox and fuzzy logic

(a)

NOVAK
(b)
PETR HAJEK
and VILEM
(a)

Institute of Computer Science, Academy of Sciences

182 07 Prague, Czech Republic
e-mail: hajek@[Link]
(b)

University of Ostrava
Institute for Research and Applications of Fuzzy Modeling
30. dubna 22, 701 03 Ostrava 1, Czech Republic
e-mail: [Link]@[Link]

Abstract
The sorites paradox (interpreted as the paradox of small natural numbers) is analyzed using mathematical fuzzy logic. In the first part, we
present an extension of BL-fuzzy logic by a new unary connective At of
almost true and the crisp Peano arithmetic extended by a fuzzy predicate
of feasibility. Then we give examples of possible semantics of At and examples of semantics of feasible numbers. In the second part, we present
an analysis of the sorites paradox within fuzzy logic with evaluated syntax
and show that under a very natural assumption we obtain a consistent
fuzzy theory. Thus, sorites is not paradoxical at all.

INTRODUCTION

One of the most striking paradoxes disturbing logical reasoning for more than
two thousand years is the sorites (heap) paradox attributed to the Aristotles
contemporary, Eubulides the Magarian philosopher who is the author of
several other known paradoxes such as the Liar one. The standard form of
sorites paradox is as follows:
One grain does not form a heap. Adding one grain to what is not
yet a heap does not make a heap. Consequently, there are no heaps.
The problem lays in the fact that in small steps proceeding little by little, we
pass from truth to falsehood. Note that other form of the same paradox is the
falakros (bald man): loosing one hair does not make a non-bald man bald, thus
it follows by induction that there are no bald men. Similarly, we can proceed
when reasoning about a lot of other kinds of properties such as few, small,

etc. Apparently, there are small numbers but not all numbers are small. A
common feature of all these examples is vagueness of the concept in concern.
This means that when passing along the possible bearers of vague property, we
are, in principle, unable to distinguish the moment when the given property
ceases to exist.
From the point of view of mathematical logic, let us consider a vague predicate feasible (following Parikh (1971) who investigated the problem in formal
arithmetic.). We will write Fe(n) for n is feasible. The the sorites para Fe(x), we
dox is clear: postulating Fe(0), (x)(Fe(x) Fe(x + 1)) and (x)
immediately come to contradiction with the induction principle.
Parikh (1971) investigated contradictory extensions of Peano arithmetic by
the above axioms strengthened even by a concrete assumption Fe(t) when t
1010

. He showed that even if in this theory one can prove

is a term like 1010
contradiction in t steps, no short proof can give contradiction (short being
precisely [crisply] dened in dependence on t).
Another analysis of this paradox is provided by Vopenka (1979) who introduced the concept of semiset. This is formally a subclass of some set, not being
necessarily a set itself. A typical semiset is that of finite natural numbers F n
containing 0 and being closed under successor, addition and multiplication, but
not containing all natural numbers. This idea gave rise to a new alternative
set theory (AST) which is equiconsistent with Zermelo-Fraenkel set theory. Induction fails for (some) formulas containing semiset variables on the constant
F n. But for each concrete natural number n, AST proves n
F n (
n being the
n-th numeral).
In this paper, we are going to analyze the notion of small natural numbers
inside fuzzy logic. We will demonstrate solution within two fundamental approaches of it, namely the basic fuzzy logic as developed in H
ajek (1998) and
fuzzy logic with evaluated syntax (FLn) (called also Pavelka logic) presented in
detail in Nov
ak et al. (1999). The paper is also a response to some publications
which appeared in the recent years and which provide more or less improper (or
ignorant to the recent achievements of) analysis of the abilities and contribution
of fuzzy logic to the solution of the sorites paradox (cf., e.g. Keefe (2000); Read
(1995)).
The rst analysis of the sorites paradox in fuzzy logic has been provided by
Goguen (1968-69). In this paper, we will make his considerations more precise.
Namely, we oer two possible solutions of the sorites paradox. The main idea of
the rst solution is the following: if x is small then it is almost true that x + 1 is
small (i.e. the truth degree of x+1 is small is only a little less than that of x is
small). In the rst case, almost true is taken to be a fuzzy unary connective
(hedge); the reader may compare it with the connective very true studied in
Hajek (2001). The second solution employs the possibility to consider axioms
true in various degrees and takes the implication (x)(Fe(x) Fe(x + 1)) as
not being fully true. Our arithmetic will be crisp except for the predicate Fe

(see also H
ajek et al. (2000))) .
The reader is assumed to be familiar with both kinds of fuzzy logics: First,
the basic fuzzy predicate logic BL, and three famous stronger logics L
 ukasiewicz
L
 , G
odel G, and product logic . In particular, recall that BL has two conjunctions: the strong conjunction & (whose standard semantics is any continuous t-norm, the residuum of the t-norm being the semantics of the implication
) and the min-conjunction , being dened as & ( ) which gives
the semantics of minimum.
Second, we suppose that the reader knows the predicate fuzzy logic with
evaluated syntax described in details in Nov
ak et al. (1999). Recall that its
basic truth structure is L
 ukasiewicz algebra LL = [0, 1], , , , , 0, 1 and it
deals with evaluated formulas, fuzzy sets of axioms and besides many-valued
interpretation as above, it employs the concepts of fuzzy theory and provability
degree of a formula in it. The completeness theorem states that the provability
and truth degrees of each formula in each fuzzy theory coincide.
In Section 2 we present an extension of BL by a new unary connective At
of almost true and the theory PAat crisp Peano arithmetic extended by the
fuzzy predicate of feasibility; we also prove a simple theorem in PAat . Section 3
contains examples of possible semantics of At and examples of semantics of Fe
(feasible numbers). Section 4 contains analysis of the sorites paradox within
fuzzy logic with evaluated syntax.
Acknowledgement. Partial support of the grant No. A1030004/00 of the
Grant Agency of the Academy of Sciences of the Czech Republic is acknowledged
by the rst author.

ALMOST TRUE AND FEASIBLE

We extend the basic predicate logic BL by a new unary connective At (At()

being read it is almost true that ) and by the following two axiom schemata.
(at1) At(),
(at2) ( ) (At() At()).
In words, if (is true) then is almost true; if implies and is almost true
then is almost true (minor reformulation using (A5)). This logic is denoted
BLat . Examples of truth functions of at are given in the next section.
Now we dene the arithmetic PAat as follows. We extend the language of
Peano arithmetic P A (see Hajek and Pudl
ak (1993)) by a new unary predicate
Fe (Fe(x) is read x is feasible) and work with the variant of BL with function
symbols (see Hajek (2000)). Thus the language consists of
(i) binary equality predicate =, unary predicate Fe,
(ii) constant
0 (zero), unary function symbol S (successor),
) Note that the notion of a large number is discussed in H
ajek (1998) 3.3.21 for L
 ukasiewicz
logic and 4.1.27 for product logic.

(iii) binary function symbols +, (addition, multiplication).

The axioms are as follows:
(i) x = y x = y (crispness axiom for =),
(ii) all axioms of Peano arithmetic (including the denition of x y as
(z)(z + x = y)),
(iii) x < y (Fe(y) Fe(x)),
(iv) Fe(x) (At(Fe(S(x)) At(Fe(x + x)) At(Fe(x x)).
Lemma 1
For each formula of P A (i.e. not containing the predicate Fe), PAat proves
(tertium non datur).

proof: For atomic this is an instance of the crispness axiom of equality;

the rest follows by induction on the complexity of formulas. Let us show the
(since 
induction step for quantiers. Assume  . Then  (x)
), thus  (x)(
(x)
) by generalization and  (x)
(x)

(x)
by the axiom (3). From the last provability we get  (x) (x) (since
(x)) (see Hajek (1998), 5.1.20).
 (x)
,
Similarly, from  we get  (x) , further  (x) (x)
and nally  (x) (x).
2
Theorem 1
A formula not containing Fe is provable in P A over classical logic i it is provable
in PAat over BLat .

proof: If P A  then PAat  thanks to the preceding lemma. Conversely,

if PAat  then inside P A dene At() for each and Fe(x) x = x.

This embeds PAat (over BLat ) non-conservatively into P A (over classical logic)
and P A  .
2
Theorem 2
PAat proves
Fe(x) Fe(y) (At(Fe(x + y)) At(Fe(x y)).

proof: We work in PAat and have the following chain of implications.

(x y & Fe(y)) (x + y y + y & At(Fe(y + y)) At(Fe(x + y))
(noticing that PAat proves u v (At(Fe(v) At(Fe(u))). Hence
x y (Fe(y) At(Fe(x + y))),
x y ((Fe(x) Fe(y)) At(Fe(x + y))).

Figure 1:
Similarly,
y x (Fe(x) Fe(y)) At(Fe(x + y))
and thus, Fe(x) Fe(y) At(Fe(x + y)) by axiom (A6) of BL.
The proof for x y is fully analogous.

Fe(x))
Observe that we have not postulated Fe(0) (thus the assumption (x)
is consistent with PAat (and makes the theory equivalent to P A).
Corollary 1
For each term t(x1 . . . xn ) of P A with the variables indicated, there is a natural
number n such that
PAat  Fe(
0) Fe(x1 ) Fe(xn ) Atn (Fe(t(x1 , . . . , xn )))
(where Atn () is At(At(. . . (At()) . . . )), n copies of At).

EXAMPLES

In this section, we give few examples of various possibilities how the connective
almost true and predicate feasible could be interpreted. Let us remark that such
examples can be produced ad libitum.

3.1

Examples of almost true

First let us collect trivial examples: At()  for all everything is almost
true; At() nothing is almost true; At() At() just says .
Clearly, these are uninteresting examples.
Second, let p be a truth constant for a non-extremal truth value and dene
At() (p ). Clearly, this satises our axioms; furthermore, the following
becomes provable:
At( ) ( At()),
(At() & At( )) At(At()).
Figure 1 shows graphs of this At for L
 ukasiewicz, Godel and product t-norms.

 (for the concept

Third, extend our logic by the square root connective

of square root see H

ohle (1995)) satisfying
& . If our conjunction
is interpreted by a continuous t-norm put x = max{y
| y y = x}. For

L
 ukasiewicz one gets (x+1)/2, for G
odel x, for product x. And dene At()

Lemma 2
For the -square root just dened from a continuous t-norm , the formula
At( ) (At() At())
is a tautology.

proof: We prove

xy =

x y. First,

x y)2 ( x)2 ( y)2 = x y

(using the fact that ((p q) & (p q)) ((p & p) (q &
q)) is a tautology).

2
2

y) then z x y, z x z 2 x y, hence
Conversely,
if z (x

y. Thus x y is the maximal z such that z 2 x y and

z x

2
hence, x y = x y.
Fourth, observe that if a : [0, 1] [0, 1] is a possible semantics of At and
c : [0, 1] [0, 1] is any non-decreasing hedge then
a (x) = max(a(x), c(x))
is also a possible semantics of At. To see this, observe the following:
x y at(x) at(y) max(at(x), c(x)) max(at(y), c(x))
max(at(x), c(x)) max(at(y), c(y))
(recall the BL-tautology (p q) (p r) (q r)). For example, c(x) may
be a constant 0 < c < 1. Or e.g. let d(x) = 0 for x 0.7, d(x) = 1 for x > 0.7.
Combine these with the example above.

3.2

Examples of feasible

(1) Let At() p (p a truth constant, whose truth value, again denoted
by p, is close to 1 but dierent from 1. Extend the standard model N of natural
numbers by the following interpretation be of Fe: Put q1 = 2, qn+1 = qn2 . Put
Fe(0) = Fe(1) = Fe(2) = 1, and for qn < k qn+1 put Fe(k) = pk = p p
where is our favorite continuous t-norm. This makes (N, Fe) to a -model of
PAat .
Indeed, let us verify the axioms for Fe, observing that it is equivalent to
Fe(0) At(Fe(1)), Fe(1) At(Fe(2)), x 2 (Fe(x) At(Fe(x2 ))).

Thus assume k > 2, qn < k qn+1 , then Fe(k) = pn , qn+1 < k2 qn+2 , thus
Fe(k2 ) = pn+1 and hence, the formula Fe(k) (p Fe(k2 )) (or equivalently
(Fe(k) & p) Fe(k2 )) has the truth value 1. For k = 0, 1, 2 the verication is
obvious. If our is L
 ukasiewicz then due to its nilpotence, for some k we shall
have Fe(k) = 0. If is product then Fe(k) > 0 for each k but inf(Fe(k)) = 0.
Fe(x). For product it veries
Thus for L
 ukasiewicz the model veries (x)
Fe(x).
(x) Fe(x) but not (x)

(2) Let at(x) = x and for simplicity work in product logic (thus at(x) =

 that we have to guarantee Fe(0)

x the usualsquare root). Recall
Fe(1), Fe(1) Fe(2) and Fe(n) Fe(n2 ) for n 2.

The simplest thing to do is to put Fe(n) = n1 for n 2 and Fe(1) = 1/ 2,

Fe(0) = 1/4 2 (or e.g. Fe(0) = 34 , Fe(1) = 23 ). If one nds this to decrease too


(
>
0
small)
for
n

2,
Fe(1)
=
Fe(2), Fe(0) =
rapidly
put
Fe(n)
=
1/n

Fe(1).
This model has disadvantage that we have Fe(0) < 1, which
does not t the

intuition. To have Fe(0) = 1, we may put at(x) = p x for some suitable

p < 1.

SORITES IN FUZZY LOGIC

WITH EVALUATED SYNTAX

Recall
that FLn works with evaluated formulas. These are couples of the form

a where is a formula and a [0, 1] is its syntactic evaluation. The latter
can also be understood as an initial information about the truth of . The
truth values are assumed to form the L
 ukasiewicz MV-algebra (based on [0, 1]).
The language of FLn is also supposed to contain truth constants a (truth value
constants as special formulas) for all the truth values a [0, 1] (alternatively,
we may conne ourselves only to all rationals from [0, 1] for the details, see
Hajek (1998); Nov
ak et al. (1999)).
A fuzzy theory T is a fuzzy set of formulas determined by a fuzzy set of
axioms (i.e. axioms need not be completely convincing and thus, their truth
may be smaller). A formula may be proved using an evaluated proof (a common
denition of a proof accompanied by its value). Then T a means that is
provable in T in the degree a where a is supremum of the values of all proofs of
. Then for each model V it holds that a V() and T |=a means that is
true in T in the degree a where a is inmum of the values of in all models.
The completeness theorem states that both degrees are equal.
Note that FLn is a full generalization of the Fregean understanding to inference in logic: the inference proceeds with truth of facts and not with the facts
themselves.
To formalize the sorites paradox, we can proceed in two ways in FLn. The
rst way is analogous to that provided for BL-fuzzy logic in the preceding section.

4.1

Introducing connective At

We introduce a new unary connective At, which must be logically fitting, i.e.
there is n > 0 such that the formula
( )n (At() At())

(1)

is an axiom with the degree 1. This means that its interpretation at : [0, 1]
[0, 1] must be a Lipschitz continuous function (cf. Mesiar and Nov
ak (1997)).
In particular, this means that at preserves arbitrary inma and suprema.
Furthermore, we will assume axioms (at1) and (at2) in the degree 1 as well
as the following special book-keeping axiom:
(B2) At(aa) at(a), a [0, 1]
where at(a) denotes the truth constant for the truth value at(a) when a is given.
Theorem 3 (Completeness)
Let J be a language extended by the logically tting unary connective At. Then
T a A i

T |=a A.

for every fuzzy theory T and a formula A FJ(T ) .

proof: This follows from the assumptions and Novak et al. (1999), Corol2

lary 4.6.
Lemma 3
For every fuzzy theory T and every formula ,
Tat  (x)(At()) At((x))

proof: This follows from the completeness theorem and the assumption (1).
2

Recall that by atn we denote the n-times composite function of at. Let us
now dene an (pseudo-)inverse of the function at by

at(1) (y) = {a | y at(a)}.
Lemma 4
For each n N and a [0, 1),
(a) at(1) (y) y.
(b) at(n) (y) = at(1) at(1) (y) =




{a | y atn (a)}.

ntimes)

(c) atn (0) = 1 i

at(n) (1) = 0.
8

(d) If atn (a) = 1 then at(n) (1) a.

(e) If atn (0) = a then at(n) (a) = 0.

proof: (a) follows from the fact that y {a | y at(a)}.

(b) For n = 2 we prove that

{b | {a | y at(a)} at(b)} = {a | y at2 (a)}
using the fact that at preserves arbitrary inma. Then use induction.
(c), (d), (e) follow immediately from (b).
2
The main disturbing fact about sorites paradox is that using correct way
of reasoning and stemming from apparent truth, we derive an apparently false
conclusion. The main problem, in our opinion, lays in the fact that after adding
one stone to something which may still not be a heap, its form very slightly (imperceptibly) changes. Classical logic has no means to distinguish such situation
and so, it must neglect it. The price it pays for such negligence is disqualication
of the inference process in all cases similar to sorites.
We will now construct a fuzzy theory Tat . Our goal is to enable to introduce
Fe(x) may be provable in the degree 1 and show
explicitly that Fe(
0), (x)
that this is not paradoxical in FLn.
The theory Tat is given by the following fuzzy set of special axioms.
(i) Axioms of the theory PAat in the degree 1.
(ii) Axiom of the path to the sorites:

) for some m0 > 0.
(at3) 1 Atm0 (
From (at3) we immediately obtain Tat  Atn () for every n m0 .
Let us comment on a somewhat surprising axiom (at3). This axiom explicitly
states that the unary connective almost true enables our theory to pass from
truth to falsity. This can be justied as follows.
The connective At says that the given truth value a can be understood as
almost true. It is clear that 1 is almost true without doubts. Lower a may
be almost true in a degree lower than 1 but still greater than a. Hence, it
seems natural to suppose that after, say n steps we may arrive at the conclusion
that a is n-times almost true with truth value 1.
Similarly, we may state that 0 is almost true in some very small positive
degree since in general, we may not be 100% sure that something is really
false. But then, after repeating, say, m steps, we may arrive at the conclusion
that 0 is m-times almost true with truth value a. But then we immediately
obtain that atn+m (0) = 1; this is reected by axiom (at3).
The following is immediate.
Lemma 5

be the fuzzy theory Tat without axiom (at3). If PAat  then Tat
 .
Let Tat
9

Lemma 6
Tat  Atn ()

(2)

for every formula and n m0 .

proof: From (at3) and the fact that  holds for every formula , we
obtain (2) by the following evaluated proof:


1 {provable formula}, 1 At( ) {(at1), rM P }, . . . ,


1 Atm0 () Atm0 () {(at1), (at2), rM P }, 1 Atm0 (
) {(at3)},


1 Atm0 () {rM P }, . . . , 1 Atn () {(at1), rM P }.
2
Theorem 4
Let the fuzzy theory TF e be obtained from Tat by





Fe(x) . (3)
0), 1 (x)(Fe(x) At(Fe(S(x)))), 1 (x)
TF e = Tat 1 Fe(
Then TF e is a consistent fuzzy theory.

proof: We will construct a model N, Fe of TF e . Therefore, we put

Fe(0) = 1,

(4)
(1)

Fe(n + 1) = at

(Fe(n)),

n N.

(5)

It follows from (5) that Fe(n) = at(n) (Fe(0)). Then N, Fe is a model of TF e .
Indeed, put B(n) = {a | Fe(n) at(a)}. Then



Fe(n)
at(a) = at
a =
aB(n)

aB(n)

= at(at(1) (Fe(n))) = at(Fe(n + 1))

because of continuity of at and thus,

(Fe(n) Fe(S(n))) = 1.
nN

Finally, by axiom (at3) we get at(n) (1) = 0 for each n m0 , and so

Fe(n) = at(n) (Fe(0)) = 0, i.e. Fe(n) = 1 by (4), (5) and Lemma 4.
2
A simple example of the denition of at is to put
at(x) = 1 (x + )
for some small > 0. Then atn (x) = 1 (x + n) and we can nd a number
m0 1 such that atm0 (0) = 1. This connective works well also in PAat (cf.
Sections 2 and 3).
10

4.2

Sorites in general fuzzy theory

The solution presented in the previous subsection does not utilize the advantage
of FLn, which is considering evaluation just in the syntax as the initial information about the truth. Therefore, let us return to the observation, that the
step Fe(x) Fe(S(x)) is not fully convincing. We cannot take this formula
as an axiom with truth value 1 but with slightly lower one. The magnitude of
decreasing of 1 depends on the concrete situation how precisely we see the
heap, i.e. where is the level of inperceptibility of changes in its size and shape.
The following theorem, whose formulation is very close to the formulation
provided by Parikh (1971) formalizes the sorites paradox using evaluated syntax
of FLn.
Theorem 5
Let TP A be a fuzzy Peano theory, i.e. its fuzzy set of special axioms consists of
Peano axioms accepted in the degree 1 and are crisp. Furthermore, let 0 < 1
and Fe  J(TP A ) be a new predicate. Then the fuzzy theory





Fe(x) (6)
TF e = TP A 1 Fe(
0), 1 (x)(Fe(x) Fe(S(x))), 1 (x)
is a consistent conservative extension of TP A .

proof: The proof is based on construction of model N, Fe in which

Fe(n + 1) = Fe(n) (1 ),

nN

(7)

where is L
 ukasiewicz conjunction (details can be found in Nov
ak et al.
(1999)).
2
According to this theorem, we can consistently add axioms of the predicate Fe
to the fuzzy theory containing Peano arithmetic.
Corollary 2
For each n N,
TF e e(n) Fe(
n),

e(n) = 0 (1 n).

proof: Starting from 1 Fe(0), we construct a proof of Fe(n) with the value

e(n). On the other hand, by construction of the model V |= TF e , we get

V(Fe(
n)) = e(n).
2

It is clear that we may take arbitrary  in (7). Alternatively, we may also

take some function such that (n) for all n N.
This solution of the Sorites paradox seems to t well the intuition and, moreover, transition from full truth to falsity without contradiction is well justied.
Namely, it can be seen from Corollary 2 that e(0) = 1 and there is a number n0
such that e(n0 ) = 0. The same holds in any model of TF e there are numbers
m0 and n0 , m0 < n0 such that Fe(m0 ) = 1 and Fe(n0 ) = 0. The number n0

11

being the apparently large number) is determined by the threshold and it

actually depends on the context. Thus, it can be, say n0 = 1015 for a computer
memory, n0 = 105 for the human hair, but n0 = 100 for a bus full of people. We
argue that this is precisely in accordance with our intuition and observation.

CONCLUSION

Our axioms appear to be reasonable desiderata for the innite fuzzy notion
of a small natural number and for the underlying concept of almost true
understood as a hedge (unary connective). Let us stress that the aim has been
just to isolate some important axioms characterizing the notion and show their
consistency by giving concrete examples (models); in Section 2, no attempt has
been made to select any particular semantics of almost true and feasible as
recommended, best etc. On the contrary, since the notion is fuzzy, it would
be unwanted to make it crisp real-valued by xing one example as the semantics.
One can suggest further, more restricting axioms (one candidate being Fe(0),
or Fe(1000000) etc.).
In Section 4, we have used fuzzy logic with evaluated syntax. First, we have
shown that the previous approach works well here, too. On the other hand,
evaluated syntax enables us to provide an initial information about transition
from truth to falsity inside sorites simply by doubting the axiom (x)(Fe(x)
Fe(x + 1)). The doubt is expressed by setting its initial truth value slightly
smaller than 1. This gives us a consistent fuzzy theory which mathematizes the
paradox.

References
Goguen, J. A. (1968-69) The logic of inexact concepts, Synthese 19, 325373.
Hajek, P. (1998) Metamathematics of fuzzy logic (Kluwer, Dordrecht).
H
ajek, P. (2000) Function symbols in fuzzy predicate logic, Proc. East West
Fuzzy Colloquium 2000 (Zittau-G
orlitz), 28.
H
ajek, P. (2001) On very true, Fuzzy Sets and Systems 124, 329333.
Hajek, P., Paris, J., Shepherdson, J. (2000) The liar paradox and fuzzy logic,
Journ. Symb. Logic 65, 339346.
Hajek, P. and Pudl
ak, P. (1993) Metamathematics of first-order arithmetic
(Springer, Heidelberg).
Hohle, U. (1995) Commutative residuated l-monoids, in H
ohle, U. and Klement, E. P., Non-Classical Logics and Their Applications to Fuzzy Subsets. A
) Note that by no means we can say that n is the rst large number; it is only the number
0
which we surely know to be large.

12

Handbook of the Mathematical Foundations of Fuzzy Set Theory (Dordrecht:

Kluwer).
Keefe, R. (2000) Theories of Vagueness (Cambridge University Press, Cambridge).
Mesiar, R. and Nov
ak, V. (1997), On Fitting Operations, Proc. of VIIth IFSA
World Congress (Academia, Prague), 286290.
Nov
ak, V. (1996) Paradigm, Formal Properties and Limits of Fuzzy Logic,
Int. J. of General Systems 24, 377405.
Nov
ak, V. and Perlieva, I. (eds.) (2000) Discovering the World With Fuzzy
Logic (Springer-Verlag, Heidelberg (Studies in Fuzziness and Soft Computing,
Vol. 57)).
Nov
ak, V., Perlieva I. and Mockor, J. (1999) Mathematical Principles of Fuzzy
Logic (Kluwer, Boston/Dordrecht).
Parikh, R. (1971), On existence and feasibility in arithmetic, Journ. Symb.
Logic 36, 494508.
Read, S. (1995) Thinking about logic. (Oxford University Press, Oxford).
Vopenka, P. (1979) Mathematics in the alternative set theory (Teubner, Leipzig).

13

Prof. RNDr. Petr H

ajek, DrSc. is a senior scientist in the Institute
of Informatics of the Academy of Sciences of the Czech Republic. He is author of coauthor of .... monographs and ..... scientic papers in the eld of
mathematical logic.
Prof. Ing. Vil
em Nov
ak, DrSc. is the director of the Institute for
Research and Applications of Fuzzy Modeling of the University of Ostrava in
the Czech Republic. He is author or coauthor of 4 monographs and more than
140 scientic papers in the eld of fuzzy logic and fuzzy modeling.

14