Jan 9, 1997 - The following section reviews quantifiers in first order logic and looks into two-valued ... In the following sections, we also denote the truth-value of a proposition P as ..... The following property shows the spe- cialty of the ...
An Overview of Fuzzy Quanti ers, Part 1: Interpretations Liu, Yaxin Laboratory of Articial Intelligence Department of Computer Science and Technology Peking University Beijing 100871, P.R.China Etienne E. Kerre Fuzziness and Uncertainty Modelling Research Unit Department of Applied Mathematics and Computer Science University Gent, Krijgslaan 281 (S9) B-9000 Gent, Belgium January 9, 1997 Abstract Quanti cation is an important topic in fuzzy theory and its applications. An overview is presented for quanti cation in fuzzy theory. After This work has been supported by the International Projects of the Flemish Community Cooperation with P.R.China (No.9604)
1
a brief review of quanti ers in rst order logic, two approaches of generalizing quantifers are given, the algebraic method and the substitution method. By distinguishing the fuzziness of predicates and quanti ers, various approaches to quanti cation in fuzzy logic can be organized. Quantiers in rst order logic can be generalized in crisp sense, and these generalized quanti ers can also be applied to fuzzy sets. Moreover, quanti ers themselves can be fuzzy, i.e., they can only be represented by a fuzzy set. These dierent kinds of quanti cations are identi ed. Quanti ers relate close to the concept of the cardinality of a fuzzy set, which is summarized before investigating fuzzy quanti cations. Dierent to classical logic, various semantics of propositions in fuzzy logic fall into dierent frameworks which are known as the possibility distribution based reasoning system and the many-valued fuzzy logics. Accordingly, numerical and possibilistic interpretation explored in literature are reviewed conforming to these two frameworks.
Keywords: fuzzy logic, cardinality of fuzzy sets, fuzzy quanti er, manyvalued logic, possibility distribution, numerical quanti er, OWA operator.
1 Introduction The expressive ability of rst order logic benets a lot from the universal quantier and the existential quantier, which enable us to make statements about
properties of a class of objects without enumerating them. But it is well known to linguists and logicians that the universal and existential quantiers are still not powerful enough to grasp all the quantications in natural language and in logic as well. Intuitionally, quantiers relate to the concept of cardinality of sets, which indicates the quantity or counting number of a given set. Logical re2
searches are mainly undertaken within the framework outlined by Mostowski 16] early in the year of 1955. Since then a large number of mathematically interesting quantiers, known as generalized quantiers, are discovered and studied in two-valued logic and many-valued logics 15, 11]. Barwise and Cooper 2] motivated the study of generalized quantiers in linguistics 12, 23], the concept of which is di erent to that of logicians and mathematicians. Since the mid-seventies, Zadeh developed his theory of approximate reasoning 32] together with PRUF 34] based on fuzzy set theory and possibility theory, and discussed at large the importance of fuzzy quantiers in natural language 29, 33, 30]. In contrast to linguists and logicians, Zadeh identies the quantiers in natural language, for example many, most, etc., as fuzzy quantiers with the insight that such quantications are fuzzily dened in nature.
The examples of fuzzy quantication are: \There are a lot of skyscrapers in New York", \Most students are single", \Few young men are fat". Also noticing the
relation between quantiers and cardinalities, Zadeh treats fuzzy quantiers, which relate close to the cardinality of fuzzy sets, as fuzzy numbers while distinguishing quantiers of the rst kind or absolute quantiers from quantiers of the second kind or relative quantiers. Examples of the former ones are much more than 10, a great number of, close to 100, etc., while those of the latter are most, little of, about half of, etc. Some quantiers such as many and few can be
used in either sense, depending on the context. Generally speaking, quantiers in logic take the generic form of QxA(x), where Q is the quantier, A(x) is a predicate with a variable x, and the quantication is over x. With the usual contrast of fuzzy versus crisp, we have the following table: 3
A crisp A fuzzy Q
crisp
I
II
Q
fuzzy
III
IV
Type I quantications, such as \All natural numbers are real", \More than half of the countries in the world competed in the Centennial Olympic Games",
are generalized quantications in the viewpoint of classical two-valued logic, usually involving cardinal numbers. Type II quantications can be seen as the extensions of Type I in many-valued fuzzy logic, and the only di erence is that the quantiers are applied to fuzzy sets. This type of quantication also relates to other many-valued logics, some examples are \Some professors are young", \No more than half of students are tall". Type III and IV quantications involve quantiers which are represented by fuzzy sets, or more precisely, possibility distributions. Such quantications are exemplied by \Almost all birds can y", \Few new pop songs can live long", etc.
As an extension of traditional logical quantiers, fuzzy quantiers are studied by various authors with similar classication and assumptions to Zadeh's, but vary greatly in the interpretation and reasoning schemas. Corresponding to the classication of the role of fuzzy sets in approximate reasoning made by Dubois et al. 9, 7], the studies of quantiers under the topic of fuzzy sets are also undertaken within two frameworks: one follows the tradition of many-valued logics, while the other is based on possibility distribution.
The following section reviews quantiers in rst order logic and looks into two-valued extensions. After that, before diving into quantications involving fuzzy sets, we rst examine the concept of cardinality of a fuzzy set. The subsequent sections discuss the non-fuzzy quantication of fuzzy sets and fuzzy 4
quantication. Reasoning with fuzzy quantiers and the applications of fuzzy quantiers are discussed in Part 2 of this paper.
2 Two-Valued Quantications
2.1 Quantications in First Order Logic First, let us recall the semantics and properties of quantiers in rst-order logic. As an extension of propositional logic, rst order logic is enriched mainly by predicates and quantications. In fact, quantications become natural when
predicates are introduced. Predicates enable us to describe if objects in a given set have a common property, while quantiers enable us to make summaries on the set without enumerate it. The complete study of rst order logic can be found in any textbook on mathematical logic 1, 5], here we only describe those parts related to quantiers. The underlying language for rst order logic consists of the following symbols: constants a b c : : : free variables u v w : : : bounded variables x y z : : : n-ary predicate symbols P Q R : :: logical connectives :
_ ^ ! and $
quantiers 8 9 and auxiliary symbols of parentheses and comma. 5
De nition 2.1 A formula A is a string of the above symbols dened recursively: If t1 t2 : : : tn are either constants or free variables and P is an n-ary predicate symbol, P(t1 t2 : : : tn) is a formula if A ia a formula, :A is a formula if A B are formulas, A _ B A ^ B A ! B and A $ B are formulas if A is a formula, u is a free variable and x is a bounded variable which does not occur in A, then 8xAx=u] and 9xAx=u] are formulas, where
Aa=b] means the formula obtained when all occurrences of a in A are substituted by b. A sub-formula of formula A is a substring of A and itself also a formula. If a formula contains no free variables, it is also called a sentence.
Semantics for rst order logic, or an interpretation
I
of a formula A will
consist of: a universe U of individuals assignment of a unique individual I C (a) 2 U to each constant a assignment of a unique n-ary relation I P (P n) U n to each n-ary predicate P and the truth evaluation mapping vI which maps each formula to the truth-value set f?
>g.
De nition 2.2 The truth evaluation mapping vI is dened recursively as: vI (P (t1 t2 : : : tn)) = > i I T (t1 ) I T (t2 ) : : : I T (tn) I P (P)
6
where
8 > < fI C (ti)g I T (ti ) = > :U
if ti is a constant if ti is a free variable
for 1 i n
vI (:A) = > i vI (A) = ? vI (A _ B) = ? i vI (A) = vI (B) = ? vI (A ^ B) = > i vI (A) = vI (B) = > vI (A ! B) = ? i vI (A) = > vI (B) = ? vI (A $ B) = > i vI (A) = vI (B) vI (8xA) = > i vI (Au=x]) = >, where u is a new free variable vI (9xA) = > i there exists a new constant a such that vI (Aa=x]) = >. In the following sections, we also denote the truth-value of a proposition P as (P) for convenience. Obviously, once an interpretation is determined, a relation on U can be derived from a formula. Furthermore, note that the interpretation of a predicate is a relation, we also can claim a formula is equivalent to a predicate, i.e.: for any formula A containing n free variables, an n-ary predicate PA can be dened by PA (x1 x2 : : : xn) $ A. So for convenience, we can write A(u) if u is a free variable occuring in formula A. Moreover, we use the notation A(t) instead of A(u)t=u] if A(u) is a formula containing a free variable u.
Proposition 2.1 8xA(x) , 8yA(y)
(1)
9xA(x) , 9yA(y)
(2)
7
8xA(x) ,
A(u) where u is a variable
(3)
9xA(x) ,
A(a) where a is a constant
(4)
8x8yA(x
y)
, 8y8xA(x
y)
(5)
9x9yA(x
y)
, 9y9xA(x
y)
(6)
:8xA(x) , 9x:A(x)
(7)
8x(A(x) ^ B(x))
, 8xA(x) ^ 8xB(x)
(8)
9x(A(x) _ B(x))
, 9xA(x) _ 9xB(x)
(9)
It is worth noticing that if the universe U is nite, the universal and existential quantiers have equivalent forms in terms of logical connectives. If we enumerate the element in U as u1 u2 : : : un, and introduce new constants u1 u2 : : : un, which are always assigned as the corresponding elements in U, then we can write P(ui). Such an interpretation is slightly di erent to the above denition, but it is easy to verify that the underlying mapping is the same. Now, from the semantics dened above, we have: 8xA(x) , A(u1 ) ^ A(u2 ) ^ ^ A(un)
(10)
9xA(x) , A(u1 ) _ A(u2 ) _ _ A(un ):
(11)
and Usually, it is convenient to interpret the truth-values ? and > as real numbers 0 and 1, respectively. We will adopt this numerical interpretation in the consequent sections.
2.2 Generalization of First Order Logic Quantiers The rst attempts to generalize quantiers in classical logic are the denitions of 9! and 9!!, which are read as \there exists exactly one" and \there exists at 8
most one" respectively, after the equality predicate is introduced. They are
dened as 4 = 4 9!!xP (x) = 9!xP (x)
9x(P (x) ^ 8y(P(y) ! x y))
(12)
8x8y(P(x) ^ P (y) ! x y)
(13)
The essence in the above denition is the distinction between individuals. Obviously, more complex extensions will involve the concept of cardinality or cardinal numbers. According to Yager 25], the extensions of quantiers can be classied into two approaches: the substitution approach and the algebraic approach. In the substitution approach, a quantied proposition is represented by an equivalent logical sentence. The sentence involves atoms which are instances of the predicate evaluated at the individuals in a universe U. Assume U is a nite set of n individuals and P is a predicate which has truth-values (P(ui)) for each ui 2 U. The truth-value of the quantied proposition QxP (x) is the truth-value of a logical sentence only involving P (ui), thus only decided by P(ui). Examples of this kind of interpretation are 8 and 9 dened in (10) and (11). An alternative approach to investigate quantied propositions is the algebraic approach. With the same assumptions as above, we interpret the quantied proposition QxP (x) by associating with Q 1. a subset SQ R, 2. a function FQ FQ((P (u1)) (P(u2)) : : : (P(un))) 2 R such that (QxP (x)) = 1 9
if FQ 2 SQ
For universal quantier, Q = 8, the association would be: SQ = fng
FQ =
n X i=1
(P (ui)):
For existential quantier, Q = 9, the association is: SQ = f1 2 : : : ng FQ =
n X i=1
(P(ui)):
This approach is comparable to Mostowski's method 16], in which a quantier Q is equivalent to a second order binary predicate TQ , whose arguments are the cardinalities of each of the two parts of a bi-partition of the universe U given by the quantied predicate P according to its truth-values at the elements of U. In this approach, the universal and existential quantiers are represented as: 4 T (jP >j jP ?j) = 4 jP ?j = 0 = 8 4 4 9xP (x) = T9 (jP >j jP ?j) = jP >j > 0
8xP (x)
(14) (15)
where P > = fu j u 2 U P(u) = >g and P ? = fu j u 2 U P(u) = ?g, and jAj indicates the cardinality of A. Obviously the relation between FQ and SQ is equivalent to the second order predicate TQ . In the case of two-valued logic, the denitions are equivalent. However in many cases, the algebraic approach gives us a briefer and easier denition as in the following example. For the quantier \a majority of", M, the denition given by the latter is: SM = fd n +2 1 e : : : ng FM =
n X i=1
(P (ui))
where dxe indicates the ceiling of x. The denition given by the substitution approach obviously is more complicated. 10
3 Cardinalities of Fuzzy Sets In this section, we mainly concentrate on nite fuzzy sets. A fuzzy set F is nite i the support of F is nite. In logicians' point of view, natural numbers are rst recognized as the cardinalities of nite crisp sets. Therefore as extensions of cardinality from crisp sets to fuzzy sets, two possibilities can be considered. One approach is to extend the set of cardinalities from natural numbers to non-negative reals, while the other extends a natural number to a fuzzy set whose universe is the set of natural numbers. These kinds of cardinalities are referred to as scalar and fuzzy cardinalities, respectively, according to Dubois and Prade 8]. In classical set theory, cardinalities are dened based on equipotency. Equipotent fuzzy sets should have the same cardinal. The equipotency of fuzzy sets can be dened as follows, a special case of Wygralak's denition 24]:
De nition 3.1 Two fuzzy sets A, B on U are said to be equipotent i for each natural number i,
inf ft j jAtj ig = inf ft j jBtj ig
(16)
supft j jAtj ig = supft j jBtj ig:
(17)
or equivalently
De nition 3.2 If the i-support suppi(F ) of a fuzzy set F is dened as suppi (F) = ft j jFtj = ig
i2N
two fuzzy sets A, B on U are said to be equipotent i
suppi (A) = suppi (B) 11
i 2 N:
(18)
3.1 Scalar Cardinalities De Luca and Termini 6] proposed the following denition for a scalar cardinality, named the power of a fuzzy set.
De nition 3.3 (De Luca and Termini) Let U be the universe, A be a fuzzy set dened on U with the membership function A : U
7! 0
1], and the support
of A, supp(A) = fu j u 2 U A(u) > 0g, is assumed to be nite, then the power of A is dened as: jAj =
X u 2U
A(u):
(19)
It is easy to verify that the denition agrees to the equipotency standard.
Example 3.1 Fuzzy set A, B and C are dened on U = fa b c d e f g hg as A = B = C =
a
b
c
d
e
f
g
h
0:1 0:3 0:6 0:6 0:9 0:7 1:0 0:2
a
b
c
d
e
f
g
h
0:2 0:6 0:7 1:0 0:9 0:6 0:3 0:1
a
b
c
d
e
f
g
h
! ! !
0:1 0:1 0:2 1:0 1:0 1:0 1:0 0:0
The cardinals of the sets are jAj = 0:1 + 0:3 + 0:6 + 0:6 + 0:9 + 0:7 + 1:0 + 0:2 = 4:4 jB j = 0:2 + 0:6 + 0:7 + 1:0 + 0:9 + 0:6 + 0:3 + 0:1 = 4:4 jC j = 0:1 + 0:1 + 0:2 + 1:0 + 1:0 + 1:0 + 1:0 + 0:0 = 4:4
Obviously, A and B are equipotent, and each of them is not equipotent to C , but all of them are of the same cardinality.
12
Obviously some properties for cardinalities of crisp sets still hold for this denition, with new interpretations of set operations for fuzzy sets:
Proposition 3.1 (Dubois and Prade 8]) Let A B be fuzzy sets on a universe U , then 1. A B ) jAj jB j (monotonicity), where A B is dened as 8x 2 U
A(x) B(x)
2. jAj = jU j ; jAj (when U nite) (coverage property), where 8x 2 U
A(x) = 1 ; A(x)
3. jA B j + jA \ B j = jAj + jB j (additivity), where
(A \ B)(x) = T (A(x) B(x))
(A B)(x) = S(A(x) B(x)) 8x 2 U
and this only holds for proper choices of denitions of T (t-norm) and S (t-conorm).
Some suitable denitions for T and S are the following: S(a b) = max(a b)
T(a b) = min(a b)
(20)
S(a b) = a + b ; ab
T(a b) = ab
(21)
T(a b) = max(0 a + b ; 1):
(22)
S(a b) = min(1 a + b) If we choose S and T as (20), AB = A\B =
a
b
c
d
e
f
g
h
0:2 0:6 0:7 1:0 0:9 0:7 1:0 0:2
a
b
c
d
e
f
g
h
0:1 0:3 0:6 0:6 0:9 0:6 0:3 0:1 13
! !
Thus, jA B j = 0:2 + 0:6 + 0:7 + 1:0 + 0:9 + 0:7 + 1:0 + 0:2 = 5:3 jA \ B j = 0:1 + 0:3 + 0:6 + 0:6 + 0:9 + 0:6 + 0:3 + 0:1 = 3:5 jAj + jB j = jA B j + jA \ B j = 8:8
Proposition 3.2 (Zadeh 29]) Let F , G be fuzzy sets on U , then: max(jF j jGj) max(0 jF j + jGj ; jU j)
jF Gj jF j + jGj
(23)
jF \ Gj
(24)
min(jF j jGj):
The following extension of scalar cardinality, referred to as p-power, attributes to Kaufmann 13]: jAjp
=
X u2U
(A(u))p
(25)
where p is a natural number. It is easy to nd out that jAj0 = jsupp(A)j
jAj1
= jAj:
The property of monotonicity is still valid for jAjp, and additivity is only valid for T = min S = max, while coverage property is violated except for p = 1. Gottwald 10] has dened the p-power in terms of -sections. The -section of a fuzzy set A is dened by s (A) = fu 2 U jA(u) = g
0 < 1:
Then the following property holds: jAjp
=
X 0
0.
De nition 3.4 (Zadeh) The fuzzy cardinality of A, such that supp(A) is nite, is denoted as jAjF , whose membership function is as follows: jAjF (n) = supf j jA j = ng
n2N
(27)
here we dene sup = 0:0.
Example 3.2 For A, B, C and U are same as dened in Example 3.1, we have jAjF
jB jF
= =
0 1
2
3
4
5
6
7
8
0:0 1:0 0:9 0:7 0:0 0:6 0:3 0:2 0:1 0:0
0 1
2
3
4
5
6
7
8
0:0 1:0 0:9 0:7 0:0 0:6 0:3 0:2 0:1 0:0 15
!
!
jC jF
=
0 1
2
3
4
5
6
7
8
!
0:0 0:0 0:0 0:0 1:0 0:2 0:0 0:1 0:0 0:0
We have jAjF = jB jF = 6 jC jF . More generally, for fuzzy cardinalities, two fuzzy sets are equipotent i they are of the same cardinal. So in the following examples, we omit B. The following property follows directly from the denition. It reects the property of the cardinalities of the -cuts of a fuzzy set.
Proposition 3.3 (Dubois and Prade 8]) A is a fuzzy set on a universe U , suppose f j 9u 2 U
A(u) = 0 < < 1g = f1 2 : : : m g
where 0 = 0 < 1 < < m < m+1 = 1. Then for 1 i m + 1 8 0
< QII (x) x a II Q; (x) = >: 1 x > a:
8> < QII (x) x b II Q+ (x) = >: 1 x < b:
(64)
Then we can have QII
= QII; and QII+ :
where QII (x) = T (QII ; (x) QII+ (x))
with T any t-norm operator. Therefore, the overall truth-value is calculated as = T (; + )
(65)
where ; and + are the truth-values calculated with the quantiers QII; and QII+ ,
respectively. 35
Besides the above interpretations, the substitution approach can also be applied 25]. The substitution approach involves the rewriting of an equivalent formula of the quantied proposition in terms of a set of predicates connected by logical connectives. Suppose we investigate the proposition P:\QX's are F", where X is the universe, F is a fuzzy set on X, acting as a predicate. Let VF be the set of all logical sentences whose atomic propositions consist of the predicate F applied to an element in X. We represent a quantier Q by a fuzzy set SFQ on the universe VF , such that for each v 2 VF , SFQ (v) indicates the degree of possibility of v as a meaning for Q. Furthermore, for each v 2 VF , let TF (v) indicate the truth-value of v resulting from its logical structure and the predicate F . From them we can calculate the truth of the quantied proposition P as: (P ) = vmax 2V min(SFQ (v) TF (v)): F
(66)
And for practical use, we should point out that the maximum can be equivalently obtained over the support of SFQ .
References 1] J. Barwise, Mathematical Logic (North-Holland, 1977). 2] J. Barwise, R. Cooper, Generalized quantiers and natural language, Linguistics and Philosophy 4 (1981) 159{219. 3] P. Bosc, L. Lietard, Monotonic quantied statements and fuzzy integrals, in: Proceedings of the First International Joint Conference of NAFIPS, IFIS and NASA (1994) 8{12.
36
4] P. Bosc, L. Lietard, On the comparison of the Sugeno and the Choquet fuzzy integrals for the evaluation of quantied statements, in: Proceedings of the Third European Congress on Intelligent Techniques and Soft Computing (EUFIT'95) vol 2 (1995) 709{716. 5] A. Church, Introduction to Mathematical Logic (Princeton Press, 1970). 6] A. De Luca, S. Termini, A denition of non-probabilistic entropy in the setting of fuzzy sets theory, Information and Control 20 (1972) 301{312. 7] D. Dubois, J. Lang, H. Prade, Fuzzy sets in approximate reasoning, part 2: logical approaches, Fuzzy Sets and Systems 40 (1991) 203{244. 8] D. Dubois, H. Prade, Fuzzy cardinality and the modeling of imprecise quantication, Fuzzy Sets and Systems 16 (1985) 199{230. 9] D. Dubois, H. Prade, Fuzzy sets in approximate reasoning, part 1: inference with possibility distributions, Fuzzy Sets and Systems 40 (1991) 143{202. 10] S. Gottwald, A note on fuzzy cardinals, Kybernetika 16 (1980) 156{158. 11] R. H"ahnle, Commodious axiomazation of quantiers in multiple-valued logic, in: Proceedings of International Symposium on Multi-Valued Logics (1996). 12] L. Hella, Denability hierarchies of generalized quantiers, Annals of Mathematical Logic (1990). 13] A. Kaufmann, Introduction #a la theorie des sous-ensembles ous, vol 4: Complement et Nouvelles Applications (Masson, 1977). 14] G.J. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic: Theory and Applications (Prentice Hall, 1995). 37
15] P. Lindstr"om, First order predicate logic with generalized quantiers, Theoria 32 (1966) 186{195. 16] A. Mostowski, On a generalization of quantiers, Fundamenta Mathematicae 44 (1957) 12{36. 17] V. Novak, Fuzzy Sets and their Applications (Adam Hilger, 1989). 18] H. Prade, A note on the evaluation of conditions involving vague quantiers in presence of imprecise or uncertain information, Bulletin for Studies and Exchanges on Fuzziness and its Applications (BUSEFAL) 32 (1987). 19] D. Ralescu, Cardinality, quantiers, and the aggregation of fuzzy criteria, Fuzzy Sets and Systems 69 (1995) 355{365. 20] N. Rescher, Many-Valued Logic (MacGraw-Hill, 1969). 21] H. Thiele, On T-quantiers and S-quantiers, in: Proceedings of the Twenty-Fourth International Symposium on Multiple-Valued Logic (1994) 264{269. 22] H. Thiele, On fuzzy quantiers, in: Fuzzy Logic and its Applications to Engineering, Information Science and Intelligent Systems (Kluwer Academic Publishers, 1995). 23] J. van Eijck, Generalized quantiers and traditional logic, in: J. van Benthem et al., eds., Generalized Quantiers, Theory and Applications (Foris, 1985). 24] M. Wygralak, Vaguely Dened Objects, (Kluwer Academic Publishers, 1995). 38
25] R.R. Yager, Quantied propositions in a linguistic logic, International Journal of Man-Machine Studies 19 (1983) 195{227. 26] R.R. Yager, Reasoning with fuzzy quantied statements: part I, Kybernetes 14 (1985) 233{240. 27] R.R. Yager, Families of OWA operators, Fuzzy Sets and Systems 59 (1993) 125{148. 28] R.R. Yager, A. Rybalov, Uninorm aggregation operators, Fuzzy Sets and Systems 80 (1996) 111{120. 29] L.A. Zadeh, A computational approach to fuzzy quantiers in natural languages, Computers and Mathematics with Applications 9 (1983) 149{184. 30] L.A. Zadeh, A theory of commonsense knowledge, in: H.J. Skala et al., eds., Aspectes of Vagueness (D. Reidel Publishing Company, 1984) 257{295. 31] L.A. Zadeh, The concept of a linguistic variable and its applications to approximate reasoning, in: R.R. Yager et al., eds., Fuzzy Sets and Applications: Selected Papers by L.A. Zadeh (John Wiley and Sons, 1987) 219{366. 32] L.A. Zadeh, A theory of approximate reasoning, in: R.R. Yager et al., eds., Fuzzy Sets and Applications: Selected Papers by L.A. Zadeh (John Wiley and Sons, 1987) 367{412. 33] L.A. Zadeh, The role of fuzzy logic in the management of uncertainty in expert systems, in: R.R. Yager et al., eds., Fuzzy Sets and Applications: Selected Papers by L.A. Zadeh (John Wiley and Sons, 1987) 413{441. 39
34] L.A. Zadeh, PRUF | a meaning representation language for natural language, in: R.R. Yager et al., eds., Fuzzy Sets and Applications: Selected Papers by L.A. Zadeh (John Wiley and Sons, 1987) 499{568. 35] L.A. Zadeh, Test-score semantics as a basis for a computational approach to the representation of meaning, in: R.R. Yager et al., eds., Fuzzy Sets and Applications: Selected Papers by L.A. Zadeh (John Wiley and Sons, 1987) 655{684.
40