Fuzzy Natural Logic: Theory and Applications › f › Conferences › FSTA2016Sli.pdfFuzzy sets in...
Transcript of Fuzzy Natural Logic: Theory and Applications › f › Conferences › FSTA2016Sli.pdfFuzzy sets in...
Fuzzy Natural Logic: Theory and Applications
Vilem Novak
University of OstravaInstitute for Research and Applications of Fuzzy Modeling
NSC IT4InnovationsOstrava, Czech [email protected]
FSTA 2016, Liptovsky Jan, January 27, 2016
�����
The concept of natural logicGeorge Lakoff: Linguistics and Natural Logic, Synthese 22, 1970
Goals
• to express all concepts capable of being expressed in natural language
• to characterize all the valid inferences that can be made in naturallanguage
• to mesh with adequate linguistic descriptions of all natural languages
Natural logic is a collection of terms and rules that come with naturallanguage and allow us to reason and argue in it
Hypothesis (Lakoff)
Natural language employs a relatively small finite number of atomicpredicates that are used in forming sentences. They are related to eachother by meaning-postulates, e.g.,
REQUIRE(x , y ,S)⇒⇒⇒ PERMIT(x , y , S)that do not vary from language to language.
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 2 / 54
�����
The concept of natural logicGeorge Lakoff: Linguistics and Natural Logic, Synthese 22, 1970
Goals
• to express all concepts capable of being expressed in natural language
• to characterize all the valid inferences that can be made in naturallanguage
• to mesh with adequate linguistic descriptions of all natural languages
Natural logic is a collection of terms and rules that come with naturallanguage and allow us to reason and argue in it
Hypothesis (Lakoff)
Natural language employs a relatively small finite number of atomicpredicates that are used in forming sentences. They are related to eachother by meaning-postulates, e.g.,
REQUIRE(x , y ,S)⇒⇒⇒ PERMIT(x , y , S)that do not vary from language to language.
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 2 / 54
�����
The concept of natural logicGeorge Lakoff: Linguistics and Natural Logic, Synthese 22, 1970
Goals
• to express all concepts capable of being expressed in natural language
• to characterize all the valid inferences that can be made in naturallanguage
• to mesh with adequate linguistic descriptions of all natural languages
Natural logic is a collection of terms and rules that come with naturallanguage and allow us to reason and argue in it
Hypothesis (Lakoff)
Natural language employs a relatively small finite number of atomicpredicates that are used in forming sentences. They are related to eachother by meaning-postulates, e.g.,
REQUIRE(x , y ,S)⇒⇒⇒ PERMIT(x , y , S)that do not vary from language to language.
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 2 / 54
�����
FNL and semantics of natural language
Fuzzy sets in linguistics: L. A. ZadehThe concept of a linguistic variable and its application to approximatereasoning (1975), Quantitative Fuzzy Semantics (1973), A computationalapproach to fuzzy quantifiers in natural languages (1983), Precisiatednatural language (2004)
Sources for the development of FNL
• Results of classical linguistics
• Logical analysis of concepts and semantics of natural languageTransparent Intensional Logic (P. Tichy, P. Materna)
• Montague grammar(English as a Formal Language)
• Results of Mathematical Fuzzy Logic
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 3 / 54
�����
FNL and semantics of natural language
Fuzzy sets in linguistics: L. A. ZadehThe concept of a linguistic variable and its application to approximatereasoning (1975), Quantitative Fuzzy Semantics (1973), A computationalapproach to fuzzy quantifiers in natural languages (1983), Precisiatednatural language (2004)
Sources for the development of FNL
• Results of classical linguistics
• Logical analysis of concepts and semantics of natural languageTransparent Intensional Logic (P. Tichy, P. Materna)
• Montague grammar(English as a Formal Language)
• Results of Mathematical Fuzzy Logic
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 3 / 54
�����
FNL and semantics of natural language
Fuzzy sets in linguistics: L. A. ZadehThe concept of a linguistic variable and its application to approximatereasoning (1975), Quantitative Fuzzy Semantics (1973), A computationalapproach to fuzzy quantifiers in natural languages (1983), Precisiatednatural language (2004)
Sources for the development of FNL
• Results of classical linguistics
• Logical analysis of concepts and semantics of natural languageTransparent Intensional Logic (P. Tichy, P. Materna)
• Montague grammar(English as a Formal Language)
• Results of Mathematical Fuzzy Logic
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 3 / 54
�����
FNL and semantics of natural language
Fuzzy sets in linguistics: L. A. ZadehThe concept of a linguistic variable and its application to approximatereasoning (1975), Quantitative Fuzzy Semantics (1973), A computationalapproach to fuzzy quantifiers in natural languages (1983), Precisiatednatural language (2004)
Sources for the development of FNL
• Results of classical linguistics
• Logical analysis of concepts and semantics of natural languageTransparent Intensional Logic (P. Tichy, P. Materna)
• Montague grammar(English as a Formal Language)
• Results of Mathematical Fuzzy Logic
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 3 / 54
�����
FNL and semantics of natural language
Fuzzy sets in linguistics: L. A. ZadehThe concept of a linguistic variable and its application to approximatereasoning (1975), Quantitative Fuzzy Semantics (1973), A computationalapproach to fuzzy quantifiers in natural languages (1983), Precisiatednatural language (2004)
Sources for the development of FNL
• Results of classical linguistics
• Logical analysis of concepts and semantics of natural languageTransparent Intensional Logic (P. Tichy, P. Materna)
• Montague grammar(English as a Formal Language)
• Results of Mathematical Fuzzy Logic
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 3 / 54
�����
The concept of Fuzzy Natural Logic
Paradigm of FNL
(i) Extend the concept of natural logic to cope with vagueness
(ii) Develop FNL as a mathematical theoryExtension of Mathematical Fuzzy Logic;application of Fuzzy Type Theory
Fuzzy natural logic is a mathematical theory that provides models of termsand rules that come with natural language and allow us to reason andargue in it. At the same time, the theory copes with vagueness of naturallanguage semantics.
(Similar concept was initiated in 1995 by V. Novak under the nameFuzzy logic broader sense (FLb))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 4 / 54
�����
The concept of Fuzzy Natural Logic
Paradigm of FNL
(i) Extend the concept of natural logic to cope with vagueness
(ii) Develop FNL as a mathematical theoryExtension of Mathematical Fuzzy Logic;application of Fuzzy Type Theory
Fuzzy natural logic is a mathematical theory that provides models of termsand rules that come with natural language and allow us to reason andargue in it. At the same time, the theory copes with vagueness of naturallanguage semantics.
(Similar concept was initiated in 1995 by V. Novak under the nameFuzzy logic broader sense (FLb))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 4 / 54
�����
The concept of Fuzzy Natural Logic
Paradigm of FNL
(i) Extend the concept of natural logic to cope with vagueness
(ii) Develop FNL as a mathematical theoryExtension of Mathematical Fuzzy Logic;application of Fuzzy Type Theory
Fuzzy natural logic is a mathematical theory that provides models of termsand rules that come with natural language and allow us to reason andargue in it. At the same time, the theory copes with vagueness of naturallanguage semantics.
(Similar concept was initiated in 1995 by V. Novak under the nameFuzzy logic broader sense (FLb))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 4 / 54
�����
Expected contribution of FNL
• Development of methods for construction of models of systems andprocesses on the basis of expert knowledge expressed in genuinenatural language
• Development of algorithms making computer to “understand” naturallanguage and behave accordingly
• Help to understand the principles of human thinking
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 5 / 54
�����
Expected contribution of FNL
• Development of methods for construction of models of systems andprocesses on the basis of expert knowledge expressed in genuinenatural language
• Development of algorithms making computer to “understand” naturallanguage and behave accordingly
• Help to understand the principles of human thinking
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 5 / 54
�����
Expected contribution of FNL
• Development of methods for construction of models of systems andprocesses on the basis of expert knowledge expressed in genuinenatural language
• Development of algorithms making computer to “understand” naturallanguage and behave accordingly
• Help to understand the principles of human thinking
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 5 / 54
�����
Current constituents of FNL
• Theory of evaluative linguistic expressions
• Theory of fuzzy/linguistic IF-THEN rules and logical inference(Perception-based Logical Deduction)
• Theory of fuzzy generalized and intermediate quantifiers includinggeneralized Aristotle syllogisms and square of opposition
• Examples of formalization of special commonsense reasoning cases
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 6 / 54
�����
Current constituents of FNL
• Theory of evaluative linguistic expressions
• Theory of fuzzy/linguistic IF-THEN rules and logical inference(Perception-based Logical Deduction)
• Theory of fuzzy generalized and intermediate quantifiers includinggeneralized Aristotle syllogisms and square of opposition
• Examples of formalization of special commonsense reasoning cases
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 6 / 54
�����
Current constituents of FNL
• Theory of evaluative linguistic expressions
• Theory of fuzzy/linguistic IF-THEN rules and logical inference(Perception-based Logical Deduction)
• Theory of fuzzy generalized and intermediate quantifiers includinggeneralized Aristotle syllogisms and square of opposition
• Examples of formalization of special commonsense reasoning cases
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 6 / 54
�����
Current constituents of FNL
• Theory of evaluative linguistic expressions
• Theory of fuzzy/linguistic IF-THEN rules and logical inference(Perception-based Logical Deduction)
• Theory of fuzzy generalized and intermediate quantifiers includinggeneralized Aristotle syllogisms and square of opposition
• Examples of formalization of special commonsense reasoning cases
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 6 / 54
�����
Higher-order fuzzy logic — Fuzzy Type Theory
Formal logical analysis of concepts and natural language expressionsrequires higher-order logic — type theory.
Why fuzzy type theory
• It is a constituent of Mathematical Fuzzy Logic, well established withsound mathematical properties.
• Enables to include model of vagueness in the developed mathematicalmodels
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 7 / 54
�����
Higher-order fuzzy logic — Fuzzy Type Theory
Formal logical analysis of concepts and natural language expressionsrequires higher-order logic — type theory.
Why fuzzy type theory
• It is a constituent of Mathematical Fuzzy Logic, well established withsound mathematical properties.
• Enables to include model of vagueness in the developed mathematicalmodels
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 7 / 54
�����
Higher-order fuzzy logic — Fuzzy Type Theory
Formal logical analysis of concepts and natural language expressionsrequires higher-order logic — type theory.
Why fuzzy type theory
• It is a constituent of Mathematical Fuzzy Logic, well established withsound mathematical properties.
• Enables to include model of vagueness in the developed mathematicalmodels
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 7 / 54
�����
Fuzzy Type Theory
Generalization of classical type theory
Syntax of FTT is an extended lambda calculus:
• more logical axioms
• many-valued semantics
Main fuzzy type theories
IMTL, Lukasiewicz, EQ-algebra
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 8 / 54
�����
Fuzzy Type Theory
Generalization of classical type theory
Syntax of FTT is an extended lambda calculus:
• more logical axioms
• many-valued semantics
Main fuzzy type theories
IMTL, Lukasiewicz, EQ-algebra
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 8 / 54
�����
Fuzzy Type Theory
Generalization of classical type theory
Syntax of FTT is an extended lambda calculus:
• more logical axioms
• many-valued semantics
Main fuzzy type theories
IMTL, Lukasiewicz, EQ-algebra
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 8 / 54
�����
Basic concepts
Types
Elementary types: o (truth values), ε (objects)Composed types: βα
Formulas have types: Aα ∈ Formα,Aα ≡ Bα ∈ Formo
λxα Cβ ∈ Formβα
∆∆∆ooAo ∈ Formo
Formulas of type o are propositions
Interpretation of formulas Aβα are functions Mα −→ Mβ
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 9 / 54
�����
Basic concepts
Types
Elementary types: o (truth values), ε (objects)Composed types: βα
Formulas have types: Aα ∈ Formα,Aα ≡ Bα ∈ Formo
λxα Cβ ∈ Formβα
∆∆∆ooAo ∈ Formo
Formulas of type o are propositions
Interpretation of formulas Aβα are functions Mα −→ Mβ
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 9 / 54
�����
Basic concepts
Types
Elementary types: o (truth values), ε (objects)Composed types: βα
Formulas have types: Aα ∈ Formα,Aα ≡ Bα ∈ Formo
λxα Cβ ∈ Formβα
∆∆∆ooAo ∈ Formo
Formulas of type o are propositions
Interpretation of formulas Aβα are functions Mα −→ Mβ
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 9 / 54
�����
Truth values for FNL
Standard Lukasiewicz MV∆-algebra
E = 〈[0, 1],∨,∧,⊗,∆,→, 0, 1〉
∨,∧ = minimum, maximum
a⊗ b = 0 ∨ (a + b − 1) ( Lukasiewicz conjunction)
a→ b = 1 ∧ (1− a + b) ( Lukasiewicz implication)
¬a = a→ 0 = 1− a ¬¬a = a
a↔ b = (a→ b) ∧ (b → a) ∆(a) =
{1 if a = 1,
0 otherwise.
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 10 / 54
�����
Semantics of FTT
FrameM = 〈(Mα,=α)α∈Types ,E∆〉
Fuzzy equality =α: Mα ×Mα −→ L
[x =α x ] = 1 (reflexivity)
[x =α y ] = [y =α x ] (symmetry)
[x =α y ]⊗ [y =α z ] ≤ [x =α z ] (transitivity)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 11 / 54
�����
Generalized completeness (Henkin style)
Theorem
(a) A theory T of fuzzy type theory is consistent iff it hasa general model M .
(b) For every theory T of fuzzy type theory and a formula Ao
T ` Ao iff T |= Ao .
FTT has a lot of interesting properties and great explication power
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 12 / 54
�����
Natural language in FNL
Two important classes of natural language expressions
• Evaluative linguistic expressions
• Intermediate quantifiers
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 13 / 54
�����
Natural language in FNL
Two important classes of natural language expressions
• Evaluative linguistic expressions
• Intermediate quantifiers
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 13 / 54
�����
Evaluative linguistic expressions
Example
very short, rather strong, more or less medium, roughly big, extremely big,very intelligent, significantly important, etc.
• Special expressions of natural language using which people evaluatephenomena and processes that they see around
• They are permanently used in any speech, description of any process,decision situation, characterization of surrounding objects; to bringnew information, people must to evaluate
FNL: Formal logical theory T Ev — semantics of evaluative linguisticexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 14 / 54
�����
Evaluative linguistic expressions
Example
very short, rather strong, more or less medium, roughly big, extremely big,very intelligent, significantly important, etc.
• Special expressions of natural language using which people evaluatephenomena and processes that they see around
• They are permanently used in any speech, description of any process,decision situation, characterization of surrounding objects; to bringnew information, people must to evaluate
FNL: Formal logical theory T Ev — semantics of evaluative linguisticexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 14 / 54
�����
Evaluative linguistic expressions
Example
very short, rather strong, more or less medium, roughly big, extremely big,very intelligent, significantly important, etc.
• Special expressions of natural language using which people evaluatephenomena and processes that they see around
• They are permanently used in any speech, description of any process,decision situation, characterization of surrounding objects; to bringnew information, people must to evaluate
FNL: Formal logical theory T Ev — semantics of evaluative linguisticexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 14 / 54
�����
Evaluative linguistic expressions
Example
very short, rather strong, more or less medium, roughly big, extremely big,very intelligent, significantly important, etc.
• Special expressions of natural language using which people evaluatephenomena and processes that they see around
• They are permanently used in any speech, description of any process,decision situation, characterization of surrounding objects; to bringnew information, people must to evaluate
FNL: Formal logical theory T Ev — semantics of evaluative linguisticexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 14 / 54
�����
Evaluative linguistic expressions
Simple pure evaluative expressions
〈Hedge〉〈TE-adjective〉(i) TE-adjectives:
Gradable ajdectives (small, medium, big; weak, medium strong,strong)Evaluative adjectives (silly, normal, intelligent; good, average, bad)
(ii) Hedges:
• Narrowing: very, extremely, significantly• Widening: more or less, roughly, very roughly• Specifying: approximately, about, rather, precisely, typically
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 15 / 54
�����
Evaluative linguistic expressions
Other evaluative expressions
(i) Compound evaluative expressionsroughly small or medium, small but not very (small)
(ii) Negative evaluative expressionsnot small, NOT very strong, NOT VERY strong
(iii) Fuzzy numbersaround two million, roughly one thousand
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 16 / 54
�����
Semantics of natural language in FNL
Fundamental concepts
(i) context/possible world
(ii) intension
(iii) extension
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 17 / 54
�����
Semantics of natural language in FNL
Possible world
parameter w ; maximal consistent set of known facts
Evaluative linguistic expressions allow a special model:characterize a certain bounded part of an ordered scaleContext = Possible world
w = 〈vL, vS , vR〉, vL, vS , vR ∈ R
Example (Distance)
w = 〈10, 100, 400〉 (Slovakia)w = 〈100, 500, 3 000〉 (USA)
Set of possible worlds/contexts W
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 18 / 54
�����
Semantics of natural language in FNL
Possible world
parameter w ; maximal consistent set of known facts
Evaluative linguistic expressions allow a special model:characterize a certain bounded part of an ordered scaleContext = Possible world
w = 〈vL, vS , vR〉, vL, vS , vR ∈ R
Example (Distance)
w = 〈10, 100, 400〉 (Slovakia)w = 〈100, 500, 3 000〉 (USA)
Set of possible worlds/contexts W
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 18 / 54
�����
Semantics of natural language in FNL
Extension of a linguistic expression A
A class of elements in a given possible world/context having the propertydenoted by AMathematical model of extension (in a context w ∈W ):
Extw (A ) ⊂∼ U
Extension changes when changing the context
Example
Nearw1(150 km)= 0.4 (w1 =Czech Republic)Nearw2(150 km)= 0.95 (w2 =Russia)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 19 / 54
�����
Semantics of natural language in FNL
Extension of a linguistic expression A
A class of elements in a given possible world/context having the propertydenoted by AMathematical model of extension (in a context w ∈W ):
Extw (A ) ⊂∼ U
Extension changes when changing the context
Example
Nearw1(150 km)= 0.4 (w1 =Czech Republic)Nearw2(150 km)= 0.95 (w2 =Russia)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 19 / 54
�����
Extensions of evaluative expressions
Context w = 〈0, 4, 10〉
Small Medium BigExtremely
Very
0 4 10
ExSm
VeSm? Small?Big ?
0 4 10
0 4 10
Roughly
(a)
(b)
(c)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 20 / 54
�����
Construction of extensions of evaluative expressions
Horizon
Each of the limit points vL, vS , vR is a starting point of a horizon runningfrom it in the sense of the ordering towards the next limit point
Horizon is determined by a special indiscernibility relation (fuzzy equality)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 21 / 54
�����
Construction of extensions of evaluative expressions
Horizon
Each of the limit points vL, vS , vR is a starting point of a horizon runningfrom it in the sense of the ordering towards the next limit point
1
LH MH RH
vL vS vR
Horizon is determined by a special indiscernibility relation (fuzzy equality)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 21 / 54
�����
Construction of extensions of evaluative expressions
Horizon
Each of the limit points vL, vS , vR is a starting point of a horizon runningfrom it in the sense of the ordering towards the next limit point
1
LH MH RH
vL vS vR
Horizon is determined by a special indiscernibility relation (fuzzy equality)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 21 / 54
�����
Construction of extensions of evaluative expressions
Extension of evaluative expression is delineated by shifting of the horizonusing linguistic hedge
Hedges — shifts of horizon
ν : [0, 1] −→ [0, 1]
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 22 / 54
�����
Construction of extensions of evaluative expressions
Extension of evaluative expression is delineated by shifting of the horizonusing linguistic hedge
Hedges — shifts of horizon
ν : [0, 1] −→ [0, 1]
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 22 / 54
�����
Construction of extensions of evaluative expressions
a
b
c
a,b,c
vL
vS
vR
1
Mea2
Mea
LHMH
RH
Context
Hedge
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 23 / 54
�����
Extensions of evaluative expressions
• Extremely small, Very small, Small, Very roughly small
• Medium
• Very roughly big, More or less big, Big, Rather big, Very big,Significantly big, Extremely big
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 24 / 54
�����
Semantics of natural language in FNL
Intension of evaluative expression
Formalization of a property; it is invariant with respect to possibleworlds/contexts gives different truth values in various possible worlds
Mathematical model of intension
Int(A ) : W −→ F (U)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 25 / 54
�����
Intension of simple evaluative expressions
Contexts Extensions
֏
֏
֏
various contexts extensions
( ([0,1]))wFWIntension:
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 26 / 54
�����
Intermediate quantifiers
P. L. Peterson, Intermediate Quantifiers. Logic, linguistics, and Aristoteliansemantics, Ashgate, Aldershot 2000.
Quantifiers in natural language
Words (expressions) that precede and modify nouns; tell us how many orhow much. They specify quantity of specimens in the domain of discoursehaving a certain property.
Example
All, Most, Almost all, Few, Many, Some, NoMost women in the party are well dressedFew students passed exam
Intermediate quantifiers form an important subclass of generalized(possibly fuzzy) quantifiers
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 27 / 54
�����
Intermediate quantifiers
P. L. Peterson, Intermediate Quantifiers. Logic, linguistics, and Aristoteliansemantics, Ashgate, Aldershot 2000.
Quantifiers in natural language
Words (expressions) that precede and modify nouns; tell us how many orhow much. They specify quantity of specimens in the domain of discoursehaving a certain property.
Example
All, Most, Almost all, Few, Many, Some, NoMost women in the party are well dressedFew students passed exam
Intermediate quantifiers form an important subclass of generalized(possibly fuzzy) quantifiers
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 27 / 54
�����
Semantics of intermediate quantifiers
Main idea
They are classical quantifiers ∀ and ∃ taken over a smaller class ofelements. Its size is determined using an appropriate evaluative expression.
Classical logic: No substantiation why and how the range of quantificationshould be made smaller
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 28 / 54
�����
Semantics of intermediate quantifiers
Main idea
They are classical quantifiers ∀ and ∃ taken over a smaller class ofelements. Its size is determined using an appropriate evaluative expression.
Classical logic: No substantiation why and how the range of quantificationshould be made smaller
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 28 / 54
�����
Formal theory of intermediate quantifiers
T IQ = T Ev+ 4 special axioms
“〈Quantifier〉 B’s are A”
(Q∀Ev x)(B,A) := (∃z)((∆∆∆(z ⊆ B)︸ ︷︷ ︸
“the greatest” part of B’s
&&&
(∀x)(z x ⇒⇒⇒ Ax))︸ ︷︷ ︸each of B’s has A
∧∧∧
Ev((µB)z))︸ ︷︷ ︸size of z is evaluated by Ev
Ev — extension of a certain evaluative expression
(big, very big, small, etc.)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 29 / 54
�����
Special intermediate quantifiers
P: Almost all B are A := Q∀Bi Ex(B,A)
(∃z)((∆∆∆(z ⊆ B) &&&(∀x)(zx ⇒⇒⇒ Ax))∧∧∧ (Bi Ex)((µB)z))
B: Almost all B are not A := Q∀Bi Ex(B,¬¬¬A)
T: Most B are A := Q∀Bi Ve(B,A)
D: Most B are not A := Q∀Bi Ve(B,¬¬¬A)
K: Many B are A := Q∀¬¬¬(Sm ννν)(B,A)
G: Many B are not A := Q∀¬¬¬(Sm ννν)(B,¬¬¬A)
F: Few B are A := Q∀SmVe(B,A)
H: Few B are not A := Q∀SmVe(B,¬¬¬A)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 30 / 54
�����
Special intermediate quantifiers
Classical quantifiers
A: All B are A := Q∀Bi∆∆∆(B,A) ≡ (∀x)(Bx ⇒⇒⇒ Ax),
E: No B are A := Q∀Bi∆∆∆(B,¬¬¬A) ≡ (∀x)(Bx ⇒⇒⇒¬¬¬Ax),
I: Some B are A := Q∃Bi∆∆∆(B,A) ≡ (∃x)(Bx ∧∧∧ Ax),
O: Some B are not A := Q∃Bi∆∆∆(B,¬¬¬A) ≡ (∃x)(Bx ∧∧∧¬¬¬Ax).
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 31 / 54
�����
Nouns and objects
Simple noun phrase:
small house, very high tree, shallow sea
We evaluate one or more specific features of nouns:height, width, depth, beauty, etc.
Features have values (type α) taken from a set Mα = R
Mathematical model of the semantics of nouns
Objects = sets (vectors) of values of featuresNouns represent (fuzzy) sets of such objects
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 32 / 54
�����
Nouns and objects
Simple noun phrase:
small house, very high tree, shallow sea
We evaluate one or more specific features of nouns:height, width, depth, beauty, etc.
Features have values (type α) taken from a set Mα = R
Mathematical model of the semantics of nouns
Objects = sets (vectors) of values of featuresNouns represent (fuzzy) sets of such objects
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 32 / 54
�����
Nouns and objects
Simple noun phrase:
small house, very high tree, shallow sea
We evaluate one or more specific features of nouns:height, width, depth, beauty, etc.
Features have values (type α) taken from a set Mα = R
Mathematical model of the semantics of nouns
Objects = sets (vectors) of values of featuresNouns represent (fuzzy) sets of such objects
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 32 / 54
�����
Extension of “Very tall Swedes”
w in
Extension of Swedes
Swedes
0
1
(=people)
Extension of Very tall
Features
Very tallSet of people
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 33 / 54
�����
Semantics of “Most Swedes are tall”
SwedesTall
Very big
0 0.5 1
Tall
0
1
(Most Swedes are tall)=
(=people)
w in
Extension of Tall Extension of Very big (size)
Features
Set of people
Extension of Swedes
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 34 / 54
�����
Intension of “Most Swedes are tall”
Intension
λwωϕ · (∃zo(αϕ))((∆∆∆(z ⊆ Swede w) &&&
(∀hαϕ)(zh⇒⇒⇒ Bi(wv)(hv)) &&& BiVe((µ(Swede w)z))
Extension is a truth value
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 35 / 54
�����
Generalized (intermediate) syllogisms
Definition
• A syllogism 〈P1,P2,C 〉 — logical argument in which the conclusion Cis inferred from two premises — major P1 and minor P2.
• Intermediate syllogism: traditional syllogism where one or more of itsformulas are replaced by some containing intermediate quantifiers.
• We say that the syllogism is valid if
T IQ ` P1 &&& P2⇒⇒⇒ C ,
(equivalently, T IQ ` P1⇒⇒⇒ (P2⇒⇒⇒ C ))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 36 / 54
�����
Generalized (intermediate) syllogisms
Definition
• A syllogism 〈P1,P2,C 〉 — logical argument in which the conclusion Cis inferred from two premises — major P1 and minor P2.
• Intermediate syllogism: traditional syllogism where one or more of itsformulas are replaced by some containing intermediate quantifiers.
• We say that the syllogism is valid if
T IQ ` P1 &&& P2⇒⇒⇒ C ,
(equivalently, T IQ ` P1⇒⇒⇒ (P2⇒⇒⇒ C ))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 36 / 54
�����
Generalized (intermediate) syllogisms
Definition
• A syllogism 〈P1,P2,C 〉 — logical argument in which the conclusion Cis inferred from two premises — major P1 and minor P2.
• Intermediate syllogism: traditional syllogism where one or more of itsformulas are replaced by some containing intermediate quantifiers.
• We say that the syllogism is valid if
T IQ ` P1 &&& P2⇒⇒⇒ C ,
(equivalently, T IQ ` P1⇒⇒⇒ (P2⇒⇒⇒ C ))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 36 / 54
�����
105 valid generalized Aristotle’s syllogisms
Figure I
Q1 M is Y
Q2 X is M
Q3 X is Y
Figure II
Q1 Y is M
Q2 X is M
Q3 X is Y
Example (ATT-I)
All women (M) are well dressed (Y ) (∀x)(M x ⇒⇒⇒ Y x)Most people in the party (X ) are women (M) (Q∀
Bi Vex)(X ,M)
Most people in the party (X ) are well dressed (Y ) (Q∀Bi Vex)(X ,Y )
Example (ETO-II)
No lazy people (Y ) pass exam (M) ¬¬¬(∃x)(Yx ∧∧∧Mx)Most students (X ) pass exam (M) (Q∀
Bi Vex)(X ,M)
Some students (X ) are not lazy people (Y ) (∃x)(X ∧∧∧¬¬¬Y )
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 37 / 54
�����
105 valid generalized Aristotle’s syllogisms
Figure I
Q1 M is Y
Q2 X is M
Q3 X is Y
Figure II
Q1 Y is M
Q2 X is M
Q3 X is Y
Example (ATT-I)
All women (M) are well dressed (Y ) (∀x)(M x ⇒⇒⇒ Y x)Most people in the party (X ) are women (M) (Q∀
Bi Vex)(X ,M)
Most people in the party (X ) are well dressed (Y ) (Q∀Bi Vex)(X ,Y )
Example (ETO-II)
No lazy people (Y ) pass exam (M) ¬¬¬(∃x)(Yx ∧∧∧Mx)Most students (X ) pass exam (M) (Q∀
Bi Vex)(X ,M)
Some students (X ) are not lazy people (Y ) (∃x)(X ∧∧∧¬¬¬Y )
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 37 / 54
�����
105 valid generalized Aristotle’s syllogismsFigure III
Q1 M is Y
Q2 M is X
Q3 X is Y
Figure IV
Q1 Y is M
Q2 M is X
Q3 X is Y
Example (PPI-III )
Almost all old people (M) are ill (Y ) (Q∀Bi Exx)(Mx ,Yx)
Almost all old people (M) have gray hair (X ) (Q∀Bi Exx)(Mx ,Xx)
Some people with gray (X ) hair are ill (Y ) (∃x)(Xx ∧∧∧ Yx)
Example (TAI-IV )
Most shares with downward trend (Y ) are from car industry (M)(Q∀
Bi Vex)(Yx ,Mx)All shares of car industry (M) are important (X ) (∀x)(Mx ,Xx)
Some important shares (X ) have downward trend (Y ) (∃x)(Xx ∧∧∧ Yx)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 38 / 54
�����
105 valid generalized Aristotle’s syllogismsFigure III
Q1 M is Y
Q2 M is X
Q3 X is Y
Figure IV
Q1 Y is M
Q2 M is X
Q3 X is Y
Example (PPI-III )
Almost all old people (M) are ill (Y ) (Q∀Bi Exx)(Mx ,Yx)
Almost all old people (M) have gray hair (X ) (Q∀Bi Exx)(Mx ,Xx)
Some people with gray (X ) hair are ill (Y ) (∃x)(Xx ∧∧∧ Yx)
Example (TAI-IV )
Most shares with downward trend (Y ) are from car industry (M)(Q∀
Bi Vex)(Yx ,Mx)All shares of car industry (M) are important (X ) (∀x)(Mx ,Xx)
Some important shares (X ) have downward trend (Y ) (∃x)(Xx ∧∧∧ Yx)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 38 / 54
�����
Square of opposition — basic relations
P1,P2 ∈ Formo are:
(i) contraries if T ` ¬¬¬(P1 &&& P2). By completeness:
M (P1)⊗M (P2) = 0
P1 and P2 cannot be both true but can be both false
(ii) sub-contraries if T ` P1∇∇∇P2. By completeness:
M (P1)⊕M (P2) = 1
P1 and P2 cannot be both false but can be both true
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 39 / 54
�����
Square of opposition — basic relations
(i) P1,P2 ∈ Formo are contradictories if both
T ` ¬¬¬(∆∆∆P1 &&& ∆∆∆P2) as well as T `∆∆∆P1∇∇∇∆∆∆P2.
By completeness:M (∆∆∆P1)⊗M (∆∆∆P2) = 0 M (∆∆∆P1)⊕M (∆∆∆P2) = 1
P1 and P2 cannot be both true as well as both false
(ii) The formula P2 is a subaltern of P1 in T if T ` P1⇒⇒⇒ P2. Bycompleteness:
M (P1) ≤M (P2)
(P1 a superaltern of P2)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 40 / 54
�����
Generalized 5-square of opposition
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 41 / 54
�����
Negation in FNLTopic—focus articulation:
Each sentence is divided into topic (known information) and focus (newinformation)
Focus is negated
Example
It is not true, that:
(i) JOHN reads Jane’s paperSomebody else reads it
(ii) John READS JANE’S PAPERJohn does something else
(iii) John reads JANE’S PAPERJohn reads some other paper
(iv) John reads Jane’s PAPERJohn reads Jane’s book (lying on the same table)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 42 / 54
�����
Negation in FNLTopic—focus articulation:
Each sentence is divided into topic (known information) and focus (newinformation)
Focus is negated
Example
It is not true, that:
(i) JOHN reads Jane’s paperSomebody else reads it
(ii) John READS JANE’S PAPERJohn does something else
(iii) John reads JANE’S PAPERJohn reads some other paper
(iv) John reads Jane’s PAPERJohn reads Jane’s book (lying on the same table)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 42 / 54
�����
Negation in FNLTopic—focus articulation:
Each sentence is divided into topic (known information) and focus (newinformation)
Focus is negated
Example
It is not true, that:
(i) JOHN reads Jane’s paperSomebody else reads it
(ii) John READS JANE’S PAPERJohn does something else
(iii) John reads JANE’S PAPERJohn reads some other paper
(iv) John reads Jane’s PAPERJohn reads Jane’s book (lying on the same table)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 42 / 54
�����
Negation in FNLTopic—focus articulation:
Each sentence is divided into topic (known information) and focus (newinformation)
Focus is negated
Example
It is not true, that:
(i) JOHN reads Jane’s paperSomebody else reads it
(ii) John READS JANE’S PAPERJohn does something else
(iii) John reads JANE’S PAPERJohn reads some other paper
(iv) John reads Jane’s PAPERJohn reads Jane’s book (lying on the same table)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 42 / 54
�����
Negation in FNLTopic—focus articulation:
Each sentence is divided into topic (known information) and focus (newinformation)
Focus is negated
Example
It is not true, that:
(i) JOHN reads Jane’s paperSomebody else reads it
(ii) John READS JANE’S PAPERJohn does something else
(iii) John reads JANE’S PAPERJohn reads some other paper
(iv) John reads Jane’s PAPERJohn reads Jane’s book (lying on the same table)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 42 / 54
�����
Negation in FNL
The law of double negation is in NL fulfilled (and so, in FNL as well)
We do not think (Th) that John did not read this paper (R(J, p))
¬¬¬Th(¬¬¬R(J, p)) ≡ Th(R(J, p))
However:It is not clear (Cl) whether John did not read Jane’s paper (R(J, p))
¬¬¬Cl(¬¬¬R(J, p)) ≡ Cl(R(J, p))?
Consequence: We suspect John to read Jane’s paper
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 43 / 54
�����
Negation in FNL
The law of double negation is in NL fulfilled (and so, in FNL as well)
We do not think (Th) that John did not read this paper (R(J, p))
¬¬¬Th(¬¬¬R(J, p)) ≡ Th(R(J, p))
However:It is not clear (Cl) whether John did not read Jane’s paper (R(J, p))
¬¬¬Cl(¬¬¬R(J, p)) ≡ Cl(R(J, p))?
Consequence: We suspect John to read Jane’s paper
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 43 / 54
�����
Negation in FNL
not unhappy 6≡ happyThis is not violation of double negation
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 44 / 54
�����
Negation in FNL
not unhappy 6≡ happyThis is not violation of double negation
unhappy is antonym of happy
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 44 / 54
�����
Fuzzy/linguistic IF-THEN rules and logical inference
Fuzzy/Linguistic IF-THEN rule
IF X is A THEN Y is B
Conditional sentence of natural language
Example: IF X is small THEN Y is extremely strong
Linguistic description
IF X is A1 THEN Y is B1
IF X is A2 THEN Y is B2
. . . . . . . . . . . . . . . . . . . . . . . .IF X is Am THEN Y is Bm
Text describing one’s behavior in some (decision) situation
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 45 / 54
�����
Fuzzy/linguistic IF-THEN rules and logical inference
Fuzzy/Linguistic IF-THEN rule
IF X is A THEN Y is B
Conditional sentence of natural language
Example: IF X is small THEN Y is extremely strong
Linguistic description
IF X is A1 THEN Y is B1
IF X is A2 THEN Y is B2
. . . . . . . . . . . . . . . . . . . . . . . .IF X is Am THEN Y is Bm
Text describing one’s behavior in some (decision) situation
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 45 / 54
�����
Perception-based logical deduction
Crossing strategy when approaching intersection:
R1 := IF Distance is very small THEN Acceleration is very big
R2 := IF Distance is small THEN Brake is big
R3 := IF Distance is medium or big THEN Brake is zero
• Gives general rules for driver’s behavior independently on the concreteplace
• People are able to follow them in an arbitrary signalized intersection
We can “distinguish” between the rules despite the fact that their meaningis vague
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 46 / 54
�����
Perception-based logical deduction
Crossing strategy when approaching intersection:
R1 := IF Distance is very small THEN Acceleration is very big
R2 := IF Distance is small THEN Brake is big
R3 := IF Distance is medium or big THEN Brake is zero
• Gives general rules for driver’s behavior independently on the concreteplace
• People are able to follow them in an arbitrary signalized intersection
We can “distinguish” between the rules despite the fact that their meaningis vague
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 46 / 54
�����
Perception-based logical deduction (PbLD)
• It imitates human way of reasoning
• It realizes algorithms based on an expert knowledge specified innatural language
• Linguistic description can be learned from datapresence of an expert is not necessary
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 47 / 54
�����
Applications of FNL
• Fuzzy/linguistic control (real applications)
• Decision making
• Classification and evaluation
• Time series forecasting (forecasting of trend or trend-cycle)
• Mining information from time series in the form of natural languageexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 48 / 54
�����
Applications of FNL
• Fuzzy/linguistic control (real applications)
• Decision making
• Classification and evaluation
• Time series forecasting (forecasting of trend or trend-cycle)
• Mining information from time series in the form of natural languageexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 48 / 54
�����
Applications of FNL
• Fuzzy/linguistic control (real applications)
• Decision making
• Classification and evaluation
• Time series forecasting (forecasting of trend or trend-cycle)
• Mining information from time series in the form of natural languageexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 48 / 54
�����
Applications of FNL
• Fuzzy/linguistic control (real applications)
• Decision making
• Classification and evaluation
• Time series forecasting (forecasting of trend or trend-cycle)
• Mining information from time series in the form of natural languageexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 48 / 54
�����
Applications of FNL
• Fuzzy/linguistic control (real applications)
• Decision making
• Classification and evaluation
• Time series forecasting (forecasting of trend or trend-cycle)
• Mining information from time series in the form of natural languageexpressions
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 48 / 54
�����
Conclusions
• The concept of fuzzy natural logic was presented
• Formalization of the semantics of two classes of natural languageexpressions
• Human reasoning: perception-based logic deduction (PbLD),intermediate syllogisms, generalized square of opposition
• FNL includes a mathematical model of the vagueness phenomenon.
Model of the Sorites/Falakros paradox appearing in the semanticsof evaluative expressions
• Many open problems:
• How surface structures can be transformed into logical formulasrepresenting their meaning
• Formalization of negation• Formalization of hedging• Formalization of presupposition and its consequence• Formalization of the meaning of verbs
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 49 / 54
�����
Conclusions
• The concept of fuzzy natural logic was presented
• Formalization of the semantics of two classes of natural languageexpressions
• Human reasoning: perception-based logic deduction (PbLD),intermediate syllogisms, generalized square of opposition
• FNL includes a mathematical model of the vagueness phenomenon.
Model of the Sorites/Falakros paradox appearing in the semanticsof evaluative expressions
• Many open problems:
• How surface structures can be transformed into logical formulasrepresenting their meaning
• Formalization of negation• Formalization of hedging• Formalization of presupposition and its consequence• Formalization of the meaning of verbs
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 49 / 54
�����
Conclusions
• The concept of fuzzy natural logic was presented
• Formalization of the semantics of two classes of natural languageexpressions
• Human reasoning: perception-based logic deduction (PbLD),intermediate syllogisms, generalized square of opposition
• FNL includes a mathematical model of the vagueness phenomenon.
Model of the Sorites/Falakros paradox appearing in the semanticsof evaluative expressions
• Many open problems:
• How surface structures can be transformed into logical formulasrepresenting their meaning
• Formalization of negation• Formalization of hedging• Formalization of presupposition and its consequence• Formalization of the meaning of verbs
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 49 / 54
�����
Conclusions
• The concept of fuzzy natural logic was presented
• Formalization of the semantics of two classes of natural languageexpressions
• Human reasoning: perception-based logic deduction (PbLD),intermediate syllogisms, generalized square of opposition
• FNL includes a mathematical model of the vagueness phenomenon.
Model of the Sorites/Falakros paradox appearing in the semanticsof evaluative expressions
• Many open problems:
• How surface structures can be transformed into logical formulasrepresenting their meaning
• Formalization of negation• Formalization of hedging• Formalization of presupposition and its consequence• Formalization of the meaning of verbs
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 49 / 54
�����
Conclusions
• The concept of fuzzy natural logic was presented
• Formalization of the semantics of two classes of natural languageexpressions
• Human reasoning: perception-based logic deduction (PbLD),intermediate syllogisms, generalized square of opposition
• FNL includes a mathematical model of the vagueness phenomenon.
Model of the Sorites/Falakros paradox appearing in the semanticsof evaluative expressions
• Many open problems:
• How surface structures can be transformed into logical formulasrepresenting their meaning
• Formalization of negation• Formalization of hedging• Formalization of presupposition and its consequence• Formalization of the meaning of verbs
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 49 / 54
�����
Thank you for your attention
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 50 / 54
�����
Falakros/sorites paradox in evaluative expressions
Theorem (〈linguistic hedge〉 small)
• Zero is “(very) small” in each context` (∀w)((Smννν)w 0)
• In each context there is p which surely is not “(very) small”` (∀w)(∃p)(∆∆∆¬¬¬(Smννν)w p)
• In each context there is no n which is surely smalland n + 1 surely is not small
` (∀w)¬¬¬(∃n)(∆∆∆(Smννν)w n &&& ∆∆∆¬¬¬(Smννν)w(n + 1))
• In each context and for every n, if the latter is smallthen it is almost true that n + 1 is also small
` (∀w)(∀n)((Smννν)w n⇒⇒⇒ At((Smννν)w (n + 1))
At(A) – “it is almost true that A”, measured by the implication(Smννν)w n⇒⇒⇒ (Smννν)w (n + 1)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 51 / 54
�����
Falakros/sorites paradox in evaluative expressions
Theorem (〈linguistic hedge〉 small)
• Zero is “(very) small” in each context` (∀w)((Smννν)w 0)
• In each context there is p which surely is not “(very) small”` (∀w)(∃p)(∆∆∆¬¬¬(Smννν)w p)
• In each context there is no n which is surely smalland n + 1 surely is not small
` (∀w)¬¬¬(∃n)(∆∆∆(Smννν)w n &&& ∆∆∆¬¬¬(Smννν)w(n + 1))
• In each context and for every n, if the latter is smallthen it is almost true that n + 1 is also small
` (∀w)(∀n)((Smννν)w n⇒⇒⇒ At((Smννν)w (n + 1))
At(A) – “it is almost true that A”, measured by the implication(Smννν)w n⇒⇒⇒ (Smννν)w (n + 1)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 51 / 54
�����
Falakros/sorites paradox in evaluative expressions
Theorem (〈linguistic hedge〉 small)
• Zero is “(very) small” in each context` (∀w)((Smννν)w 0)
• In each context there is p which surely is not “(very) small”` (∀w)(∃p)(∆∆∆¬¬¬(Smννν)w p)
• In each context there is no n which is surely smalland n + 1 surely is not small
` (∀w)¬¬¬(∃n)(∆∆∆(Smννν)w n &&& ∆∆∆¬¬¬(Smννν)w(n + 1))
• In each context and for every n, if the latter is smallthen it is almost true that n + 1 is also small
` (∀w)(∀n)((Smννν)w n⇒⇒⇒ At((Smννν)w (n + 1))
At(A) – “it is almost true that A”, measured by the implication(Smννν)w n⇒⇒⇒ (Smννν)w (n + 1)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 51 / 54
�����
Falakros/sorites paradox in evaluative expressions
Theorem (〈linguistic hedge〉 small)
• Zero is “(very) small” in each context` (∀w)((Smννν)w 0)
• In each context there is p which surely is not “(very) small”` (∀w)(∃p)(∆∆∆¬¬¬(Smννν)w p)
• In each context there is no n which is surely smalland n + 1 surely is not small
` (∀w)¬¬¬(∃n)(∆∆∆(Smννν)w n &&& ∆∆∆¬¬¬(Smννν)w(n + 1))
• In each context and for every n, if the latter is smallthen it is almost true that n + 1 is also small
` (∀w)(∀n)((Smννν)w n⇒⇒⇒ At((Smννν)w (n + 1))
At(A) – “it is almost true that A”, measured by the implication(Smννν)w n⇒⇒⇒ (Smννν)w (n + 1)
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 51 / 54
�����
Falakros/sorites paradox in evaluative expressions
Theorem (〈linguistic hedge〉 big)
• In each context there is p which surely is not big` (∀w)(∃p)(∆∆∆¬¬¬(Bi ννν)w p)
• In each context there is no n which is surely bigand n − 1 surely is not big
` (∀w)¬¬¬(∃n)(∆∆∆(Smννν)w n &&& ∆∆∆¬¬¬(Smννν)w(n + 1))
• In each context and for every n, if the latter is bigthen it is almost true that n − 1 is also big
` (∀w)(∀n)((Bi ννν)w n⇒⇒⇒ At((Bi ννν)w (n − 1))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 52 / 54
�����
Falakros/sorites paradox in evaluative expressions
Theorem (〈linguistic hedge〉 big)
• In each context there is p which surely is not big` (∀w)(∃p)(∆∆∆¬¬¬(Bi ννν)w p)
• In each context there is no n which is surely bigand n − 1 surely is not big
` (∀w)¬¬¬(∃n)(∆∆∆(Smννν)w n &&& ∆∆∆¬¬¬(Smννν)w(n + 1))
• In each context and for every n, if the latter is bigthen it is almost true that n − 1 is also big
` (∀w)(∀n)((Bi ννν)w n⇒⇒⇒ At((Bi ννν)w (n − 1))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 52 / 54
�����
Falakros/sorites paradox in evaluative expressions
Theorem (〈linguistic hedge〉 big)
• In each context there is p which surely is not big` (∀w)(∃p)(∆∆∆¬¬¬(Bi ννν)w p)
• In each context there is no n which is surely bigand n − 1 surely is not big
` (∀w)¬¬¬(∃n)(∆∆∆(Smννν)w n &&& ∆∆∆¬¬¬(Smννν)w(n + 1))
• In each context and for every n, if the latter is bigthen it is almost true that n − 1 is also big
` (∀w)(∀n)((Bi ννν)w n⇒⇒⇒ At((Bi ννν)w (n − 1))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 52 / 54
�����
Vagueness of evaluative expressions
In each context there is no last surely small xand no first surely big x
Theorem
TEv ` (∀w)¬¬¬(∃x)(∀y)(∆∆∆(Smννν)wx &&&(x <w y ⇒⇒⇒∆∆∆¬¬¬(Smννν)wy)),
TEv ` (∀w)¬¬¬(∃x)(∀y)(∆∆∆(Bi ννν)wx &&&(y <w x ⇒⇒⇒∆∆∆¬¬¬(Bi ννν)wy))
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 53 / 54
�����
Holcapek, M., Monadic L-fuzzy quantifiers of the type 〈1n, 1〉, Fuzzy Sets andSystems 159(2008), 1811-1835.
Dvorak, A. and Holcapek, M., L-fuzzy Quantifiers of the Type 〈1〉 Determined byMeasures, Fuzzy Sets and Systems, 160(2009), 3425-3452
V. Novak, On fuzzy type theory, Fuzzy Sets and Systems 149 (2005), 235–273.
V. Novak, EQ-algebra-based fuzzy type theory and its extensions, Logic Journal ofthe IGPL 19(2011), 512-542
V. Novak, A comprehensive theory of trichotomous evaluative linguisticexpressions, Fuzzy Sets and Systems 159(2008), 2939-2969.
V. Novak, A formal theory of intermediate quantifiers, Fuzzy Sets and Systems159(2008) 1229–1246
P. Murinova, V. Novak, A formal theory of generalized intermediate syllogisms,Fuzzy Sets and Systems 186(2012), 47-80
V. Novak, The Alternative Mathematical Model of Linguistic Semantics andPragmatics, Plenum, New York, 1992.
P. L. Peterson, Intermediate Quantifiers. Logic, linguistics, and Aristoteliansemantics, Ashgate, Aldershot 2000.
Vilem Novak (IRAFM, CZ) Fuzzy Natural Logic FSTA 2016 54 / 54