1. Introduction 1. Introduction 1. Introduction : What is the problem ? 2. Framework : Applicative formalism and Combinatory Logic with types 3. Four Theories of Quantification I/ Fregean Theories (A) Classical presentation : with bound variables (B) with λ-calculus (Church, Montague) (C) Illative quantifiers (Curry) without bound variables II/ A non Fregean theory : (D) Star Theory without bound variables III/ Reduction of Illative Theory to Star Theory IV/ Logical cube
75
Embed
1. Introduction · 1. Introduction 1. Introduction : What is the problem ? 2. Framework : Applicative formalism and Combinatory Logic with types 3. Four Theories of Quantification
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1. Introduction1. Introduction
1. Introduction : What is the problem ?
2. Framework : Applicative formalism and Combinatory Logic with types
3. Four Theories of Quantification
I/ Fregean Theories
(A) Classical presentation : with bound variables
(B) with λ-calculus (Church, Montague)
(C) Illative quantifiers (Curry) without bound variables
II/ A non Fregean theory :
(D) Star Theory without bound variables
III/ Reduction of Illative Theory to Star Theory
IV/ Logical cube
PROBLEM :PROBLEM :
• Classical Logic (or fregean logic) is not fully adequate toa deep logical analysis of natural languages.
• In classical logic, there are different operations for buildingpropositions :
- connection between propositions ;- negation, modalities ;- predication (or application of a predicate onto terms) ;- quantification (or application of a quantifier ontopredicate to build a proposition or another predictate).
• In natural languages, these operations are used for buildingstatements but there are other fundamental operations :
• A determination is used to bring information about anexpression (terms, predicates); different synytacticalcategories (adjectives, relatives, articles, adverbs …)express a determination operation.
• In classical logic, after Frege, determination is not takeninto account :
adjectives, common names, intransitive verbs are analysedby a predicative operation.
• With Anca Pascu, we are studying a new formalism to takeinto account the logical operations of determination innatural languages and their different consequences.
• This formalism is called Logic of Determination of Objects(LDO) (presented in other meetings).
Logic of Determination of Objects(LDO)
• We start with the fundamental distinction of G. Frege(1893) : concepts (predicates) are operators thatexpress an « unsaturation » (or « no completion ») andobjects express « saturation » (or completedexpression).
• In LDO, we study different kinds of objects :
- fully determinate objects ;- more and less determinate objects ;- typical objects ;- atypical objects ;- determination of more determinated objects from an
underterminate object …
• A concept is an operator such that it can be applied todifferent objects. If the object « falls under the concept », theresult of the application is « true ».
• To each concept ‘f’ is associated its « intension » and its« essence ».
• The « intension » ‘Int(f)’ is an ordered set of properties whosethe meanings are contained in the meaning of the concept ‘f’.
• The « essence », ‘Ess (f)’, is a part of the « intension », ‘Int(f)’.
• To each concept ‘f’ is canonically associated an uniqueabstract object ‘τ(f)’ and a determination operator ‘δ(f)’.
• The object ‘τ(f)’ represents, as an object, the concept ‘f’; it iscompletly undeterminate.
• The operator ‘δ(f)’, associated to the concept ‘f’, contributes todetermine different objects.
• The object ‘τ(f)’ is called « the typical object »; it representsthe concept ‘f’ as an completly undeterminate object.
• It is a generator of differnt more or less determinate objectswhich fall under the concept ‘f’.
• The object which falls under the concept (or, when the conceptis applied to it with the result « true ») is belongs to theexpanse of the concept ‘f’, noted ‘Exp(f)’.
• When this object is fully determinate, it belongs to theextension of f, noted ‘Ext(f)’, a part of the expanse of ‘f’.
• An object is a typical in stance of a concept ‘f’ when this object
1°) falls under the concept ‘f’
2°) it inherits all properties of the intension of ‘f’.
• An object is an atypical instance of ‘f’ when
1°) it falls under the concept ‘f’
2°) there is some property of the intension of ‘f’ such that this object does not inherit.
• All objects that fall under ‘f’
inherit all properties of the essence of ‘f’.
EXTENSION = class of fully determinate men
Typical menAn atypical man
Abraham LincolnSocrates
ConfuciusLao Tseu
General properties : « to have two eyes, two legs, two arms … » in Intension of « man »
Specific Property :« to have onlyone leg »
Cyclopewith only one eye
Different expressions of LDO are defined insidethe general framework of Combinatory Logicof Curry with functional types.
2. Combinatory Logic with types2. Combinatory Logic with types
Applicative ExpressionsApplicative Expressions • We use applicative expressions and functional types (or categories of functors) from given basic types (or basic categories). • The application is a fundamental operation; it builds all applicative expressions from more elementary expressions. • The applicative expression ‘XY’ is the result of the application ‘@’ of the operator X to the operand Y. The application is represented by means of the following applicative tree :
X Y----------------> X applies to Y and builds a result XY = X@Y
• There are different types of applicative expressions
The Curry’s Combinatory Logic is an applicative formalism with Combinators. It is a logic of operators and composition of operators.
A combinator is an abstract operator which builds a “complex operator” from more elementary operators or transform an operator in another. EXAMPLE: The combinator BB builds a complex operator ‘BBXY’ from the given applicative expressions X and Y. When this complex operator is applied to the operand ‘Z’, then the result is the applicative expression ‘X(YZ)’ :
BB X------------------> BBX Y --------------------------->
3. Four Theories of Quantification3. Four Theories of Quantification
I/ Fregean Theories with bound variables
• Theory A : Classical Theory in First Order Language
• Theory B : Montague’s Quantification expressedin Church’s λ-Calculus
II/ Fregean Theory without bound variables
• Theory C : Illative Theory expressedin Curry’s Combinatory Logic
III/ Non Fregean Theory without bound variables
• Theory D : Star Theory expressedin Curry’s Combinatory Logic
Fregean Approach Non Fregean Approach
With bound variables
Without bound variables
Theory A : classicalFirst Order Languages (LPO)
Theory B using Church’s λ-calculus(Montague, Dowty …)
Theory C using Curry’s combinatory LogicIllative Theory of Quantification
Theory D using Curry’s Combinatory Logic: Star Theory of QuantificationFor processing of natural languages
Theories of QuantificationTheories of Quantification
Questions about Fregean QuantificationQuestions about Fregean Quantification
(i) Are there variables in natural languages ?
(ii) Are necessary variables for analysing phrases withquantifiers ?Can we conceive another logical analysis of naturallanguage,different of Frege’s and Montague’s analyses ?
(iii) Is there a formalism which can adequatly formalizethe quantifications expressed by natural languages ?
(iv) Is it possible to build logical representations“without (bound) variables” such that the structures ofthese representations are more adequate to thestructures (and meanings) of these sentences ?
I/ Fregean theoriesI/ Fregean theories
Theories (A), (B) and (C)Theories (A), (B) and (C)
Predicates and QuantifiersPredicates and Quantifiers
Predicates• [Otp : P1] means “ P1 is a one-place predicate”• [OtOtp : P2] means “ P2 is a two-place predicate”
Fregean Quantifiers• [OOtpp : Q ] means “Q is a fregean quantifier” : it builds a proposition from an one-place predicate.• [OOtpOOtpp : Q ] means “Q is a fregean restricted quantifier” : it builds a proposition from two one-place predicates. No Fregean quantifiers• [OOtpt : Q* ] means “Q* is an operator (star quantifier) which builds a term from a one-place predicate” :it builds a term from a predicate.
John is happy
John is happyN* S\N*---------> [T]S/(S\N*)----------------------------------------------->
S
0. John is happy----------------------------------------------
1. [N* : John] x [S/N* : is happy]2. [S/(S\N*): C*John] x [S/N* :is-happy]3. [S : (C*John) (is-happy)]
Some girl is pretty(5”) (λP.λQ.((∀x)[P(x) & Q(x)]) (girl’) (is-pretty’)(5’) (∃x) [girl’(x) & is-pretty’(x)]
PROBLEM : Are there bound variablesin natural languages ?
Linguists must explain how natural languages capture and encode by specificstructures the different quantification and determination operations.For analysing the constitution of sentences with quantification, the formalrepresentations given by the classical predicate calculus are not adequate sincenatural languages do not use bound variables in semiotic quantification strategy. For linguistics, bound variables are only useful artefacts. Is it possible to give logical representations “without (bound) variables” so that thestructures of these representations are more adequate to the syntactical structures ofsentences?
More precisely, the logical representations must explain how the analysed sentencesare build and also how they are related to a syntactic analysis.
• Illative operators “represent” classical quantifiers insideCurry’s Combinatory Logic formalism. • Illative operators are adjoined to the “pure” applicativeformalism and their actions are defined, by means ofelimination and introduction rules in Gentzen’s NaturalDeduction style, without using bound variables.
Every student admire some lecturer O. Every student admire some lecturerU*/N N (S\N)/N V*/N N---------------> ----------------------> U* V* 1. (every student) x admire x (some lecturer)
The two quantifiers Π1 and Π2 are not independent since it is possible to defineΠ2 , inside Combinatory Logic, from Π1 by the following relation betweenoperators:
[ Π2 =def ((B(CB2)Φ) => Π1) ] This relation shows that the restricted illative quantifier Π2 is defined by means ofa Combinator B(CB2)Φ that combines the implication operator => with thequantifier Π1.
1. Π2fg hyp.
2. [ Π2 =def (B(CB2)Φ) => Π1 ] def. de Π2
3. ((B(CB2)Φ) => Π1) fg rempl. 2., 1.
4. (CB2)(Φ =>) Π1 fg [e-B]
5. B2 Π1 (Φ =>) fg [e-C]6. Π1 (Φ => fg) [e-B2]
The elimination rule [e-Π2] is deduced from [e-Π1]:
1. Π1 (Φ => fg) hyp.2. fx hyp.3. (Φ => fg) x [e-Π1],1.4. =>(fx)(gx) [e-Φ],3.5. gx modus ponens
Illative Existential QuantifiersIllative Existential QuantifiersΣΣ11 and and ΣΣ22
• Types and applicative building •• OOtpp : Σ1 OOtpp : Σ1 Otp : g
Every student admire some lecturer O. Every student admire some lecturerU*/N N (S\N)/N V*/N N---------------> ----------------------> U* V* 1. (every student) x admire x (some lecturer)
Arguments against theories (A), (B) and (C) (fregean theories)
• There are not bound variables in natural languages (against A). • Proper names and quantified phrases Subjects are not represented by the sameprocess (against A). • Logical decomposition in Subject and Predicate does not work (against A). • Common nouns, adjectives and intransitive verbs are represented by the sameprocess (against A, B, C). • Determination operations are not expressed (against A,B,C). • It is not easy to represent topicalization operations (against A,B,C). • The fregean logical represntations do not take in account the problems of typicalityand atypicality (from Rösch approach of categorization) (against A,B,C).
II/ Star Theory of QuantificationII/ Star Theory of Quantification
a non Fregean Approacha non Fregean Approachof Quantifiersof Quantifiers
in Natural Languagesin Natural Languages
Star Theory of Quantification :non Fregean Quantifiers Q*
Quantifiers considered as building terms operators and not as building propositionsfrom predicates
The expressions with a freagean restricted universal quantifier :
(Π2 student’)((Σ 222 lecturer’) admire’) are logical representations of the sentence : Every student admires some lecturer
=> Fregean Quantifiers build propositions or predicates from predicates.
We are going to define two new operators Π∗Π∗ and Σ∗Σ∗ (building terms) which aremore elementary than the quantifiers Π2 and Σ222 in the above representations:
[ Π2 =def ΒC*Π∗Π∗ ] [Σ222 = def ΒC*Σ∗Σ∗ ]
Applicative Tree with Star QuantifiersApplicative Tree with Star Quantifiers
Now, the building of the proposition is given by the applicative tree:
Logical RepresentationsLogical Representationswith Star Quantifierswith Star Quantifiers
• In the applicative expression :
admire’ (Σ∗Σ∗ lecturer’) (Π∗Π∗ student’) the quantifiers Σ∗Σ∗ and Π∗Π∗ are operators whose operands are considered asproperties with the assigned type Otp.
It follows that the types of Σ∗Σ∗ and Π∗Π∗ must be O(Otp)t .
• These new operators Σ∗Σ∗ and Π∗Π∗ build terms from properties.
We have the following β-reduction : (Π2 student’) ((Σ222 lecturer’) admire’)
Every student admires some lecturerby means of Categorial Applicative Grammar with Type Raising and N*
Every student admire some lecturer O. Every student admires some lecturerN*/N N (S\N*)/N* N*/N N ---------------> -----------------> N* N* 1. [N*:every student] x [(S\N*)/N*:admires]
x [N*:some lecturer]------------->[T] ----------------->[T] S\(S/N*) S\(S/N*) 2. [S/(S\N*):C*(every student)]x[(S\N*)/N*:admire]
x [S\(S/N*): C*(some lecturer)]------------------------->[B]
S/N* 3. [S/N*: B(C*(every student)) admire)] x [S\(S/N*): C*(some lecturer)
The classical existantial Quantifier Σ2 reduces tothe existantial Star Quantifier ΣΣ**
[ Σ2 = BC* ΣΣ** ] (law) Reduction of Σ2 to Σ*
Reduction : (Σ2f)g ->β g(ΣΣ**f)
1. (Σ2f)g hyp.
2. [ Σ2 = BC* ΣΣ** ] law
3. BC* ΣΣ** fg rempl.
4. C* (ΣΣ** f) g [B-e]
5. g( ΣΣ** f) [C*-e]
Ext(g)
Ext(f)
Referential
Extτ(f) ∩ Ext(g) ≠ ∅
Some (typical) f is gSome (typical) beaver builds dams(Σ2f)g Extτ(f) ∩ Ext(g) ≠ ∅
Extτ(f)
IF there is a f which is g THEN a fortiorithere is an f which is gExample : Some typical beaver builds dams=> There is a beaver which builds dams
Ext(g)
Ext(f)
Referential
Ext(f) ∩ Ext(g) ≠ ∅
There is a beaver which builds damsg (Σ*f) Ext(f) ∩ Ext(g) ≠ ∅
Extτ(f)
IF there is an f which is gDoes not imply necessarly that the indetermined instance which is f and is g f et gIs a typical instance Example : There is a Frenchman who does not groan does not imply that this Frenchman is a typical instance of Frenchmeen
IV/ Logical IV/ Logical « « CubeCube » » of ofFregean and non FregeanFregean and non Fregean
Quantifiers.Quantifiers.A new Interpretation ofA new Interpretation of
Fregean and non FregeanFregean and non FregeanQuantifiersQuantifiers
Meaning of star quantifiers
• The terms Π∗(f) and Σ∗(f) denote objects ofExp (f) (Expanse of ‘f’).
- Π∗(f) denotes an abstract « whatever, any » object; anytypical object of Ext(f) is determined from Π∗(f)
- Σ∗(f) denotes an abstract « undeterminate » object. Atypical object which exists in Ext(f) can be gerneratedby successive determinations from Σ∗(f).
All Frenchmen groan A (typical) Frenchman groans
There is a (typical)Frenchmanwho groans
Some Frenchman groans
There is a Frenchman
Typical and classical quantifiersTypical and classical quantifiers
(Π2) fg
(Σ2) fg
g(Π*f)
g(Σ*f)
All f are g A typical f is g
There is a typical fwhich is g
Some f is g
Σf
β-reduction
β-reduction
β-reduction
β-reduction
There is a f
AristotleAristotle’’s Squares Square
(Π2 f)g
(Σ2 f)g
(Π2 f) (Ng)
(Σ2 f) (Ng)
There is a f
contraries
contradictories
subcontraries
Square of the reduction of Square of the reduction of « « classicalclassical » » (fregean) (fregean)quantifiers toquantifiers to
star (non fregean) quantifiersstar (non fregean) quantifiers
(Π2f) g
(Σ2f) g
Σ1 f
g (Π∗Π∗f)
g (ΣΣ**f)
ΣΣ* * (f)
β-reduction
β-reduction
Square of Star QuantifiersSquare of Star Quantifiers(non Fregean Quantifiers)(non Fregean Quantifiers)
g (ΠΠ**ff)
g (ΣΣ**f)
(Ng) (ΠΠ**ff)
(Ng) (ΣΣ**f)
There isan f
contradiction
Square of star quantifiersSquare of star quantifiers
g (ΠΠ**f)
g (ΣΣ**f)
(Ng) (ΠΠ**f)
(Ng) (ΣΣ**f)
ΣΣ* (* (f)
(Π(Π22f) g
(Σ(Σ22f) g
(ΠΠ22 f) (Ng)
(ΣΣ2 2 f )(Ng)
ΣΣf
g(ΠΠ∗f)
g(ΣΣ*f)
(Ng)(ΠΠ*f)
(Ng)(ΣΣ*f)
ΣΣ*(f)
opposite
disjunction
All f are g No f is g
There isa f
A typical f is g
Some fis g
A typical f is not g
There is a fwhich is not g
There isa typical f
« « CubeCube » » of quantifiers of quantifiers
There is a typical f which is not g
Some typical f is g
All beavers build dams No beaver build dams
There is a beaver
A typical beaverbuilds dams
Some beaverbuilds dams
A typical beaverdoes not build dams
There is a beaverwhich does not build dams
There is a typical beaver
« Cube » of quantifiers : examples
There is a typical beaver which does not build damsThere is a typical beaver
which builds dams
All f are g
There is a typical f which is g
No f is g
There i a typical f which is not g
There is a f
A typical f is g
Some f is g
A f typique n’est pas g
There is a f which is not g
There is A typical f
Fregean quantificationFregean quantification
No fregeanNo fregeanquantification =quantification =star quantification star quantification
All f are g No f is g
There is f
A typical f is g
Some fis g
No typical f is g
There is a f which is not g
There is a f
Typical quantification
Aristotle’s square
Square of star quantifiers
There is a typical fwhich is g
There is a typical f which is not g
• In classical Logic
Exp(f) = Ext(f)Ext(f) ⊆ Ext(g) Int(f) ⊇ Int(g)
All objects are typical objects, no atypicalobjects
Star qunatifiers are classical quantifiers
BibliographyBibliography
BOURBAKI N., 1970, Eléments de mathématiques, théorie des ensembles, Hermann,Paris.CURRY H.B. , FEYS R. 1958 ,Combinatory Logic, Amsterdam : North Holland.DESCLES J.-P. , GUENTCHEVA Z, 1980, “Recherches sur les valeurs des indéfinisNJAKOJ et NJAKAKAV en bulgare, Opérations de détermination, Théorie etdescription, Volume 1, Laboratoire de linguistique formelle, Université de Paris 7,pp. 239-286.DESCLES, J.-P., 1981, “De la notion d’opération à celle d’opérateur ou à larecherche de formalismes intrinsèques”, Mathématiques et sciences humaines, Paris,1981, pp. 5-32.DESCLES J.-P., FROIDEVAUX C., 1982, “Axiomatisation de la notion de repérageabstrait, Mathématiques et sciences humaines, 78, pp. 73-119.DESCLES J.-P., 1986, “L’implication entre concepts : la notion de typicalité” ,Travaux de Linguistique et de Littérature, XXIV, 1, pp. 179-102.DESCLES J.-P., 1987, “Réseaux sémantiques : la nature logique et linguistique desrelateurs” , Langages, 87. DESCLES J.-P., 1988, “Approximation et typicalité”, L’à-peu-près, Aspects anciens etmodernes de l’approximation , Editions de l’Ecole des Hautes Etudes en SciencesSociales, Paris, pp. 183-195.DESCLES J.-P., 1990, Langages applicatifs , langues naturelles et cognition, Paris :Hermès.
DESCLES J.-P., 1993, “Dialogue sur la typicalité”, Modèles et concepts pour la sciencecognitive, hommage à Jean-François Le Ny, (ed. M. Denis, G. Sabah), Presses del’Université de Grenoble, pp. 139-163.DESCLES J.-P., 1993,”Auto-applicativité, paradoxes et logique combinatoire”, LesCahiers du CREA, Epistémologie et anthropologie, CREA/ Ecole Polytechnique, Paris,pp. 63-119. DESCLES J.-P., 1996," Cognition, compilation, langage"in (ed. G. Chazal, M.-N.Terrasse) , Philosophie du langage et informatique, Paris, Hermès, 1996, pp. 103-145.DESCLES J.-P.,1997, “Schèmes, notions, prédicats et termes”, Logique, discours etpensée, Mélanges offerts à Jean-Blaize Grize, Peter Lang, pp. 9-36.DESCLES J.-P., 1997, “La logique combinatoire, types, preuves et langage naturel”,Introduction aux logiques non classiques, (ed. D. Mieville), Centre de Recherchessémiologiques, Neuchâtel, pp. 91-160.DESCLES, J.-P., 2002, « Categorization : A Logical Approach of a CognitiveProblem”, Journal of Cognitive Science, Vol. 3, n° 2, 2002, pp. 85-137.DESCLES, Jean-Pierre, 2003, « Combinatory Logic, Language, and CognitiveRepresentations », in Paul Weingartner (editor) Alternative Logics. Do Sciences NeedThem ?, Springer, pp. 115-148. DESCLES J.-P.,GUENTCHEVA Z., 1997, “Quantification Without Bound variables”,International Meeting on Free Variables Semantics, Université d ‘Osnabrück,septembre 1997.
DESCLES J.-P., GUENTCHEVA Z., 1997, “Quel est l’opérande du quantificateur?”, Colloque “Analyse formelle et contrastive des langues naturelles, Université deVarsovie, mai 1997, Semantyka a konfrontacja jezykowa, II, Warszawa, 1998 DESCLES J.-P., GUENTCHEVA Z., 1998, “A Non Fregean Approach ofQuantifiers in Natural Languages”, International Meeting on Logics , Languageand Artificial Intelligence, Fortaleza, 9-13 juillet 1998.DESCLES J.-P., KANELLOS I., 1991, ““La notion de typicalité : une approcheformelle” in D. Dubois, pp. 225-244.DESCLES J.-P., PASCU A., 1998, “Quantification non frégéenne et théorie de latypicalité”, Institut des Sciences Humaines Appliquées, Université de Paris-Sorbonne, juin 1998.DUBOIS D. , 1991, Sémantique et cognition : catégories, prototypes, typicalité,Editions du CNRS, Paris.FREGE, G., 1884, Die Grundlagen des Arithmetik, Verlag von Wilhem Koebner,Breslau.FREGE, G., 1967 1893, The Basic Laws of Arithmetic, exposition of the system,translated and edited, with an introduction, by Montgomery Furth, University ofCalifornia Press, Berkeley and Los Angeles.FROIDEVAUX C., 1983, “ La fonction logique du epsilon de Hilbert à travers les‘Grundlagen der Mathematik”, Mathématiques et sciences humaines, 84, pp. 65-82.HILBERT D., BERNAYS P., 1939/1970, Grundlagen der Mathematik, II? Springer-Verlag, Berlin.
HINDLEY, J.R., LERCHER B., SELDIN, 1972, J.P. Introduction to CombinatoryLogic, Cambridge University Press.LE NY, J.-F., 1989, Science cognitive et compréhension du langage, PressesUniversitaires de France.PASCU, A., 1999, Approche logique de la typicalité, Dissertation, University of Paris-SorbonnePASCU, A., 2006, Les objets dans la représentation des connaissances. Applicationaux processus de catégorisation en informatique et sciences humaines, Habilitation àdiriger des recherches, Paris-Sorbonne.