Notes on Natural Logic

Similar documents
2 Deduction in Sentential Logic

0.1 Equivalence between Natural Deduction and Axiomatic Systems

Harvard School of Engineering and Applied Sciences CS 152: Programming Languages

Harvard School of Engineering and Applied Sciences CS 152: Programming Languages

TABLEAU-BASED DECISION PROCEDURES FOR HYBRID LOGIC

Semantics with Applications 2b. Structural Operational Semantics

1 Solutions to Tute09

5 Deduction in First-Order Logic

Gödel algebras free over finite distributive lattices

Harvard School of Engineering and Applied Sciences CS 152: Programming Languages


An effective perfect-set theorem

Lecture 2: The Simple Story of 2-SAT

Lecture l(x) 1. (1) x X

Structural Induction

TR : Knowledge-Based Rational Decisions and Nash Paths

Computing Unsatisfiable k-sat Instances with Few Occurrences per Variable

Recall: Data Flow Analysis. Data Flow Analysis Recall: Data Flow Equations. Forward Data Flow, Again

arxiv: v1 [math.lo] 24 Feb 2014

Introduction to Greedy Algorithms: Huffman Codes

Strongly compact Magidor forcing.

Another Variant of 3sat. 3sat. 3sat Is NP-Complete. The Proof (concluded)

Sy D. Friedman. August 28, 2001

Logic and Artificial Intelligence Lecture 24

Homework #4. CMSC351 - Spring 2013 PRINT Name : Due: Thu Apr 16 th at the start of class

Computing Unsatisfiable k-sat Instances with Few Occurrences per Variable

The Probabilistic Method - Probabilistic Techniques. Lecture 7: Martingales

In this lecture, we will use the semantics of our simple language of arithmetic expressions,

On the Optimality of a Family of Binary Trees Techical Report TR

SAT and DPLL. Introduction. Preliminaries. Normal forms DPLL. Complexity. Espen H. Lian. DPLL Implementation. Bibliography.

Logic and Artificial Intelligence Lecture 25

Lattices and the Knaster-Tarski Theorem

CS792 Notes Henkin Models, Soundness and Completeness

NOTES ON FIBONACCI TREES AND THEIR OPTIMALITY* YASUICHI HORIBE INTRODUCTION 1. FIBONACCI TREES

SAT and DPLL. Espen H. Lian. May 4, Ifi, UiO. Espen H. Lian (Ifi, UiO) SAT and DPLL May 4, / 59

TEST 1 SOLUTIONS MATH 1002

Strong normalisation and the typed lambda calculus

Outline Introduction Game Representations Reductions Solution Concepts. Game Theory. Enrico Franchi. May 19, 2010

Cut-free sequent calculi for algebras with adjoint modalities

Search Space and Average Proof Length of Resolution. H. Kleine Buning T. Lettmann. Universitat { GH { Paderborn. Postfach 16 21

Syllogistic Logics with Verbs

Finding Equilibria in Games of No Chance

Handout 4: Deterministic Systems and the Shortest Path Problem

The Real Numbers. Here we show one way to explicitly construct the real numbers R. First we need a definition.

Another Variant of 3sat

COMPUTER SCIENCE 20, SPRING 2014 Homework Problems Recursive Definitions, Structural Induction, States and Invariants

Essays on Some Combinatorial Optimization Problems with Interval Data

CIS 500 Software Foundations Fall October. CIS 500, 6 October 1

The Stackelberg Minimum Spanning Tree Game

R-automata. 1 Introduction. Parosh Aziz Abdulla, Pavel Krcal, and Wang Yi

THE NUMBER OF UNARY CLONES CONTAINING THE PERMUTATIONS ON AN INFINITE SET

BAYESIAN GAMES: GAMES OF INCOMPLETE INFORMATION

A relation on 132-avoiding permutation patterns

Lecture 5: Tuesday, January 27, Peterson s Algorithm satisfies the No Starvation property (Theorem 1)

TR : Knowledge-Based Rational Decisions

AVL Trees. The height of the left subtree can differ from the height of the right subtree by at most 1.

10.1 Elimination of strictly dominated strategies

CSE 21 Winter 2016 Homework 6 Due: Wednesday, May 11, 2016 at 11:59pm. Instructions

CS 4110 Programming Languages and Logics Lecture #2: Introduction to Semantics. 1 Arithmetic Expressions

Generalising the weak compactness of ω

Extender based forcings, fresh sets and Aronszajn trees

MSU CSE Spring 2011 Exam 2-ANSWERS

being saturated Lemma 0.2 Suppose V = L[E]. Every Woodin cardinal is Woodin with.

ExpTime Tableau Decision Procedures for Regular Grammar Logics with Converse

Lecture 14: Basic Fixpoint Theorems (cont.)

Levin Reduction and Parsimonious Reductions

monotone circuit value

CIS 540 Fall 2009 Homework 2 Solutions

SET 1C Binary Trees. 2. (i) Define the height of a binary tree or subtree and also define a height balanced (AVL) tree. (2)

UPWARD STABILITY TRANSFER FOR TAME ABSTRACT ELEMENTARY CLASSES

GUESSING MODELS IMPLY THE SINGULAR CARDINAL HYPOTHESIS arxiv: v1 [math.lo] 25 Mar 2019

Algorithmic Game Theory and Applications. Lecture 11: Games of Perfect Information

3 The Model Existence Theorem

Global Joint Distribution Factorizes into Local Marginal Distributions on Tree-Structured Graphs

IEOR E4004: Introduction to OR: Deterministic Models

An Adaptive Characterization of Signed Systems for Paraconsistent Reasoning

PARTITIONS OF 2 ω AND COMPLETELY ULTRAMETRIZABLE SPACES

Level by Level Inequivalence, Strong Compactness, and GCH

Tableau-based Decision Procedures for Hybrid Logic

Microeconomics of Banking: Lecture 5

The Binomial Theorem and Consequences

CATEGORICAL SKEW LATTICES

Forward Risk Adjusted Probability Measures and Fixed-income Derivatives

A Knowledge-Theoretic Approach to Distributed Problem Solving

Maximum Contiguous Subsequences

Continuous images of closed sets in generalized Baire spaces ESI Workshop: Forcing and Large Cardinals

Sequential allocation of indivisible goods

Arborescent Architecture for Decentralized Supervisory Control of Discrete Event Systems

0/1 knapsack problem knapsack problem

6 -AL- ONE MACHINE SEQUENCING TO MINIMIZE MEAN FLOW TIME WITH MINIMUM NUMBER TARDY. Hamilton Emmons \,«* Technical Memorandum No. 2.

Syllogistic Logics with Verbs

FORCING AND THE HALPERN-LÄUCHLI THEOREM. 1. Introduction This document is a continuation of [1]. It is intended to be part of a larger paper.

Ch 10 Trees. Introduction to Trees. Tree Representations. Binary Tree Nodes. Tree Traversals. Binary Search Trees

Kuhn s Theorem for Extensive Games with Unawareness

Initializing A Max Heap. Initializing A Max Heap

Yao s Minimax Principle

UNIT VI TREES. Marks - 14

Copyright 1973, by the author(s). All rights reserved.

A relative of the approachability ideal, diamond and non-saturation

Development Separation in Lambda-Calculus

Transcription:

Notes on Natural Logic Notes for PHIL370 Eric Pacuit November 16, 2012 1 Preliminaries: Trees A tree is a structure T = (T, E), where T is a nonempty set whose elements are called nodes and E is a relation on T, E T T, called the immediate edge relation, satisfying the following conditions: for all nodes, n, n, m T, Every node has a unique predecessor: If nem and n Em, then n = n, There are no cycles: If (n 1,, n k ) is a sequence of nodes where for each i = 1,, k 1, n i En i+1, then n 1 n k, and Between any two nodes there is a unique path: For each n, n T there is a unique sequence n 1, n 2,, n m such that n = n 1 En 2 En m = n Let n be a node, then succ(n) = {n nen } are the successors of n and pred(n) = {n n En} are the predecessors of n A node r is called the root of the tree provided pred(r) = A node n is called a leaf provided succ(n) = A path in a tree is a sequence of nodes connected by an edge relation: a path is a sequence (n 1, n 2,, n k ) such that for each i = 1,, k 1, n i En i+1 The length of a path is equal to the number of edges along that path (equivalently, one minus the number of nodes) The height of a tree is the length of the longest path Here is an example: n 8 n 6 n 5 n 4 n 7 n 3 n 2 n 1 The root of this tree is n 1 and the leaves are n 8, n 5, n 4, and n 7 We have succ(n 3 ) = {n 6, n 5, n 4 } and pred(n 3 ) = {n 1 } Two paths in the tree are (n 3, n 5 ) and (n 1, n 3, n 6 ) The height of the tree is 3 (the longest path is (n 1, n 3, n 6, n 8 ) which has length 3) Webpage: aistanfordedu/ epacuit, Email: epacuit@umdedu 1

2 The logic of All and Some Definition 21 (Sentences) Let V = {X, Y, Z, } be a set of variables language L AS are all the sentences of the form: The All and Some All X are Y Definition 22 (Models) A model for L AS is a pair M = W, [ ], where W is a non-empty set and [[ ] : V (W ) is a function assigning to each variable a subset of W Truth of sentences is defined as follows: 1 M = All X are Y iff [[X ] [Y ] 2 M = iff [[X ] [Y ] Example 23 Suppose M = W, [ ], where W = {a, b, c}, [X ] = {a}, [Y ] = {a, c}, [U ] = {c, b} and [[Z ] = Then, it is easy to verify that: 1 M = All X are Y and M = All Y are X 2 M =, M = Some U are Y, M = Some Y are U, and M = Some U are X 3 M = All Z are Y, M = All Z are X, and M = All Z are Z 4 M = Some Z are Y, M = Some Z are X, and M = Some Z are Z Let M be a model and Γ a set of sentences We write M = Γ provided M = S for each S Γ Definition 24 (Semantic Consequence) Let Γ be a set of sentences and S a sentence We write Γ = S (read S semantically follows from Γ ) provided for all models M if M = Γ then M = S A proof rule is a relation between a (possibly empty) set of sentences and a sentence We denote proof rules as follows: S 1 S 2 S n S Alternatively, we may write R({S 1,, S n }, S) (where R is the name of the rule) The intended interpretation is that from S 1, S 2,, and S n, one can infer S An axiom system is a set of proof rules Definition 25 (Proof Tree) Let A be an axiom system and Γ a set of sentences We write Γ A S provided there exists a tree T = (T, E) where the set of nodes T are sentences in L AS, the root of T is S, and and for each S T : 1 If S is a leaf node, then S Γ, or 2

2 there is a rule R in A such that R(succ(S ), S ) The tree T is called a proof tree We say T is a proof tree for (Γ, A) provided that it satisfies the above to properties for Γ and A The axiom system for the logic of All and Some, denoted A AS contains the following rules: All X are X All X are Y All Y are Z All X are Z Some Y are X Some X are X All X are Z Some Y are Z Theorem 26 (Soundness of A AS ) Let Γ be a set of sentences and S a setence Then, Γ AAS S implies Γ = S Proof We say that T is a proof tree for (Γ, A AS ) if T satisfies the two conditions of Definition 25 for Γ and A AS Given this notion, we want to prove a fact about all proof trees: For all proof tree T, for all sentences S if T is a proof tree for (Γ, A AS ) with root S, then Γ = S The proof is by induction on the height of trees Base Case: Suppose that T is a proof tree of height 0 Then T consists of exactly one node Since T is a proof tree for (Γ, A AS ), either S Γ or S is of the form All X are X If S Γ and M = Γ, then it is obvious that M = S If S is of the form All X are X, then any M makes S true (this follows from the fact that [X ] [X ] for any variable X) Hence, Γ = S Induction Step: The Induction hypothesis is: for any proof tree T for (Γ, A AS ) of height less than k with the sentence S as the root, Γ = S Suppose that T is a proof tree of height k with root S There are four cases Case 1 The proof tree T is of the form: T 1 T 2 All X are Y All Y are Z All X are Z where T 1 and T 2 are proof trees of height less than k By the induction hypothesis, Γ = All X are Y and Γ = All Y are Z Suppose that M = Γ Then M = All X are Y and M = All Y are Z We must show M = All X are Z Suppose that a [X ] Then, since [[X ] [Y ], a [Y ] and since [Y ] [Z ], we have a [Z ] Hence, M = All X are Z Case 2 The proof tree T is of the form: 3

T 1 T 2 All X are Z Some Y are Z where T 1 and T 2 are proof trees of height less than k By the induction hypothesis, Γ = All X are Z and Γ = Suppose that M = Γ Then M = All X are Z and M = We must show M = Some Y are Z Since [[X ] [Y ], there is a a [X ] [Y ] Then, since [X ] [Y ] [X ] [Z ], a [Z ] Since, we also have [[X ] [Y ] [Y ], we have a [Y ] [Z ] Hence, M = Some X are Z Case 3 The proof tree T is of the form: T 1 Some Y are X where T 1 is a proof tree of height less than k By the induction hypothesis, Γ = Suppose that M = Γ Then M = Hence, [Y ] [X ] = [[X ] [Y ] Therefore, MSome Y are X Case 4 The proof tree T is of the form: T 1 Some X are X where T 1 is a proof tree of height less than k By the induction hypothesis, Γ = Suppose that M = Γ Then M = Since [X ] [Y ] there is an a [X ] [Y ] [X ] So, a [X ] = [X ] [X ] Therefore, M = Some X are X qed Definition 27 Suppose that Γ is a set of sentences For U, V V, define U Γ V iff Γ All U are V When Γ is clear from the context, I write U V Lemma 28 For each Γ, the relation Γ is reflexive and transitive Proof Since All X are X is a proof tree, we have Γ All X are X for each variable X Hence, for all X V, X X For transitivity, suppose that Γ All X are Y and Γ All Y are Z Then, the following is a proof tree: 4

T 1 T 2 All X are Y All Y are Z All X are Z Therefore, Γ All X are Z qed If X is a variable, let [X] = {Y Y X} be the downset of X Theorem 29 Γ = S implies Γ S Proof Let Γ be a set of sentences and suppose that Γ = S The proof splits into two cases depending on the form of S The first case is when S is of the form Construct a model M S = W S, [ ] S as follows: W S = {Some U are V Some U are V Γ} [X ] S = {Some U are V U X or V X} We first show that M S = Γ Suppose that All X are Y Γ If [X ] S =, then clearly [X ] S [Y ] S Suppose that Some U are V [X ] S Then either U X or V X Since Γ All X are Y, we also have X Y Then, since is transitive, we have either U Y or V Y Hence, Some U are V [Y ] S Thus, M S = All X are Y Suppose that Γ Then, since X X, we have [X ] S and since Y Y, we have [Y ] S Hence, [X ] S [Y ] S Therefore, M S = Next, we show that if M S =, then Γ Suppose that M S = Then, we have Some U are V [X ] [Y ] for some Some U are V Γ Since Some U are V Γ, we have Γ Some U are V Since, Some U are V [X ] [Y ], we have U X or V X and U Y or V Y There are four cases Case 1 Suppose that U X and U Y, the Γ All U are X and Γ All U are Y Then, the following is a proof tree for : All U are X Some U are V Some U are U All U are Y Some U are X Some X are U 5

Case 2 Suppose that U X and V Y Then Γ All U are X and Γ All V are Y Then, the following is a proof tree for : All U are X Some U are V Some V are U All V are Y Some V are X Some X are V Case 3 Suppose that V X and U Y Then, Γ All V are X and Γ All U are Y Then, the following is a proof tree for : All V are X Some U are V All U are Y Some U are X Some X are U Case 4 Suppose that V X and V Y Then Γ All U are X and Γ All V are Y Then, the following is a proof tree for : Some U are V All V are X Some V are U Some V are V All V are Y Some V are X Some X are V 6

In all cases, we have Γ, as desired The proof of the statement follows from these two observations: Suppose that Γ = Then, since M S = Γ, we must have M S = Hence, Γ The second case is when S is of the form All X are Y Construct a model M A = W A, [ ] A as follows: W A = V { } where V { [X ] A [X] { } if Γ or Some Y are X Γ = [X] otherwise We first show that M A = Γ Suppose that All X are Y Γ Then X Y Suppose that V [X ] A Then, either V is a variable such that V X or V is In the first case, we have V X and X Y Since, is transitive, we have V Y, and so V [Y ] A In the second case, we have either Γ or Some Y are X Γ In either case, we also have [Y ] A Hence, [X ] A [Y ] A Suppose that Γ Then, we have [X ] A and [Y ] A Therefore, [X ] A [Y ] A, so M A = We now show that if M A = All X are Y, then Γ All X are Y Suppose that M A = All X are Y Then, [X ] A [Y ] A Since X [X ] A (this follows from the fact that X X for all variables), we have X [Y ] A Hence, X Y But this means, Γ All X are Y, as desired The proof of the statement follows from these two observations: Suppose that Γ = All X are Y Then, since M A = Γ, we must have M A = All X are Y Hence, Γ All X are Y qed 21 Completeness for a subclass of models In this section, we show how to use the above completeness theorem to prove completeness for a subclass of models Let M = {M for all X V, [[X ] } be the class of models that assign nonempty subsets to variables We write Γ = M S provided for each M M, if M = Γ then M = S Obviously, the above set of rules A AS are sound for this class of models The question is: Are they complete? It is not hard to see that the answer is no If we restrict to the models in M, then the following rule becomes valid: All X are Y Let A be the set of rules in the axiom system A AS together with the above rule We now show that this axiom system is complete with respect to M Theorem 210 Γ = M S implies Γ A S Proof Let Γ = Γ { Γ AAS All X are Y } We first prove two claims Claim 211 If Γ = M S, then Γ = S 7

Suppose that Γ = M S and M = Γ First, note that for each each X V, we have Some X are X Γ (this follows from the fact that for each X V, Γ All X are X) Since M = Γ, for each X V, M = Some X are X Hence [X ] [X ] = [X ] Therefore, M M Furthermore, since Γ Γ, we have M = Γ Since Γ = M S This implies, M = S Hence, Γ = S This completes the proof of the first claim Claim 212 If Γ AAS S, then Γ A S Suppose that Γ AAS S The proof is by induction on the height of proof trees Base Case: Suppose that T is a proof tree of height 0 This means that either S is All X are X or S Γ Suppose that S is All X are X Since All X are X is a proof rule in A, we have Γ A S If S Γ, then we are also done since this immediately implies Γ = S Suppose that S is of the from with Γ AAS S Since all the A contains all the rules of A AS, the following is a proof tree in A : T All X are Y where T is the proof tree for All X are Y using the rules from A AS Hence Γ A S Induction Step: Since A contains all the rules of A AS, all extensions of proof trees for the axiom system A using rules from A AS are still proof trees in the axiom system A This completes the proof of the claim Putting everything together: Suppose that Γ = M S Then, Γ = S by Claim 211 By the completeness theorem for A AS, we have Γ AAS S By Claim 212, Γ A S, as desired qed 8