+ _ + No mortal man can slay every dragon No mortal Dutchman can slay every dragon No mortal man can slay every animal No mortal man can decapitate

Similar documents
Semantic Entailment and Natural Deduction

10. Presuppositions Introduction The Phenomenon Tests for presuppositions

What would count as Ibn Sīnā (11th century Persia) having first order logic?

1. Introduction Formal deductive logic Overview

Quantifiers: Their Semantic Type (Part 3) Heim and Kratzer Chapter 6

Identifying Anaphoric and Non- Anaphoric Noun Phrases to Improve Coreference Resolution

Anaphora Resolution in Biomedical Literature: A

Exercise Sets. KS Philosophical Logic: Modality, Conditionals Vagueness. Dirk Kindermann University of Graz July 2014

1 Clarion Logic Notes Chapter 4

Part II: How to Evaluate Deductive Arguments

Comments on Truth at A World for Modal Propositions

Outline of today s lecture

Validity & Soundness LECTURE 3! Critical Thinking. Summary: In this week s lectures, we will learn! (1) What it is for an argument to be valid.

Facts and Free Logic. R. M. Sainsbury

Facts and Free Logic R. M. Sainsbury

Ayer on the criterion of verifiability

INTERMEDIATE LOGIC Glossary of key terms

Natural Language Processing (NLP) 10/30/02 CS470/670 NLP (10/30/02) 1

Artificial Intelligence I

MISSOURI S FRAMEWORK FOR CURRICULAR DEVELOPMENT IN MATH TOPIC I: PROBLEM SOLVING

Entailment as Plural Modal Anaphora

Semantic Foundations for Deductive Methods

THE MEANING OF OUGHT. Ralph Wedgwood. What does the word ought mean? Strictly speaking, this is an empirical question, about the

Quantificational logic and empty names

Question Answering. CS486 / 686 University of Waterloo Lecture 23: April 1 st, CS486/686 Slides (c) 2014 P. Poupart 1

Verificationism. PHIL September 27, 2011

Module 5. Knowledge Representation and Logic (Propositional Logic) Version 2 CSE IIT, Kharagpur

Introduction Symbolic Logic

INTRODUCTION TO HYPOTHESIS TESTING. Unit 4A - Statistical Inference Part 1

An Introduction to. Formal Logic. Second edition. Peter Smith, February 27, 2019

Vagueness and supervaluations

Artificial Intelligence: Valid Arguments and Proof Systems. Prof. Deepak Khemani. Department of Computer Science and Engineering

Verification and Validation

1. Introduction. Against GMR: The Incredulous Stare (Lewis 1986: 133 5).

SOME RADICAL CONSEQUENCES OF GEACH'S LOGICAL THEORIES

Introduction to Philosophy

What is an Argument? Validity vs. Soundess of Arguments

McDougal Littell High School Math Program. correlated to. Oregon Mathematics Grade-Level Standards

Artificial Intelligence. Clause Form and The Resolution Rule. Prof. Deepak Khemani. Department of Computer Science and Engineering

TWO VERSIONS OF HUME S LAW

INTRODUCTION TO LOGIC 1 Sets, Relations, and Arguments

HS01: The Grammar of Anaphora: The Study of Anaphora and Ellipsis An Introduction. Winkler /Konietzko WS06/07

Artificial Intelligence Prof. Deepak Khemani Department of Computer Science and Engineering Indian Institute of Technology, Madras

Some observations on identity, sameness and comparison

Remarks on a Foundationalist Theory of Truth. Anil Gupta University of Pittsburgh

QCAA Study of Religion 2019 v1.1 General Senior Syllabus

Georgia Quality Core Curriculum

Artificial Intelligence Prof. P. Dasgupta Department of Computer Science & Engineering Indian Institute of Technology, Kharagpur

Chapter 3: Basic Propositional Logic. Based on Harry Gensler s book For CS2209A/B By Dr. Charles Ling;

9.1 Intro to Predicate Logic Practice with symbolizations. Today s Lecture 3/30/10

SEVENTH GRADE RELIGION

Announcements. CS243: Discrete Structures. First Order Logic, Rules of Inference. Review of Last Lecture. Translating English into First-Order Logic

Logic and Pragmatics: linear logic for inferential practice

Philosophy 220. Truth Functional Properties Expressed in terms of Consistency

Argumentation Module: Philosophy Lesson 7 What do we mean by argument? (Two meanings for the word.) A quarrel or a dispute, expressing a difference

Haberdashers Aske s Boys School

In this section you will learn three basic aspects of logic. When you are done, you will understand the following:

Lecture 6 Keynes s Concept of Probability

Portfolio Project. Phil 251A Logic Fall Due: Friday, December 7

Macmillan/McGraw-Hill SCIENCE: A CLOSER LOOK 2011, Grade 4 Correlated with Common Core State Standards, Grade 4

Elements of Science (cont.); Conditional Statements. Phil 12: Logic and Decision Making Fall 2010 UC San Diego 9/29/2010

KEEP THIS COPY FOR REPRODUCTION Pý:RPCS.15i )OCUMENTATION PAGE 0 ''.1-AC7..<Z C. in;2re PORT DATE JPOTTYPE AND DATES COVERID

Announcements. CS311H: Discrete Mathematics. First Order Logic, Rules of Inference. Satisfiability, Validity in FOL. Example.

From Machines To The First Person

Complications for Categorical Syllogisms. PHIL 121: Methods of Reasoning February 27, 2013 Instructor:Karin Howe Binghamton University

Phil 413: Problem set #1

Broad on Theological Arguments. I. The Ontological Argument

Can Negation be Defined in Terms of Incompatibility?

Reconsidering Raising and Experiencers in English

Who wrote the Letter to the Hebrews? Data mining for detection of text authorship

Truth and Molinism * Trenton Merricks. Molinism: The Contemporary Debate edited by Ken Perszyk. Oxford University Press, 2011.

Exposition of Symbolic Logic with Kalish-Montague derivations

Semantics and Pragmatics of NLP DRT: Constructing LFs and Presuppositions

TEXT MINING TECHNIQUES RORY DUTHIE

World History and Geography Correlated to Common Core State Standards for Literacy in History/Social Studies, Science, and Technical Subjects

Qualitative versus Quantitative Notions of Speaker and Hearer Belief: Implementation and Theoretical Extensions

Postulates for conditional belief revision

Circularity in ethotic structures

Now consider a verb - like is pretty. Does this also stand for something?

Definite Descriptions and the Argument from Inference

Macmillan/McGraw-Hill SCIENCE: A CLOSER LOOK 2011, Grade 1 Correlated with Common Core State Standards, Grade 1

Understanding Truth Scott Soames Précis Philosophy and Phenomenological Research Volume LXV, No. 2, 2002

Grade 6 Math Connects Suggested Course Outline for Schooling at Home

Resolving Direct and Indirect Anaphora for Japanese Definite Noun Phrases

2.1 Review. 2.2 Inference and justifications

A Problem for a Direct-Reference Theory of Belief Reports. Stephen Schiffer New York University

The Externalist and the Structuralist Responses To Skepticism. David Chalmers

Macmillan/McGraw-Hill SCIENCE: A CLOSER LOOK 2011, Grade 3 Correlated with Common Core State Standards, Grade 3

Lecture 3 Arguments Jim Pryor What is an Argument? Jim Pryor Vocabulary Describing Arguments

Language, Meaning, and Information: A Case Study on the Path from Philosophy to Science Scott Soames

Logical Omniscience in the Many Agent Case

College and Career Readiness Anchor Standards for Reading. Step Into the Time 36 Step Into the Place 92, 108, 174, 292, 430

Is the law of excluded middle a law of logic?

Moore on External Relations

Grade 6 correlated to Illinois Learning Standards for Mathematics

A Linguistic Interlude

Logic Appendix: More detailed instruction in deductive logic

AliQAn, Spanish QA System at multilingual

Logical (formal) fallacies

Appendix 1. Towers Watson Report. UMC Call to Action Vital Congregations Research Project Findings Report for Steering Team

Transcription:

+ _ + No mortal man can slay every dragon No mortal Dutchman can slay every dragon No mortal man can slay every animal No mortal man can decapitate every dragon

Extending the monotonicity calculus and embedding it in a textual inference environment based on McCartney 2009

Textual Inference

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO. IR Between March and June, scientific observers say, up to 300,000 seals are killed. In Canada, seal-hunting means jobs, but opponents say it is vicious and endangers the species, also threatened by global warming. Hunting endangers seal species. QA Aeschylus is often called the father of Greek tragedy; he wrote the earliest complete plays which survive from ancient Greece. He is known to have written more than 90 plays, though only seven survive. The most famous of these are the trilogy known as Orestia. Also well-known are The Persians and Prometheus Bound. "The Persians" was written by Aeschylus. SUM A Pentagon committee and the congressionally chartered Iraq Study Group have been preparing reports for Bush, and Iran has asked the presidents of Iraq and Syria to meet in Tehran. Bush will meet the presidents of Iraq and Syria in Tehran.

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO.

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO. YES

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO. YES IR Between March and June, scientific observers say, up to 300,000 seals are killed. In Canada, seal-hunting means jobs, but opponents say it is vicious and endangers the species, also threatened by global warming. Hunting endangers seal species.

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO. YES IR Between March and June, scientific observers say, up to 300,000 seals are killed. In Canada, seal-hunting means jobs, but opponents say it is vicious and endangers the species, also threatened by global warming. Hunting endangers seal species. YES

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO. YES IR Between March and June, scientific observers say, up to 300,000 seals are killed. In Canada, seal-hunting means jobs, but opponents say it is vicious and endangers the species, also threatened by global warming. Hunting endangers seal species. YES QA Aeschylus is often called the father of Greek tragedy; he wrote the earliest complete plays which survive from ancient Greece. He is known to have written more than 90 plays, though only seven survive. The most famous of these are the trilogy known as Orestia. Also well-known are The Persians and Prometheus Bound. "The Persians" was written by Aeschylus.

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO. YES IR Between March and June, scientific observers say, up to 300,000 seals are killed. In Canada, seal-hunting means jobs, but opponents say it is vicious and endangers the species, also threatened by global warming. Hunting endangers seal species. YES QA Aeschylus is often called the father of Greek tragedy; he wrote the earliest complete plays which survive from ancient Greece. He is known to have written more than 90 plays, though only seven survive. The most famous of these are the trilogy known as Orestia. Also well-known are The Persians and Prometheus Bound. "The Persians" was written by Aeschylus. YES

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO. YES IR Between March and June, scientific observers say, up to 300,000 seals are killed. In Canada, seal-hunting means jobs, but opponents say it is vicious and endangers the species, also threatened by global warming. Hunting endangers seal species. YES QA Aeschylus is often called the father of Greek tragedy; he wrote the earliest complete plays which survive from ancient Greece. He is known to have written more than 90 plays, though only seven survive. The most famous of these are the trilogy known as Orestia. Also well-known are The Persians and Prometheus Bound. "The Persians" was written by Aeschylus. YES SUM A Pentagon committee and the congressionally chartered Iraq Study Group have been preparing reports for Bush, and Iran has asked the presidents of Iraq and Syria to meet in Tehran. Bush will meet the presidents of Iraq and Syria in Tehran.

Examples IE At the same time the Italian digital rights group, Electronic Frontiers Italy, has asked the nation's government to investigate Sony over its use of anti-piracy software. Italy's government investigates Sony. NO IE Parviz Davudi was representing Iran at a meeting of the Shanghai Co-operation Organisation (SCO), the fledgling association that binds Russia, China and four former Soviet republics of central Asia together to fight terrorism. China is a member of SCO. YES IR Between March and June, scientific observers say, up to 300,000 seals are killed. In Canada, seal-hunting means jobs, but opponents say it is vicious and endangers the species, also threatened by global warming. Hunting endangers seal species. YES QA Aeschylus is often called the father of Greek tragedy; he wrote the earliest complete plays which survive from ancient Greece. He is known to have written more than 90 plays, though only seven survive. The most famous of these are the trilogy known as Orestia. Also well-known are The Persians and Prometheus Bound. "The Persians" was written by Aeschylus. YES SUM A Pentagon committee and the congressionally chartered Iraq Study Group have been preparing reports for Bush, and Iran has asked the presidents of Iraq and Syria to meet in Tehran. Bush will meet the presidents of Iraq and Syria in Tehran. NO

Approaches

Approaches robust,! but shallow" deep,! but brittle" lexical/! semantic! overlap" Jijkoun & de Rijke 2005" FOL &! theorem! proving" Bos & Markert 2006" Imprecise, confused by negation, quantification, etc. Brittle, requires translation of NL into FOL

robust,! but shallow" deep,! but brittle" lexical/! semantic! overlap" Jijkoun & de Rijke 2005" natural! logic" FOL &! theorem! proving" Bos & Markert 2006" Imprecise confused by negation, quantification, etc. Brittle, requires translation of NL into FOL

robust,! but shallow" lexical/! semantic! overlap" Imprecise confused by negation, Jijkoun & de Rijke 2005" quantification, etc. patterned! relation! extraction" natural! logic" semantic" graph" matching"?? FOL &! theorem! proving" deep,! but brittle" Bos & Markert 2006" Brittle, requires translation of NL into FOL

Shallow approaches Bag-of-words alignment and deletion: based on the independent probabilities that each word in the premise supports the hypothesis and that each word in the hypothesis derives its support from one word in the premise. For each word pair we calculate lexical similarity, e.g. string similarity, distributional similarity, WordNet-based distance metrics,... Some words are more important than others (different weights).

The first settlements on the site of Jakarta were established at the mouth of the Ciliwung, perhaps as early as the 5th century AD. The first settlements on the site of Jakarta were established as early as the 5th century AD. Several airlines polled saw costs grow more than expected even after adjusting for inflation. Some of the companies in the poll reported cost increases. Sharon denies Arafat could be targeted for assassination Arafat targeted for assassination

Graph-matching Represent sentence as a (dependency) graph Align p and h, score each possible alignment, take the best, evaluate strength. But: finding the best alignment in this way is NP-complete, so we need heuristics. Assumptions: once a match if found the rest of the graph doesn t affect the validity of the match No dogs barked loudly No dogs barked Sharon denies Arafat could be targeted for assassination Arafat targeted for assassination alignment and inference determination are done in one go 25 of the dead were members of he law enforcement agencies and the rest of the 67 were civilians 25 of the dead were civilians Sales fell/sales rose// Sales fell/sales did not rise.

More evolved RTE systems Linguistic analysis: tokenization, phrase structure parse and/or dependency parse,... Alignment Entailment determination

Linguistic analysis subj(rose,sales) rose nnmod(sales,mitsubishi) subj sales nnmod Mitsubishi obj percent num 46 obj(rose,percent) num(percent,46)

Aside: why dependency grammar Closer to semantic relationships, but also more underspecified: no distinction between elements that modify the head of a phrase and those that modify the whole phrase. Dependency trees contain one word per node, makes parsing more straightforward? But the Stanford parser (and the XLE parser) are not pure dependency parsers.

Linguistic analysis subj(rose,sales) nnmod(sales,mitsubishi) sales subj rose obj percent obj(rose,percent) num(percent,46) S nnmod Mitsubishi 46 num NP VP NP

Alignment:MANLI In most Pacific countries there are very few women in Parliament. Women are poorly represented in Parliament. Align via EQ, SUB, DEL, INS: DEL(In), DEL(most), DEL(Pacific), DEL (countries), DEL(there), EQ(are, are), SUB(very few, poorly represented), EQ(women, Women), EQ(in,in),EQ(parliament,parliament),EG(.,.)

MANLI Alignment Scoring: Edit type features:eq higher than SUB and DEL higher than INS (why?) Phrase features: size Semantic relatedness feature for SUB based on string similarity, synonymy and the like, distributional similarity Contextual features: difference in position, similarity of neighboring material

Insure best alignment for each phrase pair (simulated annealing technique). Assigning weights to the features

Determining inferences Represent the data as features Classification task Statistical learning algorithm Features: polarity: no, not, without, except adjunct: adverbs, prepositional phrases antonymy: derived from WordNet: fall, rise,... (scores high for alignment but gets a feature that insures that inference will be low) modality: possible, not possible, actual, not actual, nessary, not necessary factivity: try ~ manage quantifiers, numbers, dates, time,...

RTE provides a rather small development/ training set; lots of features leads to overfitting, too few features to errors. Results in general rather imprecise.

Using Natural Logic for Entailment

Entailment determination premise hypothesis (= thesis to be proven) Functional view: input an ordered pair (p,h), output a Boolean value, 1 if p entails h, 0 otherwise.

X is a couch X is a sofa X is a crow X is a bird X is a fish X is a carp X is a hippo X is hungry X is a cat X is a dog

Which notion of entailment? 1. entailment as a two way classification: output labels (entailment, nonentailment) are interpreted as denoting sets of ordered pairs (relations) of declarative expressions (T): entailment (def) {(p,h) DomTx2: p h} non-entailment (def) {(p,h) DomTx2: p h} X is a crow, X is a bird: yes X is a crow, X is a canary: no X is a crow, X is hungry: no From MacCartney 2009

2. entailment as a three-way classification: difference between contradiction and compatibility entailment (def) {(p,h) DomTx2: p h} contradiction (def) {(p,h) DomTx2: p h} compatibility (def) {(p,h) DomTx2: p h p h} X is a crow, X is a bird: yes X is a crow, X is a canary: no X is a crow, X is hungry: compatible

3. a. entailment as containment (monotonicity); output space (def) {(p,h) DomTx2: p h h p} (def) {(p,h) DomTx2: p h h p} (def) {(p,h) DomTx2: p h h p} no-containment (def) {(p,h) DomTx2: p h h p} X is a crow, X is a bird: X is a bird, X is a crow: X is a sofa, X is a coach: X is a crow, X is a canary: no containment X is a crow, X is hungry: no containment From MacCartney 2009

3. b. entailment as containment; input space: not just T but also E and mappings if x,y DomT then x y iff x = false or y = true if x,y DomE then x y iff x = y if x,y DomA B then x y iff for all a DomA x(a) y(a) (one function entails another if each of its outputs entails the corresponding output of the other function) otherwise x y and y x From MacCartney 2009

Entailment relations X is a couch X is a sofa X is a crow X is a bird X is a fish X is a carp X is a hippo X is hungry X is a cat X is a dog 2-way! RTE1,2,3" Yes! entailment" No! non-entailment" 3-way! FraCaS,! PARC, RTE4" Yes! entailment" Unknown! compatibility" No! contradiction" containment! Sánchez-Valencia" P = Q! equivalence" P < Q! forward! entailment" P > Q! reverse! entailment" P # Q! non-entailment" From MacCartney 2011

Garfield is a cat Garfield is a mammal Garfield is not a fish Garfield is not a carp Which of these entailments can the monotonicity calculus do?