Table Of Content

Lecture Notes in Artificial Intelligence 1744 SubseriesofLectureNotesinComputerScience EditedbyJ.G.CarbonellandJ.Siekmann Lecture Notes in Computer Science EditedbyG.Goos,J.HartmanisandJ.vanLeeuwen 3 Berlin Heidelberg NewYork Barcelona HongKong London Milan Paris Singapore Tokyo Steffen Staab Grading Knowledge Extracting Degree Information from Texts 1 3 SeriesEditors JaimeG.Carbonell,CarnegieMellonUniversity,Pittsburgh,PA,USA Jo¨rgSiekmann,UniversityofSaarland,Saarbru¨cken,Germany Author SteffenStaab UniversitätKarlsruhe(TH),InstituteforAppliedInformatics andFormalDescriptionMethods(AIFB) 76128Karlsruhe,Germany E-mail:[email protected] Cataloging-in-Publicationdataappliedfor DieDeutscheBibliothek-CIP-Einheitsaufnahme Staab,Steffen: Gradingknowledge:extractingdegreeinformationfromtexts/Steffen Staab.-Berlin;Heidelberg;NewYork;Barcelona;HongKong; London;Milan;Paris;Singapore;Tokyo:Springer,1999 (Lecturenotesincomputerscience;1744:Lecturenotesin artificialintelligence) ISBN3-540-66934-5 CRSubjectClassification(1998):I.2.7,H.3,I.2,F.4 ISSN0302-9743 ISBN3-540-66934-5Springer-VerlagBerlinHeidelbergNewYork Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthematerialis concerned,specificallytherightsoftranslation,reprinting,re-useofillustrations,recitation,broadcasting, reproductiononmicrofilmsorinanyotherway,andstorageindatabanks.Duplicationofthispublication orpartsthereofispermittedonlyundertheprovisionsoftheGermanCopyrightLawofSeptember9,1965, initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer-Verlag.Violationsare liableforprosecutionundertheGermanCopyrightLaw. ©Springer-VerlagBerlinHeidelberg1999 PrintedinGermany Typesetting:Camera-readybyauthor SPIN:10749999 06/3142–543210 Printedonacid-freepaper Foreword If you are sitting in a basement room without a view — not to mention the bars in front of the windows — and writing a book, then you better have good company. I had the best company you could imagine. Waltraud Hiltl, KatjaMarkert,MartinRomacker,KlemensSchnattinger,AndreasKleeandI sharedverylittleofficespace,butplentyofchocolate,coffee,champagne,and enthusiasm for our research. North German coolness and creativity sprang mostly from my colleagues in the second floor. I learned a lot from and laughed a lot with Nobi Bro¨ker, Susanne (Sue) Schacht, Manfred Klenner, Peter Neuhaus, Stefan Schulz, and Michael Strube. I thank my friend and partner Angela Ro¨sch for motivational and tech- nical support and for living together with someone who cares about strange things, works too much and does not improve in any way over the years. Special thanks go to my family who sometimes wondered what was going on whenIstartedtalkingenthusiasticallyabout“semantics”,buttheyneverlet wane their encouragement for me. KornelMarcoprovidedgreatservicebyimplementingpartsofthesystem presented in this book. Joe Bush helped me polish up the text with his capabilitiesasanAmericannativespeaker.Remainingerrorsareentirelymy fault and due to my lack of diligence. This book would not have seen the light of day without the dissertation grantthroughtheGraduiertenkolleg“Menschliche&MaschinelleIntelligenz” funded by the Deutsche Forschungsgemeinschaft (DFG). Mostly, however, I must thank my advisor Udo Hahn. My perspective on research in Artificial Intelligence, Computational Linguistics, and Cognitive Science grew under his auspices. He fostered the work reported in this book in so many ways that I cannot name them all. Steffen Staab Karlsruhe, Germany, October 1999 (cid:0)(cid:2)(cid:3)(cid:4)(cid:5)(cid:6)(cid:7)(cid:2)(cid:8)(cid:9) (cid:10)(cid:11)(cid:12)(cid:13)(cid:3) (cid:11)(cid:14) (cid:15)(cid:16)(cid:17)(cid:5)(cid:18)(cid:19)(cid:6)(cid:19)(cid:2)(cid:11)(cid:8) (cid:20)(cid:21) (cid:6) (cid:18)(cid:11)(cid:22)(cid:12)(cid:19)(cid:5)(cid:3)(cid:21) (cid:11)(cid:14) (cid:23)(cid:8)(cid:9)(cid:5)(cid:4)(cid:6) (cid:24)(cid:11)(cid:25)(cid:3)(cid:18)(cid:26) Abstract “Text Knowledge Extraction” maps natural language texts onto a formal representation of the facts contained in the texts. Common text knowledge extraction methods show a severe lack of methods for understanding natural language “degree expressions”, like “expensive hard disk drive” and “good monitor”, which describe gradable properties like price and quality, respec- tively. However, without an adequate understanding of such degree expressions it is often impossible to grasp the central meaning of a text. Thisbookshowsconciseandcomprehensiveconceptsforextractingdegree information from natural language texts. It researches this task with regard to the three levels of (i) analysing natural language degree expressions, (ii) representingtheminaterminologicframework,and(iii) inferencingonthem by constraint propagation. On each of these three levels, the author shows that former approaches to the degree understanding problem were too sim- plistic,sincetheyignoredbyandlargetheroleofthebackgroundknowledge involved. Thus, he gives a constructive verification of his central hypothe- sis, viz. that the proper extraction of grading knowledge relies heavily on background grading knowledge. Thisconstructionproceedsasfollows.First,theauthorgivesanoverview oftheParseTalkinformationextractionsystem.Then,fromthereviewofrele- vantlinguisticliterature,theauthorderivestwodistinctcategoriesofnatural language degree expressions and proposes knowledge-intensive algorithms to handle their analyses in the ParseTalk system. These methods are applied to two text domains, viz. a medical diagnosis domain and a repository of texts from information technology magazines. Moreover, for inferencing the authorgeneralizesfromwell-knownconstraintpropagationmechanisms.This generalization is especially apt for representing and reasoning with natural language degree expressions, but it is also interesting from the point of view where it originated, viz. the field of temporal reasoning. The conclusion of the book gives an integration of all three levels of understanding showing that their coupling leads to an even more advanced — and more efficient — performance of the proposed mechanisms. Contents 1. Introduction.............................................. 1 1.1 Problems in Understanding Degree Expressions ............ 3 1.2 General Approach ...................................... 4 1.3 Overview.............................................. 5 2. ParseTalk — The System Context ........................ 7 2.1 An Architecture for Text Knowledge Extraction ............ 8 2.2 Syntactic Analysis...................................... 8 2.2.1 Dependency Grammar ............................ 9 2.2.2 The ParseTalk Parser ............................. 12 2.3 Conceptual System ..................................... 14 2.3.1 Description Logics................................ 15 2.3.2 Knowledge Base.................................. 20 2.3.3 Semantic System ................................. 23 2.4 Referring and Relating .................................. 25 2.4.1 Centering ....................................... 26 2.4.2 Relation Path Patterns and Metonymy.............. 27 2.4.3 An Example Text ................................ 28 3. Lexical Semantics of Degree Expressions.................. 31 3.1 Scales................................................. 31 3.1.1 Critique on Ontological Models for Degree Expressions 32 3.1.2 New Ontological Entities .......................... 33 3.2 Gradable Adjectives .................................... 36 3.2.1 Classification of Adjectives ........................ 36 3.2.2 Figurative Language .............................. 37 3.2.3 Multiple Word Senses............................. 40 3.2.4 Nominative vs. Normative Use ..................... 41 3.2.5 Two Types of Comparison......................... 42 3.3 Non-adjectival Degree Expressions........................ 44 3.4 Summary.............................................. 45 x Contents 4. Representation and Inferences ............................ 47 4.1 Requirements on Modeling Degree Relations ............... 47 4.1.1 Linguistic Stipulations ............................ 48 4.1.2 Stipulations from Vagueness ....................... 48 4.1.3 Stipulations on Inferences ......................... 51 4.1.4 The Challenge: Functions.......................... 52 4.2 Binary Relations ....................................... 53 4.2.1 Representation................................... 53 4.2.2 Inferencing ...................................... 57 4.2.3 Soundness and Incompleteness ..................... 59 4.2.4 Computational Complexity ........................ 61 4.3 Non-binary Relations ................................... 63 4.3.1 TCSPs and Allen’s Calculus ....................... 65 4.3.2 From Binary to Non-binary Relations ............... 66 4.3.3 A Formal Model of Generalized Temporal Networks (GTNs) ......................................... 68 4.3.4 Determining Consistency .......................... 72 4.3.5 Computing the Minimal Network................... 78 4.3.6 Scaling by Abstractions ........................... 81 4.3.7 Scaling by Generalizations......................... 84 4.4 Related Work .......................................... 88 4.4.1 Related Work on Representing and Inferencing with Degree Expressions ............................... 88 4.4.2 Related Work on Temporal and Spatial Reasoning.... 94 4.5 Conclusion on Representation and Inferences............... 96 5. Relative Comparisons..................................... 99 5.1 Basic Model for Interpreting Relative Comparatives......... 101 5.1.1 Comparative Interpretation as Semantic Copying..... 101 5.1.2 Core Algorithm .................................. 103 5.1.3 An Example of Semantic Interpretation ............. 105 5.2 Extension to Textual Phenomena......................... 106 5.2.1 Comparatives with Omitted Complements........... 108 5.2.2 An Example for Omitted Complements ............. 109 5.2.3 Metonymies in the Complement .................... 110 5.2.4 Metonymic Entities in the Omitted Complement ..... 111 5.2.5 An Example for Metonymic Entities in the Omitted Complement..................................... 112 5.3 Theoretical and Empirical Coverage ...................... 114 5.4 Related Work .......................................... 116 5.4.1 Generative Linguistics ............................ 116 5.4.2 Cognitive Foundations ............................ 121 5.4.3 Computational Approaches ........................ 122 5.5 Conclusion on Relative Comparisons ...................... 126