Computational Linguistics and Intelligent Text Processing: 4th International Conference, CICLing 2003, Mexico City, Mexico, February 16-22, 2003. Proceedings (Lecture Notes in Computer Science, 2588)

CICLing 2003 (www.CICLing.org) was the 4th annual Conference on Intelligent Text Processing and Computational Linguistics. It was intended to provide a balanced view of the cutting-edge developments in both the theoretical foundations of computational linguistics and the practice of natural language text processing with its numerous applications. A feature of CICLing conferences is their wide scope that covers nearly all areas of computational linguistics and all aspects of natural language processing applications. The conference is a forum for dialogue between the specialists working in these two areas. This year we were honored by the presence of our keynote speakers Eric Brill (Microsoft Research, USA), Aravind Joshi (U. Pennsylvania, USA), Adam Kilgarriff (Brighton U., UK), and Ted Pedersen (U. Minnesota, USA), who delivered excellent extended lectures and organized vivid discussions. Of 92 submissions received, after careful reviewing 67 were selected for presentation; 43 as full papers and 24 as short papers, by 150 authors from 23 countries: Spain (23 authors), China (20), USA (16), Mexico (13), Japan (12), UK (11), Czech Republic (8), Korea and Sweden (7 each), Canada and Ireland (5 each), Hungary (4), Brazil (3), Belgium, Germany, Italy, Romania, Russia and Tunisia (2 each), Cuba, Denmark, Finland and France (1 each). Computational Linguistics and Intelligent Text Processing Preface Organization Table of Contents Starting with Complex Primitives Pays Off* Introduction Lexicalized Tree-Adjoining Grammar An Alternate Perspective on Adjoining Discourse Structure Summary Things Are Not Always Equal 1 Introduction 2 Subsumption and Linguistic Phenomena 3 Partial VP Fronting in German 4 Stylistic Inversion in French 5 Discussion References GIGs: Restricted Context-Sensitive Descriptive Power in Bounded Polynomial-Time Introduction Global Index Grammars Indexed Grammars and Linear Indexed Grammars Global Indexed Grammars GILs and Dyck Languages Recognition of GILs Graph-Structured Stacks GILs Recognition Using Earley Algorithm Conclusions Total Lexicalism and GASGrammars: A Direct Way to Semantics DRT, UCG, and Total Lexicalism Definition System of GASGrammars Implementation in Prolog Pseudo Context-Sensitive Models for Parsing Isolating Languages: Classical Chinese – A Case Study Introduction The Three Models Results Conclusion Imperatives as Obligatory and Permitted Actions 1 Introduction 2.1 Physical Action and Imperatives 2.2 Obligation and Permission 2.3 Obligation in a Context 2 Analysis 3 Model for Imperatives in Terms of Obligation and Permission 3.1 Syntax of LDL 3.1.1 Definition of Sets 3.1.2 Definition of Terms 3.1.3 Syntax of LDL 3.1.4 Category of Actions 3.2 Axioms 3.3 Inference Rules 3.4 Interpretation 3.5 Correctness of Actions 3.6 Encapsulation 3.7 Defining Sets of Obligatory Actions 4 Obligation/Permission and Inference 4.1 JorgensenTMs Dilemma 4.2 Paradox of Free Choice Permission-FCP 4.3 Conjunction Elimination and Obligations 5 Feasibility of Implementation and Applications of the Model 6 Conclusions Acknowledgements. The first author would like to thank CONACYT-IIE. References Formal Representation and Semantics of Modern Chinese Interrogative Sentences 1 Introduction 2 Representation and Interpretation of Chinese Semantics 3 Classification and Querying Focus of Interrogative Sentences 3.1 Classification 3.2 Querying Focus 4 Representation and Semantics of Interrogative Sentences 4.1 Proposition Set Approach and Structured Meaning Approach 4.2 Representation and Semantics of Chinese Interrogative Sentences 4.3 fiW+nefl Interrogative Sentence 5 Discussion 6 Conclusion References Analyzing V+Adj in Situation Semantics 1 Introduction 2 Difference between V-Statement and A-Statement 3 Formalizing VA-Statements in Situation Semantics 4 Analyzing VA-Statements 4.1 Formal Model of Semantic Pointer 4.2 Automatic Analysis 5 Conclusion References Diagnostics for Determining Compatibility in English Support-Verb-Nominalization Pairs 1 Background and Problem 1.1 Semantic-Role Constraints as a Predictive Model 2 Aktionsart Classes and SVN Combinations 2.1 First Sample and Results 2.2 Testing on Corpus Data 3 Summary and Conclusion References A Maximum Entropy Approach for Spoken Chinese Understanding 1 Introduction 2 Semantic Representation and Semantic Symbol 3 The Maximum Entropy Analysis Method 3.1 The Maximum Entropy Model 3.2 Features for ME Analysis Model 3.3 Parameter Estimation and Semantic Analysis 4 Experiments and Results Analysis 5 Conclusion Acknowledgements. The authors are grateful to Dr. Mark Seligman for his helpful work. The authors also would like to say a very big thank to the anonymous reviewers for their beneficial comments. References A Study to Improve the Efficiency of a Discourse Parsing System 1 Introduction 2 Identifying Elementary Discourse Units 3 Factors Used for Recognizing Relations 3.1 Text Cohesion as RelationTMs Predictors 3.2 Cue Phrases 4 Relation Set and Relation Recognition 4.1 Relation Recognition 4.2 Scoring Heuristic Rules 4.3 LIST Relation 5 Rhetorical Parser 5.1 Rules for the Rhetorical Parser 5.2 Algorithm for Rhetorical Parser 6 Conclusion References Conversion of Japanese Passive/Causative Sentences into Active Sentences Using Machine Learning Introduction Tagged Corpus as Supervised Data Machine Learning Method (Support Vector Machine) Method of Using the Results of Unsupervised Data as Features Features (Information Used in Classification) Experiments Conclusion From Czech Morphology through Partial Parsing to Disambiguation Introduction Morphological Analyser {tt ajka} Description of Czech Morphology Implementation of the Analyser Partial Parser {sc Dis} Verb Rules Parsing Mechanism Extension {sc VaDis} Partial Automatic Disambiguation Conclusion Fast Base NP Chunking with Decision Trees – Experiments on Different POS Tag Settings Introduction Chunking Probability Estimation Combination of Pattern Types Combination of Categories Test Environments Method Probability Estimation Combination of Pattern Types Combination of Categories Experiments Discussion Guaranteed Pre-tagging for the Brill Tagger Introduction The Brill Tagger Initial State Tagger Final State Tagger Standard Pre-tagging with the Brill Tagger Guaranteed Pre-tagging Impact of Guaranteed Pre-tagging on textsc {Senseval-2} Data Experiment Results Discussion An Anomaly in Lexicalized Contextual Rules Conclusions Performance Analysis of a Part of Speech Tagging Task Introduction Classifiers Combination Mathematical Foundations Precision on Agreement Set Lower and Upper Bounds for Overall Precision Empirical Results State of the Art in POS Tagging Experiments Solutions for High Precision POS Tagging Solution 1: Highly Accurate Tagging Using Minimum Human Intervention Solution 2: Combining Taggers for Improved Precision Conclusion An Efficient Online Parser for Contextual Grammars with at Most Context–Free Selectors Introduction The Formalism of Contextual Grammars Linguistic Relevance of CG with Context-Free Selectors An Earley-Based Parser for CGs The Components of the Parser Transformation of Infinite Selector Languages Final Discussion Offline Compilation of Chains for Head-Driven Generation with Constraint-Based Grammars 1 Introduction 2 Related Work 3 Offline Compilation of Chains 3.1 Termination Criteria 3.2 Boundness Situation of Semantic Variables of a Lexical Sign Non Instantiated Variables in Non Head Daughters 4 Applications 4.1 Preventing Over- and Undergeneration 4.2 Avoiding Failing Unifications 5 Evaluation 6 Conclusion References Generation of Incremental Parsers Introduction Parser Generation Standard Parsing Incremental Parsing User Interface Experimental Results Conclusions Computing with Realizational Morphology Introduction Realizational Morphology Formal and Computational Issues Application to Lingala Features Realization Rules Rules of Referral Conclusion Approach to Construction of Automatic MorphologicalAnalysis Systems for Inflective Languages with LittleEffort* 1 Introduction 2 Some Considerations on Inflective Languages 2.1 Static vs. Dynamic Methods 2.2 Morphological Models 3 Approach 4 Conclusions References Per-node Optimization of Finite-State Mechanisms for Natural Language Processing 1 Introduction 2 Per-node Classification 3 Assignment of Polymorphic Formats to Nodes 4 Experimental Results 5 Conclusions and Future Work References An Evaluation of a Lexicographer’s Workbench Incorporating Word Sense Disambiguation Motivations The {sc waspbench} System {sc waspbench} and Machine Translation (MT) Evaluating {sc waspbench} Experimental Setup The Task Instruction and Available Time Data The Participants Evaluation of the Results Summary of the Data Discussion User Experience with the Workbench Conclusions and Further Research Using Measures of Semantic Relatedness for Word Sense Disambiguation Introduction The Lesk Algorithm WordNet Measures of Semantic Relatedness The Leacock--Chodorow Measure The Resnik Measure The Jiang--Conrath Measure The Lin Measure The Hirst--St. Onge Measure Disambiguation Using Semantic Relatedness Experimental Data Experiments and Results Analysis and Discussion Information Content Variations Window Size Variations Related Work Future Work Conclusions Automatic Sense Disambiguation of the Near-Synonyms in a Dictionary Entry Near-Synonyms Sense Disambiguation Intersection of Text and Gloss Other Words in Synsets Being Near-Synonyms Antonyms Systematic Polysemy Context Vectors Using a Decision Tree to Combine Indicators Building a Standard Solution Results and Evaluation Comparison with Related Work Conclusion and Future Directions Word Sense Disambiguation for Untagged Corpus: Application to Romanian Language Introduction A Bootstrapping Algorithm (BA) for WSD The Application for Words' Disambiguation Experiment Experimental Comparison with the NBC Algorithm Further Work Automatic Noun Sense Disambiguation* An Extension of the Conceptual Density Experimental Results and Conclusions Tool for Computer-Aided Spanish Word Sense Disambiguation* 1 Introduction 2 Requirements for a WSD Markup Tool 2 Tool We Developed 3 Conclusions References Augmenting WordNet’s Structure Using LDOCE Introduction Related Work Finding the Connection between a Noun and a Verb Augmenting {it WordNet} 's Structure Noun-Verb Pairs Recognizing Denominal Verbs Word-Sense Disambiguation in {it LDOCE} Word-Sense Disambiguation in {it WordNet} Linking the Noun with the Denominal Verb Results Word-Sense Disambiguation in {it LDOCE} Word-Sense Disambiguation in {it WordNet} Labelling the Links with Semantic Relations Conclusions Building Consistent Dictionary Definitions Introduction The Correspondencies between Syntactic and Semantic Structures of Dictionary Definitions The Syntactic Structures of Dictionary Definitions (in Czech) Noun Definions in SSJ{accent 20 C} Parsing Syntactic Structures of Dictionary Definitions (in Czech -- SSJ{accent 20 C}) Conclusions Is Shallow Parsing Useful for Unsupervised Learning of Semantic Clusters? Experiments on Extracting Semantic Relations from Syntactic Relations Introduction Syntax-Based Technique Adopted Adaptations to Portuguese Extentions to the Notion of Syntactic Context Experiment 1 Experiment 2 Experiment 3 Concluding Remarks A Method of Automatic Detection of Lexical Relationships Using a Raw Corpus Introduction Subsumption Ratio Experiment Conclusions Sentence Co-occurrences as Small-World Graphs: A Solution to Automatic Lexical Disambiguation 1 Introduction 2 The Graph-Theoretic Approach 3 Examples and Discussion of Results References Dimensional Analysis to Clarify Relations among the Top-Level Concepts of an Upper Ontology: Process, Event, Substance, Object 1 Introduction 2 Representation Formalism 3 Dimensional Analysis 4 Basic Concepts 4.1 Object 4.2 Attribute 4.3 Group 4.4 System 4.5 Time 4.6 Function 4.7 State 4.8 GrainSize 4.9 PhysicalSubstance 5 The Representation of Process and Event 5.1 Special Cases of Event 6 Discontinuous Processes 7 Substance and Object 8 Conclusion References Classifying Functional Relations in Factotum via WordNet Hypernym Associations* Introduction Background Importance of Non-hierarchical Semantic Relations Factotum Inferring Relation Markers Classifying the Functional Relations Methodology Results Related Work Conclusion Processing Natural Language without Natural Language Processing 1 Introduction 2 Confusion Set Disambiguation 2.1 Learning Curve Experiments for Confusion Set Disambiguation 3 Lexical Probabilities and Language Modeling 4 AskMSR: Data-Driven Automatic Question Answering 4.1 Utilizing Data Redundancy in Automatic Question Answering 4.2 AskMSR System Architecture 4.2.1 Query Reformulation 4.2.2 N-Gram Mining 4.2.3 N-Gram Filtering 4.2.4 N-Gram Tiling 5 What If an Annotated Corpus Is Needed? 6 Conclusions References The Design, Implementation, and Use of the Ngram Statistics Package Introduction Tokenization of Text From Tokens to Ngrams Counting Ngram Frequencies Counting Bigrams Counting Ngrams Ngram Filters Measures of Association for Ngrams Background Implementation Comparing Ranked Lists of Ngrams Applications of the Ngram Statistics Package Future Work An Estimate Method of the Minimum Entropy of Natural Languages Introduction Minimum Entropy of Natural Languages Estimation of Minimum Entropy Based on a Hypothesis of Conservation of Information Quantity Estimation of Minimum Entropy of Character in Japanese Estimation of Minimum Entropy of Character in Chinese Empirical Proof of ``Conservation of Information Quantity'' Conclusion A Corpus Balancing Method for Language Model Construction 1 Introduction 2 Lexical Analysis of the Training Corpus 2.1 Preprocessing Stage 2.2 Comparison Stage 2.2.1 Comparison of the Probability Distributions 2.2.2 Identification of the Disparate Words 3 Lexical Enrichment of the Training Corpus 3.1 The Process of Enrichment 4 Experimental Results 4.1 Description of the Corpora 4.1.1 The DIME Corpus 4.1.2 The WebDIME Corpus 4.2 Results of Lexical Comparison between DIME and WebDIME 4.3 Results of Lexical Enrichment of WebDIME Corpus 5 Conclusions and Future Work Acknowledgements. This work was done under the partial support of CONACYT (project 31128-A), the fiLaboratorio Franco-Mexicano de Informática (LAFMI)fl, and the Human Language Technologies Laboratory of INAOE. References Building a Chinese Shallow Parsed TreeBank for Collocation Extraction 1 Introduction 2 Shallow Parsing for Collocation Extraction 3 Some Issues in Treebank Annotation 3.1 Guideline Preparation 3.2 Word Segmentation and Part of Speech Tagging 3.3 Syntactic Bracketing 3.4 Annotation Process and Quality Control 4 Conclusion and Future Plans References Corpus Construction within Linguistic Module of City Information Dialogue System Introduction Corpus Construction Corpus of Recorded Sentences Corpus of Generated Sentences Corpus of Simulated Sentences Conclusion and Future Work Diachronic Stemmed Corpus and Dictionary of Galician Language* Introduction Stemming Non-normative Texts Corpus Obtained Possible Applications Conclusions References Can We Correctly Estimate the Total Number of Pages in Google for a Specific Language?* Google Is Extremely Unreliable in All Its Statistics Method of Estimation An Example: Spanish References The Word Is Mightier than the Count: Accumulating Translation Resources from Parsed Parallel Corpora 1 Introduction 2 The Corpus 3 Methods and Results 3.1 Exhaustive Generation and Dictionary Lookup 3.2 Mutual Information 3.3 Expectation Maximization(1) 3.4 Expectation Maximization(2) 3.5 Postprocessing 4 The Discussion References Identifying Complex Sound Correspondences in Bilingual Wordlists Introduction Related Work The Word-to-Word Model of Translational Equivalence Discovering Non-compositional Compounds in Bitexts Implementation of the Algorithm The Algonquian Data Experimental Evaluation Conclusion Generating Texts with Style Introduction Hard and Soft Constraints A Working Example The ICONOCLAST System Satisfying Hard Constraints Satisfying Soft Constraints Conclusions Multilingual Syntax Editing in GF Introduction Multilingual Authoring The GF Niche -- Meaning-Based Technique The Scope of the Paper The GF Syntax Editor The GF Grammar Formalism Abstract Syntax: Simple Example Concrete Syntax Semantic Control in Abstract Syntax Semantic Disambiguation Application Grammars and Resource Grammars Discussion Comparison to Other Systems Future Work QGen – Generation Module for the Register Restricted InBASE System Introduction Organization of the Generation Module QGen Knowledge Representation Planning Realization in Specific NL Linguistic Background Conclusion Towards Designing Natural Language Interfaces 1 Introduction 2 AutoPat Overview 3 User Interface 3.1 Desiderata for an AutoPat Interface Design 3.2 Interface Overview 3.3 Verbose Mode 3.4 Professional Mode 4 Lexicon Customization through the Interface 5 Conclusions References A Discourse System for Conversational Characters 1 Introduction 2 Speech Act Networks 3 Situated Action Planner 4 Conclusion References A Portable Natural Language Interface for Diverse Databases Using Ontologies* 1 Introduction 2 Previous Work of the Authors 3 Previous Work on NLIDBTMs and Ontologies 3.1 NLIDB's Projects 3.2 Ontology Projects 4 Natural Language Query Processing System 5 Current Progress 6 Final Remarks Time-Domain Structural Analysis of Speech Introduction TIDOSA Method Description TIDOSA Structural Primitives Method Application Conclusion Experiments with Linguistic Categories for Language Model Optimization* Introduction Description of Categories Language Model Evaluation Conclusions Chinese Utterance Segmentation in Spoken Language Translation 1 Introduction 2 Related Work and Our Motivations 2.1 Related Work 2.2 Our Motivations 3 Segmentation Based on Multi-level Linguistic Analysis 3.1 Splitting by Keyword Detection 3.2 Splitting by Pattern Matching 4 Experimental Results 5 Conclusion Acknowledgements. This work is sponsored by the Natural Sciences Foundation of China under grant No.60175012, as well as partly supported by the Education Ministry of Japan under Grant-in-Aid for Scientific Research (14380166, 14022237) and a grant funded by the University of Tokushima, Japan. References Using Natural Language Processing for Semantic Indexing of Scene-of-Crime Photographs Introduction Automatic Analysis Robust Parsing and Semantic Interpretation Domain Modelling Implementation Inference and Triples Extraction A Retrieval Mechanism Related Work Conclusions and Future Work Natural Language in Information Retrieval 1 Not Much Natural Language in Information Retrieval So Far 2 NLIR Œ A Natural Language Information Retrieval 3 Rich Resources and Shallow Analysis in Lexware 4 Lexware Applied in Indexing of Swedish Parliamentary Debates 5 Evaluation 6 Conclusions References Natural Language System for Terminological Information Retrieval 1 Introduction 2 Main Issues of IR 2.1 Database Structure 2.2 Searching 2.3 Expanded Searching 2.4 Ranking 2.5 Scoring 3 Experiments 3.1 Query 3.2 Searching 3.3 Ranking 4 Conclusion References Query Expansion Based on Thesaurus Relations: Evaluation over Internet Introduction Query Expansion Heuristic Expansion over the Internet Concluding Remarks Suggesting Named Entities for Information Access Introduction Linguistic Processing of the Document Collection Search Process Conclusions References Probabilistic Word Vector and Similarity Based on Dictionaries Introduction Probabilistic Word Vector Basic Idea Methods Computation of Word Vectors Word Similarity Definition and Method Simulation Evaluation Discussion Conclusions Web Document Indexing and Retrieval Introduction Related Works Web Documents Indexing Scheme Experiments and Results Conclusions Future Works Event Sentence Extraction in Korean Newspapers 1 Introduction 2 Event Sentence Extraction 3 Experiment Results 4 Conclusion References Searching for Significant Word Associations inText Documents Using Genetic Algorithms Introduction The Application of GAs to the Searching Process The Results of Experiments Conclusions Cascaded Feature Selection in SVMs Text Categorization Introduction Our Approach Experiments Data and Preprocessing Results and Discussion Summary and Future Work A Study on Feature Weighting in Chinese Text Categorization* Introduction Feature Weighting 2.1 The Related Formulae The Influential Factors of Weighting Features Our Methods of Weighting Features Experiments Training and Test Set The Multi-step Dimension Reduction The Centroid-Based Classifier Experimental Results Conclusion References Experimental Study on Representing Units in Chinese Text Categorization 1 Introduction 2 Text Representation in Automatic Categorization 3 Naïve Bayes Classifier 4 Experiments and Results 4.1 Corpus for Training and Testing 4.2 Word Segmentation and Part of Speech Tagging 4.3 Scoring Text Categorization 4.4 Experiments and Results 5 Discussion Acknowledgements. Many thanks to Mr. Feng Shicong for providing the corpus. And we gratefully acknowledge comments from two anonymous reviewers. This research was funded by National Natural Science Foundation of China (69973005 and 60173005) and 985 Projects of Peking University. References Partitional Clustering Experiments with News Documents Introduction Documents Description Experiment Description Results Conclusions Fast Clustering Algorithm for Information Organization* 1 Introduction 2 New Clustering Algorithm That Supports the Dense Area 3 Test Results References Automatic Text Summarization of Scientific Articles Based on Classification of Extract’s Population 1 Introduction 2 Related Works on Document Summarization 3 Characteristics of the Proposed Method 4 Design of ExtraGen System 4.1 Statistical Module 4.2 Discourse Module 4.3 Generation and Classification Module 5 Evaluation 6 Conclusion References Positive Grammar Checking: A Finite State Approach Introduction The Child Data System Architecture The Lexicon Lookup The Grammar Parsing and Ambiguity Resolution Error Detection The System Performance Test with Other Tools Conclusion Author Index

This book constitutes the refereed proceedings of the 4th International Conference on Computational Linguistics and Intelligent Text Processing, CICLing 2003, held in Mexico City, Mexico in February 2003.

The 67 revised papers presented together with 4 keynote papers were carefully reviewed and selected from 92 submissions. The papers are organized in topical sections on computational linguistics formalisms; semantics and discourse; syntax and POS tagging; parsing techniques; morphology; word sense disambiguation; dictionary, lexicon, and ontology; corpus and language statistics; machine translation and bilingual corpora; text generation; natural language interfaces; speech processing; information retrieval and information extraction; text categorization and clustering; summarization; and spell-checking.