Text, Speech and Dialogue: 10th International Conference, TSD 2007, Pilsen, Czech Republic, September 3-7, 2007, Proceedings (Lecture Notes in Computer Science, 4629)

This book constitutes the refereed proceedings of the 10th International Conference on Text, Speech and Dialogue, TSD 2007, held in Pilsen, Czech Republic, in September 2007. The 80 revised full papers presented in this volume cover a wealth of state-of-the-art research results in the field of natural language processing with an emphasis on text, speech, and spoken dialogue ranging from theoretical and methodological issues to applications in various fields. Title Page Preface Organization Table of Contents Language Modeling with Linguistic Cluster Constraints Some of Our Best Friends Are Statisticians Some Special Problems of Speech Communication Recent Advances in Spoken Language Understanding Transformation-Based Tectogrammatical Dependency Analysis of English Introduction Layers of Annotation of the Prague Dependency Treebank Dependency Annotation of the Penn Treebank fnTBL Evaluation The Algorithm Phase 1: Deletion of Nodes and Assignment of Functors Phase 2: Relocation, Copying and Creation of Nodes Phase 3: Creation of Leaf Nodes and val_frame.rf Assignment Phase 4: Assignment of the t_lemma Attribute Results and Closing Remarks Multilingual Name Disambiguation with Semantic Information Introduction and Related Work Name Disambiguation with Semantic Information Data Description Experimental Evaluation and Comparative Study Experimental Setup Results and Discussion Conclusions Inducing Classes of Terms from Text Introduction Related Work Clustering by Abstraction The Input: Specific Classes Generating Intermediate Classes Generating Generic Classes by Hierarchical Clustering Classifying Unknown Words Clustering by Specification Experiments and Evaluation Accurate Unlexicalized Parsing for Modern Hebrew Dimensions of Unlexicalized Parsing Dimensions of Modern Hebrew Grammar Modern Hebrew Structure The Modern Hebrew Treebank Scheme Treebank Grammars for Modern Hebrew Experimental Setup Results and Analysis Conclusion Disambiguation of the Neuter Pronoun and Its Effect on Pronominal Coreference Resolution Introduction Data Sets Experimental Setup Effect on Pronominal Coreference Resolution Concluding Remarks Constructing a Large Scale Text Corpus Based on the Grid and Trustworthiness Introduction and Related Work Methodology Constructing the Corpus Based on Grid The Framework to Construct a Corpus Collecting LMTs Automatically Checking LMTs The Trustworthiness of LMTs The Definition of the Trustworthiness TLMP and Its Algorithm TLMU Conclusion References Disambiguating Hypernym Relations for $Roget’s$ Thesaurus Introduction Semantic Distances in $Roget's$ Thesaurus Related Work on Discovering Hypernyms Potential Relations in $Roget's$ Thesaurus Identifying Relations Identifying Hypernyms in Existing Ontologies Identifying Hypernyms in Dictionaries Identifying Hypernyms in a Large Corpus Labeling a Hypernym Network in $Roget's$ Evaluation Manual Evaluation of Hypernyms Combining the Hypernyms from the Resources Evaluation Through Applications Conclusion Dependency and Phrasal Parsers of the Czech Language: A Comparison Introduction The Compared Parsers The Prague Parsers -- Basic Characteristics The Brno Parser -- Basic Characteristics The Principal Differences of the Parsers Q1: The Input Format Q2: Dependency Trees vs. Phrasal Trees Q3: One Resulting Tree vs. (Shared) Forest Q4: Projective vs. Non-projective Trees Q5: The Testing Data Set The Results Measuring Techniques Problems and Discussion Conclusions and Future Directions Automatic Word Clustering in Russian Texts Introduction AWC Techniques Computer Implementation of AWC Experimental Results on AWC with Various Parameters Automatic Clustering of Verbs in Experimental Corpus Automatic Clustering of Descriptors in Scientific Texts Automatic Clustering of Words in Parallel Texts Conclusions and Work in Progress References Feature Engineering in Maximum Spanning Tree Dependency Parser Introduction Adding New Feature Templates Evaluation Feature Space Reduction Selection of the Least Important Features Evaluation Future Work Conclusion Automatic Selection of Heterogeneous Syntactic Features in Semantic Similarity of Polish Nouns Introduction Checking Restrictions Feature Selection and Similarity Measures Evaluation Experiments Conclusions Bilingual News Clustering Using Named Entities and Fuzzy Similarity Introduction Multilingual News Clustering: Our Approach Document Representation: Cognate Named Entities Selection Document Similarity Using a Fuzzy System Evaluation Corpus Clustering Algorithm Experiments and Results Conclusions and Future Work Extractive Summarization of Broadcast News: Comparing Strategies for European Portuguese Introduction Selective Dissemination of Multimedia Contents Extractive Summarization Feature-Based Summarization MMR Summarization LSA Summarization Results Final Remarks On the Evaluation of Korean WordNet Introduction Related Works Data Resources Korean WordNet, U-WIN Korean Monolingual Dictionary, Kum Sung Experiment and Results Experiment Results and Discussion Conclusion and Future Works References An Adaptive Keyboard with Personalized Language-Based Features Introduction Related Work Experiment: Common or Personalized Dictionary User Study: Requirements A Personalized-Adaptive On-Screen Keyboard Conclusion and Discussion An All-Path Parsing Algorithm for Constraint-Based Dependency Grammars of CF-Power Introduction The Grammatical Formalism The Core Algorithm Example Introducing Constraints Example --- Continued Computational Complexity Comparison to Related Works Conclusion Word Distribution Based Methods for Minimizing Segment Overlaps Introduction Previous Work on Text Segmentation Word Distribution New Measures for Distributional Stopword Removal Distribution Difference Distribution Significance Experiments Experimental Procedure Experiment 1 - Baseline Results Experiment 2 - AI Lectures Data Set Experiment 3 - Mars Data Set Conclusions and Future Work On the Relative Hardness of Clustering Corpora Introduction Datasets Calculating the Relative Hardness of a Corpus Clustering the Datasets Correlation Between Relative Hardness and F-Measure Conclusions Further Work Indexing and Retrieval Scheme for Content-Based Multimedia Applications Introduction New Approaches Indexing Scheme: Cellular Tree Query Experiments Feature Extraction and Distance Function Experimental Results Conclusion References Automatic Diacritic Restoration for Resource-Scarce Languages Introduction Grapheme-Based Diacritic Restoration The Data Sets Experiments Experimental Setup Memory-Based Learning Experimental Results: Unknown Words Experimental Results: Plain Text Conclusion and Future Work Lexical and Perceptual Grounding of a Sound Ontology Sounds Without Identifiable Sources Features of Source Unidentifiable Sounds Proposed Ontologies for Sounds Previous Work on Sound Ontologies WordNet's Sound Classification Experiment Method and Materials Results and Discussion Conclusions Named Entities in Czech: Annotating Data and Developing NE Tagger Introduction Proposed Two-Level NE Classification Annotating Data Development of NE Tagger Task Definition Decomposition of the Task Implementation Results and Evaluation Conclusion Identifying Expressions of Emotion in Text Introduction Related Work The Emotion Annotation Task Measuring Annotation Agreement Automatic Emotion Classification Feature Set Experiments and Results Conclusion and Future Work References ECAF: Authoring Language for Embodied Conversational Agents Introduction Background The ECAF Architecture ECAF Language Case Studies Conclusion Dynamic Adaptation of Language Models in Speech Driven Information Retrieval Introduction Related Work System Description Speech Recognition Information Retrieval Language Model Adaptation Experiments Experimental Set-Up Results Analysis of Results Discussion Conclusions Whitening-Based Feature Space Transformations in a Speech Impediment Therapy System Introduction The Whitening Process Independent Component Analysis Whitening-Based Springy Discriminant Analysis Experiments and Results Conclusions Spanish-Basque Parallel Corpus Structure: Linguistic Annotations and Translation Units Introduction Related Work Bilingual Parallel Corpus Structure Adding Information to the Bilingual Parallel Corpus Environment for Consulting Translation Units and Their Linguistic Annotations Conclusion and Future Work An Automatic Version of the Post-Laryngectomy Telephone Test Introduction The Post-Laryngectomy Telephone Test Test Data The Speech Recognition System Results Conclusion and Outlook Speaker Normalization Via Springy Discriminant Analysis and Pitch Estimation Motivation Vocal Tract Length Normalization VTLN with Maximum Likelihood (ML) Criterion VTLN with Linear Discriminant (LD) Criterion Springy Discriminant Analysis VTLN with Springy Discriminant (SD) Criterion Relation Between Pitch and Optimal Warping Factor Experimental Results Future Work A Study on Speech with Manifest Emotions Introduction Existing Approaches Methodology, Database, and Analysis Tools Results of the Analysis of Speech with Manifest Emotions Discussion and Conclusions References Speech Recognition Supported by Prosodic Information for Fixed Stress Languages Introduction Exploiting Prosodic Information Acoustic Prosodic Features Acoustic Prosodic Pre-processing Speech Material Automatic Prosodic Segmentation Speech Recognition Using Prosody First Pass Recognition Lattice Rescoring Second Pass Recognition Experimental Tests Conclusions TRAP-Based Techniques for Recognition of Noisy Speech Introduction Experimental Setup TRAP-Based Feature Extraction Modifications Multi-band TRAP System Critical-Band Spectrogram Modification System Combinations Multi-stream Combination Vector Concatenation Conclusions and Discussions Intelligibility IsMoreThan a SingleWord:Quantification of Speech Intelligibility by ASR and Prosody Introduction Databases Subjective Evaluation Automatic Speech Recognition System Prosodic Features Automatic Evaluation Results Summary Appositions Versus Double Subject Sentences – What Information the Speech Analysis Brings to a Grammar Debate Introduction Double-Subject Sentences in Romanian Methodology Double-Subject Spoken Database Analysis Methodology Double-Subject Sentences Analysis Conclusions and Further Work References Automatic Evaluation of Pathologic Speech – from Research to Routine Clinical Use Introduction TE Voice and Used Database The Automatic Speech Analysis System Recognizer Training Recording Environment Experimental Results Subjective Evaluation Automatic Evaluation Discussion and Summary The LIA Speech Recognition System: From 10xRT to 1xRT Introduction The Speeral Decoder Search Strategy Acoustic-Linguistic Look-Ahead Optimizing the Computation of Acoustic Likelihoods Overview of the Broadcast News Transcription System Transcription Acoustic Modelling Lexical and Language Models Conclusion and Perspectives Logic-Based Rhetorical Structuring for Natural Language Generation in Human-Computer Dialogue Introduction Task-Independent Discourse Ontology SDRT Adaptation for Language Generation in Dialogue Dialogue SDRS Update Rhetorical Relations Computation: Extended Example Conclusions and Further Work Text-Independent Speaker Identification Using Temporal Patterns Introduction Databases $SET-M$ Verbmobil Applied Methods Features Classifier Specifications and Test Phase Training Experiments and Results Discussion Summary Recording and Annotation of Speech Corpus for Czech Unit Selection Speech Synthesis Introduction Selection of Sentences Recording of the Corpus Annotation of the Corpus Annotation Rules The 1st Phase of Annotation The 2nd Phase of Annotation -- Revision Conclusion Sk-ToBI Scheme for Phonological Prosody Annotation in Slovak Motivation Speech Material Sk-ToBI: Break Index Tier Description of Break Indices Sk-ToBI: Tone Tier Possible Realizations of Pitch Contours in Slovak Specific Features of Sk-ToBI Sk-ToBI: Miscellaneous Tier Sk-ToBI: Orthographic Tier Conclusion References Towards Automatic Transcription of Large Spoken Archives in Agglutinating Languages – Hungarian ASR for the MALACH Project Introduction Database Speech and Transcription Data Training and Test Sets Lexical Modeling Approaches The Application of Morph Segmentations Language Modeling Word-Based LM Morph-Based LM Acoustic Modeling Pronunciation Modeling Phoneme Acoustic Modeling Experimental Results Speaker Independent ASR Results Speaker-Wise ASR Results with Unsupervised Speaker Adaptations Conclusions References Non-uniform Speech/Audio Coding Exploiting Predictability of Temporal Evolution of Spectral Envelopes Introduction Encoding Parameterizing Temporal Envelopes in Critical Sub-bands Excitation of FDLP in Frequency Sub-bands Decoding Experiments Objective Quality Tests on HC Conclusions Filled Pauses in Speech Synthesis: Towards Conversational Speech Introduction Prediction of Filled Pauses Classifier Description Feature Sets Corpora Rhythm and Synthesis of Filled Pauses Evaluation Prediction Perceptual Evaluation Conclusion Exploratory Analysis ofWord Use and Sentence Length in the Spoken Dutch Corpus Introduction Related Work Type of Speech Sentence Length Part of Speech Words Gender Education Level Conclusions Design of Tandem Architecture Using Segmental Trend Features Introduction Design of Tandem Architecture Segmental-Feature Trend Features Separation of Trend and Location Trend Clustering Experiments and Results Summary An Automatic Retraining Method for Speaker Independent Hidden Markov Models Introduction The Retraining Procedure The Continuous Speech Recognizer and Automatic Segmentation Experiments and Results Conclusions and Future Work User Modeling to Support the Development of an Auditory Help System Introduction Auditory Help System Model Multiple Streams Attention Memory Auditory Help System Categories of Help Topics Pilot Study Results Summary References Fast Discriminant Training of Semi-continuous HMM Introduction Fast Discriminant Weighting for SCHMM Training Contrastive Experiments on Digit Recognition Task Training Strategy Acoustic Model Training Results Fast Discriminant Weighting on LVCSR Task Experimental Framework SCHMM Estimate Results Conclusion Speech/Music Discrimination Using Mel-Cepstrum Modulation Energy Introduction ME and MCME ME MCME Databases Databases Organization of Training and Test Set Experimental Results and Discussion Experimental Setup Results and Discussion Conclusion Parameterization of the Input in Training the HVS Semantic Parser Introduction The HVS Parser Input Parameterization Experiments Semantics Accuracy Concept Accuracy Morphological Analysis Results Conclusion A Comparison Using Different Speech Parameters in the Automatic Emotion Recognition Using Feature Subset Selection Based on Evolutionary Algorithms Introduction Related Work Study of Automatic Emotion Recognition Relevant Parameters Using Machine Learning Paradigms Corpus Emotional Feature Extraction Machine Learning Standard Paradigms Used Experimental Results Most Relevant Features Conclusions and Future Work References Benefit of Maximum Likelihood Linear Transform (MLLT) Used at Different Levels of Covariance Matrices Clustering in ASR Systems Introduction Maximum Likelihood Linear Transform (MLLT) Levels of CMs Clustering Practical Application of the MLLT for Different Levels of CMs Clustering Results of Experiments Conclusions Information Retrieval Test Collection for Searching Spontaneous Czech Speech Introduction Collection ``Documents'' Topics Relevance Judgments Evaluation Measure Initial Experiments Conclusion and Future Work Inter-speaker Synchronization in Audiovisual Database for Lip-Readable Speech to Animation Conversion Introduction Database Construction The DTW Procedure Experiments and Results Subjective Testing of the Time Warping Results of the DTW Tests Training and Testing the System by a Single Speaker Training and Testing the System by Multiple Speakers Discussion Conclusions References Constructing Empirical Models for Automatic Dialog Parameterization Introduction Problem Setting Related Works Models for Parameter Estimation Numerical Indicators Example Numerical Models The Ivakhnenko Method The Contents of the Method Application of Method Experiments Conclusions and Future Work References The Effect of Lexicon Composition in Pronunciation by Analogy Introduction Pronunciation by Analogy Aligned Lexical Database Substring Matching Building the Pronunciation Lattice Decision Function Lexical Databases BEEP CMU Dictionary Results Conclusions Festival-si: A Sinhala Text-to-Speech System Introduction Sinhala Phonemic Inventory Diphone Database Construction Natural Language Processing Modules The Phone Set Definition Tokenization and Text Normalization Letter-to-Sound Conversion Syllabification and Stress Assignment Phrase Breaking Algorithm Prosodic Analysis Integration with Different Platforms Evaluation Conclusions and Future Work References Voice Conversion Based on Probabilistic Parameter Transformation and Extended Inter-speaker Residual Prediction Introduction Speech Data Speech Analysis and Synthesis Speech Data Alignment and Selection Parameter Transformation Simple LSF Transformation Combined LSF and $F_0$ Transformation Residual Prediction Training Stage Transformation Stage Experiments and Results Objective Evaluation -- LSF and $F_0$ Transformation Subjective Evaluation -- Speaker Discrimination Test Conclusion and Future Work Automatic Czech – Sign Speech Translation Introduction Phrase-Based Machine Translation Tools and Evaluation Methodology Data Evaluation Criteria Decoders Experiments Phrases Comparison "Class-Based" Language Model Post-processing Enhancement Czech to SC Translation Conclusion Maximum Likelihood and Maximum Mutual Information Training in Gender and Age Recognition System Introduction Architecture of the Recognizer Feature Extraction -- Mel-Frequency Cepstral Coefficients Models' Training Gender Estimation Age Estimation Conclusion Pitch Marks at Peaks or Valleys? Introduction The Proposed Method Motivation Description Discussion Experiments and Results Conclusion Quality Deterioration Factors in Unit Selection Speech Synthesis Introduction Evaluation Method Terminology Evaluation Procedure Artefact Collection Factors in Unit Selection Factors Assignment Results Conclusion Topic-Focus Articulation Algorithm on the Syntax-Prosody Interface of Romanian A Tribute to Eva Hajičová’s Speech The IS-Driven Mapping on the Syntax-Prosody Interface Topic-Focus Computing for Romanian Application of the TFA Algorithm for Romanian Intonational TFA Algorithm Conclusion References Translation and Conversion for Czech Sign Speech Synthesis Introduction Sign Speech Synthesis Translation System Conversion System Sign Speech Editor Process of Continuous Sign Speech Synthesis Analysis of Symbols and Trajectories for Isolated Sign Talking Head Subsystem Synchrony of Manual and Non-manual Components Animation Model Perceptual Evaluation Test A: Isolated Signs Test B: Continuous Speech Summary and Conclusions AWizard-of-Oz System Evaluation Study Introduction Wizard-of-Oz Experiments Wizard-of-Oz System Evaluation Predicting User Satisfaction Predicting Individual User-Satisfaction Measures Conclusion New Measures for Open-Domain Question Answering Evaluation Within a Time Constraint Introduction New Approaches Evaluating QA Systems Ranking Function Independent of Time $(MRR_2)$ Ranking Function with Inverse Temporal Dependence $(MRRT)$ Ranking Function with Inverse Exponential-Like with Time Dependence $MRRT_e$ Experiment at QA CLEF-2006 Performance Evaluation Conclusions and Future Work A Methodology for Domain Dialogue Engineering with the Midiki Dialogue Manager Introduction The Methodology Proposed for Domain Dialogue Engineering Declarative Components in the Specification of a Dialogue Procedural Components in the Specification of a Dialogue Conclusions References The Intonational Realization of Requests in Polish Task-Oriented Dialogues Introduction Selection of Recordings Grammatical Forms of REA Dialogue Acts Intonational Realizations of REA Dialogue Acts Instrumental and Perception-Based Analyses Nuclear Melody Pitch Parameters Discussion and Conclusions References Analysis of Changes in Dialogue Rhythm Due to Dialogue Acts in Task-Oriented Dialogues Introduction Dialogue Corpus Recording Speech Data Acoustic Labelling Dialogue Act Tagging Analysis Analysis of Relations Between Initiate, Response, and Follow-Up Analysis Comparing Dialogue Acts Analysis Comparing Aizuchi and Repeat Response Timing for Interjections (Filled Pause) Toward Accomplishment of Natural and Smooth Human-Machine Communication Conclusion Recognition and Understanding Simulation for a Spoken Dialog Corpus Acquisition Introduction An Architecture for the Acquisition of Corpora The DIHANA and the EDECAN Corpora The Previous Task: DIHANA and the Semantic Representation The New Task: EDECAN and the Semantic Representation Understanding Simulator Conclusions and Future Work First Approach in the Development of Multimedia Information Retrieval Resources for the Basque Context Introduction Analysis of Features Across Language Development of Multimodal Digital Resources Evaluation of the Audio Resources Concluding Remarks References The Weakest Link Introduction Views on Grounding Grounding in DIT Information Exchange Through Understanding and Adoption Strengthening Weak Mutual Beliefs Empirical Support of the Strengthening Principle Concluding Remarks A Spoken Dialog System for Chat-Like Conversations Considering Response Timing Introduction Previous Literature on Chat-Like Conversation Dialog System Speech Analysis and Recognition Response Generator Response Timing Generator Speech Synthesizer Response Timing Generation Features for Timing Generation Response Timing Generation Using Decision Tree Example of Dialog with the System Experiments and Results Evaluation of Timing Generation Evaluation of Dialog System Conclusions Digitisation and Automatic Alignment of the DIALOG Corpus: A Prosodically Annotated Corpus of Czech Television Debates Introduction Original Data Storage Original Structure of the Transcription Digitisation and Programming of the DIALOG Corpus Text Processing Transcription Alignment Alignment Evaluation User Interface to the Corpus Conclusion Setting Layout in Dialogue Generating Web Pages Introduction Structure and Architecture of the System Generating a Layout Example Conclusions and Future Work Graph-Based Answer Fusion in Multilingual Question Answering Introduction Related Work Proposed Method Graph Representation Ranking Function Experimental Evaluation Experimental Setup Results Conclusions References Using Query-Relevant Documents Pairs for Cross-Lingual Information Retrieval Introduction The QRDP Probabilistic Model Maximum Likehood Estimation The EuroGOV Corpus Evaluation of the Results Conclusions Detection of Dialogue Acts Using Perplexity-Based Word Clustering Introduction Dialogue Act Detection Word Clustering Algorithm Detection of Dialogue Acts Database Description Experimental Setup and Results Conclusion References Dialogue Management for Intelligent TV Based on Statistical Learning Method Introduction System Overview Dialogue Management for Intelligent TV Dialogue Modeling Discourse Analysis Dialogue Control Empirical Experiments Conclusion References Multiple-Taxonomy Question Classification for Category Search on Faceted Information Introduction Related Work Language Modeling Framework System Description Experiments Dataset Evaluation and Results Error Analysis Conclusions and Future Work Author Index This book constitutes the refereed proceedings of the 10th International Conference on Text, Speech and Dialogue, TSD 2007, held in Pilsen, Czech Republic, September 3-7, 2007. The 80 revised full papers presented together with 4 invited papers were carefully reviewed and selected from 198 submissions. The papers present a wealth of state-of-the-art research results in the field of natural language processing with an emphasis on text, speech, and spoken dialogue ranging from theoretical and methodological issues to applications in various fields and with special focus on corpora, texts and tra Constitutes the refereed proceedings of the 13th International Conference on Text, Speech and Dialogue, TSD 2010, held in Brno, Czech Republic, September 2010.