LREC 2000 2nd International Conference on Language Resources & Evaluation

Papers and abstracts by paper title: A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Papers and abstracts by ID number: 1-50, 51-100, 101-150, 151-200, 201-250, 251-300, 301-350, 351-377.

List of all papers and abstracts

Paper Paper Title Abstract
1 The Cost258 Signal Generation Test Array This paper describes a benchmark for Analysis-Modification-Synthesis Systems (AMSS) that are back-ends of all concatenative speech synthesis systems. After introducing the motivations and principles underlying this initiative, we present here a first anonymous objective evaluation comparing the performance of 5 such AMSS.
2 Collocations as Word Co-ocurrence Restriction Data - An Application to Japanese Word Processor - Collocations, the com bination of specific words are quite useful linguistic resources for NLP in general. The purpose of this paper is to show their usefulness, exem plifying an application to K anji character decision processes for Japanese w ord processors. U nlike recent trials of autom atic extraction, our collocations were collected m anually through many years of intensive investigation of corpus. Our collection procedure consists of (1) finding a proper com bination of words in a corpus and (2) recollecting similar com binations of words, incited by it. This procedure, which depends on hum an judgm ent and the enrichm ent of data by association, is effective for rem edying the sparseness of data problem , although the arbitrariness of hum an judgm ent is inevitable. A pproximately seventy two thousand and four hundred collocations w ere used as w ord co-occurrence restriction data for deciding K anji characters in the processing of Japanese w ord processores. Experiments have show n that the collocation data yield 8.9% higher fraction of Kana-to-Kanji character conversion accuracy than the system w hich uses no collocation data and 7.0% higher, than a com m ercial word processor software of average perform ance.
5 Enhancing the TDT Tracking Evaluation Topic Detection and Tracking (TDT) is a DARPA-sponsored initiative concerned with finding groups of stories on the same topic (tdt, 1998). The goal is to build systems that can segment, detect, and track incoming news stories (possibly from multiple continuous feeds) with respect to pre-defined topics. While the detection task detects the first story on a particular topic, the tracking task determines, for each story, which topic it is relevant to. This paper will discuss the algorithm currently used for evaluating systems for the tracking task, present some of its limitation, and propose a new algorithm that enhances the current evaluation.
7 GREEK ToBI: A System for the Annotation of Greek Speech Corpora Greek ToBI is a system for the annotation of (Standard) Greek spoken corpora, that encodes intonational, prosodic and phonetic information. It is used to develop a large and publicly available database of prosodically annotated utterances for research, engineering and educational purposes. Greek ToBI is based on the system developed for American English (ToBI), but includes novel features (“tiers”) designed to address particularities of Greek prosody that merit annotation, such as stress and juncture. Thus Greek ToBI includes five tiers: the Tone Tier shows the intonational analysis of the utterance; the Prosodic Words Tier is a phonetic transcription; the Break Index Tier shows indices of cohesion; the Words Tier gives the text in romanization; the Miscellaneous Tier is used to encode other relevant information (e.g., disfluency or pitch-halving). The development of GRToBI is largely based on the transcription and analysis of a corpus of spoken Greek, that includes data from several speakers and speech styles, but also draws on existing quantitative research on Greek prosody.
8 English Senseval: Report and Results There are now many computer programs for automatically determining which sense a word is being used in. One would like to be able to say which were better, which worse, and also which words, or varieties of language, presented particular problems to which programs. In 1998 a first evaluation exercise, SENSEVAL, took place. The English component of the exercise is described, and results presented.
10 SALA: SpeechDat across Latin America. Results of the First Phase The objective of the SALA (SpeechDat across Latin America) project is to record large SpeechDat-like databases to train telephone speech recognisers for any country in Latin America. The SALA consortium is composed by several European companies, (CSELT, Italy; Lernout & Hauspie, Belgium; Philips, Germany; Siemens AG, Germany; Vocalis, U.K.) and Universities (UPC Spain, SPEX The Netherlands). This paper gives an overview of the project, introduces the definition of the databases, shows the dialectal distribution in the countries where recordings take place and gives information about validation issues, actual status and practical experiences in recruiting and annotating such large databases in Latin America.
11 Using a Large Set of EAGLES-compliant Morpho-syntactic Descriptors as a Tagset for Probabilistic Tagging The paper presents one way of reconciling data sparseness with the requirement of high accuracy tagging in terms of fine-grained tagsets. For lexicon encoding, EAGLES elaborated a set of recommendations aimed at covering multilingual requirements and therefore resulted in a large number of features and possible values. Such an encoding, used for tagging purposes, would lead to very large tagsets. For instance, our EAGLES-compliant lexicon required a set of about 1000 morpho-syntactic description codes (MSDs) which after considering some systematic syncretic phenomena, was reduced to a set of 614 MSDs. Building reliable language models (LMs) for this tagset would require unrealistically large training data (hand annotated/validated). Our solution was to design a hidden reduced tagset and use it in building various LMs. The underlying tagger uses these LMs to tag a new text in as many variants as LMs are available. The tag differences between these variants are processed by a combiner which chooses the most likely tags. In the end, the tagged text is subject to a conversion process that maps the tags from the reduced tagset onto the more informative tags from the large tagset. We describe this processing chain and provide a detailed evaluation of the results.
12 TransSearch: A Free Translation Memory on the World Wide Web A translation memory is an archive of existing translations, structured in such a way as to promote translation re-use. Under this broad definition, an interactive bilingual concordancing tool like the RALI’s TransSearch system certainly qualifies as a translation memory. This paper describes the Web-based version of TransSearch, which, for the last three years, has given Internet users access to a large English-French translation database made up of Canadian parliamentary debates. Despite the fact that the RALI has done very little to publicize the availability of TransSearch on the Web, the system has been attracting a growing and impressive number of users. We present some basic data on who is using TransSearch and how, data which was collected from the system’s log file and by means of a questionnaire recently added to our Web site. We conclude with a call to the international community to help set up a network of bi-textual databases like TransSearch, which translators around the world could freely access over the Web.
13 Semantic Encoding of Danish Verbs in SIMPLE - Adapting a Verb Framed Model to a Satellite-framed Language In this paper we give an account of the representation of Danish verbs in the semantic lexicon model, SIMPLE. Danish is a satellite-framed language where prepositions and adverbial particles express what in many other languages form part of the meaning of the verb stem. This aspect of Danish - as well as of the other Scandinavian languages - challenges the borderlines of a universal, strictly modular framework which centralises around the governing word classes and their arguments. In particular, we look into the representation of phrasal verbs and we propose a classification into compositional and non-compositional phrasal verbs, respectively, and adopt a so-called split late strategy where non-compositional phrasal verbs are identified only at the semantic level of analysis.
14 A Comparison of Summarization Methods Based on Task-based Evaluation A task-based evaluation scheme has been adopted as a new method of evaluation for automatic text summarization systems. It evaluates the performance of a summarization system in a given task, such as information retrieval and text categorization. This paper compares ten different summarization methods based on information retrieval tasks. In order to evaluate the system performance, the subjects’ speed and accuracy are measured in judging the relevance of texts using summaries. We also analyze the similarity of summaries in order to investigate the similarity of the methods. Furthermore, we analyze what factors can affect evaluation results, and describe the problems that arose from our experimental design, in order to establish a better evaluation scheme.
15 A Word Sense Disambiguation Method Using Bilingual Corpus This paper proposes a word sense disambiguation (WSD) method using bilingual corpus in English-Chinese machine translation system. A mathematical model is constructed to disambiguate word in terms of context phrasal collocation. A rules learning algorithm is proposed, and an application algorithm of the learned rules is also provided, which can increase the recall ratio. Finally, an analysis is given by an experiment on the algorithm. Its application gives an increase of 10% in precision.
16 Perceptual Evaluation of a New Subband Low Bit Rate Speech Compression System based on Waveform Vector Quantization and SVD Postfiltering This paper proposes a new low rate speech coding algorithm, based on a subband approach. At first, a frame of the incoming signal is fed to a low pass filter, thus yielding the low frequency (LF) part. By subtracting the latter from the incoming signal the high frequency (HF), non-smoothed part is obtained. The HF part is modeled using waveform vector quantisation (VQ), while the LF part is modeled using a spectral estimation method based on a Hankel matrix, its shift invariant property and SVD, called CSE. At the receiver side an adaptive postfiltering based on SVD is performed for the HF part, a simple resynthesis for the LF part, before the two components are added in order to produce the reconstructed signal. Progressive speech compression (variable degree of analysis/synthesis at transmitter/receiver) is thus possible resulting in a variable bit rate scheme. The new method is compared to the CELP algorithm at 4800 bps and is proven of similar quality, in terms of intelligibility and segmental SNR. Moreover, perceptual evaluation tests of the new method were conducted for different bit rates up to 1200 bps and the majority of the evaluators indicated that the technique provides intelligible reconstruction.
17 Terms Specification and Extraction within a Linguistic-based Intranet Service This paper describes the adaptation and extension of an existing morphological system,Word Manager,and its integration into an intranet service of a large international bank.The system includes a tool for the analysis and extraction of simple and complex terms.As a side-effect the procedure for the definition of new terms has been consolidated.The intranet service analyzes HTML pages on the fly,compares the results with the vocabulary of an inhouse terminological database (CLS-TDB)and generates hyperlinks in case matches have been found.Currently,the service handles terms in both German and English.The implementation of the service for Italian,French and Spanish is under way.
18 Semantico-syntactic Tagging of Very Large Corpora: the Case of Restoration of Nodes on the Underlying Level The Prague Dependency Treebank has been conceived of as a semi-automatic three-layer annotation system, in which the layers of morphemic and 'analytic' (surface-syntactic) tagging are followed by the layer of tectogrammatical tree structures. Two types of deletions are recognized: (i) those licensed by the grammatical properties of the given sentence, and (ii) those possible only if the preceding context exhibits certain specific properties. Within group (i), either the position itself in the sentence structure is determined, but its lexical setting is 'free' (as e.g. with a deleted subject in Czech as a pro-drop language), or both the position and its 'filler' are determined. Group (ii) reflects the typological differences between English and Czech; the rich morphemics of the latter is more favorable for deletions. Several steps of the tagging procedure are carried out automatically, but most parts of the restoration of deleted nodes still have to be done ''manually''. If along with the node that is being restored, also nodes depending on it are deleted, then these are restored only if they function as arguments or obligatory adjuncts. The large set of annotated utterances will make it possible to check and amend the present results, also with applications of statistic methods. Theoretical linguistics will be enabled to check its descriptive framework; the degree of automation of the procedure will then be raised, and the treebank will be useful for most different tasks in language processing.
19 Coreference in Annotating a Large Corpus The Prague Dependency Treebank (PDT) is a part of the Czech National Corpus, annotated with disambiguated structural descriptions representing the meaning of every sentence in its environment. To achieve that aim, it is necessary i.a. to make explicit (at least some basic) coreferential relations within the sentence boundaries and also beyond them. The PDT scenario includes both automatic and 'manual' procedures; among the former type, there is one that concerns coreference, indicating the lemma of the subject in a specific attribute of the label belonging to a node for a reflexive pronoun, and assigning the deleted nodes in coordinated constructions the lemmas of their counterparts in the given construction. 'Manual' operations restore nodes for the deleted items mostly as pronouns. The distinction between grammatical and textual coreference is reflected. In order to get a possibility of handling textual coreference, specific attributes reflect the linking of sentences to each other and to the context of situation, and the development of the degrees of activation of the 'stock of shared knowledge' will be registered in so far as they are derivable from the use of nouns in subsequent utterances in a discourse.
20 Designing a Tool for Exploiting Bilingual Comparable Corpora Translators have a real need for a tool that will allow them to exploit information contained in bilingual comparable corpora. ExTrECC is designed to be a semi-automatic tool that processes bilingual comparable corpora and presents a translator with a list of potential equivalents (in context) of the search term. The task of identifying translation equivalents in a non-aligned, non-translated corpus is a difficult one, and ExTrECC makes use of a number of techniques, some of which are simple and others more sophisticated. The basic design of ExTrECC (graphical user interface, architecture, algorithms) is outlined in this paper.
22 Creating and Using Domain-specific Ontologies for Terminological Applications Huge volumes of scientific databases and text collections are constantly becoming available, but their usefulness is at present hampered by their lack of uniformity and structure. There is therefore an overwhelming need for tools to facilitate the processing and discovery of technical terminology, in order to make processing of these resources more efficient. Both NLP and statistical techniques can provide such tools, but they would benefit greatly from the availability of suitable lexical resources. While information resources do exist in some areas of terminology, these are not designed for linguistic use. In this paper, we investigate how one such resource, the UMLS, is used for terminological acquisition in the TRUCKS system, and how other domain-specific resources might be adapted or created for terminological applications.
26 The TREC-8 Question Answering Track The TREC-8 Question Answering track was the first large-scale evaluation of domain-independent question answering systems. This paper summarizes the results of the track, including both an overview of the approaches taken to the problem and an analysis of the evaluation methodology. Retrieval results for the more stringent condition in which system responses were limited to 50 bytes showed that explicit linguistic processing was more effective than the bag-of-words approaches that are effective for document retrieval. The use of multiple human assessors to judge the correctness of the systems' responses demonstrated that assessors have legitimate differences of opinion as to correctness even for fact-based, short-answer questions. Evaluations of question answering technology will need to accommodate these differences since eventual end-users of the technology will have similar differences.
27 IREX: IR & IE Evaluation Project in Japanese We will report on the IREX (Information Retrieval and Extraction Exercise) project. It is an evaluation-based project for Information Retrieval and Information Extraction in Japanese. The project started in May 1998 and concluded in September 1999 with the IREX workshop held in Tokyo with more than 150 attendance (IREX Commettee, 1999). There is a homepage of the project at (IREX, Homepage) and anyone can download almost all the data and the tools produced by the project for free.
28 Towards A Universal Tool For NLP Resource Acquisition This paper describes an approach to developing a universal tool for eliciting, from a non-expert human user, knowledge about any language L. The purpose of this elicitation is rapid development of NLP systems. The approach is described on the example of the syntax module of the Boas knowledge elicitation system for a quick ramp up of a standard transfer-based machine translation system from L into English. The preparation of knowledge for the MT system is carried out into two stages; the acquisition of descriptive knowledge about L and using the descriptive knowledge to derive operational knowledge for the system. Boas guides the acquisition process using data-driven, expectation-driven and goal-driven methodologies.
29 The Multi-layer Language Knowledge Base of Chinese NLP This paper introduced the effort to build a multi-layer knowledge base of Chinese NLP which combined with list-based, rule-based and corpus-based language information. Different kinds of information are designed to solve different kind of problems that encountered in the Chinese NLP. The whole knowledge base is designed with theoretical consistency and can easily be put into practice in the application systems.
31 With WORLDTREK Family, Create, Update and Browse your Terminological World Companies need to extract pertinent and coherent information from large collections of documents to be competitive and efficient. Structured terminologies are essential for a better drafting, translation or understanding of technical communication. WORLDTREK EDITION is a tool created to help the terminologist elaborate, browse and update structured terminologies in a ergonomic environment without changing his or her working method. This application can be entirely adapted to the « terminological habits » of the expert. Thus, the data loaded in the software is meta-data. Links, status, property names and domains can be customized. Moreover, the validation stage is facilitated by the use of templates, queries and filters. New terms and links can be easily created to enrich the domains and points of view. Properties like definition, context, equivalent in foreign languages are associated with the terms. WORLDTREK EDITION facilitates the comparison and merging of pre-existing networks. All these tasks and the visualization techniques constitute the tool which will help the terminologist to be more effective and productive.
32 Etude et Evaluation de la Di-Syllabe comme Unité Acoustique pour le Système de Synthèse Arabe PARADIS L' étude que nous présentons dans cet article s' inscrit dans le cadre de la réalisation d' un système de synthèse de la parole à partir du texte pour la langue arabe. Notre système PARADIS est basé sur la concaténation des di-syllabes avec TD-PSOLA comme technique de synthèse. Nous présentons dans cet article l' intérêt du choix de la di-syllabe comme unité de concaténation pour le synthétiseur et son apport au niveau de la qualité de synthèse. En effet, la di-syllabe permet d' améliorer amplement la qualité de synthèse et de réduire les problèmes de discontinuité temporelle lors de la concaténation. Cependant, on est confronté à plusieurs problèmes causés par la taille considérable de l' ensemble des di-syllabes et leur adaptation aux modèles prosodiques qui sont d' habitude associés à la syllabe comme unité rythmique. Nous décrivons alors le principe sur lequel nous nous sommes basés pour réduire le nombre de di-syllabes. Nous présentons ensuite la démarche que nous avons mise au point pour la génération et l' étiquetage automatique du dictionnaire de di-syllabes. Ainsi, nous avons choisi des logatomes ayant des formes particulièrement appropriées à l' automatisation de la procédure de génération du corpus des logatomes et à l' opération de segmentation automatique. Par ailleurs, nous présentons une technique d' organisation du dictionnaire acoustique parfaitement adaptée à la forme de la di-syllabe arabe.
33 Dialogue Annotation for Language Systems Evaluation The evaluation of Natural Language Processing (NLP) systems is still an open problem demanding further research progress from the research community to establish general evaluation frameworks. In this paper we present an experimental multilevel annotation process to be followed during the testing phase of Spoken Language Dialogue Systems (SLDSs). Based on this process we address some issues related to an annotation scheme of evaluation dialogue corpora and particular annotation tools and processes.
34 Evaluation of TRANSTYPE, a Computer-aided Translation Typing System: A Comparison of a Theoretical- and a User-oriented Evaluation Procedures We describe and compare two protocols —one theoretical and the other in-situs —for evaluating the TRANSTYPE system, a target-text mediated interactive machine translation prototype which predicts in real time the words of the ongoing translation.
35 Extraction of Semantic Clusters for Terminological Information Retrieval from MRDs This paper describes a semantic clustering method for data extracted from machine readable dictionaries (MRDs) in order to build a terminological information retrieval system that finds terms from descriptions of concepts. We first examine approaches based on ontologies and statistics, before introducing our analogy-based approach that lets us extract semantic clusters by aligning definitions from two dictionaries. Evaluation of the final set of clusters for a small set of definitions demonstrates the utility of our approach.
36 Obtaining Predictive Results with an Objective Evaluation of Spoken Dialogue Systems: Experiments with the DCR Assessment Paradigm The DCR methodology is a framework that proposes a generic and detailed evaluation of spoken dialog systems. We have already detailed (Antoine et al., 1998) the theoretical bases of this paradigm. In this paper, we present some experimental results on spoken language understanding that show the feasibility and the reliability of the DCR evaluation as well as its ability to provide a detailed diagnosis of the system’s behaviour. Finally, we highlight the extension of the DCR methodology to dialogue management.
37 MHATLex: Lexical Resources for Modelling the French Pronunciation The aim of this paper is to introduce the lexical resources and environment, called MHATLex, and intended for speech and text processing. A particular attention is paid to a pronunciation modelling which can be used in automatic speech processing as well as in phonological/phonetic description of languages. In our paper we will introduce a pronunciation model, the MHAT model (Markovian Harmonic Adaptation and Transduction), which copes with free and context-dependent variants. At the same time, we will present the MHATLex resources. They include 500,000 inflected forms and tools allowing the generation of various lexicons through phonological tables. Finally, some illustrations of the use of MHATLex in ASR will be shown.
38 Dialogue and Prompting Strategies Evaluation in the DEMON System In order to improve usability and efficiency of dialogue systems a major issue is of better adapting dialogue systems to intended users. This requires a good knowledge of users’ behaviour when interacting with a dialogue system. With this regard we based evaluations of dialogue and prompting strategies performed on our system on how they influence users answers. In this paper we will describe the measure we used to evaluate the effect of the size of the welcome prompt and a set measures we defined to evaluate three different confirmation strategies. We will then describe five criteria we used to evaluate system’s question complexity and their effect on users’ answers. The overall aim is to design a set of metrics that could be used to automatically decide which of the possible prompts at a given state in a dialogue should be uttered.
39 SLR Validation: Present State of Affairs and Prospects This paper deals with the quality evaluation (validation) and improvement of Spoken Language Resources (SLR). We discuss a number of aspects of SLR validation. We review the work done so far in this field. The most important validation check points and our view on their rank order are listed. We propose a strategy for validation and improvement of SLR that is presently considered at the European Language Resources Association, ELRA. And finally, we show some of our future plans in these directions.
41 EULER: an Open, Generic, Multilingual and Multi-platform Text-to-Speech System The aim of the collaborative project presented in this paper is to obtain a set of highly modular Text-To-Speech synthesizers for as many voices, languages and dialects as possible, free for use in non-commercial and non-military applications. This project is an extension of the MBROLA project: MBROLA is a speech synthesizer, freely distributed for non-commercial purposes, which uses diphone databases provided by users (19 languages in year 2000). Euler extends this idea to whole TTS systems by providing a backbone structure (MLC) and several generic algorithms for POS tagging, grapheme-to-phoneme conversion, and prosody generation. To demonstrate the potentials of the architecture and draw developpers’ interest we provide a full EULER-based TTS in French and in Arabic. Euler currently runs on Windows and Linux, and it is an open project: many of its components (and certainly its kernel) are provided as GNU C++ sources. It also incorporates, as much as possible, components and data derived from other TTS-related projects.
43 On the Use of Prosody for On-line Evaluation of Spoken Dialogue Systems This paper focuses on the users’ signaling of information status in human-machine interactions, and in particular looks at the role prosody may play in this respect. Using a corpus of interactions with two Dutch spoken dialogue systems, prosodic correlates of users’ discon-firmations were investigated. In this corpus, disconfirmations may serve as a signal to ‘go on’ in one context and as a signal to ‘go back’ in another. With the data obtained from this corpus an acoustic and a perception experiment have been carried out. The acoustic analysis shows that the difference in signaling function is reflected in the distribution of the various types of disconfirmations as well as in different prosodic variables (pause, duration, intonation contour and pitch range). The perception experiment revealed that subjects are very good at classifying disconfirmations as positive or negative signals (without context), which strongly suggests that the acoustic features have communicative relevance. The implications of these results for human-machine interactions are discussed.
44 A Word-level Morphosyntactic Analyzer for Basque This work presents the development and implementation of a full morphological analyzer for Basque, an agglutinative language. Several problems (phrase structure inside word-forms, noun ellipsis, multiplicity of values for the same feature and the use of complex linguistic representations) have forced us to go beyond the morphological segmentation of words, and to include an extra module that performs a full morphosyntactic parsing of each word-form. A unification-based word-level grammar has been defined for that purpose. The system has been integrated into a general environment for the automatic processing of corpora, using TEI-conformant SGML feature structures.
45 The EUDICO Project, Multi Media Annotation over the Internet In this paper we dsecribe a software environment that facilitates media annotation and analysis of media related corpora over the internet. We will describe the general architecture of this environment and we will introduce our Abstract Corpus Model with which we isolate corpora specific formats from the annotation and analysis tools. The main set of tools is described by giving examples of their usage. Finally we will discuss features regarding the distributed character of this environment.
47 Towards a Strategy for a Representation of Collocations - Extending the Danish PAROLE-lexicon We describe our attempts to formulate a pragmatic definition and a partial typology of the lexical category of ’collocation’ taking both lexicographical and computational aspects into consideration. This provides a suitable basis for encoding collocations in an NLP-lexicon. Further, this paper explains the principles of an operational encoding strategy which is applied to a core section of the typology, namely to subtypes of verbal collocation. This strategy is adapted to a pre-defined lexicon model which has been developed in the PAROLE-project. The work is carried out within the framework of the STO-project the aim of which is to extend the Danish PAROLE-lexicon. The encoding of collocations, in addition to single-word lemmas, greatly increases the lexical and linguistic coverage and thereby also the usability of the lexicon as a whole. Decisions concerning the selection of the most frequent types of collocation to be encoded are made on empirical data i.e. corpus-based recognition. We present linguistic descriptions with focus on some characteristic syntactic features of collocations that are observed in a newspaper corpus. We then give a few prototypical examples provided with formalised descriptions in order to illustrate the restriction features. Finally, we discuss the perspectives of the work done so far.
48 Perceptual Evaluation of Text-to-Speech Implementation of Enclitic Stress in Greek This paper presents a perceptual evaluation of a text to speech (TTS) synthesizer in Greek with respect to acoustic registration of enclitic stress and related naturalness and intelligibility. Based on acoustical measurements and observations of naturally recorded utterances, the corresponding output of a commercially available formant-based speech synthesizer was altered and the results were subjected to perceptual evaluation. Pitch curve, intensity, and duration of the syllable bearing enclitic stress, were acoustically manipulated, while a phonetically identical phrase contrasting only in stress served as control stimulus. Ten listeners judged the perceived naturalness and preference (in pairs) and the stress pattern of each variant of a base phrase. It was found that intensity modification adversely affected perceived naturalness while increasing perceived stress prominence. Duration modification had no appreciable effect. Pitch curve modification tended to produce an improvement in perceived naturalness and preference but the results failed to achieve statistical significance. The results indicated that the current prosodic module of the speech synthesizer reflects a good balance between prominence of stress assignment, intelligibility, and naturalness.