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
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.
159	Screffva: A Lexicographer's Workbench	This paper describes the implementation of Screffva, a computer system written in Prolog that employs a parallel corpus for the automatic generation of bilingual dictionary entries. Screffva provides a lemmatised interface between a parallel corpus and its bilingual dictionary. The system has been trialled with a parallel corpus of Cornish-English bitext. Screffva is able to retrieve any given segment of text, and uniquely identifies lexemes and the equivalences that exist between the lexical items in a bitext. Furthermore the system is able to cope with discontinuous multiword lexemes. The system is thus able to find glosses for individual lexical items or to produce longer lexical entries which include part-of-speech, glosses and example sentences from the corpus. The corpus is converted to a Prolog text database and lemmatised. Equivalents are then aligned. Finally Prolog predicates are defined for the retrieval of glosses, part-of-speech and example sentences to illustrate usage. Lexemes, including discontinuous multiword lexemes, are uniquely identified by the system and indexed to their respective segments of the corpus. Insofar as the system is able to identify specific translation equivalents in the bitext, the system provides a much more powerful research tool than existing concordancers such as ParaConc, WordSmith, XCorpus and Multiconcord. The system is able to automatically generate a bilingual dictionary which can be exported and used as the basis for a paper dictionary. Alternatively the system can be used directly as an electronic bilingual dictionary.
310	SegWin: a Tool for Segmenting, Annotating, and Controlling the Creation of a Database of Spoken Italian Varieties	A number of actions have been recently proposed, aiming at filling the gap existing in the availability of speech annotated corpora of Italian regional varieties. A starting action is represented by the national project AVIP (Archivio delle Varieta di Italiano Parlato, Spoken Italian Varieties Archive), whose main challenge is a methodological one, namely finding annotation strategies and developing suitable software tools for coping with the inadequacy of linguistic models for Italian accent variations. Basically, these strategies consist in adopting an iterative process of labelling such that a description for each variety could be achieved by successive refinement stages without loosing intermediate stages information. To satisfy such requirements, a specific software system, called SegWin, has been developed by Politecnico di Bari, which: • “guides” the human transcribers in the annotation phases by a sort of “scheduled procedure”; • allows incremental addition of information at any stage of the database creation; • monitors/checks the consistency of the database during every stage of its creation The system has been extensively used by all the partners of the project AVIP and is continuously updated to take into account the project needs. The main characteristics of SegWin are here described, in relation to the above mentioned aspects.
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.
197	Semantic Tagging for the Penn Treebank	This paper describes the methodology that is being used to augment the Penn Treebank annotation with sense tags and other types of semantic information. Inspired by the results of SENSEVAL, and the high inter-annotator agreement that was achieved there, similar methods were used for a pilot study of 5000 words of running text from the Penn Treebank. Using the same techniques of allowing the annotators to discuss difficult tagging cases and to revise WordNet entries if necessary, comparable inter-annotator rates have been achieved. The criteria for determining appropriate revisions and ensuring clear sense distinctions are described. We are also using hand correction of automatic predicate argument structure information to provide additional thematic role labeling.
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.
341	Semi-automatic Construction of a Tree-annotated Corpus Using an Iterative Learning Statistical Language Model	In this paper, we propose a method to construct a tree-annotated corpus, when a certain statistical parsing system exists and no tree-annotated corpus is available as training data. The basic idea of our method is to sequentially annotate plain text inputs with syntactic trees using a parser with a statistical language model, and iteratively retrain the statistical language model over the obtained annotated trees. The major characteristics of our method are as follows: (1)in the first step of the iterative learning process, we manually construct a tree-annotated corpus to initialize the statistical language model over, and (2) at each step of the parse tree annotation process, we use both syntactic statistics obtained from the iterative learning process and lexical statistics pre-derived from existing language resources, to choose the most probable parse tree.
228	Shallow Discourse Genre Annotation in CallHome Spanish	The classification of speech genre is not yet an established task in language technologies. However we believe that it is a task that will become fairly important as large amounts of audio (and video) data become widely available. The technological cability to easily transmit and store all human interactions in audio and video could have a radical impact on our social structure. The major open question is how this information can be used in practical and beneficial ways. As a first approach to this question we are looking at issues involving information access to databases of human-human interactions. Classification by genre is a first step in the process of retrieving a document out of a large collection. In this paper we introduce a local notion of speech activities that are exist side-by-side in conversations that belong to speech-genre: While the genre of CallHome Spanish is personal telephone calls between family members the actual instances of these calls contain activities such as storytelling, advising, interrogation and so forth. We are presenting experimental work on the detection of those activities using a variety of features. We have also observed that a limited number of distinguised activities can be defined that describes most of the activities in this database in a precise way.
82	Shallow Parsing and Functional Structure in Italian Corpora	In this paper we argue in favour of an integration between statistically and syntactically based parsing by presenting data from a study of a 500,000 word corpus of Italian. Most papers present approaches on tagging which are statistically based. None of the statistically based analyses, however, produce an accuracy level comparable to the one obtained by means of linguistic rules [1]. Of course their data are strictly referred to English, with the exception of [2, 3, 4]. As to Italian, we argue that purely statistically based approaches are inefficient basically due to great sparsity of tag distribution - 50% or less of unambiguous tags when punctuation is subtracted from the total count. In addition, the level of homography is also very high: readings per word are 1.7 compared to 1.07 computed for English by [2] with a similar tagset. The current work includes a syntactic shallow parser and a ATN-like grammatical function assigner that automatically classifies previously manually verified tagged corpora. In a preliminary experiment we made with automatic tagger, we obtained 99,97% accuracy in the training set and 99,03% in the test set using combined approaches: data derived from statistical tagging is well below 95% even when referred to the training set, and the same applies to syntactic tagging. As to the shallow parser and GF-assigner we shall report on a first preliminary experiment on a manually verified subset made of 10,000 words.
61	SIMPLE: A General Framework for the Development of Multilingual Lexicons	The project LE-SIMPLE is an innovative attempt of building harmonized syntactic-semantic lexicons for 12 European languages, aimed at use in different Human Language Technology applications. SIMPLE provides a general design model for the encoding of a large amount of semantic information, spanning from ontological typing, to argument structure and terminology. SIMPLE thus provides a general framework for resource development, where state-of-the-art results in lexical semantics are coupled with the needs of Language Engineering applications accessing semantic information.
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.
170	Software Infrastructure for Language Resources: a Taxonomy of Previous Work and a Requirements Analysis	This paper presents a taxonomy of previous work on infrastructures, architectures and development environments for representing and processing Language Resources (LRs), corpora, and annotations. This classification is then used to derive a set of requirements for a Software Architecture for Language Engineering (SALE). The analysis shows that a SALE should address common problems and support typical activities in the development, deployment, and maintenance of LE software. The results will be used in the next phase of construction of an infrastructure for LR production, distribution, and access.
257	Some Language Resources and Tools for Computational Processing of Portuguese at INESC	In the last few years automatic processing tools and studies based on corpora have became of a great importance for the community. The possibility of evaluating and developing such tools and studies depends on the availability of language resources. For the Portuguese language in its several national varieties these resources are not enough to meet the community needs. In this paper some valuable resources are presented, such as a multifunctional lexicon, general-purpose lexicons for European and Brazilian Portuguese and corpus processing tools.
186	Some Technical Aspects about Aligning Near Languages	IULA at UPF has developed an aligner that benefits from corpus processing results to produce an accurate and robust alignment, even with noisy parallel corpora. It compares lemmata and part-of-speech tags of analysed texts but it has two main characteristics. First, apparently it only works for near languages and second it requires morphological taggers for the compared languages. These two characteristics prevent this technique from being used for any pair of languages. Whevener it its applicable, a high quality of results is achieved.
158	Something Borrowed, Something Blue: Rule-based Combination of POS Taggers	Linguistically annotated text resources are still scarce for many languages and for many text types, mainly because their creation repre-sents a major investment of work and time. For this reason, it is worthwhile to investigate ways of reusing existing resources in novel ways. In this paper, we investigate how off-the-shelf part of speech (POS) taggers can be combined to better cope with text material of a type on which they were not trained, and for which there are no readily available training corpora. We indicate—using freely avail-able taggers for German (although the method we describe is not language-dependent)—how such taggers can be combined by using linguistically motivated rules so that the tagging accuracy of the combination exceeds that of the best of the individual taggers.
331	SpeechDat-Car Fixed Platform	SpeechDat-Car aims to develop a set of speech databases to support training and testing of multilingual speech recognition applications in the car environment. Two types of recordings compose the database. The first type consist of wideband audio signals recorded directly in the car while the second type is composed by GSM signals transmitted from the car and recorded simultaneously in a far-end. Therefore, two recording platforms were used, a ‘mobile’ recording platform installed inside the car and a ‘fixed’ recording platform located at the far-end fixed side of the GSM communications system. This paper describes the fixed platform software developed by the Universitat Politecnica de Catalunya (ADA-K). This software is able to work with standard inexpensive PC cards for ISDN lines.
373	SPEECHDAT-CAR. A Large Speech Database for Automotive Environments	The aims of the SpeechDat-Car project are to develop a set of speech databases to support training and testing of multilingual speech recognition applications in the car environment. As a result, a total of ten (10) equivalent and similar resources will be created. The 10 languages are Danish, British English, Finnish, Flemish/Dutch, French, German, Greek, Italian, Spanish and American English. For each language 600 sessions will be recorded (from at least 300 speakers) in seven characteristic environments (low speed, high speed with audio equipment on, etc.). This paper gives an overview of the project with a focus on the production phases (recording platforms, speaker recruitment, annotation and distribution).
63	SPEECON - Speech Data for Consumer Devices	SPEECON, launched in February 2000, is a project focusing on collecting linguistic data for speech recogniser training. Put into action by an industrial consortium, it promotes the development of voice controlled consumer applications such as television sets, video recorders, audio equipment, toys, information kiosks, mobile phones, palmtop computers and car navigation kits. During the lifetime of the project, scheduled to last two years, partners will collect speech data for 18 languages or dialectal zones, including most of the languages spoken in the EU. Attention will also be devoted to research into the environment of the recordings, which are, like the typical surroundings of CE applications, at home, in the office, in public places or in moving vehicles. The following pages will give a brief overview of the workplan for the months to come.
71	Spoken Portuguese: Geographic and Social Varieties	The Spoken Portuguese: Geographic and Social Varieties project has as its main goal the Portuguese teaching as foreign language. The idea is to provide a collection of authentic spoken texts and to make it friendly usable. Therefore, a selection of spontaneous oral data was made, using either already compiled material or material recorded for this purpose. The final corpus constitution resulted in a representative sample that includes European, Brazilian and African Portuguese, as well as Macau and East-Timor Portuguese. In order to accomplish a functional product the Linguistics Center of Lisbon University developed a sound/text alignment software. The final result is a CD-ROM collection that contains 83 text files, 83 sound files and 83 files produced by the sound/text alignment tool. This independence between sound and text files allows the CD-ROM user to manipulate it for other purposes than the educational one.
262	Spontaneous Speech Corpus of Japanese	Design issues of a spontaneous speech corpus is described. The corpus under compilation will contain 800-1000 hour spontaneously uttered Common Japanese speech and the morphologically annotated transcriptions. Also, segmental and intonation labeling will be provided for a subset of the corpus. The primary application domain of the corpus is speech recognition of spontaneous speech, but we plan to make it useful for natural language processing and phonetic/linguistic studies also.
252	Sublanguage Dependent Evaluation: Toward Predicting NLP performances	In Natural Language Processing (NLP) Evaluation, such as MUC (Hirshman, 98), TREC (Harman, 98), GRACE (Adda et al, 97), SENSEVAL (Kilgarriff98), performance results provided are often average made on the complete test set. That does not give any clues on the systems robustness. knowing which system performs better on average does not help us to find which is the best for a given subset of a language. In the present article, the existing approaches which take into account language heterogeneity and offer methods to identify sublanguages are presented. Then we propose a new metric to assess robustness and we study the effect of different sublanguages identified in the Penn Tree Bank Corpus on performance variations observed for POS tagging. The work we present here is a first step in the development of predictive evaluation methods, intended to propose new tools to help in determining in advance the range of performance that can be expected from a system on a given dataset.
317	Survey of Language Engineering Needs: a Language Resources Perspective	This paper describes the current state of an on-going survey that aims at determining the needs of users with respect to available and potentially available Language Resources (LRs). Following market monitoring strategies that have been outlined within the Language Resources- Packaging and Production project (LRsP&P LE4-8335), the main objective of this survey is to provide concrete figures for developing a more reliable and workable business plan for the European Language Resources Association (ELRA) and its Distribution Agency (ELDA), and to determine investment plans for sponsoring the production of new resources.