Datasets:

taln-ls2n

/

taln-archives

like 3

[ "combinaison de scores", "Système de Questions/Réponses", "compacité", "densité" ]

[ "P", "P", "P", "U" ]

[ "", "This article discusses combinations of scores for selecting the best answer in a factual question answering system. Two major components of our QA system: (i) relevant passage selection, and (ii) answer extraction, produce a variety of scores. Here we study the expressivity of these scores, comparing our passage density score (i) to more conventional ranking techniques in information retrieval. In addition, we study varying the length (in number of sentences) of context retained in the relevant passages. We find that a three sentences window, and a mixing of Lucene or Cosine ranking with our compactness score (ii) provides the best results.", "Question Answering, compactness, density, combination of scores" ]

[ "connecteurs concessifs", "Dialogue homme-machine", "sémantique", "cohérence discursive", "pragmatique" ]

[ "P", "P", "P", "R", "U" ]

[ "", "Cue words play an important part in discourse interpretation (whether dialogues are concerned or not), hence when utterances have to be produced, the choice of the words that connect these utterances (for example, in spoken dialogue) is essential for ensuring the comprehension of the illocutionary goals of the speakers. In computational linguistics, the issue has been mitigated particularly in the context of monologue discourse interpretation, whereas for dialogues, research is usually limited to establishing an almost one-to-one mapping between rhetorical relations and cue words. In this paper we propose a mechanism for guiding concessive connectors in dialogue, at the same time from a discourse and from a semantic point of view; each considered connector will be constrained via a set of conditions that take into account discourse coherence and the semantic relevance of each word concerned. Discourse constraints, expressed in a formalism derived from SDRT (“Segmented Discourse Representation Theory”) will be mapped to semantic constraints on the connectors ; these semantic constraints are proposed by the Geneve linguistics school (Moeschler). Finally, the coherence of the discourse resulted from the use of these connectors will be assessed.", "Human-computer dialogue, discourse coherence, concessive connectors, semantics, pragmatics" ]

[ "ancrage", "robustesse", "dialogue homme-machine", "compréhension mutuelle" ]

[ "P", "P", "P", "P" ]

[ "", "This paper presents a grounding model for robustness in dialogue systems. The grounding process (Clark & Schaefer, 1989) suggests that dialogue participants provide evidence of understanding in order to reach mutual understanding. We propose here a computational definition of the grounding criterion based on understanding judgments that keeps a motivation for providing evidence of understanding, as opposed to some existing models. Eventually, we detail the grounding process on a dialogue generated by the actual implementation.", "human-machine dialogue, robustness, grounding, mutual understanding" ]

[ "Énergie textuelle", "Réseaux de neurones", "Résumé automatique", "Frontières thématiques", "Modèle de Hopfield" ]

[ "P", "P", "P", "P", "M" ]

[ "", "In this paper we present Enertex applications to study fundamental problems in Natural Language Processing. Enertex is based on textual energy, a neural networks approach, inspired by statistical physics of magnetic systems.We obtained good results on the application of this method to automatic multi-document summarization and thematic border detection in three languages : English, Spanish and French.", "Textual Energy, Neural Networks, Hopfield Model, Automatic Summarization, Thematic Boundaries" ]

[ "Résumé automatique", "algorithme génétique", "optimisation multi-objectif", "pré-filtrage de phrases" ]

[ "P", "P", "P", "P" ]

[ "", "In this paper, we present the improvements that we brought to the ExtraNews system dedicated for automatic summarisation of multiple documents. This system is based on the use of a genetic algorithm that combines sentences of the source documents to form the extracts. These extracts are crossed and transferred to generate new ones. The multiplicity of the extract selection criteria inspired us the first improvement that consists in the use of a multi-objectif optimization technique to evaluate these extracts. The second improvement consists of the integration of a sentence pre-filtering step which is based on the notion of dominance between sentences. Our objective is to reduce the sentence number of the source texts. An evaluation of the proposed improvements to our system is realized on DUC' 04 and DUC' 07 corpus.", "Automatic summarization, sentences pre-filtering, multi-objective optimization, genetic algorithm" ]

[ "recherche locale", "Traduction statistique", "post-traitement" ]

[ "P", "P", "U" ]

[ "", "Most phrase-based statistical machine translation decoders rely on a dynamicprogramming technique for maximizing a combination of models, including one or several language models and translation tables. One implication of this choice is the design of a scoring function that can be computed incrementally on partial translations, a restriction a search engine using a complete-state formulation does not have. In this paper, we present experiments we conducted with a simple, yet effective greedy search engine.We report significant improvements in translation quality over a state-of-the-art beam-search decoder, for some configurations.", "Statistical Machine Translation, local search, post-processing" ]

[ "SMS", "décodage phonétique", "modèles de langage", "transducteurs finis" ]

[ "P", "U", "M", "U" ]

[ "", "This paper presents a system aiming at normalizing the orthography of SMS messages, using techniques that are commonly used in automatic speech recognition devices. We describe a baseline system and various evolutions, which are shown to improve significantly the quality of the output normalizations.", "SMS, phonetic decoding, language models, finite-state transducers" ]

[ "grammaires à concaténation d’intervalles", "Grammaires d’arbres adjoints à composantesmultiples", "légère sensibilité au contexte" ]

[ "P", "M", "M" ]

[ "", "This paper investigates the relation between TT-MCTAG, a formalism used in computational linguistics, and RCG. RCGs are known to describe exactly the class PTIME ; « simple » RCG even have been shown to be equivalent to linear context-free rewriting systems, i.e., to be mildly context-sensitive. TT-MCTAG has been proposed to model free word order languages. In general, it is NP-complete. In this paper, we will put an additional limitation on the derivations licensed in TT-MCTAG. We show that TT-MCTAG with this additional limitation can be transformed into equivalent simple RCGs. This result is interesting for theoretical reasons (since it shows that TT-MCTAG in this limited form is mildly context-sensitive) and also for practical reasons (the proposed transformation has been used for implementing a parser for TT-MCTAG).", "Multicomponent Tree Adjoining Grammars, Range Concatenation Grammars, mild context-sensitivity" ]

[ "dépendance", "Analyseur syntaxique" ]

[ "P", "P" ]

[ "", "We describe a parser for lexicalized dependency grammar. The formalism is characterized by a factorization of the syntactic constraints, based on the separation between dependency and word order, the functional (rather than phrasal) specification of dependents, the distinction between valency and non valency dependents, and the incremental saturation of the trees. These features enable a concise formulation of the grammar at a very abstract level and eliminate syntactic information redundancy due to alternative forms of dependents and word order. Each word form produces one or more elementary dependency trees. Trees, both elementary and derived, are combined by adjoining a saturated dependent to a governing tree, after unification of shared nodes and relations. This is achieved using a bi-directional chart parser.", "Syntactic parser, dependency" ]

[ "grammaire", "modélisation", "TAG", "variation dialectale" ]

[ "P", "P", "U", "U" ]

[ "", "The task of writing formal lexicons and grammars for closely related dialects can benefit from factoring part of the modelling. Isomorphic linguistic subsystems from the different dialectsmay have a common description, while the differences are specified aside. This process leads to a layered grammar model: a kernel common to the whole family of dialects, and a superficial skin specifying the particular properties of each one of them. We apply this principle to the family of French-lexifier creole languages of the American-Caribbean area.", "TAG, grammar, modelling, dialectal variation" ]

[ "Analyseur syntaxique statistique", "Analyse syntaxique non lexicalisée", "Analyse du français" ]

[ "M", "R", "R" ]

[ "", "We show that we can acquire satisfactory parsing results for French from data induced from the French Treebank using an unlexicalised parsing algorithm.", "Statistical parser, Unlexicalised parsing, French parsing" ]

[ "Wordnet", "corpus alignés", "Wikipédia", "sémantique lexicale" ]

[ "P", "R", "U", "M" ]

[ "", "This paper describes the construction of a freely-available wordnet for French (WOLF) based on Princeton WordNet by using various multilingual resources. Polysemous words were dealt with an approach in which a parallel corpus for five languages was wordaligned and the extracted multilingual lexicon was disambiguated with the existing wordnets for these languages. On the other hand, a bilingual approach sufficed to acquire equivalents for monosemous words. Bilingual lexicons were extracted from Wikipedia and thesauri. The merged wordnet was evaluated against the French WordNet. The results are promising, and applications are already intended.", "Wordnet, aligned corpora, Wikipedia, lexical semantics" ]

[ "réseau lexical", "jeu en ligne", "sens d’usage d’un terme", "Traitement Automatique du Langage Naturel", "relations typées pondérées" ]

[ "P", "P", "P", "M", "R" ]

[ "", "Lexical information is indispensable for the tasks realized in NLP, but collecting lexical information is a difficult work. Indeed, when done manually, it requires the competence of experts and the duration can be prohibitive. When done automatically, the results can be biased by the corpus of texts. The approach we present here consists in having people take part in a collective project by offering them a playful application accessible on the web. From an already existing base of terms, the players themselves thus build the lexical network, by supplying associations which are validated only if they are suggested by a pair of users. Furthermore, these typed relations are weighted according to the number of pairs of users who provide them. Finally, we approach the question of the word usage determination for a term, by searching relations between this term and its neighbours in the network, before briefly presenting the realization and the first obtained results.", "Natural Language Processing, lexical network, typed and weighted relations, word usage, webbased game" ]

[ "LMF", "Dictionnaire électronique Arabe", "usage éditorial", "interrogation générique", "modèle normalisé" ]

[ "P", "P", "P", "P", "R" ]

[ "", "This paper is interested in the development of the Arabic electronic dictionaries of human use. It proposes a unified and standardized model for these dictionaries according to the future standard LMF (Lexical Markup Framework) ISO 24613. Thanks to its subtle and standardized structure, this model allows the development of extendable dictionaries on which generic interrogation functions adapted to the user’s needs can be implemented. This model has already been carried out on some existing Arabic dictionaries using the ADIQTQ (Arabic DIctionary Query Tool) system, which we developed for the generic interrogation of standardized dictionaries of Arabic.", "Arabic electronic dictionary, human use, standardized model, LMF, generic interrogation" ]

[ "polysémie régulière", "WordNet", "désambiguïsation lexicale", "métaphore", "métonymie" ]

[ "P", "P", "P", "U", "U" ]

[ "", "This paper presents an analysis and modeling of polysemy in the WordNet English lexical database. It exploits the concepts hierarchy (constituted by synsets), and the gloss defining each of these concepts. The result consists of rules set which enabled us to identify in a largely automated way, with a precision close to 91%, more than 2100 synsets pairs, connected by a regular polysemy relation. Our method also allows a partial word sense disambiguation of the definition associated with these synsets.", "regular polysemy, metaphor, metonymy, WordNet, word sense disambiguation" ]

[ "Triggers Inter-Langues", "Traduction Automatique Statistique", "Information Mutuelle", "Corpus parallèle", "Décodage" ]

[ "P", "R", "U", "U", "U" ]

[ "", "In this paper, we present an original approach for machine translation based on inter-lingual triggers. First, we describe the idea of inter-lingual triggers and how to determine them. Then, we present the way to make good use of them in order to integrate them in an entire translation process. We used inter-lingual triggers to estimate different translation tables. Then we compared our translation system based on triggers to a state-of-the-art system based on IBM model 3 (Brown & al., 1993). The experiments showed that automatic translations generated by our system outperform model 3 of IBM by 2.4% in terms of BLEU (Papineni & al., 2001).", "StatisticalMachine Translation, Inter-Lingual Triggers,Mutual Information, Parallel corpus, Decoding process" ]

[ "Paraphrase", "Traduction Automatique Statistique basée sur les segments", "Aide à la rédaction" ]

[ "P", "M", "M" ]

[ "", "In this article, we present a method to obtain paraphrases for short text spans that can be useful to help a writer in reformulating text. The main resource used is a bilingual phrase table containing aligned phrases, a common resource in statistical machine translation. The writer can mark a segment for paraphrasing, and this segment is first translated into a pivot language before being back-translated into the original language, which is possible without performing a full translation of the input. Our proposed framework allows integrating and combining various models for estimating paraphrase quality. We propose linguistic models which permits to conduct empirical experiments about the characteristics of paraphrases for short text spans. Application domains include, in addition to paraphrasing aids, summarization and rephrasing of text for conforming to conventional or stylistic guidelines. We finally discuss the limitations of our work and describe possible ways of improvement.", "Paraphrasing, Phrase-Based Statistical Machine Translation (PBSMT), Authoring aids" ]

[ "architecture linguistique", "Traduction Automatique", "TA", "TAO", "architecture computationnelle", "TA experte", "TA par règles", "TA empirique", "TA statistique", "TA par l'exemple" ]

[ "P", "P", "U", "U", "R", "M", "U", "M", "U", "U" ]

[ "", "Contrary to a wide-spread idea, the linguistic and computational architectures of MT systems are independent. The former concern the choice of the intermediate representations, the latter the type of algorithm, programming, and resources used. It is thus possible to use \"expert\" or \"empirical\" computational methods to build various phases or modules of systems having various linguistic architectures. We finish by giving some elements for choosing these architectures depending on the translational situations and the available resources, in terms of dictionaries, corpora, and human competences.", "Machine Translation, MT, linguistic architecture, computational architecture, expert MT, rule-based MT, empirical MT, statistical MT, example-based MT" ]

[ "Entités nommées", "Types sémantiques", "Analyse syntaxique robuste" ]

[ "P", "P", "P" ]

[ "", "In this paper we propose a new method that enables to correct and to associate new semantic types in the context of Named Entity (NE) Recognition Systems. After named entities (and more generally proper nouns) have been detected in texts, a semantic compatibility checking is performed. This checking can not only confirm or correct previous results of the NER system but also associate new NE types that have not been previously foreseen. This checking is performed using information associated to the NE by a robust syntactic analyzer and confronting this information to WordNet. After this checking is performed, final results of the NER system are better and new NE semantic tags are created. This second point is particularly useful when adapting existing NER systems to new domains.", "Named Entities, Robust Parsing, Semantic Types" ]

[ "treillis de Galois", "Désambiguïsation non supervisée", "entités nommées" ]

[ "P", "P", "P" ]

[ "", "We present an unsupervised method for named entities disambiguation, based on concept lattice mining.We perform a formal concept analysis from relations between named entities and their syntactic contexts observed in a training corpora. The resulting lattice produces concepts which are considered as labels for named entities and context annotation. Our approach is validated through a cascade evaluation which shows that supervised named entity classification is improved by using the annotation produced by our unsupervised disambiguation system.", "Unsupervised word sense disambiguation, concept lattice, named entities" ]

[ "métonymie", "Entités Nommées", "méthode hybride", "analyse syntaxique robuste", "approche distributionnelle" ]

[ "P", "P", "P", "R", "M" ]

[ "", "In this paper, we describe the method we develop in order to solve Named entity metonymy in the framework of the SemEval 2007 competition. In order to perform Named Entity metonymy resolution on location names and company names, as required for this task, we developed a hybrid system based on the use of a robust parser that extracts deep syntactic relations combined with a non supervised distributional approach, also relying on the relations extracted by the parser.We describe this methodology as well as the results obtained at SemEval 2007.", "Named Entities, metonymy, hybrid method, robust parsing, distributional approach" ]

[ "morphosyntaxe", "sémantique", "étiquetage automatique", "terminologie médicale arabe" ]

[ "P", "P", "P", "M" ]

[ "", "This paper examines the semantic relations existing in Arabic medical texts between the head and its extension in a two-noun compound, particularly when the head is a deverbal noun or a nominalization. By studying semantic relations encoded by nominalizations, this research work aims at underlining the correlation between morphosyntax and semantics notably the influence of the head noun structure on the set of semantic relations that can be established. The discovery of regularities in the functioning of the head noun allows thus to predict the type of relation that will be encoded. Such data are a pre-requisite for natural language processing and automatic part-to-speech tagging of medical Arabic texts.", "Part-of-speech tagging, Arabic medical terminology, morphosyntax, semantics" ]

[ "chunker", "Analyse syntaxique", "étiquetage morphosyntaxique", "analyseur stochastique", "analyseur symbolique superficiel" ]

[ "P", "U", "M", "M", "M" ]

[ "", "We show in this paper that a strong correlation exists between the performance of chunk parsers and the quality of the tagging task in input. This dependency becomes linear when the size of the chunks is small. Our demonstration is based on an experiment conducted at the end of the Passage 2007 shared task evaluation initiative (de la Clergerie et al., 2008). The performance of two parsers which took part in this evaluation has been investigated. The results indicate that the chunking task, for sufficiently short chunks, is similar to a super-tagging task.", "Parsing, tagging, stochastic parser, symbolic shallow parser, chunker" ]

[ "Apprentissage Artificiel", "Programmation Logique Inductive", "Analyse et Indexation/méthodes", "Medical Subject Headings" ]

[ "P", "P", "M", "U" ]

[ "", "Indexing is a crucial step in any information retrieval system. In MEDLINE, a widely used database of the biomedical literature, the indexing process involves the selection of Medical Subject Headings in order to describe the subject matter of articles. The need for automatic tools to assist human indexers in this task is growing with the increasing amount of publications to be referenced in MEDLINE. In this paper, we describe the use and the customization of Inductive Logic Programming (ILP) to infer indexing rules that may be used to produce automatic indexing recommendations for MEDLINE indexers. Our results show that this original ILP-based approach overperforms manual rules when they exist. We expect the sets of ILP rules obtained in this experiment to be integrated in the system producing automatic indexing recommendations for MEDLINE.", "Abstracting and Indexing/methods, Medical Subject Headings, Machine Learning, Inductive Logic Programming" ]

[ "proposition syntaxique", "Alignement", "études contrastives français-japonais", "similarité lexicale" ]

[ "P", "P", "M", "U" ]

[ "", "This paper is based on our observations obtained from the results of our French-Japanese clause alignment system. Structures fundamentally difficult to align were so numerous and results obtained by various word-matching methods were so unsatisfactory that we questioned the existence of equivalence at the syntactic clause level between French and Japanese. In order to compensate the defect that we discovered, we propose some operations to restore aligned clause equivalence to improve the quality of aligned corpora.", "Alignment, syntactic clause, French-Japanese contrastive study, lexical similarty" ]

[ "Unification", "arbre de dérivation", "grammaire rationnelle d’arbres", "grammaire d’arbres adjoints" ]

[ "P", "P", "P", "P" ]

[ "", "The derivation trees of a tree adjoining grammar provide a first insight into the sentence semantics, and are thus prime targets for generation systems. We define a formalism, feature based regular tree grammars, and a translation from feature based tree adjoining grammars into this new formalism. The translation preserves the derivation structures of the original grammar, and accounts for feature unification.", "Unification, tree adjoining grammar, derivation tree, regular tree grammar" ]

[ "segmentation thématique", "Méthodes d’évaluation", "segmentation de texte" ]

[ "P", "R", "R" ]

[ "", "This paper present a semantic and syntactic distance based method in topic text segmentation and compare it to a very well known text segmentation algorithm : c99. To do so we ran the two algorithms on a corpus of twenty two French political discourses and compared their results. Our two conclusions are that the two approaches are complementary and that evaluation methods in this domain should be revised.", "Evaluation methods, text segmentation, topic segmentation" ]

[ "Grammaire Sémantique", "Dialogue", "Analyse", "Génération", "Réversibilité" ]

[ "P", "P", "P", "P", "U" ]

[ "", "In this paper we present a language model for man-machine dialogue, as well as algorithms for analysis and text generation. The originality of our approach is to base analysis and generation on the same knowledge. These one is structured like a library of concepts and syntactic patterns. The algorithms are based on a principle of correspondence between offers and expectations, and calculation of a heuristic scoring.", "Semantic Grammar, Reversibility, Analysis, Generation, Dialogue" ]

[ "DRT", "Modèle sémantique", "analyse syntaxique en dépendance" ]

[ "P", "M", "M" ]

[ "", "This works present an extension of the formal representation of semantic for the natural language processing tool developped at Orange Labs. We cover here only issues relating to build the semantic representations, based on linguistic parsing. To obtain more detailed representations of the argumental structure statements, we include insights from the DRT in the system of representation based on the semantic graphs to give an account of scope.", "Semantic modelling, parsing, DRT" ]

[ "plate-forme de validation", "Lexiques", "cadres de sous-catégorisation" ]

[ "P", "P", "M" ]

[ "", "Lexicon production is essential but complex and all creation methods (automatic acquisition or manual creation) require human validation. For this purpose, we propose a freely available Web-based framework, named Sylva (Systematic lexicon validator). The main point of our framework is that it handles multi-level validations and keeps track of the resource’s history. The expert linguist task is made easier : (s)he has only to consider data on which validators disagree.", "Lexicons, validation framework, subcategorization frames" ]

[ "Offres d’emploi", "Classification de textes", "Modèle probabiliste", "Ressources humaines" ]

[ "P", "R", "U", "U" ]

[ "", "The exponential growth of the Internet has allowed the development of a market of on-line job search sites. This paper presents the E-Gen system (Automatic Job Offer Processing system for Human Resources). E-Gen will perform two complex tasks : an analysis and categorisation of job postings, which are unstructured text documents, an analysis and a relevance ranking of the candidate answers (cover letter and curriculum vitae). Here we present the work related to the second task : we use vectorial representation before generating a classification with SVM to determine the type of the attachment. In the next step, we try to classify the candidate answers with different classifiers (SVM and ngrams of words).", "Text Classification, Probabilistic Model, Human Ressources, Job Offer" ]

[ "Machine finie à états", "morphologie à deux niveau" ]

[ "M", "M" ]

[ "", "In this paper, we present a new language to write rational relations compiled into finite state automata. There are two main novelties in the language. Firstly, the descriptions may have more than two levels. Secondly, various units may be used to express the relationships between the levels. Using these features, it is possible to align finely multiple representations.", "Finite-state machine, two-level morphology" ]

[ "multimodalité", "paradigme d’évaluation", "diagnostic", "Dialogue finalisé", "évaluation pour le dialogue hommemachine", "test utilisateur", "paraphrase multimodale" ]

[ "P", "P", "P", "M", "M", "U", "M" ]

[ "", "Evaluating human-machine dialogue systems is not so efficient, objective, and consensual than evaluating other natural language processing systems. Oral and multimodal dialogue systems are still working within reduced applicative domains. Comparative and normative evaluations are then difficult. Moreover, the continuous technological progress makes obsolete and numerous the evaluating paradigms. Some solutions are still to be identified to improve existing methods and to allow a more automatic diagnosis of systems. The aim of this paper is to provide a set of remarks dealing with the evaluation of multimodal spoken language dialogue systems. Some extensions of existing paradigms are presented, in particular DQR/DCR, considering that some paradigms fit better multimodal issues than others. Some conclusions and perspectives are then drawn on the future of the evaluation of human-machine dialogue systems.", "Task-driven dialogue, multimodality, evaluating human-machine dialogue, evaluation paradigm, user test, diagnosis" ]

[ "morphologie dérivationnelle", "ressource lexicale", "traitement automatique des familles de mots" ]

[ "P", "R", "M" ]

[ "", "In this paper we present POLYMOTS, a lexical database containing eight thousand common nouns in French. Whereas most of the existing lexicons for derivational morphology take affixes as starting point for producing paradigms of words, we defend here the idea that it is possible to isolate a morpho-phonological stem and produce a paradigm of words belonging to the same family. This point leads us to describe three types of stems according to their phonological and morphological form. The article presents the different features of such a lexical database and discusses the applications and future work using and enriching this resource.", "lexical resource, derivational morphology, word families processing" ]

[ "génération", "traduction automatique", "néologisme", "morphologie", "études empiriques" ]

[ "P", "P", "P", "U", "M" ]

[ "", "Generating constructed neologisms in a machine translation system is confronted to the issue of selecting the right affixes, especially when some affixes can be used alternately. We propose here an “extensive” study that looks into large textual data collections (web) for prefixed forms that have been automatically generated, in order to find out parameters that allow the use of both prefixes or, on the contrary, that prevent one or the other prefixation. The volatility of web resources requires methodological precautions, especially in data counting.", "morphology, machine translation, generation, neologism, empirical studies" ]

[ "transducteurs", "parole conversationnelle", "segmentation", "Unitex", "Traitement Automatique du Langage Parlé (TALP)", "chunks" ]

[ "P", "P", "P", "P", "M", "U" ]

[ "", "This paper describes the use of the CasSys platform in order to achieve the chunking of conversational speech transcripts by means of cascades of Unitex transducers. Our system is involved in the EPAC project of the French National Agency of Research (ANR). The aim of this project is to develop robust methods for the annotation of audio/multimedia document collections which contains conversational speech sequences such as TV or radio programs. At first, this paper presents the adaptation of a former chunking system (Romus) which was developed in the restricted framework of dedicated spoken manmachine dialogue. Then, it describes the problems that are arising due to 1) spontaneous speech disfluencies and 2) errors for the previous stages of processing (automatic speech recognition and POS tagging).", "Spoken Language Processing, chunking, conversational speech, transducers, Unitex" ]

[ "Regroupement de documents", "Suivi d’événement" ]

[ "P", "M" ]

[ "", "This paper analyses the problem of automatic document clustering based on events. We first specify the notion of event. Then, we detail the document modelling method and the learning approach for document clustering based on k-means. We finally evaluate our document clustering system on a small corpus and discuss the quantitative evaluation for this kind of task.", "Document clustering, Event tracking" ]

[ "classification des styles de discours rapporté", "écrits journalistiques", "détection de citations", "identification du locuteur", "techniques par apprentissage et base de règles" ]

[ "P", "P", "M", "M", "M" ]

[ "", "In the application context of reported content, that includes plagiarism and impact of textual information searched, citations finding and its fundamentals is essential as it may help estimating legal value of a citation (with or without specifying original source). We propose here a comparison between automatic methods for finding up those elements and we quantitatively evaluate them. A French journalistic corpus has been manually annotated to be used as learning base and for testing.", "detection of citations, reported speech style classification, source identification, machine learning and rules-based techniques, news corpus" ]

[ "structure du dialogue", "Actes de langage complexes", "terrain commun" ]

[ "P", "M", "U" ]

[ "", "Attributing one illocutionary value to a utterance in a dialogue can be difficult. First, a utterance can include several discourse segments, each one with a specific illocutionary value. Moreover, one discourse segment can be linked to a complex speech act that groups for instance a question together with the assertion of new information. Finally, the dialogue structure with the various exchanges and sequences can be decisive when identifying the dialogue act, and can also bring an additional illocutionary value, for instance the one consisting of closing the current sequence. With the aim to determine how to react to such a composite dialogue act, we present a theoretical approach to dialogue act analysis considering the task context and the dialogue participants’ knowledge. We illustrate our choices in terms of segmentation and identification of composite acts, and we present the main features of a strategy for determining the most relevant reaction.", "Complex speech acts, dialogue structure, common ground" ]

[ "représentation évènementielle", "extraction d’information", "Sous-langage", "structure prédicative", "structure predicate-arguments" ]

[ "P", "P", "U", "M", "M" ]

[ "", "The representation of motion events is important for an automatic comprehension of disease outbreak reports. The dispersion of components in this type of reports makes it difficult to have such a representation. This paper describes an automatic extraction of event structures representation of these texts.We built an information extraction system by using cascaded finite state transducers which allowed the realization of three tasks : the named entity recognition, the component annotation and representation and the event structure representation. We obtained a recall of 80, 93% for the named entity recognition task and a recall of 97, 88% for argument representation task. Thereafter, we worked in anaphoric pronouns resolution where we obtained a precision of 81.83%.", "Sublanguage, event structure representation, information extraction, predicative structure, predicate-arguments structure" ]

[ "Traduction automatique multilingue", "approche par objets", "génération de lexiques bilingues" ]

[ "M", "M", "M" ]

[ "", "The increase of cross-cultural communication triggered notably by the Internet intensifies the needs for multilingual linguistic tools, in particular translation systems for several languages. The LATL has developed an efficient multilingual parsing technology based on an abstract and generic linguistic model and on object-oriented software design. The proposed project intends to apply a similar approach to the problem of multilingual translation (German, French, Italian and English).", "Multilingual machine translation, object design, bilingual dictionary derivation" ]

[ "Appariement", "Entités nommées", "Apprentissage supervisé", "Mesures de similarité textuelle" ]

[ "P", "P", "P", "P" ]

[ "", "Matching named entities consists in grouping the different forms under which an entity may occur. Textual similarity measures are the usual tools for this task. We propose to combine several measures in order to improve the performance. We show the relevance of supervised learning in this objective through experiences with two corpora, especially in the case of the C4.5 algorithm.", "Named entities,Matching, Textual similaritymeasures, Supervised learning" ]

[ "logique", "Graphe sémantique", "quantification", "Théorie Sens-Texte (TST)" ]

[ "P", "P", "U", "M" ]

[ "", "We discuss the meaning of semantic graphs, in particular of those used in Meaning-Text Theory. We provide a precise, possibly underspecified, meaning to such graphs through a simple translation into a Minimal Recursion Semantics formula. This translation covers cases of multiple predications over several entities, higher order predication and modalities.", "Semantic graph, logic, quantification, Meaning-Text Theory (MTT)" ]

[ "frames sémantiques", "annotation sémantique", "compréhension automatique de la parole", "système de dialogue oral", "décodage conceptuel", "inférence sémantique" ]

[ "P", "P", "M", "M", "U", "R" ]

[ "", "This paper introduces a knowledge representation formalism, used for incremental and partially automated annotation of the French MEDIA dialogue corpus in terms of semantic structures. We describe an automatic interpretation process for composing semantic structures from basic semantic constituents using patterns involving constituents and words. The process has procedures for obtaining semantic compositions and generating frame hypotheses by inference. This process is applied to MEDIA, a dialogue corpus manually annotated at the word and semantic constituent levels, and thus produces a higher level semantic frame annotation. The Knowledge Source defined and the results obtained on the automatically-derived annotation are reported.", "spoken language understanding, spoken dialogue system, semantic structures, semantic frames, conceptual decoding, semantic annotation, semantic inference" ]

[ "indexation", "recherche d’information", "langue arabe", "Google", "lemmatisation" ]

[ "P", "P", "P", "P", "U" ]

[ "", "Web search engines provide quite good results for Latin characters-based languages. However, they still show many weaknesses when searching in other languages such as Arabic. This paper discusses a qualitative analysis of information retrieval in Arabic language, highlighting some of the numerous limitations of available search engines, mainly when they are not properly adapted to the Arabic language specificities. To argue our analysis, we present some results based on quite sufficient observations and tests on various Arabic linguistic phenomena. To validate these observations, we essentially have tested the Google search engine.", "information retrieval, Arabic, indexation, lemmatization, Google" ]

[ "outils de correction linguistique", "Microsoft", "français", "correcteur orthographique", "correcteur grammatical", "réforme de l’orthographe", "féminisation des noms de métier" ]

[ "P", "P", "P", "P", "P", "P", "P" ]

[ "", "New French proofing tools were recently made available to Microsoft Office 2003 users. A new spell-checker and a new grammar checker make it possible to improve the document creation process. Based on recent external evaluations, we present specific aspects of these improved tools and discuss the more fundamental issue of how to evaluate proofing tools (What do we need to evaluate? Which criteria should be applied? Why develop a new version?). Current language changes including the spelling reform, and such innovations as feminine job titles are among the themes we discuss in this paper.", "spell-checker, grammar checker, French, proofing tools, Microsoft, spelling reform, feminine job titles" ]

[ "polysémie", "cooccurrences", "correspondances de traduction", "prédiction de traduction" ]

[ "P", "U", "R", "R" ]

[ "", "Word Sense Disambiguation has a central role in NLP applications relevant to translation. The work presented in this article is a part of a study on the overlaps and divergences between the semantic spaces occupied by the polysemous items of two languages. Correspondences between these items are rarely biunivocal and their study provides insights into the possibilities and limits of using a second language for the disambiguation of polysemous items of a source language. The aim of this work is to establish correspondences of optimal granularity between the items of two languages in translation. Such correspondences could be used for the prediction of the most adequate translation equivalents for new occurrences of the polysemous source language items.", "polysemy, cooccurrences, translation correspondences, translation prediction" ]

[ "analyseur morphologique", "morphologie à deux niveaux" ]

[ "P", "M" ]

[ "", "In this paper, we present a morphological analyzer for the Akkadian verb. This language belongs to the semitic family. Verb inflection involves modifications within the root. The analyzer is an example of the use of a multilevel formalism having powerful relational operators, notably the join. The morphology is described using relatively simple constraints which are composed together in three possible ways : they apply simultaneously or sequentially, or even in a mixed way, where the constraints apply simultaneously on some levels and sequentially on others. This mechanism is used to describe the vocalization of verbal roots as successive vowels insertions. This process is simpler than the use of fixed vocalic patterns which implies interdigitation. Moreover, it conveniently explains the forms of weak verbs.", "morphological analyzer, two-level morphology" ]

[ "sémantique quantitative", "spécificités", "cooccurrences", "analyse de régression", "polysémie", "sémantique lexicale" ]

[ "P", "P", "P", "P", "P", "M" ]

[ "", "This article discusses the methodology and results of a quantitative semantic analysis of about 5000 keywords (pivotal terms) in the domain of French machining terminology. The KeyWords Method is used in order to identify the most typical words. Next, a quantitative semantic analysis of the keywords determines their degree of monosemy, which is implemented in terms of degree of overlap between co-occurrents of co-occurrents of keywords. Finally, the quantitative data is submitted to various regression analyses, in order to check the hypothesis that the most typical terms are not always the most monosemous terms. This article presents the statistical results of this semantic analysis and provides linguistic interpretation. Building on corpus data, the investigation attempts to establish in how far keywords are polysemous and which factors are most predictive.", "lexical semantics, quantitative semantics, keywords, polysemy, co-occurrences, regression analysis" ]

[ "ressources linguistiques", "analyse syntaxique", "motifs lexico-syntaxiques", "analyse lexicale", "formats d’échange", "automates finis", "réseaux de transitions récursifs" ]

[ "P", "U", "U", "U", "M", "M", "M" ]

[ "", "The Outilex software platform, soon available to research, development and industry, comprises software components implementing all the fundamental operations of written text processing, including processing without lexicons, exploitation of lexicons and grammars, and language resource management. All data are structured in XML formats, and more compact readable or binary formats, if required. The required format converters are included in the platform ; the grammar formats allow for statistical approaches to be combined with resource-based approaches. Manually constructed lexicons for French and English, originating from the LADL, with substantial coverage, will be distributed with the platform under LGPL-LR license.", "syntactic parsing, lexico-syntactic patterns, lexical analysis, language resources, exchange formats, finite-state automata, recursive transition networks" ]

[ "grammaires multilingues", "traduction automatique de la parole", "modélisation du langage", "grammaire d’unification", "reconnaissance linguistique" ]

[ "P", "P", "U", "M", "M" ]

[ "", "Today, the most common architecture for speech understanding consists of a combination of statistical recognition and robust semantic analysis. For many applications, however, grammar-based recognisers can offer significant advantages. In this paper, we present a methodology and an Open Source platform (Regulus), which together permit rapid derivation of linguistically motivated recognisers, in either language, from a bilingual grammar of Catalan and French.", "speech to speech translation, language modelling, unification grammar, grammar-based recognition, multilingual grammars" ]

[ "disfluences", "analyse syntaxique en dépendances", "traitement automatique de l’oral" ]

[ "P", "M", "R" ]

[ "", "In this paper we describe an attempt to take speech disfluencies into account in a linguistic analysis system initially designed for written data. Using a detailed corpus analysis, we show how the lexicon and grammar can be modified to solve the simplest cases (such as filled pauses, single-word repeats, and so forth). More difficult cases such as phrasal repeats and self-repairs required the development of a semantic control mechanism in order to avoid combinatorial explosion. This study also reveals the difficulty of processing word fragments and aborted constructs, which receive no satisfactory solution in the current state of the art.", "disfluencies, parsing, automatic speech processing" ]

[ "acquisition", "concepts bilingue", "Web", "alignement superficiel" ]

[ "P", "P", "P", "M" ]

[ "", "This article describes a way of using the Web as a huge multilingual corpus to perform supervised acquisition of bilingual French/English concepts. Such acquisition is initialized with a French verb. Sentences expressed in both languages are then matched by using couples of proper nouns which are unchanged in these two languages. Such matching is automatic, highly succinct, but does not prevent sentences from being wrongly aligned. However, we show that it makes it possible to extract French and English terms that are equivalent in their context of use. These terms constitute multilingual resources that are particularly adapted to the Web, especially for cross-lingual question answering.", "acquisition, bilingual concepts, shallow alignment, Web" ]

[ "relations sémantiques", "analyse distributionnelle", "ressources lexicales" ]

[ "P", "P", "U" ]

[ "", "In this paper, we study the semantic relations that hold between nouns and verbs. We benefit from the data provided by Upery, a program that automatically extracts word associations from a 200 million words corpus by means of distributional analysis. We present the results of an experiment in which noun-verb associations are extracted by crossing two criteria : distributional proximity and cooccurrence. The 1441 couples share the same arguments in the corpus and appear at least once with the same argument within the same paragraph. We present a typology of these noun-verb couples, showing the necessity to take into account non-mophologically related couples which amounts to 70 % of the data.", "semantic relations, lexical resources, distributional analysis" ]

[ "résumés multidocuments", "synthèse d’information", "évaluation du résumé" ]

[ "P", "P", "R" ]

[ "", "We present an information synthesis system for the production of multidocument summaries tailored by a complex query.We first analyze the user profile expressed by complex questions and we then compare the similarity between the documents and the questions at two levels : global and detailed. This research shows the importance of the study of the relevance of a sentence in the thematic structure of the document. This methodology has been applied for our participation in the DUC 2005 evaluation conference where our system was judged to be among of the best ones.", "information synthesis, multi-document summarization, evaluation of the summary" ]

[ "sous-catégorisation", "lexique-grammaire", "M. Gross" ]

[ "P", "U", "U" ]

[ "", "Maurice Gross’ grammar lexicon contains rich and exhaustive information about the morphosyntactic and syntactic properties of French syntactic functors (verbs, adjectives, nouns). Yet its use within natural language processing systems is hampered both by its non standard encoding and by a structure that is partly implicit and partly underspecified. In this paper, we present a method for translating this information into a format more amenable for use by NLP systems, we discuss the results obtained so far, we compare our approach with related work and we identify the possible further uses that can be made of the reformatted information.", "Grammar Lexicon, M. Gross, subcategorisation" ]

[ "grammaire d’arbres adjoints", "calcul sémantique" ]

[ "P", "M" ]

[ "", "In this paper, we consider a linguistic formalism for which the integration of semantic information within a large scale grammar has not yet been realised, namely, Tree Adjoining Grammar (TAG). We propose a method for integrating this information and describe its implementation within a core TAG for French called FRAG. Furthermore, we show that the formalism used to specify the grammar, XMG, (Duchier et al., 2004) allows for a strong factorisation of the semantic data, thus giving better support for the development, maintenance and debugging of the grammar.", "Tree Adjoining Grammar, Semantic Construction" ]

[ "questions booléennes", "système de questions-réponses" ]

[ "P", "P" ]

[ "", "In this paper, we propose a method to answer Yes/No questions from a Questions-Answering System point of view. These questions were called “Boolean question” during the first Technolangue French question answering campaign called EQueR which used a French corpus and questions. It was also the first campaign to introduce an evaluation for this particular kind of question. We discuss some key points related to these Yes/No questions as well as report on our system used in the EQueR evaluation campaign.", "question answering, yes/no questions, boolean question" ]

[ "-ité", "-Able", "corpus journalistique", "morphologie", "productivité morphologique quantitative" ]

[ "P", "P", "P", "U", "M" ]

[ "", "In this paper, we study the quantitative productivity of French suffixes -Able and -ité in corpora. We first analyze them independently and then when they belong to the same derivational chain (the suffix -ité chooses bases with -Able in about 15 % of its formations). The productivity of these suffixations is statisticallly estimated in relation to the corpus size. Both these affixes are productive in modern French : they continue to form new lexemes throughout the corpora and are not saturated, their productivity indexes show a stable evolution although this is dependent on applied calculations. Nevertheless, -ité is more frequent in the journalistic corpora except when it is applied to adjectives suffixed with -Able. Knowing that both -Able adjective and its -ité noun convey the same property, this finding indicates that the base complexity has to be taken into account in the formation of possible lexicon.", "morphology, newspaper corpus, -Able, -ité, quantitative morphological productivity" ]

[ "coordination", "syntaxe", "développement de grammaire", "grammaires de propriétés (GP)", "grammaires de construction (CxG)" ]

[ "P", "P", "M", "M", "M" ]

[ "", "We propose to treat coordination phenomena as syntagmatic accumulations, establishing a parataxis relation among their constituents. This consideration and its implications on description and analysis represent a departure from orthodox assumptions of formal linguistics about this question. We propose a description of the syntactic characteristics of coordination, then its formal representation and its integration into a grammar for french which is used for natural language processing. This strictly syntactic description is meant to be completed by information coming from other linguistic domains, which allows us to illustrate some notable specificities of our model.", "syntax, coordination, grammar development, property grammars (PG), construction grammars (CxG)" ]

[ "thesaurus", "système de question-réponse", "similarité" ]

[ "P", "R", "U" ]

[ "", "In order to improve question answering systems, the use of semantic knowledge is essential. However for French, such knowledge is not easily available. Indeed, French thesauruses are scarse and under-developed. In this paper, we firstly explain the reason why semantic knowledge should be used in a question answering system. Then, we present the EuroWordnet thesaurus, focussing on its limitations and the improvements we have made. We add undefined relationships with the help of the English base by using Wikipedia (2006).", "thesaurus, question answering system, similarity" ]

[ "formes dérivées", "dictionnaire électronique", "flexion" ]

[ "P", "P", "U" ]

[ "", "Our study concerns the automatic recognition of certain derived forms, i.e. the diminutives and the augmentatives of simple nouns and adjectives, as well as the comparatives and the superlatives of simple adjectives of Modern Greek. These forms are generally produced by the adjunction of a suffix to the corresponding standard form. We justify our choice to add them to the electronic dictionary of Modern Greek. Their processing implies an evolution of the representation of the dictionary which now uses a system of rules in order to easily generate the derived inflected forms, to label them as such, and to clearly put them in relation to their base form. We achieve better structuring of the lexical resources and flexible modular dictionary production.", "electronic dictionary, inflection, derived forms" ]

[ "anaphore", "saillance", "co-référence", "contexte", "perception visuelle" ]

[ "P", "P", "U", "U", "M" ]

[ "", "Anaphora resolution in human-machine dialogue systems is often based on models and algorithms that were designed for text. But dialogue happens in a situation, i.e., an immediate physical environment and events that are perceived simultaneously by the participants. This situation can be at the origin of referential expressions called “anaphora without linguistic antecedents”. Attributing some referents to such expressions is difficult for two reasons. First, because situational factors are numerous and implicit. Second, because ambiguities can appear between a potential situational antecedent and a potential linguistic antecedent. In this paper we propose a model that clarifies how situational factors work and how they can be taken into account when interpreting anaphoric expressions. By integrating the notion of salience that is common to both situational and linguistic aspects, we show with numeric scores how the interference between the situation and the language can be solved.", "coreference, anaphora, context, visual perception, salience" ]

[ "adaptabilité", "traduction probabiliste", "aspiration de bitextes", "mémoire de traduction" ]

[ "P", "P", "M", "M" ]

[ "", "We present our participation in the second evaluation campaign of CESTA, an EVALDA project within the framework of Technolangue. The goal of this task consisted in testing the adaptability of translation systems.We analyze the inadequacy of a statistical phrase-based system trained on legislative texts to translate medical corpora. We describe the experiments we conducted in order to adapt our engine to the new task.", "statistical translation, adaptation, Web crawling, memory-based translation" ]

[ "systèmes de question-réponse" ]

[ "P" ]

[ "", "Question-answering systems are usually composed of three main modules, namely, question analysis, document selection and answer extraction. In this paper, we focus on the third module, more specifically when the expected answer is not a named entity because this kind of answer is more difficult to extract. We describe how the parser Cass is employed to tag the answer in the candidate sentence. Then, we evaluate the results of this approach. Beforehand, we describe and evaluate the question analysis module, as the information it produces is used during our final answer extraction step.", "question answering systems" ]

[ "transducteurs", "dictionnaires électroniques", "automates à nombre fini d’états" ]

[ "P", "P", "R" ]

[ "", "In natural language processing, dictionaries usually associate additional information with lexical entries. The most effective representation of dictionaries makes use of finite-state machines – either automata (recognizers) or transducers. In this paper, we draw our inspiration from algorithms to directly build the minimal deterministic automaton and we propose a new form of a transducer. This new form outperforms existing transducers in terms of speed while computing outputs and in terms of size calculated on the basis of the number of transitions and different outputs, as shown in our experiments.", "finite state automata, transducers, electronic dictionaries" ]

[ "sémantique inférentielle", "raisonnement non monotone", "causalité" ]

[ "P", "U", "U" ]

[ "", "We are developing a system that aims to perform the same inferences as a human reader, on car-crash reports. More precisely, we expect it to determine the causes of the accident as they appear from the text. We describe the linguistic and semantic levels of analysis, and the inference rules used by the system.", "inferential semantics, non monotonic reasoning, causation" ]

[ "résolution des références", "dialogue humain", "évaluation quantitative" ]

[ "P", "P", "R" ]

[ "", "This article studies the resolution of references to entities in cases when a computational representation of the entities is available. Our data is a corpus of human dialogues about the front pages of one or more daily francophone newspapers. The main goal is to propose a method for the detection and resolution of references made by speakers to newspaper articles. A grammar is used to detect the nominal expressions referring to documents, while statistical methods are used to detect the pronouns referring to documents. An algorithm inspired from coreference resolution is used to solve all these expressions, i.e. to find the documents and articles they refer to. Specific quantitative metrics are applied to evaluate these proposals.", "reference resolution, human dialogue, quantitative evaluation" ]

[ "classes sémantiques", "nuances de sens", "acquisition sur corpus" ]

[ "P", "R", "R" ]

[ "", "The corpus-based acquisition of lexical semantic information has given rise to numerous studies on the automatic constitution of semantic classes, clustering words with similar meanings. However, the practical interest of these classes remains limited in the absence of knowledge about the nuances of meaning differentiating the words of a same class. We present a first system to make explicit such semantic nuances, in a semi-automatic way, using data from a text corpus, thus reaching a degree of word meaning definition, to our knowledge, never attained before by automatic means, This technique exploits large contexts around word occurrences to bring together pairs of words characterised by a similar meaning nuance. The limitation of this approach is that it only provides an implicit representation of the discovered distinctions : human interpretation is still required to name them. However, it enables an important level of automation, so that the human validation work can only introduce a limited bias in the results.", "semantic classes, nuances, corpus-based acquisition" ]

[ "groupe nominal prédicatif", "extraction d’information", "nominalisation", "marqueurs prépositionnels" ]

[ "P", "P", "P", "U" ]

[ "", "The identification of predicative structures is of great interest in information extraction. Although there is abundant literature on this subject, particularly in the genomic field, the majority relates to the relations around the verb. Few are interested in the relation which can link a nominalization and its actants in a noun phrase with predicative head (NPP). Our work involves firstly showing the complexity of different types of NPPs and the paraphrastic relations which link them with the verbal forms, followed by a unified view of the nomino-verbal predicative structures. We further show how we designed a link grammar allowing the identification of each actant in the NNPs. We describe the implementation with Link Parser for information extraction in scientific articles in the field of Biology.", "nominalization, predicative noun phrase, prepositional markers, information extraction" ]

[ "fouille d’erreurs", "analyse syntaxique" ]

[ "P", "P" ]

[ "", "We introduce an error mining technique for automatically detecting errors in resources used in parsing systems.We applied this technique on parsing results produced on several millions of words by two distinct parsing systems, which share a common syntactic lexicon and pre-parsing processing chain.We were thus able to identify errors and missing elements in the resources. In particular, by comparing both systems’ results, we were able to differentiate between problems stemming from shared resources and those resulting from grammars.", "parsing, error mining" ]

[ "lexique", "syntaxe", "TAL", "lexique-grammaire", "sous-catégorisation", "standardisation" ]

[ "P", "P", "P", "U", "U", "U" ]

[ "", "Based on existing lexical resources for NLP, in particular inflected form and subcategorization lexica, this paper proposes a unified and normalized representation, required for any further use of the data out of their original context. As a starting point for our model, we chose the Lexical Markup Framework, for three reasons. Firstly, it covers various layers of linguistic description including morphology, syntax and semantics. Secondly, it allows for combining extensional (i.e. lists of forms or constructions) and intensional (i.e. reference to paradigms) lexical descriptions. Thirdly, it makes use of externally defined data categories, ensuring linguistic soundness and, ultimately, compatibility with standardized corpus annotation formats.", "lexicon, NLP, syntax, lexicon-grammar, normalization, subcategorization" ]

[ "dialogue homme-machine", "reconnaissance automatique de la parole", "apprentissage automatique à base de corpus" ]

[ "P", "P", "R" ]

[ "", "Within the framework of the French evaluation program MEDIA on spoken dialogue systems, this paper presents the methods developed at the LIA lab for the robust extraction of basic conceptual constituents or concepts from an audio message. The conceptual decoding model proposed follows a stochastic paradigm and is directly integrated into the Automatic Speech Recognition (ASR) process. This approach allows us to both keep the probabilistic search space on sequences of words produced by the ASR module and project it to a probabilistic search space of sequences of concepts. The experiments carried on on the MEDIA corpus show that the performance reached by our approach is state of the art on manual transcriptions of dialogues. By partitioning the training corpus according to different sizes, one can measure the impact of the training corpus on the decoding performance, therefore estimate both the minimal and optimal number of dialogue examples required. Finally we detail how a priori knowledge can be integrated in our models in order to increase their coverage and therefore lowering, for the same level of performance, the amount of training corpus required.", "spoken dialogue, automatic speech recognition, corpus-based methods" ]

[ "SVM", "arbres de décision", "questions-réponses", "prédiction de la difficulté" ]

[ "P", "P", "P", "R" ]

[ "", "This paper presents two methods for automatically predicting the ability for a question answering system to automatically reply to a factoid question. The context of this prediction is the determination of the need to initiate a dialog with the user in order to focus or reformulate the question. The first method is an adaptation of a document retrieval prediction system based on SVM and decision trees. The features involved include question or document text, and cohesion measures between documents or extracts from which the answer is extracted. The second method uses only expected answer type to predict the answer validity. Both methods have been evaluated with data from the participation of our QA engine in the Technolangue EQUER campaign. On the one hand, the SVM based method leads to the best results. It correctly determines which are easy questions, namely, those to which our system gives the right answer, and which are hard questions, those to which our system gives bad or no answer. On the other hand, we show that for our system, the expected answer type (proper nouns, numbers, locations ...) is not a determining factor in defining question hardness.", "question-answering, difficulty prediction, SVM, decision trees" ]

[ "variation terminologique", "évolution terminologique", "corpus diachronique", "distance", "terme complexe" ]

[ "P", "P", "P", "P", "P" ]

[ "", "This paper presents a work about terminological evolution of dated terms, extracted from diachronic corpora of scientific or technical texts. Disappearance and appearances of terms attestations are the most obvious evidence of change. Nonetheless taking into account the variant forms of terms helps describe their historic development. A distance between complex terms allows terminological variation to be integrated into diachronic analysis. Some results and comments are given at the end of the paper.", "terminological evolution, diachronic corpus, complex term, terminological variation, distance" ]

[ "ressources lexico-sémantiques", "CAH", "dictionnaire sémique", "sémantique textuelle", "classification automatique", "Jaccard" ]

[ "P", "P", "R", "M", "M", "U" ]

[ "", "This paper describes an experiment of classification, based on a corpus of dictionary definitions. Underlying this research is the building of a lexico-semantic resource based on a structuralist approach to meaning. The semantic content of a lexical item is made up of semes; the meaning of a text emerges from groupings of stabilised seme sets. The purpose of the experiment is to make up semantic classes (or clusters) with dictionary definitions by using the HCA method. Semantic classes are built from lexical items according to generic semes in that they are shared by all lexical items of the class. They are differentiated according to their specific semes. From a selection of dictionary entries sharing generic seme /arbre/ (“tree”), the distribution and the organisation of an assumed semantic class linked to the domain of Forestry will be studied.", "lexico-semantic resources, seme dictionary, text semantics, clustering, HCA, Jaccard, UPGM" ]

[ "approche par contraintes", "analyse du discours", "formalisation du discours" ]

[ "P", "R", "M" ]

[ "", "We focus on the problem of discourse structure analysis from the point of view of its formal description and automatic processing.We intend to test the hypothesis that constraint-based approaches could enable various structures and cues of their presence to be taken into account. The CDML formalism which is here introduced allows such an approach.", "discourse analysis, discourse formalisation, constraint-based approach" ]

[ "phrase nominale arabe", "analyse syntaxique", "grammaire HPSG", "schéma adapté" ]

[ "P", "P", "P", "U" ]

[ "", "In this paper, we propose a parsing approach for Arabic nominal sentences using the Head-Driven Phrase Grammar formalism HPSG. To elaborate this approach, we begin by studying the typology of the Arabic nominal sentence specifying its different forms. Then, we elaborate a HPSG grammar for this type of sentence. The elaborated grammar respects the Arabic language specificity. Next, we present a parsing approach based on a bottom-up unification process. Finally, we illustrate the implementation and experimentation of the realised prototype.", "arabic nominal sentence, HPSG grammar, parsing" ]

[ "évaluation", "segmentation automatique", "accord interjuges" ]

[ "P", "P", "U" ]

[ "", "The objective of this research is to evaluate the efficacy of algorithms in identifying thematic breaks in texts. With this aim, 32 newspaper articles were segmented by groups of 15 judges. The analysis of their answers indicates that each judge, taken individually, is not very reliable, unlike the global index of segmentation derived from all the judges. If two algorithms tested are able to find the beginning of the articles when those are concatenated, they fail in the detection of the changes of topic perceived by the majority of the judges. It should however be noted that the judges, taken individually, are themselves largely ineffective in the identification of the changes of topic. In conclusion, we examine various explanations for the low level of performance observed.", "automatic segmentation, evaluation, interrater agreement" ]

[ "Grammaires de Propriétés", "indice de corrélation", "analyseur syntaxique", "modèle des patrons", "technique d’analyse hybride" ]

[ "P", "P", "M", "M", "R" ]

[ "", "Hybrid parsing techniques based both on statistical and symbolic methods remain rare. In general, they integrate statistical information into a context-free skeleton, in order to control the selection of rules and structures. We propose a statistical method which allows the evaluation of a correlation index between two linguistic objects, namely, category, and property. We describe how to integrate this statistical information into the framework of Property Grammars. The correlation index is used for controling both the selection process of category constituents and the evaluation of properties satisfaction.", "parsing, patterns model, correlation index, Property Grammars, hybrid parsing techniques" ]

[ "modélisation statistique du langage", "combinaison linéaire", "modèles distants" ]

[ "P", "P", "R" ]

[ "", "In the framework of statistical language modeling, we show that it is possible to use n-gram models with a history different to the one used during training. Our study deals with classical and distant n-gram models and is restricted to bigram models. We present four use cases for two bigram models : distant and non distant. By using the linear combination, we show an improvement of 14 % in terms of perplexity compared to the classic bigram model. Moreover, a study has been performed in order to emphasize the histories for which our models are efficient.", "statistical language modeling, distant models, linear combination" ]

[ "dictionnaire multilingue", "banque terminologique juridique", "édition de terme" ]

[ "R", "R", "M" ]

[ "", "This paper presents the particular use of « Jibiki » (Papillon’s Web server development platform) for the LexALP1 project. LexALP’s goal is to harmonize the terminology of the Alpine Convention’s2 four languages (French, German, Italian and Slovenian) so that member states are able to cooperate effectively. For this, the project uses the Jibiki platform in order to build a term bank used to compare the specialized terminology of seven different national legal systems in four different languages, and to harmonize it, optimizing the understanding between various alpine states in environmental matters at a supranational level. We describe how a generic platform like Jibiki is used to cope with a new kind of dictionary.", "multilingual dictionary, legal term bank, term editing" ]

[ "incompréhension", "dialogue homme-machine", "malentendu", "acte de dialogue", "but de dialogue", "stratégie de dialogue" ]

[ "P", "P", "U", "M", "R", "M" ]

[ "", "Miscommunication causes inconvenience within a spoken dialogue system. In this paper, we investigate this problem with regard to achieving genericity and robustness within dialogue management. Therefore, different error types occurring in a spoken dialogue system were first identified. Two markers were then proposed to detect and to trigger misunderstanding to the dialogue manager that should then choose an adequate strategy so as to generate an intelligent and suitable answer to the user. Moreover, several issues are investigated in order to process misinterpretations in the dialogue manager. An experiment performed using our approach to the spoken dialogue system Mélina shows promising results in processing miscommunication and proves the value of our approach.", "misunderstanding, misinterpretation, dialogue act, dialogue goal, dialogue strategy, humancomputer dialogue" ]

[ "ressources linguistiques", "multilinguisme", "architecture linguistique", "gestion de lexiques", "base de données linguistique" ]

[ "P", "P", "P", "M", "R" ]

[ "", "The proper functioning of Intuition, Sinequa’s platform for information retrieval, depends on the development and integration of a large number of linguistic resources. To improve their coherence and management, these resources which contain very diverse knowledge, need to be unified. As Intuition is available in most European languages, the multilingual factor complicates this unification. To overcome the problems caused by the structural differences among languages, a new linguistic architecture has been designed and modelled using UML. This meta-model is the starting point for a new database that will be the core component of a new workbench, centred on its user, the linguistic expert. This workbench will provide a centralised management of all the linguistic resources within Intuition.", "lexicon management, linguistic database, linguistic resources, multilinguism, linguistic architecture" ]

[ "poèmes arabes", "réseau de neurones", "syllabes", "classification", "perceptron multi couches", "analyse phonétique" ]

[ "P", "P", "P", "P", "M", "M" ]

[ "", "We have created a new system to recognize the different composition modes of Arab poetry. In this article we describe the different modules of this system. The use of an artificial learning technique for grouping a phonetic sequence of syllables is explained by the fact that we imitate the human process learning.", "artificial neural networks, multi layer perceptron, classification, arabic poetry, syllabes, phonetic analysis" ]

[ "annotation automatique", "relations de contrôle", "contrôle", "spécifications des besoins", "sémantique", "extraction d’information" ]

[ "P", "P", "P", "P", "P", "M" ]

[ "", "This paper presents a technique for the automatic extraction of control sentences from software requirements specifications (SRS). Our aim is to annotate action sentences from an SRS document, and to recognize if those actions are machine controlled or user controlled, in order to better understand the limits and responsibilities of a software system and to support the Requirements Engineering process. We present a linguistic analysis of control markers and a technique to automatically annotate control sentences by means of contextual exploration rules. These rules are implemented in the contextual exploration engine Excom and applied to a set of commercial SRS.", "semantic filtering, natural language requirements, requirements engineering, control" ]

[ "mémoire de traduction sous-phrastique", "détection du domaine de traduction", "traduction assistée par ordinateur", "récupération sensible au contexte" ]

[ "P", "P", "M", "M" ]

[ "", "In this article, we present a sub-sentential translation memory sensitive to the translation topic, a first step towards a full-fledged context-sensitive memory. This system is able to recycle previous translations indexed into the memory, not only for full sentences, but also for contiguous subsegments of these sentences, through word alignment information. Interesting segments are proposed to the translator. We also describe the creation of an artificial user (a simulator), necessary to test the system performances when no human intervention is possible, as is the case for these experiments. We test it when translating a set of disparate bilingual corpora. These performances are reflected in different metrics which we define. Finally, we show that a first attempt to automatically detect the translation context can be beneficial and promises to improve such a memory, by acting as a filter on the target material proposed to the user.", "computer assisted machine translation, sub-sentential translation memory, context-sensitive retrieval, translation topic detection" ]

[ "désambiguïsation morphologique", "analyse morphologique", "TALN arabe" ]

[ "P", "R", "M" ]

[ "", "In this paper we first propose a morphological disambiguation and analysis method of non voweled Arabic texts. This method could handle morphological ambiguities, caused by both the absence of the vowel marks and the irregularity of certain words (e.g. irregularity of the plural form for nouns and adjectives). Then, we present the MORPH2 system, a morphological analyser of non voweled Arabic texts, based on the proposed method. This system is evaluated on a school book and newspaper articles. The obtained results are very encouraging. Indeed, the global measures of recall and precision are 69,77 % and 68,51 % respectively.", "morphological analysis, morphological disambiguation, arabic NLP" ]

[ "représentations ontologiques", "ontologies", "Web sémantique", "analyse syntaxico-sémantique" ]

[ "P", "P", "P", "P" ]

[ "", "Since the conception of the semantic Web an important challenge faces natural language processing: i.e. making the existing textual contents of the « classic »Web available for ontology based applications, reasoning and finally for the semantic Web. Since the majority of Web sites contain textual information, the creation of ontological representations of these will be useful if not necessary. In our article we propose a method to automate this task by using domain ontologies and a deep syntactic-semantic analysis.", "syntactic and semantic analysis, ontologies, ontological representations, semantic Web" ]

[ "évolutivité", "notions de base", "thésaurus", "vecteurs conceptuels" ]

[ "P", "P", "P", "U" ]

[ "", "In the field of Natural Language Processing, in order to arrive at a thematic representation system of general knowledge, methods leaning on thesaurus have been widely used for about fifteen years. A thesaurus consists of a set of concepts defining a system generating a vector space to model general knowledge. These concepts, often organized as a hierarchy, constitute a fundamental, but fixed tool. When concepts evolve (as in technical fields), a thesaurus can evolve, but only as a result of an arduous process, requiring the collaboration of human experts. After detailing the desired characteristics of knowledge modelling systems, we define the « basic notions ». Their construction is initially based on the concepts of a thesaurus. They constitute another generating system of this vector space. We discuss the establishment of these basic notions, which naturally leads us to the question of their number. Lastly, we clarify how, being freed from the concepts of the thesaurus, the basic notions evolve progressively in an iterative process as new texts are being analysed.", "thesaurus, conceptual vectors, basic notions, evolution" ]

[ "document scientifique", "recherche d’information translangue", "corpus comparables", "typologie de documents", "catégorisation", "document vulgarisé" ]

[ "P", "M", "R", "M", "U", "R" ]

[ "", "In this paper, we aim to ripen the notion of corpora comparability. We study especially the distinction between scientific and popularized documents in the medical domain. We suppose that this distinction can give important informations, for instance in information retrieval. In the same time, we suppose that documents reflect the context of their production and provide features necessary for this distinction. We study and present several features, linguistic, typographic, lexical and other, for the characterization of medical documents as scientific or popularized. The results presented are acquired on data in Russian and Japanese. Some of analyzed features turn out to be relevant.We give then some remarks and suggestions as for the distinction of scientific and popularized documents and their theoretical issues.", "translingual information retrieval, comparable corpora, document typology, categorisation, scientific document, popularized document" ]

[ "lexicalisation", "grammaires", "syntaxe", "réseaux de transitions récursifs" ]

[ "P", "P", "P", "M" ]

[ "", "Shifting to a lexicalized grammar reduces the number of parsing errors and improves application results. However, such an operation affects a syntactic parser in all its aspects. One of our research objectives is to design a realistic model for grammar lexicalization. We carried out experiments for which we used a grammar with very simple content and formalism, and a very informative syntactic lexicon, namely the lexicon-grammar of French elaborated by the LADL. Lexicalization was performed by applying the parameterized-graph approach. Our results suggest that most information in the lexicon-grammar can be transferred into a grammar and exploited successfully for the syntactic parsing of sentences.", "grammar, syntax, lexicalization, recursive transition networks" ]

[ "clitiques", "analyseur syntaxique", "traitement multilingue" ]

[ "P", "P", "R" ]

[ "", "In this article we present an algorithm for the computational processing of clitic pronouns in Spanish and Modern Greek by the multilingual syntactic parser Fips that has been developed at the Laboratory For Language Analysis And Technology, LATL. We provide an overview of the distribution of clitic pronouns, their similarities and particular behaviour compared to general use. Then we present a common method of processing clitics for both languages. Finally, we show that the algorithm developed can be easily extended to other languages with clitics, when treated by Fips.", "syntactic parser, clitics, multilingual processing" ]

[ "proposition syntaxique", "CFG", "DCG", "Prolog", "analyse syntaxique partielle", "subordination" ]

[ "P", "P", "P", "P", "M", "U" ]

[ "", "In this paper we present SIGLé (Clauses Identification System with Light Grammar), which is a system recognizing French clauses. Sentences are divided into chunks and tagged by an external analyzer. The system then identifies the clauses, analyzes their relations and assigns them a tag indicating their syntactic nature. The system is characterized by its context-free grammar, proposing a set of tags for the \"qu-\" words adapted to our analysis, and by the use of the DCG formalism and the PROLOG language.", "partial syntactic analysis, syntactic clause, subordination, Prolog, CFG, DCG" ]

[ "corpus arboré", "coréen", "LTAG", "LTAG avec traits", "FB-LTAG", "grammaire d’arbre d’adjoint lexicalisée", "structure des traits", "extraction automatique d’une grammaire" ]

[ "P", "P", "P", "P", "P", "M", "M", "R" ]

[ "", "We present the implementation of a system which extracts not only lexicalized grammars but also feature-based lexicalized grammars from Sejong Treebank for Korean. We report on some practical experiments, in which we extract TAG grammars. Above all, full-scale syntactic tags and well-formed morphological analysis in Sejong Treebank allow us to extract syntactic features. In addition, we modify the Treebank to extract lexicalized grammars and convert them into tree schemata to resolve limited lexical coverage problems related to extracted lexicalized grammars.", "lexicalized tree adjoining grammar, LTAG, feature-based LTAG, feature structure, treebank, automatic grammar extraction, Korean" ]

[ "métaphores conceptuelles", "visualisation de corpus", "linguistique de corpus" ]

[ "P", "R", "M" ]

[ "", "This article presents the results of recent experiments achieved within the framework of a research project dedicated to the study of conceptual metaphors. These experiments comprise in a visual study of the distribution of three source domains of conceptual metaphors in a corpus of stock exchange articles. The three domains studied are meteorology, war and health, a large number of metaphorical uses of the lexicon of these three domains has been noted in the corpus. In order to visualize the distribution of these domains in the corpus, we use the ProxiDocs cartography tool and categorization. Maps of the corpus allow us to locate some conceptual metaphors in articles and sets of articles in the corpus. Articles containing non-metaphorical uses of studied domains are also differentiated on maps. Graphic representations of the corpus showing the temporal evolution of the three domains allow us to study the bond between some conceptual metaphors and the topic.", "conceptual metaphors, corpus visualization, corpus linguistics" ]

[ "corpus parallèle", "dictionnaire de terminologie bilingue", "traduction automatique statistique basée sur les segments" ]

[ "P", "P", "R" ]

[ "", "This paper describes an approach combining different statistical models for phrase-based machine translation. Different knowledge resources are used, such as two parallel corpora with different characteristics and a bilingual dictionary of terminology, in order to improve the qualitative and quantitative performance of the translation system. We evaluate our approach on the French-English language pair and show how combining the proposed resources significantly improves results.", "statistical phrase-based machine translation, parallel corpora, bilingual dictionary of terminology" ]

[ "coordination", "forêt de dérivation", "analyse syntaxique", "ellipses", "TAG", "forêt partagée" ]

[ "P", "P", "P", "P", "U", "M" ]

[ "", "In this paper, we introduce a generic approach to elliptic coordination modelization and parsing.We show that the erased lexical items can be replaced during parsing, by information gathered from the other member of the coordination, used as a guide at the derivation level. Moreover, we show how this approach can indeed be implemented as a light extension of the LTAG formalism through a fusion operation.We provide the derivation algorithms required to parse coordination constructions which can have an arbitrary number of elisions.", "parsing, TAG, coordination, ellipsis, shared forest, derivation forest" ]

[ "segmentation de textes", "classification spectrale continue", "identification de langue" ]

[ "P", "R", "M" ]

[ "", "We propose a theoretical framework within which information on the vocabulary of a given corpus can be inferred on the basis of statistical information gathered on that corpus. Inferences can be made on the categories of the words in the vocabulary, and on their syntactic properties within particular languages. Based on the same statistical data, it is possible to build matrices of syntagmatic similarity (bigram transition matrices) or paradigmatic similarity (probability for any pair of words to share common contexts). When clustered with respect to their syntagmatic similarity, words tend to group into sublanguage vocabularies, and when clustered with respect to their paradigmatic similarity, into syntactic or semantic classes. Experiments have explored the first of these two possibilities. Their results are interpreted in the frame of a Markov chain modelling of the corpus’ generative processe(s).We show that the results of a spectral analysis of the transition matrix can be interpreted as probability distributions of words within clusters. This method yields a soft clustering of the vocabulary into sublanguages which contribute to the generation of heterogeneous corpora. As an application, we show how multilingual texts can be visually segmented into linguistically homogeneous segments. Our method is specifically useful in the case of related languages which happen to be mixed in corpora.", "soft spectral clustering, text segmentation, language identification" ]

[ "adaptation", "modèles de langage", "utilisateur", "aide au handicap", "thème" ]

[ "P", "P", "P", "P", "U" ]

[ "", "Statistical language models (LM) are highly dependent on their training resources. This makes it not only difficult to interpret evaluation results, it also has a strong impact on users of a LM-based application. This question has already been studied by others. Focussing on a specific domain (text prediction in a communication aid for handicapped persons) we want to extend it to the influence of the language register. Considering corpora from five different registers, we discuss three methods to adapt a language model to its actual language resource hereby reducing the effect of training dependency: (a) a simple cache model augmenting the probability of the n last inserted words, (b) a dynamic user dictionary, keeping every unseen word and (c) an interpolated LM combining a base model with a currently updated user model. Our evaluation is based on the results obtained from a text prediction system working on a trigram LM.", "language model, dynamic adaptation, user, theme, AAC systems" ]