Datasets:

taln-ls2n

/

taln-archives

like 3

Follow

TALN research group at LS2N lab 14

[ "Conversation", "Acte de dialogue", "Analyse discursive", "Résolution de problèmes", "Schéma d'annotation" ]

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

[ "State of the Art : analysis of problem-oriented online written conversations in terms of dialog acts", "The advent of Web 2.0 and the massive creation and consumption of user generated content it triggered allowed for new kinds of interactions among web users. Indeed, freely available archives of problem-oriented online conversations represent an invaluable resource, however it remains largely unexploited. Exploiting this resource would not only allow for the improvement of systems related to collaborative problem resolution, but also to refine the customer support channels that web-based companies put at their users' disposal. However, in order to achieve this, it is first necessary to define a formal framework for the fine-grained analysis of online written conversations bearing requests for assistance. This paper presents an overview of the state of the art for discourse analysis in online written conversations, and discusses the applicability of the different methods for the analysis of conversations set within the scope of collaborative problem-solving.", "Discourse analysis, Conversation, Problem solving, Annotation scheme, Dialog act" ]

[ "segmentation", "mandarin", "Apprentissage non-supervisé", "chinois" ]

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

[ "Unsupervized Word Segmentation", "In this paper, we present an unsupervised segmentation system tested on Mandarine Chinese. Following Harris’s Hypothesis in Kempe (1999) and Tanaka-Ishii (2005) reformulation, we base our work on the Variation of Branching Entropy. We improve on (Jin et Tanaka-Ishii, 2006) by adding normalization and Viterbi-decoding. This enables us to remove most of the thresholds and parameters from their model and to reach near state-of-the-art results (Wang et al., 2011) with a simpler system. We provide evaluation on different corpora available from the Segmentation bake-off II (Emerson, 2005) and define a more precise topline for the task using cross-trained supervised system available off-the-shelf (Zhang et Clark, 2010; Zhao et Kit, 2008; Huang et Zhao, 2007)", "Unsupervized machine learning, segmentation, Mandarin Chinese" ]

[ "Texte", "auto-organisation", "lexique", "co-occurrence généralisée" ]

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

[ "Lexical Incrementation within Texts: a Self-Organization", "We propose here a dynamic study of lexicon: we describe how it is organized progressively from the beginning to the end of a given text. We focus on the “generalized co-occurrence”, forming a graph that represents all the lemmas of the text and their mutual co-occurrence relations. The study of a corpus of 40 texts shows that these relations have a self-organized evolution: the shape and the identity of the graph of cooccurrence become stable after a period of organization finished before the first half of the text. Then they no longer change: new words and new co-occurrence relations gradually take place in the network without changing its overall shape. We show that the evolution of the “generalized co-occurrence” is the quick construction of a system, which is then flexible enough to channel the flow of information without changing its identity.", "Text, lexicon, generalized co-occurrence, self-organization" ]

[ "Relations sémantiques", "Similarité sémantique", "Similarité distributionnelle" ]

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

[ "A Study of Heterogeneous Similarity Measures for Semantic Relation Extraction", "This paper evaluates a wide range of heterogeneous semantic similarity measures on the task of predicting semantic similarity scores and the task of predicting semantic relations that hold between two terms, and investigates ways to combine these measures. We present a large-scale benchmarking of 34 knowledge-, web-, corpus-, and definition-based similarity measures. The strengths and weaknesses of each approach regarding relation extraction are discussed. Finally, we describe and test two techniques for measure combination. These combined measures outperform all single measures, achieving a correlation of 0.887 and Precision(20) of 0.979.", "Semantic Similarity, Semantic Relations, Distributional Similarity" ]

[ "Système de traduction automatique", "mesure de confiance", "champs aléatoires conditionnels", "estimation de la confiance" ]

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

[ "Integrating Lexical, Syntactic and System-based Features to Improve Word Confidence Estimation in SMT", "Confidence Estimation at word level is the task of predicting the correct and incorrect words in the target sentence generated by a MT system. It helps to conclude the reliability of a given translation as well as to filter out sentences that are not good enough for post-editing. This paper investigates various types of features to circumvent this issue, including lexical, syntactic and system-based features. A method to set training label for each word in the hypothesis is also presented. A classifier based on conditional random fields (CRF) is employed to integrate features and determine the word’s appropriate label. We conducted our preliminary experiment with all features, tracked precision, recall and F-score and we compared with our baseline system. Experimental results of the full combination of all features yield the very encouraging precision, recall and F-score for Good label (F-score: 86.7%), and acceptable scores for Bad label (F-score: 36.8%).", "Machine translation, confidence measure, confidence estimation, conditional random fields" ]

[ "Traduction statistique", "normalisation de textos", "algorithme de recherche vorace", "modèle de langue" ]

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

[ "Applying a Statistical Machine Translation Algorithm to SMS Text Message Normalization", "We report on the application of a statistical machine translation algorithm to the problem of SMS text message normalization. The technique is based on a greedy search algorithm described in (Langlais et al., 2007). A first normalization is generated, then a function that generates new hypotheses is applied iteratively to a current best guess, while maximizing a scoring function. This method leads to a drop in word error rate of 33% on a held-out test set, and a BLEU score gain of over 30%. We focus on the methods of generating the initial normalization and the operations that allow us to generate new hypotheses.", "Machine translation, SMS, text message, normalization, greedy search algorithm, language model" ]

[ "Correction automatique", "détection de langue", "données produite par l’utilisateur" ]

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

[ "Towards Automatic Spell-Checking of Noisy Texts : General Architecture and Language Identification for Unknown Words", "This paper deals with the problem of spell checking on degraded-quality corpora such as blogs, review sites and social networks. We propose a first architecture of correction which aims at reducing overcorrection, and we describe its implementation. We also report and discuss the results obtained thanks to the module that detects whether an unknown word from a sentence in a known language belongs to this language or not.", "Spelling correction, language identification, User-Generated Content" ]

[ "Expressions polylexicales", "corpus", "lexicographie", "extraction lexicale", "lexique", "mesures d’association" ]

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

[ "An Open and Generic Framework for the Acquisition of Multiword Expressions", "In this paper, we present and evaluate an open and flexible methodological framework for the automatic acquisition of multiword expressions (MWEs) from monolingual textual corpora. We start with a pratical motivation followed by a theoretical discussion of the behaviour and of the challenges that MWEs pose for NLP applications. Afterwards, we describe the modules of our framework, the overall pipeline and the design choices of the tool implementing the framework. The evaluation of the framework was performed extrinsically based on an application : computerassisted lexicography. This application can benefit from MWE acquisition because the expressions acquired automatically from corpora can both speed up the creation and improve the quality and the coverage of the lexical resources. The promising results of previous and ongoing experiments encourage further investigation about the optimal way to integrate MWE treatment into NLP applications, and particularly into machine translation systems.", "Multiword expression, lexical extraction, lexicon, association measures, corpus, lexicography" ]

[ "néologie", "génération morphologique", "morphologie constructionnelle", "incomplétude lexicale" ]

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

[ "Prediction Device of Formal Neologisms : the Case of -IER Suffixed Nouns Denoting Artifacts", "We’ll introduce here a device that can predict neologisms using an example the automatical generation of nominal neologisms suffixed by -IER denoting artifacts (saladier, brassière, thonier). The aim of this article is double. We will first address the - IER suffixation constraints in order to take them into account on the implementation of our morphological generator. Second, we will describe our method to predict formal neologisms. Such a method will permit to automatically enrich NLP lexicons.", "constructional morphology, neology, morphological generation, lexical incompleteness" ]

[ "analyse syntaxique par transitions", "structure de dépendance non-projective", "grammaire catégorielle de dépendance" ]

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

[ "Beginnings of a Transition-Based Parsing for Non-Projectives Dependency Structures", "This paper presents an extension of the traditional transition-based dependency parser adapted to discontinuous dependencies and the first results of its training on a dependency tree corpus of French. The first experimental results will be useful for the choice of parsing configuration features well adapted to discontinuous dependencies in order to ameliorate learning of head dependencies.", "transition-based parsing, non-projective dependency structure, dependency categorial grammar" ]

[ "micro-syntaxe", "Arbres syntaxiques", "unité illocutoire", "unités rectionnelles", "macrosyntaxe", "entassement" ]

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

[ "Creating a Multi-Tree from a Tagged Text : Annotating Spoken French", "This study focuses on automatic analysis of annotated transcribed speech. The annotation system considered has been recently introduced to address the several limitations of classical syntactic annotations when faced to natural speech transcriptions. It introduces many different components such as embedding, piles, kernels, pre-kernels, discursive markers etc.. All those components are tightly coupled in a complex tree structure and can hardly be considered separately because of their close intrication. Hence, a joint analysis is required but no analysis tool to handle them all together was available yet. In this study, we introduce such an automatic parser of annotated transcriptions of speech and present the corresponding framework based on multi-trees. This framework permits to jointly handle separate aspects of speech such as macro and micro syntactic levels, which are traditionnaly considered separately. Several applications are proposed, including analysis of the transcribed speech by classical parsers designed for written language.", "Syntactic trees, illocutionary unit, microsyntax, macrosyntax, piles" ]

[ "modifieurs de polarité", "fouille d’opinion", "modifieurs de valence affective" ]

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

[ "Automatic Construction of a Contextual Valence Shifters Lexicon", "The research presented in this paper takes place in the field of Opinion Mining, which is mainly devoted to assigning a positive or negative label to a text or a sentence. The context of a highly polarized word plays an essential role, as it can modify its original polarity. The present work addresses this issue and focuses on the detection of polarity shifters. In a previous study, we have automatically extracted adverbs impacting the polarity of the adjectives they are associated to and qualified their influence. The extraction system has then been improved to automatically build a lexicon of contextual valence shifters. This lexicon contains lexico-syntactic patterns combined with the type of influence they have on the valence of the polarized item. The purpose of this paper is to show how the current system works and to present the evaluation of the created lexicon.", "opinion mining, contextual valence shifters" ]

[ "diacritisation", "analyse morphosyntaxique", "langue arabe", "traitement automatique" ]

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

[ "Apport of Diacritization in Arabic Morpho-Syntactic Analysis", "This work is concerned with the automatic morphological and syntactical analysis of the Arabic language. It focuses on diacritization and on its contribution to morphological analysis. Most of existing morphological analyzers and syntactical taggers do not take diacritics into account; as a consequence, they make mistakes that could have been avoided. In this paper, we propose a method which process diacritics. We show that doing so reduces considerably the morphological error rate, depending on the diacritics rate in the input text.", "diacritization, computer processing, morpho-syntaxic analysis, Arabic language" ]

[ "formation en ligne", "TAL", "EIAH", "socio-constructivisme", "acquisition des connaissances", "apprentissage collaboratif en ligne" ]

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

[ "Collaborative Knowledge Building Indicators Extraction in Distance Learning", "Natural Language Processing techniques are still not very much used within the field of Technology Enhanced Learning. They are restricted to specific tasks or domains such as CALL (standing for Computer Assisted Language Learning) in which they are ubiquitous but do not match every linguistic aspect they could process. We are seeking to explore possibilities or performances of thoses techniques for less specific systems including a network or community aspect. More precisely, our goal is to get indicators about collective knowledge building and its modalities. This paper presents the problem and the background of our thesis problem, as well as our motivation and our first reflections.", "NLP, TEL, distance learning, socio-constructivism, knowledge aquisition, collaboration, CSCL" ]

[ "désambiguïsation lexicale non-supervisée", "mesures de similarité sémantique à base de connaissances", "algorithmes globaux de propagation de mesures locales" ]

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

[ "State of the art : Local Semantic Similarity Measures and Global Algorithmes for Knowledge-based Word Sense Disambiguation", "We present the main methods for unsupervised knowledge-based word sense disambiguation. On the one hand, at the local level, we present semantic similarity measures, which attempt to quantify the semantic proximity between two word senses. On the other hand, at the global level, we present algorithms which use local semantic similarity measures to assign the appropriate senses to words depending on their context, at the scale of a text or of a corpus.", "unsupervised word sense disambiguation, knowledge-based semantic similarity measures, global algorithms for the propagation of local measures" ]

[ "sms", "essentialisation", "résumé automatique", "compression de texte", "lisibilité" ]

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

[ "Textual Compression Based on Rules Arising from a Corpus of Text Messages", "The present research seeks to reduce the size of text messages on the basis of compression techniques observed mostly in a corpus of sms. This paper explains the methodology followed to establish compression rules. It then presents the 33 considered rules, and illustrates the four suggested levels of compression with two practical examples, automatically generated by a first prototype. This research’s main purpose is not to produce \"sms-language\", but consists in designing a textual compression process able to generate short and understandable texts from any textual source in French. The term of \"essentialization\" is proposed to describe this approach of textual reduction.", "summarization, text compression, text messaging, readability, essentialization" ]

[ "étiquetage", "séquences verbales figées", "reconnaissance automatique", "transformations linguistiques", "ressources électroniques" ]

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

[ "For an Automatic Fixed Verbal Sequence Tagging: State of the Art and Transformational Approach", "This article presents a resource-based method aiming at automatically recognizing fixed verbal sequences in French (i.e casser sa pipe, briser la glace, prendre en compte) inside a text. This resource describes each sequence from the view-point of transformational possibilities and restrictions. Fixed sequences are not totally fixed and an exhaustive description is necessary to not only extract canonical forms. We will first describe some transformational approaches that are able to extract phraseological sequences. The building of the resource will be then addressed followed by our approach to automatically recognize fixed sequences in corpora.", "fixed verbal sequences, automatic recognition, tagging, linguistical transformations, electronic resources" ]

[ "émotion", "forum de santé", "traitement automatique de la langue", "désambiguïsation lexicale" ]

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

[ "Analysis of Emotion in Health Fora", "Studies about emotion in fora are numerous in Linguistics and Psychology. This contribution approaches this subject from an Information and Communication Sciences point of view, and studies emotion as a criteron of pertinence for patients in a health forum. This paper introduces the empirical step of automatic language processing in order to answer this question, and uses data processing on the corpus of forum messages, semi-supervised categorisation of messages and use of software NooJ for Natural Language Processing.", "emotion, health forum, automatic language processing, lexical disambiguation" ]

[ "peuplement d’ontologie", "extraction de relations", "représentation des connaissances" ]

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

[ "Population of an Ontology Modeling the Behavior of an Intelligent Environment Guided by Instance Relation Extractions", "We present an approach for ontology population, which aims at modeling the behavior of software components, for enabling a transition from natural language requirements to formal specifications. The ontology was designed based on the knowledge of the domotic domain and is initialized from a description of a physical configuration of an intelligent environment. Our method focuses on extracting relation instances which allows the extraction of concept instances linked by these relations. We built extraction rules using elements coming from syntactic analysis of user need descriptions, semantic and terminological resources linked to the knowledge contained in the ontology. Our approach for ontology population, distinguishes itself by its purpose, which is not to extract all instances describing a domain but to extract instances that can participate without any conflict to one of the mutiple operation decribed by users.", "relation extraction, ontology population, knowledge representation" ]

[ "questionnaire à choix multiple", "Agent conversationnel éducatif", "intelligence artificielle", "jeu sérieux", "système d'évaluation de réponses libres" ]

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

[ "Of the Use of Natural Dialogue to Hide MCQs in Serious Games", "A major weakness of serious games at the moment is that they often incorporate multiple choice questionnaires (MCQs). However, no study has demonstrated that MCQs can accurately assess the level of understanding of a learner. On the contrary, some studies have experimentally shown that allowing the learner to input a free-text answer in the program instead of just selecting one answer in an MCQ allows a much finer evaluation of the learner's skills. We therefore propose to design a conversational agent that can understand statements in natural language within a narrow semantic context corresponding to the area of competence on which we assess the learner. This feature is intended to allow a natural dialogue with the learner, especially in the context of serious games. Such interaction in natural language aims to hide the underlying MCQs. This paper presents our approach.", "Educational conversational agent, artificial intelligence, serious game, multiple-choice questionnaire, automatic assessment of free-text answer" ]

[ "Fouille d’opinion", "Classification", "Intensité de l’Opinion", "Résumé de texte d’opinion", "Popularité" ]

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

[ "System of Customer Review Summarization using Twitter and SentiWordNet", "As E-commerce is becoming more and more popular, the number of customer reviews raises rapidly. Opinions on the Web affect our choices and decisions. Thus, it is more efficient to automatically process a mixture of reviews and prepare to the customer the required information in an appropriate form. In this paper, we present ResTS, a new system of feature-based opinion summarization. Our approach aims to turn the customer reviews into scores that measure the customer satisfaction for a given product and its features. These scores are between 0 and 1 and can be used for decision making and then help users in their choices. We investigated opinions extracted from nouns, adjectives, verbs and adverbs contrary to previous research which use only adjectives. Experimental results show that our method performs comparably to classic feature-based summarization methods.", "Opinion mining, Sentiment Classification, Opinion Strength, Feature-based Opinion Summarization, Feature Buzz Summary" ]

[ "questions à réponses multiples", "question-réponse", "question liste" ]

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

[ "Typology of Multiple Answer Questions for a Question-answering System", "The evaluation campaigns of question-answering systems are generally based on the validity of an individual answer supported by a passage (for a factual question) or a group of answers coming all from a same supporting passage (for a list question). This framework does not allow the possibily to answer with a set of answers, nor with answers gathered from several documents. This cross-checking can be needed for building an answer composed of several elements in order to be as accurate as possible. Besides a large majority of questions with a singular form seems to be answered with a single answer whereas they can be satisfied with many. We present here a typology of questions with multiple answers and an overview of problems encountered by a question-answering system with this kind of questions.", "question-answering, multiple answer questions, list question" ]

[ "Reconnaissance d’entités nommées", "portabilité entre les langues", "approche symbolique" ]

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

[ "Adapting a French Named Entity Recognition System to English with Minimal Costs", "Cross-language portability of Named Entity Recognition systems requires linguistic expertise and needs human effort. Adapting symbolic systems suffers from the cost of developing new lexicons and updating grammar rules. Porting statistical systems on the other hand faces the problem of the high cost of annotation of new training corpus. This paper examines the cost of adapting a rule-based Named Entity Recognition system designed for well-formed text to another language. We present a low-cost method to adapt a French rule-based Named Entity Recognition system to English. We first solve the problem of lexicon adaptation to English by simply translating the French lexical resources. We then get to the task of grammar adaptation by slightly modifying the grammar rules. Experimental results are compared to a state-of-the-art English system.", "Named entity recognition, symbolic approache, cross-language portability" ]

[ "Langue des signes", "Espace de signation", "capture de mouvement", "gestes de pointage", "suivi du regard" ]

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

[ "Automatic Analysis of Discourse in Sign Language : Signing Space Representation and Processing", "In sign language, signing space is used to locate and refer to entities whose locations are important for understanding the meaning. In this paper, we propose a computer-based representation of the signing space and their associated functions. It aims to analyze manual and non-manual gestures, that contribute to locating and referencing signs, and to make real their effect. For that, we propose an approach based on the analysis of motion capture data of entities’ assignment and activation events in the signing space.", "Sign language, Signing space, pointing gestures, motion capture, gaze tracker" ]

[ "État de l’art", "Fouille d’opinion", "Multi-domaines", "Cross-domaines" ]

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

[ "State of the Art : Influence of Domain on Opinion Classification", "The interest in opinion mining has grown concurrently with blogs, forums, and others platforms where the internauts can freely write about their opinion on every topic. As the amounts of available data are increasingly huge, the use of automatic methods for opinion mining becomes imperative. However, sentiment is expressed differently in different domains : words distributions can indeed differ significantly. An effective global opinion classifier is therefore hard to develop. Moreover, a classifier trained on a source domain can’t be used without adaptation on a target domain. This article aims to describe the state-of-the-art methods used to solve this difficult task.", "State of the art, Opinion mining, Multi-domain, Cross-domain" ]

[ "Acquisition de relations", "Relations sémantiques", "Synonymie", "Terminologie", "Domaine Biomédical", "Corpus de spécialité" ]

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

[ "State of the Art on the Acquisition of Semantic Relations between Terms : Contextualisation of the Synonymy Relations", "Accessing to the context of specialised texts is a crucial but difficult task. It requires automatic or semi-automatic methods dedicated to the identification of semantic relations between terms appearing in the texts. NLP approaches for acquiring semantic relations between terms can be distinguished according to the type of information : the internal structure of the terms and the term context. In order to improve the quality of the acquired synonymy relations and their reusability in other corpora, we aim at taking into account the context into an approach based on the internal structure of the terms. We present the results of a preliminary experiment taking into account the use of the terms in a English biomedical corpora. This experiment will be helpful to add semantic constraints to the already acquired synonymy relations.", "Relation Acquisition, Synonymy, Semantic Relations, Terminology, Biomedical Domain, Specialised corpora" ]

[ "PCFG", "transducteur d’arbre", "algorithme CYK", "Extraction de grammaire", "grammaire de Lambek" ]

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

[ "PCFG Extraction and Pre-typed Sentences Analysis", "This article explains the way we extract a PCFG from the Paris VII treebank. Firslty, we need to transform the syntactic trees of the corpus into derivation trees. The transformation is done with a generalized tree transducer, a variation of the usual top-down tree transducers, and gives as result some derivation trees for an AB grammar. Secondely, we have to extract a PCFG from the derivation trees. For this, we assume that the derivation trees are representative of the grammar. The extracted grammar is used, via the CYK algorithm, for sentence analysis.", "Grammar Extraction, Lambek grammar, PCFG, tree transducer, CYK Algorithm" ]

[ "Expression polylexicale", "traduction automatique statistique", "alignement bilingue" ]

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

[ "Mining a Bilingual Lexicon of MultiWord Expressions : A Statistical Machine Translation Evaluation Perspective", "This paper describes a method aiming to construct a bilingual lexicon of MultiWord Expressions (MWES) from a French-English parallel corpus. We first extract monolingual MWES from each part of the parallel corpus. The second step consists in acquiring bilingual correspondences of MWEs. In order to assess the quality of the mined lexicon, a Statistical Machine Translation (SMT) task-based evaluation is conducted. We investigate the performance of three dynamic strategies and of one static strategy to integrate the mined bilingual MWES lexicon in a SMT system. Experimental results show that such a lexicon significantly improves the quality of translation.", "MultiWord expression, bilingual alignment, statistical machine translation" ]

[ "Mesures de comparabilité", "Évaluation", "Corpus comparables" ]

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

[ "Some variations on quantitative comparability measures and evaluations on synthetic French-English comparable corpora", "Following the pioneering work by (Li et Gaussier, 2010) we address in this paper the analysis of a family of quantitative measures of comparability dedicated to the construction or evaluation of comparable corpora. After recalling the definition of the comparability measure proposed by (Li et Gaussier, 2010), we develop some variants of this measure based primarily on the consideration of the occurrence frequency of lexical entries and the number of their translations. We compare the respective advantages and disadvantages of these variants in the context of an experiments based on the progressive degradation of the Europarl parallel corpus, by replacing blocks according to the methodology followed by (Li et Gaussier, 2010). The impact of the coverage of bilingual dictionaries on these measures is also discussed.", "Comparable corpora, Comparability measures, Evaluation" ]

[ "inférence grammaticale", "clustering hiérarchique", "grammaires catégorielles" ]

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

[ "Clustering for categorial grammar induction", "In this article, we describe the way we use hierarchical clustering to learn an AB grammar from partial derivation trees. We describe AB grammars and the derivation trees we use as input for the clustering, then the way we extract information from Treebanks for the clustering. The unification algorithm, based on the information extracted from our cluster, will be explained and the results discussed.", "categorial grammars, hierarchical clustering, grammatical inference" ]

[ "extraction", "expressions polylexicales", "locution verbale", "expressions figées", "classification de textes", "transformation" ]

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

[ "Enhance Multiword Expressions Extraction and Description with Transformational Rules", "This paper presents a methodology to extract and describe verbal multiword expressions using their transformational behavior. Several objectives are targeted: 1) automatically extracting MWE and especially frozen expression, 2) describing linguistically their MWE behavior, 3) comparing statistical methods and our approach, and finally 4) showing the importance of MWE in a text classification tool.", "multiword expression, verbal phrase, extraction, transformation, text classification" ]

[ "corpus multilingues", "comparabilité", "classification", "similarité textuelle translingue" ]

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

[ "Multilingual document clustering : state of the art", "Multilingual corpora are extensively exploited in several branches of natural language processing. This paper presents an overview of works in the automatic construction of such corpora. We address this topic by first providing an overview of different perceptions of comparability. We then examine the main approaches to similarity computation, construction and evaluation developed in the field. We notice that the measurement of the textual similarity is usually based on corpus statistics or the structure of ontological resources or on a combination of these two approaches. In a multilingual framework, with the use of a multilingual dictionary or a machine translator, many problems arise. The exploitation of a multilingual ontological ressource seems to be a worthy option. In clustering, the problem of adding documents to the initial base without affecting the quality of clusters remains open.", "multilingual corpora, comparability, crosslingual textual similarity, classification" ]

[ "TreeTagger", "étiquetage morphosyntaxique", "langue arabe", "ASVM 1.0", "TALN", "corpus d'apprentissage", "segmentation de l'arabe", "arbre de décision", "lexique", "jeux d’étiquette" ]

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

[ "Development of resources for training and the use of the tagger TreeTagger on Arabic", "In this paper, we present the steps of the development of resources for training and the use of a new tool for the part-of-speech tagging of Arabic. We implemented a tagging system based on TreeTagger, a generic stochastic tagging tool, very popular for its efficiency. First of all, we began by gathering a working corpus, large enough to ensure a general linguistic coverage. This corpus has been used to implement the tokenization process, as well as to train TreeTagger. We first present our method of tokenization, then we describe all the steps of the preprocessing and training process, using ASVM 1.0 to yield a raw POS tagging that was subsequently manually corrected. Finally, we implemented a straightforward evaluation of the outputs, both in a quantitative and qualitative way, on a small test corpus. Though restricted, this evaluation showed really encouraging results.", "NLP, Arabic language, training corpus, POS tagging, tokenization, decision tree, lexicon, tagsets, TreeTagger, ASVM 1.0" ]

[ "apprentissage supervisé", "Fouille d’opinions", "Machine à Vecteur de Support (SVM)", "combinaison des classifieurs" ]

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

[ "Polarity Opinion Detection in Arabic Forums by Fusing Multiple SVMs", "This article describes our contribution on the polarity’s detection of opinions in Arabian language by supervised training. Indeed the proposed system consists of three phases: the pretreatment of the corpus, the extraction of the features and the classification. For the second phase, we use twenty features of which the main are emotionalism, the reflexivity, the adressage and the polarity. The phase of classification represents in our work the combination of the several SVMs (Support Vector Machine),to solve the multi class problem. We analyzed the two strategies of the SVMs multi class that are: \"one against all\" and \"one against one\" in order to compare the results and to improve the performance of the global system.", "Opinion Mining, supervised training, Support Vector Machine (SVM), classifiers combination" ]

[ "état de l’art", "méthodes supervisées", "extraction de termes-clés", "méthodes non-supervisées" ]

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

[ "State of the Art of Automatic Keyphrase Extraction Methods", "This article presents the state of the art of the automatic keyphrase extraction methods. The aim of the automatic keyphrase extraction task is to extract the most representative terms of a document. Automatic keyphrase extraction methods can be divided into two categories : supervised methods and unsupervised methods. For supervised methods, the task is reduced to a binary classification where terms are classified as keyphrases or non keyphrases. This classification requires a learning step which is not required by unsupervised methods. The unsupervised methods use features extracted from the analysed document (sometimes a document collection) to check properties which allow keyphrase identification.", "keyphrase extraction, supervised methods, unsupervised methods, state of the art" ]

[ "Étiquetage syntaxique", "Analyse syntaxique en dépendances discontinues" ]

[ "P", "R" ]

[ "On the Effect of Head Tagging on Parsing Discontinuous Dependencies in French", "In this paper we want to show the strong impact of syntactic tagging on syntactic dependency parsing. The rules of categorial dependency grammar used to parse French deal with discontinuous dependencies and long distance syntactic relations. Such parsing method produces a substantial number of dependency structures and takes too much parsing time. We want to show that a local tagging method can reduce these problems and help to solve the global problem of dependency parsing disambiguation. Then we adapt a tagging method to types of the categorial dependency grammar. We obtain a dependency-head pre-selection allowing to reduce parsing ambiguity and to see that we can find distant relation of dependencies through local results of such method.", "Discontinuous Dependency Parsing, Syntactic Tagging" ]

[ "arabe", "entités nommées", "règles d’extraction", "ontologie", "Analyse linguistique", "fouille de textes", "extraction d’informations" ]

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

[ "A linguistic approach for knowledge extraction from an Arabic text", "We present in this paper a knowledge extraction system for Arabic. The information extraction is based on a deep morphosyntactic analysis. It also recognizes single words, idiomatic expressions, compounds and named entities. The analysis also identifies dependency relations, verb tenses and passive/active forms. Information extraction is application-independent and uses extraction rules that rely on the result of the morphosyntactic analysis. At this level, some named entity categories can be reconsidered. This extraction is based in our experimentations on the security ontology. The Resource Description Framework (RDF) obtained is then processed to gather information concerning a single event or named entity. The information extracted can help to understand the information contained in a set of texts, to infer knowledge into a knowledge base, or be used for monitoring tools.", "Linguistic analysis, Text Mining, Arabic, named entities, information extraction, extraction rules, ontology" ]

[ "lexique", "corpus", "phraséologie", "écrits scientifiques" ]

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

[ "Extraction of academic lexicon's simple words in humanities writings", "This paper presents a first extraction of academic lexicon's simple words in french academic writings in the fields of humanities and social sciences through a corpus study of research articles using morpho-syntactic analysis. This academic lexicon resource will be used for automatic indexing as a stoplist in order to exclude this lexicon from the terminology. We try various extraction methods and show that a first simple words lexicon can be generated but that multiwords expressions and words distribution should be taken into consideration to extract academic phraseology.", "corpus, scientific writings, lexicon, phraseology" ]

[ "chiasme", "figure de style", "rhétorique", "antimétabole" ]

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

[ "Towards an automatic identification of chiasmus of words", "This article summarises the study of the rhetorical figure “chiasmus” (e.g : “Quitters never win and winners never quit.”). We address the problem of its computational identification. How can a computer identify this automatically ? For this purpose this article will provide a formal description of the phenomenon. First, we put together an annotated text for testing our hypothesis. At the end we demonstrate that the use of stopword lists and the identification of the punctuation improve the precision of the results with very little impact on the recall. We discover also that using lemmatization improves the results but stemming doesn’t. Finally we see that a French thesaurus provided us with good results on the most elaborate form of chiasmus.", "chiasmus, rhetoric, antimetabole, stylistic device" ]

[ "Graphes d’Unités", "Dictionnaire Explicatif et Combinatoire", "Théorie Sens-Texte", "Représentation de Connaissances Linguistiques" ]

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

[ "ECD Knowledge Representation : Fundamental Concepts of the Unit Graphs Framework", "In this paper we are interested in the choice of a knowledge representation formalism that enables the representation, manipulation, query, and reasoning over linguistic knowledge of the Explanatory and Combinatorial Dictionary (ECD) of the Meaning-Text Theory. We show that neither the semantic web formalisms nor the Conceptual Graphs Formalism suit our needs, and justify the introduction of a new formalism denoted Unit Graphs. We introduce the core of this formalism which is the Unit Types hierarchy, and present Unit Graphs and how one may use them to represent aspects of the ECD.", "Linguistic Knowledge Representation, Meaning-Text Theory, Unit Graphs, Explanatory and Combinatorial Dictionary" ]

[ "réécriture de graphes", "évaluation de shéma d’annotations", "parsing", "analyse en syntaxe profonde" ]

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

[ "Towards a generic graph rewriting system to enrich syntactic structures", "This work aims to present a new approach for injecting deep dependencies (subject of control verbs, subject sharing in case of ellipsis, ...) into a surfacic and projective treebank. We use a graph rewriting system with constraint programming techniques for producing generic rules which can be easily applied to a treebank. Moreover, we are testing the genericity of our rules by using output of three different parsers to evaluate how the rules behave on predicted parse trees.", "graph rewriting system, annotation schemes evaluation, deep syntax parsing" ]

[ "induction de sens", "désambiguïsation lexicale", "langues peu dotées", "ressources langagières" ]

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

[ "A State of the Art of Word Sense Induction: A Way Towards Word Sense Disambiguation for Under-Resourced Languages", "Word Sense Disambiguation (WSD), the process of automatically identifying the meaning of a polysemous word in a sentence, is a fundamental task in Natural Language Processing (NLP). Progress in this approach to WSD opens up many promising developments in the field of NLP and its applications. Indeed, improvement over current performance levels could allow us to take a first step towards natural language understanding. Due to the lack of lexical resources it is sometimes difficult to perform WSD for under-resourced languages. This paper is an investigation on how to initiate research in WSD for under-resourced languages by applying Word Sense Induction (WSI) and suggests some interesting topics to focus on.", "Word Sense Disambiguation, Word Sense Induction, under-resourced languages, lexical resources" ]

[ "Dialecte Tunisien", "lexique ASM-DT", "TDT: Tunisian Dialect Translator" ]

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

[ "Generation of tunisian dialect corpora for adapting language models", "Lately, given the serious situation in the Arab world, the Arab dialects such as Tunisian dialect became increasingly used and represented in the interviews, news and debate programs. However, this situation presents negative consequences for Natural Language Processing (NLP): Since dialects are not officially written and have no orthographic standard, it is very costly to obtain adequate corpora to train NLP tools. Therefore, it does not even exist parallel corpora between Standard Arabic (MSA) and Tunisian Dialect(TD). In this work, we propose a method for the creation of a bilingual lexicon MSA-TD and an automatic process for generating dialectal corpora. These resources will be used to build a language model for Tunisian news, in order to integrate it into an Automatic Speech Recognition (ASR).", "Tunisian Dialect, MSA-TD lexicon, TDT: Tunisian Dialect Translator" ]

[ "détection automatique de sessions de recherche", "Recherche d’information", "analyse de journal de requêtes" ]

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

[ "Automatic search session detection exploiting results similarity from an external document collection", "Search engines users apply complex search behaviours such as query reformulation and multitasking search to satisfy their information needs. These search behaviours may be observed through query logs, and constitute clues allowing a better understanding of users’ needs. In this perspective, it is decisive to group queries related to the same information need into a unique search session. We propose an automatic session detection method exploiting the WIKIPEDIA documents collection, based on the similarity between the results returned for each query pair to estimate the similarity between queries. This method shows better performance than both temporal and lexical approaches traditionally used for successive session detection, and can be applied as well to multitasking search session detection. These experiments were conducted on a dataset originating from the OpenEdition Web portal.", "Information retrieval, automatic search session detection, query log analysis" ]

[ "Reconnaissance d'entités nommées", "Analyse syntaxique", "Reconnaissance de noms propres", "Traitement automatique du chinois", "Extraction d'information" ]

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

[ "A Mixed Morpho-Syntactic and Statistical Approach to Chinese Named Entity Recognition", "This paper presents a morpho-syntactic and statistical approach for Chinese named entity recognition which is a part of an automatic system for information extraction. The process is divided into three steps : first, the generation of possible proper nouns is based on morphological rules; second a language model is used to select the best result, and last, a local syntactic parsing performs the named entity recognition. Syntactic parsing makes named entity recognition more relevant and more complete.", "Proper noun recogition, Named entity recognition (NER), Chinese Natural Language Processing, Information extraction, Syntactic parsing" ]

[ "interface syntaxe sémantique", "FrameNet", "Analyse sémantique automatique" ]

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

[ "Probabilistic modeling of the syntax semantic interface using probabilistic context free grammars, Experiments with FrameNet", "This paper presents a generative method for predicting the frame semantic structure of a sentence from its syntactic structure and describes the grammars used with their performances. This system allows to predict, for a word in the syntactic context of a sentence, the most probable frame. The generative system allows to give a frame to a word and semantic roles to a set of pathes. Although results are not yet satisfying, this parser allows to group semantic parsing tasks (frame selection, role fillers selection, role assignment) unlike previously published works. In addition, it offers a new approach to parse semantic frames insofar as it is based more on syntactic structure rather than words of the sentence.", "Automatic Semantic Parsing, syntax semantic parsing, FrameNet" ]

[ "FrameNet", "construction à attribut de l’objet", "français", "divergence syntaxe-sémantique" ]

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

[ "Addressing object predicative complements in a French FrameNet", "Within the ASFALDA project, which includes the production of a French FrameNet, we try to provide a linguistically motivated treatment for a typical example of syntax-semantics mismatch : object complement construction. In order to do so, we first give an overview of syntactic and semantic properties of object complement constructions. Next, we study the way FrameNet deals with English verbs taking part in those constructions, and finally take a stance for a homogenized treatment of the construction within the French FrameNet.", "FrameNet, French, object complement construction, syntax-semantics mismatch" ]

[ "Besoin informationnel", "tâches de recherche d’informations", "Langage naturel", "Recherche informations", "Expression et interprétation des besoins", "Formulation question", "comportement utilisateur" ]

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

[ "Distinctions in the Formulation of Information Needs in Natural Language; Construction of a Task-Based User Profile Database", "With the massive and heterogeneous web document collections, IR system must analyze the behaviors of users which are unpredictable and varied. The approach described in this paper provides a description of the verbalizations of the information need in natural language. For this, we used data collected (i.e. users’ complaints in natural language) through a search engine dedicated to economic reports in French over 5 consecutive years totaling a corpus of 1398 natural language requests. Then, we compared the expression as such of the information need according to the IR task.", "Information Retrieval, Information Need, Query formulation and Query Expression, Query Formulation, Natural Language, User Behavior, IR task" ]

[ "Verbe terminologique ou spécialisé", "sémantique des cadres", "sémantique lexicale", "structure argumentale", "étiquetage en rôles sémantiques" ]

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

[ "The descriptive approaches of the verb in Linguistics, Terminology and NLP", "As part of our research project, which aims to implement a text simplification tool for the specialized usages of verbs in medical corpora using the syntactic and semantic analysis of these verbs in context, we propose an overview of some approaches and work whose main research object is the description of verbs, within the three research areas which interface our study is : linguistics, terminology and NLP. We pay a particular attention to studies that can have an impact on our work. This state of the art allows us to better understand the theoretical framework related to our research project. Moreover, it allows us to have benchmarks and references that might be usefull for the realization of our project.", "Specialized verb, Frame Semantics, lexical semantics, argumental structure, Semantic Role Labeling" ]

[ "Réseau de Neurones Profond", "Étiquetage morpho-syntaxique", "Plongements", "TALN", "Apprentissage Automatique" ]

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

[ "Deep Neural Network applied to Part-of-Speech Tagging", "Syntactic and semantic parsing of non-canonical languages is mainly restricted by the lack of labelled data sets. It is thus essential to develop strong systems capable of combining canonical and non-canonical text corpora. Deep Learning methods proved their efficiency in domains such as imaging or acoustic process.We propose neural network architecture applied to natural languages processing. Furthermore, instead of extracting from the sentence a rich set of hand-crafted features wich are the fed to a standard classification algorithm, we drew our inspiration from recent papers about the automatic extraction of word embeddings from large unlabelled data sets. On such embeddings, we expect to benefit from linearity and compositionality properties to improve our system performances.", "NLP, Part-of-Speech Tagging, Machine Learning, Deep Neural Network, Embeddings" ]

[ "extraction terminologique", "ressources endogènes", "apprentissage automatique" ]

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

[ "Terminology Extraction: Towards Resource Minimization", "The article presents the method which aims to minimize the use of external resources for the terminology extraction task and to make this task less langage dependent. For that purpose, the method builds simplified morphological and morphosyntactic resources directly from a lemmatized corpus. These endogenous resources are used both in filters, which refine the statistical calculations, and in patterns for polylexical terms identification. The method was tested on two comparable corpora in chemistry and in telecommunication in French and in English. The precision observed on the first 100 monolexical terms fluctuates between 71% and 87% for French and between 44% and 69% in English ; for polylexical terms the precision was 69-78% in French and 69-85% in English depending on the domain.", "terminology extraction, endogenous resources, machine learning" ]

[ "grammaire", "temporalité", "LSF", "modélisation LSF" ]

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

[ "Description of structures of time (in French sign \nlanguage) based on a formal grammar", "Temporality constitutes a major issue in filed of modeling french signed language (LSF). In fact, it is very difficult to model actual discriptions of the aspect-temporal systems of LSF. In this paper we present the bases of a novel formal grammar that permits the modeling of the LSF. This paper presents a study to construct this grammar. We analysed a French SL corpus to create formel rool between the signed gesture and its signification. Our objective is to obtain rules of grammar that can generate a synthesis project using a signer avatar.", "grammar, LSF, temporality, modeling LSF" ]

[ "assistants interactifs", "apprentissage artificiel", "systèmes de question-réponse" ]

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

[ "Natural Human-Machine Interaction for Manipulating Formal Language: Bootstrapping with Mapping", "We consider the problem of mapping natural language written utterances expressing operational instructions to formal language expressions, applied to French and the R programming language. Designing a learning operational assistant, which is our long term goal, requires the means to train and evaluate it, that is, a baseline system able to interact with the user. After presenting the guidelines of our work, we propose a model to represent the problem and discuss the fit of direct mapping methods to our task. Finally, we show that, while not resulting in excellent scores, a simple approach seems to be sufficient to bootstrap an interactive learning system.", "interactive assistants, machine learning, question answering systems" ]

[ "dérivation lexicale", "raisonnement analogique", "analogie", "graphe lexical", "composantes connexes" ]

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

[ "Merging structures of lexical derivations by analogical reasoning", "This paper presents a method for merging structures of lexical derivations by analogical reasoning. Following the presentation of general features of a lexical graph from the French Lexical Network, we focus on the weak connected components of this graph. This components are grouped together in three steps : by isomorphism, by relational similarity and finally by attributional similarity. The results of the last merging are analyzed in detail.", "lexical graph, analogy, connected components, analogical reasoning, lexical derivation" ]

[ "corpus DECODA", "MMR", "ROUGE", "Résumé de conversations parlées", "résumé par extraction" ]

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

[ "Extraction methods for automatic summarization of spoken conversations from call centers", "In this paper we speak about automatic spoken conversation summaries. We use conversation from some real cases call from a call center extracted from the DECODA corpus. We test some extractive summary methods used in text summary (MMR) and some dialogue heuristics methods. It’s mainly to select the longest speaker turn in different part of the dialogue, the first quarter, the whole dialogue, and the last quarter of the dialogue. All the results are evaluated thanks the ROUGE software. The results show the limits of these classical approaches and suggest that we need some abstractive methods including some structural features of the conversation. In fact, these results show that the structural heuristics based methods are even or better than the classic method like MMR.", "spoken conversation summarization, extractive summary, ROUGE, DECODA corpus, MMR" ]

[ "Contextes Riches en Connaissances", "terminologie", "traduction terminologique", "CRC", "identification de définitions", "identification d’exemples", "énoncé définitoire" ]

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

[ "Knowledge-Rich Contexts Extraction from Comparable Corpora", "Some contexts provide only a simple explanation of a given term even if they contain it. However, others contain fragments of definitions, mention a terminological variant or use other concepts to make it easy the term understanding. In this work we focus on « definitory contexts » that would be valuable to a human for knowledge acquisition from texts, mainly in order to assist in terminological translation from comparable corpora. Indeed, provide the term with an example, makes it possible to understand its exact meaning. First, we specify three hypothesis defining the concept of a defnitory example. Then we evaluate its validity through a method based on the knowledge-Rich Contexts (KRCs) and hierarchical relationships between terms.", "Knowledge-Rich Contexts, KRCs, mining definitions, minging examples, terminology, terminological translation" ]

[ "grammaires de propriétés", "Treebanks", "langue arabe", "grammaire hors-contexte" ]

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

[ "Induction of a variable granularity property grammar from the Arabic Treebank ATB", "This paper presents an approach for building an Arabic property grammar using the treebank ATB. This approach consists in two main steps: (1) inducing a context-free grammar from a treebank and (2) inducing a property grammar. So, we acquire first a context-free grammar (CFG) from the source treebank and then, we induce the property grammar by generating automatically existing relations between grammatical units described in the CFG. The result is a new resource for Arabic, opening the way to new tools and descriptions.", "Treebanks, Arabic language, context-free grammar, property grammars" ]

[ "ontologies", "modélisation logique", "commande en langue naturelle", "analyse structurelle de surface" ]

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

[ "", "In this paper, we focus on natural language interaction for artificial agents. Our architecture relies on a command logical model to enhance the semantic interpretation. It allows us to catch the « functional structure » of the user sentence, i.e. each terms compared to each others. This paper describes a partial structural approach which relies on the agent ontology to build a logical form of the sentence. We first define an algorithm to anchor a word from the command in the ontology and we use it to make our partial analysis. Lastly, we explain briefly how to use our model for the semantic interpretation of the user command.", "natural language command, partial structural analysis, logical form, ontologies" ]

[ "ALAO", "TAL", "morphologie flexionnelle", "détection d’erreurs", "rétroaction(s)" ]

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

[ "", "We present an empirical approach to the automated evaluation of learner's answers in a CALL system (Computer Assisted Language Learning). We suggest the realization of an error parsing module using NLP techniques (Natural Language Processing). The errors stem from a language learners corpus. This article describes the implementation, from an existing linguistic model, of an inflectional context-free morphology parser.", "NLP, CALL, errors detection, inflectional morphology, feedback" ]

[ "définition terminographique", "sous-langage", "annotation automatique", "indices morphosyntaxiques", "repérage de frontière" ]

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

[ "", "This article presents a procedure to locate and tag the generic elements of terminographic definitions, taking advantage of the formal characteristics of the definition sublanguage. This four step procedure is part of a larger (semi-)automatic parser of the conceptual structure of terminographic definitions, intended to ease the tagging of a corpus for studying conceptual regularities in definition structure. The method involves the development of an automatic tagging system, based on the identification of morphosyntactic boundary markers, which does not require the use of additional linguistic resources.", "terminographic definition, automatic tagging, boundary location, morphosyntactic markers, sublanguage" ]

[ "extraction de paraphrases", "WordNet", "mesure de similarité", "fusion d’articles", "distance sémantique", "identification d’hyperonyme", "Wikipedia", "entités nommées", "analyse syntaxique", "désambiguïsation lexicale", "cadres de sous-catégorisation", "apprentissage" ]

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

[ "", "We describe here how to automatically import encyclopedic articles into WordNet. This process makes it possible to create new entries, attached to their appropriate hypernym. In addition, the preexisting entries of WordNet can get enriched with complementary descriptions. Reiterating this process on several encyclopedias makes it possible to constitute a corpus of comparable articles; we can then automatically extract paraphrases from the couples of articles that have been created. The paraphrases components can finally be disambiguated, by means of a similarity measure (using the verbs WordNet hierarchy).", "paraphrases extraction, articles merging, similarity measure, semantic distance, hypernym identification, WordNet, Wikipedia, named entities, syntactic analysis, word sense disambiguation, syntactic frames, unsupervised learning" ]

[ "DiCo", "Fonctions lexicales (non standard)", "modélisation des relations sémanticolexicales" ]

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

[ "", "In the lexicographical products developped within the framework of the Explicative and Combinatorial Lexicology such as the DiCo, the lexico-semantic links are modeled by means of lexical functions. However, only a part of them (called standard) happen to appear as a real formal encoding. The others (called non-standard), which represent links more specific to some lexical units, are written in a heterogeneous and barely formalized way. Therefore, some relations can't be taken into account in automatic processings. We propose, in this paper, a methodology for the normalization of non standard lexical functions in order to make them machine readable in applications such as text-analysis and generation. To complete this work, we discuss some theoretical assumptions drawn upon this formalism and sketch some propositions for a global and homogeneous processing of all the lexical links described in the DiCo.", "(non standard) Lexical Function, modelling of lexico-semantic links, DiCo" ]

[ "référence", "saillance", "communication homme machine multimodale" ]

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

[ "", "We are interested about multimodal systems that use the following modes and modalities : speech (and natural language) as input as well as output, gesture as input and visual as output through displaying on the screen. The user exchanges with the system by a gesture and/or an oral statement in natural language. This exchange, encoded on the different modalities, contains the goal of the user and also the designation of objects (referents) necessary to the realization of this goal. The system must identify in a precise and non-ambiguous way the objects designated by the user. In this paper, our main concern is the oral designations, without gesture, of objects in the visual context. Indeed, the whole of the multimodale context including visual mode, influences the production of the user input. In order to identify a designation based on the visual context, we propose an algorithm which uses « traditional » linguistic knowledge, knowledge about manipulated objects and perceptive aspects (degree of salience) associated to these objects.", "multimodal human computer communication, reference, salience" ]

[ "accès lexical", "profil d’utilisateur", "réseaux sémantiques" ]

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

[ "", "This paper describes a joint and comparative evaluation of 5 lexical resources (descriptive, paradigmatic and syntagmatic) from a lexical access angle, with the further perspective of constructing a tool based on a combination of these resources to avoid the \"tip of the tongue\" (TOT) phenomenon. This phenomenon characterises a person who has difficulty in saying or writing an intended word but one who is able to produce semantically associated words. The evaluation corpus is composed of 20 TOT examples each linked to 50 users’ association sets of 5 semantically associated words. The results highlight that all the tested resources are complementary. Moreover, 79% of proposed association sets contain at least one association leading to the TOT through its relative words in at least one resource (in the worse case the TOT has to be found among 2500 words). Finally the results show variations between users which could increase performance thanks to user profiles.", "semantic networks, lexical access, user profiling" ]

[ "extraction d’information", "biographies", "analyse de réseaux sociaux", "entités nommées", "représentation de connaissances" ]

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

[ "", "Information Extraction systems are widely used to create flat databases of templates filled with the data they extract from text. In this article we describe an ongoing research project that focuses on the use of automatically extracted data to create a more complex representation structure. This structure is a model of the social network underlying the relations that can be established between the entities of a corpus of biographies.", "information extraction, social network analysis, named entities, biographies, knowledge representation" ]

[ "ressource prédicative", "extraction d’information", "patrons lexico-syntaxiques" ]

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

[ "", "In this article, we present a method aiming at building a resource for an information extraction task, from an already existing French predicative lexical resource. We point out the weaknesses and strengthnesses of two predicative resources we worked with : Les tables du LADL and Volem. We present why we select Volem as the most interesting resource for the task. Thereafter, we make a list of the needs an information extraction task implies, and how we include missing information in the resource we selected. We evaluate the resource completed by those missing informations, using it in an information extraction task.", "predicative resource, information extraction, lexico-syntactic patterns" ]

[ "corpus de requêtes d’assistance", "agent conversationnel", "activité conversationnelle", "actes de dialogue" ]

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

[ "", "In order to conceive a conversational agent able to assist ordinary people using softwares, we have built up a specific corpus of assistance requests in french, and studied its characteristics. We show here that assistance requests can be clearly distinguished from the ones from other available corpora in related domains. We also show that this corpus isn’t homogenous, but on the contrary reflects various conversational activities, among which the assistance itself. Those observations allow us to discuss about the opportunity to consider assistance as a general language particular registre.", "corpus of assistance requests, conversational agent, conversational activity, speech acts" ]

[ "alignement multilingue", "corpus parrallèles", "multitextes", "multidocuments", "extraction de structures", "alignement endogène" ]

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

[ "", "For many reasons, the multilingual corporas have interested various communities. Among these corporas, the parallel texts are used as well in terminology, lexicography or as a source of informations for example-based translations. The European Union, which involved the production of legislative documents, generates these parrallel texts in more than twenty languages. Also, with the Web as a vector of diffusion, we can wonder if these parallel texts can be treated as documents. This article describes a alignment system taking account a great number of languages (> 2) and the structural characteristics of the analyzed documents.", "multilingual alignment, parallel corpora, multitexts, multidocuments, extraction of structures, endogenous alignment" ]

[ "évaluation", "système de questions-réponses", "focus", "recherche d’information" ]

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

[ "", "Evaluation campaigns take into account only the final results obtained by information retrieval systems. Our perspective is that of the glass box evaluation of a question-answering system. The processing of a question is accomplished by a series of components. The purpose of this article is to study the element in the sentence which holds the key information. This element is to be found again in the sentence containing the answer next to the answer itself, and is called the focus. We will begin by defining this concept. We will then applied it to the QALC question answering system. Finally we will demonstrate the pertinence of using glass box evaluations to enhance the global performance such systems.", "question answering system, information retrieval, evaluation, focus" ]

[ "segmentation", "segments d'information évolutive", "algorithme TextTiling" ]

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

[ "", "Our hypothesis is that the TextTiling's boundaries can be considered as clues we can use with other linguistic features to automatically detect evolving information segments. This work is developed as part of an industrial project aiming to automatically detect textual zones containing potentially evolving information.", "evolving information, segmentation, TextTiling algorithm" ]

[ "annotation", "disfluences", "corpus oraux", "prosodie", "XML" ]

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

[ "", "Disfluencies (repeats, word-fragments, self-repairs, aborted constructs, etc.) inherent in any spontaneous speech production constitute a real difficulty in terms of annotation. Indeed, the annotation of these phenomena seems not easily automatizable, because their study needs an interpretative judgement. In this paper, we present a methodology for the annotation of disfluencies (also named “production phenomena”) which frequently occur in speech corpora. Constituting such data allows not only to represent some relevant aspects of speech productions (so as to be a basis for observations and comparisons with other data), but also to improve automatic speech processing (particularly for parsing).", "speech corpora, annotation, disfluencies, prosody, XML" ]

[ "architecture modulaire", "portabilité", "génération", "dialogue", "XML" ]

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

[ "", "Natural language generation for human-computer dialogue imposes specific constraints, such as the spontaneous and fragmented character of the utterances, speaker types or constraints related to the system’s time of response. In this context, the issue of a thouroughly specified architecture stems naturally, with respect to the processing stages in the modules involved and to the interfaces between these modules as well. In order to allow for a quasi-total freedom concerning the theoretical principles driving the processing stages, such an architecture must be modular (i.e., allowing the independence of the modules of each other) and portable (i.e., allowing a certain interoperability between modules designed following this architecture and existing modules, designed following standard, reference architectures, such as the RAGS model). Thus, in this paper firstly the proposed architecture will be presented in a concise manner, comparing it then to the RAGS model and arguing for the design choices being made. Afterwards, the portability of the architecture will be described, via an extended example whose general character resides in the fact that a set of rules are obtained, that allow automatic translations between representation formats in our architecture and in the RAGS model, in both ways. Finally, a set of conclusions and pointers to further work end up the paper.", "generation, dialogue, modular architecture, portability, XML" ]

[ "résolution anaphorique", "anaphore pronominales", "anaphores zéro", "analyse dynamique", "analyse de discours", "grammaires de propriétés", "grammaire fonctionnelle", "discours oral" ]

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

[ "", "We present an anaphora resolution integrated in a discourse analysis. This study deals with pronoun anaphora and zero anaphora. Our analysis is based on three approaches. A constraint based rule analysis, a functional approach and a dynamic analysis. In order to evaluate the feasibility of our approach and its reliability we have experimented with a corpus of 97 speech stories produced by children. We present here the results of our evaluation experiment.", "discourse analysis, anaphora resolution, pronoun anaphora, zero anaphora, property grammars, functional grammar, dynamic analysis, speech discourse" ]

[ "Apprentissage automatique", "co-training", "classification" ]

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

[ "", "In the domain of supervised and semi-supervised classification, this paper describes an experimental context suitable with statistical classification. It shows an alternative method usable when learning data is unsufficient but when many unlabeled data is avaliable : the multi-classifier co-training. Two classifiers based on different classification methods replace the two independent views of the original co-training algorithm : icsiboost based on boosting and LIBLINEAR which is a logistic regression classifier.", "Machine learning, classification, co-training" ]

[ "Cohésion lexicale", "segmentation thématique", "analyse distributionnelle", "ressources lexicales" ]

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

[ "", "The present work takes place within the Voiladis project (Lexical neighborhood for discourse analysis), whose purpose is to exploit lexical cohesion markers in the study of various discursive phenomena. We want to show the relevance of a distribution-based lexical resource to locate interesting relations between lexical items in a text.We call neighbors lexical items that share a significant number of syntactic contexts in a given corpus. In order to evaluate the usefulness of such a resource, we address the task of topical segmentation of text, which generally makes use of some kind of lexical relations. We discuss here the importance of the particular resource used for the task of text segmentation. Using a system inspired by (Hearst, 1997), we show that lexical neighbors provide better results than a classical resource.", "Lexical cohesion, lexical resources, distributional analysis, text segmentation" ]

[ "extraction de lexique", "corpus spécialisé", "chinois", "unité lexicale", "lexie" ]

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

[ "", "Building lexical resources is a vital task in improving the efficiency of information retrieval systems. This article introduces a Chinese lexical unit extraction method for untagged specialized corpora. This method is based on an incremental process driven by an association score. This work features an unsupervised statistically aided linguistic approach. The extraction results — evaluated on a random sample of the working corpus — show decent precision and recall which amount respectively to 52.6% and 53.7%.", "specialized corpus, lexical unit, lexicon extraction, Chinese" ]

[ "induction de sens", "espace sémantique", "réduction de dimensions", "Locality Sensitive Hashing", "clustering de mots", "objets multi-représentés" ]

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

[ "", "Words can have many senses. In order to process information correctly, a computer should be able to decide which sense of a word is used in a given context. This unsolved problem has generated much research in word sense disambiguation and in the generation of semantic spaces in order to separate possible meanings. Here, we adapt two existing methods to automatically distinguish words uses and senses.We apply them to multiple semantic spaces produced by a syntactic analysis of a very large number of web pages. These adaptations and the results presented in this article differ from the original methods in that they use a combination of several high dimensional spaces instead of one single representation. Each of these competing semantic spaces takes part in a clustering phase in which they vote on sense induction.", "semantic space, dimensionality reduction, Locality Sensitive Hashing, Word Sense Induction, words clustering, multi-represented data" ]

[ "méta-moteur de traduction", "traduction automatique", "web", "modèle de langage" ]

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

[ "", "Considering the Web and multilingual documents development expansion, the need of fast translation has become an evidence. This paper presents a system that proposes, for a given sentence, a list of N translation hypotheses instead of a single translation, using several machine translation systems already existing. Nine free and available (on the Internet) automatic translation engines have been chosen to submit a text to be translated and to receive its translation. The translations obtained are evaluated individually with a language model adapted and a metric elaborated by us, and in this way classified by relevance order. The experiment have pointed out that this meta-translation engine is more useful than the use of one system for translation.", "automatic translation, web, language model, meta-translator" ]

[ "régression logistique", "bagging", "boosting", "lisibilité", "modèle de langue" ]

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

[ "", "Reading is known to be an essential task in language learning, but finding the appropriate text for every learner is far from easy. In this context, automatic procedures can support the teacher’s work. Some works on English reveal that it is possible to assess the readability of texts using statistical classifiers. In this paper, we present an original approach comparing various classification techniques, namely logistic regression, bagging and boosting on two training corpora. The results show a slight superiority for multinomial logistic regression over bagging or boosting.", "readability, logistic regression, bagging, boosting, language model" ]

[ "normalisation", "entités nommées", "traitement de l'information", "système complexe", "méthodes endogènes", "analyse de corpus" ]

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

[ "", "Normalization is involved in many fields of information processing. It improves performances for several applications, such as information retrieval or information extraction, and makes linguistic resources constitution more reliable. Normalization consists in standardizing each variant of a term or named entity into a unique form, and this way restricts the impact of term variation. Our work applies to named entity normalization, for which we implemented a complex system that mixes several approaches. We present here one of its components: an endogenous method to mark out and validate the normalized named entities. Moreover, we place the user in the center of our normalization process, in order to obtain fully reliable data that fit his needs.", "normalization, named entities, information processing, corpus analysis, endogenous methods, complex system" ]

[ "Etiquetage d’entités nommées", "ressources sémantiques" ]

[ "M", "M" ]

[ "", "In this paper, we present a way to use of Wikipedia as an external resource to disambiguate and detect named entities, without learning step. We explain how we build our system and why we used five linguistic editions of the Wikipedia corpus to increase the volume of potentially matching candidates. We finally experiment our system on a news corpus.", "Named entity labeling, semantic resources" ]

[ "Distance intertextuelle", "classification", "arabe", "statistique lexicale", "corpus", "lemmatisation" ]

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

[ "", "Our researche works are interested in the application of the intertextual distance theory on the Arabic language as a tool for the classification of texts. This theory handles the classification of texts according to criteria of lexical statistics, and it is based on the lexical connection approach. Our objective is to integrate this theory as a tool of classification of texts in Arabic language. It requires the integration of a metrics for the classification of texts using a database of lemmatized and identified corpus which can be considered as a literature reference for times, genres, literary themes and authors and this in order to permit the classification of anonymous texts.", "Intertextual distance, Arabic, classification, lemmatization, corpus, lexical statistics" ]

[ "Argumentation", "Insinuation", "Norme coutumières", "Justification" ]

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

[ "", "Strategic means and forms for the generation of textual descriptions of a same reality are numerous. Most definitions of the argumentation given in the literature share the idea that to argument we must provide elements in favor of a given conclusion. Unfortunately, in our task consisting in generating biased descriptions of a road crash, we do not have only elements in favor of the desired conclusion, but also elements that are against it and must nevertheless be present in the description unless we cannot be able to understand how the accident happened. To remedy this problem, we proposed a module in our system for generating biased descriptions that handles the task of presenting better the undesirable elements using argumentative techniques.", "Argumentation, Insinuation, Customary norms, Justification" ]

[ "systèmes de questions-réponses", "Recherche d’information", "Traitement Automatique des Langues" ]

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

[ "", "This paper describes an evaluation and an analysis of the results of an answers reranking method for a question-answering system. Candidate answers to a question are reordered by computing a tree transform cost. We discuss the results of our evaluation on the Clef 2004-2005 corpus and describe possible improvements to the system.", "Information Retrieval, Natural Language Processing, Question-Answering systems" ]

[ "TimeML", "Annotation temporelle", "repérage des événements" ]

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

[ "", "The processing of temporal information is crucial for the understanding of natural language texts. The specification language TimeML was developed to facilitate the identification and normalization of temporal expressions and events in texts written in English. The aim of the various TimeML projects was to formulate an annotation scheme able to be applied to free text, such as that which is found on the Web, for example. Recently, efforts have been made to apply TimeML to languages other than English, namely Chinese, Korean, Italian, Spanish and German. Some efforts have been made in this direction with respect to French, but they remain somewhat scattered. In this paper, we detail our ongoing work, which aims to establish comprehensive resources for the annotation of French texts according to TimeML - an annotation guide, a Gold Standard corpus and modules for automatic annotation.", "Temporal annotation, event recognition, TimeML" ]

[ "morphosyntaxe", "système de question-réponse", "dialogue homme-machine", "Grammaire Interactive", "modèles dynamiques", "expérimentation utilisateur", "formulation de réponse" ]

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

[ "", "In this paper we study the interrogation in Human-Computer Dialogue in order to integrate interpretative and dynamic functionnalities in a Question-Answering system using Interactive Grammar. It constitues a direct link between question and answer and assumes that the morphosyntax of interrogation depends on an anticipated answer. We propose an observation of question-answering corpora coming from the RITEL system showing the distribution of different morphosyntatic kinds of questions. Then we present an experimentation based on this observation in which French native speakers answer to those different kinds of questions. Our objective is to observe the difference between the anticipated answer and the effective one.", "Interactive Grammar, dynamic models, human-computer dialogue, question-answering system, morphosyntax, user experimentation, answer formulation" ]

[ "annotation", "entités nommées", "grammaires locales", "base de connaissances", "Nooj" ]

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

[ "", "The aim of this paper is to describe our approach for annotating of the referential mentions that refer to the entities which are the instances of Person. Our method is inspired by the recent work in information extraction and particularly the named entities recognition and classification task. Local grammars are used to identify this category of named entities and to generate an extra-linguistic knowledge base which is further used for the process of reference resolution.", "named entities, entity recongnition, local grammars, Nooj, knowledge base" ]

[ "Japonais", "Analyse Syntaxique", "Grammaire" ]

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

[ "", "We describe here the way the Japanese sentence is structured, with its minimal content, the structure of its components, the different kinds of complex sentences and the possible modal effects. The aim of this description is to analyse the Japanese sentence in a universal way while respecting the language properties. The syntactic parser FIPS is now being developed to analyse Japanese sentences following grammar rules coming from this syntactic description. Even though the parser used to be used for Western languages only, the first results of simple Japanese sentence analysis are very positive. This shows how FIPS can adapt easily to very different languages.", "Syntactic Parser, Japanese language, Grammar" ]

[ "Compréhension du langage", "langue parlée spontanée" ]

[ "M", "U" ]

[ "", "The EmotiRob project, financed by ANR, aims at realizing a robot companion for weakened children. The project decomposes into two under parts that which are the module of understanding to include what says the child and a module of emotional interaction to bring an answer by feigning feelings by the movements of the body, lines of the face and by emission of small simple sounds. The module of understanding reuses the works of the system Logus. The main difficulty is to develop the existing system of a human-machine dialogue finalized towards a wider domain and to detect the emotional state of the child. At first, we present the EmotiRob project and its specificities. Then, the system of understanding Logus, on which bases itself this work is presented. Finally, we present the adaptations of the system to his new task EmotiRob.", "spoken language understanding, spontaneous language" ]

[ "Identification d'opinions", "Fouille de textes", "Traitement automatique du langage naturel" ]

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

[ "", "We present a model for the representation of opinions, by analyzing the elements which compose them and some properties. The model has an operating counterpart, implemented in the form of a set of tags. For the automatic application of these tags on Spanish texts, we work on the writing of contextual rules. A primary subset of rules was written for the identification of some elements of the model. We analyze the first results of their application.", "Opinions Identification, Text Mining, Natural Language Processing" ]

[ "Induction lexicale", "transducteurs stochastiques", "langues apparentées" ]

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

[ "", "We apply different models of graphemic similarity to the task of bilingual lexicon induction between a Swiss German dialect and Standard German. We compare stochastic transducers using sliding windows from 1 to 3 letters, trained with the Expectation-Maximisation algorithm on training corpora of different sizes. While the unigram transducers provide good results with very small corpora, we show that bigram transducers outperform them with corpora of 750 word pairs or more. Overall, the trained models show between 7% and 15% F-measure improvement over Levenshtein distance.", "Lexicon induction, stochastic transducers, cognate languages" ]

[ "génération intégrée localisée", "architectures de génération", "SDRT", "segmentation du discours" ]

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

[ "", "In this paper, we propose a localized integrated approach for generation. In this approach, the integrated handling of linguistic decisions is limited to the production of propositions whose decisions concerning their generation are dependant. The generation is performed by groups of propositions of limited size with an integrated handling of linguistic decisions that concern the production of the propositions that belong to the same group. Our approach provides a solution for the computational complexity problem of classical integrated generation. Therefore, it provides an alternative to separated generation (sequential and interactive) that has many drawbacks but is widely implemented in today’s generation systems.", "localized integrated generation, generation architectures, SDRT, discourse segmentation" ]

[ "dictionnaire", "génération automatique", "conditions de structure morphématique" ]

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

[ "", "The objective of this article is the presentation of an automatic system of generation of dictionaries electronics of Arabic basing itself on the conditions of morphemic structure and the matrices lexical. This work returns within the framework of two projects MIRTO and OREILLODULE under development to laboratories LIDILEM of Grenoble and UTIC of Tunisia.", "automatic system of generation, conditions of morphemic structure, matrices lexical" ]

[ "citations", "contruction et étude de corpus", "genre journalistique" ]

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

[ "", "In the plagiarism detection context, finding citations and their components is essential as it may help estimating legal value of a copy (with or without original source specified). We propose here a quantitative and qualitative study of citations we extracted from a corpus we previously built. This study aims at orienting our research towards an efficient automatic citations extraction method.", "citations, corpus creation and analysis, journalistic genre" ]

[ "Question réponse enchainée" ]

[ "R" ]

[ "", "We present works realized in the field of questions answering systems(SQR) using chained questions. The search for documents in a SQR is disrupted by the absence of information on the value to be granted to the possibly useful elements of text in search of information which appear in bound questions. The recent campaigns of evaluation show that this problem is under estimated, and did not make the work of dedicated technique. To improve the search for documents in a SQR we study a new method to organize the information bound to the interactions between questions. This one is based on the operation of a structure of data adapted to the transmission of the information of bound questions up to the search engine.", "chained question answering" ]

[ "correction grammaticale", "unification", "syntagme" ]

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

[ "", "Free software users community is sorely lacking French grammar checking. With the aim of filling this gap, we have worked on the adaptation to French of a tool originally developped for a foreign language. Thanks to this work, we could show that classic natural language processing approaches used in grammar checking are not suitable. To remedy it, we suggest an evolution of grammar checkers that includes linguistic principles such as chunking and unification. Despite its efficiency, this evolution is not sufficient to get a good French grammr checker. We are then thinking of a new approach of the problem.", "grammar checking, chunk, unification" ]

[ "informations spatio-temporelles", "pages Web", "annotation", "extraction d'information", "tourisme" ]

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

[ "", "This paper presents a project for the detection, extraction and annotation of temporal and spatial information and of tourism objects in order to fill the knowledge base of a tourism Web portal. We focus on the differences that exist between extraction from structured documents and extraction from Web pages. First, the different types of information to extract are presented. We then discuss methods for linking these pieces of information together – for example relating the name of a restaurant to its opening hours – and how to annotate the extracted data. The program we have developed to perform the extraction and annotation, as well as an evaluation of this program, are presented here. Finally, we focus on the semiotic marks which appear on the Web and show they also prove useful in interpreting Web pages.", "information extraction, annotation, spatial & temporal information, tourism, Web pages" ]

[ "Corpus", "Expression régulière", "Concordancier", "TAL", "Parser" ]

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

[ "", "This paper describes a tool that enables complex queries on tagged corpora, and free formatting of the results. The formalism used is based on regular expressions, which are wellknown from most corpus linguists. As opposed to other software, the free formatting of the results enables re-use of the query results by additional tools, and proves useful for a wide range of applications well beyond that of simple concordance programs.", "Corpora, Concordancer, NLP, Parser, Regular expression" ]

[ "filtrage d'information", "analyse partielle", "grammaires locales", "TALN", "lexique-grammaire" ]

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

[ "", "We present a partial analysis approach to the problem of information filtering, based on the results of different areas of research, ranging from information retrieval to natural language processing. We lay the emphasis on the particular constraints of the information filtering activity compatible with a linguistic and user-friendly treatment (partial analysis). We also present some results measured on an actual corpus in order to illustrate the potential of the described approach.", "information filtering, NLP, partial analysis, local grammars, lexicon-grammar" ]

[ "Automates finis", "grammaires context-free", "graphes", "forme normale de Greibach" ]

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

[ "", "The amount of documents available over the Internet is exploding. This phenomenon requires the development of new electronic representations and tools to search information into these documents. This article deals with context-free algorithms applied to finite state graphs (strictly finite or with cycles). It shows new methods and representations to combine efficiently complexities in terms of memory space and processing time.", "Context-Free Grammars, Finite State Automata, Graphs, Greibach Normal Form" ]