Datasets:

taln-ls2n

/

taln-archives

like 3

Follow

TALN research group at LS2N lab 11

[ "modèle probabiliste", "contexte pertinent", "analyse sémantique", "extraction automatique", "information mutuelle moyenne" ]

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

[ "", "This work is part of a larger research project entitled « Oréodule » aiming to develop tools for automatic speech recognition, translation, and synthesis for the Arabic language. The core of our interest in this work is in improving the probabilistic model on which our semantic decoder rests. To achieve this goal, we tested the influence of the pertinent context use, and of the contextual data integration of different types, on the effectiveness of the semantic decoder. The results are satisfactory.", "semantic analysis, probabilistic model, automatic extraction, pertinent context, overage mutual information" ]

[ "Analogie", "traduction automatique", "proportion", "chaîness de symboles", "divergences entre langues" ]

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

[ "", "We position ourselves in the current trend of Natural Language Processing, i.e., corpus-based approach and “least effort” approach.We shall inquire how far it is possible to go without any preprocessing of raw data. The theoretical question in the background is: which operations are fundamental in language? Proportional analogy has been mentionned by many grammarians and linguists. We shall inspect the efficiency of such an operation by testing it against a difficult task of NLP: machine translation. We shall also show some good properties brought by the formalisation of such an operation with theoretical results in formal language theory and the adequacy of analogy with the description of natural languages. To summarize, we shall illustrate an operation that is fundamental in language in its theoretical aspects as well as in its practical peformance.", "Analogy, proportion, strings of symbols, machine translation,divergences across languages" ]

[ "français", "syntaxe", "Corpus arboré", "corpus journalistique" ]

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

[ "", "This paper presents the first linguistic results exploiting a new treebank for French (Abeillé et al., 2001). The corpus comprises 1 million words fully annotated and disambiguated for parts of speech, inflectional morphology, compounds and lemmas, and partially annotated with syntactic constituents. It is made of extracts from newspapers ranging from 1989 to 1993 and written by different authors, and covers a variety of subjects (economy, literature, politics, etc.). After explaining how this corpus was built, and how it can be used with a specific search tool, we present some linguistic results obtained when searching the corpus for lexical or syntactic frequencies and preferences, and explain why we think some of these results are relevant both for theoretical linguistics and psycholinguistics.", "Treebank, French, Newspaper corpora, syntax" ]

[ "Graphes Conceptuels", "Schèmes Sémantico-Cognitifs", "Temps linguistique", "Valeurs Aspectuelles" ]

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

[ "", "We propose here a system which deals with time, aspect and verbal meanings in natural language processing within Simple Conceptual Graphs on the basis of the Semantico-Cognitive Schemes and the aspecto-temporal theory both introduced by J-P. Desclés. Our work bears on French texts. A text is represented by two different structures both represented within the SCG model. The first one models the semantico-cognitive representation while the second one is the temporal diagram representing the temporal constraints between the situations described in the text. Linking these structures leads us to slightly extend the original SCG model.", "Linguistic time, Aspectual values, Semantico-Cognitive Schemes, Conceptual graphs" ]

[ "Banque d’arbres", "éditeur d’arbres", "analogie", "filtrage tolérant", "mesure de l’effort" ]

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

[ "", "The construction of a treebank is a Very cumbersome and time-comsuming process. To speed up this process, we see the process of building a treebank as a sequence of edition and search operations. Our purpose is to assess the effort, measured by operations (cliks and keystrokes), needed to augment a treebank. We have proposed a tool, Boardedit, which incorporates a tree editor and parsing aids. However, the effort needed will depend on the quality of the answer of the parsing aids, this can be seen as a measure of the quality of the parsing aids. Our tree editor is essential for our tools during this experiment, parsing aids will be always used with the tree editor. In the experiment, we augment a tree-bank of 5 000 sentences with 1 553 new sentences. We show that the reduction in the number of operations is more than 4/5 of the effort.", "Treebank, analogy, approximate matching, tree editor, tree-banking assessment" ]

[ "XML", "Ressources linguistiques", "TAG" ]

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

[ "", "This paper presents the ATOLL workbench for Tree Adjoining Grammars. This workbench provides several tools and resources based on the use of the markup language XML which eases the construction and the exchange of linguistic resources.", "" ]

[ "Représentation Vectorielle", "Champs de Markov", "Désambiguïsation", "Sémantique Distributionnelle", "Recherche Documentaire", "algorithme EM" ]

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

[ "", "The present contribution focuses on the integration of word senses in a vector representation of texts, using a probabilistic model. The vector representation under consideration is the DSIR model, that extends the standard Vector Space (VS) model by taking into account both occurrences and co-occurrences of words. The integration of word senses into the co-occurrence model is done using a Markov Random Field model with hidden variables, using semantic information derived from synonymy relations extracted from a synonym dictionary.", "Word Sense Disambiguation, Distributional Semantics, Vector Space Representation, Information Retrieval, Markov Random Fields, EM algorithm" ]

[ "n-grams", "multilinguisme", "classification numérique de textes" ]

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

[ "", "Real multilingual numerical classification is impossible if only words are treated as the privileged unit of information. Although it makes tokenisation (in which words are considered as tokens) relatively easy in English or French, it makes it much more difficult for other languages such as German or Arabic. Moreover, lemmatisation, typically used to normalise and reduce the size of the lexicon, poses another challenge. The notion of n-grams which, for the last ten years, seems to have produced good results both in language identification and speech analysis, has recently become a privileged research axis in several areas of knowledge acquisition and extraction from text. In this paper, we present a text classification tool based on n-grams of characters and evaluate its results and compare them with those obtained from a different classification tool based solely on the processing of words.", "" ]

[ "classes d'objets", "modèle relationnel", "interrogation de BD en langage naturel" ]

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

[ "", "We investigate how to use natural language to query a database from both the linguistic and database points of view (but without AI considerations). In order to achieve this goal, we need a natural language model which we can map on to a relational database model. We have chosen to use the word classification called « classes d'objets » as proposed by the Laboratoire de Linguistique Informatique (LLI). We present here the first results of this work.", "natural language database query, relational model, « classes d'objets »" ]

[ "prosodie", "corpus", "étiquetage" ]

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

[ "", "The manual transcription of prosody is an extremely time-consuming activity, which requires highly specialised experts, and is prone to errors and subjectivity. Full automation is not achievable in the current state of the technology, but we present in this paper a technique that automates critical steps in the process, which results in a substantial annotation time reduction, and improves the objectivity and coherence of the annotation. In addition, the necessary human phases do not require a highly specific training in phonetics, and can be achieved by syntax students and corpus workers.", "corpus, prosody, tagging" ]

[ "entité nommée", "variante terminologique", "Système de question-réponse", "recherche d'information" ]

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

[ "", "We developed a system, QALC, that participated to the Question Answering track of the TREC evaluation conference. QALC exploits an analysis of documents, selected by a search engine, based on the search for multi-words terms and their variations both to select a minimal number of documents to be processed and to give indices for comparing question and sentence representations. This comparison also takes advantage of a question analysis module and a recognition of numeric and named entities in the documents.", "" ]

[ "cohérence thématique", "Cadre thématique", "exploration contextuelle" ]

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

[ "", "To improve the results of automatic summarization or semantic filtering systems concerning thematic coherence, we propose a model which combines a statistic analysis system identifying thematic breaks and a linguistic analysis system identifying discourse frames.", "" ]

[ "liaison", "lecture", "spontané", "débit de parole", "longueur", "fréquence lexicale" ]

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

[ "", "Various studies on liaison phenomena in French have shown the influence of several linguistics as well as para-linguistics factors on liaison realization. In this study we aim at determining the relative contribution of certain of these factors by using a robust methodology and tools used in automatic language processing. In a 5 hours speech corpus, produced by 10 speakers, we study the effect of speech style (oral reading/spontaneous speech), speech rate, as well as the contribution of syntactic and lexical (word length and frequency) factors on liaison realization. Results show that even if several factors can contribute to predict some liaisons, these factors are often interdependant and do not allow a sufficient prediction of liaison realization.", "" ]

[ "Transducteur", "noms propres", "extraction de motifs" ]

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

[ "", "This article describes a fmite-state cascade for proper nouns extraction in texts. After a preprocessing (division of the text in sentences, tagging with dictionaries, etc.), a series of finite state transducers cascade is applied one after the other to the text and locate left and right contexts which indicate presence of a person name. An evaluation on a journalistic corpus (Le Monde) gives a rate of precision of 98,7% for a rate of recall of 91,9%.", "Transducer, proper nouns, pattern extraction" ]

[ "arbres stratifiés ordonnés", "Extraction de motifs", "distances d'édition", "séquences" ]

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

[ "", "This paper presents a new algorithm designed to detect recurrent syntactical patterns in natural language texts. It first describes the pattern extraction algorithm which is based on an edit model generalized to Stratified Ordered Trees (SOT). Then it focuses on experiments with french classical literature of the 18th and 19th century. One section is dedicated to the efficiency evaluation. The last provides some examples of such recurrent patterns that are typical of an 18th century author, Madame de Lafayette.", "Pattern extraction, stratified ordered trees, edit distance, sequences" ]

[ "Compréhension Automatique de la Parole", "robustesse", "Communication Homme-Machine", "analyse syntaxique partielle", "grammaires de dépendances" ]

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

[ "", "This paper presents a spoken french understanding system which aims at providing a detailed linguistic analysis as well as preserving the robustness of standard methods. The first part focusses on the importance of a richer analysis for applications that are not dedicated to a very restricted task. We then present our two-level architecture system : a robust independant-domain shallow parsing step followed by a task based analysis of the whole structure of the utterance.", "Man Machine Dialogue, speech understanding, robustness, shallow parsing, dependency grammars" ]

[ "synonymie", "Structuration de terminologie", "aide à la validation" ]

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

[ "", "The terminology building process cannot be fully automatized. Various terminological tools have been designed but human expert knoxledge is still required. This paper shows how SynoTerm, our system for the discovery of synonymy relations between terms, deals with human knowledge. The challenge for terminological tools is to complete the time-consuming tasks under human control and to combine the algorithmic results with human specific information.", "" ]

[ "analogie", "morphologie dérivationnelle", "structure du lexique" ]

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

[ "", "We present a method to extract morphological links from a synonym dictionary without supervision. The analysis of these links brings somes aspects of the morphological structure of the lexicon to light. We have then taken advantage of this structure to identify allomorphic variations of the extracted suffixations.", "derivational morphology, analogy, structure of the lexicon" ]

[ "Synonymie relative", "synonymie subjective", "approches statistiques", "distances thématiques" ]

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

[ "", "Synonymy is a pivot relation in NLP but remains problematic. Putting forward, the distinction between relative and subjective synonymy, allows us to circunvent some difficulties. In the framework of conceptual vectors, it is then possible to formalize test functions for synonymy and to experiment their use.", "Relative synonymy, subjective synonymy, statistical approaches, thematic distances" ]

[ "mémoire de traduction sous-phrastique", "traduction assistée par ordinateur", "traduction automatique à base d’exemples" ]

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

[ "", "The usefullness of translation support tools based on translation memories is often limited by the nature of the text segments that they connect, generally whole sentences. This article examines the potential of a type of system that would be able to recuperate the translation of arbitrary sequences of words.", "" ]

[ "système multi-agents", "analyse syntaxique", "syntaxe", "environnement" ]

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

[ "", "In this article, we present un multi-agent plateform for natural language experimentation and processing. After describing the APA agent model, we present the advancement in our work: an agent-based implementation of a syntactic parser according to the chunk dependency paradigm. We then show other potential implementations according to other syntactic paradigms but also further than mere syntax.", "" ]

[ "énumérations", "représentation", "modèles de texte", "classification", "structures textuelles" ]

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

[ "", "This paper is related to the study and the representation of the internal structure of enumerations. We use two types of text structure: on one hand, the Rhetorical Structure Theory (RST) which provides a framework for interpreting and representing the discursive structures of texts and on the other hand, the model of text architecture which is is mainly dedicated to the study and representation of visuo-spatial structures of texts. After a brief presentation of the theories and models, we focus the paper on the study of enumerations. We exhibit three significant examples of particular enumerations which are called non-parallel enumerations. We analyse these examples and their characteristics and we propose a general principle for the representation of these structures with these two reference models. Then, we present a classification of the different types of enumeration in the light of these models.", "Text structures, enumerations, representation, classification, text models" ]

[ "groupe nominal", "anglais médical", "adjectif", "corpus bilingue", "ambiguïté syntaxique", "traduction", "découpage", "traduction automatique" ]

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

[ "", "Syntactic ambiguity is a particularly difficult problem to solve for the morpho-syntactic analysis programs of machine translation software, especially in the case of the long noun phrases that are typical of technical or scientific writing. Using a bilingual corpus of medical research articles translated into French, we examine various ways in which disambiguation can be achieved in English adj ective-noun-noun structures, where the adjective may modify either of the following nouns.", "adjective, bilingual corpus, machine translation, medical English, noun clause, syntactic ambiguity, syntactic structure, translation" ]

[ "extraction d’information", "génomique", "transducteurs linguistiques" ]

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

[ "", "This papers describes a system extracting information about interactions between genes or proteins, from large textual databases. The system is based on a set of linguistic finite-state transducers. The paper shows how a part of the resources (namely the set of verbs expressing the notion of interaction) can be acquired semi-automatically from the corpus. A detailed evaluation is provided.", "" ]

[ "Reconnaissance Automatique de la Parole", "Modèles de Langage statistique", "Serveurs de Dialogue", "Arbre de Décision" ]

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

[ "", "Within the framework of Human-Computer dialogue in spontaneous speech, we propose in this paper a method which automatically builds, from a training corpus, a set of Language Models (LMs) organized as a binary tree. Each LM correspond to a specific dialogue state, where the general LM is attached to the root node and the more specialized ones are represented by the leaves. Such LMs can be used to automatically adapt the decoding process to the dialog situation performed. We propose a two-pass decoding strategy, which implements this idea by dynamically selecting a set of LMs according to the dialog situation detected.", "" ]

[ "Ressources linguistiques", "Dictionnaires", "Bases de données textuelles", "Web" ]

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

[ "", "This paper presents the computerized linguistic resources of the Research Laboratory ATILF (Analyses et Traitements Informatiques du Lexique Francais) available via the Web, and will serve as a helping document for demonstrations planned within the framework of TALN 2001. The Research Laboratory ATEF is the new IHVIR (Unité Mixte de Recherche) created in association between the CNRS and the University of Nancy 2 since January 2nd, and succeeds to the local component of the INaLF situated in Nancy. This considerable amount of resources concerning French language consists in a set of more than 3500 literary works grouped together in Frantext, plus a number of dictionaries, lexis and other databases. These web available resources are operated and run through the potentialities and powerful capacities of a software called Stella, a search engine specially dedicated to textual databases and relying on a new theory of textual objects. The general policy of our laboratory is to welcome and give the research and teaching world the widest access to all our resources.", "" ]

[ "Cartographie de Textes", "Apprentissage Automatique", "Recherche d’Information", "Extraction d’Information" ]

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

[ "", "We present an ongoing research project on the new field of Text Mapping that allows a novice user to explore a new domain by navigation through an homogeneous corpus thanks to interactive conceptual maps. A map is composed of concepts (the nodes) depending on the user’s request and its evolution, and semantic/lexical relations (the links). Machine Learning techniques are combined with Natural Language Processing methodologies to build the maps.", "Text Mapping, Information Retrieval, Information Extraction, Machine Learning" ]

[ "Dialogue Homme-Machine", "traitement de la référence" ]

[ "P", "R" ]

[ "", "ln this study, we propose a model for reference resolution in the context of human computer interaction. Starting from psychological consi erations on the necessity of a common system of inferential strategies in order to allow communication, we define a forma ism based on production rules associated with cognitive costs. Through examples, we will show how this malism can be used as a framework to integrate the processing of various reference nomena, and how this integration can lead to more efficient interactive systems.", "" ]

[ "Réponse au courriel", "Analyse de texte", "Analyse de corpus" ]

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

[ "", "This paper discusses the strengths and weaknesses of different NLP techniques applied to the automatic follow-up of e-mail. First, we present the most widely-used methods, then we present a set of criteria for selecting the method to be used in particular circumstances. These criteria have been developed through a case study of a corpus provided by Bell Canada Enterprises. With our corpus, it appears that although no individual method is completely satisfactory, an approach combining several techniques is more appropriate.", "" ]

[ "Extraction d’information", "Entités nommées" ]

[ "M", "U" ]

[ "", "This paper discusses the influence of the corpus on the automatic identification of proper names in texts. Rule-based techniques developed for the newswire genre are generally not sufficient to deal with larger corpora containing texts that do not follow strict writing constraints. After a brief review of the research performed on news texts, we present some of the problems involved in the analysis of informal texts by using two different corpora (the first one composed of electronic mails, the second one of hand-transcribed telephone conversations). Once the sources of errors have been presented, we then describe an approach to adapt a proper name extraction system developed for newspaper texts to the analysis of e-mail messages.", "" ]

[ "Formalisation de l’analogie", "langages de mots analogiques", "mise en correspondance d’espaces analogiques" ]

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

[ "", "The generativists had three arguments against analogy: the innate hypothesis and the conteXt-free hypothesis, and overgeneration. Theoretical and experimental results, based on a recently obtained cornputational expression of analogy, contribute to the refutation of these three points in a constructive Way.", "Formalisation of analogy, languages of analogical strings, mapping of analogical spaces" ]

[ "Identification", "interprétation", "représentation des connaissances", "base de données", "graphes", "exploration contextuelle" ]

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

[ "", "SEEK-JAVA is a identification, interpretation and representation system of knowledge extracted from texts. It gives label to the relation between concepts and automatically identify arguments of the relations. The results, capitalized in a data base, are proposed by an interface, either in the form of graphs or in the form of tables. This system, integrated in the FilText platform, is based on contextual exploration method.", "" ]

[ "gestionnaire de dialogue", "modèle de dialogue", "modèle de la tâche", "dialogue oral homme-machine" ]

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

[ "", "In this paper we describe the dialog manager of a spoken language dialog system for information retrieval. The dialog manager maintains both static and dynamic knowledge sources, which are used according to the dialog strategies. These strategies have been developed so as to obtain the overall system objectives while taking into consideration the desired ergonomic choices. Dialog management is based on a dialog model which divides the interaction into phases and a dynamic model of the task. This allows the dialog strategy to be adapted as a function of the history and the dialog state. The dialog manager was evaluated in the context of the LE3-ARISE project, where in the last user tests the dialog success improved from 53% to 85%.", "spoken language dialog, dialog management, dialog model, task model" ]

[ "Morphologie", "terminologie", "recherche d’information", "variantes de termes", "médecine", "Doc’CISMeF", "MeSH" ]

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

[ "", "", "" ]

[ "collocations", "extraction automatique", "statistiques lexicales" ]

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

[ "", "Collocations are interesting for several NLP applications such as language generation or analysis and monolingual or bilingual lexicography. The first approaches to finding collocations appeared in 1970's and were statistically based. Today, the methods adopted for the identification of collocations still include statistics but also linguistic processing. We introduce a few approaches to finding collocations in corpora. For each method, we precise the linguistic characteristic of collocation which as been taken into account.", "collocations, automatic identification, lexical statistics" ]

[ "simplification de phrases", "Extraction de relations", "apprentissage automatique" ]

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

[ "Automatic Information Extraction in the Medical Domain by Cross-Lingual Projection", "Machine learning based relation extraction requires large annotated corpora to take into account the variability in the expression of relations. To deal with this problem, we propose a method for simplifying sentences, i.e. for reducing the syntactic variability of the relations. Simplification requires the annotation of a small corpus, which will be automatically augmented. The process starts with the annotation of the simplification thanks to a CRF classifier, then the relation extraction, and lastly the automatic completion of the training corpus for the simplification through the results of the relation extraction. The first results we obtained for the task of relation extraction of the i2b2 2010 challenge are encouraging.", "Relation extraction, sentence simplification, machine learning" ]

[ "Extraction d’information", "reconnaissance des entités médicales", "projection d’annotation", "apprentissage" ]

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

[ "Automatic Information Extraction in the Medical Domain by Cross-Lingual Projection", "This research stems from our willingness to test new methods for automatic annotation or information extraction from one language L1 by exploiting resources and tools available to another language L2. This approach involves the use of a parallel corpus (L1-L2) aligned at the level of sentences and words. To address the lack of annotated medical French corpus, we focus on the French-English language pair to annotate automatically medical French texts. In particular, we focus in this article on medical entity recognition. We evaluate our medical entity recognition method on the English corpus and the projection of the annotations on the French corpus. We also discuss the problem of scalability since we use a parallel corpus extracted from the Web and propose a statistical method to handle heterogeneous corpora.", "Automatic Information Extraction, Annotation Projection, Medical Entity Recognition, Machine Learning" ]

[ "remplissage de formulaires", "Extraction d’information", "segmentation de texte" ]

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

[ "A Graph-Based Method for Template Filling in Information Extraction", "In event-based Information Extraction systems, a major task is the automated filling from unstructured texts of a template gathering information related to a particular event. Such template filling may be a hard task when the information is scattered throughout the text and mixed with similar pieces of information relative to a different event. We propose in this paper a two-step approach for template filling : first, an event-based segmentation is performed to select the parts of the text related to the target event ; then, a graph-based method is applied to choose the most relevant entities in these parts for characterizing the event. Using an evaluation of this model based on an annotated corpus for earthquake events, we achieve a 72% F-measure for the template-filling task.", "Information Extraction, Text Segmentation, Template Filling" ]

[ "théorie des types", "récit de voyage", "grammaire catégorielle", "compositionalité", "interface syntaxe-sémantique", "interface sémantique-pragmatique" ]

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

[ "Processing of a Pyrenees travel novels corpus : a syntactical, semantical and temporal analysis", "In this article, we present a type theoretical framework which we apply to the syntactic analysis and the computation of DRS semantics. Our tool, Grail, is used for the automatic treatment of French text and we use a Pyrenées travel novels corpus, Itipy, as a test case. We explain our use of categorial grammars and specifically the Lambek calculus and its connection to the simply typed lambda-calculus in connection with DRT. Flexible typing has to be allowed in some cases and forbidden in others. Some linguistic phenomena presenting some kind of meaning shifts inducing typing conflicts will be introduced. We then present our motivations in the pragmatic field to use a system with sorts and variable types in lexical semantics and then we present how we process events temporality, in the light of Verkuyl’s Binary Tense (Verkuyl, 2008).", "compositionality, syntax-semantics interface, semantics-pragmatics interface, categorial grammar, type theory, travel novel" ]

[ "Mots composés", "analyse syntaxique", "réordonnanceur", "champs markoviens aléatoires" ]

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

[ "Recognition of compound words tested against parsing and vice-versa : evaluation of two discriminative approaches", "We propose two discriminative strategies to integrate compound word recognition in a real parsing context : (i) state-of-the-art compound pregrouping with Conditional Random Fields before parsing, (ii) reranking parses with features dedicated to compounds after parsing. We show that these two approaches help reduce up to 18% of the gap between a baseline parser and parser with golden segmentation and up to 25% for compound recognition.", "Multiword expressions, parsing, Conditional random Fields, reranker" ]

[ "nom déverbal", "cadre de sous-catégorisation", "groupe prépositionnel", "analyse en dépendances" ]

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

[ "On Computing Subcategorization Frames of French Deverbal Nouns (Case of Genitive)", "Fine dependency analysis needs exact information on the subcategoriziation frames of lexical units. These frames being well studied for the verbs, we are interested in this paper by the case of the noun deverbals. Our main goal is to calculate the subcategoriziation frame of deverbals in French from that of the source verb. However, this task needs a representative list of French deverbal nouns. To obtain such a list, we use a simplified algorithm detecting deverbal nouns in texts. The obtained list attested by linguists is compared with the existing list Verbaction of deverbals expressing the action/activity of French verbs. For these deverbal nouns, we analyse the origin of their subordinate prepositional phrases in different contexts relative to their source verbs. This analysis is carried out over the corpus Paris 7 and is limited to the most frequent cases of the genitive, i.e. to the prepositional phrases headed by the prepositions de, des, etc.", "Deverbal Noun, Subcategorization Frame, Prepositional Phrase, Dependency Tree" ]

[ "vectorisation", "Calcul de similarité", "TF-IDF", "modèle vectoriel", "Okapi BM-25" ]

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

[ "Vectorization, Okapi and computing similarity for NLP : say goodbye to TF-IDF", "In this position paper, we review a problem very common for many NLP tasks: computing similarity (or distances) between texts. We aim at showing that what is often considered as a small component in a broader complex system is very often overlooked, leading to the use of sub-optimal solutions. Indeed, computing similarity with TF-IDF weighting and cosine is often presented as “state-of-theart”, while more effective alternatives are in the Information Retrieval (IR) community. Through some experiments on several tasks, we show how this simple calculation of similarity can influence system performance. We consider two particular alternatives. The first is the weighting scheme Okapi-BM25, well known in IR and directly interchangeable with TF-IDF. The other, called vectorization, is a technique for calculating text similarities that we have developed which offers some interesting properties.", "Calculating similarities, vector space model, TF-IDF, Okapi BM-25, vectorization" ]

[ "français parlé", "Etiquetage morpho-syntaxique", "ressources langagières" ]

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

[ "TCOF-POS : A Freely Available POS-Tagged Corpus of Spoken French", "This article details the creation of TCOF-POS, the first freely available corpus of spontaneous spoken French. We present here the methodology that was followed in order to obtain the best possible quality in the final resource. This corpus already is freely available and can be used as a training/validation corpus for NLP tools, as well as a study corpus for linguistic research. We also present the results obtained by two POS-taggers trained on the corpus.", "Etiquetage morpho-syntaxique, français parlé, ressources langagières" ]

[ "alignement sous-phrastique", "corpus parallèle", "traduction automatique statistique" ]

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

[ "Hierarchical sub-sentential alignment with Anymalign", "We present a sub-sentential alignment algorithm that aligns sentence pairs from an existing alignment matrix in a very easy way. This algorithm is inspired by previous work on alignment by recursive binary segmentation and on document clustering. We evaluate the alignments produced on machine translation tasks and show that we can obtain state-of-the-art results, with gains up to more than 4 BLEU points compared to our previous work, with a method that is very simple, independent of the size of the corpus to be aligned, and can directly produce symmetric alignments. When using this method as an extension of the translation extraction tool Anymalign, our experiments allow us to determine some of its limitations and to define possible leads for further improvements.", "parallel corpus, sub-sentential alignment, statistical machine translation" ]

[ "Translittération", "alignement de mots", "traduction automatique", "construction de dictionnaires multilingues", "recherche d’information interlingue" ]

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

[ "Using Arabic transliteration to improve word alignment from French-Arabic parallel corpora", "In this paper, we focus on the use of Arabic transliteration to improve the results of a linguistic word alignment approach from parallel text corpora. This approach uses, on the one hand, a bilingual lexicon, named entity and cognates linguistic properties to align single words, and on the other hand, syntactic dependency relations to align compound words. We have evaluated the word aligner integrating Arabic transliteration using two methods: A manual evaluation of the alignment quality and an evaluation of the impact of this alignment on the translation quality by using the statistical machine translation system Moses. The obtained results show that Arabic transliteration improves the quality of both alignment and translation.", "Transliteration, word alignment, multilingual lexicons construction, machine translation, cross-language information retrieval" ]

[ "compositionnalité", "Corpus comparable", "information contextuelle", "lexique bilingue" ]

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

[ "Compositionality and Context for Bilingual Lexicon Extraction from Comparable Corpora", "In this article, we study the possibilities of improving the alignment of equivalent terms monolingually acquired from bilingual comparable corpora. Our overall objective is to identify and to translate highly specialised terminology. We applied a compositional approach enhanced with pre-processed context information. Our evaluation demonstrates that our alignment method outperforms the compositional approach for translationally equivalent term discovery from comparable corpora.", "Comparable Corpora, compositionality, context information, bilingual lexicon" ]

[ "raffinement", "lexique", "verbes", "ressource", "étiquetage de rôles sémantiques" ]

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

[ "Resource Refining : « Les Verbes Français »", "This paper introduce the impovements we made to the resource « Les Verbes Français » in order to make it more usable in the field of natural language processing. Syntactic and semantic information is corrected, restructured, unified and then integrated to the XML version of this resource, in order to be used by a semantic role labelling system.", "resource, lexicon, verbs, refinement, semantic roles labeling" ]

[ "méronymie", "analyse distributionnelle", "sémantique lexicale", "évaluation" ]

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

[ "Study of meronymy in a distribution-based lexical resource", "In this paper, we study the way meronymy behaves in a distribution-based lexical resource. We address the question of the evaluation of such resources through a semantic-based approach. Our method consists in collecting meronyms from a resource which we cross with a distributionbased lexical resource made from an encyclopedic corpus. Meronyms are then sub-categorized manually : firstly following the sub-relation they bear (STUFF/OBJECT, MEMBER/COLLECTION, etc.), then following the semantic class of their members. Results show that distributional analysis identifies meronymic relations in different proportions according to the semantic classes of the words involved in the meronymic pairs.", "distributional analysis, lexical semantics, meronymy, evaluation" ]

[ "Cohésion thématique", "graphe de cooccurrences", "marche aléatoire" ]

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

[ "Topical Cohesion using Graph Random Walks", "In this article, we propose a novel metric to weight specialized lexicons terms according to their relevance to the underlying thematic. Our method is inspired by Web as corpus approaches and accumulates exogenous knowledge about a specialized lexicon from the web. Terms cooccurrences are modelled as a graph and a random walk algorithm is applied to compute terms relevance. Finally, we study the performance and stability of the metric and evaluate it in a bilingual lexicon creation context.", "Thematic relevance, cooccurrence graph, random walk" ]

[ "Wikipédia", "paraphrase", "aide à la rédaction" ]

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

[ "Assisted rephrasing for Wikipedia contributors through Web-based validation", "This works describes initial experiments on the validation of paraphrases in context. Wikipedia’s revisions are used : we assume that a set of possible rewritings are available for a given phrase that has been rewritten in the encyclopedia’s revision history, and we attempt to find the subset of those rewritings that can be considered as valid paraphrases. We tackle this problem as a classication task which we provide with features obtained from Web data. Our experiments show that our system improves performance over a set of simple baselines.", "paraphrasing, Wikipedia, authoring aids" ]

[ "simplification automatique", "lisibilité", "analyse syntaxique" ]

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

[ "Syntactic Simplification for French Sentences", "This paper presents a method for the syntactic simplification of French texts. Syntactic simplification aims at making texts easier to understand by simplifying the elements that hinder reading. It is based on a corpus study that aimed at investigating the linguistic phenomena involved in the manual simplification of French texts. We have first gathered a parallel corpus of articles from Wikipedia and Vikidia, that we used to establish a typology of simplifications. In a second step, we implemented a system that carries out syntactic simplifications based on these corpus observations. We described simplification rules in order to generate simplified sentences. A module subsequently selects the best subset of sentences. The evaluation of our system shows that about 80% of the sentences produced by our system are accurate.", "automatic simplification, readability, syntactic analysis" ]

[ "Résumé automatique", "document arabe", "approche symbolique", "approche numérique", "approche hybride" ]

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

[ "Comparative study of three approaches to automatic summarization of Arabic documents", "In this paper, we propose a comparative study between three approaches for automatic summarization of Arabic documents. Thus, we proposed three methods for extracting most representative sentences of a document. The first method is based on a symbolic approach, the second is relied on a numerical approach and the third is based on a hybrid approach. These methods are implemented respectively by the ARSTResume, R.I.A and HybridResume systems. Then, we present the results obtained by the three systems and we conduct a comparative study between the obtained results in order to highlight the advantages and limitations of each method. The evaluation results showed that the numerical approach has better performances than the symbolic approach. But, combining into a hybrid approach achieved the best results for our text corpus.", "Automatic summarization, symbolic approach, numerical approach, hybrid approach, Arabic document" ]

[ "sémantique quantitative", "unités lexicales spécifiques", "analyse des marqueurs lexicaux stables", "analyse des mots-clés", "analyse de régression" ]

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

[ "Semantic analysis of keywords and stable lexical markers in a technical corpus", "This article presents the results of a quantitative semantic analysis of typical lexical units in a specialised technical corpus of metalworking machinery in French. The study aims to find out whether and to what extent the keywords of the technical corpus are monosemous. A simple regression analysis, used to examine the correlation between typicality rank and monosemy rank of the keywords, points out some statistical and methodological problems, notably a frequency bias. In order to overcome these problems, we adopt an alternative approach for the identification of typical lexical units, called Stable Lexical Marker Analysis (SLMA). We discuss the quantitative and statistical results of this approach with respect to the correlation between typicality rank and monosemy rank.", "typical lexical units, Keyword Analysis, Stable Lexical Marker Analysis (SLMA), quantitative semantics, regression analysis" ]

[ "Fouille de graphes", "réseaux phrastiques", "analyse textuelle", "navigation textuelle" ]

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

[ "Graph Mining Under Linguistic Constraints to Explore Large Texts", "In this paper, we propose an approach to explore large texts by highlighting coherent sub-parts. The exploration method relies on a graph representation of the text according to the Hoey linguistic model which allows the selection and the binding of sentences in the graph. Our contribution relates to using graph mining techniques under constraints to extract relevant subparts of the text (i.e., collections of homogeneous sentence sub-networks). We have conducted some experiments on two large English texts to show the interest of the proposed approach.", "Graph Mining, sentence networks, textual analysis, textual navigation" ]

[ "acquisition de paraphrases", "constitution de corpus" ]

[ "P", "M" ]

[ "A study of paraphrase along 3 dimensions : corpus types, languages and techniques", "In this paper, we report a detailed study of the impact of corpus type on the task of sub-sentential paraphrase acquisition. Our experiments are for 2 languages and 4 corpus types, and involve an efficient machine learning-based combination of 4 paraphrase acquisition systems. We obtain relative improvements of more than 27% in F-measure over the best individual system on English and French, and obtain a relative improvement over the combination system of 22% for English and 5% for French when using all other corpus types as additional training data for our most readily available corpus type.", "paraphrase acquisition, corpus collection" ]

[ "correction automatique", "Étiquetage morpho-syntaxique", "qualité d’annotation" ]

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

[ "Detecting and correcting POS annotation in the French TreeBank", "The quality of the Part-Of-Speech (POS) annotation in a corpus has a large impact on training and evaluating POS taggers. In this paper, we present a series of experiments that we have conducted on automatically detecting and correcting annotation errors in the French TreeBank. Two methods are used. The first simply relies on identifying tokens with missing tags and correct them by assigning the tag the same token observed in the corpus. The second method uses n-gram variations to detect and correct conflicting annotations. The evaluation of the automatic correction is performed extrinsically by comparing the performance of different POS taggers in relation to the level of correction. Results show a statistically significant improvement in precision and indicate that the POS annotation quality can be noticeably enhanced by using automatic correction methods.", "Part-Of-Speech tagging, automatic correction, annotation quality" ]

[ "dépendances", "DMRS", "réécriture de graphes", "interface syntaxe-sémantique" ]

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

[ "Semantic Annotation of the French Treebank using Modular Graph Rewriting", "We propose to annotate the French Treebank with semantic dependencies in the framework of DMRS starting from an annotation with surface syntactic dependencies and using modular graph rewriting. The article presents some new results related to the rewriting calculus: the use of rules to make a link with lexicons, especially with the lexicon of verbs Dicovalence, and the introduction of filters to discard inconsistent annotations at some computation steps. It also presents new results related to the rewriting system itself: the system has a larger coverage (causative constructions, rising verbs, . . .) and the order between modules has been studied in a more systematic way. This system has been experimented on the whole French Treebank with the prototype GREW, which implements the used rewriting calculus.", "graph rewriting, syntax-semantics interface, dependencies, DMRS" ]

[ "Treebank hybride", "Grammaires de Propriétés", "French Treebank" ]

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

[ "Enriching the French Treebank with Properties", "We present in this paper the hybridation of the French Treebank with Property Grammars annotations. This process consists in acquiring a PG grammar from the source treebank and generating the new syntactic encoding on top of the original one. The result is a new resource for French, opening the way to new tools and descriptions.", "Hybrid treebank, French Treebank, Property Grammars" ]

[ "Corpus arboré", "analyse syntaxique statistique", "adaptation de domaine" ]

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

[ "The Sequoia corpus : syntactic annotation and use for a parser lexical domain adaptation method", "We present the building methodology and the properties of the Sequoia treebank, a freely available French corpus annotated following the French Treebank guidelines (Abeillé et Barrier, 2004). The Sequoia treebank comprises 3204 sentences (69246 tokens), from the French Europarl, the regional newspaper L’Est Républicain, the French Wikipedia and documents from the European Medicines Agency. We then provide a method for parser domain adaptation, that makes use of unsupervised word clusters. The method improves parsing performance on target domains (the domains of the Sequoia corpus), without degrading performance on source domain (the French treenbank test set), contrary to other domain adaptation techniques such as self-training.", "Treebank, statistical parsing, parser domain adaptation" ]

[ "dépendances", "annotations concurrentes", "annotation collaborative de corpus" ]

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

[ "ACOLAD: platform for collaborative dependency annotation", "This paper presents an open-source platform for collaborative editing dependency corpora. ACOLAD platform (Annotation of corpus linguistics for the analysis of dependencies) offers manual annotation services such as segmentation and multi-level annotation (segmentation into words and phrases minimum (chunks), morphosyntactic annotation of words, syntactic annotation chunks and annotating syntactic dependencies between words or chunks). In this paper, we present ACOLAD platform, then we detail the representation used to manage concurrent annotations, then we describe the mechanism for importing external linguistic resources.", "corpus collaborative annotation, concurrent annotations, dependencies" ]

[ "Préférence", "dialogue", "apprentissage automatique" ]

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

[ "Towards Preference Extraction From Negotiation Dialogues", "This paper presents an NLP based approach for preference expression extraction from negotiation dialogues. We propose a new annotation schema for preferences and dependencies among them and illustrate on two different corpus genres. We then suggest a learning approach that efficiently extracts preference expressions using a combination of local and discursive features and assess the reliability of our approach on each corpus genre.", "Preference, dialogue, machine learning" ]

[ "conflit", "wikipedia", "syntaxe", "sémantique", "interaction" ]

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

[ "Conflicts detection in online epistemic communities", "Conflicts in online epistemic communities can be a blocking factor when producing knowledge. We present a way to automatically detect conflict in Wikipedia discussions, based on subjectivity and connotation marks. Two rules are evaluated : a local rule that uses the structure of the discussion threads, connotation and subjectivity marks and a global rule that takes the whole thread into account and only subjectivity. We show that the two rules produce similar results but that the simplicity of the global rule makes it a preferred approach to detect conflicts.", "wikipedia, conflict, syntax, semantics, interaction" ]

[ "entités nommées", "annotation", "extraction d’information", "concepts métier", "données conversationnelles" ]

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

[ "What is the contribution of named entities detection for information extraction in restricted domain ?", "In the framework of general domain dialog corpora a particular focus is dedicated to Named Entities definition and recognition, which are mostly very generic (personal names, locations, etc.). Moreover, call-centre data mining is strategic for a company like EDF, the public opinion analysis playing a significant role in EDF services quality evaluation and for marketing applications. In this purpose a domain dependant definition of entities of interest is essential. In this primary work we compare two types of entities models (generic and specific to the domain) in order to observe their respective coverage. We annotated manually a sub-corpus extracted from a large corpus of oral dialogs recorded in an EDF call-centre. The respective proportion of generic vs domain-specific Named Entities is then estimated. Impact for future work on building EDF domain-specific entities models is discussed.", "named entities, business concept, information extraction, conversational data, annotation" ]

[ "textométrie", "analyse de sentiments", "linguistique de corpus", "classification automatique supervisée" ]

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

[ "About the application of textometric methods for developing classification criteria in Sentiment analysis", "Over the last ten years, NLP has contributed to applied research on subjectivity, especially in applications such as Opinion mining and Sentiment analysis. However, corpus linguistics and textometry have often addressed the issue of subjectivity in text. Our purpose is to show, !rst, what textometric analysis could bring to sentiment analysis, and second, the bene!ts of pooling linguistic/textometric analysis and automatic classification methods based on supervised learning. By processing a corpus of posts from fora, we will show that the building of criteria from a textometric analysis could improve classification results, compared to a purely lexical approach.", "corpus linguistics, textometry, sentiment analysis, supervised learning" ]

[ "Langue des signes", "synchronisation", "analyse de corpus", "modèle grammatical" ]

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

[ "Methodology for corpus exploration and grammatical rule building in Sign Language", "This paper presents a methodology for Sign Language video observation to extract and then formalise observed linguistic structure. This methodology is relevant to all linguistic layers from sub-lexical to discourse as a whole. Relying on two examples, we apply this methodology and describe the AZee model, which integrates the required flexibility for synchronising articulators, hence enables a specification of any new systematic rule observed.", "Sign Language, corpus analysis, grammatical models, synchronisation" ]

[ "annotation manuelle", "accords inter-annotateurs" ]

[ "P", "P" ]

[ "Manual Annotation of Football Matches : Inter-annotator Agreement ! Gooooal !", "We present here an annotation campaign of commentaries of football matches in French. The annotation was done from a very heterogeneous text corpus of both match minutes and video commentary transcripts. We show how the intra- and inter-annotator agreement can be used efficiently during the whole campaign by proposing a definition of the markables suited to our type of task, as well as emphasizing the importance of using it appropriately. We also show how some clues, collected through statistical analyses, could be used to help correcting the annotations. These statistical analyses are then used to assess the impact of the source modality (written or spoken) on the cost and quality of the annotation process.", "manual annotation, inter-annotator agreement" ]

[ "adaptation à un nouveau domaine", "fouille d’opinion", "auto-apprentissage" ]

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

[ "Study of various strategies for adapting an opinion classifier to a new domain", "The work presented in this article takes place in the field of opinion mining and aims more particularly at finding the polarity of a text by relying on machine learning methods. In this context, it focuses on studying various strategies for adapting a statistical classifier to a new domain when training data only exist for one or several other domains. This study shows more precisely that a self-training procedure consisting in enlarging the initial training corpus with texts from the target domain that were reliably classified by the classifier is the most successful and stable strategy for the tested domains. Moreover, this strategy gets better results in most cases than (Blitzer et al., 2007)’s method on the same evaluation corpus while it is more simple.", "opinion mining, domain adaptation, self-training" ]

[ "lexique des émotions", "profl combinatoire", "expressions polylexicales", "collocations", "cooccurrences" ]

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

[ "The Lexicoscope : an integrated tool for combinatoric profles observation and lexico-syntactic constructs extraction", "The German-French research project Emolex whose aim is the contrastive study of the combinatorial behaviour of emotion lexemes in 5 languages has led to the development of methods and tools to extract, display and compare the combinatorial profles of simple and complex expressions. In this paper, we present the overall architecture of the query platform which has been conceived to ensure efcient processing of huge annotated text corpora (consisting of several hundred millions of word tokens) accessible through a web-based interface. We put forward the concept of “complex query nodes” introduced to enable users to carry out progressively elaborated extractions of lexical-syntactic patterns. We fnally give primary results of an automated method for the retrieval of recurrent multi-word expressions, which takes advantage of the complex query nodes implementation.", "collocations, combinatorial profles, multi-word expressions" ]

[ "équivalents terminologiques", "corpus comparables", "terminologie médicale", "vecteurs contextuels", "étude qualitative" ]

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

[ "Analysis of contexts and candidates in term-translation spotting in comparable corpora", "The standard approach for identifying terminological equivalents from comparable corpora is based on the comparison of source and target language context words using a bilingual lexicon. We cary a manual analysis of the linguistic properties (parts of speech, specificity and semantic relations) of the context words and the inacurrate equivalents given by the standard approach applied to medical terminology, in order to suggest improvements based on the selection of context words.", "terminological equivalents, contextual vectors, comparable corpora, medical terminology, qualitative study." ]

[ "prototype", "mots composés", "documents comparables", "méthode compositionnelle", "terminologie bilingue", "extraction terminologique", "corpus" ]

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

[ "BiTermEx , A prototype for the extraction of multiword terms from comparable documents through the compositional approach", "We describe BiTermEx, a prototype for extracting multiword terms from comparable corpora, using the compositional method. We focus on morphology-based variations of multiword constituents and their recombinaison. We experimented our approach on scientific and popular science corpora. We record TOP1 precisions of 92% and 97,5% on French to English alignments and 94% on French to Japanese.", "term extraction, prototype, bilingual terminology, comparable documents, compositional method, multiword terms, corpus" ]

[ "événements", "Extraction d’information", "approche symbolique", "apprentissage de patrons linguistiques" ]

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

[ "Automatic events extraction by combining multiple approaches", "In this paper, we present an automatic system for extracting events based on the combination of two existing information extraction approaches : the first one is made of hand-crafted linguistic rules and the second one is based on an automatic learning of linguistic patterns. We have shown that this mixed approach leads to a significant improvement of extraction performances.", "Text mining, events, symbolic extraction, linguistic pattern learning" ]

[ "apprentissage automatique", "French Tree Bank", "CRF", "chunking" ]

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

[ "Machine Learning of a chunker for French", "We describe in this paper how to automatically learn a chunker for French, from the French Tree Bank and CRFs (Conditional Random Fields). We did several experiments, either to recognize every possible kind of chunks, or to focus on simple nominal phrases only. We evaluate the obtained chunker on internal data (i.e. also extracted from the French Tree Bank) as well as on external (i.e from a distinct corpus) ones, to measure its robustness.", "chunking, Machine Learning, French Tree Bank, CRF" ]

[ "similarité textuelle", "classification de documents", "corpus spécialisé" ]

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

[ "Deletion of dimensions of textual similarity for the exploration of collections of accident reports in aviation", "In this paper we study the relationship between external classification and textual similarity in collections of incident reports. Our goal is to complement the existing classification-based analysis strategies by automatically establishing similarity links between documents in such a way that they do not reflect the dominant organisation of the classification schemas. In order to discover such transversal dimensions of similarity, we compute association scores between terms and classes and exlude the most correlated terms from the similarity calculation. We demonstrate on a 500 document corpus that by using this method, we can isolate topics that would otherwise have been masked by the dominant dimensions of similarity in the collection.", "textual simliarity, document classification, specialised corpora" ]

[ "extraction de phrases parallèles", "traduction automatique statistique", "corpus comparables multimodaux", "Reconnaissance de la parole" ]

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

[ "Automatic Translation from Comparable corpora : extracting parallel sentences from multimodal comparable corpora", "Statistical Machine Translation (SMT) systems depend on the availability of bilingual parallel text, also called bitext. However parallel corpora are a limited resource and are often not available for some domains or language pairs. We present an alternative method for extracting parallel sentences from multimodal comparable corpora. This work extends the use of comparable corpora, in using audio instead of text on the source side. The audio is transcribed by an automatic speech recognition system and translated with a base-line SMT system. We then use information retrieval in a large text corpus of the target language to extract parallel sentences. We have performed a series of experiments on data of the IWSLT’11 speech translation task (TED) that shows the feasibility of our approach.", "Automatic speech recognition, statistical machine translation, multimodal comparable corpora, extraction of parallel sentences" ]

[ "Pronoms démonstratifs", "résolution des anaphores", "traitement de la langue arabe" ]

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

[ "Automatic recognition of demonstrative pronouns function in Arabic", "Anaphora resolution is one of the most difficult tasks in NLP. Classifying pronouns before attempting a task of anaphora resolution is important because to handle the cataphoric pronoun, the system should determine the antecedent into the segment following the pronoun. Although, for the anaphoric pronoun, the system should look for the antecedent into the segment before the pronoun. In addition, the number of demonstrative pronouns is very important in Arabic. In this paper, we describe a machine learning method for classifying demonstrative pronouns in Arabic. We have evaluated our approach on a corpus of 365585 words which contain 14318 demonstrative pronouns and we have obtained encouraging results: 99.3% as F-Measure.", "Demonstrative pronouns, anaphora resolution, ANLP" ]

[ "évaluation", "Analyse temporelle" ]

[ "P", "R" ]

[ "An Automatic Temporal Expression Annotator and its Evaluation on the French TimeBank", "In this article, we present a tool that extracts and normalises a subset of temporal expressions in French. This tool is being developed and used in the ANR (French National Research Agency) project Chronolines, applied to a corpus of provided by the Agence France Presse. The aim of the project is to semi-automatically construct event chronologies from this corpus. To do this, a detailed analysis of the temporal information conveyed by texts, is required. The system we present here is the first version of a temporal annotator that we have developed for French. We describe it in this article and present the results of an evaluation.", "Temporal processing, evaluation" ]

[ "Discours", "PDTB", "analyse discursive", "RST", "SDRT", "corpus annoté manuellement" ]

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

[ "Towards the FDTB : French Discourse Tree Bank", "We present the first steps towards creating an annotated corpus for discourse in French : the French Discourse Treebank enriching the FTB. Our methodology is based on the Penn Discourse Treebank (PDTB), but it differs in at least two points of a theoretical nature. First, our goal is to provide full coverage of a text, while the PDTB provides only partial coverage, which can not be described as discourse analysis such as the one made in RST or SDRT, two major theories on discourse. Second, we were led to define a new hierarchy of discourse relations which is based on RST, SDRT and PDTB.", "Discourse, manually annotated corpus, discourse analysis, PDTB, RST, SDRT" ]

[ "Combinaison de ressources", "RI contextuelle", "recherche web" ]

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

[ "Query Contextualization and Reformulation by Combining External Corpora", "Improving document retrieval using external sources of information has been extensively studied throughout the past. Improvements with either structured or large corpora have been reported. However, in these studies resources are often used separately and rarely combined together. We present an evaluation of the combination of four different scalable corpora over a web search task. An informative divergence measure is used to extract contextual features from the corpora and improve query representation. We use the ClueWeb09 collection along with TREC’s Web Track topics for the purpose of our evaluation. Best results are achieved when combining all four corpora, and are significantly better than the results of other approaches.", "Resources combination, contextual IR, web search." ]

[ "Repérage des entités nommées", "Adaptation non supervisée" ]

[ "P", "R" ]

[ "Named Entity Recognition for Arabic : Unsupervised adaptation and Systems combination", "The recognition of Arabic Named Entities (NE) is a potentially useful preprocessing step for many Natural Language Processing Applications, such as Machine Translation. This task is however made very complex by some peculiarities of the Arabic language. In this paper, we present a summary of our recent efforts aimed at developing a statistical NE recognition system, with a specific focus on feature engineering aspects. We also report several approaches for adapting this system in an entirely unsupervised manner to a new domain.", "Unsupervised domain adaptation, named entity recognition" ]

[ "analyse de sentiments", "ressources lexicales", "morphologie dérivationnelle" ]

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

[ "Spreading Polarities among Word Families: Impact of Morphology on Building a Lexicon for Sentiment Analysis", "Lexical resources are essential for many natural language applications (for example, opinion mining from corpora). However, building them entails different problems (cost, coverage, etc.). In this paper, we wanted to verify whether morphological information about derivation could be used to automatically annotate semantic information. Starting from a resource that groups words into morphological families in French, we have built a lexicon with polarities for 4 065 words from an initial seed set of manual annotated adjectives. The results obtained show that spreading polarities is accurate for 78.89% of the families with a unique adjective. The lexicon obtained also improves the results of the opinion mining system on different corpora.", "lexical resources, derivational morphology, opinion mining" ]

[ "transitions thématiques", "annotation", "Structure du discours", "segments thématiques" ]

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

[ "Manual thematic annotation of a journalistic corpus : first observations and evaluation", "The work presented in this paper focuses on the creation of a corpus of journalistic texts annotated at dicourse level, more precisely on a topic level. The annotation model is a classic segmentation one, to which we add transition zones between topical units. We assume that in a well-structured text, the author provides information helping the reader to move from one topic to another, where an identification of these clues is likely to improve automatic segmentation. The produced annotations have been subject of several quantitative analyses showing a set of linguistic properties of topical transitions.", "Discourse structure, topical segments, topical transitions, annotation" ]

[ "extraction", "terminologie", "langue contrôlée", "potentiel terminologique", "filtrage de termes" ]

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

[ "Multi-extraction as a strategy of optimized extraction of terminological and lexical resources", "Based on the evaluation of terminological extractors, initially to find the best tool for building a controlled language lexicon, we propose a strategy of optimized extraction of terminological resources. Our work highlights that the cooperation of several extraction tools gives better results than the use of a single one. It both reduces silence and automatically filters noise thanks to a variable related to termhood.", "terminology, extraction, controlled language, termhood, term filtering" ]

[ "Recherche d’information", "correction d’erreurs", "xml", "dysorthographie" ]

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

[ "Structured Information Retrieval Approach based on Indexing Time Error Correction", "In this paper, we focused on errors in the textual content of XML documents. We propose an approach to reduce the impact of these errors on Information Retrieval (IR) systems. Indeed, these systems rely on indexes associating each document to corresponding terms. Indexes quality is negatively affected by those misspellings. These errors makes it difficult to later retrieve documents (or parts of them) in an effective way during the querying phase. In order to deal with this problem we propose to include an error correction mechanism during the indexing phase of documents. We achieved an implementation of this spelling aware information retrieval system which is currently evaluated over INEX evaluation campaign documents collection.", "Information retrieval, misspellings, error correction, xml" ]

[ "post-édition", "adaptation aux domaines de spécialité", "Traduction automatique statistique" ]

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

[ "Statistical Post-Editing of Machine Translation for Domain Adaptation", "This paper presents a statistical approach to adapt generic machine translation systems to the medical domain through an unsupervised post-edition step. A statistical post-edition model is built on statistical machine translation outputs aligned with their translation references. Evaluations carried out to translate medical texts from French to English show that a generic machine translation system can be adapted a posteriori to a specific domain. Two systems are studied : a state-of-the-art phrase-based implementation and an online publicly available software. Our experiments also indicate that selecting sentences for post-edition leads to significant improvements of translation quality and that more gains are still possible with respect to an oracle measure.", "Statistical Machine Translation, Post-editing, Domain Adaptation" ]

[ "Corpus arboré de Paris 7", "Résolution d’entités nommées", "Corpus annoté" ]

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

[ "Referential named entity annotation of the Paris 7 French TreeBank", "The French TreeBank developed at the University Paris 7 is the main source of morphosyntactic and syntactic annotations for French. However, it does not include explicit information related to named entities, which are among the most useful information for several natural language processing tasks and applications. Moreover, no large-scale French corpus with named entity annotations contain referential information, which complement the type and the span of each mention with an indication of the entity it refers to. We have manually annotated the French TreeBank with such information, after an automatic pre-annotation step. We sketch the underlying annotation guidelines and we provide a few figures about the resulting annotations.", "Named entity resolution, Annotated corpus, French TreeBank" ]

[ "fonctions de croyance", "estimation de paramètres", "Traduction automatique statistique", "apprentissage automatique" ]

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

[ "Feature calculation for Statistical Machine Translation by using belief functions", "In this paper, we consider the translation of texts within the framework of the 7th Workshop of Machine Translation evaluation task and the COSMAT corpus using a statistical machine translation approach. This work is focused on the translation features calculation of the phrase contained in a phrase table. The classical way to estimate these features are based on the direct computation counts or frequencies. In our approach, we propose to use the concept of belief masses to estimate the phrase probabilities. The Belief Function theory has proven to be suitable and adapted for the management of uncertainties in many domains. The experiments based on our approach are focused on the language pair English-French.", "Statistical machine Translation, belief function, machine learning, feature estimation" ]

[ "Tour de parole", "Conversation", "Centre d’appels", "Reconnaissance de Parole" ]

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

[ "Turn-taking length as criterion to select call center conversations", "This article focuses on telephone conversations collected in an EDF Call Center, automatically segmented in “turn-taking” and automatically transcribed. It shows a relationship between the length of the turns and their content regarding the vocabulary and the feelings that are conveyed. After showing that there is an interest in studying these long turns, the article analyzes their content and lists some examples around the notions of argumentation and claim. It shows that the length of turns can be a useful criterion for selecting conversations.", "Call Center, Conversation, Turn Taking, Automatic Speech Recognition" ]

[ "méthodologie", "Français écrit des sourds", "syntaxe", "TAL", "Français Langue Etrangère", "linguistique de corpus", "lexique" ]

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

[ "Methodological, linguistic and computational challenges for processing written French of deaf people", "With the setup of a national emergency call-center for deaf people in France (CNRAU), some questions arise in linguistics and natural language processing about the written expression of deaf people. It is practiced by an heterogeneous population and shows morpho-syntactic, lexical and syntactic specificities which increase the difficulty, over the emergency situation, to successfully communicate between the deaf callers and the call-center operators. A first corpus (FAX-ESSU) of written French of deaf people was built with emergency conditions in order to provide linguistic and NLP solutions to the call center operators. On this corpus, we present a first study realized with the help of a natural language processing toolbox, in order to validation linguistic phenomenons described in the scientific literature and to enrich the knowledge of written French of deaf people.", "Written French of deaf people, NLP, French as a foreign language, corpus linguistics, lexicon, syntax, methodology" ]

[ "réécriture de graphes", "interface syntaxe-sémantique" ]

[ "P", "U" ]

[ "Grew: a Graph Rewriting Tool for NLP", "We present a Graph Rewriting Tool dedicated to NLP applications. Graph nodes contain feature structures and edges describe relations between nodes. We explain the Graph Rewriting framework we use, the implemented system and some experiments.", "graph rewriting, syntax-semantics interface" ]

[ "navigation enrichie", "Traitements multi-vues" ]

[ "P", "R" ]

[ "Navigation interfaces through audio and video contents", "Two types of demonstrators are shown. A first interface, with didactic purposes, allows automatic processing of video documents to be observed. Several representation levels can be viewed simultaneously, which is particularly helpful to analyse the behaviour of multi-view approaches. The second interface is an operational audio document \"consumption\" interface. It offers an enriched navigation experience through the visualisation of automatically extracted metadata.", "Multi-view processing, enriched navigation" ]

[ "Synthèse de texte", "Grammaire attribuée", "Syntaxe" ]

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

[ "Automated generation of text with Syntox", "Syntox, which includes an online user interface at URL http://www.syntox.net, is an implementation of a model based on attribute grammars, in the context of automated generation of text. The sofware is intended as a platform for experimentation with an ergonomic interface. Syntox is usable with large vocabularies and grammars to produce ambiguous texts from an explicit description of linguistic phenomena.", "Text generation, Attribute Grammars, Syntax" ]

[ "French Tree Bank", "étiquetage POS", "chunking", "apprentissage automatique", "CRF" ]

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

[ "A Segmenter-POS Labeller and a Chunker for French", "We propose a demo of two softwares : a Segmenter-POS Labeller for French and a Chunker for texts treated by the first program. Both have been learned from the French Tree Bank.", "POS tagging, chunking, Machine Learning, French Tree Bank, CRF" ]

[ "segmentation", "phonétisation", "alignement", "syllabation" ]

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

[ "SPPAS : a tool to perform text/speech alignment", "SPPAS is a new tool dedicated to phonetic alignments, from the LPL laboratory. SPPAS produces automatically or semi-automatically annotations which include utterance, word, syllabic and phonemic segmentations from a recorded speech sound and its transcription. SPPAS is currently implemented for French, English, Italian and Chinese There is a very simple procedure to add other languages in SPPAS : it is just needed to add related resources in the appropriate directories. SPPAS can be used by a large community of users : accessibility and portability are importants aspects in its development. The tools and resources will all be distributed with a GPL license.", "segmentation, phonetization, alignement, syllabification" ]

[ "entités nommées", "extraction de relations", "création d’ontologies", "similarité" ]

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

[ "How to adapt the Proxem semantic analysis engine to the Human Resources field", "Proxem develops since 2007 the NLP platform, Antelope, with which one can quickly build vertical semantic applications (for e-reputation, business intelligence or consumer reviews analysis, for instance). Antelope was used to create a Human Resources solution, notably used by APEC, making it possible (1) to extract information from resumes and offers and (2) to find the most relevant jobs matching a given resume (or vice versa). We present here how to adapt Antelope to a particular area, namely HR.", "named entities, information extraction, ontologies development, matching" ]

[ "recherche d’information", "analyse d’opinion", "génération de texte", "fouille du web" ]

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

[ "Nomao : a geolocalized search engine dedicated to place recommendation and ereputation", "This demonstration showcases NOMAO, a geolocalized search engine which recommends places (bars, shops...) based on the user’s and its friend’s tastes and on the web surfers’ recommendations", "information retrieval, opinion mining, text generation, web mining" ]

[ "Dictionnaire électronique", "verbes", "dimension bilingue", "diversité linguistique" ]

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

[ "The DictAm : An electronic dictionary of Amazigh-French verbs", "In the frame of promoting the linguistic diversity among knowledge society, we suggest to elaborate an electronic dictionary of Amazigh-French verbs (DictAm). This dictionary aims at accounting for the verbs in the Amazigh sphere as a whole: conjugation, diathesis and meaning. The DictAm also adopts a comparative perspective in the sense that it collects the lexical materials of different dialectal varieties and makes them reachable. Now, the DictAm comprises around 3000 verbs deriving from about sixty Amazigh speeches. This work is currently in progress as well as it aspires to set up a repertoire of all Amazigh verbs and their French equivalents.", "Electronic dictionary, bilingual dimension, linguistic diversity, verbs" ]

[ "aide à la prononciation", "temps réel", "retour visuel", "tête parlante", "GMM" ]

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

[ "Vizart3D: Visual Articulatory Feedack for Computer-Assisted Pronunciation Training", "We describe a system of visual articulatory feedback, which aims to provide any speaker with a real feedback on his/her own articulation. Application areas are computerassisted pronunciation training (phonetic correction) for second-language learning and speech rehabilitation. This system, named Vizartd3D, is based on the 3D augmented talking head developed at GIPSA-lab, which is able to display all speech articulators including usually hidden ones like the tongue. In our approach, the talking head is animated automatically from the audio speech signal, using GMM-based voice conversion and acoustic-to-articulatory regression.", "visual feedback, pronunciation training, GMM, real-time, talking head" ]

[ "logiciel", "corpus", "oral", "linguistique" ]

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

[ "ROCme!: software for the recording and management of oral corpora", "management of speech recordings. Users can create interfaces for metadata collection thanks to XML tags. Speakers autonomously fill in questionnaires, record, play, and save audio; and browse sentences (or other types of corpora). ROCme! can display text, optionally with HTML formatting, images, sounds, and video.", "corpus, oral, linguistics, software" ]

[ "ressources terminologiques", "ressources lexicales", "liens lexicaux", "annotations contextuelles", "corpus spécialisé", "structure actancielle" ]

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

[ "Why compile terminological resources and why do it differently?", "In this talk, I will argue that terminological resources that account for the lexico-semantic properties of terms are a necessary – perhaps even indispensable – complement to other kinds of resources. Using examples taken from the field of the environment, I will first show that general lexical resources (including those that have a large coverage) do not give a complete picture of the meaning of terms or the lexical structure observed from the point of view of a special subject field. I will then proceed to show that terminological resources (thesauri, ontologies, term banks), that often take a conceptual approach, focus on the relationship between terms and the knowledge they convey and give little information about the linguistic behavior of terms. I will present a type of resource that describes the lexico-semantic properties of terms (argument structure, lexical relationships, contextual annotations, etc.) and some methodological considerations about its compilation.", "terminological resources, lexical resources, lexical relationships, specialized corpora, argument structure, contextual annotations" ]

[ "Apprentissage par Imitation", "Étiquetage de Séquences", "Apprentissage Structuré" ]

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

[ "Imitation learning for sequence labeling: towards a formalization of easy-first labeling methods", "Structured learning techniques, aimed at modeling structured objects such as labeled trees or strings, are computationally expensive. Many attempts have been made to reduce their complexity, either to speed up learning and inference, or to take richer dependencies into account. These attempts typically rely on approximate inference techniques and usually provide very little theoretical guarantee regarding the optimality of the solutions they find. In this work we study a new formulation of structured learning where inference is primarily viewed as an incremental process along which a solution is progressively computed. This framework generalizes several structured learning ap- proaches. Building on the connections between this framework and reinforcement learning, we propose a theoretically sound method to learn to perform approximate inference. Experiments on four sequence labeling tasks show that our approach is very competitive when compared to several strong baselines.", "Imitation Learning, Structured Learning, Sequence Models" ]