Datasets:

taln-ls2n

/

taln-archives

like 3

[ "apprentissage actif", "champs aléatoires conditionnels", "CRF", "apprentissage semi-supervisé", "test statistique de proportion" ]

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

[ "Strategies to select examples for Active Learning with Conditional Random Fields", "Nowadays, many NLP problems are modelized as supervised machine learning tasks. Consequently, the cost of the ex- pertise needed to annotate the examples is a widespread issue. Active learning offers a framework to that issue, allowing to control the annotation cost while maximizing the classifier performance, but it relies on the key step of choosing which example will be proposed to the expert. In this paper, we examine and propose such selection strategies in the specific case of Conditional Random Fields (CRF) which are largely used in NLP. On the one hand, we propose a simple method to correct a bias of certain state-of-the-art selection techniques. On the other hand, we detail an original approach to select the examples, based on the respect of proportions in the datasets. These contributions are validated over a large range of experiments implying several tasks and datasets, including named entity recognition, chunking, phonetization, word sens disambiguation.", "CRF, conditional random fields, active learning, semi-supervised learning, statistical test of proportion" ]

[ "Comptes-rendus hospitaliers", "extraction d'information", "apprentissage statistique" ]

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

[ "Risk factor identification for diabetic patients from clinical records using hybrid approaches", "In this paper, we present the methods we designed to process clinical records written in English. The aim of this study consists in identifying risk factors for diabetic patients and to define the temporal relation of those medical events wrt. the document creation time. Our approach relies (i) on HeidelTime to identify temporal expressions, (ii) on CRF and post-processing rules to identify treatments, diseases and risk factors, and (iii) on rules to determine the temporal relation of each medical event. On a corpus of 514 documents, we achieved a 0.8451 global F-measure. We observe we performed best on the identification of information mentionned in the text than information inference from lab results.", "Electronic Health Records, Information Extraction, Machine Learning" ]

[ "Étiquetage Morpho-Syntaxique", "Apprentissage Statistique", "Champs Markoviens Aléatoires" ]

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

[ "Ignore what you know to better learn what you don't : a case study on type constraints for CRFs", "When information about the possible outputs of a sequence labeling task is available, it may seem appropriate to include this knowledge into the system, so as to facilitate and speed-up learning and inference. However, we show in this paper that using such constraints at training time is likely to drastically reduce performance, even when they are both correct and useful at decoding. In this paper, we study this paradox and show that the lack of contrast induced by constraints leads to a form of under-fitting, that it is however possible to partially overcome.", "Part-of-Speech Tagging, Statistical Machine Learning, Conditional Random Fields" ]

[ "Analyse Temporelle", "Extraction d'Information", "Développement d'un Corpus Annoté" ]

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

[ "An analysis of temporal expressions in Electronic Health Records in French", "References to phenomena ocurring in the world and their temporal caracterization can be found in a variety of natural language utterances. For this reason, temporal analysis is a key issue in natural language processing. This article presents a temporal analysis of specialized documents. We use a corpus of documents contained in several de-identified Electronic Health Records to develop an annotated resource of temporal expressions relying on the TimeML standard. We then use this corpus to evaluate several methods for the automatic extraction of temporal expressions. Our best statistical model yields 0.91 F-measure, which provides significant improvement on extraction, over the state-of-the-art system Heidel-Time. We also compare our medical corpus to FR-Timebank in order to characterize the uses of temporal expressions in two different subdomains", "Information Extraction, Temporal Analysis, Development of Annotated Corpus" ]

[ "Compréhension automatique de la parole", "apprentissage sans données de référence", "espace sémantique continu", "données d'apprentissage hors domaine" ]

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

[ "Spoken language understanding without reference data", "Most recent state-of-the-art spoken language understanding models have in common to be trained on a potentially large amount of data. However, the required annotated corpora are not available for a variety of tasks and languages of interest. In this work, we present a novel zero-shot learning method for spoken language understanding which alleviate the need of any annotated or in-context data. Instead, it combines an ontological description of the target domain and the use of a continuous semantic space trained on large amounts of unannotated and unstructured found data with neural network algorithms. We show that this very low cost model can reach instantly performance comparable to those obtained by either state-of-the-art carefully hand crafted rule-based or trained statistical models on reference spoken language understanding tasks (test sets of the second and the third Dialog State Tracking Challenge, DSTC2,DSTC3). Eventually we extend the approach with an online adaptative strategy allowing to refine progressively the initial model with only a light and adjustable supervision.", "Spoken language understanding, continuous semantic space, zero-shot learning, out-of-domain training data" ]

[ "désambiguïsation d'entités", "modèle génératif", "Événements", "échantillonnage de Gibbs" ]

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

[ "Entity disambiguation for event template induction", "In this paper, we present an approach for event induction with a generative model. This model makes possible to consider more relational information than previous models, and has been applied to noun attributes. By their influence on parameter estimation, this new information make probabilistic topic distribution more discriminative and more robust. We evaluated different versions of our model on MUC-4 datasets.", "Event Induction, Generative Model, Entity Disambiguation, Gibbs Sampling" ]

[ "désambiguïsation lexicale", "langues peu dotées", "traduction automatique", "clarification de texte", "portage d'annotations" ]

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

[ "Rapid Construction of Supervised Word Sense Disambiguation System for Lesser-resourced Languages", "We introduce a method to quickly build a Word Sense Disambiguation (WSD) system for a lesser-resourced language L, under the condition that a Statistical Machine Transation system (SMT) is available from a well resourced language where semantically annotated corpora are available (here, English) towards L. We argue that it is less difficult to obtain the resources mandatory for the development of an SMT system (parallel-corpora) than it is to create the resources necessary for a WSD system (semantically annotated corpora, lexical resources). In the present work, we propose to translate a semantically annotated corpus from English to L and then to create a WSD system for L following the classical supervised WSD paradigm. We demonstrate the feasibility and genericity of our proposed method by translating SemCor from English to Bangla and from English to French. SemCor is an English corpus annotated with Princeton WordNet sense tags. We show the feasibility of the approach using the Multilingual WSD task from Semeval 2013.", "clarification of texts, word sens disambiguation, under resourced languages, machine translation, annotation transfert" ]

[ "similarité sémantique", "classification supervisée", "Analyse d'opinion" ]

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

[ "A Weakly Supervised Approach for Aspect-Based Opinion Mining in a Specific Domain", "The goal of aspect-based opinion mining is to associate an opinion with fine-grain aspects of a given product. Most approaches designed in this purpose use unsupervised techniques, whereas the information of the desired targeted aspects can often be given by the end-users. We propose in this paper a new approach for targeted opinion detection that uses this minimal information, enriched using several semantic similarty measures, along with topical segmentation and supervised classification. We prove the interest of the approach on an evaluation corpus in the specific domain of video games.", "Opinion analysis, classification, semantic similarity" ]

[ "Sémantique de corpus", "Fouille d'opinion", "Analyse des sentiments", "Textométrie" ]

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

[ "A method of corpus semantics applied to opinion mining and sentiment analysis: the impact of dialogical and dialectical features on the expression of subjectivity", "This paper first aims to provide a critical overview of the relationship between NLP and linguistics, and then to sketch out a discussion on the possible contribution of corpus semantics in an application-based context based on several subjective text mining studies (sentiment analysis and opinion mining). These studies break away from traditional approaches founded on the detection of axiological markers. Instead, they use explicit criteria related to the representation of actors (dialogical component) and argumentative or narrative structures (dialectical component). We hope to highlight the benefit of a methodological dialogue between theory (text semantics), meaning-oriented methods of corpus linguistics (i.e. textometrics) and NLP current practices in terms of algorithmic (machine learning) and assessment methodology.", "Textometry, corpus semantics, opinion mining, sentiment analysis" ]

[ "similarité", "voisinage sémantique", "classification de termes" ]

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

[ "Semantic homogeneity estimation for MCQs", "Semantic homogeneity states that terms are semantically close but not similar. This notion is the focus of work related to multiple-choice test generation, and especially to automatic distractor selection. In this paper, we present a method to estimate semantic homogeneity within a framework of automatic distractor validation. This method is based on a combination of several criteria of semantic relatedness and similarity between terms, by machine learning. We will show that our method allows to obtain a better estimation of semantic homogeneity than methods proposed in related work.", "similarity, semantic relatedness, term ranking" ]

[ "relations sémantiques", "définition", "hyperonymie", "synonymie", "méronymie", "patrons lexico-syntaxiques", "prédication", "distributionnalisme" ]

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

[ "Automatic Extraction of Semantic Relations from Definitions : en experiment in French with an Hybrid Approach", "This article presents an experiment to extract semantic relations from definitions. It is based on approximately one million definitions from two general dictionaries (Trésor de la Langue Française, French Wiktionary) and from the collaborative Wikipedia. We aim at extracting from these data several semantic relations : hyperonymy, synonymy, meronymy and other semantic relations. The methodological approach combines the precision of lexico-syntactic patterns and the recall of statistical analysis. After a survey of the state-of-the-art methods in this area, we detail our system and give the overall outcomes : about 900 000 hypernymy and 100 000 synonymy relations are extracted with a precision above 90% on a sample of 500 pairs for each relation. About 2 millions of definitory predicates are also extracted.", "semantic relations, lexico-syntactic patterns, distributionnalism, predication, hypernymy, synonymy, meronymy, definition" ]

[ "thésaurus distributionnels", "similarité sémantique", "Sémantique lexicale" ]

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

[ "Downgrading non-semantic neighbors for improving distributional thesauri", "Most of the methods for improving distributional thesauri focus on the means – representations or similarity measures – to detect better semantic similarity between words. In this article, we propose a more indirect approach focusing on the identification of the neighbors of a thesaurus entry that are not semantically linked to this entry. This identification relies on a discriminative classifier trained from unsupervised selected examples for building a distributional model of the entry in texts. Its bad neighbors are found by applying this classifier to a representative set of occurrences of each of these neighbors. We evaluate more particularly the interest of this method for thesauri built from syntactic co-occurrents and we show the interest of associating this method with those of (Ferret, 2013b) following an ensemble strategy.", "Lexical semantics, semantic similarity, distributional thesauri" ]

[ "grammaire", "TAG", "D-STAG", "ACG", "discours", "grammaire d'arbres adjoints", "grammaire catégorielle abstraite", "sémantique", "Syntaxe", "D-LTAG" ]

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

[ "Sentential and Discourse TAG-Based Grammars: An ACG Approach to D-STAG", "This article presents a method to interface a sentential grammar and a discourse grammar without resorting to an intermediate processing step. The method is general enough to build discourse structures that are direct acyclic graphs (DAG) and not only trees. Our analysis is based on Discourse Synchronous TAG (D-STAG), a Tree-Adjoining Grammar (TAG)-based approach to discourse. We also use an encoding of TAG into Abstract Categorial Grammar (ACG). This encoding allows us to express a higher-order semantic interpretation that enables building DAG discourse structures on the one hand, and to smoothly integrate the sentential and the discourse grammar thanks to the modular capability of ACG. All the examples of the article have been implemented and may be run and tested with the appropriate software.", "Syntax, semantics, discourse, grammar, Tree-Adjoining Grammar, TAG, D-LTAG, D-STAG, Abstract Categorial Grammar, ACG" ]

[ "réécriture de phrase", "variations sémantiques", "fonction noyau", "reconnaissance de paraphrases", "implication textuelle" ]

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

[ "Enriching String Rewriting Kernels With Lexico-semantic Types", "Many high level natural language processing problems can be framed as determining if two given sentences are a rewriting of each other. One way to solve this problem is to learn the way a sentence rewrites into another with kernel-based methods, relying on a kernel function to measure the similarity between two rewritings. While a wide range of rewriting kernels has been developed in the past, they often do not allow the user to provide lexico-semantic variations of words, which could help capturing a wider class of rewriting rules. In this paper, we propose and implement a new class of kernel functions, referred to as type-enriched string rewriting kernel, to address this lack. We experiment with various typing schemes on two natural sentence rewriting tasks, paraphrase identification and recognizing textual entailment.", "kernel methods, semantic variations, sentence rewriting, paraphrase identification, textual entailment" ]

[ "paraphrase", "analyse morphologique", "compréhension", "Domaines de spécialité", "terminologie médicale", "composition" ]

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

[ "Automatic extraction of layman paraphrases for medical terms", "We all have health concerns and sensibility to health information available in the modern society through modern media, such as scientific research, health social media, clinical documents, TV and radio broadcast, or novels. However, medical area conveys very specific notions (e.g., blepharospasm, alexitymia, appendicectomy), which are difficult to understand by people without medical training. We propose an automatic method for the acquisition of paraphrases for technical medical terms. We expect that such paraphrases are easier to understand than the original terms. The method is based on the morphological analysis of terms, syntactic analysis of texts, and text mining of non specialized texts. An analysis of the results and their evaluation indicate that such paraphrases can indeed be found in non specialized documents and show easier understanding level. According to the setting of the method, precision of the extractions ranges between 86 and 55%. This kind of resources is useful for several Natural Language Processing applications (e.g., information retrieval for lay people, text readability and simplification, question and answering systems).", "Specialized Area, Medical Terminology, Compounds, Morphological Analysis, Paraphrasis, Understanding" ]

[ "ancien français", "apprentissage automatique", "étiquetage morpho-syntaxique", "exploration de corpus", "analyse en dépendance" ]

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

[ "Old French parsing : Which language properties have the greatest influence on learning quality ?", "This paper presents machine learning experiments for part-of-speech labelling and dependency parsing of Old French. Machine learning methods are used for the purpose of corpus exploration. The SRCMF Treebank is our reference data. The poorly standardised nature of the language used in this corpus implies that training data is heterogenous and quantitatively limited. We explore various strategies, based on different criteria (variability of the lexicon, Verse/Prose form, date of writing) to build training corpora leading to the best possible results.", "POS labelling, Dependency Parsing, Old French, machine learning, corpus exploration" ]

[ "attribution d'auteur", "multilinguisme", "répétitions maximales", "classification", "chaînes de caractères" ]

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

[ "Authorship Attribution through Character Substrings (and vise versa)", "This article tackles the Authorship Attribution task according to the language independence issue. We propose an alternative of variable length character n-gram features in supervised methods : maximal repeats in strings. When character n-grams are by essence redundant, maximal repeats are a condensed way to represent any substring of a corpus. Our experiments show that the redundant aspect of character n-grams contributes to the efficiency of character-based Authorship Attribution techniques. Therefore, we introduce a new way to weight features in vector based classifier by introducing n-th order maximal repeats (maximal repeats detected in a set of maximal repeats). The experimental results show higher performance with maximal repeats, with less data than n-grams based approach.", "authorship attribution, multilinguism, classification, character substrings, maximal repeats" ]

[ "Wikipédia", "Indexation aléatoire", "Similarité sémantique", "Relation sémantique" ]

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

[ "Semantic similarity between sentences based on Wikipedia and Random Indexing", "This paper proposes a semantic similarity measure for sentence comparison based on the exploitation of Wikipedia as the only language resource. Such similarity measure is therefore usable for a wide range of languages, basically those covered by Wikipedia. Random Indexing is used to cope with the great dimensionality and the spareness of the data vectorial representations. Furthermore, a statistical weight function is used to reduce the noise generated by the use of a multi domain corpus such as Wikipedia. This semantic similarity measure has been evaluated on SemEval 2014 dataset for English language leading to very promising results, basically above the average level of the competing systems that exploit Wikipédia in conjunction with other sources of semantic information. It has been also evaluated on a set of pairs of sentences in French that we have build specifically for the task, and made freely available for the research community.", "Semantic Textual Similarity, Random indexing, Wikipédia, Semantic Relatedness" ]

[ "Typologie", "treebank", "syntaxe", "inférence de grammaire", "Grammaire de Propriétés" ]

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

[ "Automatic Linguistic Typology from Treebanks", "Linguistic typology studies different linguistic properties or phenomena in order to compare several languages or language families. We address in this paper the question of syntactic typology and propose a method for extracting automatically from treebanks syntactic properties, and bring them into a typology perspective. We present here the method and the different tools for inferring such information. The approach has been applied to 10 languages of the Universal Dependencies Treebank. We validate the results in showing how automatic classification corrélâtes with language families.", "Typology, syntax, grammar inference, Property Grammars" ]

[ "Mesures de confiance", "traduction automatique de la parole", "paramètres joints", "redécodage de graphe" ]

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

[ "Word confidence estimation for re-decoding speech translation graphs", "Word Confidence Estimation (WCE) for machine translation (MT) or automatic speech recognition (ASR) assigns a confidence score to each word in the MT or ASR hypothesis. In the past, this task has been treated separately in ASR or MT contexts and we propose here a joint estimation of word confidence for a spoken language translation (SLT) task involving both ASR and MT. We build robust word confidence estimators for SLT, based on joint ASR and MT features. Using these word confidence measures to re-decode the spoken language translation graph leads to a significant BLEU improvement (2 points) compared to the SLT baseline. These experiments are done for a French-English SLT task for which a corpus was specifically designed (this corpus being made available to the NLP community).", "Word confidence estimation (WCE), spoken language translation (SLT), joint features, search graph re-decoding" ]

[ "cognats", "Alignement multilingue", "corpus parallèles" ]

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

[ "Multi­alignment vs bi­alignment: the more languages the better", "In this paper, we propose an original method for performing the alignment of a multi­parallel corpus, ie a parallel corpus involving more than two languages, taking into account all the languages simultaneously (and not by merging a series of independent bi­alignments). To do this, we rely on the networks of lexical correspondences formed by identical chains and cognates (related words, and we show how various tiling of language pairs allow to exploit the surface similarities due to genetic relationships between languages. We evaluate our method compared to a conventional method of bi­alignment, and show how the multi­alignement achieves both more accurate and robust results.", "Multilingual alignment, parallel corpora, cognates" ]

[ "apprentissage discriminant", "Modèle neuronal de traduction", "traduction automatique par approche statistique" ]

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

[ "Discriminative Learning of Continuous Translation Models", "This paper proposes a new discriminative framework to train translation models based on neural network. This framework relies on the definition of a new objective function that allows us to introduce the evaluation metric in the learning process as well as to consider how the model interacts with the translation system. Moreover, this approach is compared with the state of the art estimation methods, such as the maximum likelihood criterion and the noise contrastive estimation. Experiments are carried out on the English to French translation task of TED Talks . The results show the efficiency of the proposed approach, whereas the initialization has a strong impact. We show that with a tailored initialization scheme significant improvements can be obtained in terms of BLEU scores.", "Neural network based translation model, statistical machine translation, discriminative learning" ]

[ "Interjections", "analyse de sentiments", "détection des émotions", "lexique affectif", "fouille d'opinions" ]

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

[ "Using interjections for emotion detection", "Although interjections have been recognized as linguistic phenomena for a long time, they have somehow been rarely studied and continue to be left aside in works dealing with microblogs. Users of this new kind of communication plateforms have popularized widely the use of linguistic constructions, like emoticons or interjections. In spite of their grammatic status and semantic richness for describing emotional states, interjections have been mostly ignored. In this article we show the importance of the role that interjections can play for detecting emotions. We detail how using interjections we have tagged automatically a French microblog corpus with emotion labels. Then we describe how we did deduce automatically from this corpus a fine-grained affective lexicon. The usefulness of the lexicon was evaluated in an emotion recognition task where, depending on the emotion, the F1-measure improvement ranged from +0.04 to +0.21.", "Interjections, emotion recognition, affective lexicon, sentiment analysis, opinion mining" ]

[ "plongements", "Analyse syntaxique en constituants lexicalisée", "réseaux de neurones" ]

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

[ "A Comparison of Neural Network Architectures for Constituent Parsing", "The article deals with lexicalized constituent parsing in a transition-based framework. Typical statistical approaches for this task are based on an unstructured representation of the lexicon. Words are represented by discrete unrelated symbols. Instead, our proposal relies on dense vector representations (embeddings) that are able to encode similarity between symbols : words, part-of-speech tags and phrase structure symbols. The article studies and compares 3 increasingly complex neural network architectures, which are fed symbol embeddings. The experiments suggest that the information given by embeddings is best captured by a deep architecture with a non-linear layer.", "Lexicalized constituent parsing, embeddings, neural networks" ]

[ "Paraphrase", "Reformulation", "Corpus oraux", "Marqueur de reformulation paraphrastique", "Apprentissage supervisé" ]

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

[ "...des conférences enfin disons des causeries... Automatic detection of segments with paraphrase relation in spoken corpora rephrasings", "Our work addresses automatic detection of segments with paraphrastic rephrasing relation in spoken corpus. The proposed approach is syntagmatic. It is based on paraphrastic rephrasing markers and the specificities of the spoken language. The reference data used are consensual. Automatic method based on machine learning using CRFs is proposed in order to detect the segments that are paraphrased. Different descriptors are exploited within a window with various sizes. The tests performed indicate that the segments that are in paraphrastic relation are quite difficult to detect. Our best average reaches up to 0.65 F-measure, 0.75 precision, and 0.63 recall. We have several perspectives to this work for improving the detection of segments that are in paraphrastic relation and for studying the data from other points of view.", "Spoken Corpora, Paraphrase, Reformulation, Paraphrastic Reformulation Marker, Supervised Learning" ]

[ "Traduction automatique", "caractéristiques à grande échelle", "ajustement du modèle" ]

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

[ "A Simple Discriminative Training Method for Machine Translation with Large-Scale Features", "The margin infused relaxed algorithm (MIRAs) dominates model tuning in statistical machine translation in the case of large scale features, but also they are famous for the complexity in implementation. We introduce a new method, which regards an N-best list as a permutation and minimizes the Plackett-Luce loss of ground-truth permutations. Experiments with large-scale features demonstrate that, the new method is more robust than MERT ; though it is only matchable with MIRAs, it has a comparatively advantage, easier to implement.", "machine translation, model tuning, large-scale features" ]

[ "Coq", "FraCas", "Inference en Langage Naturel", "Théorie des Types", "Sémantique Formelle", "Automatisation des Preuves" ]

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

[ "", "In this paper, we present the use of proof-assistant technology in order to deal with Natural Language Inference. We first propose the use of modern type theories as the language in which we translate natural language semantics to. Then, we implement these semantics in the proof-assistant Coq in order to reason about them. In particular we evaluate against a subset of the FraCas test suite and show a 95.2% accuracy and also precision levels that outperform existing approaches at least for the comparable parts. We then discuss the issue of automation, showing that Coq's tactical language allows one to build tactics that can fully automate proofs, at least for the cases we have looked at.", "Natural Language Inference, Type Theory, Formal Semantics, FraCas, Coq, Proof automation" ]

[ "polarité", "sentiment", "réseau lexical", "crowdsourcing", "GWAP" ]

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

[ "Do you like it? or not? LikeIt, a game to build a polarity lexical resource", "The ability to analyze the feelings that emerge from a text requires having a polarity lexical resource. In the lexical network JeuxDeMots we designed LikeIt, a GWAP that allows attributing a positive, negative or neutral value to a term, and thus obtaining for each term a resulting polarity. We present a quantitative analysis of polarity data obtained, together with the comparison method we developed to validate them qualitatively.", "polarity, feelings, lexical network, crowdsourcing, GWAP" ]

[ "verbes", "noms de médicaments", "forum", "contenu généré par l'utilisateur" ]

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

[ "Study of Drug-Introducing Verbs on Health Forums", "In this paper, we combine manual/automatic annotation to identify the verbs used by the users of a health forum to say that they are taking a drug. This information is important in many aspects, one of them being the identification of the relation between drugs and side effects : the mere mention of a drug in a message is not enough to assess that the user is taking this drug, and is thus likely to provide a feedback on it. In a second part of the study, we show how the set of verbs that we identified can be used to automatically extract variants of drug names. We assume that the analysis of the variants could shed light on patterns of mistakes that users make when spelling drug names and thus, improve medical information retrieval systems.", "user-generated content, forum, verbs, drug names" ]

[ "Réseau de neurones artificiels", "Allocation latente de Dirichlet", "Initialisation de poids" ]

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

[ "Neural Network Initialization using a Topic Space", "This paper presents a method for speech analytics that integrates topic-space based representation into an artificial neural network (ANN), working as a document classifier. The proposed method consists in configuring the ANN's topology and in initializing the weights according to a previously estimated topic-space. Setup based on thematic priors is expected to improve the efficiency of the ANN's weight optimization process, while speeding-up the training process and improving the classification accuracy. This method is evaluated on a spoken dialogue categorization task which is composed of customer-agent dialogues from the call-centre of Paris Public Transportation Company. Results show the interest of the proposed setup method, with a gain of more than 4 points in terms of classification accuracy, compared to the baseline. Moreover, experiments highlight that performance is weakly dependent to ANN's topology with the LDA-based configuration, in comparison to classical empirical setup.", "Artificial neural network, Latent Dirichlet allocation, Weights initialization" ]

[ "connecteurs de discours", "annotation discursive de corpus", "grammaire et discours" ]

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

[ "FDTB1 : Identification of discourse connectives in a French corpus", "This paper presents the manual identification of discourse connectives in the corpus FTB (French Treebank) already annotated for morpho-syntax. This is the first step in the full discursive annotation of this corpus. The method consists in projecting on the corpus the items that are listed in LexConn, a lexicon of French connectives, and then filtering the occurrences of these elements that do not have a discursive use. More than 10K connectives have been identified.", "discourse connectives, discourse annotation, grammar and discourse" ]

[ "résumé automatique", "similarité sémantique", "distance d'édition" ]

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

[ "", "We here propose a sentence edit distance metric, ROBO, based on Levenshtein distance. This metric distance is weighted by words frequency and operation type. We apply ROBO on an automatic summarization system whose sentence selection metrics are on purpose restricted to sentence similarity approaches. ROBO performance can then be evaluated indirectly.", "automatic summarization, semantic similarity, edit distance" ]

[ "classification", "films", "règles", "modèle de langage", "Wikipedia", "reconnaissance d'entités nommées" ]

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

[ "Named Entity Classification for Movie Titles", "In this article, we focus on contextual classification of named entities for « movie » type. Our work is part of an application framework which aims to identify, in a text, a movie title contained in a catalog (e.g. VoD catalog). To do this, we combine two approaches : we use a rule-based system, which has good accuracy. To increase recall we couple our system with a language model. The generation of training data for the language model from Wikipedia is a crucial part of this work. We show, through the evaluation of our system, the complementarity of approaches we use.", "named entity recognition, movies, classification, rules, language model, Wikipedia" ]

[ "Traduction Automatique Statistique", "Génération Automatique de Texte", "contexte phrastique", "terminologie bilingue" ]

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

[ "Statistical machine translation adaptation through terminological enrichment based on virtual phrase generation and filtering", "We propose a technique for adding bilingual terms to a phrase-based SMT system which includes not only individual words, but also induces phrasal contexts around these words. We first generate these contexts by generalizing patterns observed for similar words in a bilingual corpus, but then filter out those contexts that fall below a certain confidence threshold, based on an original phrase-pair selection process inspired by existing sentence selection techniques.", "Statistical Machine Translation, Natural Language Generation, phrasal context, bilingual terminology" ]

[ "mesure d'intérêt", "motifs syntaxiques", "Stylistique computationnelle", "fouille de texte" ]

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

[ "An Overrepresentation-based Interestingness Measure for Syntactic Stylistic Pattern Extraction", "In this contribution, we present a computational stylistic study of the French classic literature texts based on a data-driven approach where discovering interesting linguistic patterns is done without any prior knowledge. We propose an objective measure capable of capturing and extracting meaningful stylistic syntactic patterns from a given author's work. Our hypothesis is based on the fact that the most relevant syntactic patterns should significantly reflect the author's stylistic choice and thus they should exhibit some kind of overrepresentation behavior controlled by the author's purpose. The analysed results show the effectiveness in extracting interesting syntactic patterns from classic French literary text, and seem particularly promising for the analyses of such particular text.", "Computational stylistic, text mining, syntactic patterns, interestingness measure" ]

[ "Résumé automatique", "algorithme génétique", "modèles probabilistes" ]

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

[ "Automatic summarization using a genetic algorithm", "This paper proposes a novel method for automatic summarization based on a genetic algorithm that explores candidate summaries space following an objective function computed over ngrams probability distributions of the candidate summary and the source documents. This method does not consider a summary as a stack of independant sentences but as a whole text. We compare this method to one of the best existing methods which is based on integer linear programming, and show its efficiency on TAC 2009 corpus.", "automatic summarization, genetic algorithm, probabilistic models" ]

[ "ressource termino-ontologique", "apprentissage", "similarité" ]

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

[ "Identification of units of measures in scientific texts", "The work presented in this paper consists in identifying specialized terms (units of measures) in textual documents in order to enrich a onto-terminological resource (OTR). The first step permits to predict the localization of unit of measure variants in the documents. We have used a method based on supervised learning. This method permits to reduce significantly the variant search space staying in an optimal search context (reduction of 86% of the search space on the studied set of documents). The second step uses a new similarity measure identifying automatically variants associated with term denoting a unit of measure already present in the OTR with a precision rate of 82% for a threshold above 0.6 on the studied corpus .", "onto-terminological resource, learning, similarity" ]

[ "évaluation intrinsèque", "Nettoyage de pages Web", "détourage", "évaluation extrinsèque", "collecte de corpus" ]

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

[ "Intrinsic and extrinsic evaluation of boilerplate removal tools", "In this article, we tackle the problem of evaluation of web page cleaning tools. This task is seldom studied in the literature although it has consequences on the linguistic processing performed on web-based corpora. We propose two types of evaluation : (I) an intrinsic (content-based) evaluation with measures on words, tags and characters ; (II) an extrinsic (task-based) evaluation on the same corpus by studying the effects of the cleaning step on the performances of an NLP pipeline. We show that the results are not consistent in both evaluations. We show as well that there are important differences in the results between the studied languages. We conclude that the choice of a web page cleaning tool should be made in view of the aimed task rather than on the performances of the tools in an intrinsic evaluation.", "Web page cleaning, corpus collecting, intrinsic evaluation, extrinsic evaluation, web scraping" ]

[ "Fouille d'opinion", "corpus annoté", "web social", "extraction d'information semi-supervisée" ]

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

[ "CANÉPHORE : a French corpus for aspect-based sentiment analysis evaluation", "Aspect-based sentiment analysis knows a renewed interest these last years, according to recent opinion mining evaluation series (SemEval 2014 and 2015, DEFT 2015). However, publicly available evaluation resources are scarse. This work firstly introduces a publicly available annotated French Twitter corpus for sentiment analysis evaluation on aspect, subject and opinion word levels (10 000 documents). We present baseline results on this corpus for the task of opinion word extraction and then show that these results can be improved with simple semi-supervised methods.", "Opinion mining, social web, annotated corpus, semi-supervised information extraction" ]

[ "corpus spécialisé", "CRC", "collocations", "patrons de connaisssances" ]

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

[ "Knowledge-Rich Contexts Extraction in Specialized Corpora", "The term banks and dictionaries are valuable resources that improve access to knowledge in specialized domains. These resources are often relatively poor and do not always provide, for a given term, examples of its typicall use. In this context, we implement Knowledge-Rich Contexts (KRCs) to extract examples of contexts providing illustration of terms in specialized domain. We propose a unified framework to apply at the same time knowledge pattern and collocations with acceptable quality for human review.", "specialized corpus, KRC, knowledge patterns, collocations" ]

[ "appariement des rimes", "détermination des schémas de strophes", "métrique française", "corpus poétique", "calcul du mètre" ]

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

[ "Automatic Processing of Metrical Forms in Verse Texts", "The purpose of this paper is to present the project Anamètre. The project proposes automatic processing of metrical forms in French poetry and drama from the early seventeenth to the early twentieth century. Then we present the calculation program of meters on our corpus using a deterministic approach relying on the \"métricométric\" method of B. de Cornulier. Finally, we present the procedure of maching rimes and determination of rhyme schemes in periodic sequences and specific forms.", "French metrics, poetic corpus, calculation of meter, matching rhymes, determining of rhyme schemes" ]

[ "apprentissage", "corpus de dialogues", "détection de coréférences", "paires de mentions" ]

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

[ "Machine Learning for Coreference Resolution of Transcribed Oral French Data : the CROC System", "We present CROC (Coreference Resolution for Oral Corpus), the first machine learning system for coreference resolution in French. One specific aspect of the system is that it has been trained on data that are exclusively oral, namely ANCOR (ANaphora and Coreference in ORal corpus), the first corpus in oral French with anaphorical relations annotations. In its current state, the CROC system requires pre-annotated mentions. We detail the features that we chose to be used by the learning algorithms, and we present a set of experiments with these features. The scores we obtain are close to those of state-of-the-art systems for written English. Then we give future works on the design of an end-to-end system for oral and written French.", "Dialogue corpus, Coreference resolution, Machine learning, Mention-pair model" ]

[ "diagnostic d'ambiguïté", "extraction de mot-clés", "terminologie" ]

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

[ "Towards diagnosing ambiguity of candidate terms", "Researches in the field of Word Sense Disambiguation focus on identifying the precise meaning of a lexical unit found in a text. This article tackles another kind of problem : assessing the ambiguity of a lexical unit. In other words, we try to identify if a particular unit is ambiguous or not, we define this task as ambiguity diagnosis. Our evaluation dataset contains scientific articles where ambiguous words have been tagged by experts. In order to give an ambiguity diagnosis for each term, we use two types of features : POS tags and positions in the text. We show that the position of an occurrence in the text is a strong hint for such a task.", "ambiguity diagnosis, keyword extraction, terminology" ]

[ "indexation", "propagation", "réseau lexico-sémantique", "recherche d'information", "imagerie médicale" ]

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

[ "Index augmentation through propagation over a lexical network – application to radiological reports", "Medical data being increasingly computerized, semantically effective treatment of medical reports has become a necessity. The search of radiological images can be greatly facilitated through textual indexing of the associated reports. We present here an index enlargement algorithm based on spreading activations over a general lexical-semantic network.", "lexico-semantic network, propagation, indexation, information retrieval, medical imaging" ]

[ "apprentissage supervisé", "détection de l'ironie", "Analyse d'opinion" ]

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

[ "Automatic Irony Detection in French tweets", "This paper presents a supervised learning method for irony detection in tweets in French. A binary classifier uses both state of the art features whose efficiency has been empirically proved and new groups of features observed in our corpus. We focused on negation and explicit/implicit oppositions of opinions with different polarities. Results are encouraging.", "Opinion analysis, irony detection, supervised learning" ]

[ "dictionnaire", "morphologie", "français", "ressource linguistique", "corpus" ]

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

[ "French Contemporary Morphological Dictionaries : Morfetik Database, Elements of a Model for Computational Linguistics", "In this article, we present a morphological linguistic resource for Contemporary French called Morfetik. We first detail its composition, features and coverage. We compare it to other available morphological dictionaries for French (GLAFF, LEFF, Morphalou and Dicolecte). We then study its coverage on big corpora (French Wikipedia, French version of Wacky and Le Monde 10 years). We conclude with a proposition for updating the dictionary by connecting the resource with a continuously live corpus.", "dictionary, morphology, French language, linguistic resource, corpus" ]

[ "arabe", "Morphologie", "metagrammaire", "frame semantics" ]

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

[ "A metagrammar of the morphology-semantics interface in Arabic verbs", "In this article we propose to model the derivational morphology of Arabic using the metagrammatical framework of XMG. We demonstrate that treating abstract roots and patterns as semantically underspecified atomic morphemes offers an elegant way to account for the interaction between morphology and semantics.", "Morphology, Arabic, metagrammar, frame semantics" ]

[ "Corpus arboré", "Linguistique expérimentale", "Étiquetage automatique", "Analyse syntaxique automatique", "Parsing stochastique", "Conventions d'annotation", "Outils d'annotation" ]

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

[ "Creation of a new treebank with backcovers", "We introduce 4-couv, a treebank of approximatively 3 500 trees, built from a set of literacy backcovers. It has been automatically tagged and parsed, then manually corrected and validated. It was developed in the perspective of linguistic expriment projects, and aims to be compatible with other standard treebanks for french. We present in the following the corpus itself, then the tools we used or developed for the different stages of its elaboration : texts' selection, tagging, parsing, and manual correction.", "Treebank, Tagging, Parsing, Stochastic parsing, Annotation scheme, Annotation tools, Experimental linguistics" ]

[ "Centre de contact", "interaction", "analyse lexicale et syntaxique", "conversations tchat" ]

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

[ "Comparing Written and Spoken Languages: a Descriptive Study of Chat and Phone Conversations from an Assistance Contact Center", "In this article we propose a first descriptive study of a chat conversations corpus from an assistance contact center. The analysis includes lexical, syntactic and interactional dimensions. Transcriptions of telephone conversations from an assistance call center are also studied, allowing comparisons between these two interaction modes to be drawn. The study reveals significant differences in terms of conversation flow, with an increased efficiency for chat conversations in spite of longer temporal span. The lexical and syntactic analyses also reveal differences in terms of language level with a tighter proximity between agent and customers on the phone than for chats where the style adopted by agents is different from the style adopted by customers.", "Contact center, chat conversations, interaction, lexical and syntactic analysis" ]

[ "Ressources lexicales", "Théorie Sens-Texte", "Recherche d'Information" ]

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

[ "Lexical resource building and maintenance based on the use", "Our company develops a semantic search engine based on queries rephrasing. Our search engine relies on a lexicon we built on the basis of the Meaning-Text theory. We introduce our lexical resource and explain how we enrich and update it according to the needs we detect. We also mention the customization of the Meaning Text Theory to our needs.", "Lexical Ressources, Meaning-Text Theory, Information Retrieval" ]

[ "analyse de conversation humain/humain", "extraction automatique d'unités sémantiques pertinentes", "validation d'une ontologie" ]

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

[ "Use of Semantic Annotations for Validating Mentions of Semantic Hypotheses in Telephone Conversations", "The presented work focuses on the automatic extraction of semantic units and evaluation of their relevance to telephone conversations. The corpus used is DECODA corpus. The objective of the task is to enable automatic labeling theme of each conversation. Given the spontaneous nature of this type of conversations and the size of the corpus, we propose to use a semi-supervised strategy based on the construction of an ontology and a simple active learning : a human annotator analyses not only the lists of semantic units leading to the theme, but also studying a small amount of conversations. The relevance of the relationship between the conserved semantic units, sub-theme from the ontology and annotated theme is assessed by DNN, taking into account a vector representation of the document. The integration of semantic units included in the theme classification process improves performance.", "human/human conversation analysis, automatic extraction of relevant semantic units, ontology validation" ]

[ "adaptation", "analyse morpho-syntaxique", "langue de spécialité", "dossier électronique patient" ]

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

[ "Part of Speech tagging for specialized domains : a case study with clinical documents in French", "Part-of-Speech (PoS) tagging is a core task in Natural Language Processing, often used as a stepping stone to perform more complex tasks such as information extraction or machine translation. PoS tagging of specialized documents is often challenging due to the limited availability of tools and annotated corpora dedicated to specialized domains. Herein, we present the development of a PoS annotated corpus of clinical documents in French, using annotation guidelines from the FrenchTree Bank and Multitag datasets. Through analysis of the annotated corpus, we characterize the clinical domain, including specific targets for domain adaptation. We also show the limitations of a PoS tagger trained on news documents when applied to clinical text. We expect that the domain-specific resource presented in this paper will contribute to improve PoS tagging for clinical documents in French.", "domain adaptation, part-of-speech tagging, specialized domain, EHR" ]

[ "typologie", "multimédia", "liens inter-documents", "hypertexte", "presse" ]

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

[ "Towards a typology for linking newswire contents", "In this paper, we introduce a typology of possible links between contents of a multimedia news corpus. While several typologies have been proposed and used by the community, we argue that they are not adapted to rich and large corpora which can contain texts, videos, or radio stations recordings. We propose a new typology, as a first step towards automatically creating and categorizing links between documents' fragments in order to create new ways to navigate, explore, and extract knowledge from large collections. Several examples of links in a large corpus are given.", "typology, linking documents, hypertext, multimedia, newswire" ]

[ "Corpus français", "annotation en dépendances", "dépendances non-projectives" ]

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

[ "CDGFr, a Non-projective Dependency Corpus for French", "The non-projective cases, as a part of the dependency parsing of French, are often disregarded, mainly because the tree- banks on which parsers are trained contain little or no non-projective dependencies. In this paper, we present a new freely available dependency treebank for French that includes a substantial number of non-projective dependencies. This corpus can be used to study and process non-projectivity more effectively within the context of French dependency parsing.", "Treebank for French, dependency annotation, non-projective dependencies" ]

[ "réseaux de neurones récurrents", "Multilinguisme", "transfert crosslingue", "étiquetage morpho-syntaxique" ]

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

[ "Use of Recurrent Neural Network for Part-Of-Speech tags projection from a parallel corpus", "In this paper, we propose a method to automatically induce linguistic analysis tools for languages that have no labeled training data. This method is based on cross-language projection of linguistic annotations from parallel corpora. Our method does not assume any knowledge about foreign languages, making it applicable to a wide range of resource-poor languages. No word alignment information is needed in our approach. We use Recurrent Neural Networks (RNNs) as cross-lingual analysis tool. To illustrate the potential of our approach, we firstly investigate Part-Of-Speech (POS) tagging. Combined with a simple projection method (using word alignment information), it achieves performance comparable to the one of recently published approaches for cross-lingual projection.", "Multilingualism, cross-Lingual transfer, part-of-speech tagging, recurrent neural network" ]

[ "Titrage automatique", "Segmentation thématique", "Mesures de similarité", "Pondération Okapi" ]

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

[ "Automatic Topic Segmentation and Title Assignment in TV Broadcast News", "This paper addresses the task of assigning a title to topic segments automatically extracted from TV Broadcast News video recordings. We propose to associate to a topic segment the title of a newspaper article collected on the web at the same date. The task implies pairing newspaper articles and topic segments by maximising a given similarity measure. This approach raises several issues, such as the selection of candidate newspaper articles, the vectorial representation of both the segment and the articles, the choice of a suitable similarity measure, and the robustness to automatic segmentation errors. Experiments were made on various French TV Broadcast News shows recorded during one week, in conjunction with text articles collected through the Google News homepage at the same period. We introduce a full evaluation framework allowing to measure the quality of topic segment retrieval, topic title assignment and also joint retrieval and titling. The approach yields good titling performance and reveals to be robust to automatic segmentation.", "Topic segmentation, Title assignation, Okapi weighting, Similarity measures" ]

[ "système hybride", "Analyse de sentiments associés aux aspects", "SemEval2014" ]

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

[ "An Hybrid System for Aspect-Based Sentiment Analysis", "This paper details our participation to the SemEval2014 task 4 (Aspect Based Sentiment Analysis). We present the shared task, and then describe precisely our system, which is a combination of natural processing components and classificaton modules. We also present its evaluation results and the best system results. We finally expose some new experiments aiming at improving the system.", "Aspect Based Sentiment Analysis, SemEval2014, hybrid system" ]

[ "discours", "relations causales", "annotation de corpus" ]

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

[ "A corpus specifically annotated for causal discourse relations studies : the EXPLICADIS resource", "In order to offer a characterization of causal discourse relations, we created and annotated our own corpus : EXPLICADIS (EXPlanation and Argumentation in DIScourse). This corpus was built in the continuity of a readily available corpus, the ANNODIS corpus. Providing a more precise annotation of causal relations in a set of texts that are representative of multiple textual genres, EXPLICADIS is the first corpus of its kind built specifically for causal discourse relations studies.", "corpus annotation, discourse, causal relations" ]

[ "Sémantique distributionnelle", "compositionnalité", "flexion" ]

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

[ "Separating the lexical and grammatical components of semantic vectors", "In distributional semantics, the meaning of words is modelled by vectors that represent their distribution in a corpus. Vectorial models being often built from corpora with little linguistic pre-treatment, they do not represent very well the morphological compositionality of words. We propose here a method to decompose semantic vectors into lexical and inflectional vectors.", "Distributional semantics, compositionality, inflection" ]

[ "construction de lexique morphologique", "annotation et étiquetage de corpus", "linguistique diachronique" ]

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

[ "Treatments for Preclassic French parsing", "The \"Preclassical\" French language period extends throughout the sixteenth century and the first half of the seventeenth century. This state of the written French language, which accompanies the beginnings of printing, is relatively close to the modern French, but is characterized by a large graphic variability. It is one of the most underresourced state of the French language. Here we present the construction of a lexicon, a training corpus and a language model for the Preclassic period, built from modern French resources.", "morphological lexicon construction, corpus annotation and tagging, diachronic linguistics" ]

[ "classification de texte", "Fouille de séquences", "motifs libres", "sélection de descripteurs" ]

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

[ "Sequential pattern mining for text classification", "Most methods in text classification rely on contiguous sequences of words as features. Indeed, if we want to take non-contiguous (gappy) patterns into account, the number of features increases exponentially with the size of the text. Furthermore, most of these patterns will be mere noise. To overcome both issues, sequential pattern mining can be used to efficiently extract a smaller number of relevant, non-contiguous, features. In this paper, we compare the use of constrained frequent pattern mining and δ-free patterns as features for text classification. We show experimentally the advantages and disadvantages of each type of patterns.", "Sequence mining, free patterns, text classification, feature selection" ]

[ "génération automatique de texte multilingue ; collocations ; fonctions lexicales ; théorie sens-texte" ]

[ "M" ]

[ "The treatment of collocations in multilingual text generation", "In order to develop generic natural language generators that could be reused across languages and applications, it is necessary to model the core linguistic phenomena that one finds in language. One central phenomenon that remains problematic in NLP is collocations, such as heavy rain, strong preference or intense flavour, where the same idea (here, intensity), is expressed differently depending on the lexical unit it modifies. In explicative combinatory lexicography, collocations are modelled via lexical functions, which correspond to recurrent patterns of collocations. For instance, the three expressions above are all described with the function Magn : Magn(RAIN) = HEAVY, Magn(PREFERENCE) = STRONG , etc. There are hundreds of lexical functions. In this paper, we focus on the implementation of a subset of them that are used to model support verbs and some modifiers.", "multilingual natural language generation ; collocations ; lexical functions ; meaning-text theory" ]

[ "relations causales", "forums de santé", "Pharmacovigilance" ]

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

[ "Drugs that cure, drugs that make you sick : study of causal links to identify drug side effects", "In this paper, we study the textual manifestations of the relation between drugs and side effects in online health forums. Our goal is to find relevant linguistic cues in order to improve the automatic identification of side effects by leveraging the ambiguity between actual side effects and indications (the reason for drug use). We find that the use of discourse markers can be relevant for the identification of indications – a third of indication mentions follow markers like ‘pour' (‘for') or ‘dans le but de' (‘with the aim of') – while temporal informations are not as discriminating.", "Pharmacovigilance, Health Forums, Causal Links" ]

[ "graphe", "Sémantique distributionnelle", "sémantique lexicale", "terminologie", "sémantique des cadres" ]

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

[ "Exploring distributional semantic models using a 1-NN graph", "We show how a 1-NN graph can be used to explore the semantic neighbourhoods modeled by distributional models of semantics. We check whether the connected components of the graph, which represent sets of words that occur in similar contexts, can be used to identify sets of lexical units that evoke the same semantic frame. We then illustrate different ways in which the 1-NN graph can be used to explore a model or compare different models.", "Distributional semantics, lexical semantics, graph, terminology, frame semantics" ]

[ "Représentation vectorielle continue", "Réseau de neurones", "Information contextuelle", "Word2vec", "Modèle de langue" ]

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

[ "Contribution of temporal context information to a continuous vector representation of words", "Word embedding representations are gaining a lot of attention from researchers and have been successfully applied to various Natural Language Processing (NLP) tasks. In this paper, we propose to integrate temporal context information of words into the continuous bag-of-words (CBOW) and Skip-gram architectures for computing word-vector representations. Those architectures are shallow neural-networks. The CBOW architecture predicts a word given its context while the Skip-gram architecture predicts a context given a word. However, in those neural-networks, context windows are represented as bag-of-words. According to this representation, every word in the context is treated equally : the word order is not taken into account explicitly. As a result, each word will have the same influence on the network. We then propose an original method that integrates temporal information of word contexts using their relative position. This method is inspired from Continuous Context Models. The temporal information is treated as weights, in input by the CBOW and in the output layer by the Skip-Gram. The quality of the obtained models has been measured using a Semantic-Syntactic Word Relationship test set. Results showed that the incorporation of temporal information allows a substantial quality gain of 5 and 0.9 points respectively in comparison to the classical use of the CBOW and Skip-gram architectures.", "Neural network, Continuous vectorial representation, Contextual information, Word2vec, language model" ]

[ "tweets", "CRF", "étiquettage morpho-syntaxique" ]

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

[ "Part-of-speech Tagging for Tweets with CRFs", "We are insterested in this paper in training a part-of-speach tagger for tweets in English. We first propose a reduced tagset with 17 different tags, which allows better results in accuracy than traditional tagsets which contain 45 tags. Since we have few annoted tweets, we then try and overcome this difficulty by adding data from other more standard texts into the training set. The obtained models reach better results compared to models trained with only one corpus, whether coming of Twitter or of journalistic texts.", "tweets, CRFs, part-of-speech tagging" ]

[ "corpus comparables", "analyse linguistique", "discours scientifique et de vulgarisation", "classification selon le genre" ]

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

[ "Characterizing scientific genre for academia and general public: which properties should be taken into account?", "The article focuses on the study of a set of morpho-syntactic properties for audience-based classification. The linguistic analysis of academic discourse and of popular science discourse reveals that both discourse types are characterized by specific linguistic and textual properties. This research used two French comparable corpora in regards to genre and subject matter. The corpora was composed of scientific articles and popular science texts in the domains of medicine and computer science. The experiments performed as part of our study evaluated the influence of discourse-specific morpho-syntactic properties on genre-based classification, for scientific and popular science texts.", "linguistic analysis, academic and popular science discourse, comparable corpora, genre-based classification" ]

[ "comparaisons figuratives", "règles manuelles", "analyse syntaxique profonde", "comparé", "comparant", "analyse syntaxique de surface" ]

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

[ "Extraction and automatic analysis of comparative and pseudo-comparative structures for simile detection", "This article is focused on automatic simile detection and disambiguation. It describes an algorithm which uses syntactic chunks and handcrafted rules to extract and analyse similes in a given text. This algorithm, which was evaluated on a corpus of literary texts, performs better than a system based on dependency parsing.", "simile, tenor, vehicle, chunking, handcrafted rules, dependency parsing" ]

[ "Community Management", "Multilinguisme", "Textométrie", "Fouille de texte" ]

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

[ "Methodological Proposal for Automatic Detection of Community Manager. Multilingual Study based on a Junk Food corpus", "This article describes the methodology for identifying a certain kind of speech in internet forums. The detection of the speech of a Community Manager combines recent issues in the domain of Natural Language Processing, including opinion mining and sentiment analysis, with another more abstract problem. Going beyond detecting the polarity of a message, this project targets the underlying intentions and identity of the author of the message on the forum.", "Community Management, Textometry, Multilingualism, Data Mining" ]

[ "Alignement de textes", "Génétique textuelle", "Détection d'homologies dans les séquences textuelles" ]

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

[ "MEDITE: text alignment software for the study of textual genetics", "MEDITE is an alignment software able to identifying transformations between two versions of a same text. In this paper we show the theoretical and technical aspects of this tool.", "Text alignment, Textual genetics, Homology detection in text sequences" ]

[ "Extraction de réutilisations", "analyse littéraire assistée par ordinateur", "empreintes digitales textuelles" ]

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

[ "Phoebus: a Reuse Extraction Software for Literary Text", "Phoebus is a reuse extraction software for literary text. It was developed as a computer-assisted literary analysis tool. In this context, the software automatically detects and explores textual reuse networks in classical literature.", "Reuse Extraction, textual fingerprintng, computer-assisted literary analysis" ]

[ "Qualité des données", "Applications multilingues", "Classification", "Extraction d'information", "Fouilles de données textuelles", "Recherche d'information", "Ressources du langage", "Données Ouvertes", "Données légales" ]

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

[ "TermLis : a logical information context for terminological resources", "We present TermLis a logical information context constructed from terminological resources available in XML (FranceTerme), for a flexible use with a logical context system (CAMELIS). A logical view of a context allows to explore information in a flexible way, without writing explicit queries, it may also provide insights on the quality of the data. Such a context can be enriched by other information (of diverse natures), it can also be linked with other applications (according to arguments supplied by the context). We show how to use TermLis and we illustrate, through this concrete realization from FranceTerme data, the advantages of such an approach with terminological data.", "Multilingual applications, Classification, Information extraction, Textual data mining, Information retrieval, Linguistic resources, Open Data, Information Quality, Legal Information" ]

[ "Plateforme de veille sur internet", "médias sociaux", "analyse d'opinion", "fouille de données" ]

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

[ "Study the brand image of entities as part of a social media-monitoring platform", "The work presented here is about a Web monitoring software providing powerful tools for the analysis of opinions expressed in social media about an entity, such as a celebrity, a company or a brand. The implemented tools result from the research ANR project ImagiWeb involving several industrial and academic partners.", "Web monitoring software, social media, opinion analysis, data mining" ]

[ "Entité nommée", "projection cross-lingue", "corpus parallèle" ]

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

[ "Building a Bilingual Vietnamese-French Named Entity Annotated Corpus through Cross-Linguistic Projection", "The creation of high-quality named entity annotated resources is time-consuming and an expensive process. Most of the gold standard corpora are available for English but not for less-resourced languages such as Vietnamese. In Asian languages, this task is remained problematic. This paper focuses on an automatic construction of named entity annotated corpora for Vietnamese-French, a less-resourced pair of languages. We incrementally apply different cross-projection methods using parallel corpora, such as perfect string matching and edit distance similarity. Evaluations on Vietnamese –French pair of languages show a good accuracy (F-score of 94.90%) when identifying named entities pairs and building a named entity annotated parallel corpus.", "Named entity, parallel corpus, cross-projection" ]

[ "Graphes", "Corpus", "Syntaxe de dépendances" ]

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

[ "Online Graph Matching", "We present an online tool for graph pattern matching in syntactically annotated corpora.", "Dependency Syntax, Corpus, Graph matching" ]

[ "Patient virtuel", "système de dialogue", "langage spécialisé", "langage grand public" ]

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

[ "An Interactive Virtual Patient", "This paper describes the work-in-progress prototype of a dialog system that simulates a virtual patient consultation. We describe the general architecture and specifically the mapping between technical and lay terms in the medical domain.", "Virtual patient, dialog system, specialised language, lay language" ]

[ "corpus", "corpus arborés", "environnement d'étude de corpus" ]

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

[ "Integration of the TALN proceedings treebank to the ScienQuest platform", "This demonstration shows the integration of the TALN proceedings Treebank to the ScienQuest platform. This platform was initially created for the study of the Scientext scientific texts corpus. This integration takes into account the metadata to the TALN corpus, and was done in an effort to reconcile these two corpora's sets of labels, and to convert for the TALN corpus the predefined queries designed for the Scientext corpus, in order to easily perform similar queries on the two corpora.", "corpora, treebanks, corpus study environment" ]

[ "aide à la communication", "prédiction sémantique", "génération de texte", "frame semantics", "handicap" ]

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

[ "An Augmentative and Alternative Communication with semantic prediction", "This demo shows a mobile app (for smartphone and tablet) for people with language and/or speech disorders for generating sentences from the combination of icons and verbalize text generated by Text-To-Speech (TTS). The main criticism expressed by patients using existing solutions is the time too long required to compose a sentence. Hence, existing solutions are hard to use in dialogic conditions. To allieviate this problem, we have developed a text generation engine with semantic prediction. This engine only proposes to user the relevant pictograms in view of the current entry (e.g. after the pictogram [eat], the application offers pictograms [apple] or [meat] corresponding to edible concepts). We have multiplied from 5 to 10 the speed of composing a sentence compared to existing solutions.", "natural language generation, semantic prediction, frame semantics, disability, AAC" ]

[ "analyse sémantique", "TIER", "SGRM", "Système de Gestion de Règles Métier" ]

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

[ "An Expert System Based on a Semantic Analysis for Identifying Biological Threats", "The European project TIER (Integrated strategy for CBRN – Chemical, Biological, Radiological and Nuclear - Threat Identification and Emergency Response) aims at developing a comprehensive and integrated strategy for emergency response in case of chemical, biological, radiological and nuclear danger, as well as explosives use, based on threat identification and risk assessment. In this article, we focus on the biological risks. We introduce our business rules management system based on a semantic analysis, that enables the extraction of structured data from unstructured data with the aim to make reasoning.", "TIER, BRMS, Business Rules Management System, semantic analysis" ]

[ "corpus oraux", "annotation morphosyntaxique", "unités poly-lexicales", "disfluences" ]

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

[ "DisMo: a multi-level annotator for spoken language", "In this demonstration we present the multi-level automatic annotator DisMo which is specifically designed for the challenges posed by spoken language corpora. Its output comprises of part-of-speech tagging, lemmatization, multi-word unit detection, detection of disfluency phenomena and discourse markers.", "part-of-speech tagging, spoken corpora, disfluencies, multi-word expressions" ]

[ "marqueurs multi-polaires", "Fouille d’opinion", "adaptation au domaine" ]

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

[ "Study of Domain Dependant Multi-Polarity Words for Document Level Opinion Mining", "In this article, we propose a study on the words or multi-words which are good indicators of the opinion polarity of a text but have different polarity depending on the domain. We have performed experiments on French and English corpora, which show that taking these multi-polarity words into account improve the opinion classification at text level in a domain adaptation framework. We also show that these words are useful when the corpus contains several domains. If these domains are not explicit, using a automatic domain characterization (e.g. wich Topic Modeling approaches) allows to achieve the same results.", "Opinion mining, domain adaptation, multi-polarity markers" ]

[ "domaines de spécialité", "Extraction de termes-clés", "articles scientifiques", "méthodes non-supervisées" ]

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

[ "The impact of domains for Keyphrase extraction", "Keyphrases are single or multi-word expressions that represent the main content of a document. Keyphrases are useful in many applications such as document indexing or text summarization. However, most documents are not provided with keyphrases. To tackle this problem, researchers propose methods to automatically extract keyphrases from documents of various nature. In this paper, we focus on the difficulty of automatic keyphrase extraction in scientific papers from various areas. Using five corpora representing five areas (archaeology, linguistics, information sciences, psychology and chemistry), we observe the difficulty scale and analyze factors inducing a higher or a lower difficulty", "Keyphrase extraction, scientific papers, specific domain, unsupervised methods" ]

[ "Extraction d’information", "extraction de rôles événementiels", "modèles de langage neuronaux" ]

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

[ "Event Role Labelling using a Neural Network Model", "Information Extraction systems must cope with two problems : they heavily depend on the considered domain but the cost of development for a domain-specific system is important. We propose a new solution for role labeling in the event-extraction task that relies on using unsupervised word representations (word embeddings) as word features. We automatically learn domain-relevant distributed representations from a domain-specific unlabeled corpus without complex linguistic processing and use these features in a supervised classifier. Our experimental results on the MUC-4 corpus show that this system outperforms state-of-the-art systems on this event extraction task, especially when the amount of annotated data is small.We also show that using word representations induced on a domain-relevant dataset achieves better results than using more general word embeddings.", "Information extraction, event role filler detection, neural language models" ]

[ "rôles sémantiques", "maximum d’entropie", "représentations distribuées" ]

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

[ "Using distributed word representations for robust semantic role labeling", "According to Frame Semantics (Fillmore 1976), words’ meaning are best understood considering the semantic frame they play a role in, for the frame is what gives them context. FrameNet defines about 1000 such semantic frames, along with the roles arguments can fill in this frame. Our task is to automatically label arguments’ roles, given their span, the frame, and the predicate, using maximum entropy models. We make use of distributed word representations to improve generalisation over the few training exemples available for each frame.", "semantic role labelling, distributed word representations" ]

[ "étiquetage morphosyntaxique", "descripteurs riches", "apprentissage automatique supervisé", "systèmes statistiques robustes" ]

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

[ "Better pos-tagging for \"que\" through targeted features and rules", "Robust statistical NLP tools, and in particular pos-taggers, often use knowledge-poor features, which are easily applicable to any language but do not look beyond 1 or 2 tokens to the right and left and do not make use of syntactic equivalence classes. Although pos-tagging tends to get high accuracy scores (around 97%), the remaining 3% errors systematically result in a 3% loss in parsing accuracy. Some of the easiest phenomena to target via the injection of richer linguistic knowledge are ambiguous function words, such as “que” in French. In this study, we attempt to improve the pos-tagging of “que” through the use of targeted knowledge-rich features during training, and symbolic rules which override the statistical model during analysis. We reduce the error rate by 45% using targeted knowledge-rich features, and 55% if we add rules.", "pos-tagging, supervised machine learning, knowledge-rich features, robust statistical systems" ]

[ "faisceaux", "Programmation Dynamique", "Treillis de mots", "Analyse syntaxique", "Analyse syntaxique par dépendances", "Couplage d’analyseurs" ]

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

[ "Playing with parsers", "We present DYALOG-SR, a statistical dependency parser developed for the SPRML 2013 shared task over 9 very different languages. DYALOG-SR implements a shift-reduce parsing algorithm (a la MALT), extended with beams and dynamic programming techniques. One of the specificities of DYALOG-SR is its ability to handle word lattices as input, which was used for handling Hebrew lattices and more recently French ones produced by SXPIPE. Having access to FRMG, an alternative parser for French, we also tried a coupling with DYALOG-SR, providing us the best results so far on the French TreeBank", "Parsing, Dependency Parsing, Beams, Dynamic Programming, Word Lattice, Parser coupling" ]

[ "réseau lexical", "système lexical", "Réseau Lexical du Français (RL-fr)", "fonction lexicale", "base de données lexicale", "structure du lexique", "graphe petit monde", "proxémie", "Lexicologie Explicative et Combinatoire" ]

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

[ "Principles of Lexical Network Systemic Modeling", "We introduce a new approach for manually constructing broad-coverage lexical ressources based on a specific type of lexical network called lexical system. Drawing on experience gained from the construction of the French Lexical Network (fr-LN), we begin by formally characterizing lexical systems as “small-world” graphs of lexical units that are primarily organized around the system of Meaning-Text lexical functions. We then give arguments in favor of the proposed model that are both theory- and application-oriented.", "lexical system, lexical database, structure of the lexicon, lexical network, small-world graph, proxemy, Explanatory Combinatorial Lexicology, lexical function, French Lexical Network (fr-LN)" ]

[ "complexité lexicale", "analyse morphologique", "mots gradués", "ressources lexicales" ]

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

[ "A model to predict lexical complexity and to grade words", "Analysing lexical complexity is a task that has mainly attracted the attention of psycholinguists and language teachers. More recently, this issue has seen a growing interest in the field of Natural Language Processing (NLP) and, in particular, that of automatic text simplification. The aim of this task is to identify words and structures which may be difficult to understand by a target audience and provide automated tools to simplify these contents. This article focuses on the lexical issue by identifying a set of predictors of the lexical complexity whose efficiency are assessed with a correlational analysis. The best of those variables are integrated into a model able to predict the difficulty of words for learners of French.", "lexical complexity, morphological analysis, graded words, lexical resources" ]

[ "annotation", "inférence", "radiologie", "réseau lexical" ]

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

[ "Annotations and inference of relations in a lexical semantic network : Applied to radiology", "Relations annotation and inference in a lexical-semantic network : application to radiology Domain specific ontologies are invaluable despite many challenges associated with their development. In most cases, domain knowledge bases are built with very limited scope without considering the benefits of plunging domain knowledge to a general ontology. Furthermore, most existing resources lack meta-information about association strength (weights) and annotations (frequency information like frequent, rare ... or relevance information (pertinent or irrelevant)). In this paper, we are presenting a semantic resource for radiology built over an existing general semantic lexical network (JeuxDeMots). This network combines weight and annotations on typed relations between terms and concepts. Some inference mechanisms are applied to the network to improve its quality and coverage. We extend this mechanism to relation annotation. We describe how annotations are handled and how they improve the network by imposing new constraints especially those founded on medical knowledge. We present then some results", "relation inference, lexical semantic network, relation annotation, radiology" ]

[ "Anaphores", "Pronom", "Correcteur", "Cooccurrences", "Saillance", "Approche multistratégique" ]

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

[ "Pronoun Anaphora Resolution for Automatic Correction of Grammatical Errors", "This article relates work done in order to expand the performance of a commercial French grammar checker. We try to achieve the correction of errors only detectable when an anaphora pronoun is linked with its antecedent. The algorithm searches for the antecedent within the same sentence as the pronoun as well as in previous sentences. It relies only slightly on morphology agreement, since it is what we are trying to correct, and uses instead information from a robust syntactic parsing and reliable cooccurrences. We give examples of our results on a vast corpus, and try to identify the key criteria for successful detection. We show that some types of corrections are precise enough to be integrated in a large scale commercial software.", "Grammar checker, Anaphora, Pronoun, Salience, Multi-Strategy Approach, Cooccurrences" ]

[ "chunker", "corpus oral", "étiquetage POS", "apprentissage automatique", "disfluences" ]

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

[ "Can we chunk well with bad POS labels?", "In this paper, we test two distinct approaches to chunk transcribed oral data, trying to minimize the phases of manual correction. First, we use an existing chunker, learned from written texts, then we try to learn a new specific chunker from a small amount of manually corrected labeled oral data. The purpose is to reach the best possible results for the chunker with as few manual corrections of the POS labels as possible. Our experiments show that it is possible to learn a new effective chunker for oral data from a labeled reference corpus of small size, without any manual correction of POS labels.", "chunker, POS labeling, machine learning, oral corpus, disfluencies" ]

[ "analogie", "normalisation textuelle", "correction orthographique" ]

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

[ "Analogy-based Text Normalization : the case of unknowns words", "Analogy-based Text Normalization : the case of unknowns words. In this paper, we describe and evaluate a system for improving the quality of noisy texts containing non-word errors. It is meant to be integrated into a full information extraction architecture, and aims at improving its results. For each word unknown to a reference lexicon which is neither a named entity nor a neologism, our system suggests one or several normalization candidates (any known word which has the same lemma as the spell-corrected form is a valid candidate). For this purpose, we use an analogybased approach for acquiring normalisation rules and use them in the same way as lexical spelling correction rules.", "Text normalization, Spell checking, Analogy" ]

[ "compositionalité", "sémantique distributionnelle", "sémantique lexicale" ]

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

[ "An evaluation of various methods for adjective-nouns composition", "In the course of the last two decades, numerous algorithms have sprouted up that successfully capture the semantics of single words by looking at their distribution in text, and comparing these distributions in a vector space model. However, it is not straightforward to construct meaning representations beyond the level of individual words – i.e. the combination of words into larger units – using distributional methods. Our contribution is twofold. First of all, we carry out a large scale evaluation, comparing different composition methods within the distributional framework for the case of adjective-noun composition, making use of a newly developed dataset. Secondly, we propose a novel method for adjective-noun composition, which is a generalization of the approach by Baroni & Zamparelli (2010). The performance of our novel method is equally evaluated on our new dataset.", "lexical semantics, distributional semantics, compositionality" ]

[ "génération", "G-TAG", "TAG", "grammaires catégorielles abstraites", "réalisation syntaxique" ]

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

[ "Text Generation: Reexamining G-TAG with Abstract Categorial Grammars", "G-TAG is a formalism dedicated to text generation. It relies on the Tree Adjoining Grammar (TAG) formalism and extends it with several specific notions allowing for the construction of a surface form from a conceptual representation. This conceptual representation is independent from the target language. The goal of this paper is to study G-TAG and its specific notions from the perspective given by Abstract Categorial Grammars (ACG). We use the reversibility property of ACG and the encoding of TAG they offer. We show that the key G-TAG notions of g-derivation tree and lexicalization are naturally expressed in ACG. The construction of surface forms can then rely on the general ACG algorithms and some operations that G-TAG is lacking can be freely accounted for.", "TAG, G-TAG, generation, syntactic realization, abstract categorial grammars" ]

[ "apprentissage partiellement supervisé", "transfert cross-lingue", "analyse morpho-syntaxique" ]

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

[ "Cross-Lingual POS Tagging through Ambiguous Learning: First Experiments", "When Part-of-Speech annotated data is scarce, e.g. for under resourced languages, one can turn to crosslingual transfer and crawled dictionaries to collect partially supervised data. We cast this problem in the framework of ambiguous learning and show how to learn an accurate history-based model. This method is evaluated on four languages and yields improvements over state-of-the-art for three of them, with gains up to 3.9% absolute or 35.8% relative.", "Weakly Supervised Learning, Part-of-Speech Tagging, Cross-Lingual Transfer" ]

[ "Corpus monolingue", "Corpus annoté", "Corpus comparable", "Mesure de la comparabilité", "Construction de corpus comparable", "Analyse morpho-syntaxique" ]

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

[ "Building Monolingual Comparable and Annotated Corpora: An experimental study from a pos tagged corpus", "This work is motivated by the will of creating a new part-of-speech annotated corpus in French from an existing one. In this context, we proprose a general and operational definition of the comparability relation between annotated monolingual corpora.We propose a comparability measure and a procedure to build semi-automatically a comparable corpus from a source one. We study the use of the perplexity (information theory motivated measure) as a way to rank the sentences to select for building a comparable corpus. We show that the measure can play a role but that it is not sufficient.", "Comparable corpus, Monolingual corpus, Annotated corpus,Measuring comparability, Building comparable corpus, Part-of-Speech tagging" ]

[ "Systèmes de Dialogue", "Traitement Incrémental", "Architecture des Systèmes de Dialogue" ]

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

[ "A simple approach to make dialogue systems incremental", "Incremental dialogue is at the heart of current research in the field of dialogue systems. Several architectures and models have been published such as (Allen et al., 2001; Schlangen & Skantze, 2011). This work has made it possible to understand many aspects of incremental dialogue, however, in order to implement these solutions, one needs to start from scratch as the existing architectures are inherently different. Our approach is different as it tends towards a new generation of incremental systems that behave incrementally but work internally in a traditional way. This paper suggests inserting a new module, called the Scheduler, between the service and the client. This Scheduler manages the asynchronous events, hence reproducing the behaviour of incremental systems from the client’s point of view. On the other end, the service does not work incrementally.", "Dialogue Systems, Incremental Processing, Dialogue Systems Architecture" ]

[ "Morphologie dérivationnelle", "Français", "Réseau lexical", "Famille morphologique", "Sémantique lexicale" ]

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

[ "The Démonette Lexical Database: between Constructional Semantics and Word Formation", "Démonette is a lexical database whose vertices (lexical entries) and edges (morphological relations between the vertices) are annotated with morpho-semantic information. It results from an original design incorporating two radically different approaches: Morphonette, a resource based on derivational analogies and DériF, an analyzer based on linguistic rules. However, Daemonette is not a simple merger of two pre-existing ressources: its architecture is fully compatible with the lexematic approach to morphology; its contents can be extended using data from various other sources. The article presents the Démonette model and the content of its current version, including 31,204 verbs, action nouns, agent nouns and property adjectives, where morphological links between both direct ascendants and indirectly related words have bi-oriented definitions. Finally, Démonette is assessed with respect to Verbaction with a recall of 84% and a precision of 90%.", "Lexical Network, Derivational morphology, Morphological family, Lexical semantics, French" ]

[ "thésaurus distributionnel", "recherche d’information", "graphe de k proches voisins", "fenêtre de Parzen", "algorithme hongrois", "Tnormes" ]

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

[ "Exploring the neighbor graph to improve distributional thesauri", "In this paper, we address the issue of building and improving a distributional thesaurus.We first show that existing tools from the information retrieval domain can be directly used in order to build a thesaurus with state-of-the-art performance. Secondly, we focus more specifically on improving the obtained thesaurus, seen as a graph of k-nearest neighbors. By exploiting information about the neighborhood contained in this graph, we propose several contributions. 1)We show how the lists of neighbors can be globally improved by examining the reciprocity of the neighboring relation, that is, the fact that a word can be close of another and vice-versa. 2) We also propose a method to associate a confidence score to any lists of nearest neighbors (i.e. any entry of the thesaurus). 3) Last, we demonstrate how these confidence scores can be used to reorder the closest neighbors of a word. These different contributions are validated through experiments and offer significant improvement over the state-of-theart.", "distributional thesaurus, k nearest neighbor graph, Parzen window, Hungarian algorithm, T-norms, information retrieval" ]

[ "dispersion des données", "textes de spécialité", "modèle d’espace vectoriel", "Analyse distributionnelle", "hyperonymie", "méthode hybride" ]

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

[ "Reducing data sparsity by generalising distributional contexts: application to specialised texts", "Vector space models implement the distributional hypothesis relying on the repetition of information occurring in the contexts of words to associate. However, these models suffer from a high number of dimensions and data sparseness in the matrix of contextual vectors. This is a major issue with specialized corpora that are of much smaller size and with much lower context frequencies.We tackle the problem of data sparseness on specialized texts and we propose a method that allows to make the matrix denser, by generalizing of distributional contexts. The evaluation of the method is performed on a French medical corpus, and shows that with a small graphical window and the Jaccard Index, the context generalization with lexico-syntactic patterns improves the results, while with a large window and the cosine measure, it is better to generalize with lexical inclusion.", "Distributional analysis, specialized texts, hypernymy, data sparseness, Vector Space Model, hybrid method" ]