Search

International audience; We know that word embeddings trained using neural-based methods (such as word2vec SGNS) are sensitive to stability problems and that across two models trained using the exact same set of parameters, the nearest neighbors of a word are likely to change. All words are not equally impacted by this internal instability and recent studies have investigated features influencing the stability of word embeddings. This stability can be seen as a clue for the reliability of the semantic representation of a word. In this work, we investigate the influence of the degree of concreteness of nouns on the stability of their semantic representation. We show that for English generic corpora, abstract words are more affected by stability problems than concrete words. We also found that to a certain extent, the difference between the degree of concreteness of a noun and its nearest neighbors can partly explain the stability or instability of its neighbors.

This paper presents the first results of a multidisciplinary project, the “Evolex” project, gathering researchers in Psycholinguistics, Neuropsychology, Computer Science, Natural Language Processing and Linguistics. The Evolex project aims at proposing a new data-based inductive method for automatically characterising the relation between pairs of french words collected in psycholinguistics experiments on lexical access. This method takes advantage of several complementary computational measures of semantic similarity. We show that some measures are more correlated than others with the frequency of lexical associations, and that they also differ in the way they capture different semantic relations. This allows us to consider building a multidimensional lexical similarity to automate the classification of lexical associations.

International audience; This article presents an investigation of the role of social relations in the writing of scientific articles through the study of in-text citations. Does the fact that the author of an article knows the author whose work he or she cites have an impact on the context of the citation? Since citations are commonly used as criteria for research evaluation, it is important to question their social background to better understand how it impacts textual features. We studied a collection of science articles (N=123) from five disciplines and interviewed their authors (N=84) in order to: 1) identify the social relations between citing and cited authors; and 2) measure the correlation between a set of features related to in-text citations (N=6,956) and the identified social relations. Our pioneering work, mixing sociological and linguistic results, shows that social relations between authors can partly explain the variations of citations in terms of frequency, position and textual context.

The present study concentrates on Wikipedia talk pages, which are online discussions where the authors discuss the composition and content of Wikipedia articles. These pages provide new data for describing and analysing collaborative writing processes, which often involve conflicts. Previously, many studies have explored Wikipedia conflicts, highlighting opposite editing patterns in relation to cooperation, conflicts or quality. Most of these studies belong to the field of social sciences, and linguistic analyses are not very common in this context. Therefore, the linguistic characteristics of Wikipedia conflicts in talk pages are still little described in the literature. In this context, our objective is to analyse relevant linguistic cues which may help identify and characterize conflicts on Wikipedia talk pages. To this end, we apply two automatic methods. The first consists of the supervised automatic classification of conflicting vs. harmonic discussion threads. Secondly, we apply multidimensional analysis to the data to help profile the Wikipedia talk genre, enabling us to highlight key features and oppositions at a global level. The analyses are carried out on the WikiTalk corpus, a resource based on the French Wikipedia talk pages (160M words, 3M posts, 1M threads). The corpus includes a wide range of metadata, providing extra-linguistic characterization of the Wikipedia discussions.

International audience; Dans cet article, la question des effets du prescriptivisme est abordée sous l'angle de l'implantation terminologique. Plus précisément, nous proposons une méthode permettant d'apprécier l'enracinement des termes nouveaux dans l'usage, d'une part en déterminant diachroniquement le degré d'implantation des termes dans un domaine donné et, d'autre part, en observant les usages réels des locuteurs dans le but d'identifier des formes concurrentes. Nous examinons le cas de la terminologie officielle de l'économie et des finances, en nous fondant sur un corpus extrait du Web. Nous déployons une procédure automatique visant à quantifier l'emploi des termes recommandés et à repérer l'usage d'autres formes concurrentes non encore répertoriées.

International audience; In this paper, we present a resource that consists of query features associated with TREC adhoc collections. We developed two types of query features: linguistics features that can be calculated from the query itself, prior to any search although some are collection-dependent and post-retrieval features that imply the query has been evaluated over the target collection. This paper presents the two types of features that we have estimated as well as their variants, and the resource produced. The total number of features with their variants that we have estimated is 258 where the number of pre-retrieval and post-retrieval features are 81 and 171, respectively. We also present the first analysis of this data that shows that some features are more relevant than others in IR applications. Finally, we present a few applications in which these resources could be used although the idea of making them available is to foster new usages for IR.

International audience; This paper reports on an experiment implementing a data-intensive approach to discourse organisation. Its focus is on enumerative structures envisaged as a type of textual pattern in a sequentiality-oriented approach to discourse. On the basis of a large-scale annotation exercise calling upon automatic feature markup alongside manual annotation, we explore a method to identify complex discourse markers seen as configurations of cues. The presentation of the background to what is termed " multi-level annotation " is organised around four issues: linearity, complexity of discourse markers, top-down processing, granularity and the multi-level nature of discourse structures. In this context, enumerative structures seem to deserve scrutiny for a number of reasons: they are frequent structures appearing at different granularity levels, they are signalled by a variety of devices appearing to work together in complex ways, and they combine a textual role (discourse organisation) with an ideational role (categorisation). We describe the annotation procedure and experimental framework which resulted in nearly 1,000 enumerative structures being annotated in a diversified corpus of over 600,000 words. The results of two approaches to the rich data produced are then presented: firstly, a descriptive survey highlights considerable variation in length and composition, while showing enumerative structure to be a basic strategy resorted to in all three sub-corpora, and leads to a granularity-based typology of the annotated structures; secondly, recurrent cue configurations—-our " complex markers " —-are identified by the application of data mining methods. The paper ends with perspectives for further exploitation of the data, in particular with respect to the semantic characterisation of enumerative structures.

Nous présentons ici Webaffix, un outil qui permet de constituer et d’enrichir semi-automatiquement des données lexicales en utilisant le Web comme corpus. Il permet de détecter et d’analyser morphologiquement des unités lexicales nouvelles (c’est-à-dire absentes de listes de référence telles que les dictionnaires) construites par suffixation ou préfixation. Nous présentons les techniques utilisées par Webaffix, en déclinant les différents modes d’utilisation que nous avons envisagés et mis en pratique, ainsi que des exemples de résultats produits par diverses campagnes de collecte. Les données ainsi recueillies constituent des ressources lexicales pour différentes applications en traitement automatique des langues, mais également pour l’étude à grande échelle de la morphologie dérivationnelle., This paper deals with the design and use of Webaffix, a tool for semi-automatically detecting new word forms from the World Wide Web. We focus mainly on new derived words, i.e. coined from other lexemes through suffixation and/or prefixation processes. We develop the techniques and methods used in Webaffix, along with a sample of results obtained via several studies on French. Resources such as the ones created through the use of Webaffix are useful not only for natural language processing and information retrieval tasks, but also for the linguistic study of word creation.

International audience; Web vs dictionaries. French nominalizations in-age and in-ment. This article addresses French suffixations in-age and-ment that form nominalizations. Our aim is to test the hypotheses about these two rules, based on occurrences collected from the Web and not found in contemporary dictionaries. Through a wide-scale study, we show that the differences between the two rules are not as obvious as expected, and that this data pleads for a perspective in which both suffixes follow a single word formation rule.; Le présent travail porte sur les suffixations en -age et en -ment du français formant des nominalisations. Il a pour objectif de vérifier la validité des hypothèses auxquelles ces deux suffixations ont donné lieu, en les confrontant aux données relevées sur la Toile absentes des principaux dictionnaires de langue générale contemporains. À l'aide d'une étude en contexte à grande échelle, nous montrons que la distinction théorique opérée entre les deux procédés est loin d'être aussi nette qu'attendu et que ces données plaident en faveur d'une perspective d'un traitement des suffixes-age et-ment relevant d'une même règle de construction de lexèmes.