Search

National audience; Dans cet article, nous présentons une approche de bout en bout d'extraction de concepts sémantiques de la parole. En particulier, nous mettons en avant l'apport d'une chaîne d'apprentissage successif pilotée par une stratégie de curriculum d'apprentissage. Dans la chaîne d'apprentissage mise en place, nous exploitons des données françaises annotées en entités nommées que nous supposons être des concepts plus génériques que les concepts sémantiques liés à une application informatique spécifique. Dans cette étude, il s'agit d'extraire des concepts sémantiques dans le cadre de la tâche MEDIA. Pour renforcer le système proposé, nous exploitons aussi des stratégies d'augmentation de données, un modèle de langage 5-gramme, ainsi qu'un mode étoile aidant le système à se concentrer sur les concepts et leurs valeurs lors de l'apprentissage. Les résultats montrent un intérêt à l'utilisation des données d'entités nommées, permettant un gain relatif allant jusqu'à 6,5 %. ABSTRACT Curriculum learning : named entity recognition for semantic concept extraction In this paper, we present an end-to-end approach for semantic concept extraction from speech. In particular, we highlight the contribution of a successive learning chain driven by a curriculum learning strategy. In the learning chain, we use French data with named entity annotations that we assume are more generic concepts than semantic concept related to a specific computer application. In this study, the aim is to extract semantic concept as part of the MEDIA task. To improve the proposed system, we also use data augmentation, 5-gram langage model and a star mode to help the system focus on concepts and their values during the training. Results show an interest in using named entity data, allowing a relative gain up to 6.5%. MOTS-CLÉS : Curriculum d'apprentissage, transfert d'apprentissage, bout en bout, extraction de concepts sémantiques, entités nommées.

International audience; Malgré les faiblesses connues de cette métrique, les performances de différents systèmes de reconnaissance automatique de la parole sont généralement comparées à l'aide du taux d'erreur sur les mots. Les transcriptions automatiques de ces systèmes sont de plus en plus exploitables et utilisées dans des systèmes complexes de traitement automatique du langage naturel, par exemple pour la traduction automatique, l'indexation, la recherche documentaire... Des études récentes ont proposé des métriques permettant de comparer la qualité des transcriptions automatiques de différents systèmes en fonction de la tâche visée. Dans cette étude nous souhaitons mesurer, qualitativement, l'apport de l'adaptation automatique des modèles de langage au domaine visé par un cours magistral. Les transcriptions du discours de l'enseignant peuvent servir de support à la navigation dans le document vidéo du cours magistral ou permettre l'enrichissement de son contenu pédagogique. C'est à-travers le prisme de ces deux tâches que nous évaluons l'apport de l'adaptation du modèle de langage. Les expériences ont été menées sur un corpus de cours magistraux et montrent combien le taux d'erreur sur les mots est une métrique insuffisante qui masque les apports effectifs de l'adaptation des modèles de langage.

National audience

International audience; Nous nous intéressons, dans cet article, à la tâche d'analyse d'opinions en arabe. Nous étudions la spécificité de la langue arabe pour la détection de polarité. Nous nous focalisons ici sur les caractéristiques d'agglutination et de richesse morphologique de cette langue. Nous avons particulièrement étudié différentes représentations d'unité lexicale : token, lemme et light stemme. Nous avons construit et testé des espaces continus de ces différentes représentations lexicales. Nous avons mesuré l'apport de tels types de representations vectorielles dans notre cadre spécifique. Les performances du réseau CNN montrent un gain significatif de 2\% par rapport à l'état de l'art.

International audience

National audience; The ability to deal with one’s emotions in professional situations that involve human relations is generally taught with role playing. However, role playing is expensive (comedians) or difficult to set up if the students are play-ing a role that is not theirs yet. In order to improve the training offered to Edu-cation Counselors (ECs), we propose the TGRIS virtual reality simulator, that allows ECs to naturally discuss with a virtual novice teacher (i.e. simulated by an embodied conversational agent). This tool has the particularity of using ex-changes extracted from real problematic counselling situations. In addition, the verbal and non-verbal responses as well as the emotional state of the virtual novice teachers are piloted by the teacher. This research explores how this sys-tem allows ECs to learn how to dialogue with novice teachers who resist to their help. The experimentation lead with 12 ECs shows the potential of instru-mented training with TGRIS.; Se former à la gestion des émotions dans les métiers de l'humain passe généralement par des jeux de rôle. Il est difficile de systématiser ce pro-cédé car il est coûteux (comédiens) où difficile à mettre en place si les étudiants jouent un rôle en devenir pour eux. Pour la formation de conseillers pédago-giques de l'enseignement (CP), nous proposons TGRIS, un outil de simulation d'entretien, en Réalité Virtuelle, qui permet aux CP de discuter naturellement avec un enseignant débutant virtuel (i.e. simulé par un agent conversationnel animé). Cet outil présente la particularité d'utiliser des échanges prélevés en situation réelle d'entretien de conseil problématique. De plus, les réactions ver-bales, non-verbales et l'état émotionnel de l'enseignant débutant virtuel sont pi-lotées par le formateur. La recherche explore comment ce système permet aux CP d'apprendre à dialoguer avec un enseignant débutant qui résiste au conseil. L'expérimentation avec 12 CP montre le potentiel d'un dispositif de formation instrumenté avec TGRIS. Mots-clés. Simulateur, formation, émotion, réalité virtuelle, outil auteur, agent conversationnel animé

National audience; Nous proposons dans cet article les premières étapes de l'établissement d'un canevas méthodologique permettant d'assister les acteurs de l'instrumentation de situations pédagogiques à l'aide d'outils reposant sur des technologies innovantes. Afin d'atteindre cet objectif, il a été entrepris une modélisation des différentes parties prenantes ainsi que l'environnement dans lequel s'inscrit cette instrumentation, dans ses dimensions organisationnelles et matérielles. Nous présentons le contexte de la genèse de ce modèle : l'instrumentation par un environnement outillé basé sur la transcription de la parole en direct, ainsi que d'autres technologies à l'état de prototypes de recherche. Abstract. This article presents the first steps in building a methodological framework aiming to assist contributors instrumenting pedagogical situations using tools based on innovative technologies. This framework ought to be used in various possible situations. This variety is described by a model of various stakeholders and the environment where the process is taking place, considering organizational or equipment constraints. Furthermore, we present the context leading to the creation of this model: instrumentation using a toolset based on synchronous speech transcription among other technologies in the research prototype stage.

National audience

As part of a collaborative design-based research project, flipped classroom training in a videoconferencing context was given twice to teachers. The pedagogical design of this training was structured in a framework based on the creation of a learning community, the learning presence and the real experience as a learner of a flipped classroom experience in a videoconferencing context. Following the first iteration, empirical findings, from the point of view of learning, as well as the activity of the instructors or the technopedagogical features implemented, led to the development of a first research project aimed at describing the presence created within this community, according to the learning presence model [11]. This work helped to highlight a certain lack of socio-cognitive presence and to raise questions about the pedagogical presence., Dans notre projet de recherche collaborative orientée par la conception, une formation sur la classe inversée en contexte de visioconférence a été donnée à deux reprises à des enseignants et conseillers pédagogiques. Le design pédagogique de cette formation se structurait dans un cadre misant sur la création d'une communauté d'apprentissage, la présence à distance et le vécu en tant qu'apprenant d'une expérience de classe inversée en contexte de visioconférence. Suite à la première itération, des constats empiriques, tant du point de vue de l'apprentissage, que de l'activité des formateurs ou des caractéristiques technopédagogiques mises en oeuvre, ont motivé l'élaboration d'un premier travail de recherche visant à décrire la présence suscitée au sein de cette communauté, selon le modèle de la présence à distance [11]. Ce travail a notamment permis de mettre en évidence une certaine carence en matière de présence sociocognitive et de soulever des questions sur la présence pédagogique.

Current work on multimodal machine translation (MMT) has suggested that the visual modality is either unnecessary or only marginally beneficial. We posit that this is a consequence of the very simple, short and repetitive sentences used in the only available dataset for the task (Multi30K), rendering the source text sufficient as context. In the general case, however, we believe that it is possible to combine visual and textual information in order to ground translations. In this paper we probe the contribution of the visual modality to state-of-the-art MMT models by conducting a systematic analysis where we partially deprive the models from source-side textual context. Our results show that under limited textual context, models are capable of leveraging the visual input to generate better translations. This contradicts the current belief that MMT models disregard the visual modality because of either the quality of the image features or the way they are integrated into the model., Accepted to NAACL-HLT 2019, reviewer comments addressed, camera-ready

International audience; Alternate realities through headsets, such as augmented, mixed and virtual reality are becoming part of people's everyday life. Except in some limited context, usual keyboards are inappropriate for such technological medium and alternative interfaces for text-entry must be explored. In this paper we present the keycube, a general-purpose cubic handheld device that goes beyond the text-entry interface by including multiple keys, a touch-screen, an inertial unit with six degrees-of-freedom and a vibrotactile feedback. Strong of its form factor and affordance, the keycube offers advantages with regards to mobility, comfort, learnability, privacy and playfulness. Thus, the combination creates a novel text-entry interface convenient for many use cases across the whole reality-virtuality continuum.

Comunicació presentada a: IWSDS 2019 celebrat del 24 a 26 d'abril de 2019 a Siracusa, Itàlia. Lifelong Learning in the context of Artificial Intelligence is a new paradigm that is still in its infancy. It refers to agents that are able to learn continuously, accumulating the knowledge learned in previous tasks and using it to help future learning. In this position paper we depart from the focus on learning new tasks and instead take a stance from the perspective of the life-cycle of intelligent software. We propose to focus lifelong learning research on autonomous intelligent systems that sustain their performance after deployment in production across time without the need of machine learning experts. This perspective is being applied to three European projects funded under the CHIST-ERA framework on several domains of application. This work has been partially funded by the ALLIES, DELTA and LIHLITH projects supported by the EU ERA-Net CHIST-ERA and the Spanish Research Agency (LIHLITH, PCIN-2017-118; DELTA, PCIN-2017-082).

We present a pretopological based approach to extract ego-centered communities. Classical methods often consider only one structural feature of the network, whereas pretopology enables to do multi-criteria analysis. Our approach consists in learning a logical combination of network’s descriptors to define a pretopological space. Ego-centered communities are extracted by computing the elementary closure of each node. The quality of such communities is evaluated against the ground truth communities. We show the benefits of our method by comparing it to others on both real and synthetic networks.

Multilingual (or cross-lingual) embeddings represent several languages in a unique vector space. Using a common embedding space enables for a shared semantic between words from different languages. In this paper, we propose to embed images and texts into a unique distributional vector space, enabling to search images by using text queries expressing information needs related to the (visual) content of images, as well as using image similarity. Our framework forces the representation of an image to be similar to the representation of the text that describes it. Moreover, by using multilingual embeddings we ensure that words from two different languages have close descriptors and thus are attached to similar images. We provide experimental evidence of the efficiency of our approach by experimenting it on two datasets: Common Objects in COntext (COCO) [19] and Multi30K [7].

International audience

International audience; This paper presents a tool designed to assist a virtual reality learner in a vocational training context. The benefits of VR are spreading outside the video-games and research field, leading the vocational education institutions to consider using this technology for training purposes. By simulating emblematic professional situations, teachers can train students in good conditions regarding safety, logistics and financial resources. In the French vocational training system, VR is just at its beginnings and the lack of experiments with this training context highlighted the need for new tools allowing teachers to use VR with their students. The work presented in this paper is part of a global project aiming to create, design and assess new VR tools and methodologies for this specific context. In order to guide, inform and assist the user, we are presenting a generic tool that can ease VR sessions by proposing embedded tools within easy reach of the immersed user, such as a camera, an inventory, a printer or stock management. This tool is a virtual tablet that can be grabbed by the immersed user allowing her/him to interact efficiently with the virtual environment.

Picosecond laser ultrasonics is an experimental technique for the generation and detection of ultrashort acoustic pulses using ultrafast lasers. In transparent media, it is often referred to as time-domain Brillouin scattering (TDBS). It provides the opportunity to monitor the propagation of nanometers-length acoustic pulses and to determine acoustical, optical, and acousto-optical parameters of the materials. We report on the application of TDBS for evaluating the effect of Praseodymium (Pr) substitution on the elasticity of multiferroic (Bi1&minus, xPrx)(Fe0.95Mn0.05)O3 (BPFMO) thin films. The films were deposited on Si and LaAlO3 (LAO) substrates by a sol-gel method. X-ray diffraction and Raman spectra revealed earlier that a phase transition from rhombohedral to tetragonal structure occurs at about 15% Pr substitution and is accompanied by the maxima of remnant magnetization and polarization. Combining TDBS with optical spectral reflectometry, scanning electron microscopy, and topographic measurements by atomic force microscopy, we found that the structural transition is also characterized by the maximum optical dielectric constant and the minimum longitudinal sound velocity. Our results, together with earlier ones, suggest that BiFeO3-based films and ceramics with compositions near phase boundaries might be promising materials for multifunctional applications.

National audience

National audience

National audience

This paper presents an authoring tool in the cMOOC context called MOOCAT (MOOC Authoring Tool). It is a visual authoring tool that helps teachers to design cMOOC-oriented pedagogical scenarios. MOOCAT has two main innovative features. The first is that it offers a tool for conceiving pedagogical scenarios in a simple way through graphical representation, by providing the functionality to design learning workflows. The second is related to its capacity to bridge the gap between the conception phase and its execution into different LMS (Learning Management System), by offering services that allow the automatic deployment of pedagogical scenarios to existing platforms (e.g. OpenEDX, Moodle, etc.). This facilitates the authoring process considerably, and enhances the ability of teachers to concentrate their efforts on the content. This paper presents the model underlying MOOCAT, and describes the cMOOC scenario-building process. Also, the paper highlights the findings from a case study which was organized in order to evaluate the utility and the usability of the tool, and the usefulness of its innovative feature of conceiving and deploying a cMOOC scenario into OpenEDX platform. MOOCAT seems to be an easy-to-use tool, which offers guidance and flexibility during the design process.

International audience

National audience

This article presents on the one hand the concepts to build a longitudinal evaluation process (LEP) of a learning platform in a research context and on the other hand the organization of this LEP. Evaluations' objectives are threefold: at the research level, at the technical level and at the societal level. At the research level, this LEP is inscribed in a pragmatic constructivism posture to work iteratively and incrementally from the field. At the level of project monitoring, the LEP is built on the basis of the tools of the quality approach which can guarantee a traceability of the production of data and results. The case study chosen is the LabNbook platform to learn experimental sciences in a context of higher education., Cet article présente d'une part les concepts pour construire un processus d'évaluation longitudinale (PEL) d'une plateforme d'apprentissage et d'autre part l'organisation de ce PEL. L'objectif de cette évaluation est d'inscrire ce PEL au niveau de la recherche, dans une posture constructiviste pragmatique pour travailler de manière itérative et incrémentale à partir du terrain. Le PEL est construit sur la base des outils de suivi de projet, en utilisant une démarche qualité qui garantit une traçabilité de la production des données et des résultats. Le cas d'étude choisi est LabNbook (https://labnbook.fr), qui est une plateforme pour l'apprentissage des sciences expérimentales.

International audience; In this position paper, we present two different approaches to help social acceptance of head-mounted display in social and shared spaces: (1) discreet interactions and (2) a public spectator view. On a daily basis we evolve in shared and social spaces which may have different levels of acceptance for the use of head-mounted display (HMD), either from the users or the surroundings people. In contexts where the HMD is usually accepted by the non-HMD users (e.g. public transport) but still struggle to be adopted, we will approach the potential benefits of discreet interactions. Finally, we suggest and raise questions about the public spectator view approach to address contexts where the use of HMD is perceived by the surroundings as inappropriate

National audience

National audience

This paper describes the ON-TRAC Consortium translation systems developed for the end-to-end model task of IWSLT Evaluation 2019 for the English-to-Portuguese language pair. ON-TRAC Consortium is composed of researchers from three French academic laboratories: LIA (Avignon Universit\'e), LIG (Universit\'e Grenoble Alpes), and LIUM (Le Mans Universit\'e). A single end-to-end model built as a neural encoder-decoder architecture with attention mechanism was used for two primary submissions corresponding to the two EN-PT evaluations sets: (1) TED (MuST-C) and (2) How2. In this paper, we notably investigate impact of pooling heterogeneous corpora for training, impact of target tokenization (characters or BPEs), impact of speech input segmentation and we also compare our best end-to-end model (BLEU of 26.91 on MuST-C and 43.82 on How2 validation sets) to a pipeline (ASR+MT) approach., Comment: IWSLT 2019 - First two authors contributed equally to this work

This work investigates spoken language understanding (SLU) systems in the scenario when the semantic information is extracted directly from the speech signal by means of a single end-to-end neural network model. Two SLU tasks are considered: named entity recognition (NER) and semantic slot filling (SF). For these tasks, in order to improve the model performance, we explore various techniques including speaker adaptation, a modification of the connectionist temporal classification (CTC) training criterion, and sequential pretraining.

National audience; De plus en plus d'établissements d'enseignement supérieur, que ce soit en France ou au Québec, disposent d'équipements de visioconférence et de salles de téléprésence immersives donnant aux utilisateurs le sentiment d'être présents ensemble autour d’une même table. Cependant, ces installations ne sont pas spécifiquement adaptées aux besoins de l’enseignement et de l’apprentissage, et les enseignants, les conseillers pédagogiques et les chercheurs ont besoin de comprendre comment améliorer la qualité de l'enseignement dans de tels contextes (Lameul et Loisy, 2014). Compte tenu de cette situation, il est crucial de constituer un répertoire de connaissances rigoureuses et critiques sur les approches pédagogiques adaptées à de tels dispositifs ainsi que leurs effets sur l'apprentissage (Albero, 2011) et les innovations pédagogiques émergentes (Bédard et Béchard, 2009).Afin d'atteindre cet objectif, un projet de recherche orientée par la conception (Sanchez et Monod-Ansaldi, 2015, Wang et Hannafin, 2005) appelé TOPIC (La Téléprésence comme opportunité pour l'Innovation et la Conception pédagogiques) a été mis en place. Parmi les différentes formations qui ont été conçues sur la base d'une étroite collaboration entre chercheurs et formateurs, l'objectif principal de l’une d’elles était de permettre aux enseignants de développer quatre compétences liées à l'enseignement d'une classe inversée dans un contexte de visioconférence. Cette formation, offerte à deux reprises, se déroulait sur deux semaines, avec des participants répartis sur plusieurs sites dans deux continents.Dans ce chapitre, sont présentés le modèle de conception qui a été développé, son cadre théorique et son organisation technologique, ainsi que quelques-uns des premiers résultats obtenus à partir de données recueillies avant, pendant et après chacune des itérations de la formation. Ce chapitre focalise essentiellement sur l’expérience d’apprentissage et les modalités de formation adaptées.

In this paper, we propose several specific embeddings in Arabic sentiment analysis (SA) framework. Indeed, Arabic is characterized by its agglutination and morphological richness contributing to great sparsity that could affect embedding quality. This work presents a rigorous study that compares different types of Arabic-specific embeddings. We evaluate them with 2 neural architectures: one based on convolutional neural network (CNN) and the other one based on Bidirectional Long Short-Term Memory Bi-LSTM. Experiments are done on the Large Arabic-Book Reviews corpus LABR. Our best results boost previous published accuracy by 1.9%. Moreover, we experiment combination of our individual systems defining very confident decision, reaching an accuracy of 92.2% on 98.25% of LABR test dataset.

The use of learning dashboards with analytics might help users to gain insight into their learning process and then to make decision. However, designing meaningful Learning Analytics dashboard (LAD) is still a complex process that requires an explicit understanding of the user needs. For this reason, we carried out a user-centered design (UCD) approach with the aim to provide users with adapted LADs to support their decision-making. Hence, we develop a UCD process composed of 4 steps: (i) we propose a participatory-based design tool for capturing contextualized needs (ii) these identified needs will be described using an independent formalism and LAD models in order to capitalize on them (iii) to automate these models, we propose a LAD generation process (iv) finally, we carried out an evaluation phase with the aim to review and refine our models. During our process development, an iterative user needs refinement confirmed that decision is considered as a centered element for LAD generations.

(Scimago Q1, ATIEF A+); International audience; In spite of their effectiveness, learning environments often fail to engage users and end up under-used. Many studies show that gamification of learning environments can enhance learners' motivation to use learning environments. However, learners react differently to specific game mechanics and little is known about how to adapt gaming features to learners' profiles. In this paper, we propose a process for adapting gaming features based on a player model. This model is inspired from existing player typologies and types of gamification elements. Our approach is implemented in a learning environment with five different gaming features, and evaluated with 266 participants. The main results of this study show that, amongst the most engaged learners (i.e. learners who use the environment the longest), those with adapted gaming features spend significantly more time in the learning environment. Furthermore, learners with features that are not adapted have a higher level of amotivation. These results support the relevance of adapting gaming features to enhance learners' engagement, and provide cues on means to implement adaptation mechanisms.

National audience

The work presented in this thesis is part of a broader issue of Technology Enhanced Learning (TEL) field. It is particularly interested in Massive Open online Courses (MOOC) environments and concerns more specifically the activity of designing and conceiving pedagogical scenarios by teachers in connectivist MOOC (cMOOC).The main goal is to provide an approach, models and tools to assist teachers in the scenario conception process and the operationalization of pedagogical scenarios in a cMOOC context.In order to meet these needs, the first part of this thesis is devoted to specification of the cMOOC creation and execution process describing the involved stakeholders and the characteristics of each process step. The second part of the manuscript is consecrated to the cORPS model, which describes and defines a cMOOC-oriented pedagogical scenario.To reify our proposals, we have developed an authoring environment for cMOOC context, called MOOCAT. One of the specific characteristics of our tool lies in the exploitation and reuse of the BPMN notation. This tool and the associated operationalization service are covered in the third part of this manuscript.Finally, different publics of practitioners tested MOOCAT notably from a utility and usability point of view. The obtained findings confirm that MOOCAT can be used to design connectivist pedagogical scenarios and provide all the necessary elements for the design of such courses. Feedback from participants particularly highlighted its ease of use and the benefit of a visual authoring tool.; Le travail que nous présentons dans cette thèse s’inscrit dans une problématique générale des EIAH. Il s’intéresse plus particulièrement aux Massives Open Online Courses (MOOC) et concerne plus précisément l’activité de conception de scénarios pédagogiques par des enseignants dans les environnements MOOC connectivistes (cMOOC). Le but principal est de proposer une approche, des modèles et des outils pour assister les enseignants dans le processus de scénarisation et d’opérationnalisation des scénarios pédagogiques dans le contexte des cMOOC. Afin de répondre à ces besoins, la première partie de cette thèse est consacrée à la spécification du processus de création et de déroulement d’un cMOOC. La deuxième partie du manuscrit est consacrée au modèle cORPS de description de scénarios pédagogiques orientés cMOOC.Pour réifier nos propositions, nous avons développé un outil de scénarisation pédagogique adapté au contexte des cMOOC, appelé MOOCAT. Cet outil est doté d’un service de déploiement automatique des scénarios pédagogiques et exploite la notion de workflow à travers la réutilisation et l’adaptation de la notation BPMN. Finalement, l’environnement MOOCAT a été expérimenté auprès de différents publics d’un point de vue utilité et utilisabilité. Les résultats obtenus permettent de confirmer que MOOCAT permet de concevoir des scénarios pédagogiques connectivistes. Les feedbacks des participants ont souligné particulièrement sa simplicité d’utilisation et le bénéfice d’avoir un outil de scénarisation visuelle.

In the field of NLP, word embeddings have recently attracted a lot of attention. A textual corpus is represented as a sparse words co-occurrences matrix. Then, the matrix can be factorized, for example using SVD, which allows to obtain a shorter matrix with dense and continuous vectors. To help SVD, PMI measure is applied on the initial co-occurrence matrix, assigning a relevant weight to the co-occurrences by normalizing them using both the considered words frequencies. In this paper, we follow this idea to study if weighted networks can benefit from pre-processing that can help community detection. We first design a benchmark using LFR networks. Then, we consider PMI and another NLP inspired measure as a preprocessing of the links weights, and show that PMI worsens the results while the other one improves them. By distinguishing links inside communities and links between communities into two classes, we show that this is due to the weights distributions of these links. Links between communities are in average bigger, leading to bigger values of PMI. From this analysis, we design another set of experiments that show that it is possible to classify efficiently links into these two classes, using a small set of features. Finally, we introduce the Supervised Label Propagation (SLP) algorithm that takes into account the classification results during the propagation. This algorithm clearly improves the results, leading us to a major questioning: is community detection on weighted networks a fully unsupervised task? We conclude with our thoughts on this topic.

International audience; Introduced in the context of machine learning, the Feature F-measure is a statistical feature selection metric without parameters that allows to describe classes through a set of salient features. It was shown efficient for classification, cluster labeling and clustering model quality measurement. In this paper, we introduce the Node F-measure, its transposition in the context of networks, where it can by analogy be applied to detect salient nodes in communities. This approach benefits from the parameter-free system of Feature F-Measure, its low computational complexity and its well-evaluated performance. Interestingly, we show that in addition to these properties, Node F-measure is correlated with certain centrality measures, and with measures designed to characterize the community roles of nodes. We also observe that the usual community roles measures are strongly dependent from the size of the communities whereas the ones we propose are by definition linked to the density of the community. This hence makes their results comparable from one network to another. Finally, the parameter-free selection process applied to nodes allows for a universal system, contrary to the thresholds previously defined empirically for the establishment of community roles. These results may have applications regarding leadership in scientific communities or when considering temporal monitoring of communities.

National audience

International audience; Computer programming has become a basic skill and is now part of the curriculum taught in middle school. In France, this educational reform was very sudden and many teachers are not yet trained adequately to teach this subject , leaving them in search of guidelines and tools to help them. It is in this context that we propose TurtleTable, an educational game, who has two main originalities. First of all, this game is not based on writing a computer program, like most of the other tools, but rather on executing a given program, step by step. These two approaches are complementary and essential to apprehend the logic behind computer programming. Secondly, TurtleTable is played in groups of three, by manipulating Tangible objects on an Interactive Tabletop (TIT). The players have to collaboratively move the objects on a grid, by following the instructions of the program on the screen. The game immediately indicates which instructions are executed correctly, allowing the learners to correct their mistakes. TurtleTable was tested by 59 middle school students. We developed four different interactions for TurtleTable (TIT, tabletop, tablet and computer) to compare their effects on motivation, collaboration and learning. The preliminary analysis show that the students really appreciated the game, especially the TIT version, that encouraged collaboration and gave them the impression they learned more.

Named entity recognition (NER) is among SLU tasks that usually extract semantic information from textual documents. Until now, NER from speech is made through a pipeline process that consists in processing first an automatic speech recognition (ASR) on the audio and then processing a NER on the ASR outputs. Such approach has some disadvantages (error propagation, metric to tune ASR systems sub-optimal in regards to the final task, reduced space search at the ASR output level,...) and it is known that more integrated approaches outperform sequential ones, when they can be applied. In this paper, we explore an end-to-end approach that directly extracts named entities from speech, though a unique neural architecture. On a such way, a joint optimization is possible for both ASR and NER. Experiments are carried on French data easily accessible, composed of data distributed in several evaluation campaigns. The results are promising since this end-to-end approach provides similar results (F-measure= 0.66 on test data) than a classical pipeline approach to detect named entity categories (F-measure=0.64). Last, we also explore this approach applied to semantic concept extraction, through a slot filling task known as a spoken language understanding problem, and also observe an improvement in comparison to a pipeline approach.

International audience; Augmented Reality begins to be widely mainstream among children, due to some interactive successes in video games and social networks. Based on this interest, we present CartonEd, an open and complete toolkit suitable for children dedicated to the construction of an augmented reality headset device. The toolkit let the children play and explore augmented reality with and beyond handheld devices. Inspired by the Do-It-Yourself movement, the toolkit includes different components such as blueprints, tutorials, videos, mobile apps, a software development kit and an official website. Among the mobile applications, one is implemented to guide the children through the construction process while experiencing augmented reality. To validate our solution (in particular the construction process and the guiding app) and understand its effect on children in regard to their relation to the augmented reality, we conducted four construction sessions. Our study examines the usability of the guiding app and the construction process. We report in this paper the main components of the CartonEd toolkit and the results of an evaluation among 57 children and teenagers (ages 8-16), showing a positive outcome about their own constructed device (all functional), their feelings and wishes regarding the augmented reality.

National audience; While the field of Learning Analytics is in full development, potential users are just discovering the opportunities offered by these tools. One of the major difficulties consists in proposing a visual data representation which makes sense to the users. After refining our method to identify user needs in terms of visualization, and to define the main dimensions of a learning dashboard, we propose a tool to support participatory design. This tool is based on a canvas and cards to help dashboards’ creation using Learning Analytics. It allows users to support creativity around decision making, characterize their context, and draw a dashboard’s mockup using existing content.; Si le domaine d’analyse des données d’apprentissage est en plein développement, les utilisateurs potentiels des résultats obtenus découvrent les opportunités offertes par ces outils. L’une des difficultés majeures consiste à proposer une représentation graphique des données qui fasse sens aux utilisateurs. Après avoir présenté notre démarche permettant d’identifier les besoins des utilisateurs en terme de visualisation et de définir les dimensions principales d’un tableau de bord d’apprentissage, nous proposons un outil support à la conception participative. Cet outil est basé sur un tableau et des cartes pour aider à la création de tableaux de bord exploitant les Learning Analytics. Il permet aux utilisateurs d’organiser leur réflexion autour de la prise de décision, de caractériser leur contexte, et de maquetter le tableau de bord à l’aide de contenus existants.

International audience; Dans le cadre du projet de recherche collaborative orientée par la conception appelé TOPIC (Téléprésence comme OPportunité pour l'Innovation dans la Conception) une formation portant sur la classe inversée en contexte de visioconférence a été donnée à deux reprises à des enseignants des établissements universitaires partenaires (Université de Sherbrooke et Université Bretagne Loire). Le design pédagogique de cette formation se structurait dans un cadre pédagogique misant sur la création d'une communauté d'apprentissage, la présence à distance, et le vécu réel en tant qu'apprenant d'une expérience de classe inversée en contexte de visioconférence. À la suite de la première itération vécue au printemps 2017, des constats empiriques, tant du point de vue de l'apprentissage, que de l'activité des formateurs ou encore des caractéristiques technopédagogiques mises en oeuvre, ont suscité l'élaboration d'un premier travail de recherche visant à décrire la présence à distance vécue. Ainsi, cette première itération de la recherche a permis, notamment, de mettre en évidence une certaine carence en matière de présence socio-cognitive et de soulever des questions relativement à la présence pédagogique. Cette communication présentera le design de la formation, les résultats de recherche obtenus et les adaptations réalisées en conséquence dans la planification de la seconde itération.

The rise of mobile devices (e.g. tablets, smartphones) and their educational and recreational applications have contributed to the emergence of Mobile Learning Games (MLGs). Indeed, MLGs show great potential for increasing engagement, creativity and authentic learning. Yet, despite their great potential for education, the use of MLGs by teachers, remains very limited. This is partly due to the fact that MLGs are often designed to match a specific learning context, and thus cannot be directly reusable for other contexts. In addition, existing authoring tools are either feature-rich but require a significant investment by teachers to be used, or simple to use but do not offer enough features for the design of MLGs that meet pedagogical needs. To tackle these problems, we propose JEM iNVENTOR, a MLG authoring tool, based on a nested design approach, intended for teachers, museum curators, or any person without computer skills, wishing to script their own MLG and deploy them on mobile systems.The nested design model was approved through a series of experimentations with some twenty teachers from a wide range of expertise levels and teaching fields. We also conducted field experimentations with about 1500 students and pupils in order to evaluate the quality of MLGs created with JEM iNVENTOR as well as their impact on learners.; L’essor des périphériques mobiles (ex. tablettes, smartphones) ainsi que leurs applications pédagogiques et ludiques ont contribué à la naissance des Jeux Éducatifs Mobiles (JEM). De nombreux chercheurs ont prouvé les effets positifs de ces JEM sur la motivation des apprenants et même sur certains apprentissages. Cependant, l’utilisation de JEM en contexte scolaire reste très limitée. En effet, les JEM existants, parfois assez coûteux, sont souvent conçus pour un domaine très spécifique, et n’offrent donc pas de possibilités de réutilisation. De plus, les outils auteur existants sont, soit riches en fonctionnalités mais nécessitent un investissement important des enseignants pour être pris en main, soit simples à utiliser mais ne permettent pas de concevoir des JEM qui répondent aux besoins pédagogiques. Pour s’attaquer à ces problématiques, nous proposons JEM iNVENTOR, un outil auteur de JEM, fondé sur une approche de conception gigogne, destiné aux enseignants, conservateurs de musée, ou toute personne non-informaticienne, qui souhaitent scénariser leurs propres JEM et les déployer sur les systèmes mobiles.Le modèle de conception gigogne a été validé par une série d’expérimentations auprès d’une vingtaine d’enseignants ayant des niveaux d’expertises et des domaines d’enseignement très variés. Nous avons également mené des expérimentations de terrain, auprès d’environ 1500 étudiants et élèves, afin d’évaluer la qualité des JEM créés avec JEM iNVENTOR ainsi que leur impact sur les apprenants.

International audience

Social networks are currently the main medium through which terrorist organisations reach out to vulnerable people with the objective of radicalizing and recruiting them to commit violent acts of terrorism. Fortunately, radicalization on social networks has warning signals and indicators that can be detected at the early stages of the radicalization process. In this article, we explore the use of the semantic web and domain ontologies to automatically mine the radicalisation indicators from messages and posts on social networks. Specifically, we propose an ontology for the radicalisation domain as well as an approach to automatically compute the radicalisation indicators. In our approach, social messages are annotated with concepts and instances defined formally in a domain ontology. Annotations are then exploited within an inference phase to identify the messages exhibiting a radicalization indicator. The indicators are then computed using a set of SPARQL queries. We have implemented and evaluated the proposed solution based on a pubic Tweet dataset. Obtained results show the effectiveness of our approach. The article presents also the implemented prototype.

Although much effort has recently been devoted to training high-quality sentence embeddings, we still have a poor understanding of what they are capturing. "Downstream" tasks, often based on sentence classification, are commonly used to evaluate the quality of sentence representations. The complexity of the tasks makes it however difficult to infer what kind of information is present in the representations. We introduce here 10 probing tasks designed to capture simple linguistic features of sentences, and we use them to study embeddings generated by three different encoders trained in eight distinct ways, uncovering intriguing properties of both encoders and training methods., Comment: ACL 2018

International audience; Social skills are essential and it is often necessary to train in order to react in the best way to certain difficult, stressful situations such as a job interview, dating. The best way of training for these challenging interview situations is to role play with actors, capable of simulating emotions, but this solution is not affordable for most schools and universities. We therefore propose TGRIS, a Teacher-Guided Realistic Interview Simulator that allows students to practice interviews with an emotional virtual agent in Virtual Reality. Most importantly, this tools allows teachers to create custom interview situations by changing the agent's face, emotional state and piloting the agent's verbal and non-verbal reactions during the interview. Finally, we discuss the first experimentations and show how this solution might turn out to be better than live actors.

International audience; In the field of storytelling, speech synthesis is trying to move from a neutral machine-like to an expressive voice. For para-metric and unit-selection systems, building new features or cost functions is necessary to allow a better expressivity control. The present article investigates the classification task between direct and narrative discourse phrases to build a new expressivity score. Different models are trained on different speech units (syllable, word and discourse phrases) from an audiobook with 3 sets of features. Classification experiments are conducted on the Blizzard corpus which features children English audio-books and contains various characters and emotional states. The experiments show that the fusion of SVM classifiers trained with different prosodic and phonologic feature sets increases the classification rate from 67.4% with 14 prosodic features to 71.8% with the 3 merged sets. Also the addition of a decision threshold achieves promising results for expressive speech synthesis according to the strength of the constraint required on expressivity: 71.8% with 100% of the words, 79.9% with 50% and 82.6% with 25%.