Search

Translation into a morphologically rich language requires a large output vocabulary to model various morphological phenomena, which is a challenge for neural machine translation architectures. To address this issue, the present paper investigates the impact of having two output factors with a system able to generate separately two distinct representations of the target words. Within this framework, we investigate several word representations that correspond to different distributions of morpho-syntactic information across both factors. We report experiments for translation from English into two morphologically rich languages, Czech and Latvian, and show the importance of explicitly modeling target morphology.

The work presented in this thesis is a part of the Technology Enhanced Learning domain. It focuses on the proposal of processes, methods and tools that assist teachers and trainers in the reuse and the capitalization of educational scenarios. The objective of the proposed approach named CAPtuRe is to model, evaluate and exploit the contextual information related to a scenario based on its effective observations with the aim to enhance reuse. The main concerns are: (1) the expression and the analysis of the usage context, (2) the evaluation of the relevance of the scenario in a specific context, (3) the indexing of the contexts based on criteria of success and effectiveness of the scenario to define its reuse scope and (4) the proactive suggestion of reuse. We started by specifying a global framework for the engineering and the reuse of educational scenarios. In this context, we have defined a process specifying the scenario lifecycle introducing the contextual dimension and its utilization in a "design by reuse" environment. In order to operationalize this process, we define a generic approach to model the contextual information of a scenario that is enriched by the indicators, an indexing method and an algorithm calculating contextual similarities for the selection and the recommendation of appropriated scenarios to a target learning situation. These contributions are implemented as a software platform and applied to hybrid scenarios usage cases., Les travaux de thèse s'inscrivent dans le domaine des Environnements Informatiques pour l'Apprentissage Humain (EIAH). Ils portent sur la proposition de processus, méthodes et outils pour assister les enseignants et les formateurs dans la réutilisation et la capitalisation des scénarios d'apprentissage. L'approche proposée nommée CAPtuRe a pour objectif de modéliser, évaluer et exploiter les informations contextuelles relatives à un scénario en se basant sur des observations effectives de ce dernier pour améliorer la réutilisation.Les problématiques étudiées concernent : (1) l'expression et l'analyse du contexte d'usage, (2) l'évaluation de la pertinence du scénario dans un contexte précis, (3) l'indexation des contextes sur la base de critères de réussite et d'efficacité du scénario pour la définition de son périmètre de réutilisation et (4) la suggestion proactive de réutilisation.Nous avons commencé par la spécification d'un cadre global d'ingénierie et de réutilisation de scénarios d'apprentissage. Dans ce cadre, nous avons défini un processus qui spécifie le cycle de vie du scénario explicitant la dimension contextuelle et son utilisation dans un environnement de "conception par la réutilisation". Pour opérationnaliser ce processus, nous avons défini une approche générique de modélisation de l'information contextuelle enrichie par les indicateurs, une méthode d'indexation et un algorithme de calcul de similarités contextuelles pour la sélection et la recommandation de scénarios appropriés à une situation d'apprentissage cible. Ces contributions ont été implémentées sous la forme d'une plateforme logicielle et appliquées sur des cas d'usage de scénarios hybrides.

(ATIEF A); National audience; Gamification is increasingly used to address the lack of learner motivation. However, gamification systems generally do not consider learner's preferences for game mechanics. This paper presents a model to adapt the gamification features to the learners according to their player profile. An evaluation of this model shows that adapted game features increase learner engagement, while counter-adapted game features decrease their motivation.; La ludification est une approche de plus en plus utilisée pour répondre au manque de motivation des apprenants. Cependant, chacun a des sensibilités différentes aux mécaniques de jeu, et cela n'est généralement pas pris en compte. Cet article présente un modèle pour adapter les fonctionnalités ludiques selon les profils de joueur des apprenants. Une évaluation de ce modèle a montré que l'utilisation de fonctionnalités ludiques adaptées rend les apprenants plus engagés, tandis que l'utilisation de fonctionnalités ludiques contre-adaptées diminue leur motivation.

International audience; In this paper we aim to enhance keyword search for conversational telephone speech underlow-resourced conditions. Two techniques to improve the detection of out-of-vocabularykeywords are assessed in this study: using extra text resources to augment the lexicon andlanguage model, and via subword units for keyword search. Two approaches for dataaugmentation are explored to extend the limited amount of transcribed conversationalspeech: using conversational-like Web data and texts generated by recurrent neuralnetworks. Contrastive comparisons of subword-based systems are performed to evaluatethe benefits of multiple subword decodings and single decoding. Keyword search results arereported for all the techniques, but only some improve performance. Results are reported forthe Mongolian and Igbo languages using data from the 2016 Babel program.

Deep learning is revolutionizing speech and natural language technologies since it is offering an effective way to train systems and obtaining significant improvements. The main advantage of deep learning is that, by developing the right architecture, the system automatically learns features from data without the need of explicitly designing them. This machine learning perspective is conceptually changing how speech and natural language technologies are addressed. In the case of Machine Translation (MT), deep learning was first introduced in standard statistical systems. By now, end-to-end neural MT systems have reached competitive results. This special issue introductory paper addresses how deep learning has been gradually introduced in MT. This introduction covers all topics contained in the papers included in this special issue, which basically are: integration of deep learning in statistical MT; development of the end-to-end neural MT system; and introduction of deep learning in interactive MT and MT evaluation. Finally, this introduction sketches some research directions that MT is taking guided by deep learning.

International audience; Research and industry are more and more focusing in finding automatically the polarity of an opinion regarding a specific subject or entity. Paragraph vector has been recently proposed to learn embeddings which are leveraged for English sentiment analysis. This paper focuses on Arabic sentiment analysis and investigates the use of paragraph vector within a machine learning techniques to determine the polarity of a given text. We tested some preprocessing method, and we show that light stemming enhance the performance of classification.

Mobile Learning Games (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 limited. This is partly due to the fact that MLGs are often designed to match a specific learning context, and thus cannot be reusable for other contexts. Therefore, researchers have recently designed various types of MLG authoring tools. However, these authoring tools are not always adapted to non-computer-scientists or non-game-designers. Hence, we propose in this paper to focus on five existing MLG authoring tools, in order to assess their features and usability with the help of five teachers, who are used to organizing educational field trips. In the second part of this paper, we present an approach for designing a MLG authoring tool, based on the lacks identified through the analysis, and tailored to the teachers' different profiles and needs., Comment: European Games and Learning Alliance conference, GALA, Dec 2016, Utrech, Netherlands

International audience; The early stages of Mobile Serious Game (MSG) design are very complex. That is why many MSG research projects start with a brainstorming session, involving pedagogical and game experts. However, brainstorming with large groups of participants, who come from very different backgrounds, can prove to be long and inefficient. In addition, classic brainstorming techniques are not well adapted to the large number of constraints and technical limitations that need to be taken into account when designing MSGs (e.g. educational content, number of students, duration of field trip, fences). Our hypothesis is that the first stages of design can be efficiently supported by a board game. Indeed, games have proven to be very efficient to support collaboration and enhance motivation and they can be designed in such a way to encapsulate constraints in their game rules. In order to test this hypothesis, we created a meta-SG: a game for designing a botanical MSG for a natural park. The experimentation we conducted with 17 designers was very encouraging. The analysis of the videos, questionnaires and produced scenarios showed that the meta-SG generated a very lively and productive creativity session, and the constraints related to MSG design were not perceived as limitations but as challenging game rules. Designing Serious Games (SGs) for education is a complex process that involves several actors: SG experts, who have experience in using game mechanics to enhance learning, and educational experts, who master the didactics of the target pedagogical skills. The design of Mobile Serious Games (MSGs) requires a third type of actor, who masters the MSG's use context. This context is not limited to a set of measurable physical settings (Dourish 2004); it also contains domain specific notions that can only be identified by field experts who have a deep understanding of the physical context (e.g. location, boundaries, physical limitations) in which the MSG will take place. MSGs for museums, for example, need to be designed with a museum curator, in order to fit specific rooms and exhibits (Lonsdale 2005). Another important aspect to take into account when designing MSGs is the time context. For example, if a MSG is designed for a field trip, it must not last more than the time planned for this outing. We therefore adopt a design-based research method (Whang 2005), implying deep collaboration between our team of SG experts, pedagogical experts and field experts. Unfortunately, these experts have little time to offer for work sessions. As many before us, we were confronted to this problem during the ReVeRIES 1 research project. The goal of this project is to use mobile technologies to help humans recognize the trees that surround them, in a fun and motivating way. It involves SG experts, computer scientists, several botanists and botanical park managers. The main objective of our first creativity session was to help these experts collaborate, in order to produce plausible MSG scenarios, for the Echologia Park 2. These scenarios will be used as a basis to build an authoring tool for the park owners to create their own MSGs. We therefore needed to help the designers in producing a large number and variety of MSG scenarios, adapted to Echologia's constraints, and in particular to the park's morphology (e.g. fences, trees) and the park's visitor profiles (e.g. groups of teenagers, families). In the next section, we investigate the strengths and weakness of existing tools and methods for organizing such creativity sessions.

National audience

The development of a speech recognition system requires the availability of a large amount of resources namely, large corpora of text and speech, a dictionary of pronunciation. Nevertheless, these resources are not available directly for Arabic dialects. As a result, the development of a SRAP for Arabic dialects is fraught with many difficulties, namely the lack of large amounts of resources and the absence of a standard spelling as these dialects are spoken and not written. In this perspective, the work of this thesis is part of the development of a SRAP for the Tunisian dialect. A first part of the contributions consists in developing a variant of CODA (Conventional Orthography for Arabic Dialectal) for the Tunisian dialect. In fact, this convention is designed to provide a detailed description of the guidelines applied to the Tunisian dialect. Given the guidelines of CODA, we have created our corpus TARIC: Corpus of the interaction of the railways of the Tunisian Arab in the field of SNCFT. In addition to these resources, the pronunciation dictionary is indispensable for the development of a peech recognition system. In this regard, in the second part of the contributions, we aim at the creation of a system called conversion(Grapheme-Phonème) G2P which allows to automatically generate this phonetic dictionary. All these resources described before are used to adapt a SRAP for the MSA of the LIUM laboratory to the Tunisian dialect in the field of SNCFT. The evaluation of our system gave rise to WER of 22.6% on the test set.; Le développement d'un système de reconnaissance de la parole exige la disponibilité d'une grande quantité de ressources à savoir, grands corpus de texte et de parole, un dictionnaire de prononciation. Néanmoins, ces ressources ne sont pas disponibles directement pour des dialectes arabes. De ce fait, le développement d'un SRAP pour les dialectes arabes se heurte à de multiples difficultés à savoir, l’'abence de grandes quantités de ressources et l'absence d’'une orthographe standard vu que ces dialectes sont parlés et non écrit. Dans cette perspective, les travaux de cette thèse s’intègrent dans le cadre du développement d’un SRAP pour le dialecte tunisien. Une première partie des contributions consiste à développer une variante de CODA (Conventional Orthography for Arabic Dialectal) pour le dialecte tunisien. En fait, cette convention est conçue dans le but de fournir une description détaillée des directives appliquées au dialecte tunisien. Compte tenu des lignes directives de CODA, nous avons constitué notre corpus nommé TARIC : Corpus de l’interaction des chemins de fer de l’arabe tunisien dans le domaine de la SNCFT. Outre ces ressources, le dictionnaire de prononciation s’impose d’une manière indispensable pour le développement d’un SRAP. À ce propos, dans la deuxième partie des contributions, nous visons la création d’un système nommé conversion (Graphème-Phonème) G2P qui permet de générer automatiquement ce dictionnaire phonétique. Toutes ces ressources décrites avant sont utilisées pour adapter un SRAP pour le MSA du laboratoire LIUM au dialecte tunisien dans le domaine de la SNCFT. L’évaluation de notre système donné lieu WER de 22,6% sur l’ensemble de test.

International audience; Mobile Learning Games (MLGs) show great potential for education, especially in fields that deal with outdoor learning activities such as archaeology or botany. However, the number of MLGs currently used remains insignificant. This is partly due to the fact that the current authoring tools are based on modeling languages that only allow creating very specific and rigid types of MLGs. In this paper, we therefore propose an extensive modeling language for MLGs. This model was designed, with the help of botanical experts, in order to cover the variety of MLG types they would like for their field trips. This modeling language uses high-level concepts, such as game activities and points of interest on a map that can therefore be used by teachers in any domain. Finally, we discuss how scenarios, described with this language , can be automatically transformed into executable web applications.

International audience; This position paper is dedicated to describing a preliminary model of an integrated system, called Play Management System (PMS). PMS is designed to support both players and teachers to deliver, use, manage and track play situations. This PMS model results from a design-based research methodology. Our approach focuses on (1) the learners and the situation that emerges when they play the game, rather than the system dedicated to play and (2) the teachers who want to manage a game-based learning situation. Thus, we argue for a shift from a game-based to a play-based perspective. 1 Introduction Within a context marked by the development of alternative pedagogies, this position paper aims to describe a model of an integrated system, called Play Management System (PMS), dedicated to support players and teachers to deliver, use, manage and track play situations. The purpose of this article is to propose an innovative approach for implementing a play-based learning approach by (1) focusing on the learners and taking into consideration the situation that emerges when they play rather than the artifact dedicated to play (play vs game) and (2) focusing on the teachers who want to implement and manage a play-based learning situation in their classroom (play management vs game design). Thus, we address the issue of teachers' requirements for the orchestration of a play situation within an educational context. In the first section of this paper, we advocate for a player-centered approach for game-based learning. The second section presents a game developed during the project and the design-based research methodology adopted for designing this game. The third section describes

International audience; cording to the recent technological advances, a new type of digital learning games has emerged. These games integrate virtual worlds persistence and immersion devices allowing the learners to experience more authentic and rich situations. Several studies highlighted their pedagogical value, knowledge transfer and learners’ engaged-behaviors. In this paper, we draw the characteristics of these learning games based on the integration of new technologies according to two characteristics: immersion and persistence. To investigate the impact of such technological components, we developed a game and evaluated it in ecological conditions. Four groups of fifteen high school students played the game through two testing conditions: two groups used a prototype allowing only classical interactions limited on usual devices, while the two other groups used a prototype integrating persistent and immersive interactions using Oculus Rift vision. All the interactions were recorded and their analysis suggests more engaged behaviors from students using the immersive and persistent prototype.

International audience; Mobile devices offer great opportunities in the field of collaborative learning. They are especially interesting in their ability to provide digital information while still supporting social interactions between group members, which are essential elements of coordinated and shared activities. However, in truly mobile conditions, e.g. outdoors, the high variability of groups spatial configurations can potentially modify coordination mechanisms. We designed and tested an orienteering mobile learning game to better understand how device use shaped collaboration in highly mobile conditions. The study involved four groups of three students all equipped with tablets. We focused our analysis on the relationship between participants' arrangements (F-formations), their device usage and coordination mechanisms (i.e. awareness, regulation, information sharing, and discussion). Our results emphasize the importance of considering the transitions between arrangements more than F-formations per se. We discuss the implications of these findings for the design and analysis of mobile collaborative activities.

International audience; Mobile devices offer great opportunities in the field of collaborative learning for providing digital information while still supporting social interactions between group members. We designed and tested an orienteering mobile learning game to better understand how device use shaped collaboration in highly mobile conditions. The study involved four groups of three students all equipped with tablets. We focused our analysis on the relationship between participants’ arrangements (F- formations), their device usage and coordination mechanisms (i.e. awareness, regulation, information sharing, and discussion). Our results emphasize the importance of considering the transitions between arrangements. From these observations we derive recommendations for the design of relevant interactions techniques for mobile collaborative activities.

National audience; L’urbanisation de ces dernières décennies a indéniablement fait évoluer notre rapport aux plantes dans notre vie quotidienne. Il n’en reste pas moins que, d’une part, des générations de botanistes ont accumulé des connaissances concernant les espèces végétales et que, d’autre part, nombreuses sont les personnes intéressées par la nature qui les entoure et qui sont conscientes que les ressources naturelles sont un trésor à préserver. Cependant, il faut bien avouer que pour une majorité d’individus, la botanique reste difficilement abordable. Dans le cadre du projet ANR ReVeRIES (Reconnaissance de Végétaux Récréative, Interactive et Educative sur Smartphone), nous proposons d’utiliser les technologies mobiles pour aider l’homme à connaître son environnement et plus précisément les végétaux qui l’entourent.

International audience

International audience

International audience; We are investigating the problem of speaker and face identification in broadcast videos. Identification is performed by associating automatically extracted names from overlaid texts with speaker and face clusters. We aimed at exploiting the structure of news videos to solve name/cluster association ambiguities and clustering errors. The proposed approach combines iteratively two conditional random fields (CRF). The first CRF performs the person diarization (joint temporal segmentation, clustering, and association of voices and faces) jointly over the speech segments and the face tracks. It benefits from contextual information being extracted from the image backgrounds and the overlaid texts. The second CRF associates names with person clusters, thanks to co-occurrence statistics. Experiments conducted on a recent and substantial public dataset containing reports and debates demonstrate the interest and complementarity of the different modeling steps and information sources: the use of these elements enables us to obtain better performances in clustering and identification, especially in studio scenes.

International audience; Audio segmentation is often the first step of audio indexing systems. It provides segments supposed to be acoustically homogeneous. In this paper, we report our recent experiments on segmenting music recordings into singer turns, by analogy with speaker turns in speech processing. We compare several acoustic features for this task: FilterBANK coefficients (FBANK), and Mel frequency cepstral coefficients (MFCC). FBANK features were shown to outperform MFCC on a “clean” singing corpus. We describe a coefficient selection method that allowed further improvement on this corpus. A 75.8% F-measure was obtained with FBANK features selected with this method, corresponding to a 30.6% absolute gain compared to MFCC. On another corpus comprised of ethno-musicological recordings, both feature types showed a similar performance of about 60%. This corpus presents an increased difficulty due to the presence of instruments overlapped with singing and to a lower recording audio quality.

Musical transcription is a real challenge, moreover in a folk music context. Signal visualization tools could be of interest for such music. The present paper is a comparison of a musical transcription and two signal representations (pitch and rhythm) applied to a song from the Gwoka repertoire. The study aims at finding similar elements and differences on pitch, rhythm and performance features in both the transcription and the signal visualization. Signal visualization is founded on vowel segmentation, and extraction of pitch and duration information. On the one hand transcription gives general characteristics on the music (harmony, tonality and rhythmic structure) and on the other hand, signal visualization gives performance-related characteristics. The main conclusion is that both approaches are of great interest for understanding such a music.

International audience; In this letter, the authors propose a new embedding scheme for image-based continuous face pose estimation. The main contributions are as follows. First, it is shown that the concept of label-sensitive Locality Preserving Projections, proposed for age estimation, can be used for model-less face pose estimation. Second, the authors propose a linear embedding by exploiting the connections between facial features and pose labels via a sparse coding scheme. The resulting technique is called Sparse Label sensitive Locality Preserving Projections (Sp-LsLPP). Third, for enhancing the discrimination between poses, the projections obtained by Sp-LsLPP are fed to a Discriminant Embedding that exploits the continuous labels. The resulting framework has less parameters compared to related works. It has been applied to the problem of model-less face yaw angle estimation (person independent 3D face pose estimation). It was tested on three databases: FacePix, Taiwan, and Columbia. It was conveniently compared with other linear and non-linear techniques. The experimental results confirm that the proposed framework can outperform, in general, the existing ones.

National audience; Dans cet article, nous proposons tout d’abord un état de l’art des méthodes pour la visualisation et l’interprétation des données textuelles, et en particulier des données scientifiques. Nous présentons nos contributions à ce domaine, sous la forme de méthodes originales pour la classification automatique de documents et l’interprétation facile de leur contenu grâce à des mots-clés caractéristiques et à des classes créées par nos algorithmes. Ensuite, nous focalisons notre analyse autour des données évolutives dans le temps. Nous détaillons notre approche par diachronie, particulièrement appropriée à la visualisation des évolutions. Cela nous permet de conclure en présentant Diachronic’Explorer, un outil de visualisation destiné à l’exploration visuelle des données évolutives.; In this paper, we introduce a survey about methods used to visualize and interpret textual data, in particular scientific data. We then detail our contribution to this field, describing original tools for document clustering and cluster labeling to help users understand clustering results. Afterwards, we focus on time-evolving data. We introduce and detail our diachronic approach, particularly well-fitted to visualize evolutions in the data. It allows to conclude by presenting Diachronic'Explorer, a visualization tool which aims to deal with evolving data.

National audience; Dans cet article, nous proposons tout d’abord un état de l’art des méthodes pour la visualisation et l’interprétation des données textuelles, et en particulier des données scientifiques. Nous présentons nos contributions à ce domaine, sous la forme de méthodes originales pour la classification automatique de documents et l’interprétation facile de leur contenu grâce à des mots-clés caractéristiques et à des classes créées par nos algorithmes. Ensuite, nous focalisons notre analyse autour des données évolutives dans le temps. Nous détaillons notre approche par diachronie, particulièrement appropriée à la visualisation des évolutions. Cela nous permet de conclure en présentant Diachronic’Explorer, un outil de visualisation destiné à l’exploration visuelle des données évolutives.

International audience; This paper investigates single and cross-show diarization based on an unsupervised i-vector framework, on French TV and Radio corpora. This framework uses speaker clustering as a way to automatically select data from unlabeled corpora to train i-vector PLDA models. Performances between supervised and unsupervised models are compared. The experimental results on two distinct test corpora (one TV, one Radio) show that unsupervised models perform as good as supervised models for both tasks. Such results indicate that performing an effective cross-show diarization on new language or new domain data in the future should not depend on the availability of manually annotated data.

Every learning situation is designed according to specific instructional objectives. Instructional designers evaluate the correlation between the objectives and the real progress of the learning session by practicing an observation activity. In order to be efficient, the observation must be organized. Our goal in this work aims to provide tools and methods to support the observation activity. That is why we propose the concept of Observing Strategy. Our research aims to provide the actors of a learning system (especially the instructional designers), with a homogeneous set of tools for 1/ defining observation strategies and 2/ calculating and displaying observation results with an ergonomic and intuitive visualization interface. These two tools are designed to be used before, during and/or after the learning sessions. The underlying idea is to propose two separated, but communicating, tools allowing the formalization of observation strategies independent from the chosen TEL system. With the first tool, instructional designers can specify the organization of the observation by using a language and semantics close to their pedagogical practices, handling pre-defined indicators, and available visualization tools etc. Such observation strategy language will not deal with TEL-systems-dependent elements that could restrict the potential reuse of strategies (tracks/raw data, observed data, observing means, etc.).The receiver of the observation results can use the second tool to visualize the results of the indicators’ calculation with the right format and according to the chosen strategy specification., Toute situation d’apprentissage est conçue dans le but de répondre à des objectifs pédagogiques spécifiques. Les enseignants-concepteurs évaluent la corrélation entre cesobjectifs et le déroulement réel de la situation d’apprentissage en pratiquant une activité d’observation. Pour être efficace, cette activité a besoin d’être organisée. Le but de ce travail de recherche est de fournir des outils et des méthodes afin d’améliorer cette activité. C’est la raison pour laquelle nous avons proposé le concept de Stratégie d’Observation. Le but est d’apporter aux acteurs d’un système d’apprentissage, et particulièrement à l’enseignant-concepteur, un ensemble d’outils pour 1/ définir des stratégies d’observation et 2/ calculer et restituer les résultats à l’aide d’une interface de visualisation ergonomique et intuitive. Ces deux outils séparés, mais communiquant, sont conçus pour être utilisés avant, pendant et/ou après les sessions d’apprentissage. Le premier outil permet aux enseignants-concepteurs de spécifier la manière dont ils veulent organiser l’observation en utilisant un langage proche de leur langage pédagogique habituel. Ceci en manipulant des indicateurs prédéfinis et des moyens de perception mis à leur disposition, etc. Un tel langage permet de s’abstraire de l’environnement d’apprentissage et de ses éléments qui pourraient restreindre le potentiel de réutilisation des stratégies définies (traces, données brutes, moyens d’observation, etc.). Le destinataire des résultats de l’observation peut utiliser le deuxième outil pour visualiser les indicateurs dans un format adéquat et correspondant à la stratégie spécifiée.

International audience

International audience; The use of Learning Management Systems presents many difficulties for teachers in thinking about and designing their courses. These are related to the operationalization of learning scenarios and the appropriation of corresponding tools. The research presented in this paper aims to help trainers to control the mediated learning situations. For this purpose, we study existent tools and methods to assist their instructional design process. In this paper, we are specifically interested in the identification and the formalization of LMSs implicit instructional design language that will be the basis for the development of binding solutions. These solutions simplify instructional design on platforms and insure that future scenarios will be formalized and operationalized in conformance to the LMSs language without semantics loss. This process is dedicated for the LMS community and more specifically for pedagogical engineers and designers with a competence in IT. We illustrate the proposed process on the Moodle platform.

International audience; no abstract

Exploring the use of Attention-Based Recurrent Neural Networks For Spoken Language Understanding This study explores the use of a bidirectional recurrent neural network (RNN) encoder/decoder based on a mechanism of attention for a Spoken Language Understanding (SLU) task. First experiments carried on the ATIS corpus confirm the quality of the RNN baseline system used in this paper, by comparing its results on the ATIS corpus to the results recently published in the literature. Additional experiments show that RNN based on a mechanism of attention performs better than RNN architectures recently proposed for a slot filling task. On the French MEDIA corpus, a French state-of-the-art corpus for SLU dedicated to hotel reservation and tourist information, experiments show that a bidirectionnal RNN reaches a f-measure value of 79.51 while the use of a mechanism of attention allows us to reach a f-measure value of 80.27, L'étude porte sur l'apport d'un réseau de neurones récurrent (Recurrent Neural Network-RNN) bidirectionnel encodeur/décodeur avec mécanisme d'attention pour une tâche de compré-hension de la parole. Les premières expériences faites sur le corpus ATIS confirment la qualité du système RNN état de l'art utilisé pour cet article, en comparant les résultats obtenus à ceux récemment publiés dans la littérature. Des expériences supplémentaires montrent que les RNNs avec mécanisme d'attention obtiennent de meilleures performances que les RNNs récemment proposés pour la tâche d'étiquetage en concepts sémantiques. Sur le corpus MEDIA, un corpus français état de l'art pour la compréhension dédié à la réservation d'hôtel et aux informations touristiques, les expériences montrent qu'un RNN bidirectionnel atteint une f-mesure de 79,51 tandis que le même système intégrant le mécanisme d'attention permet d'atteindre une f-mesure de 80,27

International audience; no abstract

International audience; This paper investigates self trained cross-show speaker diarization applied to collections of French TV archives, based on an i-vector/PLDA framework. The parameters used for i-vectors extraction and PLDA scoring are trained in a unsupervised way, using the data of the collection itself. Performances are compared, using combinations of target data and external data for training. The experimental results on two distinct target cor- pora show that using data from the corpora themselves to perform unsupervised iterative training and domain adaptation of PLDA parameters can improve an existing system, trained on external annotated data. Such results indicate that perform- ing speaker indexation on small collections of unlabeled audio archives should only rely on the availability of a sufficient external corpus, which can be specifically adapted to every target collection. We show that a minimum collection size is required to exclude the use of such an external bootstrap.

International audience

International audience

Conférence EIAH 2015; National audience; Le projet GraphiT vise à aider des enseignants à se focaliser sur la conception de situations d’apprentissage tout en garantissant que les modèlesproduits sauront être exploités informatiquement pour faciliter la mise en oeuvre sur la plate-forme de formation mise à leur disposition (opérationnalisation). Notre approche originale est centrée sur l’abstraction du métier de conception d’une plate-forme afin de spécifier des langages de conception spécifiques. Cette problématique est traitée dans le cadre méthodologique de l’Ingénierie Dirigée par les Modèles et concerne, à titre de cas d’étude, la plate-forme Moodle. Notre proposition consiste à augmenter l’expressivité pédagogique des modèles de conception par la spécification de blocs qui correspondraient à des usages récurrents des outils, ressources et paramétrages de la plate-forme. Nous proposons de capturer ces éléments de conception ainsi que leurs correspondances pour faciliter la conception et diriger la traduction lors de l’utilisation d’outils dédiés que nous développons.

Conférence EIAH 2015; Le projet BASAR offre une banque de scénarios d’apprentissage hybrides. Ce projet permet une réutilisation et une capitalisation des bonnespratiques d’enseignement. Pour se faire, tout enseignant-concepteur a la possibilité de choisir un scénario qui répond à ses besoins, de le modifier, de l’utiliser et de le redéposer. Or si la spécialisation du scénario à un contexte donné améliore souvent la qualité de l’apprentissage, elle rend difficile sa réutilisation dans un autre contexte, même voisin. La connaissance des contextes adaptés à un scénario est indispensable pour mieux réutiliser en partie, voire en totalité, le scénario. Comment alors indexer le scénario d’apprentissage par leurs contextes les plus adaptés en se basant sur l’observation du déroulement des sessions d’apprentissage? Cet article propose une approche d’indexation contextuelle multi-facettes des scénarios d’apprentissage par les arbres de contexte. L’objectif principal de cette indexation est de faciliter la tâche de conception de scénarios par la réutilisation.

Conférence EIAH 2015; National audience; Avec le mouvement MOOC, les acteurs de l’éducation s’inscrivent dans le cadre d’une dynamique de renouvellement pédagogique. Cette dynamique présente différents avantages tels que : la massivité des apprenants, l’ouverture à tous, l’accessibilité, l’économie d’échelle, etc. Elle soulèvecependant des questions liées notamment à la problématique de l’abandon. Pour y remédier, une piste prometteuse réside dans le soutien à la motivation et à l’engagement des apprenants. Le but de cette communication est d'examiner du point de vue de la littérature, la question de l’abandon et les solutions possibles. Nous proposons également nos propres réflexions au fur et à mesure de notre démarche.

Conférence EIAH 2015; National audience; Les systèmes de recommandation ont pour objectif de proposer des ressources pertinentes à une personne en utilisant certaines informations sur les utilisateurs et les ressources du système. Notre travail s’intègre dans le contexte des systèmes de recommandation de ressources pédagogiques et plus particulièrement des systèmes qui utilisent des informations sociales. En se basant sur les résultats de recherche dans le domaine des systèmes de recommandation, des réseaux sociaux et des EIAH, nous avons défini une approche de recommandation de ressources pédagogiques en utilisant des informations sociales telles que les liens d’amitié et les appartenances aux groupes. Cette approche est basée sur un modèle formel qui permet de calculer la similarité entre des utilisateurs de l’EIAH afin de générer trois types de recommandation, à savoir : la recommandation des ressources populaires, utiles et des ressources visualisées récemment. Nous avons également développé une plateforme d’apprentissage qui intègre ce modèle de recommandation et qui nous a permis d'évaluer notre approche.

Conférence EIAH 2015; National audience; Les établissements scolaires sont de plus en plus équipés de dispositifs mobiles. Pourtant, des études récentes ont montré que leur utilisationpar les enseignants reste encore limitée à cause d’un manque de formation, de ressources et d’applications adaptées pour des utilisations pédagogiques. Parmi les nouvelles pistes d’utilisation, les recherches sur les Jeux Éducatifs sur Mobile (JEM) semblent très prometteuses. Dans cet article, nous proposons d’identifier les JEM qui sont issus des travaux de recherches les plus référencés et qui ont été expérimentés, afin d’étudier leurs propriétés et impacts. Ceci nous permet, dans un deuxième temps, de définir les caractéristiques communes de ces JEM et d’identifier les principaux problèmes d’utilisation par les enseignants.

Many learning environments are swiftly abandoned by the learners, even if they are effective. Gamification appears as a recent game-based learning approach to enhance the learners' motivation. However, individual expectations and preferences towards game-like features may be very different from one person to another. This paper presents a model to adapt gamification features according to a player profile of the learners. Two version of this model are evaluated within a gamified online learning environment. The first version comes from experts' judgment, and the second one is induced from empirical data. Our experiments confirm that the first version can be efficient to predict the player's preferences among the gamification features., Les environnements d'apprentissage sont fréquemment délaissés par les apprenants, malgré leur efficacité. La ludification est une approche de plus en plus utilisée pour répondre au manque de motivation des apprenants. La difficulté est que chacun a des sensibilités différentes aux mécaniques de jeu, et cela n'est généralement pas pris en compte. Cet article présente un modèle pour adapter les fonctionnalités ludiques selon les profils de joueur des apprenants. Nous comparons plusieurs méthodes pour implémenter ce modèle dans un environnement d'apprentissage ludifié : en se basant sur le jugement d'experts, ou sur des données empiriques. Nous avons implémenté un modèle développé avec la première méthode dans un environnement d'apprentissage en ligne. Les expérimentations présentées confirment l'efficacité du modèle basé sur le jugement d'experts.

This paper describes the progress made in the transcription of broadcast news (BN) and conversational telephone speech (CTS) within the combined BBN/LIMSI system from May 2002 to September 2004. During that period, BBN and LIMSI collaborated in an effort to produce significant reductions in the word error rate (WER), as directed by the aggressive goals of the Effective, Affordable, Reusable, Speech-to-text [Defense Advanced Research Projects Agency (DARPA) EARS] program. The paper focuses on general modeling techniques that led to recognition accuracy improvements, as well as engineering approaches that enabled efficient use of large amounts of training data and fast decoding architectures. Special attention is given on efforts to integrate components of the BBN and LIMSI systems, discussing the tradeoff between speed and accuracy for various system combination strategies. Results on the EARS progress test sets show that the combined BBN/LIMSI system achieved relative reductions of 47% and 51% on the BN and CTS domains, respectively

Magnetic resonance (MR) diffusion imaging provides a valuable tool used for inferring structural anisotropy of brain white matter connectivity from diffusion tensor imaging. Recently, several high angular resolution diffusion models were introduced in order to overcome the inadequacy of the tensor model for describing fibre crossing within a single voxel. Among them, q -ball imaging (QBI), inherited from the q -space method, relies on a spherical Radon transform providing a direct relationship between the diffusion-weighted MR signal and the orientation distribution function (ODF). Experimental validation of these methods in a model system is necessary to determine the accuracy of the methods and to optimize them. A diffusion phantom made up of two textile rayon fibre (comparable in diameter to axons) bundles, crossing at 90°, was designed and dedicated to ex vivo q -ball validation on a clinical scanner. Normalized ODFs were calculated inside regions of interest corresponding to monomodal and bimodal configurations of underlying structures. Three-dimensional renderings of ODFs revealed monomodal shapes for voxels containing single-fibre population and bimodal patterns for voxels located within the crossing area. Principal orientations were estimated from ODFs and were compared with a priori structural fibre directions, validating efficiency of QBI for depicting fibre crossing. In the homogeneous regions, QBI detected the fibre angle with an accuracy of 19° and in the fibre-crossing region with an accuracy of 30°.

The present work fits into the context of recommender systems for educational resources, especially systems that use social information. Based on the research results in the field of recommender systems, social networks and technology-enhanced learning, we defined an educational resource recommendation approach. We rely on social relations between learners to improve recommendation accuracy. Our proposal is based on formal models that generate three types of recommendation, namely recommendation of popular resources, useful resources and recently viewed resources. We developed a learning platform which integrates our recommendation models. In this paper, we present the results of an experiment conducted during six months in a real educational context. The goal of this experiment is to measure the relevance, quality and utility of the recommended resources. We also conduct an offline analysis by using a dataset in order to compare our approach with four baseline algorithms.

International audience; In a learning situation, gamification is a way to motivate learners and enhance their participation to learning activities by adding game elements. But it still pays little attention to the individual differences among learners’ preferences as players. This paper presents a generic and adaptive gamification system that can be plugged on various learning environments. The game elements can be automatically adapted, based on an analysis of the interaction traces. The architecture of the proposed system is based on the separation between the control of the pedagogical elements and the control of the game elements. To support the adaptivity of the game elements, we refer to a user model based on a list of player types. The user model is initialized thanks to the traces of interactions, and finally used to predict which game feature will be relevant for the user.

International audience; This study explores the use of a bidirectional recurrent neural network (RNN) encoder/decoder based on a mechanism of attention for a Spoken Language Understanding (SLU) task. First experiments carried on the ATIS corpus confirm the quality of the RNN baseline system used in this paper, by comparing its results on the ATIS corpus to the results recently published in the literature. Additional experiments show that RNN based on a mechanism of attention performs better than RNN architectures recently proposed for a slot filling task. On the French MEDIA corpus, a French state-of-the-art corpus for SLU dedicated to hotel reservation and tourist information, experiments show that a bidirectionnal RNN reaches a f-measure value of 79.51 while the use of a mechanism of attention allows us to reach a f-measure value of 80.27.

Conférence EIAH 2015; National audience; Le projet JEN.lab a pour ambition d’associer les caractéristiques des « jeux épistémiques numériques » (en particulier, les situations d’apprentissage ludiques conduisant l’apprenant à résoudre des problèmes complexes et non déterministes) et les technologies « pervasives » comme la réalité mixte et la réalité alternée, en vue de concevoir des situations d’apprentissage authentiques. Dans ce contexte, cet article propose une définition de ce que nous appelons des « jeux épistémiques numériques pervasifs » (JENP). Dans cet article, nous synthétisons un état de l’art effectué sur les jeux épistémiques numériques et sur les jeux pervasifs, et nous énumérons les bases d’un JENP en mettant en exergue sa valeur ajoutée pour la pédagogie. Enfin, nous concluons en soulignant la nécessité de réfléchir à une méthode de conception de ces outils pédagogiques émergents.

International audience; The pedagogical expressiveness of designed courses using Learning Management Systems (LMS) is strongly dependent on the teachers’ knowledge and expertise about how using the LMS at their disposal. The GraphiT project aims to help teachers in focusing on the specification of pedagogically sound learning scenarios that can be technically executable for automatically setting-up the LMS. We propose to provide teachers with LMS-specific instructional design languages and editors. Such objective requires, at first, to formally capture LMSs implicit learning design semantics. We already tackled this challenge by proposing a dedicated method of identification and formalization as a metamodel. Secondly, the LMS semantics has to be raised in order to enrich the pedagogical expressiveness of the produced models. This paper deals with the proposition of a specific LMS-centered approach for abstracting the LMS low-level parameterizations and turning them into higher-level pedagogical building blocks. We chose to present and illustrate our propositions about the Moodle LMS. We focus on the first abstraction level. It consists in identifying some pedagogical activities according to recurrent uses designers make by handling the Moodle activities.

International audience; Teacher's expertise on using Learning Management Systems (LMS) is tightly coupled to how they design their online courses. The GraphiT project aims to help teachers in specifying of pedagogically sound learning scenarios that can be technically executable for automatically setting-up the related LMS course. We intend to provide teachers with LMS-specific instructional design languages and editors. To achieve this goal, we have to raise the LMS semantics in order to enrich the pedagogical expressiveness of the produced models. We propose a specific LMS-centered approach for abstracting the low-level parameteriza-tions and turning these semantics into higher-level pedagogical building blocks. We present and illustrate our propositions focused on Moodle. In this paper, we focus on the first abstraction level: identifying pedagogical activities according to recurrent uses of Moodle activities.

International audience; This chapter aims to offer a user-centred methodological framework to guide the design and evaluation of Brain-Computer Interface videogames. This framework is based on the contributions of ergonomics to ensure these games are well suited for their users (i.e., players). It provides methods, criteria and metrics to complete the different phases required by ae human-centred design process. This aims to understand the context of use, specify the user needs and evaluate the solutions in order to define design choices. Several ergonomic methods (e.g., interviews, longitudinal studies, user based testing), objective metrics (e.g., task success, number of errors) and subjective metrics (e.g., mark assigned to an item) are suggested to define and measure the usefulness, usability, acceptability, hedonic qualities, appealingness, emotions related to user experience, immersion and presence to be respected. The benefits and contributions of the user centred framework for the ergonomic design of these Brain-Computer Interface Videogames are discussed.

International audience; Many learning environments are swiftly abandoned by the learners, even if they are effective. Gamification is as a recent game-based learning approach that can enhance the learners’ motiva-tion. However, individual expectations and preferences towards game-like features may be very different from one person to another. This paper presents a model to adapt gamification features according to a player profile of the learners. Two version of this model are evaluated within a gamified online learning environment. The first version comes from experts’ judgment, and the second one is induced from empirical data. Our experiments confirm that the first version can be efficient to predict the player’s preferences among the gamification features.