Search

International audience; A critical task for tutors is to provide learners with suitable reading materials in terms of difficulty. The challenge of this endeavor is increased by students' individual variability and the multiple levels in which complexity can vary, thus arguing for the necessity of automated systems to support teachers. This chapter describes ReaderBench, an open-source multi-dimensional and multilingual system that uses advanced Natural Language Processing techniques to assess textual complexity at multiple levels including surface-based, syntax, semantics and discourse structure. In contrast to other existing approaches, ReaderBench is centered on cohesion and makes extensive usage of two complementary models, i.e., Cohesion Network Analysis and the polyphonic model inspired from dialogism. The first model provides an in-depth view of discourse in terms of cohesive links, whereas the second one highlights interactions between points of view spanning throughout the discourse. In order to argue for its wide applicability and extensibility, two studies are introduced. The first study investigates the degree to which ReaderBench textual complexity indices differentiate between high and low cohesion texts. The ReaderBench indices led to a higher classification accuracy than those included in prior studies using Coh-Metrix and TAACO. In the second study, ReaderBench indices are used to predict the difficulty of a set of various texts. Although the high number of predictive indices (50 plus) accounted for less variance than previous studies, they make valuable contributions to our understanding of text due to their wide coverage.

International audience; Natural Language Processing (NLP) became a trending domain within recent years for many researches and companies due to its wide applicability and the new advances in technology. The aim of this paper is to introduce an updated version or our open-source NLP framework, ReaderBench (http://readerbench.com/), designed to support both students and tutors in multiple learning scenarios that encompass one or more of the following dimensions: Cohesion Network Analysis of discourse, textual complexity assessment, keywords’ extraction using Latent Semantic Analysis (LSA), Latent Dirichlet Allocation (LDA) and word2vec semantic models, as well as the analysis of online communities and discussions. The latest version of our ReaderBench framework (v4.1) includes: a) new features, Application Programing Interfaces (APIs) and visualizations (e.g., sociograms, analysis of interaction between participants inside a community), b) a new web interface written in Angular 6, and c) the integration of new technologies to increase performance (i.e., spaCy and AKKA), as well as modularity and ease of deployment (i.e., Artifactory and Maven modules). ReaderBench is a fully functional framework capable to enhance the quality of learning processes conducted in multiple languages (English, French, Romanian, Dutch, Spanish, and Italian), and covering both individual and collaborative assessments.

International audience; Natural Language Processing has massively evolved during the last years and many up-to-date applications integrate different speech tools in order to create an enhanced user experiences. For obtaining a seamless integration of existing speech recognition systems, there is a trending interest for developing and improving existing speech-to-text algorithms. The aim of this paper is to improve user interaction with the ReaderBench platform, by developing and integrating a speech recognition module designed so that young pupils can dictate their self-explanations to a given text. Afterwards, the ReaderBench framework is used to automatically evaluate the employed reading strategies based on the resulted speech transcriptions. A dataset containing 160 self-explanations from students ranging from 9 to 11 years old was analysed using both original transcripts, and the ones automatically generated by our custom speech recognition system. Multiple methods designed to perform speech recognition are also compared, while a new dedicated model was trained in order to improve the quality of the existing French model for speech recognition from CMUSphinx speech recognition system. Our revised model includes a pronunciation dictionary obtained after training a Long Short-Term Memory (LSTM) Grapheme-to-Phoneme neural network. The accuracy of our system is benchmarked in relation to the automated process of identifying reading strategies implemented in our ReaderBench framework, which is applied on both manual transcriptions and automated speech-to-text inputs. The obtained results argue for the adequacy of our method as the slight decrease in terms of identification accuracy is justifiable in contrast to the effort of manually transcribing each self-explanation.

National audience; Il est de plus en plus question, et la diffusion des MOOCs (Massive Online Open Courses) le montre, de promouvoir des ressources éducatives ouvertes pour aider l’apprentissage. Pour autant, dans beaucoup de cas, « libre » signifie seulement « gratuit », et peu importe si cette gratuité s’accompagne en réalité d’une privatisation des données privées (Morozov, 2014), rendue encore plus performante par l’utilisation de techniques de Learning Analytics (LA). Ainsi, le courant des LA est souvent perçu comme gênant l’ouverture plutôt que la promouvant.Notre argument est que la puissance de traitement des LA, plutôt que d’être utilisées à assujettir un peu plus les utilisateurs en les espionnant ou les contrôlant, devrait plutôt être utilisée pour promouvoir des ressources libres au sens de Hilton et al. (2010), c’est-à-dire, des ressources non seulement gratuites, mais aussi accompagnées des permissions suivantes : retenir, réutiliser, réviser, remixer et redistribuer. Nous reprendrons ces droits en les exemplifiant avec ReaderBench, un système multilingue libre et gratuit que nous développons.

National audience; Vocabulary acquisition is considered one of the most important aspects of a language because it defines and constraints communication. In order to overcome this, students and teachers use various techniques meant to improve the learning process and facilitate word acquisition. Using conceptual networks that establish links to prior knowledge is an approach with proven good results. The game presented in this paper, Semantic Taboo, is a serious game that helps users enlarge their conceptual network. The game has two modes, one for teachers and one for students. In the teacher module, users have the option of adding playing cards. The theme of the card (the seed) is taken as input, while the list of taboo words is selected from the list of semantically related words provided by the ReaderBench framework. The student model has the same game principle as the traditional game, but the list of taboo words and the description of the seed help the learner acquire new words. The preliminary validation consisted of 15 users who were impressed by the game concept and its educational value.

International audience; Dialog plays a central role in learning processes, particularly in shared understanding and social knowledge construction. Based on the socio-cultural paradigm, the polyphonic model of dialog considers social knowledge-building as inter-animation of voices in an analogy with polyphonic music, where novelty, creativity and coherence are assured by an interplay of dissonances and consonances, evolving towards a harmonious whole. In this analogy, voices are participants' newly introduced concepts and points of view. Automated tools implementing the polyphonic model cope with the high volume and complexity of data by relying on Cohesion Network Analysis, and applying advanced Natural Language Processing techniques. Corresponding tools were successfully used for automated dialog analysis in a wide variety of educational settings, ranging from formal philosophical discussions in text-based discussion forums to informal online blog discussions. Although the presented methods are proven to reliably predict comprehension of contents and further participation in discussions, they are limited by the nature of text-based communication. Further development of the polyphonic model will be extended to gestures and body language, both in individual and group settings.

International audience; Computer Supported Collaborative Learning (CSCL) environments have become a viable learning alternative from which valuable data can be extracted and used for advanced analyses centered on evaluating participants' involvement and their interactions. Such automated assessments are implemented within our ReaderBench framework, a Natural Language Processing platform that contains multiple advanced text analysis functionalities. The ReaderBench framework is based on Cohesion Network Analysis from which different sociograms, relying on semantic similarity, are generated in order to reflect interactions between participants. In this paper, we briefly describe the enforced mechanisms used to analyze three scenarios: individual chat conversations, virtual communities of practice and MOOCs.

National audience; Nous décrivons un prototype de Manuel ouvert, online et massif (MOOT) auquel on peut intégrer diverses fonctionnalités utilisant des tech- niques de Traitement automatique de la langue, afin de promouvoir l’apprentissage auto-régulé. Nous discutons également de l’intérêt de disposer de ressources largement ouvertes, accessibles et flexibles dans l’enseignement supérieur.

International audience; In this paper we introduce the online version of our ReaderBench framework, which includes multi-lingual comprehension-centered web services designed to address a wide range of individual and collaborative learning scenarios, as follows. First, students can be engaged in reading a course material, then eliciting their understanding of it; the reading strategies component provides an in-depth perspective of comprehension processes. Second, students can write an essay or a summary; the automated essay grading component provides them access to more than 200 textual complexity indices covering lexical, syntax, semantics and discourse structure measurements. Third, students can start discussing in a chat or a forum; the Computer Supported Collaborative Learning (CSCL) component provides in- depth conversation analysis in terms of evaluating each member’s involvement in the CSCL environments. Eventually, the sentiment analysis, as well as the semantic models and topic mining components enable a clearer perspective in terms of learner’s points of view and of underlying interests.

International audience; ReaderBench is a fully functional automated software framework, designed to be an educational helper for students, teachers, and tutors. The system makes use of text-mining techniques based on advanced natural language processing and machine learning algorithms to design and deliver summative and formative assessments using multiple data sets (e.g., textual materials, behavior tracks, self-explanations).

Each domain, along with its knowledge base, changes over time and every timeframe is centered on specific topics that emerge from different ongoing research projects. As searching for relevant resources is a time-consuming process, the automatic extraction of the most important and relevant articles from a domain becomes essential in supporting researchers in their day-today activities. The proposed analysis extends other previous researches focused on extracting co-citations between the papers, with the purpose of comparing their overall importance within the domain from a semantic perspective. Our method focuses on the semantic analysis of paper abstracts by using Natural Language Processing (NLP) techniques such as Latent Semantic Analysis, Latent Dirichlet Allocation or specific ontology distances, i.e., WordNet. Moreover, the defined mechanisms are enforced on two different subdomains from the corpora generated around the keywords " e-learning " and " computer ". Graph visual representations are used to highlight the keywords of each subdomain, links among concepts and between articles, as well as specific document similarity views, or scores reflecting the keyword-abstract overlaps. In the end, conclusions and future improvements are presented, emphasizing nevertheless the key elements of our research support framework.

International audience; With the increasing spread of the social web, identifying emotions in texts has proved to have various applications in fields like opinion mining or market analysis. Emotion recognition from written statements does not only reveal information about the person who wrote them, but can also be used in predicting how the emotional state of the readers can be affected. We propose a novel automatic method for analyzing texts that predicts how reading a news article can influence in turn the emotional state of the reader. This method integrates several word-count approaches and natural language processing techniques, such as Latent Semantic Analysis. Moreover, our implemented system contains a module designed to personalize the provided feedback according to the reader's current emotional state. A preliminary validation has been performed and results are promising.

National audience

National audience; Pensum et ReaderBench, deux environnements personnels d'appren- tissage dont le but est de favoriser l'apprentissage auto-régulé, sont tour à tour présentés. Le premier évalue automatiquement, dans les résumés ou synthèses de cours produites, les sauts de cohérence, les hors-sujets, et la pertinence des phrases. Le deuxième évalue la complexité textuelle des textes soumis à la lec- ture et les stratégies de lecture à partir des auto-explications des élèves. Leur ar- chitecture et les méthodes de traitement automatique de la langue utilisées sont décrites et discutées.

International audience; The purpose of this paper is to present and discuss a Natural Language Processing-based operationalization of Bakhtin's dialogism ideas, and to insert it in the context of distance learning. Three core concepts from Bakhin are introduced and their application in the domain of distance learning discussed. Then, two computer-based systems based on these ideas, PolyCAFe and Pensum are presented.

This deliverable describes the framework, approach and early planning for validation of the software services and evaluation of the LTfLL methodology, project and work approach.