PhD / HDR

The thesis is part of a research field that seeks to reconnect with artificial intelligence as it existed with feature engineering, namely interpretability. The aim of this research field is to understand and explain neural models and their performances, by reconnecting the results to the data and to their humanly interpretable attributes. ► Read more

► Read more

Modèles, méthodes et outils afin de produire des Ressources Pédagogiques Virtuelles (RPV) pour l’apprentissage de gestes techniques et professionnels ► Read more

Analysis and automatic evaluation of the dental surgery gesture in a virtual training environment ► Read more

The purpose of this research is to develop a Khmer language model capable of correcting Khmer misspelling words in the education field. We will offer a model for correcting Khmer misspellings using a deep learning technique that enables the computer to self-learn using the BERT model utilizing data of 100,000 words from the Khmer dictionary. ► Read more

The research purpose of this thesis is resource-constrained multilingual machine translation of speech. It aims at proposing strategies for the translation and transcription of languages with few annotated data by using both phonetic and semantic information from other languages with more resources. One of the expectations of the thesis is also to show the interest of using a single model to solve the whole task compared to the use of cascade models. ► Read more

The objective of this thesis is to explore the interpretability of lexical representations, first by constructing intrinsic evaluation sets [LLBJ15], then by studying these representations in a temporal context. Finally, from the interpretability of our representation, it will be a question of analyzing the actualization of the lexicon in the utterance, by making the link between systems of attentions on these embeddings and interpretative semantics. ► Read more

Conception d’environnements éducatifs en Réalité Mixte pour aider à surmonter des obstacles didactiques : application à l’enseignement des fractions ► Read more

This project is at the intersection of situated learning and computer supported collaborative mobile learning. Its context are school trips supported by geographic maps. ► Read more

Our work takes place in the context of the adaptation of playful learning activities. The learning and gaming profiles of the learner-player must be taken into account during the adaptation. Therefore, the main focus on this research, in the context of activity generation, is the study of the interplay between pedagogy and game-based motivation. ► Read more

The main objective of the project is to propose, develop and validate methods that allow 1) to generate expressive speech from a user-given instruction using either text-to-speech systems or voice conversion; and 2) to interact with the system during learning and inference to correct the system's audio outputs. ► Read more

Word embeddings offer new possibilities for the study of textual corpora, in particular concerning the semantics of the vocabulary used in these corpora. At the crossroads of artificial intelligence and digital humanities, this project aims at providing the community with robust tools to design interpretable temporal lexical embedding and to apply them in contexts such as the detection and characterization of neologisms on large textual corpora, in the framework of the evolution of television language via transcribed Ina corpora or to evaluate the evolution of gender stereotypes over time.

Model and tool to assist the scenarisation of VR-oriented pedagogical activities ► Read more

Le projet national PIA écri+ (http://ecriplus.fr/) est dédié à l’accompagnement des étudiants dans le développement de leurs compétences écrites en français académique à l’université. Cette thèse se déroule en partenariat entre le LIUM (Laboratoire d’Informatique de l’Université du Mans) et le CREN (Centre de Recherche en Education de Nantes). La problématique de cette thèse consiste à proposer un cadre permettant d’exploiter les besoins d’observation provenant d’une quinzaine de partenaires académiques et de mettre en place une démarche de construction et de visualisation d’indicateurs dédiés à l’étude d’impact du dispositif écri+. L’objectif de livrable de la thèse sera de disposer d’un prototype d’environnement de visualisation intégrant des tableaux de bord de suivi pédagogique et de suivi de l’étude d’impact.

The subject is supported by two laboratories of Le Mans – Université: the acoustics lab (LAUM) and the computer science lab (LIUM). The aim is to enhance automatic speech processing in meetings, transcription and speaker recognition, by using a recording device and audio signal processing from a microphone array

This thesis is part of the research topics of the Language and Speech Technologies (LST) team at the Laboratoire Informatique de l'Université du Mans (LIUM). It will take place in Le Mans, as part of the ANR AISSPER project.

The work carried out in the framework of this thesis focuses on the extraction of semantic information from the audio signal (thematic segmentation, interaction graphs, role of the speaker, ...). In a first step, a speaker segmentation tool capable of automatically identifying superimposed speech zones from the audio signal will be developed. In a second step, from annotated INA data (incivilities, covid-19), an automatic characterization of speech interruptions will be carried out on a large scale in collaboration with SHS researchers. This characterization will involve the simultaneous use of acoustic, linguistic and possibly para-linguistic representations.

This thesis is about the use of Teaching Analytics to analyze teachers' behaviors and assist them in the use of Learning Management Systems (LMS). ► Read more

the objective of the proposed thesis is twofold. First, it aims at finding a privacy preserving transform of the speech data, and, second, it will investigate the use of additional personalized transforms, that can be applied on the user’s terminal, to increase speech recognition performance. ► Read more

► Read more

L’objectif de recherche principal de la thèse est de proposer une système d'évaluation en temps-réel des activités pédagogiques des utilisateurs en réalité virtuelle à partir des données sur leurs évolutions dans le temps. ► Read more

This thesis is part of the ALLIES project, whose objective is to encourage and demonstrate the development of autonomous systems, capable of maintaining sustainable performance over time under a given learning scenario.

Les principaux objectifs de cette thèse sont de concevoir, implémenter et expérimenter des approches neuronales end-to-end dédiées à des tâches d’extraction d’informations sémantiques de la parole, étendues à la recherche d’informations paralinguistiques liées aux émotions. Clairement, deux tâches sont visées : l’extraction de sens et la détection d’émotions, qui pourront être réalisées de façon jointe ou disjointe. L’accent sera mis sur le traitement de masses de données en temps réel, qui offre de véritables opportunités industrielles. ► Read more

Neural approaches for abstractive text summarization ► Read more

The aim of this thesis is to develop a named entity recognition system in an audio stream that will rely solely on a deep neural network. Until now, this task has been carried out via successive processing chains. Also, speech and named entity recognitions fully neuronal tasks have been greatly improved in recent years. ► Read more

Socio-constructivism and connectivism theories pinpoint the importance of collaboration for learning. Nevertheless, the online social interactions underlying the collaboration processes are still not well understood. As a result, learning designers have difficulties creating effective collaborative activities in MOOC. ► Read more

With the emergence of virtual reality, computing offers new experiences for users thanks to increasingly powerful interaction and immersion possibilities. These opportunities are of great interest in the field of learning. Virtual environments allow the creation of original and dynamic learning situations, detached from the constraints that may exist during real training (danger, cost, uncertainty) and bringing specific advantages (enrichment of situations, replay, etc.) ► Read more

Sentiment analysis is a growing field of research and has been subject of numerous studies. This thesis aims at designing a hybrid pragmatic approach for Arabic sentiment analysis. More specifically, it involves injecting heterogeneous knowledge for the construction of a neural architecture in order to improve the performance of the task. ► Read more

This thesis is a part of the PASTEL project (Performing Automated Speech Transcription for Enhancing Learning), which aims to explore the potential of real-time automatic transcription for the instrumentation of mixed teaching situations, where the modalities of interaction are presential or remote, synchronous or asynchronous. ► Read more

The first scientific objective of the thesis concerns the design and development of a software tool enabling the simple configuration of educational scenarios of different complexities (authoring tool). This tool will offer various specific virtual environments (3D scenes) as well as standardised tools that can be imported into applications to simulate different tasks or activities. A module for visualizing the performance and behaviour of learners will be developed in order to carry out debriefing sessions with the teacher or in group. ► Read more

La thèse proposée s’articulera autour de la reconnaissance de locuteurs à large échelle dans les archives radio et télévisuelles de l’INA (Institut National de l’Audiovisuel), qui dispose d’une quantité impressionnante de documents. L’annotation manuelle de ces documents représente une source d’information précieuse pour l’exploitation et la commercialisation de ces données, mais nécessite un temps de travail considérable. Les annotations permettent d’enrichir les documents en décrivant notamment l’identité des locuteurs, au fil du document, ou les thèmes abordés. ► Read more

The aim of this thesis is to develop variable context speaker models (scalable) that integrate phonetic information produced by the speaker. These models will be learned on a significant amount of enrollment data (> 30s) and will adapt to the test data to ensure the best possible comparison based on the phonetic context recognized in the test sample. These models will improve the performances of recognition systems and broaden the application framework of speaker recognition. ► Read more

The thesis' objective is the proposition of a methodology guiding users in the instrumentation of learning activities, resulting in new instrument uses. This instrumentation is designed by iterations, and relevant to learners' needs for information and communication. ► Read more

In a similar way than smartphones augmented reality eyewear devices are poised to become ubiquitous by providing a quicker and more convenient access to information. There is theoretically no limit to their applicative area and use cases, many of them already explored such as military, medical, industry, education, entertainment… Some interactions are becoming a standard, such as mid-air hand gesture and voice command. Paradoxically, in many use cases where these kinds of devices are currently implemented the user cannot perform these interactions without constraint: e.g. when the users are already using their hands to hold something ... in a noisy environment or the opposite where silent is required and the vocal command could not be used properly, or even in a social context where both gesture and vocal command could be seen as weird for the users. Thus, the thesis project aims to extend interactivity of augmented reality eyewear devices: 1) By providing more discrete interaction such as head gesture based on cognitive image schemas theory, metaphorical extension and natural user interfaces based on the smartwatch finger gesture. 2) By using the context of the user to provide the more convenient interface and feedback in the right space and time. Machine learning technologies will be used both for the implementation of the interactions and for the context awareness. Many user experiments are (and will be) conducted to assess the solutions. The underlying objective of this project is to facilitate the acceptance and usage of augmented reality eyewear devices. ► Read more

The objective of this thesis is machine translation through multilingual neural systems. The effectiveness of these approaches as already been shown in monolingual machine translation where you translate from a single source language to a single target language. ► Read more

Speech technology development was split through a few reseach centers, each working on a small number of languages. The problem of fast development of new languages was actually posed relatively late. This issue is now considered strategic by industry, the rapid enrichment of a catalogue of languages naturally contribute to open new markets. ► Read more

Learning and applications of distributed multilingual representations ► Read more

This thesis work is done in connection with the European project JOKER (JOKe and Empathy of a Robot/ECA: Towards social and affective relations with a robot). This project is funded by the program CHIST-ERA (European Coordinated Research on Long-term Challenges in Information and Communication Sciences & Technologies ERA-Net). ► Read more

This thesis is part of a general problem related to technology enhanced learning (TEL) and more specifically to the construction of analysis processes to accompany the decision-making of actors involved in the teaching and learning system (teachers, learners, designers, administrators, etc.). The TEL researcher is also involved in the production of concepts, indicators or models. More specifically, we are interested in the visualization stage of this analysis process by proposing models describing the different aspects of a dashboard and a generation process to produce an adaptive, contextual and interactive dashboard for operational actors. To define what a learning dashboard is, we use the following Schwendimann definition:"A learning dashboard is a single display that aggregates different indicators about learner (s), learning process (es), and/or learning context (s) into one or multiple visualizations". In this definition, we note in particular the aggregation of indicators on learners, learning processes and taking into account the learning context, which corresponds to the adaptive and contextual aspects of our dashboards. This work is part of the ANR HUBBLE project, which aims to create a national observatory for the construction and sharing of mass data analysis processes based on traces left in e-learning environments. A special feature of the project is the use of several analysis platforms, such as UTL (Usage Tracking Language), KTBS (Kernel for Trace Based System) and UnderTracks, either separately or jointly. A HUBBLE dashboard must therefore be able to integrate the visualization of indicators from these different platforms. We have sought to identify generic dynamic, adaptive and contextual dashboard structures to meet users' needs, which should be capitalizable and reusable to facilitate users' work. The ensuing questions are (i) is it possible to describe a generic dashboard, but dedicated to an observation objective? (ii) Is context an important element in the adaptive dashboard generation? (iii) is it possible to dynamically generate a dashboard adapted to a user and his activity? ► Read more

This thesis' goal is to develop a Technology Enhanced Learning (TEL) environment allowing a student to improve his motion learning, by using his motion data, with machine learning techniques (clustering). ► Read more

Differences between training and testing conditions may significantly degrade recognition accuracy in automatic speech recognition (ASR) systems. Adaptation is an efficient way to reduce the mismatch between models and data from a particular speaker or channel. There are two dominant types of acoustic models (AMs) used in ASR: Gaussian mixture models (GMMs) and deep neural networks (DNNs). ► Read more

Communication between humans across the lands is difficult due to the diversity of languages. Machine translation is a quick and cheap way to make translation accessible to everyone. Recently, Neural Machine Translation (NMT) has achieved impressive results. ► Read more

This thesis is in the field of Virtual Environments for Human Learning (VEHL). It was funded by the JEN.lab national ANR project, which aims to create Digital Epistemic Games (DEG), a category of Serious Games dedicated to solving complex, multidisciplinary and non-deterministic problems. The objective of DEGs is to propose authentic learning situations in terms of interactions so that learners can construct and anchor knowledge in their context of use. ► Read more

Ce projet de thèse s’intègre dans une problématique générale liée aux EIAH et plus spécifiquement aux MOOCs (Massive Open Online Courses). Les travaux effectués dans ce champ disciplinaire impliquent plusieurs voies de recherches complémentaires. Nous nous focaliserons principalement sur trois aspects principaux de ces recherches. ► Read more

This thesis concerns a study of continuous word representations applied to the automatic detection of speech recognition errors. Recent advances in the field of speech processing have led to significant improvements in speech recognition performances. However, recognition errors are still unavoidable. This reflects their sensitivity to the variability, e.g. to acoustic conditions, speaker, language style, etc. Our study focuses on the use of a neural approach to improve ASR error detection, using word embeddings. These representations have proven to be a great asset in various natural language processing tasks (NLP). ► Read more

La segmentation et regroupement en locuteurs (SRL) de collection cherche à répondre à la question « qui parle quand ? » dans une collection de documents multimédia. C’est un prérequis indispensable à l’indexation des contenus audiovisuels. La tâche de SRL consiste d’abord à segmenter chaque document en locuteurs, avant de les regrouper à l’échelle de la collection. ► Read more

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 approach consists in progressively revealing the functionalities of the authoring tool, according to the complexity level that corresponds to the skills and needs of each user. For our case of study, we offer three design levels in the JEM iNVENTOR tool. Thus, the standard level allows MLGs based on a generic model to be configured, the intermediate level allows customized scenarios to be created, and finally, the expert level allows customized MLGs to be programmed. 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. ► Read more

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. ► Read more

This thesis aims at developing neural machine translation architectures by integrating different types of informations to improve the quality of the generated translations. More specifically, we are interested to guide the neural network towards a multimodal representational space by making use of parallel and multimodal corpora with images and their descriptions in multiple languages. ► Read more

Cette thèse s’inscrit à la fois dans le domaine scientifique de l’ingénierie des Environnements Informatiques pour l’Apprentissage Humain (EIAH) et de l’Ingénierie Dirigée par les Modèles (IDM). Elle s’est déroulée dans le cadre du projet de recherche ANR GraphiT. Ces travaux sont fondés sur le constat que les enseignants sous-exploitent les plate-formes de formation en ligne mises à leur disposition. ► Read more

The topic structuring is an area that has attracted much attention in the Natural Language Processing community. Indeed, topic structuring is considered as the starting point of several applications such as information retrieval, summarization and topic modeling. ► Read more