Events / Seminars

Aran Mohammadamini, postdoc fellow at LIUM, will be presenting his research work during a team seminar on Friday 05 April at 10.15 am in the IC2 boardroom. ► Read more

Cécile Macaire, a PhD student in the GETALP team at the LIG under the supervision of Benjamin Lecouteux, will be presenting her research work during a team seminar on Friday 22 March at 10.15 am in the IC2 boardroom. ► Read more

The seminar of the day is given by Jane Wottawa, lecturer at the LIUM and in the Department of Literature on her work. ► Read more

Life, universe and the rest ► Read more

Innovation in medical education: from today's simulation centers to tomorrow's. ► Read more

The LIUM welcomes Rémi Uro, a PhD student at INA, who will present part of his thesis work. ► Read more

Today's seminar is given by Aghilas Sini, Senior Lecturer at the LIUM, on his previous work. ► Read more

Emerging Computational Language Technologies in Higher Education: Insights from Biology and Computer Science ► Read more

The LIUM welcomes Mathieu Fontaine, researcher at Télécom Paris. He'll discuss three recent projects: two conducted with Télécom Paris students and another in collaboration with the RIKEN Artificial Intelligence Project (AIP) Sound Scene Understanding (SSU) team in Japan. ► Read more

The members of the LST team got away from it all for 2 days of intensive seminars and project presentations. We were also able to enjoy the seminars given by two external guests, Mathieu Fontaine, from Telecom Paris and Adrien Bardet, from Atlantique Traduction, many thanks to them. ► Read more

The LIUM welcomes Olivier Crouzet, researcher at the Laboratoire de Linguistique de Nantes (LLING).
Elisabeth Delais-Rousserie and one of their students, Pascal Yapo, will also be present. Please note that the seminar will take place 1 hour earlier than usual. The seminar will take place in the boardroom of the IC2 building. ► Read more

Séminaire de Benjamin GEORGES sur l'IA pour la reconnaissance d'objet automatique ► Read more

The next LST seminars will be given by Emmett Strickland (MoDyCo) in person and Marc Evrard (LISN) remotely and will be entitled respectively :
"Prosody of Nigerian pidgin: corpus approach and experimental validation with a text-to-speech system" et
"Annotation and proposed alternative to annotation for automatic detection of expressive speech". ► Read more

Séminaire d'Anne DASTUGUE. ► Read more

Laurianne is a Future Fellow of the Australian Research Council (ARC), and Associate Professor in the school of Computer Science at the Queensland University of Technology (QUT) in Brisbane. Her seminar is entitled "Co-designing intelligent systems that can enrich inclusive pictorial communication" ► Read more

Fethi Bougares, a teacher-researcher at the LIUM on secondment to ElyaData, is currently at the lab as part of ESPERANTO. He will be giving a seminar on his work at Elyadata on 9 June at 2pm in the amphitheatre of the IC2 building. ► Read more

On Friday 2 June at 11am in the boardroom, we will be holding a double seminar on the latest evaluation campaigns (after IWSLT) in which the team has participated. ► Read more

Hugo Riguidel, PhD student at LIUM, will present his research work during a seminar entitled "ON-TRAC consortium systems for the IWSLT 2023 dialectal and low-resource speech translation tasks". ► Read more

Simon Guillot and Thibault Prouteau, PhD students at LIUM, will present their research work during a seminar entitled "Sparser is better: one step closer to word embedding interpretability". ► Read more

Martin Rodrigue Atangana Ongolo and Yannick Yomie Nzeuhangi, PhD students at the University of Yaoundé 1 and currently in secondment at the University of Grenoble Alpes in the framework of ESPERANTO, will present their research work during seminars entitled "Automatic translation for the Ewondo language" and "Automatic speech recognition for the Ewondo language". ► Read more

Séminaire de Damien BRUN ► Read more

The next LST seminar will be given by Thibault Rouxi, PhD student at LS2N, LIA and LIUM and will be entitled "Evaluation of Language Model Rescoring in Automatic Speech Recognition". ► Read more

The next LST seminar will be given by Mano Brabant, PhD student at LIUM, who will present his advances on explicability and give us a feedback on the training he followed at Jean Zay last December. ► Read more

The next LST seminar will be given by Rémi Uro, PhD student at INA and will be entitled "Anticipation of end of speech turn for interruption detection". ► Read more

Next LST seminar will be given by Salima Mdhaffar, postdoc at LIA (Avignon Université) and will address "End-to-end model for named entity recognition from speech without paired training data" ► Read more

Preregistration is the practice of registering the hypotheses, methods, and/or analyses of a scientific study before it is conducted. It’s «emergence » in diverse disciplines has gained prominence within the open science community as a solution to some issues of the replication crisis (research practices, including p-hacking, publication bias, data dredging, inappropriate forms of post hoc analysis, and HARKing). But pre-registration has other advantages, including self-reflection of own research, credibility, among other. ► Read more

Next LST seminar will be given by Julien Hauret, PhD students at LMSSC (CNAM de Paris) and will address "Extension of the bandwidth for voice signals captured by noise-proof microphones" ► Read more

► Read more

Next LST seminar will be given by Valentin Pelloin and Martin Lebourdais, PhD students at LIUM and will address "ASR-Generated Text for Language Model Pre-training Applied to Speech Tasks" and "Overlapped speech and gender detection with WavLM pre-trained features" respectively. ► Read more

L'utilisation de plates-formes de validation de programmes est de plus en plus fréquente pour l'apprentissage des bases de l'informatique. Si l'automatisation des tests de validation libère du temps pour l'enseignant et ludifie le parcours de l'apprenant, on observe également certains biais induits par ces dispositifs pédagogiques. Des recherches en Intelligence Artificielle (IA) visent à pallier certains de ces biais. Pour cela, elles se heurtent au problème de l'analyse fine de ces programmes par les algorithmes d'apprentissage automatique. Dans cet exposé nous présentons une nouvelle méthode d'apprentissage de représentations de programmes (embeddings). Dans une approche centrée 'enseignant', nous proposons également quelques usages possibles de ces représentations pour la détection automatique de réponses. ► Read more

Next LST seminar will be given by Mohamed Ettaleb, Post-doctoral fellow at LIUM and will address "Evaluation of weakly-supervised methods for aspect extraction". ► Read more

Collaboration is a 21st century super skill. The detection of collaboration quality can help to support collaboration. This presentation addresses the automatic collaboration quality detection in a face-to-face setting in 3 different steps by summarizing my doctoral thesis: (1) to define the components to detect collaboration quality using audio data; (2) to explain the design of a sensor-based set up for automatic collaboration analytics using natural language processing, topic modelling, speech content analytics and learning analytics measures; (3) to move towards quantifying the quality of collaboration by using the automatic collaboration analytics set up and show these analysis using meaningful visualizations on an interactive dashboard consisting of network graphs. In addition to it, we address the challenges and issues at hand and how can solutions be built upon the work already done in this doctoral thesis. ► Read more

Next LST seminar will be given by Martin Lebourdais, PhD student at LIUM and will address "Overlapping speech and gender, a study of annotations for audiovisual media.". ► Read more

Next LST seminar will be given by Loïc Grobol, Assistant Professor at Université Paris Nanterre and will address "Trees, knights and puppets: transfer learning for processing of historical languages". ► Read more

The next LST seminar will be presented by Frédéric Blain, lecturer at University of Wolverhampton and will be entitled "Quality Estimation: Application to Machine Translation". ► Read more

The next LST seminar will be presented in the framework of the Antract project by Franck Mazuet, filmmaker and researcher at the CHS-CNRS (Centre d'Histoire Sociale du XXé siècle) and will be entitled "Les Actualités Françaises (1945-1969): The cinema of the event". ► Read more

Next LST seminar will be given by Peter Vickers, PhD student in Sheffield and will address .Multimodal Reasoning: Datasets, Models, and even some commonsense. ► Read more

La popularisation du terme Metavers en octobre 2021 et la facilité d’accès aux nouvelles technologies de Réalité Etendue (XR) ont fait entrer les mondes virtuels dans le domaine public. De nombreuses industries s’intéressent déjà à ces technologies pour la formation de leurs collaborateurs et pour faciliter la collaboration inter-métier. Le secteur de l’éducation pourrait également bénéficier de ces nouveaux mondes en adaptant ses méthodes pour tirer parti des possibilités offertes par ce concept. Durant ce séminaire, nous observerons quelques applications actuelles des technologies de XR pour la formation et la collaboration dans différents domaines (médical, industriel, …) afin d’ouvrir une réflexion sur les avantages que ces technologies pourraient apporter à la pédagogie de demain. ► Read more

Next LST seminar will be given by Pierre Champion, Phd student at INRIA of Nancy and LIUM and will address "Privacy-Preserving Speech Representation Learning using Vector Quantization" ► Read more

Next LST seminar will be given remotly by Chenghua Lin, lecturer at the University of Sheffield and will address "Summarising Historical Text in Modern Languages" ► Read more

Next LST seminar will be given by Mélanie Lancien, from the university of Lausanne and will address "Extending the study of speech accommodation: the relationship between hypospeech and speakers’ social proximity" ► Read more

The next LST seminar, presented by Sylvain Meignier, will focus on "careers in research". ► Read more

Lucas Ondel defended his PhD at Brno and is now in Post-Doc at Université Paris-Saclay. He will present his work on Friday 12/11 at 11:30. ► Read more

Anthony Larcher will introduce some technical aspects of how research is done today. ► Read more

Cendrine MERCIER est Maitre de Conférences au CREN. Elle est également référente numérique des 5 INSPE de l'Académie de Nantes. ► Read more

On October 1st, the day of the LST team takes place. On this occasion, researchers - the young and the more experienced - present their research themes. ► Read more

A recommendation approach based on data mining and multi-layer labeled graphs.  

The objective of a recommendation system is to assist users in choosing relevant items from a large set of elements. In the current context of the boom in the number of academic publications available (books, articles, etc.) online, providing a personalized recommendation service becomes a necessity.

Aurélie Jean nous entraîne dans un nouveau voyage : au cœur de nos institutions juridiques et des algorithmes qui s’y exercent. Comment la loi est-elle pensée et appliquée au temps des algorithmes ? Comment les algorithmes sont-ils utilisés au sein du système judiciaire ? Et est-il vraiment possible de les réguler ? ► Read more

Personal data protection in the era of the RGPD  

The CNIL is the authority in charge of personal data protection in France.This presentation will review the birth of the CNIL in 1978, explain how it works and the legal framework on which it is based, and explore more specifically certain subjects on which it is currently very much in demand, in particular biometrics and artificial intelligence.

Explicability and Interpretability of AI models: concepts, approaches and adaptability to the domain of speech processing  

This presentation proposes to define the notions of explicability and interpretability in AI and to give an overview of the approaches used. The approaches are explored depending on their context in the processing toolchain and the type of information they consider.

Joint effort on data loading process and its application to VAD, speaker turn detection and overlap detection  

Diarization is the task of finding “Who spoke when?” in an audio stream. It relies on two subtasks defined as segmentation and clustering. The former most often includes voice activity detection (VAD), overlapped speech detection (OSD) and speaker change detection (SCD). This work introduces a data loading process (DataLoader) to format and distribute the speech segments used for the training of these three tasks.

The diarization team including students from ENSIM will present their work on Friday, 21th at 11:00. ► Read more

End2End Acoustic to Semantic Transduction  

We propose a novel end-to-end sequence-to-sequence spoken language understanding model using an attention mechanism. It reliably selects contextual acoustic features in order to hypothesize semantic contents.

SINr: fast computing of Sparse Interpretable Node Representations is not a sin!

While graph embedding aims at learning low-dimensional representations of nodes encompassing the graph topology, word embedding focus on learning word vectors that encode semantic properties of the vocabulary. The first finds applications on tasks such as link prediction and node classification while the latter is systematically considered in natural language processing. Most of the time, graph and word embeddings are considered on their own as distinct tasks. However, word co-occurrence matrices, widely used to extract word embeddings, can be seen as graphs.

MiniBERT: a simple and explainable BERT model  

As part of the PolysEmY project, we work with the SNCF (French railway company) to produce "polysemic-aware" word embeddings.

Multilingual neural architectures for natural language processing

Machine translation using little data leads to poor performance. The use of multilingual systems is one solution to this problem. Multilingual machine translation systems make it possible to translate several languages within the same system. They allow languages with little data to be learned alongside languages with more data, thus improving the performance of the translation system. This thesis focuses on multilingual machine translation approaches to improve performance for languages with limited data. I have worked on several multilingual translation approaches based on different transfer techniques between languages. The different approaches proposed, as well as additional analyses, have revealed the impact of the relevant criteria for transfer. They also show the importance, sometimes neglected, of the balance of languages within multilingual approaches.

Les learning analytics seraient-elles promises à l'échec ?

learning analytics (LA) are undergoing increasing development, and following the COVID-19 pandemic and the forced introduction of large-scale distance learning, it is reasonable to anticipate an acceleration in the introduction of hybrid practices and therefore a growing need for LA. However, due to an increase in their intrinsic complexity, particularly through the use of increasingly advanced statistical models and artificial intelligence (AI) techniques, they are struggling to be put into production in a real learning context. Improving the explainability of AIs for non expert users in data analysis is therefore a crucial issue for Technology Enhanced Learning.