Description

SMarT is one of  Loria’s teams, it was created on December 2013

The acronym SMART comes in two forms:

Speech Modelisation and Text

Statistical Machine Translation

The objective of SMarT is to model the written or the spoken language (nature or altered) by using Machine Learning techniques.

Research Topics

The primary objective of the SMarT team is to develop customized language representation models for machine translation and speech recognition systems. This involves employing mathematical methods to identify, extract, and propose transfer between multiple languages, facilitating translation and speech recognition tasks. Various languages, such as Arabic, French, and English, are studied using monolingual, parallel, or comparable corpora, with a specific focus on low-resourced languages like Arabic dialects. Additionally, considerable attention is dedicated to under-resourced languages such as Russian, as well as those affected by specific language disorders like Broca’s aphasia.Another focal point for the SMarT team is the analysis of data from social networks, particularly emphasizing the detection and monitoring of misinformation propagation across diverse platforms.In addition to these pursuits, another area of research within SMarT pertains to our contributions to continuous speech recognition in individuals with pathological voice conditions. To achieve this, we develop models aimed at enhancing esophageal speech using voice conversion techniques. These conversion techniques serve as a means of translating one signal into another.

One of the most remarkable research endeavors within SMarT, which yielded spectacular results, started in 2015 prior to the advent of Generative Artificial Intelligence. This project involves the generation of Arab music improvisation through neural machine translation techniques.

Keywords

  • Machine Translation for under-resourced language
  • Automatic musical improvisation
  • Recognition of pathological voices
  • Code-switching
  • Language identification
  • Large Langue models
  • Fakes news

Keywords: Language Modelling, Machine Translation, unresourced languages, Estimation quality, Mining comparable corpora, Code-switching, Deep Learning,
voice conversion, enhancement of esophageal speech, pathological speech recognition.

Members

Scientific leader

Kamel Smaïli  (Professor, Université de Lorraine)

Administrative staff

  • Anne-Marie MESSAOUDI (CNRS)

University staff

Associated members

  • Chiraz Latiri – Professor (University of Tunis – Tunisia)
  • Karima Meftouh – Assistant Professor (University of  Badji Mokhtar, Annaba – Algeria)
  • Salima Harrat – Assistant Professor (ENS – Algiers, Algeria)

PhD students

  • 2019-: Fadi ghawanmeh (co-supervision with University of Oslo, Norway) Machine translation of Music
  • 2020-2024: Yohannes Biadgligne (University of Sudan) Amharic Machine Translation
  • 2022-2025: Abdellah Hamouda Sidhoum (Algeria) Arabic Complex Question-Answering
  • 2022-2025: Ouahab Hocini Detecting misinformation  on social networks
  • 2023-2026: Yacine Toughrai Monitoring false information across social networks.