Department 4 : Knowledge & Langage Management

Team leader : Kamel Smaïli
Tél. :  +33 3 83 59 20 83
Mail : smaili@loria.fr

Website

Presentation

The main objective of SM_a^rT is to develop models for natural language modeling for machine translation and speech recognition systems. This modelisation uses mathematical methods to identify, retrieve and propose associations between items which will be used in machine translation and speech recognition. The languages under study are done through monolingual, parallel or comparable corpora for under-resoursed languages as Arabic dialect, and for more conventional languages as modern standard Arabic, French and English.  The items retrieved could be associations between multilingual words or multilingual semantic concepts. Sentiment analysis are studied over comparable multilingual corpora and multilingual social networks.

Topic axis

• Machine translation without alignment and evolutionary decoding
• Speech translation of under-resoursed languages
• Estimation of the translation quality
• Multilingual sentiment analysis

Software

• SUBWEB : Alignment of bilingual subtitle movies.
• QUEST : Machine Translation quality estimation

Collaborations

• Sheffield university
• Tunis University.
• CRSTDLA : Research center for the development of Arabic language (Algiers – Algeria).
• LRI : Computer science Lab of Annaba university, Algeria

Keywords

Statistical language modeling, Machine Translation, Speech-To-Speech machine translation, Study of under-resourced languages, Modelisation, Mining comparable corpora, Quality estimation of machine translation.