Second Language learning assisted by Artificial Intelligence – ALAIA

The ALAIA LabCom, between Archean Technologies and the Toulouse Institute of Computer Science Research (IRIT) is intended to lead an ambitious research and innovation program in the field of computer technologies for foreign language learning. It will explore advanced Artificial Intelligence (AI) techniques to help learners improve their oral skills (expression and comprehension). According to learners’ profiles, these techniques will decide which contents (exercises/corrections) are the most adapted and the most effective for the acquisition of the target language. Archean Technologies will be able, through this LabCom, to develop breakthrough technologies compared with current computer-assisted learning software and commercialize these solutions on a B2C mode. IRIT, thanks to its high level of expertise in AI and automatic speech analysis, will be able to use the data provided by the company (recordings, software usage data) in compliance with the European regulations on protection of personal data. Thus, new lines of research will emerge from these field issues. The LabCom activity fits perfectly into the innovation component of the French government’s language and multilingualism plan on March 20th, 2018, responding to the language training emergencies of migrant populations in France, for their integration and access to employment. Initially centered on French, the proposed models and technologies will be generalized to other target languages. From demonstrators to real industrial solutions, the developed technologies will follow regular milestones of development and scientific production. In the short, medium and long term, the LabCom innovation program will integrate the multiple dimensions of language learning:
• Linguistic levels: from pronunciation to lexico-grammatical, discursive and pragmatic levels;
• Authenticity: from highly constrained exercises (reading/repetition) to spontaneous production;
• Targeted precision: from a global training to more targeted exercises related to the types of errors detected and their location;
• Source/target language systems: initiated for the Japanese/French couple of languages and then generalized to other language pairs;
• Variety of learners and variety of their skills: interferences between mother tongue and target language, pronunciation difficulties specific to individuals;
• Personalization of remediation exercises, thanks to the discriminating potential of AI algorithms and notion of learner’s profile;
• Audio-visual modalities: complementary resources to help illiterate people to learn and understand some important socio-cultural aspects.

The proposed milestones will help to gradually reach strong technological and scientific challenges:
1. Automatic detection and localization of errors at different language levels;
2. Characterization of errors according to acoustic, linguistic and didactic criteria and choice of the most appropriate remediation strategy;
3. Production of feedback to enable each learner to estimate his/her performances, to better understand his/her errors to better correct them;
4. Use of perceptual saliency to help learners to strengthen their work on oral productions;
5. Automatic profiling to customize learning exercises;
6. Spontaneous production quality in interaction situations to measure the ability to communicate with a native speaker.
The first 4 issues will be addressed during the first three years. Beyond this, automatic profiling and evaluation of oral production in an interactive context will open new and innovative perspectives.