French higher education attracted nearly 450,000 international students in 2024-2025, placing France 7th in the world. They represent almost 15% of the student population in French universities.
Student mobility is constantly increasing . In Europe, there have been more organized mobility programs in ten years than in the previous twenty-seven years, with demand in 2025 up 9% compared to 2024. This acceleration is correlated with the increase in bilateral agreements between countries and an expansion of Erasmus+ , the European support program for international mobility.
Globally, the most attractive countries are unsurprisingly English-speaking ones , particularly the United States, the United Kingdom, and Australia, as English is essential for international exchange. But for those who choose France and the French language, what kind of adventure are they embarking on?
The difficulties faced by foreign students
For these students, the language barrier complicates the assimilation of technical knowledge, hinders active participation in class and immersion in student life, despite French as a Foreign Language (FLE) instruction. The complexity of the French language exacerbates this challenge , even for Francophone international students with limited exposure to academic French.
The European Union is aware of this obstacle to mobility and encourages its Member States to provide academic and extra-academic support.
In 2024 and 2025, we conducted a survey of some of the 9,000 international students welcomed each year to the University of Montpellier. It revealed clear needs for support:
- during lectures and conferences, to understand the teachers orally and to have the transcript of the content in order to better understand their training;
- after classes, with a view to memorizing, reviewing and verifying the content of the classes.
Recent advances in artificial intelligence (AI) make it possible to meet some of these needs .
A comparative analysis of AI-based simultaneous translation solutions was conducted based on three main criteria: automatic speech recognition (enabling dynamic adaptation when the speaker changes languages or when a question is asked in another language), ease of use (minimizing installation and configuration requirements for students and teachers), and the quality of the resulting translation. This analysis revealed two types of solutions: commercial tools accessible via the cloud and open-source tools that can be installed locally on dedicated servers.
Commercial tools: effective, but expensive and risky in terms of intellectual property
Regarding commercial tools, our technology monitoring indicates that the most suitable tool for the identified needs is eventCat . This tool provides a turnkey service, requiring minimal IT and audiovisual resources for deployment. It was tested in 2024 and 2025 at a joint medical conference between two universities. The speakers had diverse language profiles: researchers with a high level of English, physicians accustomed to speaking French, and patient representatives with limited English proficiency. The audience consisted of French and Dutch students, the latter being fluent in English but not understanding French.
The tool allowed speakers to express themselves in their preferred language, while English-speaking students followed along via simultaneous translation on a dedicated link. Feedback from the speakers was very positive. Displaying the translations on a large screen placed next to the speakers, rather than on the students’ smartphones or laptops, reduced their cognitive load, as they already had to simultaneously follow the spoken presentation and read the transcript. The overall quality of the translation was satisfactory, although the acronyms used in the field and the speakers’ accents presented some challenges. This highlights the need for more nuanced contextual adaptation for highly specialized content.
The drawbacks of such a solution are an annual cost of tens of thousands of euros per training session and uncertainty regarding data. For example, at the time of writing, the GDPR policy of the identified tool is to store received audio streams and their transcripts for no more than seven days and not to use this data to train its AI models, which would expose this data to malicious prompt attacks .
However, the company operating this service is located in the United States and is therefore subject to the Cloud Act , obligating it to disclose the data it stores if ordered to do so by a US administrative authority. Concerns about data protection arose repeatedly in our discussions with both teachers and students during our investigation. It is important to remember that university professors and researchers hold the copyright to the courses they offer; therefore, it is their intellectual property that can be compromised by an automatic transcription and translation tool.
A prototype open-source tool for protecting data
It is therefore preferable for teaching staff to have alternative solutions hosted within their institution. Open-source software is the natural approach in this context; unfortunately, in 2025, no such solution yet existed to meet the needs expressed by international students in our analysis.
We therefore developed the foundations of such a tool and proposed an initial prototype . This prototype already performs local translation of text content using an open-source AI model that can run on a standard server. However, the transcription of audio streams still relies on AI services offered by various cloud providers (the most efficient models for this step are either more resource-intensive or proprietary), although these services are not tied to any particular vendor. This prototype leverages RAG (Retrieval Augmented Generation) technology to adapt to technical vocabulary and has been used in courses with 20 to 30 students. The students expressed strong interest in this type of tool, which they also use for socializing discussions.
However, they all note difficulties with technical courses taught in French that rely heavily on Anglicisms. These shortcomings still need to be addressed, but the next major step is to transcribe audio into text locally, without using a cloud service . As things stand, this requires dedicated GPUs , representing an investment of several thousand euros per site. This is the price to pay for data preservation.
Learning the language of the host country is a central objective of international mobility. The digital tools mentioned above facilitate this learning and the academic success of international students, but they do not replace French as a Foreign Language (FLE) instruction. AI tools promote the acquisition of subject-specific knowledge and, to some extent, the acculturation of international students. However, they are not sufficient for successful integration. In this respect, the support provided by the home country to prepare students for mobility remains paramount .
Author Bios: Vincent Berry is Professor at the University of Montpellier, Chouki Tibermacine is Professor of Software Science at the University of Southern Brittany (UBS), Chrysta Pelissier is a Senior Lecturer in Language Sciences and Educational Sciences at the University of Montpellier, Eric Anglaret is Professor of Materials Physics, Deputy Director for International Relations, Polytech Montpellier at University of Montpellier and Vanessa Vigano is an Instructional Designer also at the University of Montpellier
