Featured image source: Yasmine Boudiaf & LOTI / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
STEWART LAWSON-HASKETT, THE BRITISH INTERNATIONAL SCHOOL OF NEW YORK, USA
The global context
British international schools are seeing a significant rise in the number of students for whom English is an additional language (EAL) (Strand et al., 2015). Teachers must balance curriculum demands with the need to ensure that these learners are not left behind, creating workload that can feel overwhelming (Barnes et al., 2019). At the same time, students face increased cognitive load as they decode new information in a language that they may not yet fully understand (Roussel et al., 2022). One promising avenue is the careful and ethical use of AI to support EAL students without compromising learning quality.
Traditional EAL support in international schools
In most international school contexts, support for students with EAL includes a mix of pull-out sessions with EAL specialists, in-class scaffolding by subject teachers, bilingual dictionaries, translated worksheets and language buddy systems.
Pull-out sessions allow targeted instruction on grammar and literacy development, while tools such as dictionaries or translated materials provide access to core subject content. However, these approaches often come with trade-offs. Leaving lessons for EAL support can limit students’ exposure to classroom discussions. They often also require significant staff input, increasing workload in contexts where teacher time is already under pressure.
I began to explore how emerging AI tools could complement these traditional approaches. The goal was not to replace EAL specialists or existing interventions, but to build an additional layer of support without increasing workload – one that worked in real time, inside the classroom and in a format accessible to students with very limited English proficiency.
Why one solution is not enough
EAL students have complex and varied needs, depending on their age, background and prior educational experience. Some may need translation support to access basic vocabulary, while others require opportunities for deeper engagement with subject content. The science of learning suggests that students benefit from accessing material through multiple pathways. Dual coding theory (Clark and Paivio, 1991), for instance, suggests that when information is presented both visually and verbally, retention and understanding improve. Similarly, the Universal Design for Learning framework (Rose, 2000) advocates for offering multiple means of representation and expression to meet learners where they are.
Establishing a multimodal support framework
To move beyond experimentation and provide consistency across classrooms, I began to frame these tools within a more formal structure: a multimodal AI support model. The framework is made up of three pillars, with each pillar addressing a major pain point that EAL learners encounter.
The three pillars of the model are:
- student–teacher access: tools that enable students to access the teacher in real time
- student–peer access: systems that reduce language barriers during collaborative activities and reduce isolation
- student–content access: resources that students can use before, after or during class to consolidate understanding.
Here, I outline three method groupings with examples of how they have been tested in a classroom and how they could be implemented in different settings.
Method grouping 1: Dual language AI chatbots to support understanding of subject-specific content
This method addresses point three: student–content access.
I implemented a series of AI-powered chatbots designed to interact with students in a Socratic style. The system that our school uses for chatbots is Mindjoy, but other systems, such as Google Gems, are available. These chatbots operate in two modes. In the default setting, they respond in English only. However, when a student inputs a preferred language, such as Japanese or Spanish, the bot switches to dual-language mode. It replies in English, followed by a translation in the student’s native language. This approach supports language acquisition while still focusing on curriculum content.
Students use the chatbots to ask questions about the lessons, clarify terminology or explore scientific concepts in more detail. For example, a student might ask for help with waves. The chatbot responds in both English and the student’s first language, reinforcing vocabulary and comprehension.
Method grouping 2: AI translators to facilitate peer work and comprehension
This method addresses points one and two: student–teacher access and student–peer access.
Group work can be difficult for EAL students when they do not yet have enough English to participate fully. To mitigate this, I introduced voice-to-text translation tools, such as Google Translate, to assist students in real-time conversation. Students are paired thoughtfully so that one fluent English speaker is partnered with an EAL learner. When the EAL student speaks into the translator, their contribution is instantly rendered in English on a shared tablet or laptop, allowing their partner to understand and respond. The native English speaker will speak into a second device that is set to be English-to-EAL language. Using the two devices together allows for a flowing conversation between the two students and can be expanded to larger groups as needed.
A second live translation service that I use is the built-in PowerPoint subtitle feature during my teaching, to translate my spoken words. This live subtitling offers immediate support, especially when introducing complex or technical vocabulary (Frumuselu, 2018). This feature has proven to be an excellent scaffold during lessons, increasing engagement and reducing confusion while allowing students to better understand when other students ask or offer answers to questions.
The next AI translation system that I implement is the use of document translators; systems such as Google Translate will allow you to upload PDFs, Word documents and PowerPoints, translate them into different languages and download the copy. This has proven very useful for assessment practice when the use of technology is discouraged. Students can be presented with both the English copy and their native language copy to assist in translation.
The final translation system that I use is full web translators. Many web browsers now come with the ability to translate any website using AI. To support all my students, I use a Google Sites system as my learning hub. Students who prefer to can use the translate system to translate the full site into their preferred language.
Method grouping 3: AI-generated learning resources
This method addresses point three: student–content access.
Another tool in the multimodal approach is generative podcasts and videos. This uses generative AI to convert my in-class resources into podcasts or videos that the students can watch or listen to. For my case, I used NotebookLM, which is integrated into our Google Workspace. After uploading my lesson PowerPoint, I was able to generate both a short podcast, usually four to five minutes, and a video explanation of the content. Both can be created in multiple different languages. I shared these with the students on Google Classroom with the proviso that the podcast was a pre-learning tool to help to prepare for the lesson, while the video was for post-lesson consolidation of the content. These are then translated into multiple languages and shared with students via Google Classroom.
These tools are particularly valuable for students who may need additional time or repetition to grasp certain concepts. The video and audio content is often used by students independently or shared with EAL teachers, who integrate the materials into their pull-out sessions. By offering the material in various formats and languages, we reduce cognitive load and allow students to process information at their own pace.
Outcomes from a combined approach
By combining the three modes of AI support, I was able to provide more effective support before, during and after lessons, rather than relying on a single approach. Each method supports access in a different way, and while no single strategy can address all needs, together they offer a more flexible and responsive system.
Student feedback has been positive. Learners have reported feeling more confident and supported in class. Several students have requested similar chatbot support in other subjects, and some have used my chatbot to discuss English and humanities lessons. Their increased participation in science lessons, both written and verbal, suggests that the tools are making a tangible difference without substantially increasing my workload.
Although the use of these systems is visible to students, no negative responses or behaviours were reported. As this is an international school where multilingualism is common, this context may have influenced student perceptions. In other settings, careful consideration would be needed to ensure that implementation supports inclusion without inadvertently singling out individual learners.
Data protection and safeguarding
All platforms used in this project are compliant with the General Data Protection Regulation and our school’s safeguarding policy. NotebookLM and Google Translate are used as part of our school’s Google Workspace of resources, with our secure login. NotebookLM operates through our Google Workspace and is only accessible by teachers. The only resources that go through NotebookLM are my PowerPoints, which contain no student data, leaving all private information secure.
For the listening translation services such as Google Translate and PowerPoint, our ICT department has been consulted and has agreed that all their privacy policies match with our own and fall within our requirements. All students are aware when the system is being used.
For chatbot development, we used the Mindjoy platform, which adheres to strict data privacy policies. Mindjoy also provides an added layer of safeguarding, with all chatbot interactions being logged and monitored. If a conversation contains language that might be flagged as concerning, the teacher receives a notification and email with a direct link to the section of the chat in question. The system has picked up concerning information in other languages as well as English. For instance, one student engaged with a chatbot on an English poetry task that referenced the theme of death. They were writing to the chatbot in Japanese and the system automatically flagged this for review, allowing for timely and appropriate pastoral support if needed. This monitoring system ensures that students are protected while still engaging in meaningful inquiry.
Ideas for the future
As I look forward to the new academic year, I am considering how to further support my EAL students. One area of interest has been the use of headphones. In one ear, the student would have a live translation of what I am saying using the Google Translate feature. This would allow them to follow along with classroom discussions more easily, without becoming isolated from the rest of the group. I will begin a small trial of the system at the start of the 2026–27 academic year. I am interested to see the outcomes from this addition, although I do see two potential barriers. First, the cost of purchasing suitable headphones that students will not lose or damage, though minor, is still a factor. Second, the students will need training on how to operate this, and the increased cognitive load could be problematic at first. Close training and monitoring will be needed to ensure that the students are benefiting from the practice.
Understanding the limitations
Despite the successes, it is important to acknowledge the limitations of using AI in our learning environments. For the sake of brevity, I will not discuss the larger ethical or environmental concerns relating to AI but will focus on practical implications for EAL usage.
The primary issue in my context is that I do not speak any of the translated languages; this makes accuracy checks extremely difficult. To mitigate this, I have tested the systems with English. Additionally, a colleague who speaks Japanese reviewed one video, one audio file and one chatbot translation for accuracy. While this is not foolproof, it helps to maintain a reasonable standard.
A secondary concern is that students could become dependent on translations rather than developing the second language. It is important to remember, however, that without these systems, the student would not be able to access the science curriculum at the same level.
Conclusion
Teaching EAL students in international settings is complex and requires thoughtful planning. When carefully selected and ethically implemented, AI tools can dramatically enhance the learning experience. The multimodal approach described here offers a flexible, evolving set of tools that can reduce barriers. By combining chatbots, live translation and AI-generated media, teachers can create more inclusive classrooms where all students can succeed.
As these technologies advance, the possibilities for supporting EAL students will grow. The key is to remain focused on pedagogy, maintain high standards of privacy and safeguarding, and never lose sight of the students at the heart of the work.
The examples of AI use and specific tools in this article are for context only. They do not imply endorsement or recommendation of any particular tool or approach by the Department for Education or the Chartered College of Teaching and any views stated are those of the individual. Any use of AI also needs to be carefully planned, and what is appropriate in one setting may not be elsewhere. You should always follow the DfE’s Generative AI In Education policy position and product safety standards in addition to aligning any AI use with the DfE’s latest Keeping Children Safe in Education guidance. You can also find teacher and leader toolkits on gov.uk .
