Interactive learning material · M3-T3-MLC4

AI-Supported Translation and Multilingual Learning Support for Linguistic Inclusion

Multilingual learning and verified vocational translation

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre

AI mediates a conversation between two people who speak different languages — AI can mediate a conversation between people who do not share a language

Learning objective

In many schools, learners with different first languages learn together – young people with refugee backgrounds, learners from minority language communities, or adults who are retraining in a new country. Not speaking the language well doesn't prevent someone from being an excellent professional, but the language gap can still be a serious barrier to integration.

The learning content explores how artificial intelligence can help overcome language barriers. By the end of the learning, teachers will be able to use AI tools to support real-time translation, simplify explanations, facilitate multilingual learning, and develop vocabulary. They will also understand the limitations of these tools and critically review their outputs for potential mistranslations and inaccuracies in subject-specific content.

Introduction

Introduction: understanding first, language along the way

A student may know a great deal about engines, cooking, or electrical circuits, yet still struggle to follow the teacher's explanations if they do not fully understand the language of instruction. Since AI tools can be a great help in acquiring technical terminology, it is worth placing the emphasis on this in the early stages of teaching. A solid command of technical language will also help students express themselves with growing confidence and nuance in everyday language.

Linguistic inclusion

Make vocational knowledge accessible

Language difficulties can make it hard to follow explanations and demonstrate knowledge. Support with vocational vocabulary helps learners learn and integrate.

Tools for translation and multilingual support

Text and camera translation

Text translation

DeepL, Google Translate and general AI assistants translate worksheets, instructions and messages. AI assistants can also explain, rephrase and answer questions about the text in the learner’s language.

Camera translation

Pointing the camera at a sign, a label or a page of a manual shows the translation on the screen (e.g. Google Translate or Google Lens).

English labels on a device are shown in Lithuanian through a phone camera — Translation with a camera · Illustration, source: ChatGPT

Tools for translation and multilingual support

Speech translation and captions

Conversation and live speech translation

Conversation modes translate spoken dialogue in turns; newer services such as live translation in Google Translate (with headphones, over 70 languages, availability depending on country) or DeepL Voice translate speech almost in real time.

A phone and earphones support translation of an English–Lithuanian conversation — Translate live while speaking · Illustration, source: ChatGPT

Translated captions

Video platforms and meeting tools (e.g. Teams, Meet) can show captions translated into another language – in meeting tools this often requires a paid licence.

Supported languages, feature availability and required subscriptions may change. Before use, check Google Translate Help (opens in a new tab) and the DeepL Voice documentation (opens in a new tab).

Advantages and Disadvantages of AI in Translation

Flexible translation of vocational language

Adapting to the learner’s language level

A traditional translation tool can translate a text into another language, but it does not take into account who will be reading it, and you cannot “have a conversation” with it. One of the main advantages of language models, by contrast, is that they can simplify the wording and use shorter sentences on request, adapting the language level to the user’s needs.

Glossaries

If we want a language model to consistently follow our preferred technical terminology when translating specialist texts, we can provide it with a glossary, which it can then use consistently throughout the translation. Consistent use of the glossary still needs checking.

Advantages and Disadvantages of AI in Translation

A learning companion during translation

Did you understand?

One of the greatest advantages of language models is that they can respond to feedback about whether the learner has understood the explanation and adjust their style and language level accordingly. During the conversation, the learner can ask follow-up questions such as, “What does this term mean?” The model can then provide a more detailed explanation, illustrate it with examples, and/or ask questions to “check” whether the learner has really understood. Whether the interaction takes place in writing or through speech makes little difference: during translation, the language model can act as a kind of learning companion.

Fluency does not prove accuracy

The general limitations of AI tools naturally also apply to translation. Language models tend to generate confident, fluent text even when the content is incorrect. Traditional machine translation can often be recognized by awkward or unnatural sentences. However, errors made by a language model can be embedded in polished, fluid prose, creating an appearance of reliability.

Teachers should therefore take particular care when using translations as teaching material. Check important information, especially vocational content, with the teacher or a reliable source.

Strategies for multilingual learning support

Simplified explanations and language-level adaptation

Often the best support is not a full translation but an explanation in simpler language, in the language of instruction. AI can rewrite a text at a given level of the Common European Framework of Reference for languages (CEFR), e.g. A2 or B1, add a short summary in the learner’s first language, or produce a bilingual version side by side.

AI adapts a text to different languages and proficiency levels — The same content adapted to different language levels

Example prompt for language-level adaptation

Rewrite the attached text about brake inspection at CEFR level A2 in Hungarian. Keep all technical terms in Hungarian, and after each term give the Ukrainian equivalent in brackets. Add a 3-sentence summary in Ukrainian at the end. Mark any term where you are not sure of the translation.

Strategies for multilingual learning support

Vocabulary scaffolding

Vocational terminology is where learners most need support – and where translation tools most often fail. A bilingual glossary for each unit, with the term in the language of instruction, its translation, a simple explanation and a picture, helps learners understand now and learn the workplace vocabulary for the future. AI can draft the glossary; the teacher or a professional should check it. Some tools (e.g. DeepL glossaries) can be told to always translate a term in a fixed way.

A bilingual glossary pairs technical terms, illustrated with a gear and a globe — A verified bilingual glossary is the backbone of vocational language support

Strategies for multilingual learning support

Multilingual learning support in practice

Pre-teach vocabulary

Pre-teach the key terms of the next lesson in both languages.

Use learners’ languages

Let learners use their first language for thinking and discussing, and the language of instruction for the final answers (translanguaging).

Check understanding

Check understanding with short questions rather than just asking “Did you understand?”

Communication with parents

Use translation for communication with parents, while keeping important documents in verified versions.

Risks and how to manage them

Vocational risks in translation

Mistranslated vocational terms

Many technical words also have an everyday meaning. A “nut” may be understood as a fruit, a threaded fastener or a strange or funny person. Such errors make instructions meaningless or dangerous.

Safety-critical content

Safety instructions, hazard warnings and operating procedures must not rely on unverified machine translation. Use officially translated versions or have them checked by a competent speaker.

A technical term can lose its meaning in translation; a magnifying glass and warning symbol emphasise checking — A fluent translation can still be wrong – especially for technical terms and safety instructions

Risks and how to manage them

Culture, autonomy and privacy

Cultural inaccuracy

Forms of address, politeness, idioms, examples and images may not fit the learner’s culture; humour and irony are especially risky.

Overreliance

If every text is always translated, the learner has less reason to develop the language of instruction. Plan how support can be gradually reduced.

Privacy and assessment

Do not paste personal data into translation tools, and agree clear rules on whether translation tools may be used in tests.

Risks and how to manage them

Quality check for translated materials

Step 1

Keep the original

Keep the original text next to the translation.

Step 2

Back-translate

Back-translate important details into the original language and compare.

Step 3

Mark uncertainty

Use a verified glossary and ask the AI to mark uncertain terms.

Step 4

Ask for human review

Ask a colleague, a learner or a parent who speaks the language to review key materials.

Takeaways

Takeaways for teachers

AI tools make language teachers’ work easier and can also be a great help to students and adults who are learning in a language other than their mother tongue. Key lessons and tips:

AI-powered translation provides instant access to content – combined with the gradual development of the language of instruction.

An explanation simplified to the right language level is often more useful than a full translation of the text.

It is worth creating a glossary of bilingual technical terms, as AI makes mistakes most often when translating terminology.

Never rely on unchecked machine translation if a text contains safety-critical information.

Respect cultural differences, support learner autonomy, and plan for the gradual reduction of support.

Quiz

Check your understanding!

Choose one answer to each question, then check your result. You will receive a short explanation for every answer.

1What is the purpose of language support? To permanently replace learning the language of instruction with translation. To provide access to vocational content while developing the language of instruction. To work only with learners who already speak fluently.

2When can a simplified explanation be more useful than a full translation? When all vocational concepts can be left out. When there is no need to check the learner’s understanding. When shorter sentences adapted to the learner’s language level help understanding.

3How can we reduce mistranslation of technical terms such as “nut”? By using a verified bilingual vocational glossary and providing context. By automatically choosing the longest translation. By checking only how fluently the sentences read.

4How should we translate a safety instruction? We can distribute fluent machine-generated text immediately. Confident AI wording is sufficient verification. Use an official translation or have it checked by a competent speaker.

5What is the purpose of back-translation in quality checking? To prove that every detail of the original translation is correct. To compare the resulting meaning with the original and identify differences. To make keeping the original text unnecessary.

References and related materials

References and further reading

References

Council of Europe (2020): Common European Framework of Reference for Languages – Companion Volume.
Google (2025–2026): Live translate with headphones in Google Translate. Google Blog and Translate Help.
DeepL: Glossary feature documentation.

Links

Image credits

Uncredited images were taken using Claude's Mac app.

The illustrations of camera translation, live speech translation and translated captions were created with ChatGPT.

Contributors and attachments

Credits and copyright

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre
Project: 2025-1-LT01-KA220-VET-000362456

Attachments

Download the English source materialM3-T3-MLC4-TEXT-EN.pdf Download the practical exercisesM3-T3-MLC4-EXC-EN.pdf

This work may be reused and adapted under the terms of the Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0).

Funded by the European Union. Views and opinions expressed are those of the author(s) only and do not necessarily reflect those of the European Union or the National Agency.