Maltese language risks falling behind in the age of AI

Artificial Intelligence has become deeply embedded in business and daily life, with Large Language Models (LLMs) rapidly reshaping communication, customer service, automation, and content creation.

Yet while these tools demonstrate extraordinary capabilities for high-resource languages like English, Maltese remains at a competitive disadvantage in the digital sphere.

Speaking at Ctrl+Esc Malta 2025, University of Malta Associate Professor Claudia Borg and AI expert Mohamed Berrimi highlighted the critical need to accelerate investment and innovation to ensure that Maltese retains its relevance in the era of AI.

A structural inequality in language technology

The experts emphasised that AI models depend on massive volumes of high-quality data. English benefits from millions of pages of native text. Maltese provides only a few thousand. This discrepancy makes it impossible for international models to perform equally across languages.

As discussed during the event, low-resource languages like Maltese face three persistent constraints:

• Too little digital presence relative to the speaking population
• Limited availability of high-quality, native-authored datasets
• Over-dependence on low-quality translations

Globally, 43 per cent of languages are endangered. Yet only 1.5 per cent are meaningfully supported in modern AI systems. More than 80 per cent of non-English training data is derived from imperfect translations rather than culturally accurate sources. This affects both linguistic accuracy and social context in model outputs.

Even multilingual benchmarks often unfairly disadvantage low-resource languages, relying on English-centric evaluation methodologies that do not capture grammatical nuance or cultural relevance.

The bottom line remains stark: LLMs work best if you speak English.

This is true for search engines too. It also comes natural to almost all of us to write in English when searching something up on Google. For example, it’s an instinct that if I am searching for a hairdresser near me, I would write ‘hairdressers near me’ and not ‘parrukkiera viċin tiegħi’. Moreover, if I write the latter, Google does not provide any useful results.

Malta’s strategic response: Building its own AI foundation

Despite the disadvantages, Maltese NLP (natural language processing) research has made significant progress, driven largely by Prof. Borg and her team.

Key achievements include:

▪ Korpus Malti v4.0

A high-quality national language corpus containing approximately 466 million tokens, sourced from 19 domains. Maltese Wikipedia accounts for less than one per cent of it, demonstrating the need to capture diverse and representative text.

▪ BERTu and mBERTu

Maltese-trained language models that outperform generic multilingual ones when used for:

• Named-entity recognition
• Sentiment analysis
• Part-of-speech tagging
• Dependency parsing

These results illustrate that language-specific models are essential for strong performance.

▪ Speech recognition and translation

Researchers are constructing a 100-hour spoken Maltese corpus. This work will underpin speech-to-text and real-time translation systems, enabling future voice assistants and call-centre automation in Maltese.

▪ Machine translation

Work is ongoing to refine translation quality and address dataset inconsistencies. Government-backed platforms such as traduzzjoni.mt now benefit from more rigorous evaluation, including new native-language testing standards.

▪ Chatbots and financial services adoption

A dedicated Maltese chatbot system is being developed for use in banking and insurance environments, achieving 96 per cent accuracy in understanding common financial queries. This underscores the commercial potential once appropriate data pipelines are established.

▪ Bias detection and mitigation

Given the risk of societal prejudice being encoded into technology, the University of Malta is also actively studying methods to detect and reduce bias in AI systems for Maltese.

A local opportunity

Multinational technology companies are releasing more than 20 new LLMs per month, pulling AI talent into global markets and raising the urgency for smaller countries to protect their linguistic identity.

If Maltese is to succeed in the digital age, businesses will need to support ongoing research and data development. Improved datasets can unlock:

• AI-enabled customer service automation
• Smarter digital government services
• Local fintech and regtech innovation
• Preserved cultural heritage in online media
• More equitable access to emerging technologies

A call to action for the private sector

Investing in Maltese AI development is no longer a purely academic exercise. It is central to future competitiveness.

Businesses that incorporate Maltese-capable AI will gain:

• Better customer engagement
• Enhanced accessibility and compliance
• Reduced reliance on English-only automation
• Stronger marketing.

Cutting corners by relying solely on English systems risks creating a two-tier digital economy in which Maltese speakers are underserved.