Technology is reshaping how people connect — that much is obvious. The real question is this: as a new technology becomes widespread, who benefits, and who gets left behind?

Viewed through an Asian lens, the problem becomes concrete: can someone who only speaks Thai or Burmese use a smart assistant smoothly, or be correctly understood by search engines?

Asia is home to more than 60% of the world’s population and is one of the most active regions online — but activity does not equal equality. Languages such as Thai, Burmese and Marathi, each used daily by tens of millions, remain on the margins of AI training datasets. Over past decades the digitization of infrastructure and content was shaped largely around English: operating systems, programming languages, search algorithms and voice interfaces were primarily developed with English patterns in mind. As a result, the massive volumes of text needed to train large models are plentiful for standard English, but public corpora for many other languages are scarce.

Put simply: while English can draw on hundreds of millions of texts for model training, corpora for languages like Khmer or Lao are orders of magnitude smaller. With insufficient data, model performance is naturally constrained. Even for larger covered languages such as Hindi, performance drops when we move to regional varieties like Bhojpuri. That means tens of millions of dialect speakers may not receive services of comparable quality.

Compounding the issue, many Asian languages follow different usage patterns and conversational logic than English. For example, code-switching is common in India — speakers mix English and Hindi in ways commonly called “Hinglish.” That fluid mixing sounds natural to native speakers, but speech-to-text and other language tools often struggle to transcribe or interpret it. Input methods are another example: predictive text and candidate lists for Marathi or Sinhala are far less comprehensive than for English. These user-experience gaps directly reduce people’s willingness to go online in their mother tongue.

The consequences are cascading. The bulk of online content remains English, while only roughly 10% of Asia’s population can use English fluently. For the remaining 90%, limited access to information and services in their native language restricts participation in the digital economy. From a commercial perspective, a local business in Southeast Asia that builds its website in Vietnamese or Thai still faces challenges obtaining strong search visibility; search engines’ multilingual indexing is improving but uneven, so content outreach and discoverability vary. AI systems trained without robust local data can also misread local sentiment and tone, producing translations that feel stiff or inaccurate.

• Google announced its 1,000 Languages Initiative to build models that support the world’s most-spoken languages and extend language technology to under-represented communities.
• Meta’s “No Language Left Behind” project developed and open-sourced models that provide high-quality translation across 200 languages.
• AI4Bharat (based at IIT Madras) follows an open approach, creating datasets and tools for India’s official languages so local startups and researchers can build on them.

About Glodom