How Singapore’s Polyglot Pilot Is Rethinking Government Website Translation

Singapore is testing whether artificial intelligence can make multilingual government information faster and easier to access. Polyglot, a translation widget developed jointly by the Ministry of Digital Development and Information (MDDI) and the Government Technology Agency of Singapore (GovTech), with Open Government Products (OGP) involved in its rollout, went live on 26 June 2026.

The pilot first covers MDDI’s website. It translates English content into Chinese, Malay and Tamil while retaining the English version, and is expected to run until October.

The larger idea is not simply to add an automatic translation button. Singapore is testing whether government agencies can provide information in all four official languages—English, Chinese, Malay and Tamil—in a way that is faster, more consistent and better suited to local usage.

What makes Polyglot different from basic machine translation?

Older automated translation systems can rely heavily on word-for-word substitution. That approach may produce grammatically plausible text while missing the meaning of a policy term, place name or culturally specific expression.

Polyglot’s translation engine uses large language models to interpret complete sentences and passages. It also combines them with neural machine translation technology, aiming to balance contextual understanding with speed and consistency.

The system draws on Government Terms Translated, Singapore’s official terminology repository. Maintained by MDDI’s Translation Department, the database contains names and terms relating to ministries, statutory boards, government agencies, organisations, initiatives, schemes and MRT/LRT stations in English, Chinese, Malay and Tamil.

That local reference layer helps Polyglot translate Singapore-specific government content rather than treating it as generic text. The development team has highlighted examples such as:

• Translating Singapore’s Chinatown as 牛车水, rather than the more general 唐人街.
• Rendering “government” in Malay as pemerintah, rather than kerajaan. In Singapore’s republican context, the former is considered more appropriate.

These examples show why public-sector translation involves more than grammar. Official language must also account for established place names, institutional terms, policy vocabulary and expressions used by local communities.

How can the public try the beta tool?

During the pilot, visitors to mddi.gov.sg see a banner directing them to beta.mddi.gov.sg. Once there, they can use the drop-down menu in the top-right corner to select English, Chinese, Malay or Tamil and view the relevant government content in that language.

Users can also report errors and suggest wording that sounds more natural or accurate. This feedback is central to the trial: the team needs to see how the tool performs across real users, policy subjects, languages and page layouts—not only in internal testing.

Why does AI translation still need human editing?

Polyglot is designed to improve access and efficiency, not to make human review unnecessary. MDDI officers can manually edit translations, allowing subject-matter and language specialists to correct mistakes involving tone, terminology, names, addresses and other details.

Pages also carry a notice that the content has been translated using AI. Large language models can still produce inaccurate results, so the warning makes clear that the translation is intended to help users understand the information and should not automatically be treated as an infallible authoritative version.

The resulting model is therefore a cycle of automated translation, human revision and public feedback. It also explains why Polyglot remains a pilot rather than a system already deployed across every government website.

From Hack for Public Good to a reusable government tool

The first Polyglot prototype was started by a five-person OGP team during Hack for Public Good in January 2026. The project was designed to be platform-agnostic: agencies can add the translation widget to a website with a front-end script, while an administrative portal manages translations in the background.

This structure could allow the tool to work across different government websites without requiring a complex content-management-system integration for each one. Whether it expands to more agencies will depend on the pilot’s performance, translation quality and public feedback. No timetable for a full rollout has been announced.

Past translation mistakes underline the stakes

A single incorrect character or spelling mistake can change the meaning of public information. Reports have cited several past examples. In 2017, a rostrum sign for the Speak Mandarin Campaign used a Chinese character associated with “disrespect” or “contempt” instead of the character for “read”. In 2020, Tamil lyrics for the National Day Parade song My Singapore contained faulty strokes and spellings that made parts of the lyrics difficult to understand; words meaning “friends” and “siblings” were also misspelt.

Such cases do not suggest that human translation is error-free. They do show why public communications require ongoing terminology management, language review and correction processes. Polyglot’s approach is to build those requirements into a reusable digital service rather than having each agency solve the same problem independently.

What could Polyglot mean for digital inclusion?

Government websites have traditionally provided most of their content in English, while producing, updating and reviewing three additional language versions can be slow and resource-intensive. If AI can quickly produce translations that use appropriate Singaporean terminology—and staff and users can then identify and correct problems—agencies may be able to publish Chinese, Malay and Tamil information more promptly.

That could make government information and digital services easier to navigate for residents who are more comfortable using a language other than English, including some older users. More dependable multilingual content would make language choice a standard part of access rather than an optional extra.

Polyglot’s long-term value will depend on the accuracy of its translations, the sustainability of human review and whether public feedback genuinely improves the system. The stated ambition is for accurate multilingual translation to become a standard feature across Singapore government websites and digital services. The current pilot is the test of whether that ambition can work in practice.