How to Choose a Translation Agency in Singapore: 12 Criteria

Why this is harder in Singapore than elsewhere

First, no accreditation exists. Singapore does not appoint sworn translators and no government body certifies translation companies. There is no register to check, so any provider can describe itself as certified without that meaning anything specific.

Second, you cannot evaluate the product. If you could read the target language well enough to assess the translation, you would not be buying it. That makes the process the only thing you can actually assess before committing. Both facts point the same way: buy the process, not the promise.

The twelve criteria

1. Independent revision by a second linguist. The single strongest predictor of quality. Ask: Is every translation revised by a second qualified linguist, and is that included in the rate?

2. Subject-matter specialisation. Fluency is not competence. Ask: What are the qualifications of the specific linguist who will handle my content in this field?

3. Disclosure on machine translation. Post-editing is legitimate when disclosed and priced as such. Ask, in writing: Is machine translation used on my content? At what post-editing level? Which engine, and does the content leave your controlled environment?

4. Ownership of translation memory and terminology. Your TM is a compounding asset built from your own content. Ask: Do I own the translation memory and termbase, and will you export them in TMX and TBX on request, at no charge?

5. A real, contactable project manager. Ask: Who is my project manager, in which office, and what is the escalation path?

6. Coverage that matches your footprint. Depth in the pairs you use matters more than breadth. Ask: How many linguists do you have in my specific pairs and field, and how long have they worked with you?

7. Handling of your file formats. Ask: Can you work directly in my source format, and do you have connectors to my CMS or repository?

8. Confidentiality in practice. Ask: Are the individual linguists under NDA? How is data transferred and stored? Is my content ever sent to a public MT engine?

9. Capacity and realistic timelines. Sustained throughput is around 2,000–2,500 words per linguist per day including revision. Ask: How many linguists will work on this, and how will you keep terminology consistent if the file is split?

10. A defined error and complaint process. Ask: What is your process if I dispute a translation? Who arbitrates, against what error typology, and at whose cost?

11. Local presence and time zone. Ask: Is there a registered Singapore entity, and where is my project manager physically located?

12. Verifiable references in your sector. Ask: Can I speak to a client in my sector with a comparable volume and language set?

Red flags

• A per-word rate dramatically below market for the pair, with no explanation of the process difference.
• Evasiveness about whether machine translation is used.
• No named project manager.
• Refusal to release the translation memory, or a charge for doing so.
• “ISO certified” without naming which standard, the certificate number or the certification body.
• A test translation done to a standard the production work will never match.
• Agreeing to any deadline without asking about volume, format or subject matter.
• Client logos with no accompanying description of the work.

Which model fits: agency, freelancer, in-house or machine?

Most organisations end up with a combination: an agency for regulated and multilingual work, a trusted freelancer for one high-frequency pair, and post-editing for internal volume. Buying one model for everything is what produces either overspend or incidents.

What matters less than buyers think

• The total number of languages offered. Almost every agency subcontracts beyond its core pairs. What matters is depth in yours.
• Office size and staff count. Translation is done by a distributed network of in-country linguists.
• Years in business, on its own. Longevity signals stability, not quality in your field.
• Sophisticated technology, on its own. Every serious provider uses CAT tools; technology does not substitute for a qualified linguist.

A workable evaluation process

1. Shortlist three providers. More produces noise, fewer produces no comparison.
2. Send the same real document to all three, with the same brief and deadline. Never use an artificial test text.
3. Require the same answers in writing — criteria 1 to 4, 8 and 9.
4. Have the output assessed by an in-country colleague or independent reviewer, against a defined error typology.
5. Compare quotes for identical scope, checking source vs target word count, whether revision is included, and the minimum charge.
6. Start with a real project of limited scope, and evaluate what happens when something changes mid-project. That is the true test.