Hiring in Bangla and English — what changes when a candidate answers in their own language

A third of the real job postings in our system are configured for Bangla, not English. That choice changes what an interview measures — and if you are not careful, it changes it in a direction you did not intend. Here is the specific failure mode, what we test on every deploy to catch it, and the part we genuinely cannot tell you yet.

What changes when someone answers in their first language

Ask a candidate to describe a problem they solved. In their second language, the answer is shaped by what they can express, not by what they know. Someone who debugged a timezone bug across three systems and can only say "there was an error with dates, I fixed it" has given you a worse answer and identical evidence of competence.

In their first language, the same person gives you the timezone, the wrong rows, the test they added afterwards. You learn more, because you removed a filter that was never part of the job.

Unless the job actually is the language. If the role is writing English client reports, English fluency is a requirement and testing it is correct. The mistake is testing it when it is not — which is most roles, most of the time.

The failure mode this creates for scoring

Any system that scores written answers can be fooled in the same direction a tired human can: fluent, confident, well-structured text that contains nothing. In one language that is a known risk. Across two languages it gets worse, because a rough answer in Bangla and a polished answer in English are now competing, and the polished one looks better for reasons that have nothing to do with the work.

If you take one thing from this article: the question to ask about any bilingual assessment is not "does it support Bangla" but "does substance still beat polish in Bangla". Those are very different claims and only one of them is worth anything.

What we test, on every deploy

We run a fairness gate before every release, and it includes a Bangla written-answer test. The same interview question, answered three ways, all in Bangla:

The release fails if A does not beat B, or if C scores above 15 out of 100. On the most recent run, A beat B by 72 points and C scored 0. The gate has run 56 times.

That is a narrow claim, deliberately. It does not say our Bangla scoring is as good as a fluent human reviewer. It says the specific failure that matters most — rewarding fluency over substance — is checked in Bangla, in the same way and to the same standard as in English, and that a release which breaks it does not ship.

What we cannot tell you

The honest gap, and it is a real one: we do not log what language candidates actually answer in. Our interview session records carry no language field. So we can tell you that 9 of 29 real postings are configured for Bangla and 7 for English, and we can tell you what our test asserts — but we cannot tell you:

• whether candidates offered Bangla actually use it;
• whether answering in Bangla changes anyone's score in practice;
• whether the option changes who applies at all.

Those are the three questions you would most want answered, and we would need to log the answer language and compare outcomes to say anything real about them. It is worth doing and we intend to. Until then, this article is about design and testing, not about results, and anybody telling you otherwise about their own product should be asked for the sample size.

Practical rules if you hire across two languages

1. Decide whether the language is part of the job, in writing, before you post. If English matters for this role, say which parts — "writes weekly client updates in English" is testable. "Good English" is not, and it will be applied inconsistently.
2. Let the candidate choose, and say so in the posting. The offer itself changes who applies.
3. Test the language separately from the work. One short written task in English, if English is genuinely required — not an entire interview conducted as an implicit language exam.
4. Watch for the polish trap in your own panel. Interviewers do this too, and unlike a scoring system, nobody runs a test on them before each round.
5. Ask your vendor the narrow question. Not "do you support Bangla" — ask what they test, how often, and what happens to a release that fails.

Where this fits with everything else

Language is one more thing a CV or an answer does not tell you reliably, and the honest response is the same as everywhere else: know exactly what the score can and cannot read, weight the things that are actually the job, and keep a person in the decision.

If you are setting up a role that will run in Bangla, the weighting question comes first: decide what this role is actually made of, and only then decide which language proves it. And whichever language the process runs in, the people you do not hire should hear from you in the language they applied in.