Nigerian Names and Places: The Proper-Noun Problem

Ask anyone who has used transcription in Nigeria what annoys them most and you will get the same answer. The names.

Not the grammar, not the general sense. The names — of people, towns, companies and streets — rendered as something phonetically nearby and completely wrong, with total confidence.

Why proper nouns are the worst case

Ordinary words come with context. If a system half-hears a common verb, the surrounding sentence constrains what it could have been, and it usually recovers.

Proper nouns have almost none of that. A name can be followed by anything. There is no grammatical pressure pushing the system toward the right answer, so it falls back on pure sound — and if that name was not well represented in what the system learned from, the nearest match will be an ordinary English word or a more familiar name.

"Chukwuemeka" becomes something shorter. "Ikorodu" becomes two words that exist. The output is fluent, which is the problem: an obvious error gets checked, a plausible one gets published.

Why it matters more than other mistakes

Names are what people search for. A transcript archive where every third name is wrong is not searchable, whatever else is right about it.

Names are also what gets quoted, filed, cited and attributed. A misspelled name in a published article is a correction. In a legal statement it is worse. In a research dataset it silently corrupts your counts, because the same person appears under three spellings.

And unlike a garbled clause, nobody notices a wrong name unless they know the right one.

The efficient fix

Do not proofread everything. Review names first, deliberately, before anything else.

Build a list before you record. For an interview, a meeting or a service, you usually know who will be there and what will come up. Having that list beside you while reviewing turns a hunt into a check.

Skim with the audio open. Names are quick to verify because you only need the one moment.

Search and replace the repeats. A name that appears forty times is usually wrong the same way forty times, which makes it one correction rather than forty.

Say names clearly at the top of the recording. A round of introductions gives you a clean reference for every name that follows, and it helps speaker labelling too.

Recording habits that reduce it

Closer microphones help disproportionately here, because proper nouns depend entirely on acoustic detail and have no context to fall back on. The same recording that produces acceptable general text at a distance will produce unusable names.

The underlying issue

This is a representation problem, not a mysterious one. Systems reproduce what they were exposed to, and Nigerian names have been under-represented in that exposure.

The related pieces are why transcription tools mangle Nigerian English and Yoruba, Igbo and Hausa-influenced English.

A free account includes transcription minutes each month with no card. Sign up and test it on a recording full of the names you actually work with.