What code-switching actually means
Code-switching is the practice of alternating between two or more languages within a single conversation — or even within a single sentence. Linguists have studied it extensively. Nigerian speakers do it instinctively, without thinking about it, in exactly the same way a multilingual person anywhere in the world defaults to whichever language best expresses a particular idea at that moment.
In Nigeria, code-switching is not a mark of confusion or lack of fluency. It is a communicative strategy that leverages the expressive range of multiple languages simultaneously. A Yoruba speaker might open a conversation formally in English, shift to Yoruba for a culturally specific request, and return to English for a precise technical term — all in the same sentence.
This is normal. It is how millions of Nigerian phone calls actually sound.
Real examples from Nigerian calls
Here are the kinds of sentences that arrive on Nigerian business phone lines every day:
(Good morning, I want to book an appointment for my child.)
Hausa-English: "Lafiya lau. Please, how much is the delivery to Kano?"
(Fine, thank you. Please, how much is the delivery to Kano?)
Igbo-English: "Nne, I need to know — ọ dị — is the drug available?"
(Sister, I need to know — does it exist — is the drug available?)
Yoruba-English mid-clause: "The thing is, mo ti pay already o — so why dem never deliver?"
(The thing is, I have already paid — so why have they not delivered?)
None of these sentences are unusual. They are the everyday speech of educated, multilingual Nigerians who switch between languages as naturally as breathing. Any AI receptionist deployed on a Nigerian business number will encounter sentences like these within the first hour of operation.
Why this breaks most AI systems
Standard AI speech recognition and natural language processing systems are built with a single-language assumption. The model is initialised with a language — English, for example — and everything it hears is processed through that lens. When the caller switches to Yoruba mid-sentence, two things happen:
First, the phonological model for English tries to interpret Yoruba phonemes. It fails, because Yoruba has sounds — the tonal vowels, the ẹ, the ọ — that do not exist in English phonology. The model does its best, producing garbled output or simply dropping the segment.
Second, the language model layer receives this garbled output and tries to construct meaning from it. With a key phrase missing or corrupted, the intent classification — what is this caller asking for? — can fail entirely. The system might respond to the wrong question, or respond with a generic "I didn't understand that" that frustrates the caller.
The result is a caller who feels unheard and hangs up. They do not call back. They go to a competitor whose phone is answered by a human — or they post on WhatsApp about how your system is terrible.
The "separate language menus" workaround — and why it fails
Some businesses try to solve this by offering a language selection menu at the start of the call: "Press 1 for English. Press 2 for Yoruba." This addresses the language problem in the crudest possible way, and it introduces new problems.
The most immediate one: callers who code-switch do not have a single language to select. A caller planning to speak mostly English but open with a Yoruba greeting is not an edge case — they are a significant portion of your callers. When they select "English," the system processes their Yoruba greeting as noise and misses the context. When they select "Yoruba," it might struggle with their English-heavy sentences.
The second problem is friction. Every second of menu navigation is a second during which an impatient caller is deciding whether to hang up. Research on Nigerian caller behaviour consistently finds that menus — particularly multi-level IVR trees — are one of the primary reasons callers abandon calls before connecting.
Maraba has no language menu. There is no "press 1." The caller speaks however they naturally speak, and Maraba handles it.
How Maraba handles code-switching technically
Maraba's code-switching support is built at the model level, not applied as a post-processing step. The language detection runs at the utterance level — analysing each speech segment independently — rather than locking in a single language for the entire call at the start.
When the caller says "Ẹ káàárọ̀, I want to book appointment for my pikin," the system detects that the first segment is Yoruba, applies the Yoruba phonological model, transcribes "Ẹ káàárọ̀" with diacritics intact, then shifts to the Nigerian English model for the remainder. The full transcript arrives at the intent classification layer with both segments correctly rendered.
The intent classifier understands the combined meaning — a booking request, for a child, from a Yoruba-English speaker — and responds in kind. Maraba's reply acknowledges the greeting appropriately and addresses the booking request in English, or a natural mix, depending on what the caller has signalled they prefer.
Why this is not optional in Nigeria
Consider the demographics. Nigeria has an estimated 250 million people and over 500 languages. The three major languages — Yoruba, Hausa, Igbo — are each spoken by tens of millions as a first language, with English as the lingua franca layered on top for education, commerce, and formal contexts.
Most urban Nigerians are functionally trilingual: their mother tongue, English, and at minimum a working knowledge of one or two other major languages. In Lagos in particular, you will find Igbo traders who open in Yoruba, Hausa businesspeople who close in English, and everything in between. An AI receptionist that cannot navigate this reality is not a useful tool for a Nigerian business. It is a source of embarrassment.
For a clinic in Surulere, a pharmacy in Enugu, a logistics company operating across Kano and Lagos, or a law firm with clients from across the country — code-switching support is not a premium feature. It is the baseline requirement for a system that will actually work.
What this means for your business
If you are evaluating AI phone answering for your Nigerian business, the code-switching question is the most important one to test. Ask the vendor to demonstrate what happens when a caller opens in Yoruba and switches to English mid-sentence. Ask them to show you the transcript. If the Yoruba portion is garbled, missing, or stripped of diacritics — you know what the system will do with your real callers.
Maraba's Starter plan, at ₦20,000 per month, includes full code-switching support across Yoruba-English, Hausa-English, and Igbo-English. You can test it on the free plan in English first — 50 calls, limited beta spots — then upgrade when you are ready to enable the full language suite.
Request beta in English, limited beta spots. Upgrade to Starter to unlock Yoruba, Hausa, Igbo, and full code-switching support.
Request beta →