Introduction
“Nawa o.”
“You don try.”
“Abeg, no dey whine me.”
For many Nigerians, these expressions communicate much more than their literal English translations suggest. Depending on the situation, they can express surprise, appreciation, frustration, disbelief, humour, or friendly teasing.
But what happens when you ask an artificial intelligence assistant to understand these expressions?
AI tools such as ChatGPT and Google Gemini can understand many languages and informal expressions. However, their ability to interpret Nigerian Pidgin may vary depending on the phrase, its context, regional usage, and the model being tested.
This raises an interesting question: which AI assistant understands Nigerian Pidgin most accurately?
Instead of assuming that one tool is better than another, we can develop a simple experiment using 30 Nigerian Pidgin prompts. The test examines meaning, slang interpretation, humour, emotional context, and the ability to respond naturally.
This article provides the testing method, example prompts, and a scoring system. The results should be completed using actual AI responses rather than invented rankings.
Why Nigerian Pidgin Is a Difficult Test for AI
Nigerian Pidgin is widely used in everyday communication across Nigeria. People use it in conversations with friends, social media posts, entertainment, music, business interactions, and online discussions.
Although it incorporates English vocabulary, its grammar, expressions, and meanings cannot always be understood by translating each word individually.
1. One Expression Can Have Different Meanings
Consider the expression “You don try.”
Depending on the context, it can mean that someone has made a good effort, deserves appreciation, or has done enough.
An AI assistant that interprets the phrase only as a statement about trying something may miss its social meaning.
2. Context Changes the Meaning
The expression “You dey craze?” can be a serious insult, a playful reaction between friends, or an expression of disbelief.
The correct interpretation depends on the relationship between the speakers, the conversation, and the tone.
A useful AI test should therefore include both the expression and the surrounding situation.
3. Humour and Sarcasm Are Difficult to Interpret
Nigerian social media often uses exaggeration, irony, teasing, and playful insults.
For example, someone might say, “You don finish work o,” after a friend makes a ridiculous mistake.
The phrase may be sarcastic rather than genuine praise. An AI assistant needs to understand the situation instead of interpreting the words literally.
4. Pidgin Varies Across Communities
Nigerian Pidgin is not used identically by everyone. People may combine it with English, local languages, regional expressions, or internet slang.
An AI assistant may understand a common expression but struggle with a less familiar phrase or a regional variation.
That is why a single sentence is not enough to determine whether an AI tool understands Nigerian Pidgin well.
The Three AI Assistants to Test
For a practical comparison, begin with three widely accessible AI assistants.
ChatGPT
ChatGPT can interpret conversational language, explain expressions, and generate responses in different writing styles.
Website: https://chatgpt.com/
Google Gemini
Gemini can help interpret informal language, explain the meaning of phrases, and respond to conversational prompts.
Website: https://gemini.google.com/
Microsoft Copilot
Microsoft Copilot is another assistant worth including in the comparison. It can be tested on the same expressions to examine how accurately it interprets context and slang.
Website: https://copilot.microsoft.com/
The comparison should use the same prompts for every assistant. Record the model name, date, relevant settings, and any browsing or reasoning features enabled, because these factors can influence the answers.
The 30-Prompt Nigerian Pidgin Test
The following prompts cover five categories: everyday expressions, slang, humour, cultural context, and conversational ability.
Each prompt should be submitted separately to every AI assistant. Where a situation is provided, include it so the model can interpret the phrase correctly.
Category 1: Everyday Expressions
These questions test whether the AI can explain commonly used expressions without relying on word-for-word translation.
- Prompt 1: What does “Abeg, no dey whine me” mean in Nigerian Pidgin?
- Prompt 2: Explain the meaning of “You don try” in a conversation between two friends.
- Prompt 3: What does “Nawa o” usually communicate, and how can its meaning change with context?
- Prompt 4: Translate “I no get your time today” into natural English and explain the possible tone.
- Prompt 5: What does “E no concern me” mean, and when might someone use it?
Category 2: Slang and Informal Communication
These questions test whether the AI can explain slang and informal expressions in context.
- Prompt 6: What does “You dey form” mean when one Nigerian tells another person this?
- Prompt 7: Explain the expression “No wahala” and give two situations where it could be used.
- Prompt 8: What might “This guy too sabi” mean when friends are discussing someone?
- Prompt 9: Explain “I go yarn you later” in natural English.
- Prompt 10: What does “Make we reason am” mean in an informal conversation?
Category 3: Humour and Sarcasm
These prompts test whether the AI can identify the difference between literal meaning and implied meaning.
- Prompt 11: A friend makes an obvious mistake. You reply, “Oga, you don finish work o.” Is this necessarily praise? Explain.
- Prompt 12: Someone spends all their money on unnecessary purchases and says they are broke. Their friend replies, “You really get financial wisdom.” What is the likely intention?
- Prompt 13: Explain how “Who send you message?” can be used as a humorous or dismissive response.
- Prompt 14: A person arrives very late and says traffic delayed them. Their friend replies, “Na traffic carry you go another state?” Explain the joke.
- Prompt 15: Explain why “You too like better thing” may be playful teasing rather than a serious criticism.
Category 4: Cultural and Conversational Context
These questions test whether the AI can understand the social meaning of an expression.
- Prompt 16: What does “How far?” usually mean when Nigerians greet one another informally?
- Prompt 17: Explain the possible meanings of “We thank God” in a conversation about a difficult experience.
- Prompt 18: What might “God abeg” communicate when someone sees an unbelievable social media post?
- Prompt 19: Explain the difference between a serious complaint and playful teasing when someone says, “You no get shame?”
- Prompt 20: A friend says, “I dey manage.” What might this mean when you ask how they are doing?
Category 5: Understanding and Responding Naturally
These prompts test whether the AI can participate in a conversation rather than merely define expressions.
- Prompt 21: Reply naturally in Nigerian Pidgin to a friend who says, “My padi, money no dey at all.”
- Prompt 22: Explain “Person wey never chop dey advise person wey dey cook” and identify the lesson it might communicate.
- Prompt 23: A friend says, “I don tire for this life,” after a stressful day. Write a supportive response in Nigerian Pidgin without making fun of the person.
- Prompt 24: Explain the difference between “I no fit come” and “I no wan come.”
- Prompt 25: Translate “If you know, you know” into a natural Nigerian Pidgin expression while preserving its meaning.
Category 6: Mixed Language and Context Challenges
These final questions test whether the AI can handle more complicated situations.
- Prompt 26: Interpret “Abeg shift small, make I see wetin dey happen” and explain the situation in which someone might say it.
- Prompt 27: A person says, “This your explanation get as e be.” Explain why the statement may express doubt or dissatisfaction.
- Prompt 28: Interpret “I no send anybody” and explain how tone affects whether it sounds confident, dismissive, or confrontational.
- Prompt 29: Write a short, friendly Nigerian Pidgin conversation between two friends discussing an upcoming football match.
- Prompt 30: Explain this message in context: “Guy, you carry last for this one o 😂.” What does it suggest, and how might the laughing emoji affect the tone?
These prompts form a starting test set. Because some expressions have multiple legitimate interpretations, the scoring should allow for reasonable alternatives rather than requiring every AI to use identical wording.
How to Score Each AI Assistant
To make the experiment meaningful, every assistant should receive the same scoring criteria.
Give each response a score from zero to four.
- 0 points: The response misunderstands the expression or gives an incorrect explanation.
- 1 point: The response identifies a small part of the meaning but misses the main point.
- 2 points: The response captures the basic meaning but misses important context or tone.
- 3 points: The response explains the meaning accurately and handles most of the context.
- 4 points: The response accurately explains the meaning, tone, and context, using natural language where a conversational reply is required.
With 30 prompts and a maximum of four points per prompt, each AI assistant can score up to 120 points.
The percentage score can be calculated using this formula:
Score percentage = (Points earned ÷ 120) × 100
For example, if an assistant earns 96 points, its score is 80%.
This is an illustrative calculation, not an actual test result.
What Should Count as a Correct Answer?
A response should not be marked wrong simply because it uses different words from the expected answer.
For example, “Abeg, no dey whine me” may be explained as “Please, stop playing with me,” “Don't try to deceive me,” or “Please, don't take me for a fool,” depending on the context.
All three can be reasonable interpretations.
The evaluator should focus on meaning, tone, and contextual accuracy rather than exact wording.
For more reliable results, two people familiar with Nigerian Pidgin could independently score the responses and discuss disagreements.
How to Make the Experiment Fair
A comparison can become misleading if each AI receives different questions or different instructions.
Use the following testing procedure:
- Submit the same 30 prompts to every assistant.
- Use the same wording and context for each prompt.
- Start each prompt in a fresh conversation to reduce the influence of earlier answers.
- Avoid correcting an AI before recording its initial response.
- Record the assistant's model name and the date of testing.
- Save screenshots or copies of the responses.
- Score the answers using the same rubric.
- Include explanations of disputed answers and reasonable alternative interpretations.
The experiment should also distinguish between understanding Pidgin and producing natural Pidgin. An assistant may explain an expression correctly in standard English but produce unnatural phrasing when asked to respond in Pidgin.
These are related skills, but they are not identical.
Results Table: What to Record
After testing all three assistants, organize the actual results in a simple record.
- ChatGPT: points earned out of 120, percentage score, strongest category, weakest category.
- Google Gemini: points earned out of 120, percentage score, strongest category, weakest category.
- Microsoft Copilot: points earned out of 120, percentage score, strongest category, weakest category.
You can also calculate category scores separately. For example, everyday expressions account for 20 possible points because there are five prompts in that category, each worth four points.
Category-level results may reveal that one assistant understands everyday expressions well but struggles with sarcasm or natural conversational replies.
Do not publish a winner until the answers have actually been collected and evaluated. If two assistants perform similarly, report the difference rather than presenting a small score gap as proof that one is universally better.
Why This Test Matters for Nigerian AI Users
Understanding Nigerian Pidgin is useful beyond entertainment.
Customer Service
Businesses may receive customer messages containing informal expressions, slang, and mixed English. An assistant that interprets these messages correctly may be more useful for drafting responses.
However, business owners should still verify important information and review sensitive conversations.
Education
Students may prefer explanations that use familiar expressions and everyday examples. An AI tool that can explain difficult concepts in accessible language may improve the learning experience.
The quality of the explanation still matters, and technical terms should remain accurate.
Content Creation
Creators may use AI to brainstorm captions, explain jokes, draft dialogue, or develop scripts for Nigerian audiences.
Understanding local context can help reduce awkward wording and improve the relevance of the content.
Accessibility and Communication
Conversational AI may help people communicate across different language preferences. More accurate interpretation of informal expressions could make these tools easier to use for some Nigerians.
Nevertheless, Pidgin varies across speakers and communities, so no single test can represent every user's experience.
Limitations of the Test
A 30-prompt experiment is a useful starting point, but it cannot establish which assistant is best for every Nigerian Pidgin conversation.
The selected expressions may be more familiar to some models than others. Results may also change after model updates or when the prompts are presented in different contexts.
Some expressions have multiple meanings, and cultural interpretation can be subjective. A model may produce an acceptable answer even if it does not match the evaluator's preferred wording.
For stronger evidence, a larger study would need more prompts, speakers from different backgrounds, independent evaluators, and repeated testing.
Publishing these limitations makes the comparison more trustworthy rather than less valuable.
Final Verdict
Which AI understands Nigerian Pidgin best? The honest answer is that it should be determined through testing rather than assumption.
ChatGPT, Google Gemini, and Microsoft Copilot are useful candidates for a side-by-side comparison, but their relative performance on Nigerian Pidgin should be established by their actual responses to the same prompts.
A carefully conducted 30-prompt experiment can reveal how well each assistant handles everyday expressions, slang, sarcasm, cultural context, and natural conversation.
The results could also help Nigerian students, businesses, and content creators choose tools that better suit their communication needs.
For ToolPillot, the real opportunity is to publish the test responses, scoring methodology, screenshots, and findings. That would turn a general AI comparison into an original resource that readers can evaluate for themselves.
The most useful conclusion will not simply identify a winner. It will show which expressions each assistant understands, where misunderstandings occur, and what Nigerian users should know before relying on AI to interpret their language.