The Register Gap: A Meaning Intelligence Framework for Nigerian Public Discourse

Researchers introduced the Meaning Intelligence Framework (MIF), a nine-dimension evaluation schema that improves AI systems' ability to understand Nigerian public discourse by separating surface sentiment from true communicative intent. The framework increased register classification accuracy from 33.3% to 73.3% when applied to frontier language models, revealing that context failure—not translation failure—is the primary limitation of current AI systems on Nigerian languages.

The Meaning Intelligence Framework addresses a critical blind spot in natural language processing: existing sentiment classification systems treat language as a polarity detection problem, ignoring the pragmatic reality that meaning emerges from context, speaker identity, audience relationship, and situational factors. Nigerian discourse presents a particularly complex case, spanning Standard English, Nigerian English, Nigerian Pidgin, and code-mixed utterances where identical statements carry opposite meanings depending on these contextual variables. This research demonstrates that the failure mode plaguing AI systems isn't linguistic translation but semantic interpretation—the inability to map surface-level expressions onto their true communicative intentions.

The Register Gap—a 40-point accuracy improvement when models receive the MIF schema—reveals that frontier language models like Gemini 2.5 Flash possess latent capabilities to understand nuanced discourse but require explicit guidance through structured prompting. This finding has direct implications for AI development in non-Western linguistic contexts, where cultural communication norms diverge significantly from training data dominated by Western English. The framework's nine-dimension approach operationalizes previously informal analytical practices, converting subjective interpretation into reproducible evaluation metrics.

For the AI industry, this work signals growing sophistication in how models handle culturally-embedded communication. The 5.4-point improvement in composite Meaning Intelligence Score, concentrated in coded-subtext detection and strategic action recommendation, suggests practical applications for content moderation, risk assessment, and communication analysis in African markets. Releasing the calibration dataset and annotation guidelines creates infrastructure for benchmarking AI performance on non-English discourse, establishing accountability standards.

• →Register Gap discovery shows zero-shot AI performance on Nigerian discourse jumps 40 points with schema-informed prompting, indicating recoverable rather than fundamental model limitations.
• →Context failure, not translation failure, is the primary failure mode for AI systems analyzing Nigerian languages across multiple registers.
• →The MIF framework's nine-dimension annotation schema operationalizes pragmatic language analysis, making cultural communication patterns measurable for AI evaluation.
• →Coded-subtext detection and strategic action recommendation showed largest improvements (+10 points), indicating high potential for content moderation and risk analysis applications.
• →Released public calibration dataset establishes reproducible benchmarks for evaluating AI performance on non-Western linguistic contexts.