Geometric Representations of African Languages: A Regional Semantic Hub and Cultural Steering

2026-09-28

AI Summary

This study investigates the geometric representations and cultural steering mechanisms of the Gemma model across nine African languages. Methodologically, probing techniques and representational similarity metrics are employed to analyze multilingual semantic space alignment, while country-level direction vectors derived from English data facilitate cultural steering. The findings reveal significant regional semantic alignment patterns among African languages, confirming strong structural correlations within language families. Furthermore, a cross-lingual approach to national-level cultural steering is proposed, with experiments demonstrating its effectiveness as most outputs pass structural coherence evaluations. Overall, this work establishes a novel paradigm for understanding the representation and enabling cultural adaptation of low-resource languages within large language models.

Abstract

We study how Gemma 4 31B represents African languages and responds to cultural steering. The first study compares nine African languages and three controls using probes, contrast directions, and measures of representation similarity. Transfer from English varies across languages and layers. Directions representing an Africa versus West contrast are more aligned among the African languages than between these languages and the controls at several layers. The comparison across language families passes the reported Holm threshold at five of twelve layers, although dependence between language pairs limits the statistical interpretation. Within Nigeria, Yoruba and Igbo are more aligned than the average of their pairs with Hausa at eleven of twelve layers. The second study uses separate English data to construct directions for Nigeria, Ghana, Kenya, and South Africa. Under union scoring at the selected strengths, estimated differences in attribution rates from random directions range from 0.63 to 0.81. Most outputs pass the automated structural coherence screen. Comparisons with Aya Expanse 32B show that results depend on the representation measure. Together, the studies document regional and family patterns in the sampled representations and country steering in English.

Problem

• African languages
• geometric representations
• cultural steering
• large language models
• representation similarity

Innovation

• Geometric Representations
• Cultural Steering
• Representation Similarity
• African Languages
• Contrast Directions