analiza aiqbench
This research investigates how the Gemma 4 31B model encodes African languages and responds to cultural steering, revealing distinct regional alignment patterns across nine languages and multiple neural layers. The study demonstrates that representations of an "Africa versus West" contrast are significantly more aligned among African languages than with control groups, while also quantifying the efficacy of country-specific steering directions in English. These findings provide critical insights into the geometric structure of multilingual models regarding underrepresented linguistic families and their susceptibility to cultural bias.
Geometric Representations of African Languages: A Regional Semantic Hub and Cultural Steering
arXiv:2609.36205v1 Announce Type: new 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.
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.
Computation and Language (cs.CL)
arXiv:2609.36205 [cs.CL]
(or arXiv:2609.36205v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.36205 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 280 cuvinte
Subiecte conexe

BRICS Draws a New Global Line: Full Text of New Delhi Declaration
slguardian.org · 12 sept. 2026
LLM Safety Alignment in Low-Resource Languages: A Systematic Literature Review
arXiv · 19 aug. 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026
CAMMAR: Culture-Aware Matryoshka for Metaphorical Arabic Representations
arXiv · 20 iul. 2026