# Rethinking Cross-lingual Alignment: Balancing Transfer and Cultural Erasure in Multilingual LLMs

arXiv: https://arxiv.org/abs/2510.26024 · HTML: https://mlsanity.com/paper/2510.26024/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 6 of its 52 references that are in the map.
- First posted on arXiv 29 Oct 2025.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 6 of 6)

- [Language-Specific Latent Process Hinders Cross-Lingual Performance](https://mlsanity.com/paper/2505.13141/) (2025)
- [Paths Not Taken: Understanding and Mending the Multilingual Factual Recall Pipeline](https://mlsanity.com/paper/2505.20546/) (2025)
- [Localized Cultural Knowledge is Conserved and Controllable in Large Language Models](https://mlsanity.com/paper/2504.10191/) (2025)
- [Learn and Unlearn: Addressing Misinformation in Multilingual LLMs](https://mlsanity.com/paper/2406.13748/) (2024)
- [Can you map it to English? The Role of Cross-Lingual Alignment in Multilingual Performance of LLMs](https://mlsanity.com/paper/2504.09378/) (2025)
- [Beyond the Rosetta Stone: Unification Forces in Generalization Dynamics](https://mlsanity.com/paper/2508.11017/) (2025)

## Built on by (most cited first, 7 of 7)

- [Tiny Aya: Bridging Scale and Multilingual Depth](https://mlsanity.com/paper/2603.11510/) (2026)
- [Location Not Found: Exposing Implicit Local and Global Biases in Multilingual LLMs](https://mlsanity.com/paper/2604.19292/) (2026)
- [Time Travel Engine: A Shared Latent Chronological Manifold Enables Historical Navigation in Large Language Models](https://mlsanity.com/paper/2601.06437/) (2026)
- [The Culture Funnel: You Can't Align What isn't in the Data](https://mlsanity.com/paper/2606.13808/) (2026)
- [Cross-Lingual Representation Alignment by Token-Level Optimal Transport in a Language-Agnostic Space](https://mlsanity.com/paper/2609.06381/) (2026)
- [μ2-Bench: A Multilingual Machine Unlearning Benchmark](https://mlsanity.com/paper/2609.20945/) (2026)
- [Cross-Lingual Alignment for Decoder-Only Models using MoE Routers](https://mlsanity.com/paper/2610.01921/) (2026)

## Abstract (arXiv)

Cross-lingual alignment (CLA) aims to align multilingual representations, enabling Large Language Models (LLMs) to seamlessly transfer knowledge across languages. While intuitive, we hypothesize, this pursuit of representational convergence can inadvertently cause "cultural erasure", the functional loss of providing culturally-situated responses that should diverge based on the query language. In this work, we systematically analyze this trade-off by introducing a holistic evaluation framework, the transfer-localization plane, which quantifies both desirable knowledge transfer and undesirable cultural erasure. Using this framework, we re-evaluate recent CLA approaches and find that they consistently improve factual transfer at the direct cost of cultural localization across all six languages studied. Our investigation into the internal representations of these models reveals a key insight: universal factual transfer and culturally-specific knowledge are optimally steerable at different model layers. Based on this finding, we propose Surgical Steering, a novel inference-time method that disentangles these two objectives. By applying targeted activation steering to distinct layers, our approach achieves a better balance between the two competing dimensions, effectively overcoming the limitations of current alignment techniques.