# SARA: Unlocking Multilingual Knowledge in Mixture-of-Experts via Semantically Anchored Routing Alignment

arXiv: https://arxiv.org/abs/2606.25821 · HTML: https://mlsanity.com/paper/2606.25821/

## Key facts

- No citing papers in the map yet (data through 13 Oct 2026).
- Placed in Expert routing strategies, one of 12 directions in Mixture-of-experts.
- Builds on 19 of its 38 references that are in the map.
- First posted on arXiv 24 Jun 2026.

## Placed in

- Mixture-of-experts / [Expert routing strategies](https://mlsanity.com/moe/expert-routing-strategies/)

## Builds on (most cited first, 19 of 19)

- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models](https://mlsanity.com/paper/2309.12284/) (2023)
- [Language Models are Multilingual Chain-of-Thought Reasoners](https://mlsanity.com/paper/2210.03057/) (2022)
- [The Belebele Benchmark: a Parallel Reading Comprehension Dataset in 122 Language Variants](https://mlsanity.com/paper/2308.16884/) (2023)
- [Global MMLU: Understanding and Addressing Cultural and Linguistic Biases in Multilingual Evaluation](https://mlsanity.com/paper/2412.03304/) (2024)
- [On the Representation Collapse of Sparse Mixture of Experts](https://mlsanity.com/paper/2204.09179/) (2022)
- [Not All Experts are Equal: Efficient Expert Pruning and Skipping for Mixture-of-Experts Large Language Models](https://mlsanity.com/paper/2402.14800/) (2024)
- [FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Language](https://mlsanity.com/paper/2506.20920/) (2025)
- [Do Multilingual Language Models Think Better in English?](https://mlsanity.com/paper/2308.01223/) (2023)
- [Diversifying the Expert Knowledge for Task-Agnostic Pruning in Sparse Mixture-of-Experts](https://mlsanity.com/paper/2407.09590/) (2024)
- [XMoE: Sparse Models with Fine-grained and Adaptive Expert Selection](https://mlsanity.com/paper/2403.18926/) (2024)
- [HyperMoE: Towards Better Mixture of Experts via Transferring Among Experts](https://mlsanity.com/paper/2402.12656/) (2024)
- [Multilingual Routing in Mixture-of-Experts](https://mlsanity.com/paper/2510.04694/) (2025)
- [MoE-LPR: Multilingual Extension of Large Language Models through Mixture-of-Experts with Language Priors Routing](https://mlsanity.com/paper/2408.11396/) (2024)
- [MIT-10M: A Large Scale Parallel Corpus of Multilingual Image Translation](https://mlsanity.com/paper/2412.07147/) (2024)
- [Rethinking Multilingual Continual Pretraining: Data Mixing for Adapting LLMs Across Languages and Resources](https://mlsanity.com/paper/2504.04152/) (2025)
- [Unsupervised Domain Adaptation Schemes for Building ASR in Low-resource Languages](https://mlsanity.com/paper/2109.05494/) (2021)
- [Routing Manifold Alignment Improves Generalization of Mixture-of-Experts LLMs](https://mlsanity.com/paper/2511.07419/) (2025)
- [H-AES: Towards Automated Essay Scoring for Hindi](https://mlsanity.com/paper/2302.14635/) (2023)

## Abstract (arXiv)

Sparse Mixture-of-Experts (MoE) architectures have emerged as an increasingly influential paradigm as they offer a strategic balance between parameter scalability and computational efficiency. However, low-resource languages, which suffer from a scarcity of high-quality training data, often have their tokens routed to different experts than those predominantly activated by high-resource inputs, which limits cross-lingual expert sharing. This cross-lingual routing divergence consequently hinders their efficacy in multilingual contexts. To address this issue, we propose SARA (Semantically Anchored Routing Alignment), a framework designed to transfer specialized capabilities from high-resource languages as anchors to low-resource languages. SARA explicitly aligns the routing distribution of multilingual inputs with high-resource semantic anchors using a symmetric Jensen-Shannon (JS) divergence constraint. Unlike traditional distillation methods that operate on output logits, SARA directly aligns the internal routing distributions of MoE layers, encouraging mechanistic consistency in expert selection across languages. We conduct experiments on 2 LLMs across 5 low-resource languages and 3 benchmarks. Experiment results demonstrate that SARA outperforms standard instruction tuning, e.g., +0.8% on Qwen3-30B-A3B and +1.2% on Phi-3.5-MoE-instruct on Global-MMLU. Further analyses show that SARA effectively addresses performance bottlenecks in low-resource languages, providing a scalable pathway to enhance multilingual capabilities in sparse architectures.