# Soft Token Alignment for Cross-Lingual Reasoning

arXiv: https://arxiv.org/abs/2606.26466 · HTML: https://mlsanity.com/paper/2606.26466/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 24 of its 79 references that are in the map.
- First posted on arXiv 25 Jun 2026.

## Placed in

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

## Builds on (most cited first, 24 of 24)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Decoupled Weight Decay Regularization](https://mlsanity.com/paper/1711.05101/) (2017)
- [Gemma 3 Technical Report](https://mlsanity.com/paper/2503.19786/) (2025)
- [No Language Left Behind: Scaling Human-Centered Machine Translation](https://mlsanity.com/paper/2207.04672/) (2022)
- [Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoning](https://mlsanity.com/paper/2505.16782/) (2025)
- [Crosslingual Reasoning through Test-Time Scaling](https://mlsanity.com/paper/2505.05408/) (2025)
- [LLMs are Single-threaded Reasoners: Demystifying the Working Mechanism of Soft Thinking](https://mlsanity.com/paper/2508.03440/) (2025)
- [Soft Tokens, Hard Truths](https://mlsanity.com/paper/2509.19170/) (2025)
- [Learn Globally, Speak Locally: Bridging the Gaps in Multilingual Reasoning](https://mlsanity.com/paper/2507.05418/) (2025)
- [Language-Specific Latent Process Hinders Cross-Lingual Performance](https://mlsanity.com/paper/2505.13141/) (2025)
- [Think Natively: Unlocking Multilingual Reasoning with Consistency-Enhanced Reinforcement Learning](https://mlsanity.com/paper/2510.07300/) (2025)
- [A Formal Comparison Between Chain of Thought and Latent Thought](https://mlsanity.com/paper/2509.25239/) (2025)
- [Revisiting Multilingual Data Mixtures in Language Model Pretraining](https://mlsanity.com/paper/2510.25947/) (2025)
- [A Comprehensive Evaluation of Multilingual Chain-of-Thought Reasoning: Performance, Consistency, and Faithfulness Across Languages](https://mlsanity.com/paper/2510.09555/) (2025)
- [TAPO: Translation Augmented Policy Optimization for Multilingual Mathematical Reasoning](https://mlsanity.com/paper/2603.25419/) (2026)
- [Cross-Lingual Activation Steering for Multilingual Language Models](https://mlsanity.com/paper/2601.16390/) (2026)
- [Aligning Multilingual Reasoning with Verifiable Semantics from a High-Resource Expert Model](https://mlsanity.com/paper/2509.25543/) (2025)
- [Parallel Scaling Law: Unveiling Reasoning Generalization through A Cross-Linguistic Perspective](https://mlsanity.com/paper/2510.02272/) (2025)
- [Less Data Less Tokens: Multilingual Unification Learning for Efficient Test-Time Reasoning in LLMs](https://mlsanity.com/paper/2506.18341/) (2025)
- [Gradual Code-Switching as Inference-Time Cross-Lingual Representational Alignment for LLMs](https://mlsanity.com/paper/2510.05678/) (2025)
- [Align to the Pivot: Dual Alignment with Self-Feedback for Multilingual Math Reasoning](https://mlsanity.com/paper/2601.17671/) (2026)

## Abstract (arXiv)

Multilingual large language models often produce inconsistent reasoning and answers for semantically equivalent prompts in different languages. Prior work suggests that intermediate representations can be relatively language-agnostic, but generation becomes increasingly language-specific as models commit to discrete output tokens. This is problematic because language-specific lexical choices can cause semantically equivalent reasoning paths to diverge across languages. These divergences motivate searching for a cross-lingual alignment signal that is less tied to any single vocabulary item or script. We propose SOLAR, an auxiliary objective for supervised fine-tuning that aligns soft-token representations across languages, using English as a pivot. Soft tokens are probability-weighted mixtures over the vocabulary embeddings, yielding continuous representations that can aggregate information from semantically related tokens across languages. We then align each non-English soft-token summary to its English counterpart in the shared embedding space. Across four multilingual reasoning benchmarks, SOLAR improves accuracy by up to +17.7 points over the base model and +3.8 over standard supervised fine-tuning, with the largest gains on low-resource languages. SOLAR also strengthens final-layer cross-lingual similarity and substantially reduces language-cluster separability, suggesting that aligning soft-token representations helps preserve shared semantic structure during multilingual reasoning.