Soft Token Alignment for Cross-Lingual Reasoning
Quick Answer
The proposed SOLAR method enhances multilingual large language models by aligning soft-token representations across languages, improving accuracy by up to 17.7 points on reasoning benchmarks.
Quick Take
This approach reduces language-specific divergences, particularly benefiting low-resource languages and preserving shared semantic structures during reasoning.
Key Points
- SOLAR aligns soft-token representations using English as a pivot for multilingual models.
- Achieves up to +17.7 accuracy improvement on four multilingual reasoning benchmarks.
- Largest gains observed in low-resource languages, enhancing their performance.
- Reduces language-cluster separability, preserving semantic structure across languages.
- Strengthens final-layer cross-lingual similarity in multilingual reasoning tasks.
Paper Resources
📖 Reader Mode
~2 min readAbstract: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.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2606.26466 [cs.CL] |
| (or arXiv:2606.26466v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.26466 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jiayi He [view email]
[v1]
Thu, 25 Jun 2026 00:01:58 UTC (391 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →TriAgent: Divergence-Aware Committees for Cost-Efficient Financial Sentiment Analysis
TriAgent introduces a cost-efficient multi-agent system for financial sentiment analysis, combining VADER, FinBERT, and Qwen2.5. It achieves an F1 score of ~0.87 with significant savings of $9.3M/year at a 10M-user scale compared to GPT-4o-mini, while also detecting hallucinations with an AUC of 0.90.