Want Better Synthetic Data? Steer It: Activation Steering for Low-Resource Language Generation
Quick Answer
This study introduces activation steering for generating synthetic data in low-resource languages, enhancing diversity and downstream performance.
Quick Take
Evaluating four open-source , the authors find that early-layer steering improves sentiment and topic classification tasks, outperforming traditional few-shot prompting methods.
Key Points
- Activation steering improves synthetic data generation for low-resource languages.
- Two strategies: Language Steering for linguistic identity and Quality Steering for well-formedness.
- Evaluated on four open-source LLMs across 11 diverse languages.
- Early-layer steering enhances data diversity and downstream model performance.
- Results show significant improvements in sentiment and topic classification tasks.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Large language models (LLMs) have become an effective tool for synthetic data generation, including for low-resource languages, where generated data can improve downstream task performance. Current best-performing approaches typically rely on few-shot prompting with target-language examples, which increases inference costs and may reduce diversity through lexical anchoring. In this work, we investigate activation steering as an alternative for low-resource synthetic data generation. We study two steering strategies: Language Steering, which targets the linguistic identity of a language, and Quality Steering, which captures well-formedness by contrasting human-written and backtranslated text representations. We evaluate these methods across four open-source LLMs, multiple layers, and 11 typologically diverse languages by generating sentiment and topic classification data and finetuning smaller classifiers. Steering is applied in both zero-shot and few-shot prompting settings and compared against non-steered counterparts. Our results show that steering on early layers consistently improves the diversity of generated data while often yielding stronger downstream model performance, particularly for low-resource languages.
| Comments: | 25 pages |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2606.18389 [cs.CL] |
| (or arXiv:2606.18389v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.18389 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jan Cegin [view email]
[v1]
Tue, 16 Jun 2026 18:34:21 UTC (1,318 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.