Researchers have developed a new benchmark and parallel corpus to improve Arabic-Russian scientific translation. The benchmark includes approximately 27,000 sentence pairs compiled from scientific abstracts and general texts. Fine-tuning multilingual language models like Qwen2.5-7B-Instruct with LoRA techniques resulted in significant improvements in translation quality, demonstrating the necessity of domain-specific fine-tuning over few-shot prompting. AI
IMPACT This work facilitates knowledge exchange between Arabic and Russian scientific communities, potentially accelerating research collaboration and innovation.
RANK_REASON The cluster describes a new academic paper presenting a parallel corpus and benchmark for a specific language pair, along with fine-tuned models.
- Arabic
- arXiv
- Hugging Face
- LoRA
- mT5-base
- Mullosharaf Arabov Am
- NLLB-200-distilled-1.3B
- QLoRA
- Qwen2.5-7B-Instruct
- Russian
- Sustainable Development Goal 9
- UN SDG 17
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
