Cross-lingual Relation Extraction with Large Language Models: Zero-Shot, Few-Shot, and Fine-Tuned Evaluation on Romanian

2026-06-30 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

A study investigated cross-lingual relation extraction (RE) for Romanian, a low-resource language, by combining automatic dataset translation with large language model (LLM) inference. Researchers translated the SemEval-2010 Task 8 benchmark from English to Romanian using an LLM-based pipeline. They evaluated Gemma 4 31B in zero-shot, few-shot, and QLoRA fine-tuned configurations, alongside four encoder baselines ranging from 125M to 560M parameters, including XLM-RoBERTa and Romanian BERT. Results showed Romanian incurred a 3 to 5 percentage point (pp) drop relative to English in prompt-only settings, with few-shot prompting offering marginal gains. QLoRA fine-tuning significantly improved macro F1-Score by over 22 pp in both languages and reduced the cross-lingual gap from 3.3 to 1.4 pp. Notably, smaller encoder baselines performed within 1-4 pp of QLoRA Gemma on Romanian, suggesting a weak case for large 31B models in compute-sensitive deployment scenarios for single-task RE. The translated dataset, evaluation code, and trained models are released.

Key takeaway

For Machine Learning Engineers developing relation extraction systems for low-resource languages, you should prioritize QLoRA fine-tuning over zero-shot or few-shot prompting. While QLoRA significantly improves macro F1-Score by over 22 percentage points and reduces cross-lingual performance gaps, consider that smaller encoder models can achieve comparable results within 1-4 percentage points of a 31B LLM. This suggests optimizing for compute efficiency by evaluating smaller, specialized models for deployment.

Key insights

QLoRA fine-tuning significantly boosts cross-lingual relation extraction performance for low-resource languages, outperforming prompt-only LLMs.

Principles

• Low-resource RE benefits from translated datasets.
• Fine-tuning LLMs outperforms prompt-only methods.
• Smaller encoders can rival large LLMs for specific tasks.

Method

Cross-lingual RE for Romanian was achieved by translating the SemEval-2010 Task 8 benchmark from English using an LLM-based pipeline, then evaluating Gemma 4 31B and encoder baselines.

In practice

• Use QLoRA for cross-lingual RE tasks.
• Consider smaller encoders for compute-constrained RE.
• Translate English benchmarks for low-resource languages.

Topics

• Cross-lingual Relation Extraction
• Large Language Models
• QLoRA Fine-tuning
• Low-Resource Languages
• Romanian NLP
• SemEval-2010 Task 8

Best for: Research Scientist, AI Engineer, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• Few-Shot Biomedical Relation Extraction with Large Language Models: A Viable Alternative to Supervised Learning? — LLMs offer a viable few-shot alternative for biomedical relation extraction, outperforming prior few-shot methods, especially for rare relations.
• Improving Answer Extraction in Context-based Question Answering Systems Using LLMs — Fine-tuning pre-trained LLMs on benchmark QA datasets significantly improves answer extraction accuracy and contextual understanding in QA systems.
• Zero-Shot to Full-Resource: Cross-lingual Transfer Strategies for Aspect-Based Sentiment Analysis — Cross-lingual transfer strategies for ABSA vary in effectiveness based on model architecture and resource availability.
• Complex & Intelligent Systems, Volume 12, Issue 3, March 2026 — The issue highlights advancements in AI, optimization, and security across diverse applications and system complexities.
• Improving Answer Extraction in Context-based Question Answering Systems Using LLMs — Targeted fine-tuning of pre-trained LLMs significantly improves context-based question answering accuracy and relevance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.