AI-assisted cultural heritage dissemination: Comparing NMT and glossary-augmented LLM translation in rock art documents

2026-05-14 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, quick

Summary

A study compared three English machine translation (MT) setups for a Spanish academic rock art text to address challenges in multilingual cultural heritage dissemination. The setups included DeepL as a Neural Machine Translation (NMT) baseline, Gemini-Simple using a Large Language Model (LLM) with a basic prompt, and Gemini-RAG, which augmented the same LLM with glossary-based prompting via term-pair retrieval. Human evaluation, conducted using multi-way Direct Assessment (0-100) and targeted terminology auditing with an MQM taxonomy, revealed that Gemini-RAG achieved the highest exact-match terminology accuracy at 81.4%, significantly outperforming Gemini-Simple (69.1%) and DeepL (64.4%). Gemini-RAG also maintained high overall quality with a mean Direct Assessment score of 85.3, comparable to Gemini-Simple's 85.2, and superior to DeepL's 80.3. These findings suggest that glossary-augmented prompting is an effective, low-overhead method for enhancing terminology control in specialized translation.

Key takeaway

For cultural heritage institutions translating specialized documents, integrating glossary-augmented prompting with LLMs can dramatically improve terminology accuracy and overall translation quality. You should prioritize maintaining concise terminology resources and establishing lightweight evaluation procedures to implement this low-overhead method effectively, ensuring consistent and accurate dissemination of complex information to a global audience.

Key insights

Glossary-augmented LLM prompting significantly improves terminology accuracy in specialized translation without sacrificing overall quality.

Principles

• Terminology control is crucial for specialized domain translation.
• Glossary augmentation enhances LLM translation accuracy.

Method

The study compared DeepL NMT, a basic LLM prompt (Gemini-Simple), and a glossary-augmented LLM prompt (Gemini-RAG) for Spanish-to-English rock art text translation, evaluated by human direct assessment and terminology auditing.

In practice

• Maintain minimal terminology resources for specialized domains.
• Implement lightweight evaluation for translation quality.
• Use RAG with LLMs for terminology-dense content.

Topics

• Cultural Heritage Dissemination
• Rock Art Translation
• Neural Machine Translation
• Large Language Models
• Glossary-Augmented Prompting

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

Related on AIssential

• AI-assisted cultural heritage dissemination: Comparing NMT and glossary-augmented LLM translation in rock art documents — Glossary-augmented LLMs significantly improve terminology accuracy in specialized translation without sacrificing overall quality.
• Label-Aware Pseudo-Training Sample Generation for Text Classification — A new data augmentation method uses LLMs to subtly alter sentences, improving NLP performance with limited data.
• Adaptive Intellect Unleashed: The Feasibility of Knowledge Transfer in Large Language Models — Knowledge transfer can improve LLM generalization in software engineering tasks, but its implementation requires careful design.
• Most Influential ArXiv (Computation and Language) Papers (2026-04 Version) — Citation-based ranking reveals key trends in NLP research, emphasizing LLM reasoning, efficiency, and multimodal capabilities.
• Disentangling meaning from language in LLM-based machine translation — LLM machine translation functions are disentangled into distinct, specialized attention head sets for language and meaning.
• The Evolving Landscape of LLM Evaluation — Recent AI advancements include top open-weights models, zero-shot MT, and evolving job market trends.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.