Scripts Through Time: A Survey of the Evolving Role of Transliteration in NLP

2026-04-22 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, extended

Summary

This survey paper, "Scripts Through Time: A Survey of the Evolving Role of Transliteration in NLP" by Jayakumar, Halder, and Dabre, comprehensively analyzes the application of transliteration in cross-lingual Natural Language Processing. It addresses the "script barrier" that hinders transfer learning between languages with different writing systems by increasing lexical overlap. The authors present a taxonomy of motivations for using transliteration in language models, including handling named entities, code-mixed text, leveraging language family relatedness, and improving training/inference efficiency. The paper also overviews various integration approaches, such as data-level, input-level, architecture-level, and inference-level methods, discussing their evolution, effectiveness, and trade-offs. It concludes with recommendations for researchers on selecting appropriate transliteration strategies, emphasizing the practical advantages of romanization and its implicit presence in modern LLMs.

Key takeaway

For research scientists developing multilingual NLP models, you should strategically evaluate transliteration as a core technique to overcome script barriers and enhance cross-lingual transfer. Prioritize methods like multi-source self-ensembling or alignment objectives for their architectural simplicity and effectiveness, especially when dealing with low-resource languages. Be mindful of potential information loss in generative tasks and consider reversible transliteration frameworks or multi-script aware architectures to maintain native script fidelity.

Key insights

Transliteration bridges the "script barrier" in NLP, enhancing cross-lingual transfer and efficiency, especially for low-resource languages.

Principles

• Transliteration increases lexical overlap between languages.
• Romanization often reduces token fertility and API costs.
• Architectural minimalism is preferred for model reusability.

Method

Transliteration can be integrated at data, input, architecture, or inference levels, with methods like direct transliteration, embedding fusion, script adapters, or multi-source ensembles.

In practice

• Use transliteration for low-resource languages with high-resource script relatives.
• Consider reversible transliteration for generative tasks to preserve native script output.
• Employ script-based adapters to prevent gradient interference between scripts.

Topics

• Transliteration Strategies
• Cross-lingual NLP
• Script Barrier
• Large Language Models
• Romanization

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

Related on AIssential

• Scripts Through Time: A Survey of the Evolving Role of Transliteration in NLP — Transliteration bridges script barriers in cross-lingual NLP, enhancing lexical overlap and transfer learning.
• Building Singlish2Sinhala: A Machine Learning Approach to Sinhala Transliteration — Informal transliteration systems like Singlish present unique NLP challenges due to spelling inconsistencies.
• Decoding the Maghreb Diaspora: A Multilingual NLP Framework for Social Integration — Specialized NLP architectures are crucial for handling code-switching in under-resourced, hybrid languages.
• Modelling the Morphology of Verbal Paradigms: A Case Study in the Tokenization of Turkish and Hebrew — Tokenization strategy critically impacts Transformer models' ability to represent complex morphology, especially in non-concatenative languages.
• KIT's Submission to Cross-Lingual Voice Cloning in IWSLT 2026 — Explicit language tags and lexical matching significantly improve cross-lingual voice cloning performance and reduce accent leakage.
• Translating the Untranslatable: An Operationalizable Ontology for Untranslatability — A new ontology and dataset enable strategy-informed machine translation for untranslatable linguistic phenomena.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.