Transliteration is the conversion of text from one script to another based on phonetic similarity, without translating the meaning — e.g., writing Hindi words using the Latin alphabet ("Namaste") or Japanese names in English ("Tokyo").

NLP Challenges

• Ambiguity: "Mein" in Latin script could be German ("my") or transliterated Hindi ("in").
• Variation: No standard spanning — "Qubool", "Kubool", "Qabul" might all mirror the same Urdu word.
• Bridge: Transliteration is often used to bridge high-resource scripts (Latin) to low-resource scripts.

Why It Matters

• Input Methods: Many users type their native language using QWERTY keyboards (Latin script).
• Preprocessing: Often necessary to normalize text before feeding it to a model, or to train models to handle both scripts.
• U-Roman: Universal Romanization is a strategy to train multilingual models by converting EVERYTHING to Latin script first to maximize vocabulary sharing.

Transliteration is script swapping — writing a language in a different alphabet, creating a unique challenge of phonetic mapping vs. semantic meaning.