transliteration

**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.