What is transliteration, and why does it matter for name fields?

Transliteration is the process of converting characters from one script into another, typically into the Latin alphabet, while trying to preserve how the original word sounds rather than what it means. It's distinct from translation, which converts meaning.

Translating "Иван" would give you the equivalent concept in another language, while transliterating it gives you "Ivan," representing the same sounds using Latin letters. For personal names, place names, and similar identifiers, transliteration is almost always the relevant operation, since a name doesn't really have a "meaning" to translate in the first place.

This matters directly for identity document processing because many countries issue documents in scripts other than Latin, Cyrillic, Arabic, Hebrew, Greek, and others, while many downstream systems that consume extracted document data (databases, compliance checks, booking systems, forms) expect names in Latin characters. Without transliteration, a name extracted in its original script would need manual conversion before it could be used in most Latin-character systems, defeating much of the purpose of automated extraction.

Complicating things further, transliteration isn't always perfectly standardized. Different transliteration systems can render the same original name slightly differently, and some documents actually include a Latin transliteration printed directly on the document itself, a common practice for passports, precisely so international systems can read the name without needing to transliterate it independently, which can sometimes differ slightly from what a fully automated transliteration process would produce.

A well-designed extraction system needs to account for this, ideally by using any transliteration already printed on the document when it's available, and applying a well-established, standard transliteration methodology when it isn't.

Getting this right matters for practical reasons beyond just data neatness. A mismatch between how a name is transliterated in your system versus how it appears on other records associated with the same person, such as a previous booking or an existing account, can cause false negatives when trying to match records that actually refer to the same individual, creating friction or errors in downstream processes like fraud checks or repeat-customer recognition.

ScanDoc's visual zone extraction includes transliteration handling for names in non-Latin scripts, prioritizing any Latin transliteration already printed on the document itself where available, so extracted name data is both accurate and consistent with international conventions for downstream use.