liveBook · Manning

Transliteration

As explained in Natural Language Processing in Action, Second Edition

Transliteration is the conversion of characters and spellings from one writing system into another while preserving the source language's linguistic content as closely as possible.

Transliteration converts characters and spellings between writing systems so words can be represented in a shared alphabet, often Latin. It differs from translation: नमस्कार may be transliterated as namaska¯ra before being translated as hello. Accent changes such as résumé to resume provide a simpler case, while non-Latin alphabets are harder. section 8.2

Names across scripts and languages

A multilingual surname system may normalize names into Latin characters before processing them with an RNN. This supports shared pipelines, but reused spellings across national borders mean transliterated forms do not uniquely identify language or nationality. section 8.2

How Natural Language Processing in Action, Second Edition covers it

How deep it goes
Defines transliteration and demonstrates normalization, but stops short of explaining non-Latin conversion algorithms or RNN internals.
Examples
Uses accented French words, a Nepalese greeting, multilingual surnames, and Python Asciifier outputs.
What it assumes
Assumes familiarity with alphabets and distinguishes transliteration from translation before connecting both to NLP models.
What you can do after
Apply transliteration through the nlpia2 Asciifier to normalize accented Latin text.
How it connects
Connects transliteration with translation, recurrent neural networks, multilingual NLP, and nationality classification.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Natural Language Processing in Action, Second Edition. The section links point to the parts of the book it draws on.