Transliteration and cross-script search

A small library that tells you which script a text is in, writes it in Latin letters by a named library standard, and reduces every spelling of a name to one form, so that one search finds them all.

What it is for

A library catalogue with Armenian, Georgian, Russian or Syriac material records the same name several ways: in its own script, and in two or five Latin spellings depending on who catalogued it and which standard they followed. Armenia’s national poet is Tumanyan, Tumanean, Toumanyan and Tumanian on four book covers, and all four are correct. No stock search engine treats them as one word, so a reader who types one spelling misses the records filed under the others.

tetrak-translit solves that in two directions. For readers, it writes a name in Latin letters exactly as a library standard prescribes. For search, it folds every spelling, Latin or script, to one key, so that an index built with it finds a record however the name was typed.

Who it is for

Catalogue and discovery systems that hold material in these scripts (Koha, Solr and Elasticsearch analysis chains, aggregators); digitisation projects that want a romanised field beside the original; and anyone who has to convert between a script and its romanisations without guessing. It is a dependency-free Python library with a command-line tool, so it fits into a pipeline or a spreadsheet job as easily as into a search engine.

What it does today

Three jobs, in plain words:

1. Detection. Which script a string is in, and some things about how it is written: for Armenian, whether it uses the classical spelling (printed before the 1922 reform, and still used in the diaspora) or the reformed one; for Russian, whether it is pre-1918 spelling; for Syriac, whether it carries vowel points.
2. Transliteration, between a script and its named Latin schemes. Each scheme is a tested table written from its published standard, and the library says which ones can be reversed exactly and refuses to guess at the ones that cannot.
3. The fold. One function that maps any spelling, in the script or in any scheme, to a single search key, so that the index and the query agree. The key is deliberately coarse and never shown to a reader.

Syriac is the newest and the most limited: the script is normally written without vowels, so the fold works on consonants alone. The article on searching Syriac explains what that buys and what it costs. The current list of schemes, with every letter of every table, is in the documentation.

What it does not do yet

• Translate. Transliteration is the same words in different letters.
• Supply vowels for unpointed Syriac. The library renders what is written and folds on the consonants.
• Reverse a lossy scheme. Where a standard maps two letters to one, the library says so rather than guessing.
• Fold every popular spelling. Spellings that came through another language (Khachaturian via Russian, Edessa via Greek) or that insert an unwritten vowel do not fold to the script. The published test sets are the claim: what is not in them is not promised.
• Every script. Four so far. Adding one is a package of tables, and the documentation explains how.

Get it

pip install tetrak-translit

import tetrak_translit as tt

tt.transliterate("Երևան", "ala_lc")       # 'Erevan'
tt.transliterate("Երևան", "bgn_pcgn")     # 'Yerevan'
tt.fold("Yerevan", script="hy") == tt.fold("Երեւան")   # True

Articles about it

• Searching Syriac, a script that leaves out its vowels: the consonantal fold, and where we would welcome help.
• Measuring the fold, release by release: how the Armenian fold is tested against a real catalogue on every release.
• Transliterate for display, fold for finding: why a transliteration is for the reader and never for the index.
• What changed in ISO 9984 and ISO 9985 in 2026: the new editions of the standards, and the apostrophe that moved.
• Telling classical from reformed Armenian by counting one letter: the orthography detector.