One person, ten spellings: transliterating names

A name written in Arabic, Russian, or Chinese script has many Latin spellings because no single mapping to Latin letters exists. For an analyst matching people across documents, one person can appear under several spellings that are each correct under some published system, and a search for one spelling can miss the rest. Several romanization systems disagree on individual letters, each language has its own tables, and some scripts leave out information that a Latin spelling needs. A system that compares names has to treat each spelling as one rendering of a name, not as the name itself.

Published systems disagree on the same letters

The UN working group on romanization systems publishes side-by-side comparisons. The Arabic page sets the current UN system next to the earlier UN 1972 system, BGN/PCGN 1956, the I.G.N. 1973 system, ISO 233:1984, and the Survey of Egypt System. BGN here means the U.S. Board on Geographic Names, and PCGN the Permanent Committee on Geographical Names. The page describes BGN/PCGN 1956 as almost identical to UN 1972. The other systems diverge more:

• The letter jim is rendered j in the UN system, dj or j in I.G.N., ǧ in ISO 233, and g (with j in parentheses) in the Survey of Egypt System.
• The letter shin is sh in the UN system, ch in I.G.N., and š in ISO 233.

The Russian page shows the same pattern. The UN system of 1987 is based on a system known as GOST 1983. The letter ю is ju in the UN system, yu in BGN/PCGN 1947, and û in ISO 9:1995. The letter я is ja, ya, or â, and щ is šč, shch, or ŝ. A person whose surname contains those letters can appear in three documents under three spellings, each correct under its own system.

For Chinese, the UN page says Pinyin was approved in 1977 and is used in China and in international cartographic products almost universally (UN Chinese page). It also describes the Modified Wade-Giles transcription of 1892 as previously among the best-known systems, especially in English-language texts. Older documents may use it.

Scripts leave choices open

Three properties of the source languages add variation beyond the choice of system.

Omitted vowels. The Arabic page states that the script usually omits vowel points and diacritical marks, which makes uniform results difficult to obtain. The romanizer must supply vowels the written name does not show.

Name elements that change form. The same page notes that the Arabic definite article assimilates to the first consonant of a following word that begins with a sun letter, as in Ash Shāriqah, so one article can be romanized in more than one form.

Incomplete reversibility. Chinese uses a logographic script, and the UN page says the phonetic notation of a syllable cannot be unambiguously converted back to its written form. For Russian, the UN system is reversible with rare ambiguities. The page's example is a name romanized Bajuba that is reconstituted as Байуба, not Баюба. A Latin spelling therefore cannot always be traced back to the original script.

Which table a document followed

Different bodies publish tables for different uses, and each revises them. The Library of Congress tables serve cataloging, and the BGN/PCGN and UN systems serve geographical names. Their comparison tables work at the level of letters, so they bear on any name written with those letters. A document's spelling depends on which table, and which edition, its author followed.

The Library of Congress page lists the ALA-LC tables, which it says are approved by the Library of Congress and the American Library Association. The Arabic, Persian, Russian, and Chinese tables are dated 2012 and the Pushto (Pashto) table 2013. Earlier versions, including 2011 and 1997, remain posted for reference and are marked as not for current cataloging.

The GOV.UK collection holds the systems approved by the PCGN and BGN for romanizing geographical names. It was last updated on 28 January 2025 and lists separate documents for Arabic (November 2022), Persian (December 2019), Pashto (March 2018), Russian (November 2022), and Chinese (March 2018). The UN working group site describes itself as an archive of working group documents that will not be updated.

What this means for matching

Research on name matching starts from this variation. Pouliquen and colleagues describe a tool for the NewsExplorer news analysis system that groups person-name variants across Greek, Cyrillic, and Arabic scripts. The authors report that they use an internal standard representation for names rather than a conventional bilingual transliteration approach (Pouliquen et al.). The system clusters an average of 25,000 news articles per day.

The design choice follows from the systems above. Comparing two Latin strings directly depends on both having been produced under the same romanization choices. Mapping every variant to a shared internal representation treats each spelling as one view of the name.

What to check

• Which romanization system does each source follow, and does the tool record it?
• Does name matching compare Latin strings directly, or map every variant to a shared internal form?
• Does it match a name across different vowel choices and across the forms of the Arabic article, as in Ash Shāriqah?
• Does it return a Latin spelling to the original script, and flag the cases where the conversion is ambiguous, as with Bajuba?
• Does it recognize older systems, such as the Modified Wade-Giles transcription of 1892, in older documents?

Azcuna's BABEL reads, translates, and searches foreign-language documents entirely on-device, with no cloud and no connection, and keeps every claim tied to the passage it came from.

Sources

1. 1ALA-LC Romanization Tables · Library of Congress
2. 2Romanization systems · Ministry of Defence, GOV.UK (Permanent Committee on Geographical Names)
3. 3UN Romanization of Arabic for Geographical Names (1972) · UNGEGN Working Group on Romanization Systems (archive)
4. 4UN Romanization of Russian for Geographical Names (1987) · UNGEGN Working Group on Romanization Systems (archive)
5. 5UN Romanization of Chinese for Geographical Names (1977) · UNGEGN Working Group on Romanization Systems (archive)
6. 6Multilingual person name recognition and transliteration · arXiv, published in CORELA (Cognition, Representation, Langage), December 2005