Jewish Genealogy Name Variants: One Person, Many Spellings
Written for genealogists, archivists who build name indexes, and researchers working with Jewish and Eastern European records.
The same man can be Moshe on his gravestone, Movsha in a Russian metrical book, Moszek in a Polish civil register, Moses on a passenger list and Morris on a census form a few years later. To a search engine these are five different people. To a genealogist they are one person who has not yet been found in four of his records.
The working rule is simple: search for the person, not for the spelling. The overview of AI in genealogy covers the wider field. This post explains where the variants come from, which tools find them, and where AI has to stop.
Why one name becomes many
Script and spelling conventions
Hebrew script usually leaves out vowels, so a name written in Hebrew letters can be read in more than one way. Yiddish writes its vowels, but spelling varied widely before it was standardised. Then every clerk rendered what he heard in the orthography of his own state:
- The sh sound: German sch, Polish sz, Russian ш, English sh. One surname becomes Schwarz, Szwarc, Шварц and Shvarts.
- The ts sound: German z, Polish c, Russian ц, English tz or ts.
- The kh sound: ch in German and Polish, х in Russian, kh or h in English. Chaim, Хаим, Khaim and Haim are the same name.
- Russian has no h, so Hebrew and Yiddish h is usually written with г: Hirsh becomes Girsh, and Cohen becomes Kogan or Kagan.
Polish registers add their own forms of given names, such as Icek for Yitzchak and Rywka for Rivka. And when records are read by machine, recognition errors add one more layer of variants, which is why the quality of the text layer matters, as the Yiddish handwriting OCR guide discusses.
Hebrew, Yiddish and civil names
Many Jews in Eastern Europe carried more than one name at once. A Hebrew name was used in religious contexts: being called to the Torah, the marriage contract, the gravestone. A Yiddish name was used every day. A civil name appeared in state records.
Often these were linked by a kinnui, a customary secular name paired with a Hebrew one, frequently with the same meaning: Tzvi Hirsh (both "deer"), Dov Ber ("bear"), Aryeh Leib ("lion"), Ze'ev Wolf ("wolf"). A record may use both halves, only the Hebrew one, or only the Yiddish one.
Patronymics and "ben" and "bat"
Russian Imperial records usually identify a person with the father's name as a patronymic. Hebrew records and gravestones use the form Moshe ben Yitzchak, Rivka bat Avraham. The father's name is often the most useful piece of evidence you have, because it separates two men of the same name in the same town.
Fixed hereditary surnames came relatively late to many Jewish communities in the region, so in the earliest records a patronymic may be the only identifier.
Women's names
A woman may be recorded under her maiden name, her married name, or only as "wife of" her husband. In Galicia, many couples married religiously without civil registration, so children could appear in civil records under their mother's surname. A search that looks only for the father's surname misses them.
Emigration and Hebraization
Names changed after emigration, usually by the immigrants themselves, over years. The popular story that officials at Ellis Island changed names does not fit how passenger lists were made; they were compiled at the port of departure.
In Israel, and in the Jewish community of Palestine before 1948, many immigrants Hebraized their surnames. David Grün became David Ben-Gurion. A Hebraized name written in Hebrew letters then has to be transliterated again to find it in foreign records.
Places
Borders moved, and each state wrote towns in its own language. Vilnius is also Wilno, Vilna, Вильна and, in Yiddish, Vilne. Lviv is also Lwów, Lemberg and Львов. A record uses the name that was official for whoever kept it, in that year. Gazetteers such as the one on JewishGen list alternative names and help place a town on a modern map.
Tools: phonetic codes and fuzzy matching
Daitch-Mokotoff Soundex was created in 1985 by the Jewish genealogists Gary Mokotoff and Randy Daitch as a refinement of the older American Soundex, which handled Slavic and Yiddish surnames poorly. It turns a surname into a six-digit code so that names which sound alike share a code. Where a letter combination can be pronounced in two ways, a name receives more than one code. JewishGen uses it in its database searches.
Beider-Morse Phonetic Matching, developed by Alexander Beider and Stephen P. Morse, takes a different approach and aims to reduce false positives, at the cost of missing some true matches.
Fuzzy matching measures how many characters differ between two strings. It catches a misread letter or a clerk's slip; phonetic codes catch what it misses, such as Schwarz and Szwarc, which look different but sound the same. Wildcards cover endings that vary.
None of these tools crosses scripts on its own. Phonetic codes work on Latin letters. Records in Hebrew or Cyrillic need either a search in the original script or a transliteration first.
A workflow that works
- Build a variant list for each person. Hebrew, Yiddish and civil given names; the kinnui pair; the surname in each script and each spelling convention; the father's name; the spouse's names, including the wife's maiden name; every name of every town.
- Search wide. Use phonetic codes, wildcards and native-script search in every database that covers the region.
- Log each search. Where you searched, with which form, and what came back. A documented negative result saves repeating it.
- Verify each hit against context. Dates, allowing for ages that were rounded or estimated; parents' names; place; spouse; occupation. Accept a match only when several independent details agree.
- Keep the rejected candidates with the reason they were rejected. They are evidence too.
Where AI helps
- Proposing variants and transliterations. A language model can suggest spellings across scripts and conventions, including forms that are easy to forget, such as the Russian g for h. Treat its list as things to search, not as facts about your ancestor.
- Reading the records. Handwriting recognition turns scanned registers into searchable text in the original script.
- Clustering records. Entity resolution groups records that may describe the same person into candidate clusters, each with a confidence score and the reasons behind it: same father's name, same town, consistent age. The post on knowledge graphs explains how this works at archive scale.
Where AI must not decide alone
Merging two people is a human decision. In one town there may be several men with the same name and father's name, born within a few years of each other. A false merge contaminates every tree and index built on it, and it is harder to detect than a missed link. The system proposes; a person decides, and the decision is logged and reversible.
Holocaust-era identifications need particular care. Linking a victim record, a testimony or a photograph to a specific family carries weight for the people involved. The discipline described in the post on cataloguing Holocaust archives applies here: a name enters the record only with verifiable evidence. And because such records may name living relatives, it matters where they are processed, as the post on sensitive archives explains.
For archives that build name indexes
- Store the name exactly as written, in the original script.
- Store a transliteration, and record which system was used.
- Store a normalised search form and phonetic codes, generated by rule, not typed by hand.
- Never overwrite the original. Transliterations and normalised forms can be regenerated; the original cannot.
- Keep relationships: father, mother, spouse, and the role of the person in the record.
- Keep the place name as written, plus a standard identifier for the place.
- Mark machine-generated fields and links as such, with confidence, and record who confirmed each link.
A short checklist for genealogists
- A variant list per person, in every script and spelling convention.
- Hebrew, Yiddish and civil names, and the kinnui pair.
- Father's name and the wife's maiden name on the list.
- All historical names of every town.
- Phonetic codes, fuzzy search and wildcards.
- A search log, including negative results.
- Several matching details before accepting a record.
- No merges on AI confidence alone.
If you are building a name index from registers, community record books or card files and want each name kept in its original form while still being findable in all its variants, see how we approach AI-assisted archive cataloguing, or get in touch with a sample of your records.
Frequently Asked Questions
Why can't I find my ancestor under the spelling my family uses?
Because the family spelling is usually the last one in a long chain. Each record was written by a clerk in his own language and alphabet, often by ear, and the person may have used a Hebrew, a Yiddish and a civil name at different times. Search for the whole variant list, not one spelling.
What is Daitch-Mokotoff Soundex?
A phonetic coding system created in 1985 by the Jewish genealogists Gary Mokotoff and Randy Daitch. It turns a surname into six-digit codes so that spellings which sound alike are found together, and it was designed to handle Slavic and Yiddish surnames better than the older American Soundex. JewishGen uses it in its database searches.
Can AI decide whether two records are the same person?
It can propose it, with a confidence score and the evidence behind it, and that saves a great deal of searching. The decision to merge two records into one person should stay with a human, and for Holocaust-era identifications it must.
How should an archive store names in a name index?
Keep three forms side by side: the name exactly as written in the original script, a transliteration with the system named, and a normalised search form. The original is never overwritten; the other two can be regenerated when rules improve.