Every Spelling of One Name

One query that would have done the work of twenty-five

This whole table is a list of guesses about how a clerk might have spelled a name, and the site it was measured on does not need the guesses. The surname box takes wildcards. `SNG*` asked of Gloucestershire marriages for the three years 1827-1829 returns 53 rows, and in them:

Thomas Sanigre, 1828, Martha, Bristol — the marriage this archive searched for under eight spellings and recorded as not existing. And John Saniger, 1827, Ann, Rodborough standing beside John Sanigor, 1827, Ann, Rodborough — one wedding indexed under two forms, one of them the rare spelling that had already hidden Hannah's baptism for two searches.

It is not a free win. The same 53 rows hold Sanger, Shearing and Sheering, and the same pattern run nationally across the whole baptism set would return hundreds of strangers. The wildcard buys completeness and spends it on noise, so it works where a question has a place and a decade and fails where it has neither. That is the opposite trade from this table, which asks twenty-five narrow questions across four centuries. Both are worth having; this archive had only one of them.

And it found two spellings this table did not have. The same 122 rows hold SINIGOR — Sarah, Stinchcombe, 1801 — and SINNIGER, the same marriage indexed a second way. Put back to the baptism set on the table's own basis, SINNIGER returns two: Margrett at Slimbridge in 1629 and Thomas, son of Abraham, at Dursley in 1720 — Abraham being the scarp patriarch this page already names. SINIGOR returns nought, like SANIGRE. Eleven years of one county, one query, two forms missing from a table that had just been rebuilt and called complete. Both are in it now, and the honest reading is that this table is a floor and not a ceiling.

And the trade is not the same for every surname, which is worth saying before anyone takes this away as a rule. The Defranceski session ran the same method on an Italian surname the same afternoon: the exact spelling returns 5 in the England and Wales indexes, and the wildcard `DEFRANC` returns 608 — almost all of them de France, de Franca, Barker or Defrance*, a Huguenot-derived English surname with no connection to theirs. Their signal was in the exact five, and it would not have been easier to see among six hundred.

The variable is whether the stem is native to the archive's language. Snig- is rare in England, so the noise here stays readable: 122 rows for eleven years of one county, six of them this family. Defranc- in England is everywhere. The method is the same and the arithmetic is not, so the honest form of the rule is: use the wildcard where the stem is foreign to the records, and use it inside a narrow window in any case.

And there is a way to know which case you are in, which costs one query. The rule as first written here — use the wildcard where the stem is foreign to the records — asks the searcher to judge something they usually cannot. Measure it instead. Run the exact spelling first, then the wildcard, and read the ratio. Here, `SNG` over eleven years of one county returns 122 rows of which six are this family: one in twenty, and the other nineteen are legible in the result set and can be discarded by reading. The Defranceski archive's is one in a hundred and twenty, because Defranc-* is a native English stem and theirs is not an English family.

The governing variable is the rarity of the thing sought in the corpus it is sought in, and the instrument does not know that number. Neither does a spelling list. The difference is that the wildcard can be made to report it, for the price of one extra search, before anything is built on what comes back.

documented

One of these rows is a register and the rest are an index

the disagreement is documented; neither spelling is

And one row in this table does not begin with an S

documented