Method

How the text of the scans was made and processed. The original OCR was never overwritten; every automatic change is recorded.

1. Dividing into households

version 2026-11-04.1 · 2026-11-04

Households were determined from the OCR lines. A household ends at a blank line, at a line clearly shorter than the full line width that ends in an age (the typewriter gives every letter the same space, so line length equals width), or when the next line starts with a word that opens a household (colonus, vidua, Kötter, incolae, …). Never after a line ending in a comma or hyphen. Folio and letter headings are skipped. For volumes with one person per line (column OCR from step 3) a household starts at a blank line or at a position that is not a family member or servant (occupation, Einwohner, colonus, vidua …); lines containing only noise are removed first.

Counts

Measured

Checked on two random pages without blank lines: Warendorf (SAA_WadM, page 13) all boundaries in the checked part correct; Greven (SAA_GreM, page 9) all but a few correct. Known weakness: a household whose last line happens to fill the full width is not separated from the next. List volumes (after step 3) checked against the scan: Sendenhorst AS_019 16 of 19 households correct; Osterwick AS_009, where the shadow of the spine caused much noise, about 7 of 10; Harsewinkel and Beckum correct in the parts checked.

2. OCR correction

version 2026-11-04.1 · 2026-11-04

Automatic correction in three layers; the original OCR is always kept. (1) Name rules: systematic errors at the start of common names, mostly caused by capital letters typed slightly raised ('laria' → 'Maria', 'Pernardus' → 'Bernardus', 'enricus' → 'Henricus'); see the table below. (2) One character: a rare word is replaced by a much more frequent one if the difference is a known OCR confusion; e↔a deliberately excluded because of genuine variants (Anne/Anna). (3) Stray characters (|, », ^) removed and digit errors fixed (4C → 40, ll → 11).

Counts

Measured

On two pages transcribed by hand: Sendenhorst AS_019 from 80% to 87% correct words, Ahaus AS_001 from 92% to 94%. In a random check of 70 name-rule corrections almost all were right; of 70 one-character corrections about one in seven was doubtful or wrong (e.g. 'GroBe' → 'Grote' where 'Große' was meant). Spellings of the typist can also be 'corrected' ('Hennricus' → 'Henricus').

3. Column OCR of the list volumes

version 2026-11-04.1 · 2026-11-04

Volumes with one person per line were re-read column by column: columns are found from vertical white space, each column is read separately, and the lines are reassembled by their height on the page. The column with the most text is the name, to its left the position in the household, the first column to the right with mostly numbers the age, the rest remarks. Two lists side by side (or a double page) are read as two blocks. A household starts at a blank line or at a position that is not a family member or servant (occupation, Einwohner, colonus, …). Lines containing only noise (for instance from the shadow of the spine) are removed first.

Counts

Measured

Tested on 8 sample pages from different list volumes; columns recognised correctly on 7 (Lüdinghausen page 4 has a different structure). On a page from Sendenhorst transcribed by hand (AS_019): 150 of 183 words in the name and age columns correct, and the position column (Ehefrau, Kinder, …) largely readable, where ordinary OCR made it unusable.

4. Suggestions by users

version 2026-11-05 · 2026-11-05

Corrections sent in by visitors through the form, taken from the public tab of the Google Sheet. They are shown next to the text as a user suggestion and have not been checked; the original OCR and the automatic correction stay unchanged. For each household the newest suggestion applies.

Counts

Name rules

Systematic OCR errors in common names and what they were corrected to, with the number of times each rule was applied.