wondarchive

wondarchive › Transcription accuracy

How far can you trust an automatic transcription?

Very good on a sharp photo: wondarchive’s transcription is excellent on typewritten text, print and Fraktur, and its extra-strength reread reads handwriting to a high standard, old cursive included. The rare errors are mostly in proper names and numbers, which you check at a glance against the photo shown next to the text.

What makes a good transcription?

First the photo: sharp, flat, the whole page, under even light (the photography guide). Then the model: each page is read from the full-resolution original, never from a reduced copy, and a handwritten page goes through an extra-strength reread with our most precise model. At maximum quality, every page of the upload goes through it.

Where do errors creep in?

  • proper names and rare surnames, whose spelling varies from one document to the next;
  • figures, amounts and dates in a dense table;
  • words split at the end of a line or caught in the gutter of a double page;
  • blurred, shadowed or stained areas;
  • old abbreviations.

How do you check a transcription quickly?

  1. Read in two panes: the photo on the left, the text on the right.
  2. Check first what you will cite: names, dates, figures.
  3. Search for the variants of a name: the index brings spellings together and offers to merge them.
  4. Correct in the pane: the machine’s version stays available, and a button restores it.
  5. Rerun the transcription of a page whose photo has been retaken.

What does wondarchive refuse to invent?

  • a text on a blank verso or a near-empty page: it is recognised as such;
  • a certain date: an estimated date is marked as such, a doubtful date flagged;
  • a machine correction over yours: what you type is never overwritten;
  • a verified record: it stays “unverified” until you have reviewed it.

Do you need to proofread every page?

For working through and searching a collection, no: fragment search and the name index, which brings spellings together, find a name even when a letter has been misread. For citing, you check the quoted passage against the photo, as you would with a human transcription.

The account takes a minute to open, a hundred pages free for a year.

Is automatic transcription as good as human transcription?

On sharp typewritten and printed text, it comes very close, in far less time; on handwriting, the extra-strength reread gives a high-quality text. As with a human transcription, you check what you cite against the original.

Is handwriting read well?

Yes: the extra-strength reread reads cursive handwriting in European languages, in the Latin, Cyrillic and Greek alphabets. Judge on your own pages: 100 free pages, no credit card.

What do you do with a badly read page?

Correct it in the pane, or rerun it with one button after retaking the photo; maximum quality sends every page of an upload through the extra-strength reread.

Is the translation just as reliable?

It starts from the transcription, page by page, with the context of neighbouring pages and a collection glossary that keeps the same translation of a term from start to finish; the original text stays displayed alongside.

See it on your own pages

A hundred pages transcribed and translated, for a year, no card required.