I'm inclined to think that the 1/3 figure, AT MOST, may be closer to the truth. I've been working on a massive (300,000 word) publication for a number of years now. (It's FINALLY in pre-press, hooray!) My workflow was: handkey text (except for 16 pages I OCRd as a test.) Proofread 1 Attestation* 1 Proofread 2 Attestation* 2 Skimread (very superficial, but often you find stupid errors that way. Like the one on PAGE 1!!!) Software spellcheck Proofread 3 Readback** 1 Readback 2 And IIRC, Readback 3 * Attestation: Comparing my typescript to the original, word-by-word, phrase-by-phrase. ** Readback: After the HUGE error rates I was still getting after each prevous pass, I bought voice synthesis software, and had the work read back to me, while I followed along in the original. I've kept stats somewhere on the error catch-rate at each stage; I'll dig them up later. The caveat, of course, is that the only way for me to get "fresh eyes" on the project was to put it aside for a few weeks or months; I can't afford to hire someone else. The error rate on the last pass was so small that, even if I had only caught 30% of the remaining errors, the few that are statistically expectable are no longer worth it on the law of diminishing returns curve. ----- Original Message -----

From Michael Hart <hart@pglaf.org> Date Fri, 28 Jan 2005 10:06:10 -0800 (PST) Subject [gutvol-d] Error Correction Data Needed

[Please excuse cross-posting.] However, my most recent research, in conjunctions with the head of error correction at a major publisher, leads me to think 1/3 of errors might be found per pass, instead of the previous 1/2. If any of you have any suggestions as to what these figures are, please let me know.