yes. the paper is really just sampling the OCR software's underlying distributions, combined with perl's acceptance of a wide range of things that look like line noise.