The Shame of Bad Digitizations
In a recent thread on the Govdoc-l mailing list about a Congressional Publications Hub (or “pub hub” — more of the thread here), one commenter said that The American Memory project’s digital surrogates of the pre-Congressional Record publications "probably aren’t salvageable" because the TIFFs were captured at 300 ppi resolution and then converted to 2-bit bitonal black and white and that most of the text is too faded or pixelated to be accurately interpreted by optical character recognition (OCR) software. He concluded that this was "Kind of a shame."
It is indeed a "shame" that many of the American Memory Project’s "digital surrogates" probably are not salvageable. But the real shame is that we keep making the same mistakes with the same bad assumptions today that we did 10-15 years ago in regard to digitization projects.
The mistake we keep making is thinking that we’ve learned our lesson and are doing things correctly today, that our next digitizations will serve future users better than our last digitizations serve current users. We are making a series of bad assumptions.
- We assume, because today’s digitization technologies are so much better than yesterday’s technologies, that today’s digitizations will not become tomorrow’s obsolete, unsalvageable rejects.
- We assume, because we have good guidelines (like Federal Agencies Digital Guidelines Initiative (FADGI)) for digitization, that the digitizations we make today will be the "best" by conforming to the guidelines.
- We assume, because we have experience of making "bad" digitizations, that we will not make those mistakes any more and will only make "good" digitizations.
Why are these assumptions wrong?
- Yes, digitization technologies have improved a lot, but that does not mean that they will stop improving. We will, inevitably, have new digitization techniques tomorrow that we do not have today. That means that, in the future, when we look back at the digitizations we are doing today, we will once again marvel at the primitive technologies and wish we had better digitizations.
- Yes, we have good guidelines for digitization but we overlook the fact that they are just guidelines not guarantees of perfection, or even guarantees of future usability. Those guidelines offer a range of options for different starting points (e.g., different kinds of originals: color vs. B&W, images vs. text, old paper vs. new paper, etc.) and different end-purposes (e.g., page-images and OCR require different specs) and for different users and uses (e.g. searching vs reading, reading vs. computational analysis). There is no "best" digitization format. There is only a guideline for matching a given corpus with a given purpose and, particularly in mass-digitization projects, the given corpus is not uniform and the end-point purpose is either unspecified or vague. And, too often, mass-digitization projects are compelled to choose a less-than-ideal, one-size-does-not-fit-all, compromise standard in order to meet the demands of budget constraints rather than the ideals of the "best" digitization.
- Yes, we have experiences of past "bad" digitizations so that we could, theoretically, avoid making the same mistakes, but we overlook the fact that use-cases change over time, users become more sophisticated, user-technologies advance and improve. We try to avoid making past mistakes, but, in doing so, we make new mistakes. Mass digitization projects seldom "look forward" to future uses. They too often "look backward" to old models of use — to page-images and flawed OCR — because those are improvements over the past, not advances for the future. But those decisions are only "improvements" when we compare them to print — or more accurately, comparing physical access to and distribution of print vs digital access to and distribution over the Internet. When we compare those choices to future needs, they look like bad choices: page-images that are useless on higher-definition displays or smaller, hand-held devices; OCR that is inaccurate and misleading; digital text that loses the meaning imparted by layout and structure of the original presentation; digital text that lacks markup for repurposing; and digital objects that lack fine-grained markup and metadata that are necessary for accurate and precise search results finer than volume or page level. (There are good examples of digitization projects that make the right decisions, but these are mostly small, specialized projects; mass digitization projects rarely if ever make the right decisions.) Worse, we compound previous mistakes when we digitize microfilm copies of paper originals thus carrying over limitations from the last-generation technology.
So, yes, it is a shame that we have bad digitizations now. But not just in the sense of regrettable or unfortunate. More in the sense of humiliating and shameful. The real "shame" is that FDLP libraries are accepting the GPO Regional Discard policy that will result in fewer paper copies. That means fewer copies to consult when bad digitizations are inadequate, incomplete, or unusable as "surrogates"; and fewer copies to use for re-digitization when the bad digitizations fail to meet evolving requirements of users.
We could, of course, rely on the private sector (which understands the value of acquiring and building digital collections) for future access. We do this to save the expense of digitizing well and acquiring and building our own public domain digital collections. But by doing so, we do not save money in the long-run; we merely lock our libraries into the perpetual tradeoff of paying every year for subscription access or losing access.