If there’s one keyboard shortcut I know how to use, it’s Ctrl-F. Having been raised in an age of instant access, I’ll admit to sometimes perceiving the ability to locate precise information at my convenience as something of a fundamental right. As online reading and research methods become increasingly more streamlined, it’s hard to appreciate the value of manually searching through text when a search bar can do it for you.
When I first began planning my DH project on the Young Ladies’ Journal [https://www.hastac.org/blogs/jennamly/2015/10/28/reading-digital-humanities-introduction-digitizing-young-ladies-journal], one of my main goals was to create accessible digital content. The majority of inventory for the YLJ consists of Victorian fashion plates and supplements, which are fragile and must be handled with caution. The supplements, which were sent as an attachment with each magazine issue, unfold to about three-by-four feet of newsprint, and are double sided, consisting of a series of fifty to seventy-five images and descriptions for small needlework and craft projects on one side, and one to three full-size clothing patterns on the other that were meant to be cut and traced to sew.
The content on first side is frustrating to search through manually. The 50+ illustrations are placed wherever there is space, and their descriptions – which comprise of two or three sentences about the process and/or materials related to the object of illustration– are isolated in the form of two giant text blocks placed in opposite corners along the bottom of the supplement. This means that in order to draw connections between the illustrations and the text, I have to memorize the illustration number, search for a description that may literally be feet away, and then memorize the description before returning to the illustration and assessing the actual value of the description. I usually have repeat this process multiple times for each illustration, as the time I spend manually searching for each element distracts me from the process of forming observations. While taking inventory for a few illustrations this way may seem tedious at best, taking inventory for 2500 illustrations is pretty much impossible; there simply isn’t enough time to manually record that amount of detail.
This realization is right about where I started missing the all-powerful search bar. While I could eventually upload a picture of the supplement as part of a digital exhibit, anyone who wanted to search through it would still have to click and drag their way across the supplement in the same manner as searching manually. The most efficient way to process the illustrations and text together stems from two options – clicking on the image and having its description appear in a popover, and being able to search through the text blocks in PDF format via OCR recognition [http://www.abbyy.com/finereader/about-ocr/what-is-ocr/].
For those who don’t rely on Ctrl-F as heavily as I do, OCR is the technology that (thankfully) allows readers to search through text in PDF files. While computers are smart, they aren’t as adept at understanding language as they are at recognizing it. When converting printed text (especially the old, ink-stamped typeface of the YLJ), OCR software must process the text as an image, and it does so by translating what it sees as literally as it can into readable text. Computers don’t inherently understand the linguistic and practical value of the letter m or the word blue so much as they recognize them as characters attributed to a certain set of meanings. So, in the case of translating text on the giant supplements, a faded lowercase a might be produced as (|, or a T can become 'l', as it’s the closest instance of text that relates to the visual image of what it sees. Throw in some 1870s newspaper fonts and formatting, and your OCR results can start looking like hieroglyphics:
The end results my OCR scans are making me appreciate, if not the process of manually searching text, the process of manually converting text into something searchable. I have to start each translation by taking a picture of the text blocks, which are more often than not printed on crumpled and fading paper. If I can flatten the newsprint so that a relatively horizontal picture can be taken, I then have to play with the exposure for each individual shot in order to enhance the text. The picture then has to be transferred to Photoshop, where it is cropped and realigned. It then needs to be saved as a PDF that’s compatible with Adobe, which has a tool for recognizing OCR-suspects and converting them into text. If enough of the image converts, I can cross my fingers and open it in Microsoft Word, where I can then edit all of the translation mistakes. Word allows you to write macros [http://techwelkin.com/how-to-write-a-macro-in-ms-word], which can be used to find-and-replace multiple errors without manually searching for them. In order to create macros, however, you have to anticipate what the mistakes will be before they’re actually made. For instance, OCR can reproduce the word “the” correctly nine times out of ten. The tenth time, however, it can reproduce “the” as “t he”…or “th e”…“tlnc” – you just won’t know until you find it.
While this process has its own frustrations, it is ultimately more productive than transcribing the text by hand. Going through the process step-by-step ensures that there are multiple formats and copies of the text, and creates something that can be continually reassessed. I’m definitely gaining a newfound appreciation for the information that is continually accessible to me online. Every subject I search yields a compilation of someone else’s definition of correct. As a reader and a maker, it is up to me to remain vigilant of the work in front of me.