The term “accessibility” in a DH or cultural heritage context most often refers to the ease of discovery of digital content by sighted people. Such content includes scholarly editions, digitized images of original sources, and datasets derived from cultural heritage materials, to name a few types. Transcription of images of handwritten manuscripts, such as letters, diaries, and other documents, and subsequent encoding efforts with TEI or other markup schemes, are steps frequently understood as enabling machine-readability and reuse. Accessibility in the DH context is less often used to mean discovery by and legibility to people who are blind or have low vision and use assistive technologies such as screen readers and braille devices to access digital content. We argue that mark-up for this kind of machine readability requires greater consideration by the DH community.
This short paper presents findings from “Crowdsourced Data: Accuracy, Accessibility, and Authority (CDAAA)”, a 3-year IMLS-grant-funded project that investigates the sociotechnical barriers that Libraries, Archives, and Museums (LAMs) and community groups face in making crowdsourced transcriptions discoverable and accessible to blind and low vision users. 1 Crowdsourced transcriptions are the most common data solicited by LAMs,because they theoretically provide widened discovery and access for sighted and blind users alike when integrated into content management systems such as Islandora, Internet Archive, or data repositories like Zenodo or Dataverse.
A core hypothesis of CDAAA was that LAMs experience significant challenges in integrating crowdsourced data into content management systems. We have found this to be less problematic for 9 of our 12 USA-based LAM Partners than anticipated and were able to test the accessibility of transcriptions in 9 systems including college and university libraries, community groups, a state museum, and two large federal repositories. Additionally, we tested several texts with altered markup schemes from those applied in crowdsourced transcription or by subsequent editing by LAM Partners. This abstract relates to findings for one altered text, which falls into one of three CDAAA research question areas: RQ3, “When crowdsourced transcription data is successfully integrated into a CMS or database, is it accessible to people who are print-disabled, e.g. blind or low vision, and use assistive technology? If not, what is required to make the data legible? What are print-disabled users’ experiences of searching for and reading transcription data?”
To test RQ3 we recruited 12 print-disabled testers (11 blind people and one person with low vision) who use a range of assistive technologies including Job Access With Speech (JAWS), NonVisual Desktop Access (NVDA), and Zhengdu (ZDSR) screen readers, Mantis40 braille keyboards, magnifiers, and/or speech to text tools. Each participant tested 3 LAM platforms in a randomized order. Their tasks included: to seek each LAM site through a web-based search; to search for transcriptions in general within the LAM site; and then to search for a specific line of text from a transcription we knew to be in each system. These findings will be published in future.
All 12 blind or low-vision CDAAA testers have or are pursuing degrees in higher education across a range of subjects, and many use primary sources and advanced resources for their work. Each interview took place on Zoom, and participants used their preferred devices, assistive technologies, operating systems, and other settings. None had knowingly encountered crowdsourced transcriptions prior to participation in CDAAA, and none had deep familiarity with the scholarly conventions that govern many scholarly editions and DH projects.
In addition to the randomized LAM system testing, 11 of the 12 accessibility testers were asked to read and comment upon a page from the diary of Ada Blackjack (Iñupiat, 1898-1983) , the sole survivor of the Wrangel Island expedition. 1 participant’s testing session ran too long to include the Blackjack test. The diary is part of the Dartmouth College Library special collections and has been digitized along with related items from the expedition. Blackjack’s diary was transcribed by volunteers through the FromThePage crowdsourced transcription platform and integrated into the Dartmouth Libraries CMS, where it is discoverable through text string search and other methods. 3 users tested the Dartmouth system, and 11 tested an additional page of text that we rendered into simplified markup that differs from what is available on the Dartmouth Libraries site.
Early inspection of the Blackjack transcriptions and other texts selected for randomized controlled testing for RQ3, suggested that users of assistive technologies would encounter issues stemming from the markup applied to crowdsourced transcriptions. For instance, many transcriptions and subsequent TEI encoding efforts preserve how words were written in the original, i.e. preserving original spelling, punctuation, words broken over two lines, deletions, and insertions. After testing multiple systems and a simplified markup of the Blackjack diary, one participant remarked: “If you want to be faithful to the text, that should be indicated. And that's important. But without an explanation… it didn't make sense.” [ACC-6] None of our LAM Partners had provided transcription convention information at the time of testing, so users were often left to feel or sound out transcriptions with their preferred devices and relied on their existing knowledge and mental models to parse unfamiliar documents from the seventeenth-century to the twentieth. Users of braille devices frequently switched from their screen readers to braille in order to feel each letter, and the markup applied for special characters, deletions or insertions. Several participants likened deletion and insertion markup to tracked changes in word processing programs.
Ada Blackjack’s diary contains nonstandard spelling and punctuation, which when combined with the somewhat unconventional use of a span rather than break tags at the ends of each line caused further confusion. So, “I Caught a fox today the fox has trap <span class="line-break"> </span>,” causes words at the end of the line to run together with the word at the start of the next line for most screen readers. One participant was unsure whether this was a feature of the original text and authorial style or an issue with the online presentation: “Some words are run together and some aren't. There doesn't seem to be a consistent pattern inside of the text” [ACC4] To address this issue, we converted the CSS line break spans into semantic line breaks, i.e. “I Caught a fox today the fox has trap<br />” resulting in a clean line break. We also modified TEI spans for deletions and insertions and substituted semantic deletion and insertion HTML tags, as in the diary entry when Blackjack deleted the word “five” and inserted “six” we converted <span class="tei_del"> five</span> and <span class="tei_added_text"> six</span> to <del>five</del> to <ins>six</ins>. These relatively small interventions significantly increased the readability of the test page. As one participant put it “Having marks this way is more helpful than the original format that was presented.” [ACC-9]
Testers appreciated the use of headings as applied by the Dartmouth Libraries team and preserved by team CDAAA, because headings are easily navigable landmarks for most blind screen reader users. The heading names were not always clear, however. Several people were unsure what “unpaginated” meant, and others interpreted the dated headings i.e. “28” and “29th” as separate pages rather than separate entries. The short lines prompted several testers to wonder if the diary was a form of poetry. Some asked questions about the genre, style, and wanted information from us as sighted users about the differences between Blackjack’s original document and the transcription or further textual edits provided by volunteers and practitioners. A brief description of the conventions and the genre or content of a transcribed text would provide significant value to blind and low vision users. Additional information about materiality–i.e. document layout, color, signs of use or wear would also be welcome by some users. ACC-11 stated "It's good to have that piece of information [on materiality] ... It adds some kind of a human touch. And I would honestly like to have access to that information just because sighted people can have access to that information.”
The use of relatively simple markup such as standard semantic HTML can significantly improve the accessibility of transcribed texts and online scholarly editions. Based on the Blackjack case study, we believe that greater accessibility is in reach for many practitioners and organizations seeking to make their collections semi-diplomatic, and regularized markup for digital editions, DH project teams and LAM practitioners that make transcriptions available in online repositories should consider implementing a lightly encoded accessible version of transcriptions, including headings, simple tags, and words that do not break over lines or pages. A brief description of the conventions used, and, in an ideal world, a material description of the original content, would go a long way to rendering the cultural heritage and digital humanities landscapes more navigable and usable by people who are blind or have low vision.
V. Van Hyning, “Crowdsourced Data: Accuracy, Accessibility, and Authority (CDAAA),” (2022-2025) https://github.com/VVH/CDAAA ; Institute of Museum and Library Services, early career grant, RE-252344-OLS-22.