Digital Humanities Approaches to a Recipe Book Corpus
If you’re reading this blog post, you’re probably a voracious reader of recipe books. And, if you’re anything like me, you might even be partial to manuscripts. If so, the Folger Shakespeare Library’s holdings—which include the largest collection of early modern English manuscript recipe books in the world—is truly a feast for the mind. But one could spend years digging into this buffet of books, with its indecipherable early modern hands, challenging orthographies, and pages and pages of intriguing discoveries. How then might one quickly consume such a breadth of material? As one solution, I present to you a metaphor to go with this feast—a digestive, if you will. Digital humanities (DH) tools and methods can help an ambitious reader make sense of a lot of text at once—even when that reader has no programming skills—allowing them to detect patterns or isolate the points of interest.
Lucky for us, the Folger has anticipated the needs of the digital recipe book reader. As part of the three-year, Mellon-funded Before ‘Farm to Table’: Early Modern Foodways and Cultures project (2018–2021) and its legacy, the library has aimed to photograph, transcribe, and encode the entire collection. All its recipe manuscripts (with the exception of the most recent acquisitions) are photographed, and of these, 50 are also fully transcribed and encoded, with more in the queue. The encoding makes the recipe books available for study not only as manuscripts but also as digital objects, in the form of lightly edited PDFs (“reading copies,” made for ease of reading) and plain text files. It is the latter that we’ll use for our DH analysis.
Using the Recipe Book Corpus
There are many ways one might use or peruse the Folger’s corpus of recipe books, but here I’ll illustrate a couple of the easiest using Voyant, a free, browser-based software with many language and visualisation tools (Figure 1). All one must do is upload the plain text files, and Voyant will immediately present them with an interface displaying the results of the default tools.
From these default tools, one can dig deeper. Let’s use the “Contexts” tool (bottom right hand) as an example of what kind of analysis is possible. In the search bar, typing a word, such as “digestive” will search across all the uploaded documents for that spelling, listing all the instances and contexts of that term. If one then selects the term, the “Reader” tool (top middle) will display the document in which that spelling appears. Essentially, using these two tools together permits the reader to search across the entire corpus for key terms and immediately read the recipes in which those terms appear. (N.B.: these tools only search spellings, and not words, and so users should expand their search options by trialling different spellings or using wildcard searches for more comprehensive results.) For instance, searching “digestive” found three instances of that spelling across the entire corpus: two in manuscript E.a.5 and one in V.a.260. This search took seconds whereas reading all 50 books might have taken weeks.
Another neat tool is “Cirrus” (top left hand), which creates word clouds of commonly used spellings: the bigger the word, the more common the spelling. Hovering over a word will give the user the number of times that spelling appears in the corpus. To view a word cloud for a specific document, rather than the entire corpus, one can use the “scale” function (found just below the word cloud) and select the manuscript of interest. By comparing resultant word clouds, the user can gain an immediate sense of the focuses of each document. For instance, compare these word clouds from two different manuscripts in the corpus:
The word cloud in Figure 2 —featuring words like “god,” “grace,” and “lord,”—suggests that the manuscript is more likely a commonplace book or miscellany than a receipt book, whereas the word cloud in Figure 3 —featuring words like “butter,” “sugar,” and “salt”—suggests the manuscript is likely a cookery book. Reading word clouds is one way a reader can quickly identify documents that might be of greater interest.
While Voyant is a surprisingly effective tool, it’s important to note its limitations and the quirks of uploading a corpus to a tool not strictly designed for it. For instance, if using Voyant’s default settings, the most frequent spellings that will appear in the corpus’s word cloud are “folio”, “recto”, “verso”, and “page,” not because these are common words appearing in the manuscripts, but because they are common artefacts (encoding tags) of the digital documents. Still, Voyant is flexible enough that creative problem solving can expand its possibilities. For example, to get around the issue of encoding tags cluttering the cloud, the user can modify the stop words list (i.e., the words that Voyant automatically filters out of its analyses and visualizations, which include function words like “the” and “a”) to include these tags so that the remnants of the digital artefacts do not appear.
These examples are but tastes of what is possible with the Folger’s corpus. The simplicity and accessibility of the corpus and DH tools like Voyant make them ideal interdisciplinary teaching resources in addition to effective research tools. There is something for everyone, whether your focus is the big picture or the fine detail. Dig in!