April 22nd, 2012
Yes, it would seem so. Or at least, with the help of a team of brilliant researchers made up of humanities and digital technologies scholars, computers can at least be asked to try and differentiate Gertrude Stein texts from each other, from other highly sophisticated western literary works, and from a cookbook. The punchline? One such computer consistently confused Stein’s Tender Buttons with The New England Cookbook, which was highly popular during Stein’s day. While this may not seem like much of an accomplishment to the average humanities scholar (indeed, as someone on the verge of PhD Qualifying Exams, I’d be quite pleased if this were my qualifying task, instead), it represents a significant accomplishment in the digital humanities, and provides compelling evidence for the ongoing use value of computing in the study of literature.
This brief anecdote was a small portion of a talk given several weeks ago by Dr. Tanya Clement, of UT Austin, to the Critical Digital Humanities Research Group at UCR. Sponsored in part by The Center for Ideas and Society Mellon Workshop, the event was entitled “Sounding it Out: Modeling Aurality for Large-Scale Text Collection Analysis.” Dr. Clement presented research from her collaboration with a computing center at UT Austin, a project that creates computer models for analyzing sound in literary texts. It emerged several years ago out of a drive to expand the affordances of digital archives to include models that could recognize and work through texts based on the way we read them, not just based on the ways we’re currently able to upload and store them (images, text, etc). In this case, the analysis sought to understand how we might argue for the meaning of sound in literary works—not simply sound as part of form or delivery, but sound as meaning.
The project leaders created a set of protocols that would allow them to come up with a usable “knowledge representation” of texts based on sound, even taking into consideration the subjective nature of sound. For example, different readers will read the same text quite differently, based on numerous factors including regional dialect, speed, intonation, enunciation, etc. A central question then became how to make the best guess at how a text actually “sounds,” given these considerations.
Dr. Clement and the research team selected as their primary text, which they then put through a series of digital translations, from text to code to representation to “best guess at sound.” Now, this is obviously a serious reduction and is admittedly a paraphrase, so please forgive my skipping over most of the technical details, as it’s quite elaborate and I probably wouldn’t rehash it very faithfully anyways. I can say this much, however—that the poems were cycled first through a program called OpenMary, then SEASR, and then through ProseVis (someone out there (myself not included) is reading this and knows exactly what this means, and is probably excited about it—I hope so, at least). The end result of this process was a series of data maps that allowed not only some fascinating analysis (which I’ll detail shortly), but which also functioned for Dr. Clement and team as “hypothesis generators,” which allowed them to ask questions such as “Do we care about sound in this poem? Does it matter to meaning?”
So I’ll skip now to flesh out the opening anecdote. The researchers entered 9 texts into the database, including Tender Buttons and two other Stein texts, plus the cookbook, some Joyce, the Iliad and the Odyssey, and more. Just to try to make sure the computer focused on sound, and not just noting the repetition of words, they eliminated words themselves from the analysis, focusing on data about the words and phrases, such as part of speech, accented syllables, phoneme, tone, and more. What they found was pretty remarkable.
Stein scholars have charged that Tender Buttons was written to reflect Stein’s domestic pleasures shared with Alice B. Toklas, her partner of many years. Thus, it’s been argued, Stein modeled the poems on a cookbook. As it turns out, when the computer was asked to recognize which phrases and patterns belonged to which of the 9 texts, it identified them correctly with a fair degree of accuracy. However, the text that it confused most frequently for Tender Buttons was The New England Cookbook. It also regularly confused the Iliad and the Odyssey for each other, which should indicate the relatively sound (ahem) results.
In any case, what all this may mean is that sound does mean. In other words, Stein wrote Tender Buttons not merely to evoke a cookbook through food references and word choice, but she made her poems sound like the recipes. As if we needed another testament to the genius that Stein herself so often and roundly proclaimed.