The Sonnet Stretcher

We often think about the placement of a particular word in its context in a poem. Constraints of syntax, meter, and rhyme interact with choices about emphasis. It is more difficult to compare an instance of a word to other instances of the same word across a collection. Which words appear early in poems and which late? Which words occur at the beginning of lines, and which at the end? This work presents a method for uncovering the average positions of words. We consider a collection of 10,000 16th to 20th century English-language sonnets from the Chadwyck Healey
English Poetry collections, whose primary bibliographic source is the
New Cambridge Bibliography of English Literature (1969-72). Although limited to published poems (not manuscript poems or those published in newspapers, broadsides or journals), and skewing towards the canonical, the texts are accurate and offer a sizeable sample of poetry published during this period.

We create a series of two-dimensional views, one for each distinct word. Imagine that each poem has been stretched to a fixed, square shape, with each line fully justified left-to-right. The first word of the first line sits at the top left of the square, and the last word of the last line sits at the bottom right corner. The following figure illustrates this process using Shakespeare Sonnet 91.

Figure 1: For each word, we find the location of all instances of the word in all poems, after stretching all poems to a standard, idealized square.

In the first panel, we highlight all seven instances of the word
some. The second panel shows a reduced view, showing only a symbolic representation of those words. The third panel shows the final aggregate view, with all instances of the word
some from all 10,000 poems, with a slight random jitter to show density. This corpus-level view surfaces patterns:
some appears frequently at the beginning of lines and rather often about two thirds of the way through a line, but rarely at the end. The word occurs slightly more frequently in the second half of poems, but the primary pattern is left-to-right, not top-to-bottom.

The layered visualization reveals surprisingly deep relationships between metrics and meaning. The pattern of
some across the sonnet is both conceptual and functional: as a single-syllable word without a strong set of phonetic or graphic rhymes, the word is less valuable at the end of lines, while its conceptual flexibility allows it to be placed in either a stressed or unstressed position depending on its emphasis. As such, its utility in either establishing iambs or disrupting them (in, for example, trochaic inversions) is equally as likely to place it near the beginning of the line as its measured indeterminancy as either a pronoun or adjective.

The Sonnet Stretcher provides a comprehensive, abstracted view, yet requires few theoretical commitments. Though scholars such as Natalie Houston have performed distant reading on a corpus of poems from the Victorian period, focusing in particular on Elizabeth Barrett Browning in her recent paper “a distant reading of EBB’s rhymes,” her work focuses only on end-rhyme and not the way that certain words might appear more frequently in internal rhymes, at the beginning of poems, or throughout the poem (Houston, 2014). Poemage attempts to show all sonic elements of a single poem in a visual form, and presents a tremendous amount of information that many scholars disagree on, including assumptions about rhymes and metrical “perfection” (as in “a perfect iamb”) (McCurdy et al. 2016). Programs modeling poetic meter are now quite common, from Ryan Heuser’s Prosodic to programs for Russian poetics (Thorsen and Birnbaum, 2017), Medieval High German poetry (Hench and Estes, 2018), English and Italian (Delmonte 2014), Bengali, Sanskrit (N. and Lakhshmanan, 2010).

Figure 2: A sample of selected words illustrating typical patterns.
Scholars have long wrestled with whether, how, and why algorithmic tools and visualizations might aid in the reading of poetry. After all, scansion itself (the series of diacritical marks that appear above a word to indicate a stressed or unstressed syllable in traditional accentual-syllabic scansion) is a form of visualization. But visualization tools can only do so much with prosody; prosody is pronunciation and versification, and though there are rules to both pronunciation and versification there are also hundreds of variants and exceptions to those, not to mention disagreements. It is for this reason that the carefully rendered visualizations of these word positions in The Sonnet Stretcher are so evocative. The Sonnet Stretcher returns a crucial visual aspect of poetry frequently lost in the digitized reproduction of the text: the position of the words on the page. From Herbert’s “Easter Wings” to Plath’s “Bell Jar,” the poem’s spatial arrangement – what words can appear where – remains a crucial axis of meaning. Not only do these visualizations appear as conglomerate poetic texts in their own right, creating a new poetic object for observation and critical commentary, but they both confirm and complicate several commonplaces about rhyme and metricality by the very fact of this accumulative effect. That is, scholars of poetry take for granted the ways that their reading practices, shaped by education and the memorization of the very rules we mention above, impact or bias their approaches to a poetic text. With this tool, scholars are evocatively challenged to think through these assumptions.
The visualization reveals both presence and absence. We are not surprised to see that
apart,
start, and
heart commonly appear as end-rhymes, with
heart appearing more often throughout the line than the other two. But what of
summer, which unlike
winter, begins in the second and penultimate positions most often (whereas
winter,
spring, and
fall are resolutely at the end);
summer seems to have overtaken
spring as having both a beginning and an end. The evocative emptiness of
farther begs to be read as a text on its own right, showing a constellation of positions with no adherence to one or the other, much like
floating, which hovers across the cumulative sonnet image. Similarly,
rich texturally imitates the word’s meaning in its stretched form. The interpretive possibilities for reading the visualizations are myriad, and this paper will seek to think through what we might learn from position outside of the realm of rhyme (though of course rhyming-end words are impossible to ignore).

Bibliography

Houston, N. (2014)
Toward a Computational Analysis of Victorian Poetics. Victorian Studies, vol. 56, no. 3, pp. 498–510.

McCurdy, N., Lein, J., Coles, K., Meyer, M.. (2016)
Poemage: Visualizing the Sonic Topology of a Poem. IEEE Transactions on Visualization and Computer Graphics (Proceedings of InfoVis 2015), pp. 439-448, January.

Thorsen, E., Birnbaum, D. (2017)
The computationally assisted analysis of formal features in Russian poetry. http://poetry.obdurodon.org/

Hench, C., Estes, A. (2018)
A Metrical Analysis of Medieval German Poetry Using Supervised Learning". Front. Digit. Humanit., 18 July

N. R., Lakshmanan, M. (2010)
A Computational Algorithm based on Empirical Analysis, that Composes Sanskrit Poetry. IJCSIS, vol. 7, no. 2, , pp. 56-62.

Delmonte, R. (2014)
A Computational Approach to Poetic Structure, Rhythm and Rhyme.