Sentiment analysis of the volta in Shakespeare’s sonnets

Abstract

This article examines William Shakespeare’s sonnet form by using both qualitative and quantitative methods, particularly with the aid of digital technologies such as sentiment analysis. It focuses on the volta, or “turn,” in Shakespeare’s sonnets, exploring its place, degree, and characteristics in a more methodologically enriching manner than extant criticisms do. While most critics have analyzed the volta through a qualitative approach (i.e. a close reading of poems), I consider the volta as a statistical problem of identifying a turning point in a sequence of fourteen values—in this case, sentiment scores assigned to each line in a sonnet. Furthermore, this study enhances the results of quantitative and computational analysis with a qualitative analysis of the sonnets’ structure and emotions, incorporating the human-in-the-loop approach. By integrating digital methods with close reading practices, this article both confirms and complicates a standard view that the volta in Shakespeare’s sonnets tends to occur before the final couplet. In addition, this study offers new insights into the function of the Shakespearean volta, expanding the focus of formalist inquiry from where it occurs to how it works: the volta induces not so much a line-to-line transition as a section-to-section transition.

1. Introduction

This article examines the poetic form of William Shakespeare’s sonnets by using both qualitative and quantitative methods, especially with the aid of digital technologies such as sentiment analysis. The sonnet is a fourteen-line poem with regular metre and rhymes, and it has been one of the most popular poetic forms in the English language since the early modern period (Brogan et al. 2012). Since Shakespeare published 154 sonnets in 1609, his poems have become a representative example of the English sonnet, so much so that a sonnet form favored by him has been named the “Shakespearean” sonnet (Lennard 2006: 49). This study conducts a computational and formalist inquiry into one of the key issues involving Shakespeare’s sonnets that could have far-reaching implications in the history of poetic form: namely, the volta.

The volta, an Italian word meaning a “turn,” refers to a shift in logic or argument in a sonnet. This turn can occur in different places, depending on the type of sonnet form, whether it is a Petrarchan sonnet (also called the Italian sonnet), or a Shakespearean sonnet (also called the English sonnet) (Fuller 1972: 1–26; Post 2017: 78–9). The Petrarchan sonnet form, which originated from medieval Provençal poets and was perfected and popularized by the fourteenth-century Italian poet Petrarch, consists of two sections: the first eight lines (called “octave” or “octet”) that have a recurring rhyme scheme (abbaabba); and the next six lines (called “sestet”) that have a more flexible rhyme scheme (e.g. variations of cdecde). In the Petrarchan sonnet, the volta takes place between the octave and the sestet, reflecting the 8–6 division of this sonnet form (Brogan 2012). Meanwhile, the Shakespearean sonnet form, devised by early 16th century English poets such as Sir Thomas Wyatt and Henry Howard, Earl of Surrey, and later mastered by Shakespeare, maintains a different structure. It consists of three sets of four lines (called “quatrains”) that use three corresponding sets of rhymes with a similar pattern (ababcdcdefef), and the last two lines (called “couplet”) with a different rhyme pattern (gg). For the Shakespearean sonnet, the volta is usually considered to take place between three quatrains and the couplet, following the 12–2 division that is characteristic of this sonnet form (Fussell 1965: 127–8). Thus, the place of the volta differs for these two types of the sonnet: in the Petrarchan or Italian sonnet, the volta occurs in the beginning of line 9 between the first eight and the next six lines; in the Shakespearean or English sonnet, the volta takes place in the beginning of line 13 between the first three quatrains and the last couplet.

However, critics have debated whether all 154 sonnets of Shakespeare’s follow Shakespearean sonnet form in terms of the placement of the volta. The most typical explanation that is found across textbooks (Greenblatt 2018: 1061) and popular educational resources (Poetry Foundation) posits that the volta almost always takes place in the beginning of line 13. However, more scholarly accounts present a more complicated view on the issue, suggesting that the volta could occur in the beginning of line 9, as they acknowledge the “vestigial remains of the octave of the continental sonnet” (Booth 1969: 36) in Shakespeare’s sonnets—albeit to varying degrees. For those who detect the weak traces of the Petrarchan sonnet, Shakespeare’s sonnets mainly feature the “double” volta: a strong one in line 13 and a weak one in line 9 (Brooks and Warren 1938: 648; Edmondson and Wells 2003: 52). For those who discover a strong continuity between the Petrarchan and the Shakespearean sonnet, his sonnets tend to retain the volta in line 9, while possibly featuring a minor turn in line 13 (Booth 1969: 36; Muir 1979: 84; Duncan-Jones 1997: 96; Lennard 2006: 49; Post 2017: 79).

This article revisits these formal issues involving the volta in Shakespeare’s sonnets in a methodologically enriching and innovative way by integrating qualitative and quantitative analysis, distant and close reading, human interpretation and AI-assisted machine reading. Most criticisms of Shakespeare’s sonnets have approached the volta often in an impressionistic manner by conducting close readings of a few selected poems.¹ Indeed, while critics have diverged on the issue of the influence of the Petrarchan sonnet form on Shakespeare’s sonnets, their language remains similarly vague. For some, when his sonnets feature the volta in line 9, it is “often” weak (Edmondson and Wells 2003: 52); and for others, these sonnets present a strong turn in line 9 “sometimes” (Post 2017: 79), or “much more often than not” (Duncan-Jones 1997: 96). However, it is precisely the critic’s task to explore how often Shakespeare’s sonnets present the volta in the ninth line, if they do, and to assess how strong or weak the turn is.

Unlike previous criticisms, this study takes a more subtle and systematic approach to the Shakesperean volta by expanding and shifting the focus of inquiry. Rather than simply trying to specify where the volta occurs, it also provides more nuanced insights into how it works by addressing the following questions.

• Research Question 1 (the place of the volta): Where is the volta in Shakespeare’s sonnets? Does the turn occur in the ninth line like the Petrarchan sonnet? Or does it take place in the thirteenth line as it is usually believed to do in the Shakespearean sonnet?

• Research Question 2 (the degree of the volta): How much shift in thought or feeling does the volta entail in Shakespeare’s sonnets? How can we measure degrees of the turn or transition in his sonnets?

• Research Question 3 (the nature of the volta): How does the volta function in Shakespeare’s sonnets? If the volta involves a shift in thought or argument, as critics have suggested, does this logical transition entail emotional change as well? Moreover, what kind of emotional or intellectual transition does it offer? Does the volta contain a turn between individual lines (i.e. between a single line and the one that follows it), or between sections (i.e. between the two groups of consecutive lines)?

In exploring these questions, this article incorporates the approaches that have not often been adopted in the criticism of Shakespeare’s sonnets—namely, a quantitative method and Computational Literary Studies. More specifically, this study employs the technique of sentiment analysis—a computational method of evaluating and measuring sentiments in texts (Lei and Liu 2021)—by a way of examining sonnet form. While there have been a few computational analyses of Shakespeare’s sonnets (Kohonen et al. 2005; Greene et al. 2010; Peng et al. 2021), none has yet utilized sentiment analysis to explore the sonnets’ formal characteristics. By performing sentiment analysis, this work examines the Shakespearean volta through a quantitative approach.² I approach the volta as a statistical problem of identifying a turning point in a sequence of fourteen values—in this case, sentiment scores assigned to each line in a sonnet.

Moreover, this study contributes to Computational Literary Studies and its subfield of sentiment analysis in two ways. First, it is one of the few attempts to apply sentiment analysis to the problem of literary form. Sentiment analysis, initially developed as a tool for analyzing online reviews (Turney 2002), has recently been used to analyze a wide range of data such as social media posts (Mäntylä et al. 2018) and literary texts (Nalisnick and Baird 2013; Reagan et al. 2016; Elkins 2022). While these literary studies perform sentiment analysis mainly to explore narrative structure or characters traits, mine is distinctive in addressing poetic form. Furthermore, this study not only applies a computational method to new materials and new problems but also supplements it with human interpretation. Recent years have witnessed important computational literary studies of poetic form, most of which have focused on meter and rhythm (Plamondon 2006; Greene et al. 2010; De Sisto, et al. 2024; Shang and Underwood 2024). This study differs by adopting the “human-in-the-loop” approach (Yeruva et al. 2020), incorporating a qualitative examination of the sonnets with computational analysis.

In what follows, Section 2 presents a qualitative analysis of the volta in Shakespeare’s sonnets, examining their themes and syntactic elements; and it classifies the sonnets as having “hybrid,” “shadow,” or “novel” form, depending on the degree and location of the volta. Section 3 supplements this qualitative analysis with a quantitative analysis of the volta, examining the sentiment scores of Shakespeare’s sonnets through the calculation of moving difference, cumulative sum, and effect size. By comparing and synthesizing quantitative and qualitative analysis, Section 4 proposes that the volta in Shakespeare’s sonnets tends to occur before the final couplet rather than before the sestet, providing new insights into its emotional and “sectional” aspects.

2. The structure of Shakespeare’s sonnets: semantic and syntactic analysis

This section provides a qualitative analysis of the volta in Shakespeare’s sonnets by identifying its place, degree, and functions. For this purpose, I examine 152 sonnets of Shakespeare’s in the Folger Library’s electronic edition (Mowat and Werstine 2006)—that is, all his sonnets excluding the two sonnets with irregular line numbers, such as Sonnet 99 with fifteen lines and Sonnet 126 with twelve lines. This analysis is conducted based on the following principles:

1. Every sonnet presents the volta or turn before the closing couplet—that is, in the beginning of line 13. As we have seen above, the rhyme structure of the Shakespearean sonnet (ababcdcdefefgg) entails a division between the first three quatrains and the final couplet, thereby inducing a section break to function as a natural place for presenting a turn.

2. Some sonnets provide an additional turn in the beginning of line 9. For Shakespeare’s sonnets often retain the vestige of the continental sonnet form, in which the volta occurs between the octave and the sestet. In deciding whether a sonnet has the volta in line 9 or not, I consider both syntactic and semantic elements. I presume that the turn always takes place after the first eight lines whenever the ninth line begins with (or contains near the beginning) what I call transition markers—conjunctions or interjections that indicate some kind of transition (e.g. “so,” “then,” “but,” etc). When there is no transition marker, I detect the volta in line 9 based on thematic analysis of the relationship between the octave and the sestet. Only if there occurs an obvious change in thought, argument, or attitude, I consider that there exists a volta in line 9.

3. In identifying the functions of the volta, I mark one or two functions depending on the context. Critics suggest that the volta in Shakespeare’s sonnets plays various roles: it concludes, explains, summarizes, reverses, ironizes, questions, elaborates, resolves a problem, and so forth (Muir 1979: 90–1; Vendler 1997: 25; Edmondson and Wells 2003: 57; Cousins 2011: 128). In my analysis that follows, I divide these diverse functions into two groups: the primary functions and the secondary functions.

4. The primary functions of the volta are concerned with a logical relationship between two sections (whether between the first 12 lines and the last 2 lines, or between the first 8 lines and the next 6 lines), considering that the volta primarily serves to mark “a decisive ‘turn in thought’” (Brogan 2012: 1527). I propose that the primary function of the volta can be identified as one of the four types below (Table 1), each of which is often accompanied by characteristic transition markers. Although not every volta has a transition marker, whenever it has one, I indicate the primary function accordingly. For example, the ninth line of Sonnet 18 begins with “But”; hence, the primary function of the volta there is identified as “reversal.”

5. I indicate the primary function of the volta in line 13 for every sonnet whereas I identify the primary function of the volta in line 9 only if it is accompanied by a transition marker. Considering the structure of the Shakespearean/English sonnet, the placement of the volta before the couplet is natural and expected while the placement of the volta after the first two quatrains is optional. In the latter case, transition markers can serve as indisputable evidence for the existence of the turn in the beginning of line 9.

6. Meanwhile, the secondary functions of the volta are mainly concerned with clarifying a rhetorical relationship between two sections divided by the volta. The volta is considered to perform the secondary function in addition to its primary function only when its rhetorical function is obvious and conspicuous. The secondary function of the volta can be described as one of the four types below (Table 2).

Table 1.

The primary functions of the volta.

Type	Function	Transition markers
Conclusion	Explains the consequences or implications of the preceding lines	then, so, therefore, thus
Explanation	Explains the reasons for events, thoughts, or emotions that have been expressed before	for
Summary	Repeats and summarizes what has come before
Reversal	Develops ideas or emotions in the direction that is opposite to the preceding lines	but, no, unless, yet1

1

Note that the word “yet” expresses not only contrast (Oxford English Dictionary, III.10) but also addition (like “in addition” [OED, I.1]), while having other meanings related to time (like “still” [OED, II.5.a], or “until now” [OED, II.7.a.]). This study does not consider all the different meanings of “yet” because it tends to signify contrast when it appears in the beginning of a line in Shakespeare’s sonnets.

Table 2.

The secondary functions of the volta.

Type	Function
Resolution	Provides a solution to the problem or difficulty that has been stated in the preceding lines
Irony	Draws out or amplifies the paradox or irony that has been implied in the preceding lines
Analogy	Either provides a metaphor for the situation or ideas that have been expressed before; or explains the meaning of a metaphor that has appeared before
Supplement	Provides an additional insight that subtly modifies or complicates what has come before, adding a minor twist to an apparently simple summary

In analyzing the structure of Shakespeare’s sonnets, I suggest that his sonnets can fall into one of the following three categories: (1) hybrid form; (2) shadow form; and (3) novel form.

• Hybrid form designate a sonnet in which the volta occurs in both lines 9 and 13. In this case, the turn after the second quatrain is indisputably clear because there is a transition marker in (or near) the beginning of line 9 (e.g. “so,” “for,” “but,” etc). These sonnets that feature a double volta can be seen as a hybrid of the Shakespearean/English sonnet and the Petrarchan/Italian sonnet: their rhyme scheme (ababcdcdefefgg) suggests a twelve-line section followed by a two-line section, while like the Petrarchan sonnet, they also present a clear shift between the octave and the sestet. Among 152 sonnets, fifty-six sonnets (37 per cent) can be categorized as employing hybrid form.

• Shadow form refers to the structure of a sonnet in which the volta occurs in both line 9 and line 13, but the first turn is “weak,” lacking a transition marker in or near the beginning of line 9. I designate this structure—essentially a hybrid form without a transition marker—as “shadow” form because it retains the influence of the Petrarchan sonnet only in a subtle way. A sonnet in shadow form presents a turn before the sestet like the Petrarchan sonnet; however, this turn is not conspicuous or obvious because it occurs only in a semantic sense, not in a syntactic sense. In other words, the volta in line 9 can only to be interpreted through an analysis of a thematic shift between the octave and the sestet, rather than being highlighted conjunctions that signal such a shift (as is the case with hybrid form). Since the volta in line 9 is not so evident or strong, my analysis identifies its secondary function, not its primary function. Among 152 sonnets, twenty-seven sonnets (18 per cent) belong to the shadow form, making it the smallest group in Shakespeare’s sonnets.

• Novel form characterizes the structure of a sonnet in which the volta occurs only before the terminal couplet (i.e. in the beginning of line 13). This sonnet form can be described as “novel” in that it most explicitly departs from the Petrarchan/Italian sonnet in terms of structure. Considering its rhyme scheme, the Shakespearean/English sonnet form can be divided into two sections: three quatrains (ababcdcdefef) and a couplet (gg). While not all Shakespeare’s sonnets demonstrate the 12–2 division, as we have seen in the hybrid and shadow forms, a majority of his sonnets—sixty-nine sonnets (45 per cent)—exhibit a structure that closely reflects the rhyme scheme of the English sonnet, presenting a “turn” between three quatrains and the couplet.

I provide a list of sonnets that belong to each category in Supplementary Appendix 1, along with a detailed analysis of representative sonnets of each sonnet type in Supplementary Appendix 2.

2.1. Preliminary Conclusions

The semantic and syntactic analysis of the volta has identified three types of sonnet forms in Shakespeare’s sonnet sequence. This qualitative analysis of the sonnet structure provides preliminary answers to the research questions formulated above. First, regarding the placement of the volta (RQ1), I have demonstrated that the volta occurs most frequently in line 13 and less often in line 9. All of Shakespeare’s sonnets under discussion in this section present a turn before the couplet, and the function of this turn can be clearly identified and classified. Hence, this analysis suggests that Shakespeare’s sonnets typically feature the volta in line 13, confirming the standard account of the placement of the volta in his sonnets. Moreover, a close examination of individual sonnets supports the view that the volta can be found in line 9 as well, as we have categorized eighty-three sonnets (55 per cent) sonnets as hybrid and shadow forms. In other words, a slight majority of Shakespeare’s sonnets have a double volta occurring in both line 9 and line 13, although my estimate is a little lower than that of Stephen Booth’s (Booth 1969: 36); he claims that ninety-six sonnets (63 per cent) present a turn before the sestet (see Supplementary Appendix 1).

Secondly, this qualitative analysis suggests a method of measuring the degree of the volta (RQ2). For example, we can assess the degree of the turn in line 9 by considering the numbers and proportions of the three types of sonnet forms. Fifty-six sonnets (37 per cent) in hybrid form present a strong turn in line 9, as it is indicated by such conjunctions as “therefore,” “for,” “but,” and so forth. Meanwhile, twenty-seven sonnets (18 per cent) in shadow form exhibit a weak turn in line 9, lacking any transition markers between the octave and the sestet. The remaining sonnets—sixty-nine sonnets (45 per cent) in novel form—do not appear to display a turn in line 9. As for the volta in line 13, however, qualitative analysis does not readily provide a method for measuring the degree of the turn.

Finally, the preliminary analysis of the volta can provide substantial insights into the nature of the Shakespearean volta (RQ3), particularly regarding its transitional function. By identifying one or two functions of each volta, we can examine whether the volta in effect entails a dramatic shift in thoughts and emotions, as critics have long believed. According to my analysis, the voltas in Shakespeare’s sonnets have 336 functions in total, considering that one volta may perform multiple functions (see Table 3).

Table 3.

The frequency of the volta’s functions for each type of Shakespeare’s sonnets.

Function	Hybrid form		Shadow form		Novel form		Total
	Number	Percentage	Number	Percentage	Number	Percentage	Number	Percentage
Conclusion	50	31.1	9	13.2	30	26.5	83	26.1
Reversal	36	22.4	8	11.8	16	14.2	60	17.3
Summary	20	12.4	12	17.6	19	16.8	51	15
Resolution	17	10.6	24	35.3	7	6.2	48	14.1
Explanation	13	8.1	1	1.5	19	16.8	33	9.7
Irony	11	6.8	2	2.9	16	14.2	29	8.5
Supplement	8	5	5	7.4	6	5.3	19	5.6
Analogy	6	3.7	7	10.3	0	0	13	3.8

Among them, 137 functions (41 per cent) indicate a drastic shift, accounting for functions such as “reversal,” “resolution,” and “irony.” In addition, 135 functions (40 per cent) suggest a moderate shift, including “conclusion,” “explanation,” and “supplement.” Finally, sixty-four functions (19 per cent) imply little or no shift, involving “summary” or “analogy.” Thus, in most cases (81 per cent), the voltas in Shakespeare’s sonnets mark either a drastic or modest transition in argument or rhetoric. The volta initiates a meaningful shift within a sonnet, rather than merely serving as a functional device that divides two sections. However, how drastic or significant is the shift induced by the volta in each sonnet and in the sonnet sequence as a whole? The degree of the volta and other formal issues can be further examined through a quantitative method, as I will demonstrate in the next section.

3. The structure of Shakespeare’s sonnets: sentiment analysis

3.1 Methods

3.1.1 Sentiment analysis

Sentiment analysis refers to a computational technology for measuring sentiments in texts (Lei and Liu 2021). Although it is inherently challenging for computational algorithms to evaluate sentiments, which are fundamentally subjective and fluid, this study employs two strategies to maximize the efficacy of sentiment analysis: establishing an ensemble model and employing the “human-in-the-loop” approach. An ensemble model incorporates various tools of sentiment analysis to improve on the quality of analysis (Chun 2021). This study utilizes an ensemble model on a small scale, which is however unique in incorporating the results of sentiment analysis conducted by human readers to compare, validate, and supplement those produced by computers. Specifically, this article uses the following three datasets that identify a sentiment of each line in Shakespeare’s sonnets: (1) the sentiment_m dataset generated by manual tagging; (2) the sentiment_v dataset created by a lexicon-based sentiment analysis model called VADER; and (3) the sentiment_r dataset produced by a deep-learning-based sentiment analysis model using RoBERTa.

The first dataset was created by a team of researchers including myself and five graduate students. We assigned a sentiment score to every line of all 154 sonnets by giving 1 for a positive sentiment, 0 for a neutral sentiment, and −1 for a negative sentiment. We tried to collectively evaluate sentiments through multiple rounds of debate while I made final decisions. In the process, I have established basic principles by which to process ambiguity, irony, negation, and other textual complexities that characterize Shakespeare’s sonnets (see Supplementary Appendix 3). In this way, we aimed to produce not so much an objectively accurate dataset as the most consistent one as possible.

The second and third datasets were generated by computational tools for sentiment analysis that have been considered effective by experts. In order to diversify the methods, this study uses two models that operate on different principles: VADER, a lexicon-based model that is included in the popular NLTK (Natural Language Toolkit) package (Hutto and Gilbert 2014), ³ and a RoBERTa-based model that utilizes deep learning.⁴ The sentiment_v dataset and the sentiment_r dataset contain the sentiment scores of each line of Shakespeare’s sonnets, processed by VADER and the RoBERTa model, respectively, which then are normalized to have values ranging between −1 and 1.

Note that this study performs sentiment analysis of Shakespeare’s poems line by line, considering the regularity of his metrical lines. A majority of lines in his sonnets are “end-stopped,” meaning that each metrical line contains a complete sentence or phrase, or it ends with punctation. While some lines in the sonnets are “enjambed” (that is, a line does not feature a whole sentence or phrase because it extends over multiple lines), enjambment in his sonnets is mostly soft and not disruptive. For example, in many cases of enjambed lines, a line break divides a verb and an object (or object phrase): “And [I can] heavily from woe to woe tell o'er/The sad account of fore-bemoaned moan” (Sonnet 30, lines 10–11);⁵ “One on another's neck do witness bear/Thy black is fairest in my judgment's place” (131.11-2). In these cases, enjambment is soft because a line break separates elements of a sentence where it is naturally divided, such as the subject-and-verb complex and the rest of the sentence. In the latter case (Sonnet 131), the line division is even more appropriate, as the object phrase form a complete clause with its own subject and verb.

Meanwhile, there are a few cases of hard enjambment where a line break divides a sentence in more unconventional places. For instance, it divides a verb phrase (“they that level/At my abuses” [121.9-10]; “he may/Triumph in love” [151.7-8]), a “how-to” phrase (“I teach thee how/To make him…” [101.13-14]), a “where” phrase (“Where, alack,/Shall Time's best jewel…” [65.9-10]), and a subject and a verb (“love's face/May still seem…” [93.2-3]; “Three winters cold/Have from the forests shook…” [104.3-4]; “lust/Is perjured” [104.3-4]). Barring these few exceptions, however, most lines in Shakespeare’s sonnets convey complete meaning in themselves, whether they are end-stopped or softly enjambed. In other words, a metrical line for Shakespeare almost always serves as a semantic unit (if not always being a syntactic unit), thereby making it a suitable unit of sentiment analysis.

Nonetheless, the challenge of performing line-by-line sentiment analysis should be acknowledged, along with other limitations of sentiment analysis. Since most sentiment analysis models (including those used here) are trained on modern and contemporary English, they may be limited in evaluating early modern texts like Shakespeare’s sonnets. Moreover, sentiment analysis models typically use sentences as their basic unit of analysis. It remains unclear how these models are adept at evaluating the sentiment of a metrical line that functions like a sentence. However, one may argue that these models can produce approximate results due to the relatively simple and uniform structure of lines in Shakespeare’s sonnets. Furthermore, the limitations of sentiment analysis tools can be mitigated by incorporating insights from sentiment analysis performed by humans. Compared to these tools, humans are better equipped to assess the sentiment of a metrical line because we can conceive of it as a unit of meaning, even if it is not a complete sentence, especially when we have knowledge of early modern English usage. Thus, the “human-in-the-loop” approach, which integrates human interpretation and machine reading, can reinforce the validity of sentiment analysis of the sonnets.

3.1.2 Statistics

This study analyzes three datasets mentioned above (sentiment_m, sentiment_v, sentiment_r) to examine the volta’s place, degree, and characteristics. In the process, it treats the problem of the volta as a statistical issue of change point detection. My analysis aims to detect a turning point among fourteen values in a sequence—that is, sentiment scores assigned to each line of a sonnet—to understand where and how the volta occurs in a fourteen-line sonnet. To measure differences among sentiment scores within each sonnet, I deploy three statistical methods: (1) analysis of moving difference; (2) CUSUM (cumulative sum control chart); and (3) effect size calculation. I will explain these methods by using an example of the quantitative analysis of the sentiment scores of Sonnet 18 in the sentiment_m dataset. Figure 1 plots the sentimental scores of each line of the sonnet (blue line), which are overlaid by diverse statistical values produced by the three statistical methods that will be elaborated below. (For more detailed data about the quantitative analysis of Sonnet 18, see Supplementary Appendix 4.)

Figure 1.

Quantitative analysis of the volta: Sonnet 18 (sentiment_m).

First, moving difference refers to the difference between current and previous values (⁠$X_i-X_{\left\{i-1\right\}}$⁠). In the context of a sonnet, I calculate the difference between the sentiment score of a line (say, line 2) and that of a previous line (line 1). The process begins by measuring the difference between line 2 and line 1, and then sequentially repeats until the difference between line 14 and line 13 is measured. (For line 1, I take the value of the sentiment score of line 1 as the difference value itself because there is prior line for comparison.) Examining moving differences is the simplest way of analyzing a turn or transition within a sonnet on the level of individual lines. For example, in Sonnet 18 (sentiment_m), the difference in sentiment scores between lines 2 and 3, as well as between lines 8 and 9 are greater than any other differences between two consecutive lines. This indicates that the greatest shifts in sentiments occur between lines 2 and 3, and lines 8 and 9. Moving difference thus provides a simple index of the line-to-line transition.

The CUSUM is a technique used to detect notable deviations in a time series by measuring the degree of change in relation to the cumulative sum (Page 1954; Barnard 1959). In analyzing sonnet structure, the CUSUM measures the difference between sentiment scores of two consecutive lines like moving difference, but it also considers the sum of sentiment scores that have accumulated up to a certain line. Hence, this method effectively delineates the trajectory or overall tendency of sentiment scores throughout the poem. In Sonnet 18 (sentiment_m), the CUSUM value (i.e. the greater of the SH or SL values) peaks at line 8.⁶ This value indicates that Sonnet 18 exhibits the greatest shift in sentiments between lines 7 and 8, which best reflects the overall trajectory of the sonnet’s sentiments—in this case, a downward movement in the sentiment curve. Thus, the CUSUM measures the relative strength of the line-to-line transition.

Finally, effect size calculation measures the difference between two groups by comparing the standardized difference between the means of each group (Ellis 2010). The most well-known index for effect size calculation is Cohen’s d.⁷ In analyzing sonnet structure, Cohen’s d helps assess the difference between the sentiment scores of two sections in a sonnet. I calculate eleven values of Cohen’s d by sequentially grouping sentiment scores from line 3 to line 13. For example, Cohen’s d in line 3 measures the difference of sentiment scores between lines 1–2 and lines 3–14; Cohen’s d in line 4 refers to the difference between lines 1–3 and lines 4–14, and so forth. In Sonnet 18 (sentiment_m), Cohen’s d in line 9 is the highest among all Cohen’s d values. This indicates that the greatest change in sentiments takes place between the group of lines 1–8 and the group of lines 9–14. Thus, Cohen’s d provides an effective index of the section-to-section transition in contrast to the two other measurements discussed above, which focus more on the line-to-line transition. 

3.2 Analysis

By applying these three statistical methods to three diverse datasets of sentiment scores (sentiment_m, sentiment_v, sentiment_r), this study elucidates the place (RQ1), the degree (RQ2), and the nature (RQ3) of the volta in Shakespeare’s sonnets. For this purpose, my analysis focuses on two aspects of data analysis: (1) the average of moving difference, the CUSUM, and Cohen’s d for each line across 152 sonnets (i.e. all sonnets bar sonnets 99 and 126); and (2) the frequency of the number of cases when a certain line has highest values of moving difference CUSUM, or Cohen’s d. The former—the average of change-indicative values—indicates an overall tendency of major turns in Shakespeare’s sonnets. Meanwhile, the latter—the frequency of the greatest change-indicative value for each sonnet—provides a comprehensive view of the inner dynamic of each individual sonnet. The former and the latter often coincide with one another, as we can see in Table 4 which lists the top three lines with the greatest average or frequency for each dataset.

Table 4.

The top three lines with the greatest moving difference, the CUSUM, and Cohen’s d

	Rank	sentiment_m		sentiment_v		sentiment_r
		frequency (percentage)	average	frequency (percentage)	average	frequency (percentage)	average
Moving difference	1	13 (12.1%)	13	14 (13.6%)	14	3 (11.2%)	13
	2	5 (10.5%)	5	11 (13.0%)	11	9 (10.5%)	14
	3	9 (9.9%)	9	9 (10.7%)	9	13 (9.2%)	9
CUSUM	1	14 (19.1%)	14	14 (16.4%)	14	14 (17.8%)	14
	2	4 (15.1%)	13	13 (12.5%)	13	12 (11.2%)	12
	3	12 (13.2%)	12	12 (9.9%)	12	10 (9.9%)	13
Cohen’s d	1	13 (28.3%)	13	13 (22.4%)	13	13 (27.6%)	13
	2	9 (17.1%)	9	3 (19.1%)	3	3 (16.4%)	12
	3	3 (13.2%)	11	12 (13.8%)	12	12 (13.8%)	3

3.2.1 Moving difference

The analysis of moving difference demonstrates that we can detect the greatest turning point in either line 13 or line 14. As one can see in Table 4, line 13 ranks at the top in terms of the average moving difference and the frequency of the greatest moving difference in the sentiment_m dataset, as well as in the average in the sentiment_r dataset. Likewise, line 14 ranks at the top in terms of the average and the frequency in the sentiment_v dataset, while it is ranked second in the average in the sentiment_r dataset.

However, these results provide fairly weak evidence for both views that the volta occurs before the closing couplet, and that it takes place between the octave and the sestet. Moving differences of sentiment scores do not strongly support the former because they are relatively evenly distributed among all lines. Consider the frequency and the proportion of the lines that have the biggest moving difference for each sonnet in the sentiment_m dataset. Top five lines that most often demonstrate the greatest shift from the previous line are line 13 (12.1 per cent), line 5 (10.5 per cent), line 9 (9.9 per cent), line 3 (9.4 per cent) and line 11 (8.8 per cent). We find a similarly even distribution in other datasets: the percentage of the top five lines ranges from 13.5 per cent to 10.7 per cent in the sentiment_v and from 11.2 per cent to 8.6 per cent in the sentiment_r. Thus, although line 13 could be one of the most likely places where the greatest shift occurs, it is only slightly less likely that this drastic shift can take place in other locations.

For similar reasons, the examination of moving difference offers uncertain grounds for the view that finds the volta in line 9. It is meaningful that line 9 is ranked third in the average and the frequency in both the sentiment_m dataset and the sentiment_v dataset; it is also ranked second in the frequency and third in the average in the sentiment_r dataset. Yet there is not a significant difference in the frequency of the greatest turn among the top five lines. The analysis of moving difference thus indicates that the volta can take place virtually anywhere in a sonnet, not necessarily in line 13 or line 9, if one were to understand it to entail the maximum line-to-line transition.

3.2.2 CUSUM

The analysis of the CUSUM suggests that one can identify the greatest shift in line 14 when we consider a line-to-line difference in sentiment scores in relation to a trajectory of these values up to that line. As we can see in Table 4, line 14 ranks at the top in terms of both the average of CUSUM values and the frequency of the greatest CUSUM values in all three datasets.

These results provide weak evidence for both views that identify the volta in the beginning of line 13 and in that of line 9. After all, it is in line 14, not in line 13, that we find the greatest shift in Shakespeare’s sonnets according to the CUSUM. Line 13 involves a fairly significant transition in some datasets as it is ranked second in the frequency and the average in the sentiment_v dataset. But it does not display the most meaningful shift in other datasets: it is ranked third in the average in the sentiment_r dataset, and it is ranked seventh in the frequency in the sentiment_m dataset (5.9 per cent). Moreover, the distribution of the frequency of the greatest turn is relatively even, not unlike the case of moving difference, thereby making it difficult to assert with confidence which individual line entails the volta at all. For the sentiment_v, the top five lines that illustrate most often the greatest shift are line 14 (16.4 per cent), line 13 (12.5 per cent), line 12 (9.9 per cent), line 11 (9.9 per cent), and line 8 (9.2 per cent). Although the distribution of the frequency is less even for other datasets (the most impactful turn occurring most frequently in line 14 for both the sentiment_v [17.8 per cent], and the sentiment_m [19.1 per cent]), these results testify to the presence of the volta in line 14, rather than in line 13.

Moreover, the examination of the CUSUM provides even weaker grounds for the view that the volta occurs in line 9. Line 9 does not rank among the top three for all three datasets (sentiment_m, sentiment_v, and sentiment_r) in terms of both the average CUSUM values and the frequency of the greatest shift. In fact, line 9 is one of those lines in which a very minor change takes place; for example, for the sentiment_m, line 9 is ranked eleventh in the average, and the twelfth in the frequency. Thus, the analysis of the CUSUM suggests that the volta can be placed either in line 14 or anywhere in a sonnet—if we consider the volta as involving the greatest line-to-line transition relative to the trajectory of a whole poem.

3.2.3 Cohen’s d

Effect size calculation locates the greatest turn in line 13. Considering the function of Cohen’s d that is used to measure the difference between two groups, this result means that we find a greater shift in sentiments when we group fourteen lines of a sonnet into two sections consisting of lines 1–12 and lines 13–14 than otherwise. As we can find in Table 4, the Cohen’s d of line 13 ranks at the top across all datasets and in all measurements: the average of Cohen’s d and the frequency of the greatest Cohen’s d in the sentiment_m dataset, the average and the frequency in the sentiment_v dataset, and the average and the frequency in the sentiment_r dataset.

This result offers strong evidence for both views that support the presence of the volta in line 13, and that places it in line 9. For Cohen’s d is greatest when dividing a sonnet into two groups in the beginning of line 13 for all three datasets. Moreover, in contrast with moving difference and the CUSUM, the distribution of the frequency of the greatest shift is not even: the frequency of a sonnet in which Cohen’s d is greatest in line 13 is significantly higher than that of others. Consider the top five lines that yield the greatest Cohen’s d: for the sentiment_m, they are line 13 (27.6 per cent), line 9 (16.7 per cent), line 3 (12.8 per cent), line 5 (10.3 per cent), and line 11 (8.3 per cent); for the sentiment_v, they are line 13 (21.8 per cent), line 3 (18.6 per cent), line 12 (13.5 per cent), line 4 (9.6 per cent), and line 11 (7.7 per cent); and for the sentiment_r, they are line 13 (27.6 per cent), line 3 (16.4 per cent), line 12 (13.8 per cent), line 11 (8.6 per cent), and line 9 (7.9 per cent). Across all datasets—and especially for the sentiment_m and the sentiment_r—the percentage of sonnets where the greatest turn occurs in line 13 remains over 20 per cent. For comparison, the same type of data (i.e. the largest percentage of the frequency where the greatest change-indicative value appears in a certain line) remains often well below 20 per cent in the cases of moving difference and the CUSUM (see Table 4).

Moreover, the analysis of Cohen’s d supports the view that discovers the volta in line 9. For the sentiment_m, line 9 is ranked second in terms of the average Cohen’s d and the frequency of the greatest Cohen’s d. It is true that in other datasets, Cohen’s d of line 9 is not so great: for the sentiment_v, line 9 is ranked seventh both in the average and the frequency; for the sentiment_r, line 9 is ranked seventh in the average and fifth in the frequency. Yet although I do not here attempt to compare or test the validity of three datasets, close analysis of Cohen’s d in the sentiment_m demonstrates its reliability: selective examination of representative samples from each of three sonnet forms (hybrid, shadow, and novel form) illustrates the degree to which Cohen’s d has strong explanatory power in the sentiment_m (see Supplementary Appendix 5). Indeed, the average of Cohen’s d is greatest in line 13 (1.04) and second greatest in line 9 (0.81) in the sentiment_m (see Figs. 2 and 3).

Figure 2.

The average of Cohen’s d in Shakespeare’s sonnets (bar plot).

Figure 3.

The average of Cohen’s d in Shakespeare’s sonnets (box plot).

Finally, the effect size calculation of sentiment scores locates the volta in line 13 and line 9 also because the result of quantitative analysis closely matches that of qualitative analysis. Recall, for example, the result of the syntactic and semantic analysis in Section 2: novel form where the volta occurs in line 13 includes sixty-nine sonnets (45 per cent), whereas hybrid form where the turn takes place in line 9 encompasses fifty-six sonnets (37 per cent). Compare this analysis, then, with the frequency of the largest Cohen’s d. For the sentiment_m, Cohen’s d is greatest in line 13 for the greatest number of sonnets: forty-three sonnets (27.6 per cent). Moreover, it is greatest in line 9 for the second greatest number of sonnets: twenty-six sonnets (16.7 per cent). Indeed, the frequency and the proportion of the greatest Cohen’s d (i.e. approximately 28 per cent for the volta in line 13 and 17 per cent for the volta in line 9) are smaller than those of the volta according to qualitative analysis (i.e. 45 per cent for the volta in line 13 and 37 per cent for the volta in line 9). Yet considering a difference in the range of values, the results match in a more significant manner than it seems. For qualitative analysis, candidates for the volta are only restricted to two lines (i.e. line 9 and line 13) whereas for effect size calculation, candidates for the greatest turn include not just two but ten lines (i.e. line 3 to line 13). Hence, it is fair to suggest that despite apparent differences in numbers, quantitative analysis of the volta’s place effectively reflects the contours of qualitative analysis. Thus, effect size calculation demonstrates that the volta occurs most likely in line 13, and second most likely in line 9 in a sonnet—if we consider the volta as indicative of the greatest section-to-section transition.

4. Conclusion

4.1 The place of the volta (RQ1)

Both qualitative and quantitative analyses of Shakespeare’s sonnets yield a relatively consistent conclusion: the volta, or the greatest shift in sentiments, mostly likely occurs in line 13 and secondarily in line 9. Semantic and syntactic analysis in Section 2 validates critical views that identify the volta in line 13 and line 9. All of Shakespeare’s sonnets present a turn in argument or rhetoric in line 13, whose functions can be identified often in reference with a transition marker. At the same time, a slight majority of Shakespeare’s sonnets—eighty-three sonnets (55 per cent)—adopt the “hybrid” and “shadow” forms that involve a turn in line 9, whether strong or minor. Furthermore, the sentiment analysis of Shakespeare’s sonnets in Section 3 complements the qualitative analysis by demonstrating that the volta occurs in line 13 and line 9, albeit in an equivocal manner. The calculation of moving difference and the CUSUM provides weak evidence for the presence of the volta in line 13 and even weaker proof for that of the volta in line 9. However, the effect size calculation of the sentiment scores offers strong evidence for the placement of the volta in both lines, indicating that the most drastic shift in sentiments occurs most likely in line 13, and secondarily in line 9, particularly in the sentiment_m dataset.

4.2 The degree of the volta (RQ2)

Compared to most criticisms that rely on qualitative analysis, this study offers diverse methods for quantifying and measuring the degree of the volta. My analysis supports a critical view that the volta in line 13 is strong in Shakespeare’s sonnets, while the volta in line 9 is relatively weak. Moreover, this study enables us to assess precisely the strength or the weakness of the voltas in these lines by employing sentiment analysis technology. For instance, we can compare the degrees of the volta in line 13 and line 9 by analyzing the moving difference, the CUSUM, and Cohen’s d of sentiment scores related to these two lines across three datasets (sentiment_m, sentiment_v, and sentiment_r). Notably, in the sentiment_m dataset, the average value of Cohen’s d in line 13 is 1.04, while in line 9, it is 0.81. Thus, we can assert that the degree of the turn in line 13 is approximately 1.3 times greater than that in line 9 in Shakespeare’s sonnets.

4.3 The nature of the volta (RQ3)

The integration of sentiment analysis with traditional close reading provides some of the most insightful conclusions regarding the nature of the volta. It has demonstrated, first, that the volta in Shakespeare’s sonnets involves a shift of both logical and emotive nature. Qualitative analysis of Section 2 illuminates how the volta serves various logical and rhetorical functions, while quantitative analysis of Section 3 elucidates how it marks a shift not only in argument or thought but also in sentiments, incurring the greatest change of emotions in either a positive or negative direction.

Moreover, this article shows that the volta in Shakespeare’s sonnets signals not so much a line-to-line transition as a section-to-section transition. As observed in Section 3, the analysis of moving difference and the CUSUM yields inconclusive results regarding the volta, while effect size calculation consistently indicates that the volta occurs in line 13. However, this incongruence does not necessarily reflect the limitations of statistical methods; rather, it enhances our understanding of the workings of the volta. Moving difference and the CUSUM measure differences between two consecutive values, albeit in different ways, suggesting that the volta for Shakespeare does not involve a shift between a current line and a previous one. In contrast, Cohen’s d indicates a difference between two groups, and therefore, the consistent results from effect size calculation illustrate that the volta entails a shift between one group of lines and the next. Thus, this study, which integrates digital methods with close reading practices sheds new light on the volta, expanding the focus of formalist inquiry from where it occurs to how it works. The Shakespearean volta induces a section-to-section transition rather than a line-to-line transition.

Author contributions

Haram Lee (Conceptualization, Formal analysis, Investigation, Methodology, Visualization)

Funding

Funding support for this article was provided by the the New Faculty Startup Fund from Seoul National University.

Supplementary data

Supplementary data are available at DSH online.

Notes

1

An exception to this observation is the work of Stephen Booth, who analyzes all sonnets of Shakespeare’s to offer a comprehensive account of the volta (Booth 1969: 36-51). My analysis of the volta builds upon Booth’s, providing a more detailed discussion by dividing the sonnets into three types (rather than two), and embracing digital technologies and statistical methods. For detailed comparison with Booth’s work, see Supplementary Appendix 1.

2

In doing so, this study contributes to a recent trend in Shakespearean studies that deploys digital and/or quantitative methods. For the digitalization of early modern texts, see Estill 2020; for computational stylistics and digital authorship studies of Shakespeare, see Craig and Greatley-Hirsch 2017; Taylor and Egan 2017; and for quantitative studies of Shakespeare’s versification, see Tarlinskaja 2016; Bruster 2023: 43, 49–50.

3

Yeruva et al. (2020) suggest that among six lexicon-based sentiment analysis tools, VADER produce results that most closely match human interpretation and that it is particularly suited to analysis of literary texts (3106, 3107).

4

https://huggingface.co/cardiffnlp/twitter-roberta-base-sentiment-latest. This popular RoBERTa-based model evaluates sentiments based on the deep learning of 124 million tweeter feeds. A previous version of this model has been used in literary and narrative studies (Lee 2023; Lal et al. 2024).

5

Hereafter, citations of Shakespeare’s sonnets will be given with a sonnet number and line number(s) in the text.

6

To calculate the CUSUM, I convert sentiment scores of each sonnet into Z-scores for normalization. I use the equation (⁠$Z_i=\frac{X_i-\mu }{\sigma }$⁠) where $\mu$ and $\sigma$ stand for the mean and the standard deviation of all sentiment scores of a sonnet, respectively, so that Z-scores would have a mean of 0 and a standard deviation of 1. Then, I measure SH and SL values—which indicate the degree of change in an upward or positive direction and a downward or negative direction, respectively—as follows:

$$S{H}_i=\max (0, S{H}_{\left\{i-1\right\}}+Z_i-\omega )$$

 

$${SL}_i=\max (0, S{L}_{\left\{i-1\right\}}-Z_i-\omega )$$

Here $\omega$ is set as 1.

8

To calculate Cohen’s d, I use the standard metric in Ellis (2010: 10, 26). It may be more accurate to use another index called Hedge’s g, instead of Cohen’s d, to perform the effect size calculation of sonnets. Hedge’s g is known to perform well when the size of groups is dissimilar (Ellis 2010: 10), and this is the case with sonnets: we have a grouping of lines 1–2 and lines 3–14, lines 1–3 and lines 4–14, etc up to that of lines 1–12 and lines 13–14. However, my purpose is to measure relative differences between the effect size of diverse groupings, not to analyze effect size in absolute terms; hence, it does not make a critical difference in the end whether I use Cohen’s d or Hedge’s g.

Data availability

The data underlying this article are available in Zenodo, at https://doi.org/10.5281/zenodo.14965925.