![Sample our Language & Literature journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5941/TOC-Subject-Sample-2021-Language-Literature.jpg"})
Abstract
Scholarly consensus as to defining the genre of autobiography has largely coalesced around the determination that there can be no fixed and overarching definition of genre. In this study, we turn to the autobiographical texts themselves in order to discern large-scale conventions and commonplaces of writing about the self. Employing the machine learning technique of statistical topic modeling, this study analyzes over 83,000 pages of eighteenth- and nineteenth-century autobiographical texts written in English. In doing so, it reveals common autobiographical conventions, subjects, and discourses, while shedding light on current scholarly understandings of autobiography.
Disclosure Statement
No potential conflict of interest was reported by the authors.
Notes
1 See, for instance, Lejeune, Moi aussi, 13–35; and Signes de vie, 11–30. Lejeune’s original efforts to define the genre (which, he is careful to note, refer only to European texts post-1770) appear in L’ Autobiographie en France, 14; “Le Pacte autobiographique,” 138.
2 Due to the difficulties of defining autobiography, we refer to “autobiography” when examining efforts to analyze or codify the genre, while we refer to “autobiographical texts” in all other instances in order to indicate the slipperiness of a precise definition of autobiography.
3 In suggesting the alternate word “theme,” Matthew Jockers and David Mimno note that both topics and themes signify “a type of literary content that is semantically unified and recurs with some degree of frequency or regularity throughout and across a corpus.” “Significant Themes in 19th-Century Literature,” 751.
4 We note that our preliminary conclusions apply to our subset of autobiographical texts—those written in English in the eighteenth and nineteenth centuries—and we cannot presume to generalize more broadly to texts written in other languages and time periods. We also cannot generalize from the authors of the autobiographical texts to the larger population, which was much more diverse than the 273 authors in this study. Instead, the corpus sheds light specifically on published (and archived) accounts of self in English in the eighteenth and nineteenth centuries.
5 Olney, “Autobiography and the Cultural Moment,” 3.
6 Anderson, Autobiography, 1–16; Smith and Watson, Reading Autobiography, 218.
7 DiBattista and Wittman, “Introduction,” 2; Smyth, “Introduction: The Range, Limits, and Potentials of the Form,” 6.
8 Rousseau, Oeuvres complètes, 1:5; The Collected Writings of Rousseau, 5:5.
9 Frow, Genre, 10.
10 Whereas Lejeune, in examining the proper name, asserts that the place of this name is “liée, par une convention sociale, à l’engagement de responsabilité d’une personne réelle” [linked, by a social convention, to the pledge of responsibility of a real person], Paul de Man (controversially) analyzes autobiography’s “illusion of reference,” declaring: “Does the referent determine the figure, or is it the other way round: is the illusion of reference not a correlation of the structure of the figure, that is to say no longer clearly and simply a referent at all but something more akin to a fiction which then, however, in its own turn, acquires a degree of referential productivity?” Lejeune, Le Pacte autobiographique, 23; Lejeune, On Autobiography, 11; de Man, “Autobiography as De-Facement,” 69.
11 For an examination of the various forms and representations of self in autobiography, see Eakin, How Our Lives Becomes Stories: Making Selves, especially chapter 1.
12 Roach examines public intimacy as a public illusion of familiarity with and proximity to a celebrity individual through public representations of the individual. It, 3, 16–17. While most readers may not interact with an autobiographer in person, they may have a sense of “knowing” the autobiographer—and thus may feel comfortable evaluating the “truth” of the autobiographical representation—through information about the autobiographer gained from other sources.
13 Frow, Genre, 125.
14 Nancy K. Miller explores the “truth” of autobiography and distinctions between autobiography and fiction in “The Entangled Self: Genre Bondage in the Age of Memoir.”
15 G. Thomas Couser has examined the proliferation of forms of life writing, of which autobiography would be one subset. Couser even questions the term “life writing,” noting that “life narrative” would be more inclusive of non-written forms, while “life representation” would then be more inclusive of non-narrative forms. Carole Boyce Davies writes of an “autobiographical continuum,” which “ranges from the most private, personal revelation to the most austere, historical narrative,” and Anna Poletti explores “the media forms in which the information and stories relating to our identity and our lives are created and shared,” with these autobiographical forms including cardboard boxes and cameras. Couser, The Work of Life Writing: Essays and Lectures, 12–13; Davies, “Private Selves and Public Spaces: Autobiography and the African Woman Writer,” 267; Poletti, Stories of the Self: Life Writing after the Book, 9.
16 For instance, in addition to the numerous eighteenth- and nineteenth-century texts titled Autobiography (e.g. Autobiography of an Actress), the popular review journal the Quarterly Review devoted part of an 1827 issue to new editions in “the field of autobiography.” Mowatt, Autobiography of an Actress; or, Eight Years on the Stage; Rev. Article “Autobiography,” 156.
17 Lejeune, Moi aussi, 31; On Autobiography, 131–32.
18 Underwood, “Machine Learning and Human Perspective,” 106.
19 Eakin, Living Autobiographically, 17. See chapter 1 for an analysis of the conventions of self-narration.
20 As the term “autobiography” itself did not appear in English until 1797 (see Folkenflik, “Introduction: The Institution of Autobiography”), authors’ characterizations of their works in the eighteenth century often included terms such as “memoirs,” “the life of,” etc.
21 On distant reading and non-dominant discourses and subject positions, see Klein, “Distant Reading After Moretti.”
22 On the unique possibilities of data-driven literary study, see Rosenthal, “Introduction.”
23 We selected these dates, while acknowledging that they impose rather arbitrary starting and ending points, in order to include both the eighteenth-century rise in the production of autobiographical texts (thus the 1700 start date), as well as texts in the public domain and accessible via the HathiTrust Digital Library (thus the 1900 end date). Furthermore, dating the autobiographical texts posed a challenge, as autobiographical texts are at times published many years (or centuries) after authors’ lifetimes, with the dates not reflecting the milieus in which they were written. As our goal was to analyze historically specific depictions of self, we selected the publication date as the date of the text if it occurred during the author’s lifetime. For a text that was published after an author’s death, we chose the year of the author’s death as the date of the text, as that year would have been the latest year in which the text could have been written.
24 We note that these metrics necessarily involved interpretation. As we detailed earlier in the essay, in constructing the corpus we looked to scholarly determinations and frameworks of genre, with the goal of including texts that constructed coherent autobiographical selves for public consumption.
25 The difference in numbers between volumes and authors is due to authors who penned multiple autobiographical texts, as well as authors whose single autobiographical texts include multiple volumes.
26 Underwood, “DataMunging.”
27 Despite the 1700–1900 time period of our study, no texts from 1700–1709 met the criteria for inclusion in the corpus.
28 Studies in book history have documented and analyzed the exponential increase in print production in Britain and America between the beginning of the eighteenth century and the end of the nineteenth century; see, for instance, Raven, Publishing Business in Eighteenth-Century England, 33–62; St. Clair, The Reading Nation in the Romantic Period, especially 172–173; Warner, The Letters of the Republic: Publication and the Public Sphere in Eighteenth-Century America, 1–33; Eliot, “Some Trends in British Book Production, 1800-1919”; David D. Hall, ed., A History of the Book in America, vols. 1–4. Autobiographical texts also proliferated during this time period. By 1827, in reviewing a group of ten autobiographies, the British Quarterly Review noted (and lamented) “the mania” for writing autobiographies, and Louis Kaplan’s A Bibliography of American Autobiographies reflects the increase in publication as well. Rev. article “Autobiography”; Kaplan, A Bibliography of American Autobiographies.
29 McCallum, “MALLET: A Machine Learning for Language Toolkit.”
30 In this way, we focus on the discursive space of autobiography, or topics in all published pages of autobiographical texts.
31 We based these gender identities on authorial self-identifications and scholarly identifications.
32 Words that appear frequently in a language have the potential to interfere with topic modeling. For example, “the” and “a” co-occur with many nouns and consequently would be included in any topic with such nouns. These words do not convey semantic information or aid in the understanding of a topic, however. MALLET maintains a list of these high frequency words, called a “stoplist,” for removal prior to computing topics. We used MALLET’s default English language stoplist, consisting of 499 words and the letters of the alphabet. For more information on the stoplist, see http://iesl.cs.umass.edu/mallet/import-stoplist.php.
33 For a brief discussion of topic numbers and coherence, see Tangherlini and Leonard, “Trawling in the Sea of the Great Unread,” 731–732.
34 We examine the constructed nature of the textual autobiographical “I,” or autobiographical self, as a narrative creation that attempts to unify the text while purporting to refer to the embodied author outside of the text. The autobiographical “I” reflects historically and culturally specific narratives and frameworks of identity available to the author at the time of writing.
35 We number the topics according to prevalence in the corpus, with topic 1 being the most prevalent. Unless otherwise indicated, we refer in the text to the results of topic modeling all texts in our corpus, rather than topic modeling our two subsets of texts.
36 To determine word frequencies, we ran a Python script and removed all words from MALLET’s stoplist.
37 We note that “time” is currently one of the most common words in the English language. To investigate whether the prominence of the term in the corpus was simply due to its prominence in the English language, we compared the frequency of the term in our corpus, with stoplist words included, to the frequency of the term in the Google Books Ngram Viewer (https://books.google.com/ngrams) for the same time period (1700–1900). We chose the Ngram Viewer as an approximation of published texts on a larger scale. Though our numbers are only approximations, the term “time” appeared in our corpus roughly sixty percent more than in the Google Books corpus (taking the average of frequencies sampled every twenty years), suggesting the prominence of the term in our corpus was due to more than simple English usage. Michel et al., “Quantitative Analysis of Culture Using Millions of Digitized Books.”
38 Pascal, Design and Truth in Autobiography, 72.
39 Coleridge, Biographia Literaria, 1:93.
40 Wright, A Review of the Missionary Life and Labors of Richard Wright, 439.
41 The prevalence of certain of these verbs (e.g. “made”) mirrors their prominence in the Google Books corpus, while others (e.g. “heard”) appear in the autobiographical corpus notably more frequently.
42 Mowatt, Autobiography of an Actress, 149; Littlechild, The Reminiscences of Chief-Inspector Littlechild, 4.
43 Hassall, The Narrative of a Busy Life.
44 Leavitt, Autobiography of the Hon. Humphrey Howe Leavitt, 276; Martineau, Harriet Martineau’s Autobiography, 1:133.
45 We note that the subject category groupings involved subjective determinations, as we interpreted the topics then grouped them by similar subject areas.
46 The Percentage of Pages in Corpus column was computed by summing the percentage of each page assigned to an individual topic, dividing by the sum of the percentages of all pages assigned to all topics, then grouping together the topics into subject categories. The calculation approximates topic and subject category prevalence, as it takes pages, rather than words, as the basis of analysis. Additionally, the number of topics assigned to each category may not correspond to subject category prevalence, as a subject category may include many topics that represent a small percentage of the corpus, or few topics that represent a large percentage of the corpus. Finally, the table does not include the five general topics analyzed in the previous section (included in ).
47 The “non-meaningful topics” category consists of topics formed by similarities that are not meaningful for our analysis, such as groups of non-English words, contractions, names, and word misspellings or OCR errors. Matthew Jockers has noted this type of limitation of topic modeling, observing that topic models of fictional texts may be dominated by character names (see Macroanalysis: Digital Methods and Literary History, chapter 8). This 6.38% of our corpus encounters similar problems, with, for instance, four topics consisting primarily of proper names. An additional limitation of topic modeling is the potential association of one topic with one volume or author, thus repeating the structure and divisions of the corpus rather than identifying themes across the corpus. In our experiment, certain topics, due to their specificity, seemed to be correlated strongly with particular authors. However, further analysis revealed that strong correlations, or instances when one author’s pages contributed to over 50% of the topic (probabilistically speaking), occurred in less than 10% of the 100 topics (and only among the 20 least prevalent topics). A one-to-one association of topics to authors was thus not a concern for the vast majority of topics.
48 Leeson, Memoirs of Mrs. Margaret Leeson, 2:213; Bamford, Passages in the Life of a Radical and Early Days, 2:279.
49 We include quadratic trendlines (the dashed lines) on the figures to show trends over time.
50 In a wide-ranging analysis of American autobiographies of the nineteenth and twentieth centuries, Diane Bjorklund codes a sample of the autobiographies listed in Louis Kaplan’s 1961 A Bibliography of American Autobiographies and Mary Louise Briscoe’s 1982 American Autobiography 1945-1980 by “occupation or primary experience.” Looking at this sample, Bjorklund calculates the percentage of American autobiographies by clergy and/or indicating a religious focus (by the title or according to Kaplan or Briscoe’s annotation) to be 51.9% of those published in the first decade of the nineteenth century, decreasing to 31.5% by the last decade of the nineteenth century. While Bjorklund’s study differs from this study in focus (American autobiographies) and methodology (categorizing entire texts by occupation/primary experience), the study finds overall trends (decreases) in religion-focused autobiographies in the nineteenth century that roughly correspond with the decline in religion-focused discourse in this study. See Bjorklund, Interpreting the Self: Two Hundred Years of American Autobiography.
51 Here we examine historically specific societal constructions of gender, analyzing the prevailing binary (male and female) categories of the time. While our focus is on autobiographical texts, we note the computational work that has been done on gender identity, authorship, and fictional texts, such as Underwood, Bamman, and Lee, “The Transformation of Gender in English-Language Fiction”; Jockers and Mimno, “Significant Themes in 19th-Century Literature.” We also note the extensive non-computational work that has appeared on gender and authorship, detailed in studies such as Batchelor and Kaplan, eds., The History of British Women’s Writing, vols. 5-6; Nelson, “Women in Public.” Work on anonymity and pseudonymity also highlights the role of gender positions in the eighteenth- and nineteenth-century literary spheres; see, for instance, Griffin, The Faces of Anonymity: Anonymous and Pseudonymous Publication from the Sixteenth to the Twentieth Century.
52 The many scholarly studies examining considerations of gender in the context of autobiography and life writing are too numerous to list here but include, for instance: Smith and Watson, eds., Women, Autobiography, Theory: A Reader; Braxton, Black Women Writing Autobiography: A Tradition Within a Tradition; Lionnet, Autobiographical Voices: Race, Gender, Self-Portraiture; Nussbaum, The Autobiographical Subject: Gender and Ideology in Eighteenth-Century England; Davies, “Private Selves and Public Spaces: Autobiography and the African Woman Writer”; Gilmore, Autobiographics: A Feminist Theory of Women’s Self-Representation; Torres, “The Construction of the Self in U.S. Latina Autobiographies”; Miller, “The Entangled Self: Genre Bondage in the Age of the Memoir”; Mintz, Unruly Bodies: Life Writing by Women with Disabilities; DiBattista, “Women’s Autobiographies.”
53 With regard to the subject-specific topics we detail here, we have rejected the null hypothesis that the topics have the same weighting in the female- and male-authored texts in the corpus using a two-sample Student’s t-test (p < 0.01). We also note, however, the limitations of this method, with p-values necessarily indicating probability rather than certainty; see, for instance, Vidgen and Yasseri, “P-Values: Misunderstood and Misused.” Additionally, we note that the difference between female- and male-authored texts in the religion subject category was particularly pronounced likely due to our approach of topic modeling by page. When we experimented with normalizing the topic probabilities by volume (with each volume weighted equally) and then performed a two-sample Student’s t-test, the results for the religion subject category did not allow us to reject the null hypothesis (at p < 0.01), suggesting that certain female-authored texts contained a large number of religion-oriented pages. All results for other subject categories remained statistically significant (at p < 0.01).
54 On the prevalence and societal role of female-authored religious texts, see Melnyk, “Religious Genres”; Brekus, “Writing Religious Experience: Women’s Authorship in Early America.”
55 Dumond, Annie Nelles; or, the Life of a Book Agent, 223.
56 Coghlan, Memoirs of Mrs. Coghlan, 40–41; Fairfield, The Autobiography of Jane Fairfield, 204.
57 Fairfield, The Autobiography of Jane Fairfield, 5.
58 For instance, Elizabeth Gooch describes purchases (“an old villainous cabriolet, for which I paid four times its value”); Mary Eleanor Bowes, Countess of Strathmore recounts her thoughts on financial agreements (“to Mr. Mylne I wrote a civil letter, telling him…that if it was absolutely necessary, my regard for Mrs. Mylne…would induce me to join with him in being security to his creditors”); and Harriet Martineau details preferred financial arrangements (“I told all the tradesmen that I would not deal with them on any other terms than ready money payments…. I began with the house itself, offering to pay down £100 every alternate month, on condition that the workpeople were paid weekly”). Gooch, The Life of Mrs. Gooch, 2:18; Bowes, The Confessions of the Countess of Strathmore, 87; Martineau, Harriet Martineau’s Autobiography, 2:231.
59 Warren, Women, Money, and the Law, 3.
60 We applied the stoplist noted above to the word frequencies analysis.
61 Underwood, Bamman, and Lee, “The Transformation of Gender in English-Language Fiction.”
Additional information
Notes on contributors
Whitney Arnold
Whitney Arnold is Assistant Professor of Comparative Literature and Medicine at the University of California, Los Angeles (UCLA). She works primarily on eighteenth- and nineteenth-century autobiographical texts written in English and French, as well as in the area of health humanities. Currently, she is at work on a book project exploring embodiment and authorial narration in autobiographical texts, as well as a collaborative clinical project investigating the use of life narratives in medical spaces.
Benjamin Niedzielski
Benjamin Niedzielski is a Senior Software Engineer at B Capital, where he splits time between Machine Learning and Data tasks. He has MAs in Indo-European Studies (University of California, Los Angeles) and Classics (University of Kansas), and a Sc.B. in Mathematics-Computer Science (Brown University).
Nick Schwieterman
Nick Schwieterman is a PhD student at University of California, Los Angeles, where he studies Information Studies while working with Digital Humanities projects through the Scholarly Innovation Lab. He has a Master of Library and Information Studies (University of California, Los Angeles) and a BA in Philosophy (University of Cincinnati).