Quantitative Analysis of Gendered Assumptions in a Nineteenth-Century Women’s Encyclopedia:

Introduction
This paper quantifies textual patterns relating to gendered assumptions in a fairly unique text, an entire “women’s encyclopedia” from 1830’s Germany, which at 10 volumes and 1,461,000 word tokens was of comparable size to contemporary general encyclopedias, but written and marketed for a female audience. We perform experiments on classifying gender of biographical entries and querying a specific textual feature, calendar dates, with context from comparison 19
th-20
th century encyclopedias from the EncycNet corpus.

Encyclopedias in the European tradition were “an element of culture and peoples’ lives” (Loveland, 2019), but while encyclopedias invite interpretation (Einbinder, 1966), their length challenges non-digital scholarship; digital humanities thus “holds promise for the study of encyclopedias” (Loveland, 2019).
The
Damen Conversations Lexikon (“Ladies’ Conversations Encyclopedia,” hereafter
DamenLex) has been the subject of little scholarly analysis. Roßbach (2015) writes that the first edition of the
DamenLex “primarily aims to act as a behavioral guide for virtuous women,” Translation of quotations by Roßbach (2015) and Schaser (2016) in this paragraph are by us. but Schaser (2016) asserts that this “little known” encyclopedia is “a treasure trove for questions of cultural history.”
DamenLex editor Carl Herloßsohn (1804-1849) explained its content selection in gendered assumptions and value judgments, and the extent to which the
DamenLex actually followed Herloßsohn’s stated goals is the starting point of our research questions.

Figure 1. A sample page and illustration from the
Damen Conversations Lexikon

Related work
Distant reading the
DamenLex for ideological traces relates to issues of women and gender in 19
th-century Europe, women’s education and achievement, and the history of books not
by, but marketed
for women, in which women’s access to the written word, by controlling literacy and access to reading material, has been a source of anxiety (Jack, 2012). We would thus expect the
DamenLex to display evidence of two opposing forces: women’s education and controlling ideology through explicit or implicit gender presumptions within the content, stylistics, and selection of topics.

Perceptions of women readers can be traced through such texts as Ovid’s books addressed to women (e.g.
Ars Amatoria) and the Confucian
Four Books for Women (Mingqi, 1987). Eighteenth- and nineteenth-century texts for women on etiquette and conduct were prescriptive, supporting notions of “ideal womanhood” (Hemlow, 1960; Darby, 2000). The nineteenth century in which the
DamenLex was published was “a golden age for reading, and for women’s reading in particular,” per Jack (2012), as the growth of industrialization, printing and publishing were “accompanied by wide-ranging debates about what women [...] should be encouraged to read, or discouraged — even prevented — from reading.”

Gendered assumptions
Gender bias in biographical entry subjects is an unfortunate theme in the history of the encyclopedia, with the 11
th Britannica, for instance, including an entry on Pierre Curie but not Marie Curie (Thomas, 1992), and Bamman and Smith (2014) estimated that only 14.8% of biographical entries in Wikipedia have women as subjects.

The
DamenLex’s editor explicitly included many biographies of women to appeal to women readers, suggesting experiments to classify the gender of the ~800 biographical entries in the
DamenLex and almost 44,000 in comparison encyclopedias. The encyclopedias for our experiments are part of a larger set of historical reference works converted to TEI (Hagen et al., 2020): . To classify biographical entries, we trained a bag-of-words-based SVM classifier to label entries as either biography, place, object or abstract concept (with an accuracy of 0.92 for biographies). To classify the gender of each biographical entry, a rule-based approach based on Reagle and Rhue (2011) compares the ratios of male and female personal pronouns in the entry (e.g. sein/his, ihr/her). Only entries longer than 20 tokens were classified, and we only proceeded with entries for which a gender was identified. The amount of unclassified entries is low for
Brockhaus 1837 (about 1%), where entries tend to be relatively long, but higher in
Brockhaus 1911 (about 37%), for example.

Brockhaus
1809

DamenLex
1834

Brockhaus
1837

Herder
1854

Meyer
1905

Brockhaus
1911

Wikipedia
2014

Wikipedia
2015

Male
948 (95%)
480 (60%)
957 (96%)
5,952 (95%)
26,223 (94%)
7,356 (94%)
85.20%
84.50%

Female
52 (5%)
329 (40%)
42 (4%)
345 (5%)
1,599 (6%)
479 (6%)
14.80%
15.50%

Table 1. Estimate of male and female biographies in historical German encyclopedias. Wikipedia results reported by Bamman and Schmith (2014) and Graells-Garrido et al. (2015)

Figure 2. Boxplot visualizing the entry lengths of female biographies in tokens in all encyclopedias. The solid, orange line marks the median. Additionally, the median of male biography entry lengths (dotted, blue line) and the median of all entry lengths, including biographies, (dashed, green line) are given

From these results,
DamenLex contains a much higher percentage of female biographies than all other comparison texts (Table 1) including Wikipedia, with female biographies around 40%. Two chi-squared tests (Table 4 in the appendix) reveal that there is a highly significant relationship between encyclopedia and the amount of entries on women only when the
DamenLex is included in the test, confirming our hypothesis. A similar gender disparity is observed in entry lengths of of male and female biographies in the
DamenLex (Figure 2). We calculated Mann-Whitney U tests for article lengths of female biographies compared to similar sized samples from male biographies for all encyclopedias (see Table 3 in the appendix). Only the
DamenLex's and
Brockhaus' 1911 lengths of female biographies are not significantly shorter than the male entries.

Did
DamenLex devote biographies to the same “notable” women as contemporary encyclopedias? The overlap is quite low: of 329 entries on women in
DamenLex, only about 65 appear in at least one other encyclopedia, confirming that the editors of the comparison encyclopedias had a different perception of who "important women" were.

Finally, most frequent words in
DamenLex biographical entries provide insight into content differences. Among the 15 most frequent nouns in female biographies are role labels such “daughter,” “wife,” and “mother,” and family relations such as “husband” and “child.” In male biographies, in contrast, only “father” and “son” appear in the 20 most frequent nouns, while references to artistic production such as “poet,” “poem,” “opera,” and “writing” fill out the rest. Such artistic terms can also be found in female biographies, only at lower frequency ranks. Among the 50 most frequent adjectives in male entries, only a handful do not appear in the 100 most frequent adjectives in female entries: “tremendous,” “glowing,” “musical,” and “exquisite.” The first two words are typical terms to describe sublime aesthetic experiences, which aligns with contemporary gendered assumptions about aesthetics.

Herloßsohn wrote that a “romantic representation” of the subjects in the
DamenLex was desired: “not a tiresome enumeration of facts and the course of time, but a lively, rapidly gliding painting [...] should be given.” We thus hypothesize that the amount of calendar dates will be far lower in its entries. To investigate, we tagged encyclopedia entries with heideltime,

a multilingual temponym tagger. As Table 2 shows, the amount of dates is indeed far lower in
DamenLex than comparison encyclopedias, confirming the gendered assumption that hard facts such as calendar dates were considered undesirable for women readers. Verification via the chi-squared test (Table 4 in the appendix) results that there is no statistical difference between the encyclopedias concerning the amount of dates, however.

21
37
37
37
37
37
37

Brockhaus
1809

DamenLex
1834

Brockhaus
1837

Herder
1854

Meyer
1905

Brockhaus
1911

Dates
1.29%
0.73%
1.38%
1.87%
3.81%
4.32%

Table 2. Relative amount of dates found in a sample of 256 entries of similar length (about 100,000 tokens overall) per encyclopedia

Conclusion
By quantifying the ratios of male/female biographical entries in the
DamenLex and comparison encyclopedias, comparative length of biographical entries, and a query of calendar dates in the texts, we provide new knowledge and add historical context to vibrant ongoing research on gender bias in encyclopedias (including Wikipedia). We agree with Schaser (2006) that the “little known” encyclopedia of the
DamenLex is “a treasure trove for questions of cultural history,” and have presented evidence that distant reading of gender distribution in biographical entries and content presentation can reveal gendered assumptions in the text. This paper will include these and other experiments to quantify gendered assumptions in encyclopedia texts, and could support future work in gender bias in not only historical but also contemporary encyclopedias.

Appendix

Brockhaus 1809

U = 737.0,
n1 =
n2 = 52,
p < .001

Brockhaus 1837

U = 758.5,
n1 =
n2 = 42,
p < .001

Brockhaus 1911

U = 112124.5,
n1 =
n2 = 479,
p = 0.27

Herder 1854

U = 49367.0,
n1 =
n2 = 345,
p < .001

Meyer 1905

U = 728612.0,
n1 =
n2 = 1599,
p < .001

DamenLex 1834

U = 57677.5,
n1 =
n2 = 329,
p = 0.93

Table 3. Results of the one-sided Mann-Whitney U tests (
p<.01) to confirm or reject the hypothesis whether male biography entries are significantly longer than female biography entries per encyclopedia

Number of entries (incl. DamenLex)
χ2 (5,
N = 44762) = 1635.9,
p < .001

Number of entries (excl. DamenLex)
χ2 (4,
N = 43953) = 7.7,
p = .1

Number of dates
χ2 (5,
N = 600) = 5.03,
p = .41

Table 4. Results for the chi-squared tests (
p<.01) for the amount of entries on men and women as well as the amount of dates (and non-dates, in percent) per encyclopedia. For the dates, we opted for choosing the percentages over the raw counts, as the sample size makes the interpretation of the otherwise very low p-values difficult

Bibliography

Bamman, D. and Smith, N. A. (2014). Unsupervised Discovery of Biographical Structure from Text.
Transactions of the Association for Computational Linguistics, 2: 363-76.

Darby, B. (2000). The more things change… the rules and late Eighteenth-Century Conduct Books for Women.
Women’s Studies, 29(3).

Einbinder, H. (1966). The Myth of the Britannica.
Science and Society, 30(1).

Graells-Garrido, E., Lalmas, M. and Menczer, F. (2015), First Women, Second Sex: Gender Bias in Wikipedia.
Proceedings of the 26
th
ACM Conference on Hypertext & Social Media, pp. 165-74.

Hagen, T., Ketzan, E., Jannidis, F. and Witt, A. (2020). Twenty-two Historical Encyclopedias encoded in TEI: a new Resource for the Digital Humanities.
Proceedings of the 4
th
joint SIGHUM Workshop on Computational Linguistics for Cultural Heritage, Social Sciences, Humanities and Literature, pp. 112-20.

Hemlow, J. (1960). Fanny Burney and the Courtesy Books.
PLMA, 65(6).

Jack, B. (2012).
The Woman Reader. New Haven: Yale University Press.

Loveland, J. (2019).
The European Encyclopedia: From
1650 to the Twenty-first century. Cambridge: Cambridge University Press.

Mingqi, Z. (1987). The Four Books of Women: Ancient Chinese Texts for the Education of Women.
B.C. Asian Review, 1(1).

Reagle, J. and Rhue, L. (2011). Gender Bias in Wikipedia and Britannica.
International Journal of Communication, 5(0).

Roßbach, N. (2015).
Wissen, Medium und Geschlecht: Frauenzimmer-Studien zu Lexikographie, Lehrdichtung und Zeitschrift. Frankfurt am Main: Peter Lang International Academic Publishers.

Schaser, A. (2006). Rezension zu: Herloßsohn, Carl (Hrsg.): Damen Conversations Lexikon. Neusatz und Faksimile der 10-bändigen Ausgabe Leipzig 1834 bis 1838. Berlin 2005.
H-Soz-Kult.

Thomas, G. (1992).
A Position to Command Respect. Woman and the Eleventh Britannica. Metuchen, NJ: Scarecrow Press.