Novels (n = 158) by lesbian authors showed minor signs of psycholinguistic masculinisation; novels (n = 167) by homosexual men had a female-typical psycholinguistic pattern

Sexual Dimorphism in Language, and the Gender Shift Hypothesis of Homosexuality. Severi Luoto. Front. Psychol., May 31 2021. https://doi.org/10.3389/fpsyg.2021.639887

Abstract: Psychological sex differences have been studied scientifically for more than a century, yet linguists still debate about the existence, magnitude, and causes of such differences in language use. Advances in psychology and cognitive neuroscience have shown the importance of sex and sexual orientation for various psychobehavioural traits, but the extent to which such differences manifest in language use is largely unexplored. Using computerised text analysis (Linguistic Inquiry and Word Count: LIWC 2015), this study found substantial psycholinguistic sexual dimorphism in a large corpus of English-language novels (n = 304) by heterosexual authors. The psycholinguistic sex differences largely aligned with known psychological sex differences, such as empathising–systemising, people–things orientation, and men’s more pronounced spatial cognitive styles and abilities. Furthermore, consistent with predictions from cognitive neuroscience, novels (n = 158) by lesbian authors showed minor signs of psycholinguistic masculinisation, while novels (n = 167) by homosexual men had a female-typical psycholinguistic pattern, supporting the gender shift hypothesis of homosexuality. The findings on this large corpus of 66.9 million words indicate how psychological group differences based on sex and sexual orientation manifest in language use in two centuries of literary art.

Discussion

A corpus of 694 novels comprising 66.9 million words spanning more than two centuries of literary art was compiled to determine the extent to which heterosexual male and female authors, and homosexual male and female authors as well as a small sample of bisexual female authors, produced psycholinguistic outputs that differed in predictable ways. The results indicated significant sexual dimorphism6 in the language used in literary fiction written by heterosexual male and female authors, consistent with predictions based on cognitive neuroscience, psychology, and evolutionary science, while also providing support for the gender shift hypothesis of homosexuality (Abé et al., 2021; Luoto et al., 2019a; Luoto, 2020a). The gender shift hypothesis of homosexuality was strongly supported in homosexual males—who produced female-typical psycholinguistic outputs—whereas the evidence among homosexual female authors was substantially weaker, as they showed only a minor psycholinguistic shift in the heterosexual male direction.

While writers and readers, and speakers and listeners, have long been interested in how men and women may use language in slightly-to-vastly different ways, this study helps to clarify the existence, magnitude, and possible psychological underpinnings of sex differences in language use, which appear in areas over which writers would not be exercising sex-conscious psycholinguistic control. It would be difficult to conceive, for instance, how male authors might consciously increase the frequency with which they use articles (‘a,’ ‘an,’ and ‘the’) because they associate such language use with some nebulously “desirable” characteristics related to their ideas of “masculinity.” It is difficult, in other words, to explain the findings with the social role theory of gender roles, which would further struggle to provide a plausible explanation for homosexual male authors’ female-typical language use. If homosexual males were socialised into the male gender role, why do they use language in a way that resembles heterosexual women’s language use? To the extent that these findings represent non-conscious, natural ways of using language, they also suggest that homosexuality is not a conscious choice (Luoto et al., 2019a; Swift-Gallant et al., 2019; Bogaert and Skorska, 2020). It is highly unlikely, after all, that homosexual male authors have consciously chosen to write in a more female-typical way, of which they could have had limited notion at the level of psycholinguistic minutiae.

While some people argue that socialisation into gender roles underlies sex differences in humans, this hypothesis becomes implausible when considering the biological, developmental, neuroscientific, and cross-cultural evidence more broadly (Christov-Moore et al., 2014; Schmitt, 2015; Janicke et al., 2016; Archer, 2019; Del Giudice, 2019; Luoto et al., 2019a; Atari et al., 2020; Stoet and Geary, 2020; Luoto and Varella, 2021). Most sex differences in personality are of a higher magnitude in more gender-egalitarian countries than in less gender-egalitarian countries, which is the opposite of what the gender role hypothesis would predict (Schmitt et al., 2008; Falk and Hermle, 2018; Atari et al., 2020; Stoet and Geary, 2020). Furthermore, since evolutionary processes pre-date social conceptualisations of gender roles by millions of years, a full explanation of socialisation into gender roles and the effects it has on sexually differentiated traits and behaviours would need to account for how evolutionary processes act as precursors to gender roles (Janicke et al., 2016; Archer, 2019; Luoto and Varella, 2021; Luoto et al., 2021).

Ultimately, psychobehavioural sex differences arise from sexual selection, sexual differentiation of the mammalian brain, sexual division of labor, and their interactions (Figure 7) (Luoto and Varella, 2021). Sexual selection and sex differences in parental investment have exerted and currently exert selection pressures on status-striving and power-seeking among men more than in women (Luoto, 2019), contributing to men’s higher competition, aggression, risk-taking, sociosexuality, and men taking on more leadership positions than women, particularly at higher organisational and societal levels (Luoto and Varella, 2021). Sex differences in parental investment and mating competition coevolve with parental care specialisation, which can partially contribute to such psychobehavioural sex differences as found in empathising, people orientation, risk-taking, neuroticism, mate choice, sociosexuality, aggression, violence, leadership, and dominance (Archer, 2019; Henshaw et al., 2019; Luoto et al., 2019a; Luoto and Varella, 2021). Sexually dimorphic ultimate evolutionary functions exert an influence on psychobehavioural sex differences via various biological mechanisms, leading to sexually dimorphic language use which, further down the evolutionary–developmental trajectory, also reflects other known psychobehavioural sex differences (Figure 7).

Figure 7. The evolutionary–developmental origins and proximate mechanisms underlying psychobehavioural sex differences, including those in language use. Figure adapted from Luoto and Varella (2021).

Comparative research provides further evidence against social role theories of human sex differences. Evidence of sex-biassed treatment by others (equivalent to what proponents of social constructionist hypotheses think of as socialisation into gender roles in humans) is lacking in non-human animals. Behaviours of mothers toward female and male offspring show little to no difference in the few species that have been studied (Lonsdorf, 2017), yet such species show sex differences in behavioural, physical, and social development that resemble those found in infant humans (Christov-Moore et al., 2014; Lonsdorf, 2017; Archer, 2019). These include sex differences in species-typical behaviours such as grooming, playing, object manipulation, and extractive foraging (Lonsdorf, 2017). Immature chimpanzee males, for instance, engaged in more object-oriented play than females (Koops et al., 2015). Under 5-week-old newborn rhesus macaque females that were raised in a controlled postnatal environment looked more at computer-generated faces of other rhesus macaques and engaged in more affiliative behaviour with a human caregiver than newborn rhesus macaque males did (Simpson et al., 2016). Similar findings have been reported in humans: 12-month-old female infants showed a higher relative preference for a moving face over a moving car than males did (d = −0.64) (Lutchmaya and Baron-Cohen, 2002). In humans, vervet monkeys, and rhesus macaques, females have been observed playing longer with dolls and plush toys, while males play longer with wheeled toys (Christov-Moore et al., 2014). Asian elephant females have a tendency to be more social and gregarious than males (Seltmann et al., 2019). In humans and non-human primates, females engage in social grooming more often than males (Lonsdorf, 2017). In hamsters and humans, females find same-sex social interactions more rewarding than males do. Oxytocin plays a similar mechanistic role in social reward processing in a number of species, suggesting that sociality and sex differences in sociality may arise from a common evolutionary origin (Feng et al., 2015; Hung et al., 2017; Borland et al., 2018).

Furthermore, evolutionarily conserved hormonal mechanisms, such as testosterone, are associated with language use and other sexually dimorphic phenotypes (Hoskin and Meldrum, 2018; Mascaro et al., 2018; Archer, 2019; Luoto et al., 2019a), providing a biological basis for the emergence of sexually differentiated traits. Many lines of research, including longitudinal research in humans, support this theory. While hormone exposure significantly predicted gender development in girls, mothers’ socialisation to feminise the daughters had negligible effects: women exposed to more testosterone in prenatal development showed masculinised behaviours in adulthood despite parents’ socialisation efforts to have the daughters behave in a more feminine way (Udry, 2000).

Evidence for the relationship between testosterone and many sexually dimorphic phenotypes spans several different areas of research (Björkqvist, 2018; Hoskin and Meldrum, 2018; Luoto et al., 2019a; Muñoz-Reyes et al., 2020). It is noteworthy that psychological research has not found reliably occurring differences in anger frequency; instead, sex differences have been found in verbal and physical aggression, both being higher in men (Archer, 2019). Thus, the slightly higher frequency of anger-related words in male authors’ novels (d = 0.32, Figure 1) does have some equivalents in psychological research. The use of anger-related words is positively correlated with circulating testosterone levels and with polymorphisms in the androgen receptor gene (Mascaro et al., 2018), which make cells more susceptible to the masculinising influence of testosterone. These findings indicate the existence of a plausible biological mechanism (Geniole et al., 2019; Luoto et al., 2019a) which creates sex differences in anger-related language use as well as other psychobehavioural sex differences, including people–things orientation, risk-taking, and theory of mind (Khorashad et al., 2018; Luoto, 2020b; Vaskinn et al., 2020; Luoto and Varella, 2021). Furthermore, the finding of higher anger-related words and sexual words in lesbian authors relative to heterosexual women is consistent with existing findings on psychobehavioural masculinisation in non-heterosexual women, including higher sociosexuality, sensation-seeking, psychopathy, and incarceration rates compared with heterosexual women (Luoto et al., 2019a,b) (though see Gil-Llario et al., 2015 who reported lower sexual sensation seeking in self-identified lesbians than in heterosexual women).

An important contribution of this study was the ability to predict and explain sexual dimorphism in language using psychology and cognitive neuroscience. A related major result is that prior research on sex differences and sexual orientation differences in these fields have clear equivalents in the psycholinguistic outputs of authors writing literary fiction decades and centuries ago, suggesting that psychological sex differences may be relatively stable across time and across different domains—that is, they manifest not only via questionnaires, psychological tests, and behavioural measures, but also in the artistic and linguistic forms of imaginary self-expression enabled by literary fiction; and they manifest not only in contemporary population-based samples, but also in the highly specialised sample of writers of canonical literary fiction from decades and centuries ago. This coherence across different areas of research and across different time periods allays concerns that could be raised about the generalisability of the current findings.

Limitations

A clear limitation of this study was that the analyses were conducted only on English-language material. Future studies are therefore encouraged in other languages to provide an estimate of the generalisability of these findings across other languages. Corresponding results have, however, been reported in a number of languages using various literary and non-literary sources, though few studies have distinguished between writers of different sexual orientations (cf. Argamon et al., 2009; Johannsen et al., 2015; Chen et al., 2018; Koolen, 2018).

Another potential limitation of this study is that effect sizes can become biassed because of range restriction, which refers to a process in which the participants of a study are, directly or indirectly, selected from the original population on the basis of their personal characteristics (Del Giudice, 2019). In the current case, all samples of novels are likely to suffer from range restrictions as the novels were not sampled at random from all novels ever written by heterosexual or homosexual men and women; rather, canonical and prizewinning novels were mostly used, although the non-heterosexual samples also included less well-known novels because of the necessity to reach a large enough sample size. What is more, it may not be possible to directly extrapolate these findings on novelists to the respective groups of all lesbian women or all gay men or all heterosexual women and men. That is because only a small subset of each of these groups is likely to write and publish novels, particularly novels that reach a canonical status; thus, the sampling of such individuals may not be generalisable to the full sample of non-novelists in each group. This limitation can be addressed by comparing the present findings with existing findings on similar group differences that have been acquired using other kinds of methodologies and sampling protocols on non-novelists. Thus, to the extent that the current findings are consistent with the findings of other sex difference and sexual orientation difference studies (which they generally tended to be), the sampling problem of focussing only on novelists is mitigated.

This study was also limited in the sense that the heterosexual sample was drawn from canonical and prize-winning authors’ works: these culturally esteemed works may not generalise to the other 99% of literature ever written (Moretti, 2005, 2013). Furthermore, as most of the non-heterosexual sample comprised works that were not canonical nor prize-winning (necessarily so because of the difficulty of obtaining such samples that would have been large enough for adequate statistical power), I cannot rule out the possibility that the psycholinguistic differences observed in this study between authors of different sexual orientation could have been partially driven by the differences in canonicity and/or literary prestige between the samples. Nevertheless, the likelihood of this possibility is somewhat attenuated as the findings largely aligned with predictions which arose from existing psychological and linguistic research as well as theory from evolutionary human science. To explain the findings as resulting from differences in canonicity, it would be necessary to posit how the sampling strategy used for homosexual male and female authors biassed language use in opposite directions in each sample in a manner which is consistent with the theoretical hypotheses and predictions. Although the non-heterosexual samples comprised novels that were published much more recently than the novels in the heterosexual samples, those differences in publication year were controlled for in all analyses. Correlations between publication year and all psycholinguistic outcome variables are available in the Supplementary Materials, as are correlations between authors’ age at publication and all the psycholinguistic outcome variables (Supplementary Tables 8, 10, 11).

The group differences reported in the study could be somewhat attenuated because of the diversity of author demographics included in the samples of novelists. For example, authors were sampled from more than five countries. Authors’ age in the heterosexual sample of 304 novels varied from 24 to 68, while year of publication varied from 1801 to 2017 (Luoto and van Cranenburgh, 2021). Likewise, although the sample comprised mainly Caucasian authors, the full sample included authors whose racial backgrounds were Latino, African–American, Asian, Native American, and mixed (see Supplementary Tables 2–6 for details). Though making the sample more representative of the respective authors’ populations, this sample diversity may have caused more variation in the psycholinguistic outcome variables than studying more homogenous author populations, and this higher variation could have resulted in smaller effect sizes (as in Newman et al., 2008). Thus, the effect sizes reported in this study could be underestimates, and having less variation in publication year, age, race, ethnicity, and nationality can lead to detecting larger effect sizes.

The authors’ sexual orientation was determined based on biographical information, including information on the sex of any partners (married or otherwise) that the authors had or any self-identification related to sexual orientation that the authors may have made publicly known (Luoto and van Cranenburgh, 2021). The authors’ sexual orientation for the purposes of this study is therefore based on both manifest sexual behaviour as well as self-identification; however, both sexual behaviour and sexual orientation may undergo various changes over time, especially in non-heterosexual women (Luoto et al., 2019a,b), which is why the use of an aggregate measure of lifetime sexual behaviour and sexual orientation may not accurately track a person’s sexual behaviour or sexual orientation at any single point in time. Sexual orientation is used in this study as an instructive overall indicator of an author’s sexual behaviour and attractions over their lifetimes, and as such may be limited by the availability of such information in biographical material (Luoto and van Cranenburgh, 2021).

One reason why the gender shift hypothesis was not strongly supported in homosexual female authors could have been because it was not possible to control for butch/femme differences in the sampled authors. This would have been an important addition to the study. After all, there can be significant variation in the masculinity/femininity of non-heterosexual women, and research on non-heterosexual women should take this variation, conceptualised, e.g., via butch/femme categories, into account by analysing different groups of non-heterosexual women separately (Luoto et al., 2019a,b). However, in this research on literary fiction, it would have been difficult (if not impossible) to study women’s self-identification as masculine butches or feminine femmes because many of the authors had passed away.