Men use more complex language than women, but the difference has decreased over time: a study on 120 years of written Dutch

: Differences in language use between men and women have been studied intensely. We take stock of the findings and venture into less charted territory. First, we broaden the scope from well-known lexical features to the domain of syntax. Second, we take a diachronic perspective, looking at changes between 1880 and 1999. Based on a corpus of written Dutch by prolific writers, we find statistically robust differences: men ’ s style can be characterized as more complex, with the exception of lexical diversity. Through the years, however, there is growing convergence for all linguistic metrics. In the discussion section, we look at different explanations for the observed trends

The upshot of the vast literature is that there are noticeable, statistically robust differences, both at the phonetic, lexical, and morphosyntactic levels.Indeed, gender has been identified as a pervasive predictor of differences in language use (Coates 1998), surpassing many other social variables (Biber et al. 1998;Härnqvist et al. 2003;Newman et al. 2008;Rayson et al. 1997).Concentrating on stylistic differences in the use of lexicon and grammar, various attempts have been made to summarize the differences into a restricted number of dimensions, preferably one.In her pioneering work, Lakoff (1973) argues that women are more 'considerate' to the interlocutor, making more extensive use of hedges, overly polite forms and apologies.This is echoed by later research claiming that women's language is overall characterized by tentativeness and uncertainty (Newman et al. 2008): women tend to use more uncertainty verbs (Hartman 1976;Mulac and Lundell 1994;Poole 1979), modal auxiliary verbs (McMillan et al. 1977;Pennebaker et al. 2003), psychological process references (Newman et al. 2008), emotion words (Gleser et al. 1959;Mehl and Pennebaker 2003;Mulac and Lundell 1994;Mulac et al. 1986) and sentence-initial adverbs (Mulac and Lundell 1994;Mulac et al. 1986Mulac et al. , 1988Mulac et al. , 1990)).According to Tannen (1990;1994), women engage more in 'rapport talk', attempting to set up emotional and social bonds, whereas men engage more in 'report talk', centering around the exchange of information.This rapport versus report division is akin to Biber and Burges (2000) characterization of female 'involved' style versus male 'informative' style (Brownlow et al. 2003;Keune 2013).Most empirical research confirms men's speech to be characterized by this 'information emphasis' (Biber and Burges 2000).This general label captures a wide range of fairly consistent differences: on the lexical level, men use more references to quantities and locations (Gleser et al. 1959;Mulac and Lundell 1986).Men tend to use more prepositional phrases, and more articles (Gleser et al. 1959;Mehl and Pennebaker 2003).Women on the other hand use more intensifying adverbs, as part of an involved discourse (Crosby and Nyquist 1977;Lapadat and Seesahai 1978;McMillan et al. 1977;Mulac and Lundell 1986;Mulac et al. 1986).
A related approach is to look not at a monodimensional continuum (rapport vs. report, involvement vs. engagement), but to look at complexity.At the lexical level, it has been repeatedly established that men use longer words, less frequent words, and display a higher type-to-token ratio (Biber et al. 1998;Brownlow et al. 2003;Härnqvist et al. 2003;Keune 2013;Mulac et al. 1988;Newman et al. 2008).It is currently not clear whether the higher complexity extends to the morphosyntactic level as well, though a recent study on adolescents' online language indeed finds this to be the case (Verheijen and Spooren 2017).
In spite of this vast number of studies and the agreement in the dimension reduction that different authors propose (rapport vs. report, involved vs. informative), one may wonder how robust these differences are across time and across contexts.Doubts have been raised to the heterogeneity of the data on which the analyses have been based.Indeed, the results have been gathered from quite different contexts, ranging from interviews (see for instance Haas 1979), public speeches (Mulac and Lundell 1986), descriptions of photographs (Mulac and Lundell 1994;Mulac et al. 1986), secondary school essays (Mulac et al. 1990), role plays and spontaneous interaction (see for instance Mulac et al. 1988), and social media conversations (Hilte et al. 2020(Hilte et al. , 2022;;Verheijen and Spooren 2017).Some authors have been eager to proclaim stable gender differences on the basis of artificial data jeopardizing ecological validity.Furthermore, most studies have concentrated on present-day language only, and have ignored the question of whether the observed differences are peculiar to this day and age, or are timehonored differences.Diachronic studies often focus on either a very particular construction (e.g., Palander-Collin 1999), focus on a short time span and/or a limited number of writers (Alonso-Almeida and González-Cruz 2012).A rare fullblown quantitatively oriented diachronic study, by Biber and Burges (2000), shows a development towards a more 'involved' style in recent times.By contrast, another, very recent study (Degaetano-Ortlieb et al. 2021) shows that women converge to men, at least in the 18th century.
In this study, we set out to apply a diachronic perspective, to measure both lexical and grammatical features.Looking at 120 years of genre-stable written language, we want to gauge to what extent the differences noted in the literature are stable: have men and women converged or diverged over the past years?We will focus on complexity, both at the level of the lexicon and at the level of morphosyntax.In the discussion section, we will reflect on what these trends reveal about the explanations that have been proposed for gender differences.
Our aim is to combine research on gender differences in language use and research on the diachrony of stylistic complexity.Gauging stylistic complexity is greatly helped by the existence of NLP tools to automatically quantify a range of metrics, both at the lexical and at the syntactic level, of texts.Meanwhile, the increasing availability of well-documented reference corpora, enriched with metainformation about the authors, allows for research designs where statistically robust effects of age, gender and other socially relevant variables can be measured on these linguistic metrics.
To our knowledge, no research has been conducted so far looking specifically at the stability of gender differences in stylistic complexity through time.Previous research on the complexity of literary material focused mainly on (i) the simplification of journalistic prose (see Westin and Geisler 2002 for a complexity analysis on editorial comments in British newspapers) and (ii) the comparison of the complexity of different genres (journalistic prose vs. novels) (see Stajner and Mitkov 2012 for a comparative study of both fiction and newspaper material published in Britain and the US).For Dutch, Ham et al. (2018) present a diachronic, comparative analysis of stylistic complexity of literary novels and newspapers, focusing on four dimensions of complexity: (a) word complexity, (b) sentence complexity, (c) cohesion, (d) references to people.They conclude that the complexity of Dutch newspapers decreased through time, both at the word level and at the sentence level.The situation turns out to be more complex in novels, where there is no significant decrease in complexity.

Methods
To answer the question on the interaction of gender and time on lexical and syntactic complexity, we made use of the newly compiled Dutch C-CLAMP corpus (Piersoul et al. 2021).This corpus consists of articles published in Dutch literary and cultural periodicals, from 1837 to 1999.These periodicals include journalistic texts, shading into academic essays.From this corpus, we focused on the time span 1880-1999.The reason to clip the corpus and ignore the first 43 years is that there is a marked underrepresentation of women, which impedes comparison.
We divided the corpus in 12 decades, and for each decade we extracted one randomly selected text written by each of the five most prolific male and female authors.Though a binary operationalization of gender arguably does not do full justice to the real-life complexity, we stuck to a male versus female division to enhance comparability with other studies, and to avoid unresolved issues in attributing gender in historical writers.We retrieved or attributed gender on the basis of the Statbel dataset, see statbel.fgov.be.The motivation to opt for the most prolific authors per decade is twofold: first, the number of published articles serves as a proxy for the author's popularity.We assume that more prolific authors are more representative for diachronic trends, as they are more likely to weigh on the normative rules.The second reason has to do with feasibility: the software we relied on runs on a remote server, is computationally heavy, as the output standardly calculates a wide range of metrics, and runtime increases rapidly when longer or more text are fed, precluding the use of the entire corpus for the analysis.To make the linguistic measures reliable, we set a threshold of 500 words per article.This yielded 117 texts, consisting of 80,611 tokens in total.Of these 117 texts 59 were written by female authors and 58 by male authors (see Appendix A for an overview of the authors and their publication rate per decade).
These texts were then analyzed for various complexity metrics, relying on fully automatic data annotation using Tscan software, a software tool for analyzing Dutch text (Pander Maat et al. 2014).
As said, Tscan gives a plethora of linguistic metrics.From these we selected those that are straightforwardly associated with complexity, both on the lexical and the syntactic level.Some of these metrics have to do with the semantic distinction between 'concrete' and 'abstract'.We assume general, abstract words to be more complex.They tend to be chronologically secondary in acquisition, both at the ontogenetic and at the phylogenetic level.Concreteness ratings and estimated age of acquisition for Dutch (Brysbaert et al. 2014), correlate with −0.42 (Pearson correlation, p < 0.001), and abstract words tend to derive from concrete words (Kronasser's Law, see Collinge 1985).Abstract words also demand greater cognitive effort (Gianico-Relyea and Altarriba 2012).
We also assume a negative correlation between complexity and word frequency, and a positive correlation between the length and the number of morphemes a word consists of, and its complexity.At the syntactic level, we assume a positive correlation between the number of finite verbs and the degree of multiple subordination on the one hand, and complexity on the other.
The metrics under scrutiny are the following: (a) At the level of lexical complexity: (i) SIZE, WORD LENGTH IN CHARACTERS: average word length in characters (ignoring compounds).(ii) SIZE, MORPHEMES PER WORD: average number of morphemes per word.(iii) WORD FREQUENCY: average word frequency of content words (logarithmically transformed).(iv) LEXICAL DIVERSITY: measure of textual lexical diversity (MTLD McCarthy 2005), which employs a sequential analysis of a sample to estimate a lexical diversity score.It is, in effect, a correction on the Type-Token-Ration (TTR), as the latter is known to be affected by text length, and cannot be used to directly compare texts of different sizes.MTLD goes sequentially through the text and looks at how long it takes, on average, before the TTR drops below a certain threshold, often set at 0.72.Every time the TTR drops below this value, it is reset at 1. Throughout the text, there will be a number of resets.The number of words passed before every reset is then averaged.The higher this number, the longer the type accumulation persists over time, and the higher the lexical diversity can be assumed to be.1 (v) SEMANTICS, GENERAL NOUNS: proportion of general nouns (abstract nouns, referring to non-domain specific language use).For metrics calculating sentence length and word length, we excluded proper nouns.For the word length, we also ignored compounds, for two reasons: first, the demarcation between compounds and phrasal units is blurry.Second, compounds are often transparent, so their length is no longer a faithful representation of their complexity.We added a metric for the number of morphemes per word (a-ii), to account for morphological complexity.
These metrics form the dependent variables.As they are averaged over different sentences per text, they are not integers, but rational numbers, with a different range, depending on the scale of measurement.For comparability of the effect sizes, we z-transformed the linguistics measures.For each of these dependent variables, we built a linear mixed model, the current preferred method in corpus linguistics (see Gries 2015), as well as for experimental studies (Baayen et al. 2008;Jaeger 2008).
Statistical analyses were carried out with the open-source software R (R Core Team, 2020), the package lme4 (Bates et al. 2015) and the package lmerTest (Kuznetsova et al. 2017).Effect plots were drawn with the package effects (Fox 2003;Fox and Weisberg 2013).For data preparation, we used the packages dplyr (Wickham et al. 2021) and VIM (Kowarik and Templ 2016).Effect sizes are represented with two decimal places, p-values to three decimal places.P-values below 0.001 are represented as p < 0.001.
The mixed models tap into our interest in the diachronic dimension of stylistic complexity: we had the gender of the author participate in an interaction effect with the year of publication.We incorporated the individual authors as a random effect, to account for individual differences in style, which is likely to play an important role (Barlow 2013;Guy 2013).We used random intercepts, not slopes, because the short time span of some of the authors' attested texts makes it impossible for the model to detect any notable changes over time within the individual.
Looking at the main effects for GENDER and YEAR, we will be able to detect whether men and women show differences and whether there is a diachronic trend.The interaction term will tell us whether the potential diachronic trend is different for the genders.Rather than relying on a stepwise variable selection procedure on the basis of AIC or significance, which is a dangerous procedure (see Harrell 2015: 67-72;Winter 2020: 276-279), we maintain a theory-driven variable selection, including the interaction term, even when it is not significant.
As we have different dependent variables, we ran a MANOVA model to check whether there is an association of the aggregate of the linguistic measures with GENDER (this is the mirror image of what Bonferroni does for multiple independent variables).This is indeed the case (p = 0.018) (the variable MULTIPLE SUBORDINATION was left out of the MANOVA, because of the many NA values [42.7%]).
For the analysis, we give p-values.As the T-scan metrics need to be calculated per text, each text stands for one datapoint in the linear mixed models.This means that the models will have to be fitted on 117 texts.We refrained from using additive models for non-linearity, as these methods are more data-hungry than what our data allow for, may be more prone to overfitting, and do not yield straightforwardly interpretable coefficients.The random structure will explain much of the variation so that there is less variance left to be explained by the fixed effect.This is a conservative approach, to avoid Type I errors, but it comes at the cost of an increased Type II error risk.

Results
Tables 1 and 2 present the results of the linear mixed models for the lexical and the syntactic outcome variables, respectively.The columns show the estimates and p-values for the explanatory variables GENDER, YEAR, and the interaction between the two.

Stylistic gender differences in language
In Table 1, the main effect of GENDER shows that men use, on average, longer and more general and abstract words.For FREQUENCY, significance was not obtained.Women, on the other hand, score higher on LEXICAL DIVERSITY.Effect sizes are comparable, hovering around 0.5, expressed in standard deviations.The main effect of YEAR shows an increase in all metrics, except for LEXICAL DIVERSITY.Again, effect sizes are comparable, between 0.01 and 0.02.This means that the difference between men and women is equal to the change in half a century to a full century.There was no significant effect of the interaction, except for WORD FREQUENCY.This seems to suggest that the differences between men and women do not change and that both show an increase over time.Visual inspection of the effect plots (Figure 1), shows a different picture, however.We see a converging trend for GENDER over time, even for LEXICAL DIVERSITY, where women scored higher than men.The interaction term is only significant for WORD FREQUENCY because we have a 'crossing interaction'.This also explains why the main effect was not significant.In the low region of the x-axis, WORD FREQUENCY is lower for females than for males.
To test this, we looked at the text from the first 60 years , and the last 60 years  separately, to see how the gender differences play out.We expect to see larger effect sizes in the early years, and no significance in the recent years.This seems to be the case, see Table 3.
For the lexical measures, there seems to be a trend: men tended to use longer and more abstract words, and conversely, women tended to use less frequent words, and have higher lexical diversity.
Table 2 shows the estimates and significance of the syntactic measures.The main effect of GENDER shows that men use more complex syntax.The difference is non-significant for the sentence length, but for the other measures, the effects are similar in size to what we have seen for the lexical measures.For the D-level and the multiple subordination, there is a significant main effect of YEAR, as well as a significant interaction between GENDER and YEAR.Visual inspection of the effect plots  shows that we see diachronic convergence between the two genders on the syntactic measures that reach significance.This is confirmed in Table 4, where we applied the same procedure as with the lexical measures, splitting the dataset in two for the early period versus the later period.Effect sizes are larger for the early period.The difference disappears in the later period.

Discussion
Our case study on stylistic complexity in 120 years of Dutch periodicals partially confirms, partially questions and partially extends previous findings.The main effects for GENDER in the linear mixed-models offer corroboration for the existence of robust gender-related differences.These differences mount to about 0.5 standard deviation between the genders, controlling for the differences in temporal situation.Overall, there is of course no full sexual dimorphism, but the effect sizes are far from negligible and are in line with other cognitive-behavioral traits observed for gender differences (Lippa 2005).Furthermore, the results are in consonance with earlier studies on gender and style, which have characterized the difference as rapport talk versus report talk.Male authors make more use of longer, morphologically more complex, abstract words.Our results include syntactic measures, which have been far less looked at.Taking the lexical and syntactic measures together, the difference can also be viewed as a difference in complexity.With the exception of lexical diversity, male speech is more complex, at different levels.
Note that other studies have found equivocal results for the relation between gender and lexical richness (see for instance Hilte et al. 2020;Yu 2009).One interesting interpretation that Hilte et al. (2020) offer is that the complexity differences are balanced out between the sexes, with men using longer and morphological more complex words, and women using shorter but more varied words.We could rephrase that to read: men exploit syntagmatic complexity in their lexicon, and women exploit paradigmatic complexity.Two factors were non-significant: on the lexical side, WORD FREQUENCY was not statistically discernibly different for the two genders, and on the syntactic side, CLAUSE LENGTH was not different either.For word frequency, the non-significance is puzzling, in view of the crosslinguistic 'Zipfian' correlation between word size and frequency (Pustet 2004;Zipf 1949).Given the significant differences for words size and abstract meaning, we can conclude that female authors in our sample make higher use of infrequent short words with concrete meaning.It remains to be seen what sort of words these are.The effect does not appear to be attributable to the higher incidence of proper names.We ran a separate linear mixed model for the use of proper names (post-hoc), but could detect no significant main effect for GENDER (p = 0.191) and YEAR (p = 0.324), or for the interaction between them (p = 0.628).For clause length, the nonsignificance is not entirely unexpected: other studies report equivocal results as well (Hilte et al. 2020 and references cited there).Clause length is also the least straightforward complexity measure: it can be boosted by the concatenation of parataxis, which is attested at an early age and does not really add to complexity per se.Hypotaxis is arguably more indicative of complexity, and of 'report' style.
What lies behind these gender differences?The precise nature and cause remain elusive and are subject to controversy (see Cameron 1992 andCameron et al. 1989 for a critique).Berryman-Fink and Wilcox (1983) have argued that context plays an important role.Koolen and van Cranenburgh (2017) warn of a serious risk of stereotyping, essentialism and preemptive categorization (for instance, dataset bias, not controlling for confounding linguistic and extralinguistic variables, interpretation bias, taking gender divisions for granted) (see also Koolen 2018).
Different explanations have been proposed in the literature.At the risk of oversimplifying, earlier studies have generally focused on the gender-as-a-socialconstruct account.The differences are explained as the effect of cultural norms on gender and on power dynamics (Cameron 1992;Tanaka 2015).Women's language use is linked to their lack of power (Eckert 1989;Labov 1990;Trudgill 1972).Females are supposed to behave more politely, more considerately, and more compliant with norms (Labov 1990).In the narrative of gender studies and feminism, these norms are enforced by patriarchy, downplaying the role and ambition of women.In such a view, the differences in style are partially held responsible for a wage gap (Mulac 1998), and for the oppression or discrimination of women (Lakoff 1973;Togeby 1992).We cannot rule out the existence of these forces and find them certainly plausible to some extent, but they are often assumed a priori and do not take heed of alternative accounts, which may explain part of the observed tendencies.Moreover, politeness often leads to an increase in linguistic complexity (e.g., Brown and Levinson 1987). 2  Less popular, at least in (socio)linguistic circles, are explanations drawing on insights from evolutionary psychology. 3Under such a view, gender-differentiated behavior can often be related to 'mating strategies'.Though humans are not an across-the-board 'tournament-oriented' species (Prum 2017), and favor long-term bonds, there is extensive evidence for sex differences (see the overview in Buss 2016 and references cited there).Given the highly communicative nature of human behavior, it comes as no surprise that language plays an important role here, as a signal of fitness by an ostentatious display of a costly trait (Darwin 1871;Foolen 2005;Miller 2002;Rosenberg and Tunney 2008).This line of reasoning can explain some of the observed differences in the language use of men and women, and more particularly the complexity difference.On average, men are more likely to use language display, in an attempt to impress women in courtship, even in nonconversational language use.The important part is the presence of an audience, not the presence of an audience that takes turns in the performance.Supporting evidence for the latter statement comes from the fact that non-conversational artistic forms of language, like poetry or stand-up comedy, play a role in courtship and verbal display for mating purposes (Miller 2002: Ch. 10).The role of language in mating strategies is clear from behavioral studies showing that linguistic complexity goes up in mating conditions (Dunbar et al. 1997;Essers and Van de Velde 2020;Rosenberg and Tunney 2008) and that many women select men as sexual partners in considerable part on the basis of verbal skill (Lange 2011).
Applied to our results, an evolutionary psychology account could help explain two observations.First, it gives a reason for the direction of most of the differences, with men's higher complexity scores on most of the linguistic measures.Second, it gives a potential reason for the intriguing difference, with men having more 'syntagmatic' lexical complexity, and women having more 'paradigmatic' lexical complexity: the former kind of complexity is more visible directly, whereas lexical diversity measures can only be gauged indirectly.The same goes for the syntactic measures: building impressive hypotactic edifices can be evaluated directly.If language is indeed partially used for ostentatious display, it is not unreasonable to expect that men exploit the more visible metrics.
We should, however, be careful with evolutionary accounts.As the main effect of YEAR and the interaction of GENDER and YEAR show, the differences between the genders diminish over time.They are by no means deterministic, and there is no ground for 3 In part, the unpopularity of evolutionary accounts may be due to the fact they are reminiscent of obsolete views posited on shaky grounds, often with what we would now perceive as a condescending attitude vis-à-vis women.Even reputable scholars like Otto Jespersen have rather overstretched psychological interpretations of the differences between men and women (see Ch. XIII 'The woman' in Jespersen 1922).For Dutch, similar opinions have been voiced by pioneering psycholinguist Jacques van Ginneken (1913: 521).
biological reductionism.However, this does not imply a refutation of an account in terms of evolutionary psychology.The reduction to biological determinism is a fallacy (see Lange et al. 2014).Like many other traits, there is a complex dynamic between nature and nurture (see e.g., Elliott 2017;Guy 2013;Lippa 2005).Crosscultural or diachronic differences suggest a scenario in which what are in origin biologically explainable differences become culturalized.Moreover, the fact that some differences can be explained by evolutionary psychology, and biological factors, does not imply that there is a difference in linguistic ability.The stylistic differences are a function of the difference in behavior, not of the difference in ability.
The difference in actual verbal abilities is negligible (Elliott 2017: 31).If anything, women appear to outperform men (Miller 2002).What differs is the motivation to use the different constructions, strategies, and stylistic options.This is supported by the observation mentioned above that the effect sizes we observe in our study are more in line with those from gender-based personality differences (Elliott 2017) than with those from verbal intelligence, the latter typically being much smaller. 4 The general pattern that emerges from our data, as visualized in the effect plots of the linear mixed models (Figures 1 and 2) is a diachronic convergence between the sexes.This is presumably related to the historical convergence between sex roles (Carmichael et al. 2014;Delap and Morgan 2013;Holloway 2005;Mahood 1995).In our data, the convergence at the lexical level has been mainly brought about by women converging with men, in line with recent findings by Degaetano-Ortlieb et al. (2021), but in contrast to what one would expect on the basis of Biber and Burges's (2000) finding that recent times have witnessed a development towards a more 'involved' style, which is associated more with women than with men.The convergence by women towards men in our data is presumably related to the sociolinguistic observations that women are more inclined to accommodate to changing norms, especially when these are overtly prescribed (Labov 2001: 293;Palomares et al. 2016: 133;Surkyn et al. forthc.),and in formal texts (rather than conversations or letters to family members) (Degaetano-Ortlieb et al. 2021).The syntactic measures reveal a different picture: here the convergence is apparently due to a change in both genders, or even to the change in men's style over time.
4 An example is the custom of foot binding in historical Chinese culture.Though evidently a tradition that developed through cultural evolution, witness the idiosyncrasy of the habit crossculturally, it seems rooted at least partially in neotenous beauty standards for women (Winegard and Deaner 2014).The same goes for other traits that are relevant to mating: men have a preference for younger women, but exactly how young differs considerably by culture (Prum 2017).In the domain of language, women have higher-pitched voices than men, but the degree to which women overemphasize their pitch and the degree to which men prefer higher-pitched voices in women show notable differences by culture (Cheng 2020;Ohala 1983Ohala , 1984;;Pépiot 2014;Pisanski et al. 2018).

Stylistic gender differences in language 5 Conclusion
In this article, we set out to see whether men and women differ in their language use in journalistic prose (periodicals).We compiled a Dutch corpus of 80 thousand tokens, from 59 female and 58 male prolific authors, from 1880 to 1999.We then looked for a wide variety of lexical/morphological and syntactic measures and inspected possible gender differences and the diachronic stability of these differences Our results indicate that men used a more complex style a century ago, especially in those lexical and syntactic measures that have a direct, visible effect.Over time, the differences diminish, to the point that the two genders are stylistically indistinguishable in the recent part of our corpus.
Our results are compatible with earlier research.First, we replicated earlier studies that find higher complexity in male language.Second, we showed that lexical diversity has a worse fit with this general pattern.Third, we corroborate sociolinguistic insights that women tend to accommodate more to changing norms than men, especially in prestigious change.Overall, our findings mesh well with the 'Tannen-Biber position that men engage more in 'report' style, while women tend to engage in 'rapport' style, though we do not witness a trend toward a more involved style in men. 5 Rather, women increasingly used more report style, at least in cultural magazines.
For a deeper explanation, we point to two explanations.The explanation that seems to fare best among sociolinguists sees the difference in culturally entrenched gender roles.Our results do not refute such an account, but we point to a (not necessarily mutually exclusive) alternative, which is currently somewhat of a blind spot in the current sociolinguistic literature: explanations in terms of evolutionary psychology.While at first blush reminiscent of long-refuted views on gender differences that reek of misogyny, current evolutionary accounts give, in fact, a far more nuanced account: neither do evolutionary accounts advocate biological determinism, in which behavior is inexorably linked to biological gender, nor are they lacking in empirical support.Another difference between current evolutionary accounts and reductionist accounts of the (early) 20th century is that current accounts do not (implicitly) suppose that male language is the norm, and female language is a deviation, of lesser alloy.The total absence of an evolutionary perspective, in overview chapters on language and gender such as Tanaka (2015), for instance, is remarkable.This second explanation might offer additional insights into how the difference arose in the first place, and why they play out more clearly in the 'syntagmatic' measures at issue than in the 'paradigmatic' measures.The convergence of the genders through time does not refute such an explanation, as is sometimes thought.Surely, there is a cultural determinant as well.Gender differences in language are malleable through time.

(
vi) SEMANTICS, GENERAL VERBS: proportion of general verbs.(vii) SEMANTICS, ABSTRACT VERBS: proportion of abstract verbs.(b) At the level of syntactic complexity: (i) FINITE VERBS: average number of finite verbs per clause.(ii) D-LEVEL: acquisition-based sentence complexity scale.This measure judges syntactic complexity based on the sequence in which children acquire the ability to use various types of sentences; the most complex sentence types, by definition, are those that children acquire last (see Covington et al. 2006; Rosenberg and Abbeduto 1993).(iii) CLAUSE LENGTH: average clause length, computed as words per clause.(iv) MULTIPLE SUBORDINATION: The level of multiple subordination per sentence.

Figure 1 :
Figure1: Effect plots for the linear mixed models with the interaction between gender and year for the lexical measures.

Figure 2 :
Figure 2: Effect plots for the linear mixed-models with the interaction between gender and year for the syntactic measures

Table  :
Main effect and interaction of GENDER and YEAR on the lexical measures.Linear mixed-effect models, with a random effect for the author.

Table  :
Main effect and interaction of GENDER and YEAR on the syntactic measures.Linear mixedeffect models, with a random effect for author.

Table  :
Output of linear mixed model on the years - and on the years -.

Table  :
Output of linear mixed model on the years - and on the years -.