1. Introduction
This article investigates the reading of novel morpho-syntactic forms, specifically a form of gender-inclusive writing. In French, all nouns are categorized by grammatical gender (feminine or masculine), which governs agreement. While gender is arbitrary for inanimate nouns (e.g., la table ‘the.F table’ or le bureau ‘the.M desk’), it usually aligns with social gender for human nounsFootnote 1 (Corbett, Reference Corbett1991). Among the nouns referring to people, some share the same form for feminine and masculine (e.g., le/la pianiste ‘the.M/the.F pianist’), while others mark gender with suffixes as in étudiant/étudiante ‘studentM/F’, comédien/comédienne ‘actorM/F’, chanteur/chanteuse ‘singerM/F’, directeur/directrice ‘directorM/F’Footnote 2 .
This article focuses on role and profession nouns whose gender is determined by their endings. While feminine forms refer only to women, the masculine form can refer to both men and women when used as generic (see e.g., Burnett & Bonami, Reference Burnett and Bonami2019; Elmiger, Reference Elmiger2021). However, psycholinguistic research on gender processing shows that role nouns written with the masculine form encourage male mental representations over female (or non-binary) ones (e.g., Brauer & Landry, Reference Brauer2008; Chatard et al., Reference Chatard, Guimont and Martinot2005; Gabriel et al., Reference Gabriel, Gygax, Sarrasin, Garnham and Oakhill2008; Garnham et al., Reference Garnham, Gabriel, Sarrasin, Gygax and Oakhill2012; Gygax et al., Reference Gygax, Gabriel, Sarrasin, Oakhill and Garnham2008, Reference Gygax, Gabriel, Lévy, Pool, Grivel and Pedrazzini2012, Reference Gygax, Schoenhals, Lévy, Luethold and Gabriel2019; Richy & Burnett, Reference Richy and Burnett2021). Gender-inclusive writing, also called gender-fair writing, corresponds to a variety of linguistic practices providing alternatives to using the masculine as a generic. In this article, we use the term “inclusive writing” (IW) to refer to écriture inclusive, the most commonly used expression in French (see Abbou, Reference Abbou2023).
Different inclusive alternatives to the masculine form (1a) coexist in French: a repetition of the masculine and feminine forms (1b), a contraction of these two forms using a typographic symbol, such as the mid-dot in (1c), or a gender-unmarked form (1d) (Abbou, Reference Abbou2017; Elmiger, Reference Elmiger2022).
In this article, we examine the reading of forms as in (1c). Although they are still debated in France and banned in government documents, contracted forms have become quite frequent in written French since 2017Footnote 3 (Burnett & Pozniak, Reference Burnett and Pozniak2021). These contracted forms appear as morpho-syntactic innovations undergoing grammaticalization (Haspelmath, Reference Haspelmath1999). Similar processes exist in other gendered languages, using different symbols. In German, the capital “I” in the feminine ending (2a) or the gender star form between masculine word and the feminine ending (2b) replaces the masculine. In Spanish, it is “-e” (3a) or “-x” (3b) (Stetie & Zunino, Reference Stetie and Zunino2022). In Italian, the schwa (4) is used instead (Abbondanza et al., Reference Abbondanza, Bonomi, Durante, Reverberi and Foppolo2024).
While the morphology of the inclusive forms varies across languages, all these new forms emerged as alternatives to the masculine used as a generic, and their processing is being currently investigated (e.g., Friedrich et al., Reference Friedrich, Drößler, Oberlehberg and Heise2021; Kruppa et al., Reference Kruppa, Fenn and Ferstl2021; Körner et al., Reference Körner, Glim and Rummer2024; Stetie & Zunino, Reference Stetie and Zunino2022, Reference Stetie and Zunino2024; Vergoossen et al., Reference Vergoossen, Pärnamets, Renström and Gustafsson Sendén2020). Different symbols are employed to transcend the binary male-female interpretation and to encompass all genders, as the gender star in German (Körner et al., Reference Körner, Glim and Rummer2024). In French, contracted forms has been primarily perceived as a strategy for increasing the visibility of women in language, also referred to as re-feminization (see e.g., Sauteur et al., Reference Sauteur, Gygax, Tibblin, Escasain and Sato2023).Footnote 4
A growing amount of sociolinguistic and psycholinguistic studies have investigated the interpretation of IW (e.g., Tibblin et al., Reference Tibblin, Van De Weijer, Granfeldt and Gygax2023; Vervecken et al., Reference Vervecken, Gygax, Gabriel, Guillod and Hannover2015; Xiao et al., Reference Xiao, Strickland and Peperkamp2023). Overall, these works show that inclusive forms tend to feminize mental representations compared to the masculine form, although effects vary depending on the type of inclusive form (contracted vs. gender-unmarked forms in Spinelli et al., Reference Spinelli, Chevrot and Varnet2023) and the linguistic context (contracted vs. non-contracted forms in Pozniak et al., Reference Pozniak, Corbeau and Burnett2023). While this existing body of research has studied the interpretation of IW, some non-empirical contributions criticized IW, arguing that it slows down reading and is hard to understand (Manesse, Reference Manesse2022). However, few psycholinguistic studies have focused on the reading of these novel morpho-syntactic forms in French.
To our knowledge, the first study on the effects of contracted forms on reading was conducted by Gygax and Gesto (Reference Gygax and Gesto2007), who ran a self-paced reading experiment in which participants read five texts, sentence by sentence, with masculine, feminine or inclusive forms. They manipulated two forms of IW: the complete double form as in (1b) and the contracted form using the dash symbol. Reading times were longer only at the first occurrence of contracted forms, suggesting a rapid adaptation to reading contracted forms. Girard et al. (Reference Girard, Foucambert, Le Mené and Tarahomi2022) later used eye-tracking, comparing masculine forms with different contracted forms (mid-dot, parenthesis and period). They observed no significant increase in reading times, neither in the comparison between masculine and contracted forms nor between the different types of contracted forms. However, their sample was small (18 students) and the experimental material included target words with a variety of suffixes, such as marchand·e·s ‘merchant.M·F·PL’, électeur·trice·s ‘voter.M·F·PL’, and sportif·ive·s ‘athlete.M·F·PL’. Liénardy et al. (Reference Liénardy, Tibblin, Gygax and Simon2023) conducted a self-paced reading experiment in which participants read short texts containing sentences with either masculine, double forms or contracted forms with mid-dot. Contrary to previous studies, they manipulated the type of suffixes in the words subject to contraction. They divided target nouns into two categories: words ending with an additive feminine suffix such as “-e” or “-ne” (5), and words ending with a substitutive suffix like “-rice” or “-euse” (6). They distinguished between additive contracted forms, which are created by simply adding an “e” or “consonant + e”, and substitutive contracted forms, which involve the replacement of the masculine suffix with a feminine suffix. This distinction is also present in Elmiger (Reference Elmiger2022), though under different terminology: continuous forms, where the base and the feminine ending are written without any abbreviation, and discontinuous forms, where the ending does not complete the base but instead replaces a part of it. Regardless of the terminology used, it is agreed that forms like those in (5) are simpler than those in (6).
The results showed no increase in reading times for inclusive forms compared to masculine forms, suggesting that IW does not impact reading speed. Unexpectedly, reading times were longer for simpler additive forms (5) than for substitutive forms (6) during their first occurrence. The authors attribute this difference to the high pronounceability of substitutive forms, which can be read as a single noun, by combining the feminine suffix to the masculine suffix as in (7). Some media and communities support this hypothesis by writing and pronouncing the substantive form as a single noun, creating a neologism. Whereas for the contracted forms ending in “·ne”, Liénardy and colleagues argue that although there is no graphical substitution, a phonological substitution causes the increased reading time. During the feminization of nouns such as /kɔmedjε̃/ (comédien ‘actor.M’), the final nasal vowel becomes denasalized phonologically, resulting in the last syllable being pronounced as an oral vowel followed by a nasal consonant /kɔmedjεn/ (comédienne ‘actor.F’). This process of denasalization may impede subvocalization during reading and increase reading times for forms like (5b). However, as both additive forms, “·e” and “·ne”, were presented under the same condition, they could not determine whether the increase in reading time was due to one form rather than the other.
Zami and Hemforth (Reference Zami and Hemforth2024) conducted a self-paced reading experiment comparing contracted forms with mid-dot and masculine forms. Participants from France read six workplace rule texts. They manipulated the morphology of the nouns using three different feminine suffixes: “-rice” as in (6a), “-e” as in (5a), but also mute “-e” as in employé·e ‘employee.M·F’. The results showed that, even though reading contracted forms initially took more time than masculine forms, it became easier over time during the experiment, regardless of the type of suffixes.
Overall, studies have not provided a clear picture of the online processing of the variation of contracted inclusive forms and a thorough comparison between the different suffixes is missing. Analyzing the different suffixes could provide new insights into how morpho-syntactic innovations are processed and the factors at play, such as frequency of IW usage or linguistic factors like the degree of pronounceability of the contracted form.
2. The present study
We conducted four experiments to investigate the reading of contracted inclusive forms. Three experiments were conducted using eye-tracking methodology, while the fourth employed a self-paced reading technique. Eye-tracking allows for the observation of readers’ cognitive processing in real time, within natural reading conditions. This technique provides precise temporal information about when manipulations influence reading and how long these effects last. We can thus compare the reading difficulty of contracted forms using the mid-dot with feminine and masculine forms and examine when and for how long any difficulties persist. In previous studies, the reading time for inclusive forms was always compared to the reading time of the masculine form. However, role nouns are often used more frequently in the masculine form than in the feminine, especially when they are gender-neutral or stereotypically masculine. The frequency of contracted forms is likely closer to that of feminine forms than of masculine forms. Moreover, in terms of length, contracted forms are more similar to feminine forms than to masculine ones. For these reasons, we chose to compare, on the one hand, contracted forms to feminine forms, and on the other hand, feminine forms to masculine forms, rather than directly comparing contracted forms to masculine forms. Furthermore, we studied the processing of different suffixes. In previous studies, the various feminine suffixes were often conflated or not manipulated at a fine-grained level. We manipulated the gender of plural role or profession nouns. The decision to use plural nouns was primarily driven by the fact that determiners are generally marked for gender in the singular but not in the plural. By using plural nouns, we could manipulate gender on the noun itself only, rather than on both the noun and its determiner. For all the contracted forms manipulated, we used plural nouns with two mid-dots: one before the feminine suffix and one before the plural suffix “-s”. We made this choice for several reasons. Firstly, since contracted forms are not yet standardized in the language and are continually evolving, forms with a single mid-dot, such as étudiant·es, have only recently emerged. Furthermore, recent corpus studies suggest that the use of two mid-dots is more frequent than the use of a single mid-dot in plural nouns (Flesch & De Beaumont, Reference Flesch and De Beaumont2023). Finally, we hypothesized that the form we tested, with two mid-dots, is the most complex. If this form does not increase reading time, a single mid-dot is unlikely to have a greater effect.
Experiments 1 and 2 tested whether contracted forms ending in “·e” and “·ne” take longer to read than their feminine counterparts. The “·e” form is the simplest additive form and is likely the most frequently used in French (Flesch & De Beaumont, Reference Flesch and De Beaumont2023). The two forms were confounded in Liénardy et al. (Reference Liénardy, Tibblin, Gygax and Simon2023). Experiments 1 and 2 aim to disentangle the respective effects of each form that contribute to the increased reading times. Liénardy and colleagues suggested that the difficulty with “·ne” forms may be due to the phonological substitution involved in these forms. To investigate the role of the pronounceability, Experiment 3 tested complex yet highly pronounceable substitutive forms, such as (6a). Finally, Experiment 4 tested nouns with the feminine suffix “-euse”, written in two inclusive forms, one highly pronounceable, chanteur·euse·s, and one less so, chanteur·se·s.
2.1. Experiment 1
2.1.1. Method
Participants
Forty-two participants (age mean = 34, SD = 13) with French as their first language were recruited via the RISC platform or social media. Thirty-three were women and nine were men. All had normal-to-corrected vision, no language impairment and were not early bilinguals. They received a 10-euro voucher. Three participants were excluded because they answered fewer than 70% of the comprehension questions correctly; analyses were thus conducted on 39 participants. The experiments reported in this article were approved by the Ethics Committee of University Paris 8 (Approval number: CE-P8-2022-09-14). Participants’ educational background for all experiments is provided in Table 20 (Supplementary Materials).
Materials and design
Twenty-four role nouns ending in “e” in the feminine form and “·e” in the inclusive form were selected. Stereotypes associated with these nouns were controlled based on stereotypical role noun scores (Misersky et al., Reference Misersky, Gygax, Canal, Gabriel, Garnham, Braun, Chiarini, Englund, Hanulikova, Öttl, Valdrova, Von Stockhausen and Sczesny2014). The stereotypes of these role nouns were either neutral or masculine, as stereotypically feminine role nouns were excluded due to the few numbers of available forms. The mean frequency for (combined singular and plural) masculine role nouns was 10.6, while for feminine role nouns, the mean was 4.1.Footnote 5 Using these role nouns, 24 sentences were constructed with an identical structure as in (8). Each sentence started with a feminine or masculine first name, with gender names being counterbalanced (12 feminine first names and 12 masculine first names). The subject name was followed by a transitive verb and a plural role noun as object, ending with a prepositional phrase. As a within-participants factor, the plural role noun was presented either in masculine, feminine or inclusive. Following a Latin square design, three experimental lists were created to ensure that each participant read only one condition of each item, while across the three lists, each item appeared in all three conditions. There were eight items per condition in each list.
Twenty-six filler items were added. They contained words using the French spelling reform from 1990 such as the removal of the letter “i” in the word oignon ‘onion’ as in (9) as well as word abbreviations (e.g., ONU, ‘UN’). We included abbreviations and reformed spelling to incorporate rare and unconventional forms, other than the contracted inclusive forms, into our materials. This was intended to divert participants’ attention from the inclusive forms under investigation.
For one third of the trials, there was a true/false comprehension question after the sentence, so that participants had to pay attention during the experiment. These questions did not concern gender, and error rates were comparable across all gender conditions in all experiments reported in the article. Materials for all experiments are available in the OSF repository (https://osf.io/fvb57 )
Procedure
Participants were tested individually in a quiet room, seated approximately 50 cm away from the screen. The experiment was conducted using an HP EliteBook (LCD), with a 14-inch screen. The display resolution was 1920×1080 pixels, and the refresh rate was 64 Hz. All sentences were presented in black 22-point Times New Roman. Sentences were left-centered on the screen and presented in full. Eye-movements were recorded during reading using an Eyelink Portable Duo eye-tracker, with a sampling rate of 1000 Hz. A head-and-chin rest was used to prevent any involuntary movement. Only the dominant eye was tracked, identified by an eye-dominance test. A 9-points calibration process was used at the beginning of the experiment and then, every ten sentences. Participants were advised that sentences used new spelling forms and contained no spelling mistakes. To familiarize them with the task, the experiment started with three practice trials. Experimental and filler sentences were then presented in a random order. After reading a sentence, participants pressed a button and thereby prompted the next item. On 33% of the trials, participants answered a true/false comprehension question. The total duration of the experiment was around twenty minutes.
2.2.2. Results
Data analysis
For data analysis, the experimental sentences were divided into four regions, illustrated here by slashes (/): Agathe a sélectionné les / étudiant·e·s / pour le concours / d’éloquence. Analysis was conducted on two regions: the region of interest, corresponding to the plural role noun, and the region thereafter as a potential spillover region. Because processing a word or phrase is not always completed before the eyes move forward in reading (Rayner, Reference Rayner1998), the effects of gender can be detected on the role noun itself as well as in the subsequent region. Fixations shorter than 100 ms and longer than two standard deviations from the mean were removed (4.4 % of the data).
Three eye-tracking measures were analyzed: first-pass reading time, the probability of regressions into the region of interest and total reading time. First-pass reading time is defined as the duration of all fixations in a region until the eyes exit from it either to the left or right. This measure reflects early processes during reading, such as lexical access (Clifton et al., Reference Clifton, Staub and Rayner2007). It has been shown that gender recognition can be retrieved during early measures (Irmen, Reference Irmen2007; Irmen & Schumann, Reference Irmen and Schumann2011; Redl et al., Reference Redl, Eerland and Sanders2018, Reference Redl, Frank, De Swart and De Hoop2021). The probability of regressions into a region is the likelihood the region is looked back to after exiting it. Longer fixation times and a higher probability of regressions indicate greater difficulty in processing. Total reading time is the sum of all fixations made into a specific region. This late measure reflects subsequent reading processes such as syntactic and semantic integration (Clifton et al., Reference Clifton, Staub and Rayner2007). Both first-pass and total reading times were log-transformed prior to analysis to improve the normality of their distributions, in line with common practice in psycholinguistics (e.g., Baayen & Milin, Reference Baayen and Milin2010).
Data were analyzed using Bayesian models with the package brms (Bürkner Reference Bürkner2017; Bürkner & Charpentier, Reference Bürkner and Charpentier2020) in R (R version 4.3.0, R Core Team). We chose Bayesian over frequentist models for several reasons (Vasishth, Reference Vasishth, Zufferey and Gygax2023a). Rather than the null hypothesis, Bayesian models directly test the likelihood of our research hypothesis, given the observed data. They allow us to go beyond the binary decision threshold and focus on the probability distribution of the parameters. They are particularly convenient when dealing with a limited number of data as they generate a posterior distribution with a sufficient number of data points, which allows for the integration of a maximal random effects structure without convergence failure. Even with relatively few observations, such models thus yield reliable estimates for the parameters of interest.
For the interpretation of Bayesian models, each model generates a posterior distribution for the model parameters of interest (i.e., independent variables). We report the probability that the effect of the parameter on the dependent variable is less than zero (for negative estimates) or greater than zero (for positive estimates) and the 95% credible interval (indicating the range within which the true parameter value lies with 95%). These parameters and their estimated coefficients are also reported in the Supplementary Materials. We refer to the probability of the effect to estimate the influence of the parameter on the dependent variable (Vasishth, Reference Vasishth2023b). We do not interpret the results as being significant or not, but we rather focus on the probability distribution of the effect. The probabilities reported are interpreted as follows. If the probability of being different from zero is:
-
• high (P > 0.95) and/or the credible intervals do not cross zero, we interpret this as strong evidence for an effect on the dependent variable;
-
• moderate (0.80 < P < 0.95), we interpret this as weak evidence for an effect on the dependent variable;
-
• low (P < 0.80), the effect is not discussed.
The fixed effects in the model included gender (with three levels: masculine, feminine, and inclusive), using the feminine form as the reference level (deviation coding), and the order of trial presentation during the experiment (as a continuous variable). In accordance with the maximal model justified by the experimental design (Barr et al., Reference Barr, Levy, Scheepers and Tily2013), for all random variables (participants and items), we included random slopes for gender and trial order, as well as their interaction. As for the priors we included in our model, we had little previous reliable information about the distribution of all the parameters. Therefore, due to the exploratory nature of our study and to avoid overfitting resulting from strongly informative priors, we used weakly informative priors (available in the OSF repository).
First-pass reading time
Figure 1 shows the mean first-pass reading times—the sum of all fixations in a region until the eyes exit it—across the four regions of the sentence for the three gender conditions (masculine, feminine, inclusive).
Mean first-pass reading times for the suffix “-e” depending on gender condition. In all the figures, error bars represent the 95% confidence intervals.
In the region of interest, there was a moderate probability of a difference between the feminine and inclusive forms (P(coef.>0) = 0.810, CrI = [−0.043, 0.114]), with inclusive forms being read more slowly. There was also a moderate probability of an interaction between trial order and the feminine-inclusive difference (P(coef.>0) = 0.902, CrI = [−0.003, 0.013]), indicating that the feminine-inclusive difference tended to increase as the experiment progressed.
In the subsequent region, there was no more meaningful feminine-inclusive difference. A moderate probability of an effect of trial order was observed (P(coef.>0) = 0.925, CrI = [−0.001, 0.007]): first-pass reading times tended to increase as the experiment progressed. A strong probability of an interaction between trial order and the feminine-inclusive difference was observed (P(coef.>0) = 0.970, CrI = [−0.001, 0.020]), indicating that the feminine-inclusive difference tended to increase over time, with higher reading times for the inclusive condition.
Regressions into a region
Figure 2 shows the probability of making a regression into the region of interest for the three gender conditions.
Probability of regressions into the region of interest for the suffix “-e” depending on gender condition.
A high probability of a difference between feminine and masculine conditions was observed (P(coef.<0) = 0.999, CrI = [−0.966, −0.235]), indicating fewer regressions into the region of interest in the masculine condition. There was also a high probability of an effect of trial order (P(coef.<0) = 0.995, CrI = [−0.034, −0.005]), with regressions in the region of interest decreasing over the course of the experiment. However, a moderate probability of an interaction between trial order and the masculine-feminine difference was observed (P(coef.<0) = 0.850, CrI = [−0.057, 0.016]). This suggests that while the probability of making a regression in the region of interest decreased over the course of the experiment in the masculine condition, it remained relatively stable in the feminine condition. Furthermore, there was a moderate probability of an interaction between trial order and the feminine-inclusive difference (P(coef.<0) = 0.949, CrI = [−0.064, 0.006]). While the number of regressions in the feminine condition tended to remain relatively stable throughout the experiment, it decreased in the inclusive condition.
Total reading time
Figure 3 shows the mean total reading times across the four regions of the sentence for the three gender conditions.
Mean total reading times for the suffix “-e” depending on gender condition.
In the region of interest, there was a high probability of a difference between the feminine and masculine conditions (P(coef.<0) = 0.996, CrI = [−0.248, −0.047]), indicating slower reading times for feminine forms. There was also a high probability of an effect of trial order (P(coef.<0) = 0.993, CrI = [−0.010, −0.001]). Total reading times decreased as the experiment progressed, regardless of the gender condition.
In the following region, there was a moderate probability of a difference between masculine and feminine conditions (P(coef.<0) = 0.817, CrI = [−0.110, 0.040]), indicating slower reading times for feminine forms. In addition, a high probability of an effect of trial order was observed (P(coef.<0) = 0.999, CrI = [−0.010, −0.003]), indicating that total reading times decreased over time.
2.1.3. Discussion
This first experiment showed that total reading times were longer for feminine forms compared to masculine forms. Consistently, the probability of regression was higher for the feminine condition than for the masculine condition (especially towards the end of the experiment, as indicated by the interaction with the trial order). While the probability of making a regression in the region of interest decreased over the course of the experiment in both the masculine and inclusive conditions, it remained stable in the feminine condition. Previous studies, which compared inclusive forms only to masculine ones, could not capture this difference between masculine and feminine forms, as the feminine condition was not included. The fact that feminine forms are read more slowly than their masculine counterparts may be attributed to a lexical frequency effect. Indeed, we observed that masculine nouns in our experiment were more frequent than their feminine counterparts (mean frequency of 10.6 vs. 4.1). This difference in frequency between feminine and masculine forms is likely accentuated by the use of role nouns stereotypically biased masculine or neutral.
Regarding the difference between feminine and inclusive forms, there is some evidence of increased first-pass reading times. However, there is no evidence of longer total reading times or a higher probability of regressions. The lack of increased total reading times for inclusive forms aligns with findings from previous studies, which similarly did not report an increase in reading times for inclusive forms in French, except at the first occurrences (Gygax & Gesto, Reference Gygax and Gesto2007; Girard et al., Reference Girard, Foucambert, Le Mené and Tarahomi2022; Liénardy et al., Reference Liénardy, Tibblin, Gygax and Simon2023; Zami & Hemforth, Reference Zami and Hemforth2024). As previously mentioned, the inclusive form ending in “·e” is by far the simplest and most commonly used in French and is recommended in IW guides, unlike more complex forms such as “·rice” (Viennot, Reference Viennot2018). In this context, it is not surprising that it does not impede reading.
Across all gender conditions combined, we observed strong evidence for a decrease in total reading times over the experiment. However, there was weak evidence indicating that the difference in first-pass reading times between the feminine and inclusive conditions, tended to increase over the course of the experiment. The opposite was expected, as previous experiments reported a familiarity effect with inclusive forms, where the difference in processing times between masculine and inclusive conditions decreased over the course of these experiments (Gygax & Gesto, Reference Gygax and Gesto2007; Liénardy et al., Reference Liénardy, Tibblin, Gygax and Simon2023; Zami & Hemforth, Reference Zami and Hemforth2024). It is possible that the increase in first-pass reading time for inclusive forms during the experiment is due to participants not processing the mid-dot the first few times they encountered it. We discuss this issue further in the general discussion.
In the study by Liénardy and colleagues (Reference Liénardy, Tibblin, Gygax and Simon2023), it was observed that, surprisingly, additive inclusive forms were read more slowly than substitutive forms. However, in their analysis, the additive forms “·e” and “·ne” were conflated, and according to the authors, the increase in reading time was likely due to the “·ne” forms. In the following experiment, we tested whether, unlike the forms with “·e”, the inclusive forms ending in “·ne” were read more slowly than the feminine forms.
2.2. Experiment 2
2.2.1. Method
Participants
Forty-one other participants (age mean = 30, SD = 16) with French as their first language were recruited via the RISC platform or through social media. Thirty-one were women, nine were men and one was nonbinary. They all had corrected-to-normal vision and no language impairment. None of them were early bilinguals. They all received a 10-euro voucher. One participant, whose accuracy on comprehension questions was below 70%, was removed from the analysis.
Material and design
The design was the same as in the first experiment. Twenty-four role nouns, either stereotypically masculine or neutral, with their feminine forms ending in “ne”, were selected. The mean frequency for masculine role nouns was 4.2, while for feminine role nouns, the mean was 1.8. Critical role nouns were presented in masculine, feminine, or inclusive forms ending in “·ne” as in (10). Both experimental and filler items followed the same structure as in Experiment 1.
After the reading experiment, participants completed a short questionnaire assessing their attitudes toward IW. The first question did not directly address attitudes towards IW, but rather on the spelling reform (on a 6-point scale from “not problematic at all” to “very problematic”). Two questions asked how important they found the feminization of role and profession nouns and the use of IW (on a 6-point scale from “not important at all” to “very important”). Participants were also asked whether they thought that IW could hinder reading (yes/no/no opinion), how often they encountered inclusive forms (4-point Likert scale from “never exposed” to “frequently exposed”), whether they considered it overused (yes/no), whether it should be banned (yes/no/only in certain contexts), and whether its use should be promoted (yes/no/only in certain contexts). Materials as well as the questionnaire are available in the OSF repository (https://osf.io/fvb57 ).
Procedure
The procedure was the same as in Experiment 1, except that participants had to answer the questionnaire after the reading task. The total duration of the experiment, including the completion of the questionnaire, was around twenty-five minutes.
2.2.2. Results
Data analysis
Data analysis was the same as in Experiment 1. Fixations shorter than 100 ms and longer than two times from the standard deviation were removed (4.2 % of the data).
First-pass reading time
Figure 4 shows the mean first-pass reading times across the four regions of the sentence for the three gender conditions.
Mean first-pass reading times for the suffix “-ne” depending on gender condition.
In the region of interest, there was a high probability of a difference between the masculine and the feminine conditions (P(coef.<0) = 0.995, CrI = [−0.198, −0.026]), indicating shorter first-pass reading times for masculine forms. There was also a moderate probability of an effect when comparing inclusive and feminine forms (P(coef.>0) = 0.918, CrI = [−0.023, 0.142]), with inclusive forms being read more slowly. There was also a moderate probability of an interaction between trial order and the masculine-feminine difference (P(coef.<0) = 0.847, CrI = [−0.009, 0.003]). While the first-pass reading time decreased over the course of the experiment in the masculine condition, it remained relatively stable in the feminine condition. Additionally, we found a moderate probability of an interaction between trial order and the difference between inclusive and feminine conditions (P(coef.>0) = 0.889, CrI = [−0.002, 0.010]), indicating that the feminine-inclusive difference increased as the experiment progressed.
In the subsequent region, we found a moderate probability of an interaction between trial order and the feminine-inclusive difference (P(coef.<0) = 0.821, CrI = [−0.012, 0.004]). This indicates that the difference between feminine and inclusive conditions decreased as the experiment progressed.
Regressions into a region
Figure 5 shows the probability of making a regression into the region of interest for the three gender conditions.
Probability of regressions into the region of interest for the suffix “-ne” depending on gender condition.
When comparing feminine and masculine conditions, a moderate probability of an effect was observed (P(coef.<0) = 0.934, CrI = [−0.584, 0.081]), showing lower regression probability in the masculine condition. Additionally, there was a moderate probability of an effect of trial order (P(coef.<0) = 0.935, CrI = [−0.017, 0.002]). This indicates that, regardless of the gender condition, the probability of making a regression into the region of interest decreased as the experiment progressed.
Total reading time
Figure 6 shows the mean total reading times across the four regions of the sentence for the three gender conditions.
Mean total reading times for the suffix “-ne” depending on the gender condition.
To examine the potential effect of the attitude towards IW on total reading times and avoid correlated variables in statistical analyses, we tested for correlations between all the variables in the questionnaire using the Hmisc and corrplot packages (Harrell Jr, Reference Harrell2024; Wei & Simko, Reference Wei and Simko2024). These correlations are reported in Figure 1 of the Supplementary Materials.
The importance assigned to the use of IW correlated positively with the importance assigned to the feminization of role nouns (r = 0.7) and the support for IW development (r = 0.72). Conversely, the greater the importance placed on using IW, the less problematic the orthographic reform is perceived (r = −0.41), the less the IW is considered overused (r = −0.47), and the less support there is for its prohibition (r = −0.65). In contrast, the self-reported exposure to IW was not correlated with any of the variables. We therefore included both the importance placed on using IW (as a measure of attitudes towards IW) and the self-reported exposure in the statistical analysis to assess their effects on total reading times.
In the region of interest, we found a high probability of a difference between the masculine and feminine conditions (P(coef.<0) = 1, CrI = [−0.286, 0.089]), indicating that masculine forms were faster to read than feminine forms. There was also a high probability of an effect when comparing feminine and inclusive forms (P(coef.>0) = 1, CrI = [0.087, 0.252]), suggesting that inclusive forms were read more slowly than feminine forms. Additionally, a moderate probability for an interaction between the self-reported exposure to IW and the difference between feminine and inclusive forms was found (P(est<0) = 0.923, CrI = [−0.224, 0.034]). The greater the participants reported exposure to IW, the smaller the difference in reading times between feminine and inclusive forms. A high probability of an effect of trial order was observed (P(coef.<0) = 0.995, CrI = [−0.008, −0.001]), indicating that total reading times were shorter as the experiment progressed, regardless of the gender conditions.
In the subsequent region, the difference between feminine and inclusive conditions persisted with a moderate probability of an effect (P(coef.>0) = 0.945, CrI = [−0.015, 0.149]), as did the effect of trial order with a high probability of an effect (P(coef.<0) = 1, CrI = [−0.007, −0.002]). Additionally, there was a moderate probability of an effect of the self-reported exposure to IW (P(coef.>0) = 0.865, CrI = [−0.075, 0.249]).
2.2.3. Discussion
As in Experiment 1, feminine nouns were read more slowly than masculine nouns. However, inclusive forms ending in “·ne” were read more slowly than feminine forms in first-pass as well as in total reading times. Compared to feminine forms, “·ne” inclusive forms were more difficult to process than “·e” forms. This is consistent with the hypothesis of Liénardy et al. (Reference Liénardy, Tibblin, Gygax and Simon2023), who suggested that the difficulty they observed with additive inclusive forms was due to the “·ne” forms. According to these authors, the difficulty associated with “·ne” forms may be phonological, forms like comédien·ne cannot be read as a single word, which could explain the longer reading times for these forms.
Although not a primary aim of the experiment, we examined whether there was any correlation between attitudes towards IW and the total reading times of inclusive forms, and found no evidence of such a relationship. This result aligns with Kruppa et al. (Reference Kruppa, Fenn and Ferstl2021) who showed in a lexical decision task in German that recognition times for inclusive forms such as student*in ‘student.M*F’ were not correlated with participants’ attitudes towards IW. Attitudes towards IW do not influence its recognition or reading times. This observation is also consistent with the findings of Liénardy and colleagues (Reference Liénardy, Tibblin, Gygax and Simon2023), who noted that while participants judged inclusive forms as more difficult to read compared to masculine forms after the reading experience, these same inclusive forms did not take longer to read than the masculine forms. There appears to be a discrepancy between perceived difficulty and objective difficulty. Overall, these results suggest that reading inclusive forms involves some level of automatic processing, independent of participants’ attitudes.
Interestingly, we observed that the more participants reported being exposed to inclusive forms, the smaller the difference in reading times between feminine and inclusive forms. This indicates that greater familiarity with inclusive forms is associated with faster reading times. The only measure of exposure to inclusive forms available here is the participants’ self-reported experience. Additional measures for assessing exposure frequency would be necessary to accurately determine the role of frequency in reading times of contracted forms.
Finally, strong evidence of a trial order effect was observed in total reading times, which decreased throughout the experiment, independent of gender condition, in both the region of interest and the subsequent region. Weak evidence for an interaction between trial order and the difference between the feminine and inclusive conditions was observed in first-pass reading times. However, similarly to Experiment 1, this interaction occurred in the opposite direction from what was expected.Footnote 6 As the experiment progressed, the difference between the feminine and inclusive conditions increased. In contrast to Experiment 1, this effect did not persist in the subsequent region (weak evidence for an interaction in the opposite direction was observed). In both experiments, the order effect no longer interacted with the feminine-inclusive difference in total reading times, suggesting that the experimental context was insufficient to observe a familiarization effect with inclusive forms.
Returning to the phonological hypothesis proposed by Liénardy et al. (Reference Liénardy, Tibblin, Gygax and Simon2023), the absence of increased reading times for substitutive contracted forms like auteur·rice may stem from their easy pronunciation as a single word. In the following experiment, we examine the reading time of substitutive contracted forms such as agriculteur·rice·s, pronounceable as a single word.
2.3. Experiment 3
2.3.1. Method
Participants
Thirty-eight other participants with French as their first language (age mean = 33, SD = 12) were recruited via the RISC platform or through social media. Twenty-five were women, twelve were men and one did not specify a gender identity. They all had corrected-to-normal vision and no language impairment. None of them were early bilinguals. They all received a 10-euro voucher. One participant was excluded for answering fewer than 70% of the comprehension questions correctly. The subsequent analyses were therefore conducted on 37 participants.
Materials and design
The design was the same as in previous experiments. Twenty-four role nouns, either stereotypically masculine or neutral, with their feminine forms ending in “rice”, were selected. The mean frequency for masculine role nouns was 4, while for feminine role nouns, the mean was 1.9. Experimental sentences were constructed following the same structure as in previous experiments, as in (11). Critical role nouns were presented in masculine, feminine, or inclusive forms ending in “·rice”. The segmentation for the contracted forms (·rice) corresponds to the graphical cut, as defined by Elmiger (Reference Elmiger2022); that is, the ending corresponds to the graphic element that distinguishes the feminine form from the masculine form. This segmentation also aligns with the morphological cut when considering the long stem, in which case “-rice” corresponds to the feminine suffix. Fillers and practice items were identical to those used in Experiment 1.
2.3.2. Results
Data Analysis
Data analysis was the same as in the previous experiments, with the addition of word length (in number of characters) being included as a fixed factor in the model. This was done because the difference in length between the feminine and the inclusive form (agricultrices vs. agriculteur·rice·s) was greater than in the previous experiments. In the previous experiments, the number of characters differed by one character between the masculine and feminine forms in Experiment 1, two characters in Experiment 2, and by two characters (the two mid-dots) between the feminine and contracted forms in both experiments. As these variations in length were confounded with the manipulation of the gender factor, we did not include length in the statistical models, to avoid masking the potential effect of the gender factor. This decision was made based on the fact that the differences in length were not substantial enough to account for changes in reading times. Fixations shorter than 100 ms and longer than two times from the standard deviation were removed (4.7% of the data).
First-pass reading time
Figure 7 shows the mean first-pass reading times across the four regions of the sentence for the three gender conditions.
Mean first-pass reading times for the suffix “-rice” depending on gender condition.
In the region of interest, we observed a moderate probability of a difference between the masculine and the feminine conditions (P(coef.<0) = 0.918, CrI = [−0.173, 0.033]), indicating slower reading for feminine forms. We also observed a high probability of a difference between the feminine and the inclusive forms (P(coef.>0) = 0.980, CrI = [−0.007, 0.273]). This difference interacted with trial order, with a moderate probability of an effect (P(coef.<0) = 0.859, CrI = [−0.010, 0.003]). The feminine-inclusive difference decreased as the experiment progressed. Additionally, a high probability of a length effect was found (P(coef.>0) = 0.969, CrI = [−0.001, 0.037]), suggesting that longer forms were read more slowly.
In the subsequent region, a difference between the feminine and inclusive conditions was found with a moderate probability of an effect (P(coef.>0) = 0.907, CrI = [−0.029, 0.146]). The length effect persisted (P(coef.>0) = 1, CrI = [0.018, 0.045]).
Regressions into a region
Figure 8 shows the probability of making a regression into the region of interest for the three gender conditions.
Probability of regressions into the region of interest for the suffix “-rice” depending on the gender condition.
There was a high probability of a difference between the masculine and the feminine forms (P(coef.<0) = 0.968, CrI = [−0.641, 0.018]), indicating that there were fewer regressions in the masculine condition compared to the feminine condition. A high probability of an effect of trial order was also observed (P(coef.<0) = 0.950, CrI = [−0.018, 0.002]), indicating that there were fewer regressions as the experiment progressed. The masculine-feminine difference interacted with trial order, with a high probability (P(coef.<0) = 0.976, CrI = [−0.047, 0.0001]). This suggested that the probability of making a regression in the region of interest decreased in the masculine condition while remaining relatively stable in the feminine condition throughout the experiment.
Total reading time
Figure 9 shows the mean total reading times across the four regions of the sentence for the three gender conditions.
Mean total reading times for the suffix “-rice” depending on gender condition.
In the region of interest, results showed a high probability of a difference in total reading times between masculine and feminine conditions (P(coef.<0) = 0.997, CrI = [−0.271, −0.051]), with feminine forms being read more slowly. A moderate probability of a difference between inclusive and feminine forms was also observed (P(coef.>0) = 0.947, CrI = [−0.028, 0.277]), with inclusive forms being read more slowly. Additionally, a high probability of an effect of trial order was found (P(coef.<0) = 0.999, CrI = [−0.002, −0.002]), showing that total reading times decreased throughout the experiment. Moreover, a moderate probability of an interaction between trial order and the masculine-feminine difference was observed (P(coef.<0) = 0.832, CrI = [−0.012, 0.004]), indicating that the difference tends to increase, total reading times decreasing more rapidly for masculine forms than for feminine forms, as the experiment progressed. Similarly, a moderate probability of an interaction between trial order and the feminine-inclusive difference was observed (P(coef.<0) = 0.881, CrI = [−0.010, 0.003]). This suggests that the feminine-inclusive difference tended to decrease over the course of the experiment, with reading times for inclusive forms decreasing more rapidly. Finally, a high probability of length was observed (P(coef.>0) = 0.998, CrI = [0.014, 0.063]), with the longer forms being read the most slowly.
In the subsequent region, the effect of the length as well as of the trial order persisted, both with a high probability of an effect (respectively, P(coef.>0) = 0.998, CrI = [0.010, 0.037]; P(coef.<0) = 0.994, CrI = [−0.006, −0.001]). Additionally, the interaction between trial order and the feminine-inclusive difference persisted (P(coef.<0) = 0.939, CrI = [−0.012, 0.001]), indicating that this difference decreased as the experiment progressed.
2.3.3. Discussion
As in previous experiments, we observed strong evidence for longer total reading times for role nouns when they were feminine compared to when they were masculine. This increase in reading times was accompanied by a higher number of regressions in the region of interest when the noun was feminine compared to when it was masculine. As in Experiment 2, inclusive nouns were read more slowly than feminine nouns. Experiment 2 showed that, with ending in “·ne”, the increase in reading time for the inclusive condition was observed both in the region of interest as well as the following region for total reading times, and only in the region of interest for first-pass reading times. Here, the pattern is reversed. We found evidence for a difference in first-pass reading times in both the region of interest and the following region. While considering total reading times, we observed evidence for a difference between the feminine and the inclusive conditions in the region of interest only. This suggests that the effect occurred earlier with the forms ending in “·rice” than with the forms ending in “·ne”. One reason for that would be that reading contracted forms with “·rice” forms is easier to pronounce than those with “·ne”, as Liénardy et al. (Reference Liénardy, Tibblin, Gygax and Simon2023) hypothesized that the latter triggers the process of denasalization which impedes subvocalization. The advantage in reading times of highly pronounceable contracted forms compared to those that are less pronounceable is further discussed in the general discussion. Moreover, we observed evidence that the difference in reading times between the feminine and the inclusive conditions decreased throughout the experiment, suggesting a habituation to inclusive forms, which was not present in the previous experiment.
To summarize, contracted forms in “·rice” take longer to read compared to the feminine forms. Nevertheless, this processing cost compared to feminine forms was more transitory over time than that observed for the contracted forms ending in “·ne” (Experiment 2). One reason for this effect might be the highly pronounceability of contracted forms in “·rice”. As mentioned in the introduction, inclusive forms ending in “·rice” can be read as a single word, and these neologisms such as auteurice ‘author.MF’ are even used by some media and communities. It is therefore possible that, once accustomed to these forms during the experiment, participants may have read these forms as a single word, while they could not with inclusive forms ending in “·ne”. Another explanation might be the frequency of inclusive forms ending in “·rice” compared to those ending in “·ne”. However, corpus data on the frequency of these inclusive forms are required to answer this question.
To further examine the difference in processing based on the degree of pronounceability of contracted forms, the following experiment compared two potential contracted forms of the same feminine suffix, namely “-euse”. Specifically, with the inclusive forms using this suffix, depending on the placement of the mid-dot, one can obtain either a fully pronounceable neologism (chanteur·euse) or a shorter, yet a less pronounceable form (chanteur·se)Footnote 7 . Given the larger sample size required for this comparison, the experiment was conducted online using a self-paced reading paradigm, which allowed for more efficient and appropriate data collection.
2.4. Experiment 4
2.4.1. Method
Participants
Seventy-eight speakers (mean age = 33, SD = 10), with French as their first language, were recruited on the Prolific platform (www.prolific.co), where the cost is approximately £9 per hour per participant. Thirty-one were women, forty-six were men and one did not specify a gender identity. None of them were early bilinguals. Thirty-nine participants were assigned to the highly pronounceable form in “·euse”, while the remaining thirty-nine were assigned to the less pronounceable form in “·se”.
Material and design
Twenty-four role nouns, either stereotypically masculine or neutral, with their feminine forms ending in “euse”, were selected. The mean frequency for masculine role nouns was 2.4, while for feminine role nouns, the mean was 1.3. Based on the role nouns, 24 experimental sentences were constructed following the same structure as in the previous experiments. The gender of the role nouns was manipulated as a within-factor, as in previous experiments, with the three same conditions: masculine, feminine or inclusive, as in (12). Two versions of the experiment were created, one in which the role nouns used the inclusive “·euse” form was presented to half of the participants, while the other version, using the “·se” form, was presented to the other half of the participants. Fillers had an identical structure as in previous experiments.
Procedure
The experiment was administered online using Ibexfarm (Drummond, Reference Drummond2013) hosted on university servers (for evidence of the validity of internet-based data collection in psycholinguistics, see e.g., Enochson & Culbertson, Reference Enochson and Culbertson2015). The experiment used a non-cumulative moving-window self-paced reading task. Participants read each sentence one segment at a time, pressing the space bar to move on to the next segment. Sentences were presented into four presentation segments as illustrated here by “/” in (13).
Detailed instructions and three practice trials were presented prior to the experiment. One-third of the trials, participants answered a true/false comprehension question.
2.4.2. Results
Data analysis
The dependent variable was the mean reading time in milliseconds. Reading times shorter than 200 ms and longer than two standard deviations were removed (2.2% of the data). Reading times were analysed using linear-mixed models with a lognormal likelihood. The levels of the between-participants factor “Type of inclusive form” (chanteur·euse vs. chanteur·se) were coded as follows: 1 for the “·euse” level and 0 for the “·se” level. We applied scaled mean-centered coding, such that 0 coding approached −0.5 and 1 coding approached 0.5. In the same way as for the inclusive forms ending in “·rice” (Experiment 3), the inclusive form ending in “·euse” differs by 5 characters from the feminine form (e.g., chanteuses vs. chanteur·euse·s), and the form ending in “·se” differs by 3 characters from the feminine form. Due to these differences in length, we included the effect of length in the model.
Figure 10 illustrates the mean reading times across the four regions of the sentence for the three gender conditions and for both endings, “·euse” and “·se”.
Mean reading times for both endings, “·euse” and “·se”, depending on the gender condition.
Region of interest
A high probability of a difference between masculine and feminine forms was observed (P(coef.<0) = 0.966, CrI = [−0.058,0.002]), with feminine forms being read more slowly. A high probability of a difference between the feminine and the inclusive conditions was also observed (P(coef.>0) = 0.980, CrI = [0.002, 0.104]), with inclusive forms being read more slowly. There was a high probability of an effect of trial order (P(coef.<0) = 1, CrI = [−0.016, −0.011]), which interacted with the feminine-inclusive difference (P(coef.<0) = 0.988, CrI = [−0.010, −0.001]). This interaction suggests that the feminine-inclusive difference decreased as the experiment progressed. In addition, a high probability of an effect of length was observed (P(coef.>0) = 0.998, CrI = [0.005, 0.022]), with longer forms being read more slowly.
Spillover region
The difference between the feminine and the masculine conditions, observed in the preceding region, persisted with a high probability of an effect (P(coef.<0) = 0.974, CrI = [−0.072, −0.0001]), indicating slower reading for feminine forms. A moderate probability of a difference between feminine and inclusive conditions was also found (P(coef.>0) = 0.874, CrI = [−0.013, 0.050]), with inclusive forms being read more slowly. Reading times decreased over the course of the experiment, as shown by a high probability of a trial order effect (P(coef.<0) = 1, CrI = [–0.012, –0.007]). This decrease was more marked for masculine than for feminine forms, as suggested by a moderate probability of an interaction between trial order and the masculine-feminine difference (P(coef.<0) = 0.914, CrI = [–0.007, 0.001]). Finally, a high probability of an effect of length was observed (P(coef.>0) = 1, CrI = [0.009, 0.005]), with longer forms being read more slowly.
Interestingly, there was a high probability of an interaction between the feminine-inclusive difference and the type of inclusive form (“·euse” vs. “·se”) (P(coef.>0) = 0.974, CrI = [−0.0002, 0.116]), suggesting that “·se” forms produced larger reading time differences. There was also a moderate probability of an interaction between the type of inclusive form and trial order (P(coef.>0) = 0.844, CrI = [−0.002, 0.007]). Finally, there was a three-way interaction between the type of inclusive form, trial order, and the feminine-inclusive difference (P(coef.<0) = 0.935, CrI = [−0.016, 0.002]). To better understand these interactions, we computed separate models for each type of inclusive form (“·euse” vs. “·se”).
In the “·se” condition, there was a moderate probability of a difference between masculine and feminine conditions (P(coef.<0) = 0.898, CrI = [−0.079, 0.018]), with masculine forms being read more slowly. A high probability of a difference between the feminine and the inclusive conditions was also observed (P(coef.>0) = 0.985, CrI = [0.005, 0.092]), with inclusive forms being read more slowly. A high probability of trial order was observed (P(coef.<0) = 1, CrI = [−0.012, 0.005]), indicating reading times decreased as the experiment progressed. A moderate probability of an interaction between trial order and the masculine-feminine difference (P(coef.<0) = 0.880, CrI = [−0.010, 0.003]) suggested that this difference increased as the experiment progressed. There was also a moderate probability of an interaction between trial order and the feminine-inclusive difference (P(coef.<0) = 0.894, CrI = [−0.011, 0.002]), suggesting that this difference decreased as the experiment progressed. Finally, a high probability of an effect of length was found (P(coef.>0) = 0.999, CrI = [0.005, 0.017]), with longer forms being read more slowly.
In the “·euse” condition, we observed a high probability of a difference between masculine and feminine conditions (P(coef.<0) = 0.955, CrI = [−0.090, 0.006]), with feminine forms being read more slowly. A high probability of an effect of trial order was found (P(coef.<0) = 1, CrI = [−0.014, 0.008]). A moderate probability of an interaction between trial order and the feminine-inclusive difference (P(coef.<0) = 0.821, CrI = [−0.003, 0.009]) suggests a larger decrease for feminine forms. A high probability of an effect of length (P(coef.>0) = 1, CrI = [0.001, 0.019]) showed that longer forms were read more slowly.
2.4.3. Discussion
Experiment 4 replicated previous findings that feminine forms took longer to read than masculine forms, and contracted inclusive forms were read more slowly than feminine ones. This difference is observed for both forms ending in “·euse” and “·se”. However, while the results showed a difference between the feminine and the inclusive forms on the region of interest for both endings together, this effect persists in the following region only for contracted forms ending in “·se”. The difficulty encountered with less pronounceable contracted forms ending in “·se” appears to be more persistent over time than with highly pronounceable forms ending in “·euse”. Interestingly, this result aligns with results from Experiment 3 for the other highly pronounceable form ending in “·rice”. The difference between inclusive and feminine forms was evident only in the region of interest regarding total reading time. Despite the different methodologies used, reading times obtained with the self-paced reading method correspond to the total reading time recorded with eye-tracking, rather than to first-pass reading time (see e.g., Just, Carpenter & Wooley, Reference Just, Carpenter and Woolley1982; Witzel et al., Reference Witzel, Witzel and Forster2012).
3. General Discussion
A first finding is that role nouns were read more quickly in the masculine form than in the feminine form.Footnote 8 While production studies have shown a bias against feminine forms in the use of pronouns in both gender-marked and gender-unmarked languages (von der Malsburg et al., Reference Von Der Malsburg, Poppels and Levy2020; Pozniak & Burnett, Reference Pozniak and Burnett2021), to our knowledge, a higher processing time for feminine role nouns compared to their masculine counterparts has not yet been observed in comprehension. This result, initially surprising, can be explained by usage frequency. In French, most role nouns are used in masculine forms considered as being generic. Although a feminization process began in the late 20th century (Burnett & Bonami, Reference Burnett and Bonami2019; Arbour et al., Reference Arbour, De Nayves and Royer2014; Houdebine, Reference Houdebine1998, among others), the use of feminine forms remains debated and less frequent than the use of masculine forms as suggested in our experiments (mean frequency of 2.3 for feminine role nouns vs. 5.3 for masculine ones across all experiments).
The second important result of our study concerns the difference in reading times between contracted inclusive forms and feminine forms. Contracted forms ending in “·ne”, “·rice”, “·euse”, and “·se” were generally read more slowly than their feminine counterparts, while “·e” forms differed only in first-pass reading time. Furthermore, differences were observed among these inclusive forms. The distinction between additive and substitutive forms (Liénardy et al., Reference Liénardy, Tibblin, Gygax and Simon2023; or continuous vs. discontinuous following Elmiger, Reference Elmiger2022) does not account for all of our results: “·ne” forms yielded longer total reading times than feminine forms, unlike “·e” forms, though both are additive. Differences also appeared within substitutive forms, with some appearing to be slightly more difficult to read than others.
Regarding the absence of difference in total reading time between the contracted “·e” forms and their feminine counterparts, one possibility is that the “·e” forms were not read more slowly than their feminine counterparts, because participants might not have processed the mid-dot and might have read the feminine form instead. However, our study tested contracted forms with two mid-dots, which should increase the likelihood of the contracted form being perceived. Moreover, we observe evidence for longer first-pass reading times in the contracted condition compared to the feminine condition, which could not be explained if participants had not considered the contracted form at all.
We intentionally avoided questions about gender before or during the experiment, to prevent drawing attention to our focus on IW. This was intended to observe the reading of contracted inclusive forms while minimizing the development of reading strategies. Notably, Experiment 2 showed that participants’ opinions on IW did not influence their reading times. This finding suggests that reading inclusive forms was relatively automatic and not influenced by conscious strategies.
However, it is possible that the contracted “·e” form was not immediately noticed by the participants, but only after a few trials. This could explain the weak evidence for an interaction between trial order and the difference between the feminine and the inclusive conditions that we observed, which was in the opposite direction of what was expected. Indeed, instead of decreasing over the course of the experiment, the difference between the feminine condition and the inclusive condition increased. Similar weak evidence for this interaction was also found for first-pass reading times in the region of interest in Experiment 2. Furthermore, the contracted forms tested in these two experiments were the least visually salient, which may explain why participants did not immediately notice them. It is only in Experiments 3 and 4, where the contracted forms were visually very distinct from the feminine forms, that we observed strong evidence for an interaction suggesting a habituation effect, with the difference decreasing as the experiment progressed.
Further research is necessary to determine whether the mid-dots were actually processed in the “·e” inclusive condition and, importantly, whether the contracted form activated the representation of mixed-gender groups or not—something our study cannot establish. Our findings still indicate that they were processed without important difficulty. This is consistent with previous studies conducted in French on the reading of contracted inclusive forms, which tested various inclusive forms, including those ending in “·e”, and found no clear difficulty—or only transient ones, with rapid habituation (Gygax & Gesto, Reference Gygax and Gesto2007; Girard et al., Reference Girard, Foucambert, Le Mené and Tarahomi2022; Liénardy et al., Reference Liénardy, Tibblin, Gygax and Simon2023; Zami & Hemforth, Reference Zami and Hemforth2024). The “·e” form is also considered the morphologically simplest form (Elmiger, Reference Elmiger2022) and is recommended by IW guides (see e.g., Viennot, Reference Viennot2018). For these reasons, it is likely the most frequently used contracted inclusive form. Supporting this assumption, corpus studies on IW showed that the most frequent words used with an inclusive form ended in “·e” (Flesch and De Beaumont, Reference Flesch and De Beaumont2023). Finally, it is possible that “·e” forms are simpler to process because there is only one possible placement of the mid-dot when contracting both masculine and feminine forms with the feminine suffix “-e”, which is before the latter. In contrast, with suffixes such as “-rice” or, to a lesser extent, “-euse”, there are numerous possibilities for forming the contracted inclusive form, as illustrated in (14) for “-rice” and (15) for “-euse” (see e.g., Elmiger, Reference Elmiger2022). The number of possible contracted forms may influence the frequency with which readers are exposed to an inclusive form. For feminine nouns ending in “e”, readers are always exposed to the same form, whereas for feminine nouns ending in “rice”, they may encounter various contracted forms. As inclusive forms are morpho-syntactic innovations currently undergoing a grammaticalization process, the variety of forms available may slow down the grammaticalization process for “-rice” and “-euse” forms. In contrast, the inclusive form with the suffix “-e” is likely more grammaticalized, due to the more limited range of choices. However, further corpus studies focusing on the variety of endings for contracted forms are needed to address this question.
Besides frequency, pronounceability may influence the reading time of contracted forms. Indeed, among forms read more slowly than their feminine counterparts, some were easier to pronounce than others. As noted in the introduction, consistent with Liénardy et al. (Reference Liénardy, Tibblin, Gygax and Simon2023), the forms ending with “·ne” may be affected by the process of denasalization, making them potentially harder to subvocalize. Forms ending in “·rice” and “·euse” can be pronounced and read as a single word (and can give rise to neologisms), while this is not the case with forms ending in “·ne” and “·se”. If the four types of forms were read more slowly than their feminine counterparts, the more pronounceable forms appeared to pose slightly less difficulty. The results of Experiments 2, 3, and 4 together showed that, the difficulty observed with the highly pronounceable inclusive forms, “·rice” and “·euse”, was earlier and more transient compared to the difficulty observed with non-pronounceable inclusive forms “·ne” and “·se”.
According to the dual-route model of reading (Coltheart et al., Reference Coltheart, Curtis, Atkins and Haller1993, Reference Coltheart, Rastle, Perry, Langdon and Ziegler2001), familiar, frequent forms are processed via the lexical route, which facilitates rapid and automatic word recognition. In contrast, novel forms are processed via the grapheme-to-phoneme conversion route, which is more effortful and results in slower reading times for nonwords and unfamiliar words. Consequently, reading unfamiliar forms via the sublexical phonological route should be even more cognitively demanding when the forms are difficult to pronounce, such as chanteur·se. However, we observed in Experiment 2 that the more participants reported being exposed to inclusive forms, the faster they read the difficult-to-pronounce forms, such as comédien·ne, compared to the feminine form. This observation suggests that as contracted inclusive forms become familiar, readers are able to access their lexical orthographic representation directly, without going through grapheme-to-phoneme decoding. Nevertheless, further research is necessary to more precisely measure the link between frequency of exposure to contracted forms and reading time. Upcoming studies should employ a more objective measure of exposure frequency and not only rely on participants’ self-reported frequency of their exposure to contracted forms.
In conclusion, our experiments demonstrated that reading new morpho-syntactic forms, like gender-inclusive contracted forms in French, are not inherently unreadable. However, reading difficulty varies with factors such as the morphological suffix, the phonological realization of the contracted form, and its frequency. It should be noted that the sample sizes in our experiments may not be considered sufficiently powered to robustly account for the strength of all the factors manipulated (Brysbaert, Reference Brysbaert2019). Results must be interpreted with caution, and more data are needed to better understand how these new linguistic forms are processed in general. To ensure replicability, future research should investigate more deeply the role of these factors with more data, as well as examine other new morpho-syntactic forms. Nonetheless, we believe that this work provides valuable initial insights and a useful empirical basis for future investigations on this emerging linguistic phenomenon.
Acknowledgments
We would like to thank Suzanne Lesage for her work on the creation and data collection of the first experiment, as well as the Laboratoire de Linguistique Formelle for providing their servers to host our online experiment. ChatGPT (OpenAI, GPT-5) was used to assist in revising the English of the manuscript. The authors reviewed and edited all outputs and take full responsibility for the final text.
Competing interests
The authors declare that they have no competing interests.
Open access
