Reading and comprehension of the masculine used as generic in Italian professional role names

This study investigated the influence of grammatical information-specifically, the use of masculine forms as generic—on sentence comprehension in Italian. A sentence evaluation paradigm was used. In each trial, a first sentence introduced a professional role name (e.g., Gli architetti uscivano...), followed by a second sentence providing gender information (e.g., una delle donne...). Participants were asked to judge whether the second sentence was a sensible continuation of the first. Results indicated that participants' judgements were biased by the masculine plural form used as generic. The analysis of reaction time data showed that female continuations led to longer reaction times than male continuations, consistent with the judgement data. These findings align with align with previous research showing that in gender-marked languages, such as German and French, masculine generics bias mental representations in an androcentric way.

Theoretical frameworks for the cognitive processes underlying sentence and discourse comprehension widely recognize the interaction between linguistic and extra-linguistic factors. Linguistic factors encompass the core components of the language system, such as phonology, syntax, and morphology, along with the lexical processes of word identification and parsing (Perfetti et al., 2005, p. 229). These linguistic mechanisms are critically enhanced by extra-linguistic factors, which include the comprehender's general and conceptual knowledge and the inferential processes required to build a coherent situation model, or mental representation, of the text (Dijk and van Kintsch, 1983; Perfetti et al., 2005, p. 229). These inferences are the relevant processes for achieving deeper comprehension; specifically, they consist of cognitive operations that bridge disparate elements within the text or link textual information with the reader's prior knowledge to maintain the coherency necessary for understanding (Perfetti et al., 2005, p. 231). A specific and fundamental type of inference involves anaphoric resolution—the process by which a linguistic expression (the anaphor) is linked to a preceding element in the text (the antecedent), taking that antecedent as its referent. Crucially, gender information provides a robust cue for this mapping process. Such information can be conveyed through formal linguistic features, such as grammatical gender, or through the activation of social expectations, known as stereotypical gender. Within this mechanism, gender serves as a primary cue for establishing text coherence, as it guides the reader in the identification and interpretation of the intended referent (Sturt, 2003). The impact of gender information on mental model construction is famously illustrated by the “surgeon riddle” (Sanford, 1985, p.

311), where the inability to immediately resolve a gender mismatch reveals the strength of stereotypical associations. While this riddle highlights a general bias in role-assignment, such stereotypical expectations directly constrain anaphoric resolution; indeed, gender information is also retrieved through extra-linguistic sources, specifically via elaborative inferences1 based on social stereotypes. In this context, professional role names (e.g., surgeon, nurse) can trigger a mental representation of a specific gender even in the absence of explicit grammatical marking. Garnham et al. (2002) demonstrated the immediate nature of these inferences, showing that readers experience significant processing costs—such as reading slowdowns and regressions—when they encounter anaphoric pronouns or lexical elements that mismatch the stereotypical gender of the previously mentioned discourse referent. This processing cost, reflected by longer durations in self-pace reading (Carreiras et al., 1996), has been termed the mismatch-cost by Kreiner et al. (2008), who examined the different effects of definitional and stereotypical nouns (Kreiner et al., 2008, p. 256–257).

The Italian language offers prolific ground for examining the relationship between specific morphosyntactic choices and the resulting mental representations of the referents' gender. Italian is classified as a grammatical gender language, a system where every noun is assigned to a grammatical gender and human-referent role nouns typically possess distinct masculine and feminine forms [e.g., infermiere/infermiera (“nurse”)] (Gygax et al., 2019a,b). In this system, gender is an obligatory morphosyntactic category, meaning it must be expressed through agreement morphology: articles, adjectives, and past participles2 are structurally required to agree in gender with the noun they modify. This constant linguistic marking constrains the reader to assign a specific gender to discourse entities. Crucially, the masculine form in Italian operates through a system of semantic and functional duality. In standard Italian, the masculine plural is traditionally considered “neutral” grammatical gender (D'Achille, 2021). As an explicit grammatical convention, it is used as a generic form to refer to individuals of unknown sex, groups of mixed-sex composition, or cases where the gender of the referent is considered irrelevant (Gygax et al., 2008, 2021). This rule introduces systematic ambiguity within the language. While the feminine plural exclusively denotes female referents, the masculine plural must fulfill two distinct roles: a specific use (referring only to men) and a generic use (referring to a mixed or neutral group). Consequently, during real-time processing, readers are required to perform a mental inference to determine the intended meaning based on the context (Gygax et al., 2021). However, this cognitive decision-making process is not neutral. Empirical research suggests that the masculine plural often triggers a male-biased interpretation rather than a truly neutral representation (Gygax et al., 2008; Garnham and Yakovlev, 2015; Redl et al., 2018; Anaya-Ramírez et al., 2022; Xu and Sturt, 2023; Rothermund and Strack, 2024).

The investigation of gender information in professional role names—a central theme in the studies by Kreiner et al. (2008)—has been a primary focus of psycholinguistic research in Italian as well. Specifically, Cacciari and Padovani (2007) evaluated the extent to which gender stereotypes associated with 36 agent nouns exert a priming effect during the processing of anaphoric pronoun links. This line of inquiry in the Italian context follows the influential work of Banaji and Hardin (1996), who first demonstrated the automatic activation of gender stereotypes in English.

In Cacciari and Padovani's study, participants were presented with stimuli such as teacher–he or engineer–she, where the pronoun either matched or conflicted with the gender stereotype of the profession. While their initial experiment yielded no significant effects—likely due to the insufficient processing time allowed for the relatively long Italian prime nouns3 (Cacciari and Padovani, 2007, p. 285)—their second experiment, which employed increased stimulus presentation times and a longer stimulus onset asynchrony (SOA), provided evidence for stereotypical priming. The results demonstrated that reading a professional role name significantly influences the processing of the subsequent pronoun. Specifically, when the prime conveyed a female stereotype, an inhibitory effect was observed for masculine pronouns (e.g., teacher–he). Conversely, no such inhibition was detected for feminine pronouns following masculine-stereotyped primes (e.g., engineer–she) (Cacciari and Padovani, 2007, p. 289). However, they noted that from Corbett (1991) to the more recent psycholinguistic studies on anaphora resolution (e.g. Irmen and Roßberg, 2004), it has been demonstrated that the use of the generic masculine (the male form used for mixed group of men and women or when the subject of the sentence is unknown) favors a male interpretation (Cacciari and Padovani, 2007, p. 289).

Based on their previous study, Cacciari et al. (2011) hypothesized that when the antecedent is gender-marked as an epicene noun4 (e.g., vittima, “victim”), pronoun resolution is primarily guided by grammatical gender, while when the antecedent is a double- and multiple-gender noun5 (e.g., assistente, “assistant”) context is responsible for assigning it a gender value (Cacciari et al., 2011). Their study on pronoun resolution demonstrated that for nouns with a fixed grammatical gender (epicenes), the initial cognitive phase of anaphor resolution —defined as “bonding”—is temporarily impervious to contextual information. According to these authors, this suggests that the cognitive system prioritizes morphosyntactic agreement cues immediately during the early stages of processing, and only in a later “resolution” stage is the mental model adjusted according to contextual or stereotypical information. On this view, these findings suggest that the pervasive nature of gender marking in Italian creates a powerful grammatical anchor that is computationally difficult to override through purely semantic or contextual means.

Some psycholinguistic studies conducted in English, French, German, Spanish, Dutch, Russian and Chinese investigated the relationship between the stereotypical and/or grammatical information conveyed by a particular linguistic choice (in this case, the generic masculine) and the interpretation by the reader, using professional role names as stimuli (Gygax et al., 2008; Garnham and Yakovlev, 2015; Redl et al., 2018; Anaya-Ramírez et al., 2022; Xu and Sturt, 2023; Rothermund and Strack, 2024).

The study by Gygax et al. (2008), which serves as a foundational framework for the present research, compared German, French, and English to determine how grammatical cues interact with stereotypical information within the process of anaphoric resolution. Crucially, these languages represent two distinct linguistic systems: while French and German are grammatical gender languages, the English condition is gender-unmarked, as its professional role nouns (e.g., beauticians) do not carry formal gender morphology. The task involves pairs of sentences, in which the first sentence of each pair describes a group of people, presented through a professional role name declined in the plural (e.g., politicians, authors, beauticians), while the second sentence specifies whether there are women or men in the group, using a quantifier. Among the 36 professional role nouns used in each experiment pair, according to the normative study of Misersky et al. (2014), 12 are considered by the relevant speaking community as stereotypically feminine, 12 as stereotypically masculine, and 12 as neutral. An example of the sentence pairs, in which the professional role name is stereotypically female, is provided below:

(1a) The social workers were walking through the station.

(1b) Since sunny weather was forecast several of the women weren't wearing a coat (Gygax et al., 2008, p. 472).

After reading each pair of sentences, participants were asked to determine if the second sentence was “a sensible continuation of the first one” (Gygax et al., 2008, p. 475). In addition, the authors used 36 filler texts to ensure that the participants remained engaged with the task. These filler pairs were similar to experimental stimuli, but included a semantic mismatch between the first and second sentence, which was intended to elicit a clear “no” answer. The only difference among the three languages lies in the fact that the professional role nouns in French and German are declined in the masculine (e.g., les assistants sociaux, die Sozialarbeiter), according to the generic masculine convention, the English nouns remain gender-neutral (e.g., the beauticians). This design allows the authors to examine how the presence or absence of grammatical gender marking interacts with social stereotypes.

In the analysis of the proportion of positive responses among test trials, different trends emerged between French and German on the one hand and English on the other. In the first two languages, a significant main effect of continuation was recorded, when the continuation sentences mentioned men increased likelihood of correctly identifying the second sentence as a sensible continuation of the first (French: M = 0.78; German: M = 0.69) compared to those with the condition “women” (French: F = 0.58; German: M = 0.40), in response to the question of whether the second sentence was a sensible continuation of the first one. No effect was noted regarding the stereotypicality of professional role names. Conversely, in the English experiment, no main effect of continuation or the stereotype of the role nouns was recorded. However, an interaction between the two variables was found, such that, given a male stereotyped role noun, continuations with “men” were judged as more sensible than those with “women,” with the reverse pattern for female stereotypes, and no significant difference for neutral stereotypes (Gygax et al., 2008, p. 476–477). In conclusion, if grammatical gender information is present (as in French and German), it overrides the reader's stereotypical knowledge. In contrast, where grammatical marking is absent (as in English), the inferences are based on extralinguistic information, namely stereotypes, associated with that name (Gygax et al., 2008, p. 480).

The conclusions of Gygax et al. (2008) for English were corroborated by an equivalent study conducted by Xu and Sturt (2023) on Mandarin Chinese. Although it is a typologically different language from the Indo-European ones, Chinese shares with English the lack of gender marking in nominal morphology (Xu and Sturt, 2023, p. 4). The Xu and Sturt's data reveal that in the presence of a role name considered stereotypically feminine, the proportion of positive responses is higher when the second sentence is referring to women (0.79) compared to when it is “men” (0.65). Conversely, after reading a stereotypically masculine noun, a higher number of positive responses is recorded when the continuation indicates the condition “men” (0.71) compared to when “women” is encountered (0.44) (Xu and Sturt, 2023, p. 7). This contrasts with grammatical gender languages, where, as mentioned above, formal linguistic cues often override these social stereotypes. This pattern, demonstrated by Gygax et al.'s (2008) results on French and German, has been extensively confirmed across different language families. For instance, Garnham and Yakovlev (2015) demonstrated that in Russian— a language in which ≪all nouns are marked grammatically and agree with adjectives, verbs (past-tense) and pronouns […] ≫ (Garnham and Yakovlev, 2015)—grammatical gender overrides stereotypical information. Their findings show that the masculine plural, even when it is intended as generic, biases a male-specific representation, regardless of whether the professional role is stereotypically feminine or neutral. Similarly, evidence from Spanish—which has a grammatical gender system closely related to that of Italian—aligns with the Gygax et al.'s results (2008) for French and German. Anaya-Ramírez et al. (2022) found that Spanish speakers interpret the masculine plural as predominantly male-specific, and that accessing a mixed-gender interpretation requires significantly greater cognitive effort. The psychological nature of this bias in gender-marked languages has been further elucidated by Rothermund and Strack (2024) in a recent study on German. While Gygax et al. (2008) established the existence of the male bias, Rothermund and Strack investigated whether it stems from a misunderstanding of the speaker's communicative intention or from automatic associative processes. Their findings reveal that even when participants are continuously reminded of the generic intention through a novel grammatical marker, the male bias remains robust. This suggests that the androcentric representation triggered by the masculine plural is driven by associative processes that are largely immune to explicit disambiguation, posing a significant challenge to the effectiveness of purely contextual or prescriptive reminders of gender neutrality.

The pervasiveness of the male bias is the dominant finding for languages like French, German, and Spanish. Therefore, a further body of research has sought to establish a causal link between the implementation of non-sexist linguistic forms and the reduction of androcentric bias during inference formulation. For instance, Sato et al. (2016) investigated nominalized adjectives and participles in German, finding that gender-fair alternatives effectively alter male-dominant representations not only in first-language speakers but also in second-language learners. Further clarifying the cognitive impact of specific gender-fair strategies, Körner et al. (2022) examined the gender star form (e.g., Athlet * innen) in German. Their findings revealed a significant shift in mental models: while the generic masculine consistently elicited a male bias, the gender star form triggered a female bias, with participants more quickly identifying female referents as consistent with the preceding noun than male referents. Interestingly, the study found that only pair forms (e.g., readers and female readers) yielded a truly balanced representation of both genders. These results suggest that while innovative contracted forms are effective at increasing female visibility, they may do so by over-orienting the reader's attention toward the feminine component, whereas explicit doubling remains the most neutral and equitable strategy. Similarly, Tibblin et al. (2023) demonstrated in a study on French that the use of gender-fair forms significantly attenuate the male bias during the resolution of anaphoric expressions. Their results suggest that explicit linguistic cues for inclusion can successfully compete with masculine-driven defaults. Furthermore, Spinelli et al. (2023) provided a critical comparison between different gender-fair strategies in French. Their findings reveal that while neutralization strategies (e.g., using epicene nouns or collective terms) may appear theoretically inclusive, they often fail to significantly increase the visibility of women in the speaker's mental model. In contrast, explicit feminization—such as the use of split forms—proved to be far more effective in counteracting the masculine default. This distinction suggests that for a linguistic strategy to successfully rebalance gender inferences, it must proactively challenge the associative strength of the masculine generic rather than simply attempting to bypass it. The hypothesis that gender representations are dynamically shaped by the specific linguistic system in use is further corroborated by research on bilingualism. Sato et al. (2013) investigated how bilinguals of French (a grammatical gender language) and English (a natural gender language) navigate gender inferences. Their results revealed a striking “cognitive switch”: while processing in French, both native and non-native readers tended to rely on morphological cues, leading to a consistent male bias. Conversely, when switching to English, the same individuals shifted their reliance toward social stereotypes. Crucially, Sato et al. (2013) found that this shift is modulated by L2 proficiency: as bilinguals become more advanced in their second language, their mental models more closely mimic the processing patterns of native speakers. These findings support the “thinking-for-speaking” hypothesis (Slobin, 1996), suggesting that the morphosyntactic structures of a language act as attentional drivers that direct the speaker's focus toward either grammatical or stereotypical information. This reinforces the notion that the male bias in languages like French or Italian is not merely a byproduct of social knowledge, but a direct consequence of the gender markedness of language.

This interpretation is supported by recent findings in Italian. Safina (2024) specifically investigated this through a sentence-evaluation paradigm in Italian (Study 4), which directly compared different gender-sensitive strategies. Her results demonstrate that while gender-neutral forms (standard neutralization) show limited success in mitigating male bias, explicit feminization through double forms remains the most effective strategy for ensuring unbiased anaphor resolution. Furthermore, Safina (2024) highlights that individual attitude toward gender-fair language significantly modulates real-time processing, with positive attitudes increasing the likelihood of inclusive mental models even when faced with innovative, non-standard forms like the schwa.

However, the availability and relative weight of these cues vary across languages. While in languages like German, French, Spanish and Russian grammatical information—specifically the use of the plural masculine—appears to override extralinguistic knowledge, English and Chinese exhibit a different tendency. In these languages, which lack pervasive grammatical gender, stereotypical bias appears to play a more central role in mental model construction, as there is no formal grammatical marker to override social expectations.

Building on the framework outlined above, the present study aims to complement recent findings in Italian (Safina, 2024). Both studies adopt the sentence-evaluation paradigm and stimulus construction model established by Gygax et al. (2008), providing a robust and comparable methodological foundation. However, while Safina (2024) focused on the effectiveness of different gender-sensitive strategies (e.g., neutralization and explicit feminization) in reducing male bias in anaphor resolution using non-independent nouns (see below for a detailed morphological of this category), our research shifts the focus back to the “baseline” condition: the generic masculine. In fact, we decided to conduct a sentence-evaluation study on symmetrical (see below for a detailed morphological of this category) professional role nouns in Italian within the convention of the generic masculine. This allows us to investigate how the morphological structure of Italian interacts with professional stereotypes in the absence of gender-fair mitigation. Therefore, according to the theoretical-methodological framework outlined thus far, similar results are expected for Italian as those obtained in the experiments on French and German (Gygax et al., 2008). The findings described above suggest that grammatical information prevails over stereotypical information in the construction of inferences, thus leading to a mental representation influenced by an androcentric bias.

The role nouns were selected based on Misersky et al.'s (2014) gender stereotype norms for 422 Italian nouns. Misersky et al.'s study asked participants to estimate, on an 11-point scale, the proportion of women and men in the presented social and professional groups, based on their actual composition and not on how they should be according to their own view. These ratings were then converted to proportions, such that”100% women and 0% men” was represented with the value 1, “50% women and 50% men” with the value 0.5, and “0% women and 100% men” was indicated with the value 0. In the role names selected for our stimuli, the 12 stereotypically masculine nouns had an average perceived proportion of 0.31 (range 0.25–0.44), the 12 neutral nouns had an average proportion of 0.5 (range 0.45–0.55), and the 12 feminine nouns had an average proportion of 0.76 (range 0.56–0.86). Example 1a, b above shows one of the stereotypically male items. For the selected role nouns, a further selection criterion was applied, related to a characteristic intrinsic to the Italian language. To understand the linguistic constraints on professional role nouns, it is essential to frame them within the universal theory of gender assignment. According to Corbett (1991), gender systems are governed by assignment rules that can be either semantic (based on the referent's biological sex) or formal (based on morphological or phonological properties of the word). While many languages exhibit a convergence of these criteria, Corbett notes that if there are conflicting factors at work, semantic factors usually take precedence' (1991, p. 68).

Crucially, the categories used to describe these phenomena are cross-linguistically valid and not exclusive to the Italian language. Within the international typological framework, Thornton (2022) provides a detailed classification of Italian professional role names based on their relationship with the referent's sex. She identified five categories of names: epicenes (e.g., la persona, “the F person”), nouns with a fixed grammatical gender that can refer to individuals of any sex without changing their form or agreement; common gender (e.g., il cantante-la cantante, “the F/M singer”), nouns that have a single morphological form but trigger different gender agreements (masculine or feminine) depending on the referent's sex; independents (e.g., il padre-la madre,”the father-the mother”), lexically distinct terms for male and female referents that do not share the same root; symmetrical (e.g., il ragazzo-la ragazza, “the boy-the girl”), pairs of nouns where the feminine form is systematically derived from the masculine through inflectional or derivational changes; asymmetrical (e.g., il professore-la professoressa, “the F/M professor”), pairs where the feminine form is marked by specific suffixes that do not follow the standard symmetrical paradigms. This specific morphological organization in Italian dictates how the “generic masculine” is formed and processed, directly influencing the cognitive costs associated with gender-mismatch effects.

Within the list proposed in the normative study by Misersky et al. (2014), only the nouns defined as symmetric by Thornton (2022) were selected in this study. This choice was driven by the necessity to limit variables concerning the interaction between suffixes and the perception of gender, thereby reducing the experimental linguistic input to solely the ending that conveys the masculine plural. The list of selected professional role names is provided in Table A1 in Appendix.

Participants

The participants consisted of 77 native Italian-speaking students from various regions of Italy, enrolled in different Italian universities. There were 35 males and 40 females, as well as two participants who preferred not to reveal their gender. The mean age was 24.36 years (range 18–52).

Materials

The experimental stimuli consisted of 36 sentence pairs similar to 2a, b, below (see 2a', b' for an English translation):

2a) I chimici entravano nell'ufficio.

2b) Dato il brutto tempo, la gran parte [degli uomini]/[delle donne] indossava un impermeabile.

2a') The chemists were entering the office.

2b') Given the bad weather, most [of the men]/[of the women] were wearing a raincoat.

Within each sentence pair, the first sentence introduced a role name in the masculine plural form, like i chimici (“the chemists”) in 2a. The second sentence introduced the gender condition, providing either a male or a female referent; this factor was manipulated within-item (example 2b above shows both male and female continuations). The gender stereotype of the role noun in the first sentence was manipulated between-items, such that of the 36 experimental items, 12 used stereotypically masculine nouns (e.g., architetti), 12 used stereotypically neutral nouns (e.g., biologi/biologists), and 12 used stereotypically female nouns (e.g., bibliotecari/librarians). The role nouns were selected based on Misersky et al.'s (2014) gender stereotype norms for 422 Italian nouns.

In addition, there were 36 filler items. In both experimental and filler items, the first sentence of a given item presents a group of people using a professional noun in the masculine plural, while in the second sentence of each pair, an indefinite quantifier specifies that there are subsets of women or men in the group,6 thus establishing the condition. In the first sentence of each pair, following the structure described by Gygax et al. (2008), p. 474), people are described as performing actions related to specific verbal categories of meaning: (1) uscire da un posto (coming out of a place), (2) aspettare in un posto (waiting somewhere), (3) entrare in un posto (going into a place), (4) essere da qualche parte (being somewhere), (5) camminare (walking), or (6) attraversare (going across) un posto (a place). Within the continuations, the content revolves around three distinct aspects: (1) one based on different emotions (rabbia7, buon umore, calma, tristezza—anger, good mood, calm, sadness), (2) one based on weather conditions (indossare un impermeabile, non avere una giacca, avere un ombrello, indossare un cappello—wearing a raincoat, not having a jacket, having an umbrella, wearing a hat), (3) one based on intentions and will (voler andare via, voler restare, voler una pausa—wanting to leave, wanting to stay, wanting to take a break). The experimental items were modeled after those developed by Xu and Sturt (2023), which were themselves modeled on those of Gygax et al. (2008), but our items were systematically adapted to the grammatical rules of the Italian language. This adaptation ensured that each sentence maintained logical and grammatical coherence according to the specific type of professional role noun used in the introductory phrase. Furthermore, the role nouns were selected based on their gender stereotypicality specifically within the Italian context, drawing from the norms established by Misersky et al. (2014). Consequently, a strict literal translation was avoided, as it would have failed to capture the morphosyntactic requirements of Italian—such as gender agreement and lexical nuances—and would have compromised the ecological validity of the stimuli. The task (described in the procedure section below) involved participants judging whether the second sentence was a sensible continuation of the first. For all experimental items, the target response was “yes,” while for all filler items, the target response was “no,” as the fillers were intentionally formulated with logical and pragmatic inconsistencies to ensure careful reading by the participants and to monitor the validity of the responses. Below is an example of a filler item (see 3a' and 3b' for an English translation):

3a) Gli animatori stavano prendendo il sole.

3b) Dato il maltempo, la gran parte delle donne indossava un maglione.

3a') The entertainers were sunbathing.

3b') Due to the bad weather, most of the women were wearing a sweater.

Procedure

The male and female continuation conditions of the 36 items were distributed into two lists, using a Latin Square design, such that each list contained one continuation for each condition (men or women), with an equal total number of continuations featuring women and men across all items in each list. Each of the two experimental lists was combined with a full list of 36 filler items. In this way, the sample was essentially divided into two groups, and each group was presented with 72 sentence pairs (36 experimental and 36 filler items).

The experiment was administered on-line using the PCIbex platform (Zehr and Schwarz, 2018), and participants took part via a web link). The activity was administered using a sentence-by-sentence non-cumulative self-paced reading technique. On each trial, participants pressed the space bar to see the first sentence, and then pressed the space bar again, at which point the first sentence was replaced with the second sentence. Under the second sentence, the two words “sì” (yes) and “no” (no) were displayed, and participants were instructed to press the “s” or the “n” key with the left (resp. right) index finger to indicate their judgement of whether the second sentence was a sensible continuation of the first (Figures 1A, B and Figures 2A, B).

The experiment was preceded by a consent form, and some questions gathering basic demographic data, after which two practice trials were given.

Analysis

Below, we report the analyses of proportion of “yes” responses to the judgment of the second sentence, followed by the analyses of reaction time, excluding trials with “no” responses. Fillers were not included in the analysis8.

Proportions were analyzed using the Generalized Linear mixed Effects models with a binomial link function. We used the lme4 package in R, with the bobyqa optimizer. The fixed effects included (a) continuation (male vs. female), which was centered prior to analysis, such that the predictor had a mean of 0 and a range of 1 (i.e., a form of sum contrasts, which is appropriate for a factor with two levels), and (b) stereotype, which used treatment coding, with the neutral stereotype as the reference level. Therefore, the results presented below detail the continuation effect at the neutral stereotype level, as well as interactions of this factor with the male and female stereotypes, to determine the difference in the effect of the stereotypical factor between neutral and masculine nouns, and between neutral and feminine nouns. The random effects structure included intercepts for both participants and items. Per-participant random slopes were included for both main effects and their interactions. Per-item random slopes were included for the continuation effect only, as this was the only experimental factor that was manipulated within-item. In cases of singular fit, or convergence failure, random slopes were removed beginning with those explaining the smallest variance, until convergence or non-singularity was achieved (Barr et al., 2013). Random effects did not include correlation parameters.

For the reaction time analyses, we took the reaction times for the coherence judgments of the second sentence, and removed trials with a “no” response. Following this, we removed outliers (defined as trials where the RT was less than 500 millseconds or greater than 30 s). Outlier removal resulted in the loss of less than 2% of the data. Then, we calculated linear mixed effects models using models that were analogous to those of the analysis of “yes” proportions, except that the millisecond-scale reaction-time data were first log-transformed, and then fitted to a Gaussian model. We used nmkbw, from the dfoptim package, for the optimizer in the reaction time analysis. Visual inspection using the R function qqnorm suggested that the residuals of the resulting model were a good fit to a normal distribution.

Analyses for both the yes-proportions and reaction times were followed up with analyses evaluating the overall interaction between stereotype and continuation, and the main effects of stereotype and continuation. These analyses are conceptually similar to omnibus interaction and main effect analyses in ANOVA. Given that one of the factors (stereotype) contained three levels, these analyses cannot be evaluated by inspecting the coefficients of the main model. Thus, we used model comparisons with a log-likelihood chi-square test. The model comparison for the overall interaction compared the fit of full model (described above) with a model that included all effects except the stereotype-by-continuation interaction. The interaction was judged to be significant if removing the interaction led to a significant reduction in model fit. The analysis of the main effects began with a model containing both main effects (but no interaction), and then compared this model (a) with a model that removed the effect of stereotype (to test the main effect of stereotype), and (b) with a model that removed the effect of continuation (to test the main effect of continuation). The main effect was judged to be significant if removing the effect of interest led to a significant reduction in model fit.

Proportion of positive judgements

In Table 1, the means and standard errors derived from the analysis of the proportion of positive responses are reported.

The analysis showed an effect of continuation, showing that for the (reference-level) neutral stereotypes participants were significantly more likely to respond “yes” when the continuation was male (M = 0.81) rather than female (M = 0.65) (b = −1.01, Z = −4.70, p < 0.001).9

The analysis did not show significant interactions between the continuation factor and either the male stereotype (b = 0.09, Z = 0.33, p = 0.74) or the female stereotype (b = 0.49, Z = 1.87, p = 0.06), thus we did not find evidence for a difference in the size of continuation effect between the neutral and male stereotypes, or between the neutral and female stereotypes. Follow-up pairwise comparisons were then conducted to test the effect of continuation for the male and female stereotypes separately. The analysis for the male stereotypes again confirmed a significant continuation effect in the same direction as the neutral stereotypes (male continuation: M = 0.83; female continuation: M = 0.69; b = −0.97, Z = −4.69, p < 0.001). The analysis for the female stereotypes again replicated this effect (male continuation: M = 0.74; female continuation: M = 0.65; −2.85, p < 0.001).

We then conducted follow-up analyses to evaluate the overall interaction of stereotype by continuation, and to evaluate the overall main effects of stereotype and continuation, respectively. The overall interaction model did not reach significance [ χ (2) 2 = 3.57, p = 0.17]. The main effect analysis showed a significant main effect of continuation [ χ (1) 2 = 22.33, p < 0.001], reflecting the general preference for male continuations across stereotypes, but no main effect of stereotype [ χ (1) 2 = 3.12, p = 0.21].

The results for the proportion of yes-responses demonstrate that people perceive the female continuations as incongruous, even though the meaning of the second sentence is logically coherent with the first. It can be inferred that for all three stereotypes (male, female, and neutral), there is a greater percentage of positive responses, with a value ranging between 9% and 16%, when the continuation is “men” compared to when it is “women.)”

Reaction times

We now report the analysis of Reaction Times for the judgment of the coherence of the second sentence. Before proceeding, we removed trials with outliers (RT greater than 30 s or less than 500 milliseconds). We then excluded all trials where the participant answered “no” to the judgment question, so that the analysis only considered trials with “yes” answers. Filler trials were also removed. In Table 2, we show the resulting means and standard errors.

Reaction times in milliseconds were log-transformed before analysis, and the fixed and random effects were defined as for the models for proportion of “yes” responses, except that the log reaction times were fitted to a Gaussian distribution. This analysis revealed no significant fixed effects (continuation for neutral stereotype: b = 0.05, t = 1.26, ns; interaction of continuation with male stereotype: b = 0.004, |t| <1; interaction of continuation with female stereotype: b = 0.05, |t| <1). In addition, we conducted model comparisons analogous to those for the response data described above, to evaluate the overall interaction of stereotype by continuation, and the main effects of stereotype and continuation. These analyses confirmed that the overall interaction and the main effect of stereotype were both non-significant [interaction: χ (2) 2 = 0.97, p = 0.62; main effect of stereotype: χ (1) 2 = 3.45, p = 0.18]. However, the main effect of continuation was significant [ χ (1) 2 = 8.02, p < 0.01], reflecting generally faster reading times for the continuations about men than for those about women. Note that this analysis effectively collapses over all three levels of stereotype, so may have more power than the fixed effects of the main analysis described above, where the continuation effect is tested for the neutral stereotype individually, before testing whether the size of this effect differs between the neutral and other stereotypes.

This research was inspired by Alma Sabatini's essay, “Il sessismo nella lingua italiana“ (1987), particularly its reflections on the syntactic and semantic ambiguity generated using the generic masculine form (Fusco, 2012, p. 85–94; Sabatini, 1993, p. 79). Over 30 years after the publication of the Sabatini's Raccomandazioni,10 and at a time when the Accademia della Crusca notes that “the number of queries related to gender has become numerous“ (D'Achille, 2021), we aim to determine if the androcentric interpretation of professional role names, when we use the generic masculine, remains predominant.

In addition, the experiment presented here was conducted with the aim of comparing the data obtained with those from the theoretical–methodological framework outlined by Gygax et al. (2008) and Xu and Sturt (2023), thereby expanding observations on the interaction between stereotypical extralinguistic knowledge and the generic masculine. This practice, among the languages examined in the reference literature, is shared by Italian, French, and German. We predicted that the answers of the Italian sample would be more closely aligned with French and German than with English and Chinese, where gender marking is rare in nominal morphology. Specifically, we predicted that, like in French and German, people would tend to interpret masculine plural role name descriptions as referring to males, and that, unlike in English and Chinese, these interpretations would not be strongly affected by gender stereotypes of the individual role nouns. Evidence of the interaction in Italian between grammar and stereotypes in professional role names was provided by Cacciari and Padovani (2007). In addition, these findings are further strengthened by recent evidence from Safina (2024), who observed a similar persistence of male bias in Italian using a comparable sentence-evaluation paradigm. Although the two studies focused on different morphological classes, Safina (2024) tested non-independent nouns while the present study focused on symmetrical nouns, the convergence of results is striking. Both datasets indicate that the masculine plural continues to function as a primary cue for male-biased mental representations, even when gender-fair strategies or neutral stereotypes are involved.

Overall, the results obtained for Italian indicate that the proportion of positive responses is higher when the continuation refers to men, with no evidence that this tendency differed by stereotype (Figure 3). Thus, as expected, the results align more with French and German than they do with English and Chinese.

The experiments conducted by Gygax et al. (2008), by comparing languages with different grammatical gender systems, already outlined a clear picture of the possible configurations in the interaction between grammatical and extralinguistic information: first, when profession names or the definite articles accompanying them do not convey gender, the mental representation elicited is shaped by the participant's extralinguistic knowledge. This conclusion emerges from Gygax et al.'s observations in the English conditions (Gygax et al., 2008, p. 480) and is corroborated by the study on Chinese by Xu and Sturt. The latter study shows that participants produced a higher number of positive judgements when the stereotypicality of the agentive nouns matched the continuations, compared to conditions where the two gender cues did not match (Xu and Sturt, 2023, p. 9). By contrast, when a language features gender marking in nominal morphology, grammatical information appears to play a dominant role in shaping mental representations, effectively overriding the influence of stereotypical information. This is evident from the analysis of results in the French and German experiments: “[…] we demonstrated that the use of the generic masculine does not lead to a gender-open or gender-spread representation, but to a specifically male representation” (Gygax et al., 2008, p. 480).

Regarding the examination of reaction times, this study showed a main effect of continuation in the overall analysis, showing that sentences that referred to women took longer to read than those that referred to men. There was no evidence for an effect of stereotype, or an interaction of stereotype-by-continuation. Observing analyses from Gygax et al. (2008) on languages with grammatical gender marking, a slight continuation effect was registered in French, but neither significant effects from stereotypes nor interactions were found. German showed a significant continuation effect, but no stereotype effects or interactions between variables were recorded. We should be cautious in interpreting these null effects, as they may well simply reflect insufficient power to detect the effects, given the noisier nature of reaction time data relative to judgement data. Looking at affirmative judgement analyses of languages where responses were influenced by stereotypical knowledge, such as English and Chinese, we note that in the former, neither a continuation effect nor an interaction between the two variables was found. However, in the latter, reaction times for positive responses were shorter when the stereotypes associated with the professional name and the continuations aligned compared to when they did not. Therefore, participants exerted more cognitive effort to evaluate whether perceived incongruities were acceptable, especially when trying to understand how a man could perform a job considered stereotypically feminine (Xu and Sturt, 2023, p. 9). Evidence for cognitive effort in cases of incongruity between expectations about social gender and grammatical gender had already been highlighted by Kreiner et al. (2008) regarding the differences in the interpretation of definitional and stereotypical nouns, and by Anaya-Ramírez et al. (2022) concerning the semantic ambiguity generated using the generic masculine form in agent nouns.

The judgement data makes it clear that the use of generic masculine influences comprehension processes. Consequently, in certain contexts, this linguistic choice can create significant problems of textual coherence and cohesion. In addition, studies on gender-fair language in French, for instance, Kim et al. (2022), found that switching from the masculine form to gender-neutral nouns can reduce masculine biases without producing large stereotype effects or requiring more cognitive resources for processing (Kim et al., 2022, p. 23). Similarly, Tibblin et al. (2023) demonstrated that using extended or epicene forms significantly increases the percentage of women perceived in reference to non-stereotypical agentive nouns, when compared to the traditional generic masculine (Tibblin et al., 2023, p. 21).

To conclude, building upon the current understanding of gender processing in Italian (e.g., Safina, 2024), a promising avenue for future research is to expand data collection on how feminine professional role names are read and understood. By focusing Thornton (2022)'s classification of Italian personal nouns, we could determine whether morphological distinctions within the same language influence processing.

Specifically, we expect that the degree of morphological transparency and the nature of gender assignment will modulate cognitive effort during discourse integration. For epicene nouns the “grammatical bonding” (Cacciari et al., 2011) might be particularly resistant to contextual or stereotypical overrides, potentially leading to increased processing costs when the referent is male. Conversely, common gender nouns may allow for greater flexibility. Here, the lack of a gendered suffix might attenuate the initial masculine bias, facilitating the integration of feminine referents compared to standard masculine-inflected nouns.

Furthermore, investigating independent or heteronymous forms (e.g., genero/nuora—son in law/daughter in law) could provide a baseline for “maximal transparency,” where gender is lexically encoded. Comparing these categories would allow us to test whether the “mismatch cost” observed in previous studies (Kreiner et al., 2008; Anaya-Ramírez et al., 2022) varies in intensity based on how gender is morphosyntactically marked. This would clarify whether the androcentric bias in Italian is a monolithic effect or if it is sensitive to the specific formal properties of the professional role name, thereby providing more nuanced evidence for the interplay between grammatical structures and social constructs.