1. Introduction

1The studies on how sexes are represented in language pertain to a transdisciplinary field of research where linguistic aspects intersect with psychological and social issues (Stahlberg et al. 2007). The various types of gender representations in language, along with their asymmetries, is a matter widely studied in linguistics (Hellinger and Bußmann 2001) as well as in social psychology (Horvath et al. 2016; Hodel et al. 2017). Some of these studies have also affected Italian (Lepschy, Lepschy, and Sanson 2001; Marcato and Thüne 2002; Mucchi-Faina 2005; Maturi 2020), where a renewed debate has spread in the recent past on the use of a more gender-inclusive language.12

2The presence of gender biases and stereotypes has drawn much attention even in the Natural Language Processing community.3 Research in this field mainly focuses on the study of a model’s performance on data associated with a certain gender, or rather on the association between gender and certain concepts as found in language models (Sun et al. 2019).

3The present work, instead, aims at giving an exploratory account of the linguistic visibility of women in Italian language, with a focus in particular on job titles. For this purpose, we analyze the use of feminine forms used for job titles and professional roles in Twitter.

4Studies on corpus-based discourse analysis have already focused on gender issues with respect to job titles in Italian. They either quantitatively evaluate the mostly used gendered forms in texts when referring to female referents (Formato 2016, 2019; Voghera and Vena 2016), or rather assess, by means of a survey among native speakers, the degree of acceptability of some feminine job titles (Castenetto and Ondelli 2020).

5From a theoretical point of view, such works revolve (overtly or more indirectly) around the notion of markedness in language, that can be intended here as the “contrast between the unmarked (general, usual, non-salient) and the marked (special, emphatic)” (Clyne et al. (2009) cited in Formato (2019b, p.50)).4 In the present context, the “general, usual, non-salient” case is represented by masculine forms when used to express a generic reference. This means that grammatical masculine nouns are perceived and used as unmarked terms (for both men and women) based on the idea that they represent how the world is, opposing marked feminine terms which are seen as new, ungrammatical and ‘sounding bad’.

6While sharing with the studies mentioned above the same theoretical premise, the present work addresses the issue of women visibility in Italian language relying on user-generated data retrieved from Twitter: its peculiar nature as language data source, along with the opportunity it offers to extract and filter data based on specific keywords and time spans, makes this platform particularly useful for our purposes.

7More precisely, we aimed at studying the distribution over time of a selection of feminine job titles, distinguishing in particular the following cases:

• the use of marked forms, i.e. feminine forms referring to female professionals (e.g. la sindaca Raggi (‘mayor_fem Raggi’));

• (for a restricted set of examples) the use of semi-marked forms (Formato 2016), i.e. the combination of masculine forms and feminine modifiers when referring to female professionals (e.g. la neo-ministro è incinta (‘the_fem new_masc minister_masc is pregnant’)).

8We thus provide some background knowledge on the main linguistic conventions of Italian language in the assignment of grammatical gender, also mentioning some of the well-known studies that have challenged such conventions over the years, towards a more inclusive use of feminine forms, especially for professions. We then describe how data has been collected and filtered, and show the distribution of the selected job titles in both forms and across a 15-year time span.

2. Background

9Italian is a grammatical gender language5 and provides for the mandatory classification of the noun and its respective targets in agreement (modifiers, such as the adjective or the article) according to two values: masculine and feminine. The gender value is assigned according to phonological and semantic criteria (Thornton 2005). In assigning gender to nouns denoting human referents, there is a strong tendency to semantically match grammatical gender with the sex of the referent (e.g., la maestra è arrivata vs. il maestro è arrivato - ‘the teacher arrived’).

10Typically, the masculine is ‘overextended’ in reference to mixed groups (e.g. tutti i candidati ammessi - ‘all_masc admitted_masc candidates’_masc) or abstract functions (e.g. le elezioni a sindaco - ‘the mayoral_masc elections’), as well as in the case of individuals whose gender is not (yet) known (e.g. assumeremo un nuovo impiegato - ‘we’ll hire a new employee’_masc). However, there are cases in which, despite the existence of the feminine form, the masculine is also preferred to refer to a woman, especially when the person holds a prestigious position (Voghera and Vena 2016). In such a case, the assignment of grammatical gender does not follow this semantic criterion: unmarked expressions referring to a woman (Thornton 2009, 126) or semi-marked expressions6 are well attested. If we consider gender not only as a morphological category, but also as a semantic category, we can understand that, in the symbolic horizon within which the preceding examples move, masculine gender is taken as a neutral (or unmarked) form.

11The assumed neutrality of masculine forms has already been questioned from several points of view (Cavagnoli 2013; Thornton 2016; Voghera and Vena 2016). The seminal work by Alma Sabatini (1987), and the one proposed, more than two decades later, by Cecilia Robustelli (2012), have clarified the existence and use of feminine forms already provided for by the Italian linguistic system, and allowed the formulation of recommendations and guidelines for a more inclusive gendered language.

12While such reform proposals went largely unheeded (Merkel, Maass, and Frommelt 2012), more recent studies seem to reveal a slight change in linguistic habits among Italian native speakers (Castenetto and Ondelli 2020). Hence the choice to verify, by means of an analysis of user-generated content retrieved from Twitter, if a paradigm shift can be found with respect to the use of more gender-inclusive forms.

3. Data Collection

13Starting from the proposals presented in the recommendations of Sabatini (1987) and Robustelli (2012), we selected a shortlist of 11 job titles with both masculine and feminine endings. The selection is based on morphological criteria, more precisely on the different categories of gender suffix pairs that can be added to the root of a noun. We thus included the following terms:

14Job titles ending in -o_masc / -a_fem:

• ministro/ministra (‘minister’),

• sindaco/sindaca (‘mayor’).

15Job titles ending in -tore_masc / -trice_fem:

• rettore/rettrice (‘rector’).

16Job titles ending in -ere_masc / -era_fem:

• ingegnere/ingegnera (‘engineer’).

17Job titles ending in -o_masc / -a or -essa_fem7

• avvocato/avvocata/avvocatessa (‘lawyer’).

18Twitter recently introduced APIs (v2) that allow to access the full history of public conversations since the first tweet was created on March 21st, 2006. Accordingly, we take advantage of Twitter’s full-archive search endpoint8 for retrieving each tweet written in Italian and containing at least one of the words listed above, from March 21st, 2006 to March 21st, 2021 aiming at depicting the scenario of their use diachronically through a span of 15 years.

3.1 Data Cleaning

19A preliminary data analysis shows several noisy tweets in the dataset. Some keywords are indeed particularly affected by homonymy and polysemy. For example, the keywords sindaco and sindaca are also inflections of the verb sindacare (‘to judge, criticize, inspect’). A particular example is also the homonymy of the word rettore (‘rector’_masc) with the surname of a famous Italian singer and songwriter (Donatella Rettore), and of the word avvocata (‘lawyer’_fem) with a homonymous district of the city of Naples, Italy.

20Other keywords are also affected by the use of figurative language. Particularly relevant is the use of the keywords ministro and avvocata in a religious context. Indeed, in Christianity, priests are also called ministri di Dio (‘ministers of the Lord’), while avvocata nostra (‘most gracious advocate’) is part of the prayer ‘Hail Holy Queen’. These few examples help to catch a glimpse of the difficult task of cleaning and removing noisy tweets from this dataset automatically. Therefore, we performed a semi-automatic data cleaning by using filters tailored for each word.

21The final dataset consists of around 9.7 million tweets overall; Table 1 reports the number of tweets per keyword, as resulted after the cleaning process.9 Drawing inspiration from studies in demography, where male to female ratio is a common parameter, we report the proportion of masculine (M) and feminine (f) forms in terms of m/f ratio, where the higher the value the greater the unbalance between the two forms at the expense of the latter.

22On the numerical front we can see that the number of tweets containing the masculine form is greatly dominant. This is especially evident in the case of the keyword pair ingegnere/ingegnera (m/f ratio of 61.22) despite the fact that the ratio of male and female engineers in Italy is 5.38.10

23On the other hand, the feminine words that seem to be used in the most balanced way with respect to their masculine counterpart are ministra (m/f ratio of 12.32) and sindaca (m/f ratio of 15.62).

4. Data Analysis and Discussion

24The first step of our data analysis consists of observing the trends of the frequency of use of the six women’s job titles explored in this work.

25In Figure 1 we represent the frequency of feminine job titles with respect to the total of terms used to describe the profession (fem / fem + masc). We observe that from 2006 to 2021 there is a tendency to a more frequent use of female forms in general. However, relevant spikes are present on the left side of the chart. We believe they are caused by the scarcity of data before 2010, which is also imputable to the low popularity of the microblogging platform in Italy before that year. Furthermore, among the 6 sixfeminine keywords used as case study in the present work, 2 of them do not even have any occurrence in the totality of the year 2006. Their use starts with a few occurrences only from the year after (avvocatessa and rettrice).

Figure 1: Frequency trend of women’s job titles from 2006 to 2021.

26The purple  line (see Figure 1), illustrating the trend of the word ministra (‘minister’_fem) shows how the word has been increasingly used around 2016-2017, and then again around 2019-2020. The use of this term seem to increase during the election period and to decrease immediately afterwards. This outcome is indeed in line with the periods in which governmental changes occurred in Italy. In particular, in both those time spans there have been female ministers who have been highly politically exposed.11

Figure 2: Ratio between the marked forms and the semi-marked forms for the terms ministra/o (‘minister’, on the left) and avvocata/o (‘mayor’, on the right).

27Another fact worth mentioning is the trend of the mustard-yellow  line in Figure 1 depicting the use of the word sindaca (‘mayor’_fem). The word seems to have started to be used more frequently in conjunction with the election of two female mayors in two large Italian cities.12 Also the relationship between red  and blue  lines in the same figure presents a notable trend. Those lines respectively show the use of avvocata and avvocatessa (both: ‘lawyer’_fem). It is peculiar how the two lines show the same tendency throughout the years with the preference for the term avvocatessa on top of avvocata, until the year 2017. From that moment on, there is an inversion of trend and the occurrence of first term starts decreasing (blue  line), favoring the use of the second one (red  line). The oscillation between avvocatessa and avvocata therefore remains, but it seems that the latter has been increasingly gaining some ground.

28The word rettrice (‘rector’_fem), marked by the orange  line in Figure 1, has an averagely growing distribution through time (around 2%), with a spike of increase in 2020, when – for the first time – a woman has been elected as rector in the biggest university of Europe: La Sapienza in Rome.

29Finally, ingegnera (‘engineer’_fem) is the only one among the six terms taken into consideration with a low, though constant, trend throughout the temporal span of 15 years (around 1.6%), with only one recent spike around 2020-2021 (green  line).

4.1 Analysis of N-grams

30In a second step of our analysis, we aimed at investigating on the use of semi-marked forms (see Section 1). We focused on the two terms that presented the most balanced distributions with respect to their masculine counterpart (see Table 1), i.e. ministra and sindaca, and studied when and how the masculine form has been used to refer to a female referent in the real world. To do so, we extracted n-grams where one of the two tokens is one of the masculine words selected for the study and the second token is a feminine determiner or nominal modifier.

31Hence, we selected the following 2-grams of interest:

• la ministro/sindaco
  (‘the_fem minister/mayor’_masc)

• ministro/sindaco donna and donna ministro/sindaco
  (‘female minister/mayor’_masc)

• signora ministro/sindaco
  (‘Madame minister/mayor’_masc)

32In Figure 2 we show two charts (one for the word ‘minister’, and one for the word ‘mayor’) illustrating the ratio between the selected marked forms and the sum of such forms with semi-marked forms.

33In both cases it is once again evident that the data collected before 2010 is very scarce, and that relevant statistics are, therefore, to be considered valid only after that year.

34For both charts it is shown how the tendency of using marked forms (la ministra and la sindaca) is growing throughout the years; on the other hand, expressions where the female attribute is explicitly mentioned – such as signora ministra (‘Madame minister’_fem) and signora sindaca (‘Madame mayor’_fem) – are still very frequent (red  lines in both charts).

35Despite the outcomes derived from the analysis of n-grams, we acknowledge that the procedure described in this subsection is fairly limited. Beside the fact we studied the distribution of only two words out of the six selected for the present study, the availability of the same data enriched with part-of-speech tagging and parsing information would be highly beneficial for the automatic identification of marked and semi-marked forms.

5. Conclusion and Future Work

36In this paper, we reported the results of a corpus-based account of the linguistic visibility of women in Italian language, with a focus in particular on job titles, and using Twitter as data source. From a preliminary analysis of a selection of profession nouns, we found that some marked forms are increasingly being preferred in spite of semi-marked expressions. Besides extending and systematizing this analysis to other case studies, we also aim to observe the usage of such forms by Italian native speakers by tackling the issue as a stance detection task, so to assess how the users value a given marked form and, more in general, the adoption of more gender-inclusive linguistic habits. Furthermore, the messages leveraged on this topic might overlap with the task of misogyny detection and hate speech detection as well, broadening the horizons of three different NLP detection tasks. This design choice can also be motivated with regard to contextual stance detection (Cignarella et al. 2020; AlDayel and Magdy 2021), to investigate how supporters/opponents of inclusive language strategies are segregated in different online social network communities.

37Finally, due to its preliminary and exploratory nature, this work only reports the distribution of feminine and masculine forms, which are the two values for gender assignment taken in consideration for the analysis. We are well aware, however, that a comprehensive study of gender-inclusive language must necessarily cover all those linguistic forms that refer to the multiple and diverse identities in the gender spectrum.

38With respect to this point, innovative forms have been proposed in the last years, in order to overcome the binary opposition, even in a grammatical gender language as Italian, such as the schwa (Ə), the asterisk (*), the ‘at’ sign (@), and other graphic solutions. This is another aspect that is worth exploring in a stance detection perspective, so to assess users’ stance regarding the use of such linguistic innovations and their spread in everyday language.