The grammatical gender system can influence the way the semantic gender is perceived. Italian is a grammatical gender language, in which nouns are classified for gender. In this work, we investigate the usage of gender-specific forms of occupational titles in a diachronic corpus of 3 billion tokens extracted from two Italian newspapers. The hypothesis is that the usage of gender-specific forms might be inlfuenced by socio-cultural aspects, such as changes in the employment policy. We automatically collect a set of occupational titles and perform a diachronic analysis exploiting the frequency of gender-specific forms. Results show a correlation between changes in the usage of gender-specific forms and socio-cultural events.
Throughout history, the prerogative use of specific gender forms over particular professions can fade away by introducing changes in the language lexicon (e.g., neologisms) or in the language usage (e.g., word frequencies). The way the lexicon is affected by those changes depends on the grammatical gender system, i.e. the set of rules that define the agreement between noun classes forms and the other parts-of-speech. Grammatical gender systems can vary dramatically from one language to another. Gygax et al. (2019) propose a classification of languages based on their grammatical gender system. In this work, we focus on the Italian language, a grammatical gender language in which all nouns must be classiifed for gender. The Italian gender system admits
Copyright © 2021 for this paper by its authors. Use
permitted under Creative Commons License Attribution 4.0
International (CC BY 4.0).
three categories for nouns: gender-specific ending
nouns, mobile gender nouns, and nouns where the
gender is specified through determiners and
adjectives
The objective of this paper is to monitor how the use of gender-specific occupational titles has changed in the Italian language over the years through the use of diachronic analysis tools. We would like to emphasize that the goal is not to map the composition of men and women for each profession over time, as this cannot be reliably inferred from text. Instead, we are interested in gauging the cultural relevance of the genderspecific titles over time, as reflected in the news domain. Accordingly, the contributions in this paper can be summarized as follows: (i) We analyze emerging trends in the use of gender-specific occupational titles in the Italian language in a corpus of newspaper articles.1 (ii) We perform a deep-dive analysis of the figures that have guided a significant shift for two professions in particular.
Large diachronic corpora have already been
used to study social and cultural phenomenons that
affected language in a significant way. The Google
Ngrams Dataset
Occupational titles occurrences are extracted from
a diachronic corpus that comprises two
subcorpora. The former corpus is the “L’Unita`”
corpus
The first step of our investigation consists of
extracting a list of occupational titles from a
common Knowledge Base. Specifically, we have
exploited Wikidata
Once we have acquired the set of occupational titles T , the next step of the analysis consisted of measuring the frequency with which each term w ∈ T occurs for each year in the corpus described in Section 2. We also make use of the lexical information contained in said corpus in order to eliminate any remaining ambiguity in the words. In fact, for each occupational title, we counted a hit in the corpus if it appears with the NOUN tag. This allows us to avoid counting occupational titles that can be confused with verbs or adjectives, such as impiegato/impiegata, which can refer to the noun employee in Italian, but also to the past participle conjugation of the verb to employ.
Moreover, we only counted a hit if the word has ptwf odds(w)t = log ptwm
Operationally, odds(w)t specifies the probability that the feminine variant will appear in a text relative to the masculine form in the specified year t. We then obtain the time-series by concatenating the odds(w) values computed for each year: (odds(w)1948, odds(w)1949, .., odds(w)2004). Assuming a linear course of the time-series, three different scenarios can occur: (i) the occurrences of the female form are growing; (ii) the occurrences of the male form are growing; (iii) the ratio of the male and female form of an occupational title are stable over time. We combeen registered with the singular form. This is done for two reasons: first, occurrences of the plural form are outside the scope of this investigation, because in Italian the male plural form is traditionally used as the default, while the female variant of the plural is only used in exceptional cases, such as when referring to a group that is composed entirely of women. Second, this strategy filters out cases where the plural form shares the same lexical root as one of the gender variants. An example of this is the word infermiere (i.e. nurse), which can refer to both the singular masculine form (as in l’infermiere), or the plural feminine form (as in le infermiere).
Since the objective of this study is to observe the trends in the use of masculine and feminine forms for occupational titles, we are interested in analyzing how their frequency changes from one year to the other. However, measuring the absolute frequency in each year for both forms would be misleading, as it heavily depends on the amount of data that is available for each year in the corpus. Instead, we compute the smoothed relative frequency ptw for each word w and each year t using the following formula: (1) (2) ptw =
f wt + 1
Ct+ | V t |
where f wt is the frequency of word w in the year
t, Ct is the count of tokens occurring in the
corpus the year t and |V t| is the vocabulary length
computed on the year t. We compute ptw for both
gender forms of each occupational title. Then we
compute odds(w)t which represents the log ratio
of the smoothed relative frequency of the female
and male forms respectively:
puted the regression line of the time-series,
using the linear least-squares regression method
provided by the SciPy library3. We use the slope
of the regression line to determine whether the
values of odds(w)t are changing over time. If
the slope is positive/negative, odds(w)t is
increasing/decreasing over time, which means that the
frequency of wf is increasing/decreasing faster
than that of wm, or that the frequency of wm is
decreasing/increasing faster than that of wf . For
each regression line, we also compute the
statistical significance of the slope parameter relying on
the Wald Test
Figure 1 describes the value of the slope for each occupational title. Depending on the sign of the slope, we can identify two distinct groups of occupational titles. Green bars indicate that the slope of odds(w)t is positive, i.e. the frequency of the feminine form is increasing relative to that of the masculine form. On the other hand, red bars indicate that the slope is negative, thus the frequency of the feminine form is decreasing relative to that of the masculine form. Out of 35 occupational titles, 22 have a positive slope, while 11 result in a negative slope. In particular, the most positive slope is the one associated to marciat-ore/-rice (i.e. racewalker), while the most negative slope is fotomodell-o/-a (i.e. fashion model).
For many of these titles, the resulting slope can be mapped to specific social changes. An interesting example in this regard is infermiere (i.e. 3https://www.scipy.org/ nurse), to which a negative slope is recorded: indeed, in Italy the position of nurse has been opened to men starting from 19714. The odds(w) time series of infermiera/infermiere is reported in Figure 2.
Moreover, results show that managerial roles such as funzionaria (i.e. civil servant), ispettrice (i.e. inspector), direttrice (i.e. director) are associated to a positive slope, which is indicative of a stronger perception of women in such roles.
A similar push can be observed also in the scientific domain, with a positive trend for the words biologa (i.e. biologist), scienziata (i.e. scientist), as well as the artistic one. On the other hand, we observe an increase in the usage of the masculine form for segretario (i.e. secretary), ballerino (i.e. dancer), and stenografo (i.e. stenographer).
In the second part of the experiment, we attempt to identify the people that have driven the change in the usage of the feminine and masculine forms of an occupational title. To do this, we retrieve 4https://www.gazzettaufficiale.it/eli /id/1971/04/03/071U0124/sg (a) ballerino. (b) poetessa. the Named Entities (NEs) to which the occupational titles refer for each year, and monitor their frequency. In particular, we exploit the UDPipe annotations to extract valid NEs, i.e. entities that are directly connected to an occupational title via a dependency relation.
In Figure 3, we report the NEs extracted for two particular occupational titles: ballerino (i.e. male dancer) and poetessa (i.e. female poet). We have chosen these titles because they feature the largest number of occurrences of NEs in the corpus. The data is presented in the form of stacked line charts, which report the absolute frequency of each NE so that the height of a coloured line represents how many times a NE has been mentioned within a specified period. The dotted black line reports the overall smoothed relative frequency for the occupational title. Both the absolute frequency of NEs mentions and the overall smoothed relative frequency are aggregated in bins of 5 years.
Three male dancers are referenced over a wide period due to their historical role in the field: Rudolf Nureyev, Antonio Gades and Gene Kelly. However, the last years have seen a rise in popularity of new figures such as Raffaele Paganini, Joaquin Cortes, Andre´ de La Roche and Roberto Bolle.
Occurrences of specific female poets in the corpus keep low until the late ’70s. Ignoring a spike in 1953-1957, probably due to the quality issues in the data collected, the individual absolute frequency of NE mentions seems to agree with the overall smoothed relative frequency of the noun poetessa. In the 1988-2002 period, four figures overwhelm the scene: Joy Grisham, Elena Carasso, Maria Luisa Spaziani and Alda Merini. Even though the first work of Maria Luisa Spaziani dates back to 1954, we observe a significant rise in the occurrences in the early ’90s, when she is nominated three times for the Nobel Prize for Literature 5. The increase in NE mentions over time is even more apparent in this case, however, it follows a different trend compared to that of the overall frequency of the noun poetessa, which suggests that the word may have been used differently in the earliest period. 6
This paper investigates the usage of genderspecific forms of occupational titles in the Italian language in a diachronic corpus of 3 billion tokens extracted from two popular Italian newspapers. Through this analysis, we show that there are significant changes in the way newspaper articles refer to the masculine and feminine form of an occupational title and that they are consistent with socio-cultural events, such as changes in the employment policy. Moreover, we performed a more ifne-grained analysis by extracting the most influential figures that have guided this shift for two occupational titles (male dancers and female poets).
5https://en.wikipedia.org/wiki/Maria L uisa Spaziani As future work, we propose to continue work on this field by increasing the size of the corpus and by including sources other than news, such as social media, job applications, and legal documents. This can help reduce any form of linguistic bias that may have been introduced by journalists and increase the significance of the results. Moreover, we will extend the list of occupational titles, as well as group titles together based on category. Finally, we propose to improve the process used to extract named entities that are associated with occupational titles in text.
This research has been partially funded by ADISU Puglia under the post-graduate programme “Emotional city: a location-aware sentiment analysis platform for mining citizen opinions and monitoring the perception of quality of life”.