Μεταπτυχιακές Εργασίες
Μόνιμο URI για αυτήν τη συλλογήhttps://pyxida.aueb.gr/handle/123456789/7
Περιήγηση
Πλοήγηση Μεταπτυχιακές Εργασίες ανά Συγγραφέα "Andrianopoulos-Kalogeropoulos, Foivos"
A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω
Τώρα δείχνει 1 - 1 από 1
- Αποτελέσματα ανά σελίδα
- Επιλογές ταξινόμησης
Τεκμήριο Detecting gendered language patterns in Greek political discourse using NLP(2026-02-13) Andrianopoulos-Kalogeropoulos, Foivos; Ανδριανόπουλος-Καλογερόπουλος, Φοίβος; Konstantinidou, Maria; Androutsopoulos, Ion; Pavlopoulos, John; Korre, KaterinaGendered discourse refers to the way language reflects, reinforces and sometimes even challenges the already created stereotypes associated with masculinity and femininity. In other words, it is not just about how people use language differently, but about how language constructs the roles of “femininity” and “masculinity” and urges people to perform an identity. Discourse used in conversations, writings and even institutional speeches can depict how society “sees” gender and in particular gender roles. Even in official institutions, like the Greek Parliament, language used by politicians reflects how gender is perceived, and how one should use language to present a more assertive or friendly image. At the same time, this idea of “performative” discourse influences minority groups, such as the LGBTQ+ community, that can become the target of “Pinkwashing” strategies, namely a form of performative activism where companies or political parties promote themselves as “pro-LGBTQ+” to legitimize negative behaviours. This study is primarily concerned with identifying gendered speech in the proceedings of the Greek Parliament and how it can be detected both by humans and machines. In particular, the study examines whether certain linguistic traits, such as the word choice, hedging devices and the use of evaluative adjectives, can highlight if the speech contains or not gendered language. To achieve that, a group of two expert annotators was formed that was asked to label a dataset of 200 examples of speeches of the proceedings of the Greek parliament based on the existence or not of gendered discourse. Simultaneously, a Large Language Model, Llama3-70b-Instruct, was requested to annotate the same speeches to evaluate its performance compared to human annotations. In addition, lexicons that reflected word choices of males and females, in the context of political discourse were created, by prompting Llama3-70b-Instruct based on sociolinguistic theory, to capture the lexicons’ term frequency in the corpus and the semantic relations between female and male-associated terms. To achieve the creation of a common semantic space Greek-Bert was selected. Another type of lexicon containing phrases that could possibly indicate “Pinkwashing” strategies was also created and the prominence of these phrases per political party was computed, using TF-IDF, aiming at revealing how “Pinkwashing” is reflected through language and what political parties tend to take advantage of this strategy more often. The results showed that humans could detect gendered speech patterns, whereas the language model had a hard time recognizing gendered discourse, meanwhile the semantic representation of the lexicon emphasizes the performance of gender roles. As far as “Pinkwashing” is concerned, contextualization of the suggested lexicon pointed out that the terms were used in the broader theme of supporting human rights.