Publications

We develop legislative games of empowerment to demonstrate that a supranational alliance between the European Commission and the European Parliament is at the core of the conferral of policy authority to the EU executive. Leveraging a machine-annotated corpus of EU directives and regulations from 1958 to 2019, we show that parliamentary involvement and ideological proximity are key drivers of this process, with important implications for EU legitimacy.

We introduce a computational method for identifying delegating and constraining provisions in European Union (EU) laws. Leveraging the syntactic structures employed by legislators, we developed a set of extraction rules applied through a custom-built computational linguistics pipeline. We run through the pipeline more than 600,000 legal sentences that we extracted from 9319 laws adopted between 1958 and 2019. The application performs very well vis-à-vis human annotation and outperforms transformer models. The produced patterns of authority delegation and constraint resonate with our knowledge of the policymaking and history of the EU. Our approach provides valuable insights for designing transparent and adaptable rule-based computational linguistic methods of legal text analysis. We also release the comprehensively annotated dataset and the fine-tuned transformer models developed for this task.

We present a project investigating the impact of social media on ideological structuring of political attitudes in a generational perspective, with Italy (around the 2024 European Parliament elections) as a case study. We introduce an innovative mixed-methods, sequential, quantitative-driven, multi-stage design combining surveys, qualitative interviews, and social media data, integrating qualitative and quantitative components to analyze the effect of social media influencers on political attitudes across generations. We present the project design along with interaction and integration among components (including a "Swipe" module for respondent-influencer linkage and the use of AI for classifying social media posts), individual components with first empirical results, an intermediate convergent assessment stage, and directions for data analysis.

Gender stereotypes result in distinct socialization processes for men and women, leading to differences—among many others—in their communication styles. Women are often socialized to emphasize emotional aspects in their speech, while men are taught to focus on rationality. Empirical evidence from non-political contexts supports this distinction, often described as rapport versus report talk. However, it remains unclear how these dynamics manifest in parliamentary settings. We argue that women in parliament face conflicting incentives: while gender stereotypes might lead them to adopt an emotionally expressive style, the parliamentary context favours rational, fact-based debate and is dominated by men who are expected to adhere to this style. We investigate women’s tone choices in parliamentary speech through a longitudinal study covering seven decades of parliamentary debates in Italy: a typical case of modern parliamentary systems, where women’s enhanced access to the floor still takes place in a gendered environment. Our findings underscore the ongoing relevance of gendered expectations in shaping parliamentary speech, challenging assumptions about the neutralizing effects of experience and institutional socialization in legislative contexts.

Scholars employ two main measures of the executive constraints embedded in European Union laws: one is based on the variation in the use of different types of restrictions, and the second is based on the frequency of such use. They reflect two alternative conceptualizations of bureaucratic control. We label them, respectively, as the "toolbox perspective" and the "design perspective". We illustrate that the constraint frequency measure poses fewer validity problems in estimating legislators’ intent to constrain implementation and tends to produce less severe measurement errors. We then evaluate the performance in estimating constraint variation of a recent computational application and identify potential drawbacks of automated learning from hand-coded provisions. We lastly introduce a skeletal framework for a machine-learning approach based on the syntactic structures employed by legislators that could improve the performance of this innovative technique.

This paper investigates how parties spread news from traditional media on their social network accounts to engage followers. In particular, we address why parties would choose to include in the ‘media digest’ they carefully craft for their followers also content from ‘hostile’ media outlets and journalists. We analyze 4,586 Facebook posts published by the main Italian parties during the 2019 European election campaign, assessing the impact on readers’ engagement of counter-attitudinal and pro-attitudinal content from either hostile or friendly media. Our results show that parties skillfully exploit friendly and hostile sources alike to alternatively engage or outrage their followers.

Recent political upheavals, such as Brexit, Trump’s election, the rise of challenger parties in Europe, and Italy’s first "populist" government in 2018, challenge traditional theories of voting behaviour and party competition. Conventional explanations often point to populism and voter irrationality. However, recent research highlights that challenger parties leverage issue opportunities that transcend traditional ideological boundaries, suggesting a de-ideologized context where voters are drawn to post-ideological platforms. Despite these insights, a possible de-ideologization process remains insufficiently understood. The POSTGEN project seeks to fill this gap by analysing de-ideologization mechanisms and dynamics, particularly in Italy—a key case of contemporary political transformation. It adopts a generation-focused approach, emphasizing the role of younger generations in the evolving political landscape. The project examines how political attitudes develop and impact individuals, generations, and society over time. It investigates whether political issues integrate into coherent ideological structures or remain fragmented, and how non-political influencers affect these dynamics amid the crisis of traditional epistemic authorities. The project employs a mixed-method, multi-method, longitudinal strategy that features multiple components: from mass surveys, to social media content (also analysed through automated methods), to dedicated surveys aimed at secondary-school students, to qualitative in-depth interviews to young adults. By integrating these methods within a robust theoretical framework, POSTGEN aims to provide insights into the evolution of the Italian political systems, also to help understand broader changes in democratic representation in Western political systems.

An ongoing study investigating how populist radical right parties achieve distinct visual consistency in their social media campaigns. The paper combines computer vision, multimodal captioning, and topic modeling to identify the mechanisms—such as systematic bundling of visual themes—that underpin this consistency.

Parliamentary proceedings represent a rich yet challenging resource for computational analysis, particularly when preserved only as scanned historical documents. We propose a pipeline based on Vision-Language Models for the automatic transcription, semantic segmentation, and entity linking of Italian parliamentary speeches: a specialised OCR model extracts text while preserving reading order, and a large-scale Vision-Language Model performs transcription refinement, element classification, and speaker identification by jointly reasoning over visual layout and textual content. Extracted speakers are linked to the Chamber of Deputies knowledge base through SPARQL queries and a multi-strategy fuzzy matching procedure. Evaluation against an established benchmark demonstrates substantial improvements in both transcription quality and speaker tagging.

This paper advances a novel framework for detecting the political capture of bureaucracies under conditions of democratic backsliding by applying machine learning to Italian legislation.

We investigate how autocratizing leaders manipulate democratic language to sustain regime legitimacy, and how this rhetorical strategy is conditioned by international exposure and domestic institutional quality. The paper leverages word embeddings and embedding regression to trace semantic drift in political speeches.