Machine Learning Classifiers Reveal How European Politicians Justify Their Immigration Stances
Authors: Finn Connolly, Clara Eggenhuizen, Mathias Franchino, Clodagh Lynch
Date: 5 May 2025
Political justifications reveal the hidden architecture of parliamentary debates. Our research uncovers a striking pattern in immigration discourse by using cutting-edge machine learning models to classify both the positions politicians take and the justification strategies they employ in immigration debates. Politicians strategically select justification types based on their stance, with pro-immigration speakers leveraging value-based arguments and anti-immigration speakers focusing on the negative consequences of immigration. These findings offer unprecedented insights into how European political discourse is shaped and how accountability in democratic institutions can be enhanced through computational analysis.
When debating immigration in the European Parliament, politicians who support immigration are 3.7 times more likely to justify their stance by appealing to liberal democratic values than those who oppose it. This difference reveals how politicians strategically choose their justification based on their stance. Our computational analysis of over 3,000 parliamentary speeches shows a clear pattern: when Members of the European Parliament (MEPs) support immigration, they typically appeal to liberal democratic values like human rights and equality; when they oppose it, they emphasize potential threats to security and economic stability. These choices are not random, as they represent linguistic strategies that shape public perception and policy outcomes.
In an ever-changing political landscape, where policies are shifting in unpredictable directions, holding politicians accountable is increasingly important. To examine how MEPs justify their positions in immigration debates, we aim to contribute to the transparency and accountability of the MEPs, as well as provide an innovative way to study justifications in political science.
In our research, we classify both MEPs’ stances and the justifications they use to support them in speeches made in the European Parliament between 2000 and 2019. This period includes the peak of the migration crisis, which offers an ideal environment to study a policy domain as polarized as immigration. Specifically, we aim to examine the differences between pro-immigration speakers and anti-immigration speakers when it comes to justification, as well as the effect of Euroscepticism.
To classify stances on immigration, we use a pre-existing language model (Political DEBATE) to determine whether a speech is pro-immigration, anti-immigration or neutral. We also use a separate language model, capable of understanding the context and language patterns in political speeches, which we trained to be able to identify four different justification types by providing it with over 1,000 hand-annotated sentences. We distinguished between the following types of justifications:
- Value-based:
- Based on liberal-democratic values (ex. “Asylum is a universal human right, so refugees should be welcomed to Europe.”)
- Based on national values (ex. “Migrants pose a risk to our national security, as we might let in terrorists by mistake.”)
- Consequence-based:
- Based on positive consequences (ex. “Migration increases our cultural diversity.”)
- Based on negative consequences (ex. “We cannot let migrants form a financial burden on the state.”)
Pairing these two models allows us to determine how the stance on immigration relates to the justification used to support that stance.
Figure 1: Effect of Pro-Immigration Stance on Choice of Justification
Caption: Compared to anti-immigration speeches, pro-immigration speeches are much more likely to contain liberal-democratic values and much less likely to contain negative consequences.
We find that, indeed, pro-immigration speeches are more likely to use liberal-democratic values as justifications, and anti-immigration speeches are more likely to rely on the negative consequences of immigration. The differences for national values-based justifications and those based on positive consequences are not as stark. Additionally, speakers from pro-EU political parties use value-based justifications more overall, and female speakers use any kind of justification more often than male speakers. It is unclear why; perhaps women are more likely to support their positions than men are. During the migration crisis years (2014-2017), justifications of any kind were used less, possibly because speakers may have felt a sense of urgency and did not provide support for their stance because of it, or because they felt the Parliament was already united.
Figure 2: How Euroscepticism Affects Stance and Justification
Caption: While anti-immigration speeches are more likely to contain negative consequences, this difference is much smaller for speakers who are more in favour of European integration.
We also looked into the effect of Euroscepticism, the political position of being critical of European integration, on the relationship between stance and justification, as anti-EU speakers might not have shown a difference in the use of value- or consequence-based justification. This was shown to be not the case, as anti-EU speakers still tend to use either type of justification based on their stance. However, Euroscepticism still has an effect. As mentioned, pro-immigration speakers are less likely than anti-immigration speakers to use negative consequences as justifications. Interestingly, this difference is smaller for speakers that are more pro-EU. This suggests that pro-EU politicians adopt a more balanced rhetorical approach regardless of their stance on immigration. This finding reveals how political discourse is shaped in ways that go beyond simple ideological divides.
Overall, our research demonstrates that machine learning can effectively detect and measure justification patterns in political discourse, offering an innovative method to analyze political communication. Beyond just identifying stances, these models reveal the underlying reasoning structures that politicians use to persuade their audiences, and thereby contribute to political transparency for researchers and citizens alike.
While we focused on immigration debates in the European Parliament, the methodology we have developed can be applied to any political discourse with sufficient data, from national parliaments to supranational organizations, and across policy domains. Our methods give future researchers ample opportunity to track political justification over time, across political systems and along the ideological spectrum, especially as computational tools continue to advance.
About the authors
Finn Connolly, Clara Eggenhuizen, Mathias Franchino and Clodagh Lynch are students in the MSc Politics and Data Science programme at University College Dublin. This blog post is part of a collaborative research project, supervised by Dr. James Cross, for the module “Connected_Politics”.