Tuesday 08 April 2025
The quest for a deeper understanding of political texts has long been a challenge for researchers and linguists alike. With the advent of large language models, the task of analyzing party manifestos and parliamentary speeches has become more accessible than ever. A recent study published in a leading scientific journal takes this endeavor to the next level by developing a novel approach to segmenting and classifying statements from these texts.
The authors’ goal was to create a system that could not only identify individual statements within political texts but also assign them meaningful labels, such as ‘freedom and human rights’ or ‘traditional morality.’ To achieve this, they leveraged the power of large language models, specifically the XLM-RoBERTa model, to encode input sequences and then employed a conditional random field (CRF) to predict statement boundaries and labels.
The team trained their model on a dataset comprising over 41,000 statements from the UK House of Commons, covering a period spanning four decades. This comprehensive corpus allowed them to analyze the political stances and ideologies of major parties, including the Conservative Party, Labour Party, Liberal Democrats, and Scottish National Party.
One of the key insights gained from this analysis was the identification of distinct trajectories for each party over time. For example, the Conservative Party consistently scored high on right-wing labels, while the Labour Party trended towards more left-leaning stances. The study also revealed that the Scottish National Party’s political stance shifted in response to events such as the 2014 referendum on Scottish independence.
Furthermore, the researchers applied their approach to an Australian dataset, analyzing speeches from four major parties between 1998 and 2005. This analysis yielded similar results, with the Liberal Party and National Party exhibiting right-wing tendencies, while the Australian Democrats and Labor Party showed more left-leaning biases.
The implications of this research are far-reaching, as it provides a valuable tool for understanding the political ideologies and stances of major parties over time. By analyzing party manifestos and parliamentary speeches, researchers can gain deeper insights into the complex dynamics of politics, potentially informing policy decisions and public opinion.
In addition to its theoretical significance, this study also demonstrates the potential applications of large language models in political analysis. As these models continue to evolve and improve, it is likely that we will see even more sophisticated approaches to analyzing political texts, further illuminating the complexities of human politics.
Cite this article: “Unpacking Political Sentiment: A Deep Dive into Language Models and Manifesto Data”, The Science Archive, 2025.
Political Texts, Large Language Models, Party Manifestos, Parliamentary Speeches, Crf, Xlm-Roberta, Uk House Of Commons, Political Stances, Ideologies, Australian Dataset.
