A natural language measure of ideology in the Brazilian Senate

Abstract

We estimate a measure of political ideology using as data a corpus of over two decades of speeches delivered by Brazilian Federal Senators across five legislatures. We employ a computational technique that analyses political speech by extracting the dictionaries that best translate the content of each ideology. Through this supervised learning method, we calculate the classification accuracy over these political texts and show that polarization is increasing across the legislatures. The method also reveals the evolving patterns of political ideologies over a period of deep change in Brazilian society. We further investigate the political dynamic across legislatures by comparing our results with current approaches to ideology in the political science literature.