Placing machine learning into the hermeneutic circle: a combined computational-interpretive method for text analysis

Scholars are increasingly turning to machine learning text analysis (MLTA) to make sense of world politics, but the question of how computational power and interpretive expertise should work together remains underexplored. This gap stems from a lack of engagement between those who treat text as data to be computed and those who approach it as language to be interpreted. In this article, we bridge this divide by proposing a methodology that cycles between computational analysis and interpretive moments, placing machine learning within the hermeneutic circle. We argue that by iterating between these dual tasks, researchers can harness the strengths of both approaches, reducing the dimensionality of text while preserving its pragmatic structure of meaning. To illustrate our approach, we apply it to the UN General Debate Corpus (UNGDC), demonstrating how machine learning can identify coherent rhetorical intervals that are then interpreted using expert knowledge. Our primary objective is pedagogical, but our application also highlights the potential empirical payoffs of combining MLTA and interpretive analysis in the era of big data.