Unveiling Affective Polarization Trends in Parliamentary Proceedings

TL;DR

The paper tackles the growth of affective polarization in political discourse by moving beyond ideological differences to analyze the emotional style of language. It develops Hebrew Valence-Arousal-Dominance (VAD) resources and a two-stage modeling pipeline that predicts sentence-level VAD using a domain-adapted encoder, applied to the Knesset corpus. The authors demonstrate government–opposition emotional style differences and provide robust evidence of increasing polarization over time, validated through behavioral hypotheses and time-series analyses. They also release a suite of Hebrew VAD resources and demonstrate cross-language applicability of their approach for studying parliamentary discourse at scale.

Abstract

Recent years have seen an increase in polarized discourse worldwide, on various platforms. We propose a novel method for quantifying polarization, based on the emotional style of the discourse rather than on differences in ideological stands. Using measures of Valence, Arousal and Dominance, we detect signals of emotional discourse and use them to operationalize the concept of affective polarization. Applying this method to a recently released corpus of proceedings of the Knesset, the Israeli parliament (in Hebrew), we find that the emotional style of members of government differs from that of opposition members; and that the level of affective polarization, as reflected by this style, is significantly increasing with time.

Figures (7)

• Figure 1: VAD prediction combined model flow. The Encoder LLM receives a sentence as input and generates its sentence embedding. This embedding is then used as input for the regression models, which predict the Valence (V), Arousal (A) and Dominance (D) scores for the sentence.
• Figure 2: The Knesset-multi-e5-large model, the chosen LLM for extracting sentence embeddings (see Figure \ref{['fig:flow']}). It was created by fine-tuning the encoder part of the multilingual-e5-large model on the Knesset data, and then combining the tuned encoder with the original sentence-transformer.
• Figure 3: Average Valence, Arousal and Dominance values over all sentences in each of the committees.
• Figure 4: Average A-mean and V-var over all committees, over time (Knesset session). Dashed lines represent the automatically computed trend lines.
• Figure 5: Average Valence score over all sentences in each Knesset session.