Umm... With Transformers? Insights from Filled Pause Use across Four Slavic Parliaments

2026-07-07 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A study analyzed approximately 4,000 hours of parliamentary speech across Croatian, Czech, Polish, and Serbian to investigate filled pause (FP) usage. Utilizing transformer-based automatic detection for FP occurrence and Generalised Estimating Equations (GEE) with Mundlak correction for FP rate modeling, the research aimed to overcome limitations of small, single-language corpora. Findings replicated a negative association between age and speech rate with FP rate. However, gender effects were found to be language-specific and contrary to most prior literature. Novel analyses revealed a consistent positive association between sentiment and FP rate. Additionally, political orientation and power status modulated FP rates in a parliament-specific manner, with opposition speakers generally exhibiting lower FP rates than those in governing coalitions.

Key takeaway

For research scientists analyzing spontaneous speech, you should consider the significant linguistic and contextual factors influencing filled pause rates. Your models for speech analysis or generation must account for language-specific gender effects and the positive correlation between sentiment and FP rate. Furthermore, when interpreting speech patterns, recognize that political orientation and power status can modulate FP usage, requiring nuanced, context-aware approaches beyond simple demographic variables.

Key insights

Transformer-based analysis of Slavic parliamentary speech reveals nuanced, context-dependent patterns in filled pause usage beyond prior single-language studies.

Principles

• Age and speech rate negatively correlate with FP rate.
• Gender effects on FPs are language-specific.
• Sentiment positively associates with FP rate.

Method

Transformer-based automatic detection identifies filled pause occurrences. Generalised Estimating Equations (GEE) with Mundlak correction models filled pause rates, distinguishing within- from between-speaker effects.

In practice

• Apply transformer models for speech analysis.
• Use GEE for complex linguistic modeling.
• Analyze FPs to infer speaker sentiment.

Topics

• Computational Linguistics
• Filled Pauses
• Transformer Models
• Parliamentary Speech
• Generalised Estimating Equations
• Sentiment Analysis

Best for: AI Scientist, Research Scientist

Related on AIssential

• Mapping the Political Discourse in the Brazilian Chamber of Deputies: A Multi-Faceted Computational Approach — Computational methods can reveal hidden patterns in parliamentary discourse beyond voting records.
• A Sociolinguistic Analysis of Automatic Speech Recognition Bias in Newcastle English — ASR errors are not random but socially patterned, driven by dialectal variation and demographic factors.
• The Dynamics of Human and AI-Generated Language: How Semantics Fluctuates across Different Timescales — The ACW-0 semantic timescale analysis effectively distinguishes temporal organization of generic versus specific content in human and AI speech.
• Does Topic Sentiment Cause Perceived Ideology? Comparing Human and LLM Annotations in Political News Articles — Fine-tuning LLMs for ideology prediction can create spurious sentiment-ideology causal links not present in human judgment.
• Gendered Stylistic Variation in Brazilian Portuguese Google Play Reviews: A Large-Scale Study — Gender-associated stylistic differences exist in Brazilian Portuguese Google Play reviews, particularly in paralinguistic expressivity.
• A corpus-assisted metaphor analysis in portraying businesswomen: Diachronic changes in Hong Kong English news — Corpus-assisted metaphor analysis reveals diachronic shifts in media portrayals of businesswomen.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.