Source-linked AI summary
Speaker or Language? Explaining Variance in Charismatic Prosody Across Luxembourgish and French
Nina Hosseini-Kivanani, Nafiseh Taghva, Peter Gilles, Oliver Niebuhr
TL;DR
The paper examines whether language or speaker identity better explains charisma-related prosody in bilingual public speaking. It compares matched spontaneous Luxembourgish and French speech from 10 politicians using 41 acoustic-prosodic features and mixed-effects models. Speaker identity dominates variance, while language produces specific spectral and voice-quality differences without a global charisma advantage.
Problem
It remains unclear whether language systematically changes charisma-related prosody in spontaneous bilingual speech and whether language effects exceed speaker-specific differences.
Method
The study analyzes 400 spontaneous utterances from 10 bilingual Luxembourg politicians in comparable contexts using 41 features and mixed-effects models.
Results
Speaker identity accounts for more than half of the variance across most features, while French shows higher shimmer and phrase-final F0 and Luxembourgish stronger mid-frequency spectral energy.
Takeaways & Limitations
Charisma-related prosody is primarily speaker-specific, with language switching associated with specific voice-quality and spectral shifts aligned with the languages’ sociolinguistic roles.
Takeaways & Limitations
The evidence is limited to 10 high-profile political speakers from one speech community and acoustic analyses without direct listener judgments.
Abstract
from arXiv · showhide
Charismatic speech is shaped by language and speaking style, yet their relative contribution in bilingual public speaking remains unclear. We analyzed spontaneous speeches of 10 politicians who address audiences in Luxembourgish and French, in highly comparable communicative contexts across languages. From 400 utterances, we extracted 41 acoustic-prosodic features linked to vocal charisma and fitted mixed-effects models to separate speaker- and language-related variance. Speaker identity accounted for most variance, whereas language explained less, but still showed systematic differences: French productions showed higher shimmer and phrase-final F0, indicative of a polite, respectful voice, while Luxembourgish productions exhibited stronger mid-frequency spectral energy, suggesting a more vocally present profile. These patterns align with the sociolinguistic roles of Luxembourgish as an informal identity language and French as a high-prestige institutional variety.
1. Introduction
The paper asks whether charisma-related prosody differs between French and Luxembourgish and whether language or speaker identity explains more variation. It addresses these questions using comparable spontaneous bilingual political speech from the same speakers.
- Background: Prior research links perceived charisma to multidimensional acoustic-prosodic patterns involving pitch, intensity, timing, and voice quality.These patterns include both positive and negative relationships with charisma across different acoustic measures.
- Research gap: Most previous studies examined monolingual or prepared speech, leaving spontaneous bilingual prosody in sociolinguistically asymmetric languages comparatively understudied.Bilingual prosody research nevertheless reports language-dependent differences in pitch, timing, voice quality, and spectral characteristics for the same speaker.
- Research questions: Charisma-related prosody may differ between French, a prestige language, and Luxembourgish, an identity language associated with affiliation and authenticity.The study frames competing expectations about whether prestige alignment or in-group identity produces stronger charisma-related cues.
- Research questions: A second question is whether language-related differences outweigh stable speaker-specific prosodic signatures within otherwise comparable bilingual contexts.The paper tests whether cross-language differences within speakers exceed between-speaker differences within a language.
- Study design: The study compares 400 spontaneous utterances from 10 Luxembourg politicians, with 20 utterances per language for each speaker, using matched public, political, or institutional contexts.The dataset is designed for within-speaker comparison while retaining ecologically valid spontaneous speech.
- Study aim: Using established acoustic correlates of vocal charisma, the analysis estimates the relative contributions of speaker and language to prosodic variability.The study also tests systematic prestige-related modulation across languages.
2. Methods
The study constructs a balanced bilingual corpus of spontaneous political speech and extracts standardized acoustic-prosodic measurements. It then uses PCA and feature-wise mixed-effects models to compare language effects with speaker-related variance.
- Speakers and material: The corpus contains 400 sentence tokens from 10 functionally bilingual Luxembourg politicians, evenly divided between 20 Luxembourgish and 20 French sentences per speaker.The material comes from spontaneous speeches, press briefings, interviews, or parliamentary debates.
- Feature extraction: The analysis covers 41 acoustic-prosodic features, including F0, intensity, duration, timing, voice quality, spectral measures, jitter, shimmer, harmonicity, and frequency-band energy.Features were extracted through a systematic ProsodyPro pipeline.
- Preprocessing: Features were checked for artifacts, log-transformed when strongly skewed, and standardized to zero mean and unit variance before modeling.Standardization makes coefficients comparable across features.
- Statistical analysis: PCA summarizes the 41-feature prosodic space, while separate linear mixed-effects models estimate language, gender, duration, and speaker contributions for each feature.Speaker is modeled as a random intercept; language effects are evaluated with marginal R2 and type III sums of squares.
3. Results
Speaker identity explains substantially more prosodic variance than language, although language produces systematic feature-specific differences. French has higher shimmer and phrase-final F0, whereas Luxembourgish has stronger spectral energy; the composite charisma index shows no overall language advantage.
- Prosodic space: The PCA shows partially overlapping French and Luxembourgish bands, with Luxembourgish shifted toward higher PC1; PC1 and PC2 explain 31.4% and 16.8% of variance.Language therefore produces a systematic but comparatively modest displacement in prosodic space.
- Variance partitioning: Speaker identity explains more than half of the variance for most features, with median ICCSpeaker = 0.56 and median Language variance of 0.5%.Speaker-related variance reaches roughly 70–80% for several stable F0 and energy cues.
- Language effects: French is reliably higher in shimmer and phrase-final F0, with Cohen’s d = 0.90 and d = 0.68, respectively.These effects indicate greater amplitude perturbation and higher phrase-final pitch in French productions.
- Language effects: Sixteen features are higher in Luxembourgish, including low-frequency and long-term spectral energy bands, with d = −1.67 at 2750 Hz and d = −1.62 at 2500 Hz.The pattern indicates stronger spectral energy and a brighter, more vocally present balance in Luxembourgish.
- Global charisma index: The six-cue composite charisma index shows no reliable language effect, with Cohen’s d = −0.03, p = 0.89, and Language R2 approximately 0.Individual speakers differ markedly, but their relative charisma ordering across languages is inconsistent.
4. Discussion and Outlook
Language produces systematic differences in charisma-related prosody, but speaker identity remains the dominant source of variation. French and Luxembourgish show distinct voice-quality and spectral patterns, although their interpretation is constrained by phonology and the study’s limited production-only sample.
- French has higher phrase-final F0 and shimmer, whereas Luxembourgish has stronger mid-frequency spectral energy and slightly earlier F0 peaks.
- The higher French phrase-final F0 may reflect language-specific intonational phonology rather than a straightforward stylistic choice.
- The spectral and voice-quality differences align Luxembourgish with a more projected profile and French with a more considerate, institutionally oriented setting.
- Speaker identity accounts for more than half of prosodic variance across most features, while language explains less.
- The conclusions are limited to ten high-profile politicians from one speech community and political speech, without direct listener judgments.
6. Generative AI Use Disclosure
Generative AI was used only to refine text and improve clarity. The authors retain responsibility for the paper’s scientific content.
- Generative AI was limited to text refinement and clarity improvements, while the authors remain responsible for all scientific content.