Source-linked AI summary
Speaker-Specific and Language-Dependent Temporal Organization in Bilingual Political Speech
Nina Hosseini-Kivanani, Nafiseh Taghva, Peter Gilles, Oliver Niebuhr
TL;DR
The study asks how bilingual politicians organize rhythm across Luxembourgish and French, addressing limited evidence from naturalistic public speech. It analyzes 400 sentences from ten politicians with duration-based rhythm metrics, variance decomposition, and within-speaker tests. Consonant timing retains more speaker-specific structure, while French vowel timing and vocalic intervals expand and vary more, with no robust language-by-gender interactions.
Problem
How bilingual politicians organize rhythm across Luxembourgish and French in public speech, and how timing reflects speaker, language, and gender, remains understudied.
Method
The study analyzes 400 spontaneous sentences from ten bilingual politicians using segment and pause annotations, duration-based rhythm metrics, variance decomposition, and paired within-speaker tests.
Results
Consonant-based metrics retain speaker-specific signatures, whereas vowel-based metrics are largely language-driven: French has longer, more variable vowels and vocalic intervals, and no robust language-by-gender interactions emerge.
Takeaways & Limitations
Language choice systematically reorganizes rhythmic timing in bilingual political speech while the balance between speaker-specific and language-dependent structure varies by metric family.
Takeaways & Limitations
The modest sample of elite politicians and speaker-by-language aggregation limits generalization and reduces sentence-level information for variance and interaction analyses.
Abstract
from arXiv · showhide
Speech rhythm helps structure persuasive speech, but most empirical work examines monolingual English. This study asks how politicians organize timing when speaking Luxembourgish and French. We analyze 400 sentences from ten politicians, annotated for segments and pauses. We compute rhythm metrics, including means, variability, and pairwise variability indices for consonants and vowels. We quantify speaker and language contributions and test within-speaker language effects with paired t-tests. Results show that consonant-based metrics retain speaker-specific signatures, whereas vowel-based metrics are largely driven by language choice. French tokens display longer and more variable vowels and vocalic intervals, while consonant timing differences are smaller. No robust language by gender interactions emerge. These findings show that language choice systematically reorganizes rhythmic timing in bilingual public speech.
1. Introduction
The study addresses the understudied organization of rhythm across languages in public bilingual speech by examining how Luxembourgish and French shape timing, speaker signatures, and gender strategies.
- Research gap: Public bilingual speech remains understudied, especially regarding whether rhythm reflects speaker identity, language choice, or gender.Luxembourgish and French provide a socially meaningful contrast because they occupy distinct roles in Luxembourg’s public life.
- Study focus: The study analyzes spontaneous political speech from ten Luxembourgish politicians who regularly speak both languages in comparable public contexts.It uses 20 spontaneous sentences per language for each speaker and derives rhythm metrics from consonant-vowel and pause annotations.
- Research questions: RQ1 tests how much rhythmic variance is attributable to stable speaker differences versus language choice using speaker ICC and language R2.This separates speaker-specific signatures from language-dependent timing patterns.
- Research questions: RQ2 tests whether speakers systematically change vocalic, consonantal, pause, and speaking-rate patterns when switching between Luxembourgish and French.The analysis targets duration distributions, variability measures, pairwise variability indices, pause counts, and speaking rate.
- Research questions: RQ3 tests whether language-specific rhythm strategies differ between female and male politicians through Language × Gender interactions and effect sizes.The study frames gender as a possible moderator of bilingual rhythm organization rather than assuming gendered strategies.
- Scope: The study describes rhythmic timing across two languages without inferring perceived charisma directly.Its scope is bilingual production and temporal organization, not listener evaluation of charisma.
2. Methods
The study uses a balanced within-speaker corpus of spontaneous Luxembourgish and French political speech, manually annotated for segmental structure and pauses, then analyzes rhythm with paired comparisons and variance decomposition.
- Corpus: The corpus contains 400 spontaneous sentence tokens from ten bilingual politicians, with 20 Luxembourgish and 20 French sentences per speaker.The speakers are evenly split by gender and the material comes from parliamentary recordings, press conferences, and interviews.
- Annotation: Researchers manually created consonant-vowel tiers and labeled pauses of at least 200 ms after correcting language-specific automatic alignments.Identical consonant-vowel criteria were applied across languages using waveform and spectrogram evidence.
- Measures: The analysis computes means, variability, raw and normalized pairwise variability indices, syllable and intensity-peak durations, and speaking rates.Metrics are derived separately for vocalic, consonantal, voiced, and silent-pause intervals.
- Aggregation: Metrics were aggregated to one value per speaker per language to create a balanced within-speaker design.This aggregation supports paired language comparisons while reducing sentence-level variability and degrees of freedom for some analyses.
- Statistical analysis: Speaker-specific variance was estimated with ICC, language-driven variance with language R2, and within-speaker language effects with paired t-tests and paired Cohen’s d.These procedures directly correspond to the study’s speaker-versus-language and language-switching questions.
- Statistical analysis: Language, Gender, and their interaction were modeled with two-way ANOVA, with Benjamini–Hochberg correction across 23 rhythm metrics.The reported interaction p-values assess whether language contrasts differ by gender while controlling the expected false-discovery proportion.
3. Results
Language effects are concentrated in vowels and speaking rate, whereas consonant measures preserve more speaker-specific structure; language-by-gender interactions are generally absent.
- Language effects: French vowels and vocalic intervals are longer and more variable than Luxembourgish vowels, with large within-speaker effects and FDR-corrected p < .01.Mean V and ∆V have Cohen’s d = −1.71 and −1.41, while rPVI.V has d = −1.57; vocalic interval effects are all |d| ≥1.44.
- Language effects: Luxembourgish is spoken faster than French for both syllable rate (d = +1.12, p = .020) and segment rate (d = +2.10, p = .002).The rate contrast is reported for comparable political audiences and is stronger for segment rate.
- Speaker versus language: Language explains substantial variance in vowel metrics and speaking rate, while consonant metrics show higher speaker ICC and weaker language effects.Language accounts for R2 values of 0.38–0.43 in key vowel measures and about 0.50 for Rate Seg; consonant measures retain stronger speaker-linked variability.
- Speaker fingerprints: Within-speaker radar profiles remain similar across languages, but French shows outward shifts on vowel axes while consonant axes largely overlap.The magnitude of vowel shifts varies across politicians, supporting stable rhythmic fingerprints alongside language-dependent adjustment.
- Gender: Language-by-gender interactions are almost uniformly non-significant, indicating comparable Luxembourgish–French rhythm contrasts for women and men.The same pattern includes speaking rate, although languages differ substantially in segment durations and global tempo.
4. Discussion and conclusions
Bilingual politicians reorganize rhythm across Luxembourgish and French, with language effects dominating vowel timing and speaking rate while consonant measures retain stronger speaker signatures. No robust language-by-gender interactions emerged, and the findings are limited to a specific sample and production setting.
- Speaker and language contributions: The balance between speaker identity and language choice depends on the rhythm-metric family, partly supporting the speaker-signature research question.Consonant measures such as Mean C, ∆C, VarcoC, and rPVI.C show moderate language effects alongside higher ICC and smaller language-explained variance.
- Language-dependent timing: French vowels and vocalic intervals are longer and more variable than in Luxembourgish, while Luxembourgish speech is faster overall.Vowel and vocalic-interval effects are large and FDR-corrected; syllable-rate and segment-rate differences favor Luxembourgish.
- Speaker and language contributions: Vocalic metrics are primarily language-driven, whereas consonant-related metrics show stronger speaker-specific clustering.Key vocalic metrics have R²_Lang values of 0.38–0.43 with near-zero ICC, while consonantal metrics show the opposite profile.
- Gender: No Language × Gender interactions survived FDR correction, indicating comparable Luxembourgish–French rhythm contrasts for women and men in this sample.The reported non-significant interactions include speaking rate and nearly all other metrics.
- Limitations: The conclusions are bounded by a modest sample of elite politicians, speaker-by-language aggregation, and the absence of perceptual or charisma testing.Aggregation enables paired comparisons but collapses sentence-level variability and reduces degrees of freedom for variance and interaction analyses.
- Discussion and conclusions: Within naturalistic, high-stakes bilingual political speech, language choice itself systematically shapes temporal organization.The study extends prior work on within-language stylistic rhythm by identifying language choice as an additional source of rhythmic organization.
6. Generative AI Use Disclosure
Generative AI use was limited to text refinement and clarity improvements. The authors retain responsibility for the paper’s scientific content.
- Generative AI use was limited strictly to text refinement and clarity improvements.
- The authors remain solely responsible for the methodology, experiments, analysis, and conclusions.
- The disclosure does not attribute scientific content to the generative AI tool.