Languages in China link climate, voice quality, and tone in a causal chain

The study investigates whether linguistic sound systems adapt ecologically to climate via a causal chain: humidity affects voice quality, which in turn influences the number of lexical tones. Prior global correlations between aridity and reduced tonality lacked direct evidence from natural speech regarding two premises: that dry air compromises voice quality and that a desiccated larynx constrains tonal contrasts. Using China as a natural testbed due to its climatic diversity and rich variation in tonal systems, the authors aim to directly test two links—humidity → voice quality and voice quality → number of tones—and to re-evaluate the humidity–tone relationship with continuous measures. Establishing these links would strengthen the ecological adaptation hypothesis for speech and clarify mechanisms beyond confounds such as genealogy and contact.

Previous work reported associations between ecological factors and phonological patterns (e.g., ejectives and altitude/pressure; sonority and climate). Everett et al. (2015) found an association between desiccation and reduced lexical tone, arguing that tone production requires precise laryngeal control compromised by dry air. Commentaries broadly supported ecological adaptation but questioned the specific desiccation–tone link, citing lack of natural speech evidence and potential confounds (genealogy, areal contact). Studies with continuous tone measures found humidity–tone correlations that disappeared when controlling for genealogical distance (Hammarström, 2016; Roberts, 2018). Clinical and laboratory studies show hydration affects vocal fold viscoelasticity and phonatory effort, with effects on jitter/shimmer, but effects were measured under short-term provocation or special populations (e.g., athletes, singers) and not naturalistic speech of typical speakers. Tone complexity is multifaceted and not solely pitch-based; languages may use voice quality cues (breathiness, creakiness). Prior suggestions proposed using voice quality as an intermediate variable to test the causal chain humidity → voice quality → tone, but lack of standardized cross-linguistic speech data hindered direct tests.

• Data source: China’s Language Resources Protection Project (YuBao), providing standardized, high-quality recordings across 1718 locations; 997 language varieties used here, each with about 1200 lexical items. Total analyzed recordings: 1,174,686. Sampling rate 44.1 kHz, 16-bit.
• Recording standards: Controlled environments (studios/quiet rooms), background noise below −60 dB and ≤48 dB, speech volume between −18 and −6 dB; one native male speaker aged 55–65 per location; consistent equipment/protocols.
• Tonal data: Number of lexical tone oppositions per variety (0–14; mean 5); 48/997 (4.81%) non-tonal; sources: Language Atlas of China (2017a,b).
• Voice quality measures: Jitter and shimmer extracted from all files via Praat; averaged per location; expressed as percentages. Jitter/shimmer log-transformed for modeling.
• Ecological variable: Specific humidity (g/kg) from WheatA database (1982–2021); averaged across years and months per location. Range 2.24–16.24 (mean 9.158).
• Linguistic grouping: Six family groups (Altaic, Austroasiatic, Sinitic, Hmong-Mien, Tibeto-Burman, Kam-Tai) used as random effects.
• Statistical models:
  1. Linear mixed-effects models (lme4 in R) for log-jitter and log-shimmer with fixed effect of humidity, random intercepts by family, and by-family random slopes for humidity.
  2. Generalized linear mixed-effects model (Poisson) for number of tones with jitter and shimmer as fixed effects (no interaction), random intercepts by family, and by-family random slopes for jitter and shimmer.
  3. GLMM (Poisson) for number of tones with humidity as predictor; final model used polynomial terms of humidity with family random intercepts and random slopes for humidity. Parallel within-family analyses were also performed.

• Humidity → voice quality:
  • Lower humidity is associated with higher jitter and shimmer (poorer voice quality). Fixed effect of humidity significant:
    • Jitter: χ² = 160.68, df = 1, p < 0.0001
    • Shimmer: χ² = 42.58, df = 1, p < 0.0001
  • Pattern visible across China; families in humid regions (Kam-Tai, Hmong-Mien, Austroasiatic) tend to show lower jitter/shimmer; Altaic (drier regions) shows higher jitter/shimmer. Within-family significance strongest in Sino-Tibetan; non-significance in some families likely due to small sample sizes or limited humidity variance.
• Voice quality → number of tones:
  • Across all families, adding jitter/shimmer as fixed effects did not significantly improve model fit for tone counts (χ² = 2.1275, df = 5, p = 0.8312).
  • Within-family regressions: jitter negatively associated with tone count in Sino-Tibetan and Austroasiatic (significant); shimmer not a significant predictor.
  • Non-tonal languages (notably Altaic, plus some Austroasiatic and Tibeto-Burman) occur in low-humidity regions (mean 4.972 g/kg) and have higher mean jitter (2.478; above 78.94% of locations) and shimmer (12.10; above 83.85%).
• Humidity → number of tones:
  • Significant nonlinear positive relationship: locations with higher humidity tend to have more tones. Polynomial humidity terms significant in GLMM with family random effects:
    • I(Humidity²): χ² = 10.59, df = 1, p = 0.001
    • I(Humidity³): χ² = 8.06, df = 1, p = 0.005
  • Within-family significance observed in Sino-Tibetan.
• Descriptive context:
  • 997 varieties; tones range 0–14 (mean 5); humidity 2.24–16.24 g/kg (mean 9.158); 4.81% non-tonal. Tonally rich families (Hmong-Mien, Kam-Tai) tend to occur in humid southern China; non-tonal/low-tone-count varieties more common in drier northern/northwestern regions.
• Mechanistic interpretation:
  • Dry air reduces airway surface liquid and alters vocal fold vibration, increasing perturbation (jitter/shimmer) and reducing precision of laryngeal control; maintaining fine tonal contrasts becomes harder under increased phonatory effort, leading to fewer tone distinctions.

The findings empirically support a causal chain linking climate to phonological systems via physiology. First, ambient humidity significantly affects natural-speech voice quality (jitter/shimmer), demonstrating that desiccation effects are large enough to surface in everyday speech of typical speakers, not just in lab settings or special populations. Second, poorer voice quality aligns with fewer or absent tonal contrasts, with significant jitter–tone relationships within key families (Sino-Tibetan, Austroasiatic). Third, humidity shows a significant nonlinear positive relationship with tone counts, consistent with previous hints but now grounded in direct speech-derived measures. Potential confounds, including genealogical relatedness, are controlled through random effects; areal/language-contact histories are discussed and found insufficient to explain away the systematic humidity effect on phonatory capabilities. Alternative explanations via acoustic transmission (humidity affecting high vs low frequency attenuation) are considered less plausible for tone-range frequencies compared to the demonstrated production-side mechanism. The results underscore ecological adaptation in speech, where climate-mediated physiological constraints shape the synchronic distribution of tone and may trigger diachronic tone-system changes.

This study provides the first direct experimental evidence for the full causal chain humidity → voice quality → number of tones using a large, standardized cross-linguistic speech database from China. Dry climates degrade vocal fold vibratory efficiency (increased jitter/shimmer), making fine tonal distinctions harder to maintain, which corresponds to fewer tonal categories; conversely, humid environments do not hinder laryngeal control and allow maintenance or development of complex tone systems. Objective phonatory measures help disentangle climatic effects from historical and contact influences, verifying ecological adaptation of speech in the Chinese context and suggesting a mechanism for diachronic tonal change. Future directions include expanding sampling to more extremely arid regions and building global, high-quality standardized speech databases to explore geo-phonetic correlations worldwide.

• Geographic/climatic coverage: The dataset contains relatively few extremely dry regions; authors note this limits inference at the driest end and plan expansion to more arid areas.
• Within-family analyses: Humidity effects on voice quality were not significant in some families (Kam-Tai, Hmong-Mien, Austroasiatic, Altaic), likely due to small sample sizes and/or limited humidity variance.
• Speaker sampling: One speaker per location (male, 55–65) may limit within-variety variability capture and generalizability across genders/ages.
• Modeling scope: Global jitter/shimmer effects on tone were not significant across all families, with effects clearer within specific families, indicating heterogeneity and potential ceiling effects in highly tonal families.