Effects of Ambient Air Conditions on Vocal Fold Physiology

The biomechanical properties of the vocal folds are influenced directly by hydration levels (12). For instance, dehydration of the vocal folds results in decreased amplitude of vocal fold vibration (13). Fundamental frequency is impacted by hydration or dehydration according to a variety of metrics obtained from in vivo and ex vivo laboratory studies. Increased hydration is associated with heightened vocal fold viscosity and facility of phonation (14). In contrast, dehydration of the vocal folds is associated with increased phonation threshold pressure (PTP) and increased perceived phonation effort (PPE) (14), although the extent of PTP increase remains a matter of inquiry (15). Laryngeal dehydration is also associated with alterations to the ionic composition of the vocal fold surface liquid, more specifically an increase in Na⁺ concentration and associated modulation to vocal fold elasticity (16, 17).

Most relevant for our purposes, a number of studies have demonstrated that the mere inhalation of dry air (rather than the artificial or ex vivo dehydration of the larynx) impacts vocal fold physiology and results in clear effects on phonation. Crucially, in the context of the present discussion, these in vivo studies demonstrate that even brief exposure to desiccated air can negatively impact precise phonation via increased rates of jitter (imprecise pitch) and of shimmer (varying amplitude) (18). The former finding is particularly pertinent in the context of a discussion of complex tone, which necessitates relatively precise phonation, and therefore reduced rates of jitter, given that three or more pitch variances are used in a phonemically contrastive manner in languages with complex tone (19).

As noted in ref. 18, “even after a short provocation with dry air, a significant increase in perturbation measures” is evident in phonation. This increase is due at least in part to the decreased viscosity of the laryngeal mucus and associated alterations to the cohesive forces at work during the contact portion of vocal fold oscillations. Inhalation of desiccated air resulted in subjective impressions of impaired voicing abilities among healthy subjects, impressions that were consistent with objective findings on jitter, shimmer, and noise-to-harmonic ratios (18). In the case of some subjects, jitter measurements increased over 50% after merely 10 min of inhalation of very dry air. Similar studies have replicated these findings with different methodologies, demonstrating that oral breathing (which, when contrasted to baseline nasal breathing, results in less treated/humidified air reaching the larynx) is directly associated with increases in PTP and PPE (20). Hydration and increased ambient humidity are positively associated with reduced PTP and PPE and, crucially, with decreases in jitter and shimmer (21⇓⇓–24). The effects of aridity on the vocal tract have long-term effects, as evidenced by the fact that otorhinological practice patients often exhibit throat ailments associated with dry working conditions. Low humidity is reported to be the cause of sore throat in 30% of cases (25). Also, it is well known that singers maintain tones less effectively in dry conditions (18). Clearly, the human larynx is susceptible to environmental aridity. Although nasal breathing mitigates such aridity, it does not do so completely, and furthermore, oral breathing is required at higher rates of respiration.

Frigid air is always dry, regardless of saturation level (i.e., relative humidity), due to its reduced water vapor capacity (26). Air at –10 °C reaches water saturation at 2.3 g/m³, whereas air at 30 °C does so at 30.4 g/m³. In other words, a difference of 40 degrees translates into a 13-fold difference in water vapor capacity. Independent of its desiccation, cold air inhaled via the nasal cavity yields greater bronchoconstriction and increased nasal fluid (27), two factors that promote oral breathing. Oral breathing, even of nondry air, promotes the desiccation of the vocal tract and the evaporation of the airway surface liquid of the vocal cords, and the latter has been shown to yield a variety of difficulties vis-à-vis precise phonation (14). The effects of breathing cold air on the vocal tract and the larynx have been supported by numerous studies of those active outdoors in frigid contexts (28, 29). These effects are not easily compensated for, and factors such as laryngitis and xerostomia (dry mouth) are recurrent among human populations in very cold regions (28). In short, under both laboratory and naturalistic conditions, the inhalation of very cold and arid air yields short-term and long-term desiccating effects on the vocal tract and the vocal cords. And desiccation of the vocal cords has been clearly tied to imprecision of phonation (14).

In addition to these findings, a recent study has found that muscle tension dysphonia (MTD), a laryngeal ailment that can be provoked or exacerbated by dry ambient air, has demonstrable negative effects on the production of phonemic tone. Various metrics offered in ref. 30 demonstrated that MTD-diagnosed Vietnamese speakers were more likely than speakers of nontonal languages to have anomalous patterns of laryngealization during the production of tones. MTD-characteristic influences on phonation included incomplete glottal closure and impacted pitch quality (two factors that influence pitch discrimination), and the authors concluded that the tonal status of Vietnamese had an effect on the heightened expression of some MTD symptoms during speech. So there is already work demonstrating that a pathology that is more prevalent in dry ambient conditions is associated with the impairment of the production of lexical tone among native speakers of a tonal language.

These deleterious effects of desiccated air on precise phonation have also been confirmed in a series of ex vivo studies with excised canine and ovine larynges (31⇓–33). This confirmation suggests plainly that the sensitivity of speakers’ vocal cords to desiccated air is genotypically and phenotypically normal and operative across all human populations. In short, it is well established that desiccated ambient air results in compromised voice quality, as evidenced by increases to jitter, shimmer, PPE, and PPT. In contrast, conditions of humidity and hydration are positively associated with facilitated phonation and more specifically facilitated precise phonation, given the associated comparatively reduced rates of jitter and shimmer.

Languages with phonemic tone necessitate voicing at relatively precise pitches throughout an individual’s normal range. As noted in cross-linguistic surveys of fundamental frequency, the typical pitch range for most human males is about 100 Hz (34⇓–36). The just-noticeable difference between lexical tones is about 10 Hz (37), and a cross-tone pitch difference of at least 20–30 Hz is considered marginally sufficient to achieve phonemic contrast (38). These figures suggest that languages with more than three level phonemic tones present articulatory and perceptual challenges, a suggestion that is supported by work on the acquisition of tonality (39, 40). Some research suggests that the upper limit on level phonemic tones in a language is five (41), and languages with this many contrasts tend to resort to minor differences in laryngeal setting (creaky, breathy) to express tonal contrasts (34, 42) In short, producing complex tonality requires precise vocal control. The imprecision of phonation characteristic of desiccated contexts makes the task more challenging.

Distribution of the World’s Tonal Languages

Many languages of the world use phonemic tone, in which pitch is used to contrast lexical meaning. Some estimates (49) suggest that over half the world’s languages are tonal at least to a moderate extent, and of course, all spoken languages require the modulation of fundamental frequency for some purpose, be it word-level stress, phrasal stress, or general pragmatic functions. Only a smaller portion of the world’s languages use a system of complex tone, typically defined as a system containing three or more pitch-based phonemic contrasts (50). In contrast to other linguistic uses of pitch modulation, systems of complex tone generally require greater precision of pitch modulation at the level of individual segments of sound, as phonemic categorization and associated semantic judgments can be directly impaired by imprecision of pitch in such cases.

We examined the geographic distribution of languages with phonemic tone, paying particular attention to those with complex tone in two large independently coded phonological databases, the World Atlas of Linguistic Structures online (WALS) of the Max Planck Institute and the Phonotactics Database of the Australian National University (ANU). These databases have been constructed by prominent linguistic institutions and are based on careful sampling of source materials collected by field linguists throughout the world (see refs. 51 and 52 for details on the creation of these databases). The 527 languages in the WALS database were chosen so as to somewhat equitably represent world regions and linguistic families, although rigorous phylogenetic controls were not applied (51). In contrast, the creators of the ANU database simply strived to represent as many languages as possible (52). For that reason, we relied on the ANU database in our Monte Carlo analysis below, which allows for stricter controls of genealogical relationships.

The language locales in these databases represent the best estimates of the places to which the respective languages are indigenous. Fig. 1 is a map representing the distribution of languages with and without systems of complex tone, for the larger ANU database (3,756 languages). As we see in the figure, languages with complex tone (n = 629) are located primarily in tropical regions and are generally absent in extremely desiccated environments, whether high latitudes or regions such as the Atacama, Sahara, Gobi, Arabian, and Australian deserts. Two alternate maps, including a heat map of tone “hot spots,” are presented in Figs. S1 and S2.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Distribution of languages with complex tone (red dots) and without complex tone (blue dots) in the ANU database. Darker shading on map corresponds to lower MH.

Phoneticians have also noted that tone tends to be more heavily associated with particular linguistic areas than with linguistic families (50). Such areas are, in line with our prediction, largely restricted to major warm regions of high humidity, most notably Southeast Asia, sub-Saharan Africa, and New Guinea. In the Americas, languages with complex tone are found primarily in the southern regions of North America and, in the few cases they exist in South America, in Amazonia.

The prediction is further supported by the distribution of languages according to the mean specific humidity (MH) and mean annual temperature (MAT) values of their locales. (Specific humidity refers to the ratio of water vapor to total air content.) We extracted MH data for each locale from the climate database in ref. 53, which includes monthly averages for regular geographic points across the earth from 1949 to 2013. For each point and each month, the average MH was calculated over all years. Then, the mean of each monthly mean was used as the MH for that geographic point. We used specific humidity for our global analysis, rather than relative humidity, as our predictions are strongest for very frigid regions and the absolute water content of subfreezing air approaches zero, or is otherwise negligible, regardless of saturation rate (26). MAT data for each pair of coordinates were gathered via the bio-clime package accessible through ArcGIS, a geographic information system platform (www.arcgis.com). MH and MAT were collected for the language locales in both the WALS and ANU databases. For Fig. 2 and Fig. S3, we relied on the WALS data, as those data fairly equitably represent different regions and language families.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Empirical cumulative distribution function for languages according to the MH of their locations, WALS sample. The bottom quartile of language locales (by MH) is shaded.

Fig. 2 shows the cumulative proportion of languages observed as the specific humidity increases. This distribution increases more slowly for complex tone languages than languages without tones. Languages with complex tone are clearly unlikely to occur in regions with extremely low MH. MH data are the most crucial for our case, as they factor in all arid locales—whether hot deserts or frigid regions. The languages with complex tone clearly distribute primarily in regions with higher humidity values and are absent in the most desiccated regions. Furthermore, there is a clear step-wise gradation whereby languages with simple tone occur in desiccated regions at a higher rate than languages with complex tone, but at a lesser rate than languages without any tone. The findings vis-à-vis MAT also reflect this pattern, as evident in Fig. S3. Languages in very cold regions uniformly avoid complex tone and, to a lesser extent, simple tone. This is particularly true in subfreezing contexts. Consider that, of the 629 languages with complex tone in the larger ANU database, there are only two clear instances of languages spoken in subfreezing (MATs of less than 0 °C), very desiccated regions, or 0.32%. In contrast, 101 of the 3,127 (3.2%) remaining languages are found in such regions; they occur at exactly 10 times the rate in such contexts.

However, analyses of aggregated data, even those that strive to represent diverse regions and families (like the WALS data), might be misleading. No rigorous genealogical control was performed on the typological data thus far examined, so larger families could be better represented than smaller ones or language isolates. Also, the number of languages with complex tone is smaller than that with simple or no tone, so one could argue their narrower distribution around warmer regions (where most of the languages of the world exist) is accidental.

To assess the significance of our results while more carefully controlling for these factors, we conducted an analysis of range in a Monte Carlo framework. This type of analysis allows us to test for differences in the shape of the distributions of language types while also controlling carefully for genealogical influences. In every run of the simulation, a balanced sample of languages with complex tone was chosen at random from the ANU database by taking only one representative of each linguistic family among those in which complex tone was attested; the same was done for languages with simple or no tone. The goal was to compare whether the distributions of these samples differed for the two groups of languages. We ran 5,000 simulations at the 15th, 25th, 50th (median), and 75th percentiles of MH and MAT values, for each of the two language types; this yielded a total of 40,000 simulations. The lower percentiles are particularly crucial, as our predictions are strongest for low MH and MAT values.

If the tendencies apparent in the density distributions were mere artifacts of genealogy or areal effects, then every pair of Monte Carlo samples should exhibit similar climatic properties. In other words, sampled languages with complex tone would have higher MHs and MATs than languages without complex tone in about 50% of the cases, at each percentile. The distributions of the languages with and without complex tone clearly differ. The 15th, 25th, 50th, and 75th MH percentiles of balanced samples of languages with complex tone have higher MHs in 89%, 88%, 43%, and 49% of the sample cases, respectively, when contrasted to the simulations’ balanced samples of the same percentiles of remaining languages. This is exactly in line with our predictions, as languages with complex tone overwhelmingly have higher MHs in the lower percentile ranges, suggesting clearly that such languages are extremely infrequent in very arid contexts, regardless of temperature. The 15th, 25th, 50th, and 75th MAT percentiles of balanced samples of languages with complex tone have higher MATs in 93%, 77%, 17%, and 19% of the sample cases, respectively, when contrasted to the equivalent simulations’ balanced samples of remaining languages. These results also pattern neatly in the direction of our prediction, as languages with complex tone are clearly particularly avoided in very cold regions but also in very hot, arid regions. In Fig. 3, the Monte Carlo simulations based on the lower percentiles of MH and MAT, for which our account makes the strongest predictions, are summarized graphically. For our Monte Carlo results, the difference distributions have location parameters outside the 95% confidence interval for the null hypothesis. This is unsurprising given the large effect sizes depicted in Fig. 3. Figs. S4 and S5 present all of the Monte Carlo simulations. (Fig. S5 also includes some additional comments regarding the avoidance of complex tone in the hottest regions, also consistent with our account.)

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Monte Carlo results. Distribution of the differences in the (A) 15th and (B) 25th percentiles of MH as well as the (C) 15th and (D) 25th percentiles of MAT. Each chart represents 5,000 samples. For each sample, one language per family among all languages with complex tone is considered, and the same number of languages without complex tone is sampled, again with only one representative per family. The values at each percentile are then contrasted, for sampled languages with and without complex tone. Blue results represent cases in which the MH and MAT for languages with complex tone were higher than that of the remaining languages. The x axis describes the disparity in MH for A and B and in degrees Celsius for C and D. The y axis represents number of samples.

The distribution of the seven language isolates with complex tone is also restricted to warm, comparatively humid regions: Amazonia (Ticuna, Pirahã), sub-Saharan Africa (Laal, Banggime) and New Guinea (Damal, Lepki, Morori). In contrast, the 108 isolates without tone are spoken throughout the Americas, Eurasia, Africa, and Australia. A number are spoken in high latitudes in North America, South America, and Eurasia, as well as in numerous other arid regions. For both MH and MAT, there are clear disparities across isolates with and without complex tone, respectively. The average MH for isolates with complex tone is 0.017, whereas the average for other isolates is 0.013. This cross-group disparity is significant (P = 0.02, Mann–Whitney). Similarly, the average MAT for isolates with complex tone, 23.7 °C, is greater than the average for the remainder of isolates, 19.1 °C (P = 0.07, Mann–Whitney). These disparities correspond neatly with the predictions of our account, despite the small number of isolates with much tonality. Given the geographic breadth of the isolates, the disparity cannot be ascribed to areal effects either.

We also considered the language families with complex tone that straddle extremely diverse ecological zones: Afro-Asiatic, Sino-Tibetan, Nilo-Saharan, and Niger-Congo. Languages without tone in these language families are much more likely to be found in comparatively arid regions, when contrasted to those with tone and especially complex tone, and this disparity is statistically significant in each case (SI Text, Intrafamily Analysis).

Much as it is not simply due to phylogenetic confounds, the global pattern in question is not driven by one or a few linguistic areas. It surfaces clearly within Africa and Eurasia. Also, because some languages with complex tone do exist in southern North America (particularly Central America) and Amazonia but are generally absent elsewhere in the Americas, the pattern surfaces in the New World despite the generally reduced reliance on tonality in that part of the world. In addition, one of the other major noncontiguous regions with relatively robust levels of complex tonality is New Guinea. The statistically significant within-continent correlations are described in SI Text, Intracontinental Analysis.

Our intrafamily, cross-isolate, and Monte Carlo analyses suggest that the distribution in question is not due to phylogenetic factors. Our finding that the pattern surfaces within continents, in addition to our balanced Monte Carlo and cross-isolate results, suggests strongly that influences associated with specific regional patterns of linguistic contact also do not account completely for the patterns. From our perspective, we must ask why linguistic tone is so pervasively transferred across languages in some regions and why those regions tend to have arid borders. We believe that our account offers a natural explanation for this tendency and directly explains why the regions in which interlinguistic contact has led to pervasive use of tone—for example, sub-Saharan Africa and Southeast Asia—are warm humid regions. Put differently, it seems likely that tone spreads across languages more effectively via interlinguistic contact in regions with favorable ambient conditions, and very cold/dry regions apparently serve as barriers to the spread of (complex) tone. Both external and internal diachronic processes may be impacted to some degree by such ambient conditions, in ways that, at the least, merit serious consideration.

Discussion and Conclusion

This paper used findings from vocal fold physiology to predict that climatic factors constrain the use of phonemic tone. Our investigation showed that the distribution of tonal languages is in line with this prediction.

Phonetic accounts of tonogenesis, focusing on the role of laryngeal consonants in conditioning the fundamental frequency of adjacent vowels (54, 55), are crucial to the understanding of the development of tone but fail to explain the climatic patterns we find. After all, prevocalic voicing contrasts are extremely common in the world’s languages, but tone and more particularly complex tone only develop in some cases. Whatever the etiology of contrastive pitch in a given language, maintaining and augmenting the phonemic burden of pitch requires the relatively precise manipulation of the vocal folds, which is not easily learned even during normal acquisition (35, 36). Given the requisite absolute precision of vocal fold manipulation in languages with phonemic tone (especially complex tone), it seems natural that intricate tonal contrasts would be avoided when environmental pressures also work against such manipulation. This avoidance could be facilitated diachronically through, for example, reduced accuracy/increased jitter in the production of borrowed words with complex tones.

Previous work has suggested that aspects of language appear to be adapted to ecological or social niches (56). For example, linguistic diversity is correlated with biodiversity (57) and pathogen prevalence (58), and morphological complexity is related to population size (59). Our results are further suggestive of external influences on language. In contrast to previous work on putative geographic/climatic influences (2⇓⇓⇓⇓⇓⇓⇓–10), however, the causal relationship we are suggesting is supported by many experimental results. It is also supported by patterns in a much larger sample of the world’s languages, including patterns within language families and among language isolates. Like the communication systems of avian species (60, 61), it appears that human languages are environmentally adaptive. This is not necessarily surprising given the general adaptability of human behavior (62, 63) and human biology (64, 65) to cold climes and given the existence of other ecological effects on nonconscious patterns in human behavior (66).

We have not made specific claims vis-à-vis the time scale involved in this ecological adaptation. Numerous complex factors contribute to diachronic sound shifts (67, 68), and it is possible that in the short term some sound changes may be inconsistent with this adaptation. Nevertheless, the synchronic distribution of the world’s languages suggests that the adaptation is at work over the long term.

The physical components of speech articulation all represent exapted portions of the upper respiratory and digestive tracts that have more primal functions. In other words, speech articulation is ultimately a biological phenomenon. We think that, absent any of the distributional data we have considered, it is actually natural to expect that phonetic patterns are susceptible to environmental pressures. The fraction of research in laryngology we have surveyed suggests in and of itself that the expectation of such pressures is more than reasonable. Particularly in light of the ubiquity of speech in the human experience, we suspect that human sound systems are susceptible to ecological pressures. In cultures in which speech volume has been assessed quantitatively, it has been found that humans produce on average 15,000+ words per day (69). So, if speakers rely on a sound pattern that is maladaptive (even in minor ways) in particular ambient conditions, they do so ubiquitously. In the case at hand, languages with complex tone require the pervasive implementation of a particular behavior, very precise phonation, which is according to all extant pertinent experimental work simply ill-suited to desiccated contexts.

We submit that our central claim is the most plausible available conclusion in light of the relevant linguistic, physiological, and quantitative evidence, and furthermore that the latter two sorts of data are crucial for shedding light on the distribution of language types. Although we believe our account is extremely plausible, we recognize that more research is required to fully explore these issues. We hope that experimental phoneticians and others examine the effects of ambient air conditions on the production of tones and other sound patterns, so that we can better understand this pivotal way in which human sound systems appear to be ecologically adaptive.