Signal brevity and compression in loud calls of the white-headed langur Authors: Penglai Fan, Yinshu Liu, Guichuan Zeng, Ruoshuang Liu, Michael J Lawes, Qihai Zhou, Ming Li Published: 2026 ABSTRACT SummaryZipf’s law of brevity and Menzerath’s law are linguistic principles, predicting that coding length is minimized for optimal coding efficiency. We tested whether two linguistic laws apply to vocal communication of white-headed langurs inhabiting karst forests. Adult male white-headed langurs produce three loud call types in long-distance sequences. Frequent call types were shorter in duration, supporting Zipf’s law of brevity. Patterns related to Menzerath’s law were more nuanced. Inter-call intervals shortened in longer sequences, but only wahoo calls showed duration compression. Physiological constraints caused calls to shorten and intervals to lengthen later in sequences. Longer sequences contained proportionally more short snorts and fewer long roars and wahoos. Signal compression reflects a trade-off between energy cost and communication efficiency. Constant durations of snorts and roars may aid long-distance transmission and caller recognition. These findings provide evidence for signal brevity and compression under physiological limits, offering insights into vocal communication evolution in non-human primates.Subject areas: wildlife behavior, zoology, evolutionary biology Introduction Human language has a sophisticated and complex design, which is typically characterized by the economical use of words to convey complex information.1,2,3 Studies have argued that human languages comply with several universal statistical patterns or linguistic laws, reflecting the principle of information-theoretic compression that minimizes coding length for maximum coding efficiency.3,4 Specifically, Zipf’s law of brevity posits that as the duration of a vocal element increases, its frequency of use decreases.4,5,6 Thus, less costly shorter vocal elements should be used more frequently than longer vocal elements.7 Empirical evidence for Zipf’s law of brevity is ubiquitous in human languages.8,9 Similarly, Menzerath’s law states that vocal elements should be shorter in longer vocal sequences.8,10,11 This law describes a trade-off between efficient information transfer and energy expenditure,12,13 being derived from the structure of written language where syllable size decreases with the size of the word, which is corroborated in human languages.3,14 Menzerath’s law was refined by Altmann (1980)14 who argued that the size of a whole vocal sequence is mathematically related to the sizes of its components (Menzerath-Altmann’s law), with parameters that may differ across languages.15,16 Some studies have indicated that there are numerous parallels between human language and communication systems of other animals.4,17 Thus, adopting a comparative approach to vocalization behavior in both humans and non-human primates may provide insights into the evolution of human language.3 Several studies have examined whether Zipf’s law of brevity applies to the vocal communication of different species,3 demonstrating that not all animals conform to this law. Zipf’s law of brevity seems to arise from a confluence of various intricate elements. While the efficiency of physical articulation is a significant factor, with frequent words tending to be briefer with more effortlessly pronounced phonemes, elements such as predictability and the value of information conveyed could also be contributing factors.18 For example, adherence to Zipf’s law of brevity with more frequent use of acoustic elements of shorter duration is associated with short-distance vocalizations of animals, such as black-capped chickadees (Poecile atricapillus),19 Formosan macaques (Macaca cyclopis),7 and four insectivorous bat species (black-bearded tomb bat, Taphozous melanopogon; Mexican free-tailed bat, Tadarida brasiliensis; greater horseshoe bat, Rhinolophus ferrumequinum; and least horseshoe bat, R. pusillus).20 Recent studies have found that the structures of long-distance vocal communications are consistent with the predictions of Zipf’s law of brevity for the morning loud calls of two species of crested gibbons (cao vit gibbons, Nomascus nasutus; and western black-crested gibbon, N. concolor)21 and the phrases emitted by indris (Indri indri).22 These findings suggest that a cost-benefit trade-off in the energy efficiency of specifically loud vocalizations governs the laws of vocalization brevity. In contrast, there is no compelling support for Zipf’s law of brevity in the complete vocal repertoires of the common marmoset (Callithrix jacchus), golden-backed uakari (Cacajao melanocephalus), and Hainan frilled treefrog (Kurixalus hainanus), as the most frequently uttered calls are not the shortest acoustic signal in their vocal repertoire.23,24 Thus, coding efficiency is not the sole factor influencing the structure of signal elements in vocal communication. Very brief signals can impede true and accurate information transmission in situation-specific contexts.23 Accordingly, the solos of Müller’s Bornean gibbons (Hylobates muelleri) do not support Zipf’s law of brevity,25 suggesting that in addition to the size of the vocal element, the social context, such as dominance displays or mate choice behavior, may affect adherence to linguistic laws. Indeed, studies of Menzerath’s law in vocal sequences show varying and context-dependent support for signal compression across taxa. It has been hypothesized that Menzerath’s law is fundamentally rooted in physical factors, with natural selection favoring a more efficient vocal mechanism.8,26 In vocal sequences, this law can manifest in two ways that are not mutually exclusive: through compression, the shortening of individual units in longer sequences, or through compositional change, an increased proportion of calls that are inherently brief.12,22 A study on close-range vocal sequences of male geladas (Theropithecus gelada) revealed a negative correlation between call duration and sequence size, aligning with the principles of Menzerath’s law.12 Furthermore, the study found that factors such as breathing patterns and energy constraints were the underlying proximate mechanisms of this law, leading to shorter calls occurring later in the sequence. Concurrently, the study also found evidence of compositional change, with larger sequences containing a higher proportion of the shortest call type and a lower proportion of the longest call type.12 The general patterns of signal compression predicted by Menzerath’s law have been also observed in the long-distance pant hoots of chimpanzees (Pan troglodytes), the loud morning songs of the cao vit gibbon and western black-crested gibbon,21 the solos of Müller’s Bornean gibbon,25 the phrase of indri,22 the display songs of African penguin (Spheniscus demersus),27 and songs of 15 songbirds.26 However, when selection pressures like sexual selection or urgent social contexts surpass the pressure for acoustic compression, Menzerath’s law does not apply to vocal sequences where compression is unnecessary, such as the close-call sequences of mountain gorillas (Gorilla beringei) and long-distance vocal sequences of tarsiers (Tarsius spectrumgurskyae), titi monkeys (Plecturocebus cupreus), and gibbons (Hylobates funereus).13,28 Notably, Menzerath’s law also applies to temporal intervals as well as vocal units, as meerkat (Suricata suricatta) calls show compression between, but not within, calls.29 Its manifestation, thus, depends on the specific structural level under examination. A possible explanation is that a shorter interval duration increases the number of calls in a single breath, which can improve the efficiency of vocal transmission.12 A complete test of the law should, therefore, consider both compression and compositional changes as mechanisms that may jointly or independently enhance signal efficiency in vocal sequences. Zipf’s law of brevity and Menzerath’s law of compression are thought to be universal in animal communication, but the extent to which vocalizations in non-human primates, especially those with complex social system, adhere to linguistic laws is still uncertain. Studies have indicated that vocal communication plays an important role in large and relatively complex Colobine societies.30,31 White-headed langurs (Trachypithecus leucocephalus) have a rich vocal repertoire, characterized by unique age/sex-specific call types, with 11 call types emitted by adults and six by juveniles and infants.31 The resident adult male produces long-distance vocal sequences that include three acoustically distinct loud call types, namely snorts, roars, and wahoos (Figure 1). Snort has the shortest duration and the greatest bandwidth, roar is intermediate in both duration and bandwidth, and wahoo has the longest duration and the narrowest bandwidth.31 Typically, the time interval between consecutive snorts is longer than that between consecutive roars or wahoos (Figure 1). These three types of loud calls are combined into sequences in various ways, with different sizes and arrangements of calls.31 Vocal sequences are uttered at any time of the day but most frequently in the early morning and late afternoon (unpublished data, Penglai Fan). They are primarily used in vigilance and disturbance contexts associated with inter-group spacing or resource defense.31 Thus, these long-distance loud calls of white-headed langurs provide an opportunity to simultaneously test Zipf’s law of brevity and Menzerath’s law of compression. Despite evidence that call compression compromises information in long-distance calls,21,22 it remains unclear how white-headed langurs resolve the challenge of maintaining signal communication efficiency under energy/breath constraints. Example spectrogram of a vocal sequence emitted by an adult male white-headed langur Call types are labeled A (snort), B (roar), and C (wahoo). Numeric prefixes (e.g., 4B and 2C) denote the call counts. Here, we conducted a study on white-headed langurs, aiming to determine whether the long-distance loud calls of adult males adhere to the laws of brevity and compression, as well as to identify the potential mechanisms underlying these linguistic laws in loud-call vocal sequence. For Zipf’s law of brevity, we tested whether the duration of each call type is inversely proportional to its frequency of use. For Menzerath’s law, we tested whether longer vocal sequences contain shorter (compressed) call elements. We also applied Menzerath’s law to the duration of inter-call intervals within a vocal sequence to determine if these decrease as the sequence size increases. Patterns consistent with Menzerath’s law could arise from proximate energetic or respiratory constraints on vocal production, leading to shorter vocal units being produced later in sequences or more shorter calls being emitted in longer vocal sequences.12 To further explore the underlying mechanisms of Menzerath’s law, we examined (1) whether the duration of calls within a vocal sequence is shorter later in the vocal sequence, (2) whether the duration of inter-call intervals becomes longer later in the vocal sequence, and (3) whether shorter call types are emitted more frequently in longer vocal sequences. Given potential energetic or respiratory constraints in loud call production, we predict that later calls within a sequence will be shorter in duration and followed by longer inter-call intervals to facilitate breath recovery. Additionally, because shorter call types (e.g., snorts) likely incur a lower energetic cost, we further predict their proportional increase in longer vocal sequences. Results Zipf’s law of brevity Zipf’s law predicts that more frequently used linguistic elements or calls in a vocal sequence will be of shorter duration. In this study, the mean durations were 0.135 ± 0.040 s (x¯ ± SD; range: 0.025–0.316 s) for snorts, 0.328 ± 0.087 s (range: 0.070–0.653 s) for roars, and 0.333 ± 0.099 s (range: 0.028–0.791 s) for wahoos. The average individual usage frequency (x¯± SD) was 936 ± 579 calls for snorts, 391 ± 346 for roars, and 398 ± 315 for wahoos. The three loud call types differed significantly in their duration (Tukey’s post-hoc test: snort versus roar: estimate = −0.1968, z = −155.78, p < 0.001; snort versus wahoo: estimate = −0.2023; z = −158.97, p < 0.001; roar versus wahoo: estimate = −0.0055, z = −3.69, p < 0.001). The duration of the loud call type was negatively associated with the frequency of its use (full vs. null: χ2 = 9.655, p = 0.002; estimate = −0.0001, t = −3.25, p = 0.003, 95% confidence interval [CI: −0.00016, −0.00004]), supporting Zipf’s law of brevity (Figure 2). Relationship between the number and duration of each call type Data are represented as the mean ± SD. p values are from linear mixed-effects models. Three asterisks denote p < 0.001. The intensity of colors indicates the density distribution of data points. Darker areas represent higher concentrations of data points at those locations, meaning that the observed values occur more frequently. Menzerath’s law Menzerath’s law predicts that larger linguistic sequences will have shorter (i.e., more compressed) constituent calls. Vocal sequence size of white-headed langurs ranged from 2 to 331 calls (x¯± SE = 26 ± 1 calls in a vocal sequence). As the sequence size increased, the duration of call significantly decreased (full vs. null: χ2 = 722.82, p < 0.001; estimate = −0.0498, t = −7.98, p < 0.001, 95% CI [−0.06209, −0.03760]) (Figure 3), consistent with Menzerath’s law of compression. Specifically, as the sequence size increased, the duration of the snort (full vs. null: χ2 = 71.52, p < 0.001; estimate = −0.0038, t = −0.189, p = 0.851, 95% CI [−0.04365, 0.03600) and roar (full vs. null: χ2 = 1.95, p = 0.377; estimate = −0.0103, t = −1.28, p = 0.267, 95% CI [−0.02608, 0.00547]) remained relatively stable, while the duration of wahoo shortened (full vs. null: χ2 = 331.07, p < 0.001; estimate = −0.0431, t = −4.60, p < 0.001, 95% CI [−0.06151, −0.02478]), indicating compression of wahoo calls with larger vocal sequence size. The interval duration was significantly shortened with increasing vocal sequence size (full vs. null: χ2 = 328.71, p < 0.001; estimate = −0.6843, t = −5.14; p < 0.001, 95% CI [−0.94542, −0.42310]) (Figure 3). The interval durations after a snort (full vs. null: χ2 = 376.34, p < 0.001; estimate = −0.4904, t = −14.81, p < 0.001, 95% CI [−0.55506, −0.42579]) and a wahoo (full vs. null: χ2 = 480.09, p < 0.001; estimate = −0.2400, t = −6.03, p < 0.001, 95% CI [−0.33891, −0.13238]) were significantly shortened as the vocal sequence size increased. However, the interval duration after a roar (full vs. null: χ2 = 1.74, p = 0.418; estimate = −0.0528, t = −1.30, p = 0.222, 95% CI [−0.13243, 0.02676]) remained relatively stable. Relationships between call duration and sequence size and between call duration and position within sequences for all call types (snort, roar, wahoo) in adult male white-headed langurs p values are from linear mixed-effects models. One asterisk denotes p < 0.05, and three asterisks denote p < 0.001. Trend lines with different colors are displayed for different call types. Shaded gray area represents 95% confidence interval. The intensity of colors indicates the density distribution of data points. Darker areas represent higher concentrations of data points at those locations, meaning the observed values occur more frequently. Compression strategies of loud calls When all call types were analyzed together, call duration was found to be shorter in later positions in a sequence (estimate = −0.0659, t = −10.76, p < 0.001, 95% CI [−0.07790, −0.05389]). However, we found that the duration of snorts was shorter in later positions (estimate = −0.0177, t = −8.43, p < 0.001, 95% CI [−0.02177, −0.01355]), the duration of roars was independent of the call position (estimate = 0.0047, t = 0.61, p = 0.545, 95% CI [−0.01055, 0.01998]), and the duration of wahoos was longer in later positions in a vocal sequence (estimate = 0.0875, t = 18.58, p < 0.001, 95% CI [0.07831, 0.09678]) (Figure 4). Relationships between interval duration and sequence size and between interval duration and position within sequences for all call types (snort, roar, wahoo) in adult male white-headed langurs p values are from linear mixed-effects models. One asterisk denotes p < 0.05, three asterisks denote p < 0.001. Trend lines with different colors are displayed for different call types. Shaded gray area represents 95% confidence intervals. The intensity of colors indicates the density distribution of data points. Darker areas represent higher concentrations of data points at those locations, meaning the observed values occur more frequently. The interval duration was longer in later positions in a vocal sequence (estimate = 0.4961, t = 18.08, p < 0.001, 95% CI [0.44229, 0.54983]). The interval durations after a snort (estimate = 0.1795, t = 16.43, p < 0.001, 95% CI [0.15811, 0.20094]) and a wahoo (estimate = 0.4213, t = 22.96, p < 0.001, 95% CI [0.38532, 0.45723]) increased in the later parts of a sequence. The interval duration after a roar (estimate = 0.0135, t = 0.70, p = 0.483, 95% CI [−0.02421, 0.05119]) remained relatively stable in the later parts of a sequence (Figure 4). Longer vocal sequences were significantly more likely to contain more snorts (full vs. null: χ2 = 265.74, p < 0.001; estimate = 0.1444, t = 15.65, p < 0.001, 95% CI [0.12632, 0.16249]) (Figure 5). In contrast, the proportions of roars (full vs. null: χ2 = 373.17, p < 0.001; estimate = −0.5101, t = −8.06, p < 0.001, 95% CI [−0.63413, −0.38610]) and wahoos (full vs. null: χ2 = 2654.1, p < 0.001; estimate = −1.0038, t = −37.84, p < 0.001, 95% CI [−0.105576, −0.95177]) were significantly less in longer vocal sequences (Figure 5). Thus, the proportions of the short call type (snorts) increased and of the long call types (roars and wahoos) decreased with the increasing vocal sequence size. Relationship between the proportions of call types and sequence size in adult male white-headed langurs Trend lines with different colors are displayed for different call types. Shaded gray area represents 95% confidence interval. Discussion In this study, we simultaneously examined Zipf’s law of brevity and Menzerath’s law of compression in long-distance loud calls of white-headed langurs. The three call types comprising the long-distance vocal sequences differed significantly in their duration. In support of Zipf’s law of brevity,6 our results indicated that the frequency of use of the call types declined when their duration increased (Figure 2). When examining Menzerath’s law,11 a more nuanced pattern emerged. While inter-call interval shortened in longer vocal sequences, call duration compression was evident only for wahoos. Furthermore, in larger sequences, call duration became shorter and interval duration longer toward the end of sequence (Figure 4). In addition to these temporal changes, sequence composition also shifted with size, with the proportion of snorts being increased while the proportion of roars and wahoos being decreased in larger sequences (Figure 5). These findings provide the first line of evidence that long-distance loud call communication in the subfamily Colobinae follows the principles of linguistic laws that apply to humans as well as non-human primates. In white-headed langurs, there was a negative correlation between the use frequency and duration of loud call types, which is consistent with Zipf’s law of brevity. Our results indicated that the different loud call types varied in duration (Figure 2), suggesting they have different energy costs.7,32 Accordingly, we thought that shorter snorts should be the most energy-efficient loud call and the most used, and longer wahoos would be the least energy-efficient loud call type emitted by white-headed langurs, revealing a trade-off between the energy costs of a call and its effectiveness in communicating a message.4 Studies have indicated that Zipf’s law of brevity could apply to the loud call vocal sequences of many animals such as gibbons,21 African penguins,27 and indris,22 but this law was not detected to be applicable to the vocal repertoire of the common marmoset and golden-backed uakari23 or to the solo song of Müller’s Bornean gibbons.25 On the basis of factors highlighted in previous studies, such as coding efficiency and the species-specific, context-dependent costs and benefits of syllable duration,23,25 we suggest that these contrasting findings likely stem from the variations in how these factors manifest across different species. Our findings revealed that the overall organization of larger call sequences in white-headed langurs is consistent with Menzerath’s Law. At the call type level, the call duration of neither snorts nor roars showed compression with increasing sequence size, while wahoos exhibited a reduction in duration in larger sequences. Therefore, sequential compression in this species is a call type-specific adjustment, rather than a universal rule. Similarly, in the vocal sequences of gelada baboons,12 the loud morning calls of the cao vit gibbons and western black-crested gibbons,21 and the displays songs of the African penguin,27 the duration of specific call types declined with the increase in vocal sequence size. In animal communication, information compression reflects an evolutionary trade-off between coding efficiency and energy expenditure.4,5 However, in white-headed langurs, this principle is expressed differently across the three loud call types of white-headed langurs, reflecting a functional divergence. Wahoo calls are the longest and least frequent call type, and they exhibited duration compression in larger sequences, consistent with an energy-saving function. By shortening the costliest calls, langurs may optimize energetic output while preserving their effectiveness in long-distance broadcasting. In contrast, snort and roar calls maintained stable duration regardless of the sequence length. These calls, which are physiologically demanding and maintain stable durations due to substantial lung and air sac capacity, may function as honest indicators of male quality in contexts such as territory defense and individual recognition.30,33 Thus, the invariance in duration may align with sexual selection theory, which emphasizes signal reliability over coding efficiency.12,24,34 Taken together, the loud call vocal system of white-headed langurs appears to achieve a strategic balance. Energy-saving compression targets the most metabolically expensive wahoo calls, while sexual selection likely preserves the stable duration of signals essential for intrasexual competition and mate attraction, namely snorts and roars. This study found that inter-call intervals in white-headed langurs’ vocal sequences decreased significantly as the sequence lengthened, aligning with the predictions of Menzerath’s law (Figure 3). This trend is not unique to animal vocal communication. For instance, in meerkats vocal sequences, while no compression in call duration was observed, the inter-call intervals similarly shortened in accordance with Menzerath’s law.2 Notably, the occurrence of such interval compression, even when intervals can span dozens of seconds, suggests that the underlying principle may not be driven solely by an acute energetic need for brevity but could instead reflect a more general optimization for rhythmic regularity or efficient sequencing in vocal delivery. Our results further indicated that this principle held across all loud call types in white-headed langurs. As the overall vocal sequences increased in size, the duration of intervals following each of the three loud call types decreased accordingly. This implied that white-headed langurs employ the same pause strategy tailored to all loud call types, with the interval duration being compressed in longer vocal sequences. While repetition of the loud call in a vocal sequence may generate information redundancy, it can compensate for signal attenuation and improve individual recognition by receivers during long-distance communication.35 Consequently, more repetition of call types should be associated with their compression in longer vocal sequences, similar to those in crested gibbons21 and African penguins.27 However, whether adjustments in the interval duration reflect varying levels of competition or defense capabilities among different white-headed langurs needs to be further investigated. Considering all call types together, we found a negative relationship between call duration and call position but a positive association between interval duration and call position in a vocal sequence. That call duration, interval duration, and call position within a vocal sequence adhered to Menzerath’s law is attributed to the physiological or energetic limitations on vocal production, resulting in shorter calls and extended intervals as vocal sequences progress (Figure 4). Male geladas can maintain consistent call durations and reduced interval durations throughout their vocal sequences,12 while male gibbons prolong their call duration in the later parts of vocal sequences.21 Being positioned at the end of a vocal sequence, the duration of snorts shortened, while roar durations remained stable, and wahoo durations extended. These results suggested that white-headed langurs prioritize emitting longer and energy-intensive signals as lung capacity diminishes. A more detailed examination of wahoo duration, however, revealed a complex, dual-level pattern. While overall wahoo duration decreased in longer sequences, consistent with Menzerath’s law across sequences, individual wahoo grew longer in later positions within a sequence. This suggests that two competing principles may be operating simultaneously. At the level of sequence structure, efficiency-driven compression predominates, whereas at the level of within-sequence positioning, emphasis-driven prolongation becomes relevant. The latter may serve to enhance signal clarity or communicative function, analogous to how human speakers lengthen important phonemes under challenging listening conditions.26,36,37 Furthermore, extended post-call intervals following a snort and wahoo in terminal positions indicate increased time allocated for respiratory recovery. Notably, the prolonged intervals observed at the end of sequences align with the final lengthening phenomenon extensively documented in human speech, music, and bird songs.38,39 A similar pattern of prolonged intervals in later sequence positions has recently been reported in meerkats.29 It has been suggested that this may stem from a universal motor production characteristic. Specifically, movements tend to decelerate toward the end of sequences. Our findings in the primate context appear to support this notion. Together, these results illustrate how white-headed langurs may balance efficiency constraints with the need to maintain the duration of information-rich calls, thereby optimizing long-distance communication efficacy. We also found that the proportion of the short call type (snorts) increased and the proportion of long call types (roars and wahoos) decreased in longer vocal sequences in white-headed langurs (Figure 5). We suggested that the proportion of call types in vocal sequences should reflect multiple trade-offs constrained by the energy costs of loud calling on signal clarity and efficiency.12 Specifically, the production of long, costly calls such as roars and wahoos could be energetically demanding. By incorporating shorter, lower-cost calls, males can extend the overall sequence length without a prohibitive rise in total energy expenditure, effectively balancing per-call intensity against sequence-level sustainability. The same mechanisms underlying signal compression in white-headed langurs have been reported for morning loud calls in cao vit gibbons (Nomascus nasutus), which also inhabit karst landscapes and display vocal territory defense.21 Accordingly, we suggest that the environmental context could also drive selection for the convergent evolution of signal compression in long-distance vocal communication in primates. In conclusion, our findings strongly support Zipf’s law of brevity in white-headed langurs, demonstrating an inverse relationship between the duration of loud calls and their frequency of use. Patterns related to Menzerath’s law were more nuanced. While inter-call intervals shortened in larger sequences, call duration compression was evident only in one (wahoos) of the three call types. We also documented positional effects on the call and interval durations, as well as changes in the proportional use of call types within larger sequences. Together, these results suggest a vocal system in which cross-species linguistic principles are expressed variably. This variability likely reflects a balance negotiated between the transmission demands of vocal efficiency, energetic cost, and the distinct functional roles of different calls. We note as a primary limitation that our inferences regarding physiological constraints and functional trade-offs rest on correlational data. Although consistent with theoretical expectations, these findings would benefit from direct physiological measurement and experimental validation. Additionally, because our analysis focused on the sequence level, further research examining individual variation and contextual usage in finer detail could yield a deeper insight. Limitations of the study Our study has two main limitations. First, inferences about physiological constraints are based on correlational data and require direct validation through physiological measurements. Second, we analyzed only three types of long-distance loud calls, leaving the applicability of our findings to the species’ entire vocal repertoire unclear. Resource availability Lead contact Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Penglai Fan (fanpl@mailbox.gxnu.edu.cn). Materials availability This study did not generate new unique reagents. Data and code availability Data The datasets generated and analyzed during this study are publicly available in the Figshare repository at https://doi.org/10.6084/m9.figshare.29595269. Code This paper does not report original code. Other items Any additional information required to reanalyzed the data reported in this paper is available from the lead contact upon request. Acknowledgments We are grateful to the Guangxi Chongzuo White-Headed Langur National Nature Reserve for granting permission to carry out this study. We thank Dr. Zidi Wang from the School of Life Sciences, Sun Yat-sen University, for their advice on manuscript writing. We also thank Jianbao Wu, Dengpan Nong, and Shijun Wu for their support in the field study. This study was financially supported by the Key Project of Guangxi Natural Science Foundation of China (2022GXNSFDA080004), the National Natural Science Foundation of China (32470497, 31900335, and 32170492), and the Strategic Priority Research Program of Chinese Academy of Sciences (XDA23080000). Author contributions Data curation, Y.L., G.Z., and R.L.; formal analysis and visualization, Y.L.; writing – original draft, M.J.L. and P.F.; writing – review & editing and supervision, M.L., Q.Z., and P.F; project administration, P.F. All authors edited the manuscript. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERDeposited dataRepository dataThis paperhttps://doi.org/10.6084/m9.figshare.29595269Software and algorithmsR software (version 4.3.1)R Core Teamhttps://www.r-project.org/Praat software (version 6.3.11)Praat: doing Phonetics by Computerhttps://www.fon.hum.uva.nl/praat/ Experimental model and study participant details Study subjects and study site The white-headed langur, endemic to China, is critically endangered species inhabiting in karst mountain regions.40 This species has a matrilineal social organization, comprising an adult male and several adult females and their offspring, with an average group size of 12 individuals.41,42 The home range of each family group is about 15–52 ha and the resident male defends an exclusive territory.43,44,45,46 This study was conducted in the Banli area of the Guangxi Chongzuo White-headed Langur National Nature Reserve (22°24-22°46 N, 107°22–107°42 E) in Southwest Guangxi, China. The reserve’s karst landscape is covered by tropical and subtropical evergreen or seasonally deciduous forest.47,48 The forest cover is severely fragmented by agriculture, roads, and settlements.49 Ethical note All procedures in this study adhered to the wildlife protection laws of the Chinese People’s Republic of China, the regulations of the Guangxi Chongzuo White-headed Langurs National Nature Reserve and the Animal Care and Use Committee of Guangxi Normal University (GXNU-202309-006). Method details Acoustic recordings From September 2021 to December 2025, we recorded the vocalizations from 13 resident males, including both tenured and newly established males. Data collection spanned 3–5 consecutive days per month for each group, with daily observation sessions conducted from 05:30 to 12:00 and 15:00 to 20:00. We employed a combined sampling strategy. Focal animal sampling was used to systematically record all vocalizations and associated behaviors of the resident male, enabling quantitative data of call rates and sequence structure. Simultaneously, ad libitum sampling was applied to opportunistically record notable vocal events from any group member that fell outside the focal follows. All vocalizations were recorded using a Zoom F6 digital recorder (Zoom, Japan) (44.1 kHz and 16 bits sampling rate) with a Rode NTG8 directional microphone (Rode, Australia). Recordings were saved in “wav” format. Acoustic parameters Call types (snort, roar, and wahoo) were manually classified by visual inspection of sound spectrograms, based on the quantitative acoustic criteria established in our previous study and sound spectrogram on this population.31 Classification relied on key spectral features such as call duration, frequency bandwidth, and tonal structure evident in the spectrograms. The present study focused on these three long-distance call types. The remaining eight call types in the species’ repertoire, while not uncommon, are typically softer and emitted in close-range social interactions. Reliably obtaining high-quality, uncontaminated recordings of these calls under field conditions was exceedingly difficult, and the sample size was insufficient for robust quantitative analysis. To determine whether single calls belonged to the same sequence, we applied a maximum-likelihood method50,51 using the R package “diveMove”.52 Prior to analysis, we preprocessed the data by identifying and labeling very short intervals. This step was necessary because roar and wahoo calls are often produced in rapid succession based on behavioral observations. Such brief intervals within sequences may differ fundamentally from the longer intervals that likely separate distinct sequences. This approach yielded a 40-s criterion, where intervals longer than 40-s defined the boundary between independent vocal sequences. Acoustic analysis was performed using Praat software (version 6.3.11), with the spectrogram configured as follows: frequency range 0–10,000 Hz, dynamic range 60 dB, time step 0.002 s, and frequency step 20 Hz. For each resident male, we determined the number of each of the three call types in a vocal sequence. We measured the interval duration between calls, defined as the time from the end of the previous call to the beginning of the next call (Gustison et al., 2016). Vocal sequence size was determined by the total number of calls in a sequence. The duration of snort, roar and wahoo calls and their proportions in a vocal sequence were used to estimate the extent of abbreviation and compression in a call type. Quantification and statistical analysis We observed white-headed langurs over a period of 611 days between September 2021 to December 2023. Vocalizations were recorded from 13 adult males, yielding 1,128 vocal sequences. To mitigate potential bias from highly uneven sample sizes, two males with less than four sequences was excluded from all subsequent analyses (range:10–238 sequences per male). The remaining sequences consisted of 18,984 single calls, which included 10,294 snort calls, 4,306 roar calls, and 4,384 wahoo calls. For the investigation of Menzerath’s law, which relates the size of a whole to its constituent parts, we further filtered these sequences to include only those containing at least two calls. After this filtering, the dataset comprised 18,728 single calls including 10,039 snort calls, 4,305 roar calls, and 4,384 wahoo calls. Test of Zipf’ law of brevity The relationship between call duration and usage frequency across call types was evaluated using a linear mixed-effects model (LMM). In this model, mean call duration served as the dependent variable, the number of occurrences of each call type was included as a fixed effect, and individual identity was incorporated as a random effect. To further assess whether the three loud call types differed significantly in duration, an additional LMM was fitted. This model also treated call duration as the dependent variable, with call type as a fixed effect and individual as a random effect. Pairwise comparisons among call types were subsequently conducted using Tukey’s honest significant difference test. Test of Menzerath’s law To evaluate whether call and interval durations decrease with vocal sequence size and to investigate potential physiological constraints on loud calls in white-headed langurs, we fitted LMMs. In these models, the durations of calls and their subsequent intervals within a sequence served as the dependent variables. Sequence size and the position of the call/interval within the sequence were treated as fixed effects, while individual identity and sequence identity (nested within individual) were included as random effects. To assess potential respiratory constraints specifically, we examined how call and interval durations varied by position within a vocal sequence. The underlying hypothesis, based on prior research (Gustison et al., 2016),12 posits that respiratory limitations would lead to shorter calls toward the end of a vocal sequence. We then fitted separate LMMs for each call type (snort, roar, wahoo) to determine whether different calls exhibit distinct temporal patterns in relation to vocal sequences. Across all models, call and interval duration were the dependent variables, sequence size and position were fixed effects, and individual and sequence identity (nested) were random effects. To meet model assumptions and improve fit, call and interval durations were log-transformed, and sequence size and position were z-scored to standardize their scales. To quantify the relative usage of each call type within vocal sequences, we calculated call proportions by dividing the count of each call type by the total sequence size. For each call type, we fitted a generalized linear mixed model (GLMM) with a binomial distribution, specifying the call type proportion as the dependent variable, sequence size as a fixed effect, and individual identity as random effects. We assessed model fit using the “DHARMA” package to test for overdispersion. For all models, we also established null models containing only random effects and compared them with the full models using the maximum likelihood ratio method. To control the false discovery rate (FDR) across multiple tests, we adjusted the p-values using the Benjamini-Hochberg method, with the level for statistical significance set at α < 0.05. This correction reduces the likelihood of Type I errors while maintaining statistical power. All statistical analyses were done using RStudio (4.3.1) with the R-packages “lmerTest”, “lme4” and “glmmTMB” for linear mixed models, and “emmeans” for the pairwise comparisons. We employed the “performance” package53 to evaluate the normality and homogeneity of the residuals. Confidence intervals (95%) for model parameters were estimated using the Wald approximation. The majority of data visualization was performed with the “ggplot2” and “ggpointdensity” packages. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.116236. Contributor Information Penglai Fan, Email: fanpl@mailbox.gxnu.edu.cn. Qihai Zhou, Email: zhouqh@mailbox.gxnu.edu.cn. Ming Li, Email: lim@ioz.ac.cn. Supplemental information Document S1. Figures S1 and S2 References Associated Data Supplementary Materials Document S1. Figures S1 and S2 Data Availability Statement Data The datasets generated and analyzed during this study are publicly available in the Figshare repository at https://doi.org/10.6084/m9.figshare.29595269. Code This paper does not report original code. Other items Any additional information required to reanalyzed the data reported in this paper is available from the lead contact upon request.