Cooperation in humans requires groups of individuals to successfully coordinate and work together toward common or mutually beneficial goals. With the transition to agricultural societies, populations have become larger and more complex (Richerson and Boyd, 1999; Moffett, 2013). Thus, in these large, multi-ethnic societies assorting with others with similar social norms or behaviors can facilitate successful coordination and cooperation (Hamilton, 1964a,b; Axelrod and Hamilton, 1981). In evolutionary biology, an abundance of research has focused on mechanisms that allow individuals to interact preferentially with each other, showing that cooperation can evolve and stabilize when individuals preferentially interact with close kin or have a recognition mechanism to interact with other individuals with cooperative traits or shared norms (McElreath et al., 2003; Gintis, 2014). Thus, the ability to recognize others with similar social norms or behaviors is presumably under selection (Riolo et al., 2001; Jansen and Van Baalen, 2006).

As a medium of communication, human art might reflect fitness-relevant information on shared norms and behaviors of other individuals thus facilitating successful coordination and cooperation. Distinctive styles or patterns of artistic design could signify migration history, different groups with a shared interaction history (e.g., kin, neighbors, or members of the same caste) or individual-level variation and expertise (i.e., duration of practice). To investigate the long standing and growing interest in quantitative detection of ethnic markers across disciplines, we present an Bayesian analysis of a large corpus of material art created by women from Tamil Nadu in South India, called kolam drawings. Using this corpus of kolam art, we test whether stylistic variations in art can be mapped onto caste boundaries, migration and neighborhoods, and how much variation in styles can actually be accounted for by these factors. Specifically, we illustrate how we can exploit the Markovian nature of the art system to our advantage to build a hierarchical model that is able to describe and partition the variation in the complex, sequential drawing compositions in order to better understand the role of art for social coordination.

According to Pseudo-Bayesian model averaging and stacking weights in leave-one out cross-validation, our full model has the best predictive performance (see Supplementary Section 3 for more details).

On the population-level, our results illustrate that although artists are unconstrained in their patterns or stylistic variation in kolam drawings and they can freely transition back and forth between geometric spaces and gestures, artists have evident preferences and biases toward certain gestures and geometric spaces. Kolam patterns that arise in orthogonal geometric space are predicted to stay in orthogonal geometric space with a probability of 0.99 and transitioning to a different geometric space from orthogonal space to access a greater diversity of gestures hardly occurs with a probability of 0.01 (see Table 1). As seen in Table 1, if an artist draws a kolam artwork in diagonal space, they are predicted to equally likely switch to transitional space (probability of 0.51) or remain in the current diagonal space (probability of 0.49), while artists that draw a kolam artwork in transitional space are predicted to remain in the current space with a probability of 0.30 and switch to orthogonal or diagonal space with a probability of 0.51 and 0.19, respectively. Therefore, when an artist draws kolam patterns in orthogonal space, they are unlikely to transition between different geometric spaces and only draw patterns with different gestures within the orthogonal space. However, if artists draw kolam patterns in diagonal or transitional space, they tend to use a diverse set of gestures that span across multiple different geometric spaces. For a more in-depth analysis of the population-level tendencies in gesture compositions that result in specific gesture equilibria and styles in artistic design, please consult the Supplementary Section 4.2.

Figure 4 shows the estimated individual-level transitions between orthogonal, diagonal, and transitional gestures from two artists and an example kolam artwork each from their repertoire. As seen in Figure 4, artists' styles in drawing kolam artwork can be very divergent between artists. While the artist corresponding to the estimated transitions and example kolam artwork in Figure 4A would draw kolam artworks that arise and remained in orthogonal space with a probability of 0.846 at equilibrium, she would only occupy diagonal (0.041) and transitional (0.113) spaces at equilibrium. The artists corresponding to Figure 4B would create kolam artworks that occupy orthogonal (0.53), diagonal (0.35), and transitional (0.12) spaces at equilibrium. Furthermore, in contrast to the artist corresponding to Figure 4A, the artist corresponding to Figure 4B draws diagonal gestures in isolation without connecting their gestures to gestures from other geometric spaces. In fact, the kolam designs displayed in Figure 3 correspond to the artists in Figures 4A,B, respectively. Thus, the vastly different transition probabilities between gestures among artists can thus give rise to diverse patterns or motifs in kolam artwork.

Even though the estimated population-level transition matrices reveal a tendency in how kolam patterns arise, our results show that there is substantial variation in kolam drawing styles and patterns between artists. Whether styles in kolam drawings comprise of one or more transitions between geometric spaces or only arise and remain in the same geometric space, our results reveal that most of the variation in styles or patterns of kolam artwork is driven by individual differences and their expertise (see Figure 5 and Table 2). Figure 5 shows the size of the variance estimates (i.e., random effects) in transitioning to the next gesture from a given geometric space (Figure 5A) or gesture (Figures 5B–D). The variance parameters for the transition probabilities in kolam artwork are dominated by artist-level variation, whereby the artist-level variation has consistently the largest estimate on all four transition matrices. As illustrated in Figure 5, caste membership and neighborhood have smaller estimates on the four transition matrices. The duration of practice shows large estimates on the gesture transitions within diagonal and transitional space and moderately large estimates in transitions between diagonal and transitional space (see Supplementary Section 4.3). Former migration history further only shows small estimates on the transition between geometric spaces, while the estimates for within orthogonal and transitional space transitions are moderately large (see Supplementary Section 4.4).

According to the computed ICCs, kolam drawing styles with one or more transitions between different geometric spaces largely map onto individual variation (63%), so that the variance of predictions from a model with just individual artists' intercepts make up 63% of the total variance; the expertise of the artist accounts for additional 25% of the total prediction variance (see Table 2). In other words, on average 63% of the variation in the predicted transition patterns between geometric spaces can be attributed to the variation between distinct artists unexplained by other predictors, and a further 25% can be attributed to their expertise, measured as the duration of practice. Individual variation (62%) paired with artist's expertise (20%) further primarily accounts for styles of kolam artwork that arise and remain in orthogonal space (see Table 2). Caste membership can account for 12% of the variation in transitions between transitional gestures in kolam artwork, but can only account for very little variation in the rest of the estimated transition probabilities (see Table 2). Similarly, previous migration history or the neighborhood of the artist can explain 10% of the variation in transitions between orthogonal gestures, but only little in the rest of the estimated transition probabilities in kolam artwork (see Table 2).

Our statistical model revealed that styles or patterns in kolam artwork can only be very weakly mapped onto group affiliations, such as caste boundaries, neighborhoods, or previous migration. Styles or patterns in kolam artwork show group-level variation; however, the group-level variation is limited. Hence, it is unlikely that styles or patterns in this artistic tradition operate as ethnic markers. Albeit artists can relatively freely choose their styles or patterns in a kolam artwork, at the population-level, artists prefer and are biased toward specific (sequences of) gestures and prefer to remain in orthogonal gesture space. While a general tendency for specific gestures and geometric spaces exists on the population-level, artists can still widely differ in their stylistic variations or patterns in kolam art. In fact, the variation in styles and patterns in kolam art is dominated by artist-level variation and expertise. Thus, styles and patterns in kolam art can be mapped onto artists and their expertise and presumably their preferences.

Group-level variation can emerge naturally and become embedded in art through strategic decisions to communicate group affiliations and even through an unconscious population-level process of iterative learning and performance (McElreath et al., 2003). In contrast to predictions from ethnic marker theory that ethnic markers should be most prominent along cultural boundaries (McElreath et al., 2003), we show that kolam artworks only weakly covary along caste, neighborhood, or migration boundaries using a statistical approach. While the mapping between styles in kolam artworks and group affiliations is very limited, we still provide empirical evidence that there is scope for styles to become embedded in artwork as ethnic markers to signify group boundaries (for more details, see Bell et al., 2009). However, how much group-level variation is actually needed to function as an ethnic marker in a population is still unresolved.

Although art is a medium of communication and assorting with others with similar social norms or behaviors can facilitate successful coordination and cooperation, we believe that styles in art in this context do not play a coordinating role. One possible reason for the lack of covariation of artistic styles along cultural boundaries is that artists might have grown more similar to each other over time due to increased interactions (Healey et al., 2007; Granito et al., 2019). Due to the cross-sectional nature of our data, we might have not been able to see such a change over time. However, since the kolam tradition is considered community knowledge and artists share their designs with each other (Nagarajan, 2018), the development of styles might have been influenced by frequent intergroup contact, and shared attention and learning (Tomasello et al., 1993; Granito et al., 2019). Precisely, the notion of kolam art as community knowledge might have led to pressures to make artistic styles accessible and transparent to any potential audience.

The lack of migration history reflected in styles in kolam artwork could further indicate a pressure to conform to the dominant style in the region and the pressure to conform to the patrilocal style in the household. (Postmarital) Relocated kolam artists might adopt stylistic choices that are dominant in the region to communicate their belonging and similarity for successful group cooperation (Helbich and Dietler, 2008; Wallaert-Pêtre, 2008). Furthermore, the Tamil community in South India has a strongly patrilocal system of postmarital residence, thus relocated kolam artists might want to avoid communicating dissimilarity by adopting the local dominant style from their in-laws.

Avoiding conflicts with dissimilar others might be yet another reason why art, specifically styles in kolam art, have not developed to strongly map onto group boundaries and only show a limited scope for group coordination. Since individuals from different castes are often forced to cooperate with each other in the domain of farming in South India (Waring, 2012a), foreclosing valuable partnerships by using overt signals might not be desirable for successful cooperation (Smaldino, 2016). Art, specifically kolam art, might not be the preferred domain for ethnic markers to signal group membership since individuals have a multitude of (social) identities and group affiliations. Different situations and audiences might require different social identities that the artist can deliberately choose to occupy and display when suitable.

Instead, art might be a stage specifically reserved to promote artists' knowledge and expertise as well as their aesthetic preferences for symmetry, specific motifs, or gesture sequences through individual patterns and styles. Albeit artists may optimize their artistic displays toward a specific complexity “sweet spot” (Tran et al., 2021), our findings illustrate that they additionally strive to leave their individual mark on the artwork using their unique and distinct style for optimal distinctiveness (Brewer, 1991; Pickett et al., 2011). In fact, research in music has already shown that the probability of adopting a specific style disproportionately can be determined by frequency-based biases like conformity and novelty or prestige, success, and content biases (Brand et al., 2019; Youngblood, 2019). Since kolam art is culturally transmitted between artists, patterns or motifs might be subject to these transmission biases in social learning. For example, a novelty bias toward favoring unique patterns or motifs in kolam art could explain the substantial variation in kolam drawing styles between artists. Specifically, counter-dominance signaling could be a driving mechanism behind the cross-sectionally observed distinct styles between artists, since this mechanism posits that lower-status artists use highly unique styles to counter the currently dominating styles (Klimek et al., 2019). In order to investigate the cultural transmission of artistic traditions like kolam art and to disentangle the different biases at play, longitudinal data, explicit generative models, and careful consideration of the cost of adoption are required. Thus, our current data and analyses are insufficient to draw conclusions about cultural transmission processes and future research should explore this avenue in more detail.

How art can be mapped onto ethnic and cultural boundaries for the purpose of group coordination, and how much group-level variation is necessary to function as an ethnic marker are still vital questions to elucidate the coordinating role of art, and they require further investigations. Certainly, art and different aspects of art can actually play a crucial coordinating role depending on the context and the community. However, in Tamil Nadu, kolam designs might not be used as ethnic markers because there are already a variety of other ethnic markers that signify group membership. For instance, clothing, grooming, names, or the pottu (also known as bindi; see Davis, 1992, p. 25, for more information) can signify a woman's caste, religion, regional origin, class, and marital status (Mosse, 2018; Nagarajan, 2018). Furthermore, kolam designs might not be preferred as visual ethnic markers because their primary audiences are neighbors and visitors who most likely already know each other. This factor could explain a limited utility of kolam artwork serving as ethnic markers and further explain the lack of group mappings onto styles in kolam art. Another important consideration would be whether the groups are sufficiently distant to entail information about ethnic and cultural boundaries in the present sample. While we collected data from three different neighborhoods, the maximum distance between households was only slightly above 3 km. Additionally, caste endogamy persists, and thus women often still stay in the same caste community after postmarital relocation, even though artists in our sample migrated from different villages or cities of the states Karnataka, Kerala, and Tamil Nadu to Kodaikanal. Furthermore, although our sample consists of a mix of Scheduled Castes, Backward Castes, and Forward Castes, some caste groups within each of these categories are closely associated with each other or branches of the same caste community. Thus, our research only serves as an example of a systematic investigation of how styles in art can be mapped onto ethnic and cultural boundaries for group coordination by decomposing sequential behavior into states in a Markov chain. We demonstrate how describing a cultural system as a Markov chain can help us partition variation in styles to gain new evolutionary insights on the role of art for social coordination.

Certainly, loop patterns and motifs in kolam art are not limited to a sequential description using Markov chains, since ethnomathematicians have been intrigued by the many mathematical properties in kolam art (Layard, 1937; Ascher, 2002; Nagata, 2015). Kolam loop patterns are often structured by bilateral symmetry (i.e., vertical or horizontal symmetric) and rotational symmetry (see kolam drawings in Figures 2, 3). Furthermore, some complex kolam designs even exhibit fractal scaling by continuously repeating specific patterns. Symmetry, fractal scaling, geometric complexity, canvas size as well as the density of a kolam are all appreciated aesthetic qualities by Tamil women (Nagarajan, 2018, p. 165–167, 189), but not considered in the current analyses using Markov models. While the relation between complexity, density, and canvas size to individual and group-level variation has already been investigated (Tran et al., 2021), the relation between symmetrical and fractal properties of kolam designs to individual and group-level variation is still elusive and requires future explorations.

Future research should focus more on a synthesis between theoretical and statistical models and ethnographic methods in the field. For instance, classification tasks can be applied to motifs in art to gain a better understanding of the (cultural-historical) significance of specific motifs in art for social coordination (Bell, 2020). Bell (2020) demonstrated how triad classification tasks could be used to measure the signaling value of specifically selected Tongan motifs that are not constrained by functional sufficiency and thus elucidate their role for social coordination and the implied signaling dynamics. Similarly, specific kolam motifs with specific meaning, such as the “temple lamp” (see Figure 2A; Waring, 2012b) or auspicious symbols like the swastika (see Figure 1, left; Thomas, 1880) could be selected to measure their signaling value using the triad classification tasks. Our statistical approach and the methods proposed by Bell and Paegle (2021) could complement each other in future investigations of the coordinating roles of other art forms and material culture products. Another promising future avenue would be to expand ethnic marker theory and fieldwork to explain the dynamics between costly investments and ethnic markers. Since complex kolam designs require years of learning and practice, they can be considered costly. Similarly, Polynesian tattoos are markers of identity and political status, but costly due to their permanent and painful nature (Gell, 1993; Schildkrout, 2004). In these ritual examples, the evolutionary dynamics and trajectory of ethnic markers could deviate from current predictions of ethnic marker theory, and thus future research needs to consider costly ethnic markers.

In archaeology and evolutionary anthropology, a substantial amount of research has been dedicated to studying the step-by-step production of lithics (e.g., stone tools) and pottery (chaîne opératoire; Sellet, 1993). Our state-based Markov analysis could be extended and applied to investigate the technical processes and operational sequences involved in the production of material culture. Stylistic variations in kolam art and other artistic traditions can transcend specific motifs and patterns, and manifest themselves in methodological and technical processes of the art production. For instance, kolam artists can systematically differ in their use of rice powder or chalk or their sequence of loop additions with the potential aid of applying scaffolding techniques. According to Gosselain (1999), different technological stages in pottery production can be associated with different stylistic displays of groups of Bafia potters in Cameroon. Thus, different sequences or stages of the production can further reflect stylistic expressions along ethnic or cultural boundaries due to shared learning. In the context of art, methodological and technological differences, such as brushstroke, pigment or contouring sequences and technique could be investigated by describing the different behavioral sequences as states in a Markov system to elucidate whether stylistic variations follow cultural and ethnic boundaries. We believe that the potential application of our state-based Markov approach to material culture and art are vast and could lead to compelling new insights into the evolutionary importance and coordinating role of art and other cultural artifacts.

Using a state-based Markov approach to systematically study the link between styles or patterns in art and group affiliations, our findings inform discussions on the evolutionary role of art for group coordination by encouraging researchers to use systematic methods to measure the mapping between specific objects or styles onto groups.

We show on a Tamil artistic tradition case study that although art can be a medium of communication, it is not necessarily dominated by group affiliation. While distinctive styles or patterns of artistic design are not apparently linked to caste boundaries, neighborhoods, or previous migration, they are linked to artist-level variation, such as expertise and presumably preferences. Our findings corroborate our understanding that artistic traditions and behavior in this context are primarily a domain where individuals carve out a unique niche for themselves to differentiate themselves from others.

The kolam data and code for this study are available and can be found on GitHub: http://github.com/nhtran93/kolam_coordination. The R package to analyze kolam drawings can be further found on GitHub: http://github.com/nhtran93/kolam.

The studies involving human participants were reviewed and approved by the Max Planck Institute for Evolutionary Anthropology and the University of California, Davis. Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

N-HT and ŠK designed the analysis. N-HT, TMW, and BAB wrote data processing software. N-HT wrote the statistical software, conducted the analysis, and wrote the manuscript. ŠK reviewed the code. TMW collected the data and designed the kolam lexicon. SA led the data transcription team. BAB and TMW provided critical feedback. All authors provided edits and revisions.

This work was supported by US NSF Doctoral Dissertation Research Improvement Grant #0823416 and the Max Planck Institute for Evolutionary Anthropology.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.