Grammars of yoga: an exploratory text analytics approach to classical, pedagogical, scholarly, and scientific corpuses of yoga

Introduction: Originating in Indian contemplative and musical traditions, yoga has transformed into a globalized set of psychophysiological and psychospiritual practices. Its worldwide spread has generated diverse formats and frameworks within which it is understood, including classical texts, modern practice manuals, medical research developing yoga-based interventions, and the academic field of yoga studies. However, it is unclear to what degree, if at all, this multiplication of formats and frameworks of yoga leads to a convergence in the articulation of the practices and aims of yoga between these different epistemic standpoints of conceptualizing yoga. Our paper addresses this question by outlining an inductive and computational approach to provide an empirical map of the grammars of yoga of these four corpuses of yoga.

Methods: Two hundred text extracts (50 per source) were compiled from translated classical scriptures, recent biomedical research articles, modern yoga practice manuals, and academic yoga studies publications. Each extract comprised a locatable “yoga is…”-style statement plus surrounding context for natural language processing. We cleaned and tokenized the data, then applied both an unsupervised LDAMallet topic model and a semi-supervised GuidedLDA model seeded with anchor terms, allowing the topics to emerge inductively from the corpus rather than from predetermined categories. The number of topics was selected using coherence-score sensitivity analysis. We visualized results with pyLDAvis and t-SNE and tested them statistically using chi-squared analysis and bias-reduced multinomial logistic regression to assess associations between topics and source types.

Results: Text analytics produced four well-separated, stable topics in our sample of 200 texts, each dominated by vocabulary with a suggested thematic orientation: (1) interventionist/outcome-focused, (2) experiential/phenomenological, (3) genealogical/historical, and (4) soteriological/contemplative. Statistical testing confirmed strong, non-random associations between each topic and a specific corpus, with predicted assignment probabilities exceeding 0.7, and named and weighted the specific terms distinguishing each grammar.

Discussion: The presence of stable semantic signatures revealed signals of a latent sufficiently distinctive grammar for each of the four corpuses of yoga. The strong association between the source type and grammar showed that despite the vast heterogeneity across the corpuses of yoga, there are still four identifiable semantic patterns in which its practices and aims are framed. We discuss the stability of distribution as the result of the possible canalizing effects that the institutional location of the yoga practice has on the way in which its practices and goals are framed, even as their might be a more open-ended borrowing of themes and ideas at the level of actual practice. In providing a preliminary empirical validation of this pattern, our paper provides an avenue for future work that aims to create a dialogically plural and mutually enriching field of yoga studies and practice across its different institutional sites.