Three Buddhist Vocabularies: Computational Stylometry of the English Pali Canon across Sutta, Vinaya, and Abhidhamma

Joy Bose

arxiv: 2606.25372 · v1 · pith:AUUCIXQ6 · submitted 2026-06-24 · cs.CL · cs.IR

Three Buddhist Vocabularies: Computational Stylometry of the English Pali Canon across Sutta, Vinaya, and Abhidhamma

Joy Bose This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-25 21:20 UTCgrok-4.3pith:AUUCIXQ6 record.json open to challenge →

classification cs.CL cs.IR

keywords stylometrypali canonzipf lawlexical diversitybuddhist textsvocabulary overlaptipitakacomputational linguistics

0 comments

The pith

English translations of the Pali Canon's three Pitakas follow Zipf distributions but the Abhidhamma shows distinctly higher lexical diversity than the Sutta and Vinaya.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper computes Zipf rank-frequency plots, moving-average type-token ratios, numeral densities, and vocabulary overlaps on large English translations of the Sutta, Vinaya, and Abhidhamma sections. All three divisions produce strong Zipf fits, yet the Abhidhamma corpus deviates most from the ideal slope and registers markedly higher lexical diversity and numeral use. The Sutta and Vinaya sections yield nearly identical diversity scores. Overlaps between Theravada and other Vinaya traditions, as well as between different translations of the same text, are also quantified. A reader would care because these metrics supply quantitative signatures of the distinct textual functions each Pitaka serves within the canon.

Core claim

All examined corpora obey Zipf-consistent rank-frequency distributions with R-squared values above 0.989; the Vinaya lies closest to the ideal slope of -1 while the Sangaha corpus deviates most, placing the word 'consciousness' at rank 8. MATTR-500 lexical diversity is statistically indistinguishable between the Theravada Sutta and Vinaya (0.399 and 0.400) yet rises to 0.560 for the Sangaha, a difference preserved under size-controlled subsampling. The Sangaha also exhibits the highest numeral-word density at 3.26 percent, consistent with its enumerative character. Cross-tradition and cross-translation vocabulary overlaps are reported as point estimates without significance tests.

What carries the argument

Zipf rank-frequency distributions fitted by ordinary least squares together with Moving Average Type-Token Ratio (MATTR-500) computed on segmented English translations of the Tipitaka.

If this is right

The Mulasarvastivada Vinaya shares 20 percent Jaccard vocabulary and 49 percent overlap coefficient with the Theravada Vinaya, indicating measurable continuity of legal terminology across traditions.
Two English translations of the same Vinaya source share only 24 percent of their vocabulary, with diagnostic shifts such as 'musing' versus 'absorption' for jhana.
The Sangaha's elevated numeral density directly tracks its systematic listing of mental and material categories.
Size-controlled subsampling confirms that the Sangaha's higher MATTR value is not an artifact of corpus length.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The observed lexical signatures could be used to test whether other systematic Abhidharma compendia across Buddhist traditions exhibit comparable elevations in diversity and numeral use.
Translation-divergence measurements supply a baseline for quantifying how doctrinal terminology drifts when rendered into new languages over long periods.
Extending the same metrics to non-Theravada Abhidharma literature might reveal whether the higher diversity pattern is genre-specific rather than school-specific.

Load-bearing premise

The frequency and overlap statistics obtained from the selected English translations meaningfully reflect properties of the underlying Pali source texts rather than translator-specific word choices.

What would settle it

Re-running the same Zipf and MATTR pipelines on the original Pali texts or on a fresh set of independent translations of the same Pitaka sections and obtaining substantially different exponents or diversity scores.

read the original abstract

We present a computational stylometric analysis of the Tipitaka across all three Pitakas in English translation, extending earlier work on the Sutta Pitaka alone. The corpus spans 134,831 segments from Bhikkhu Sujato's Sutta Pitaka (114,591 segments, CC0), Bhikkhu Brahmali's Vinaya Pitaka (7,923 segments, CC0 2026), I.B. Horner's 1938 Vinaya translation (2,826 segments), three English translations of the Abhidhammattha Sangaha compendium (2,077 segments), and cross-tradition Vinaya texts from the Dharmaguptaka and Mulasarvastivada schools. We compute Zipf rank-frequency distributions with OLS-fitted exponents, Moving Average TTR (MATTR-500), numeral-word density, and vocabulary overlap (Jaccard and Szymkiewicz-Simpson coefficients). Main findings: (1) all corpora show Zipf-consistent distributions (R2 > 0.989); the Vinaya is closest to ideal Zipf slope -1 and the Sangaha corpus deviates most, with 'consciousness' displacing grammatical particles at rank 8; (2) MATTR-500 shows the Sutta and Vinaya Theravada are nearly identical in lexical diversity (0.399 and 0.400), while the Sangaha corpus is genuinely more diverse (0.560), confirmed by size-controlled subsampling; (3) the Sangaha corpus has the highest numeral-word density (3.26%), consistent with its systematic enumeration of mental and material categories; (4) the Mulasarvastivada Vinaya shares 20.0% vocabulary (Jaccard) and 49.1% (overlap coefficient) with the Theravada Vinaya, reflecting shared legal heritage across two millennia; (5) two English translations of the same Vinaya source text share only 24.2% of their vocabulary across 88 years, with 'musing' versus 'absorption' for jhana and 'defeat' versus 'expulsion' for parajika as the most diagnostic shifts. All results are point estimates; no significance testing is conducted. Code and data are released as open-source extensions to the Darshana Graph corpus (arXiv:2606.18222).

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This is a descriptive extension of stylometric metrics to the full English Pali canon with one useful diagnostic on translation variance, but the core comparisons are undercut by translator effects and lack of testing.

read the letter

The main takeaway is that the work adds Vinaya and Abhidhamma sections to prior Sutta-only stylometry and reports concrete numbers like 3.26% numeral density in the Sangaha and 24.2% vocabulary overlap between two Vinaya translations. It also includes cross-school Vinaya overlap figures.

What stands out is the direct comparison of two English Vinaya versions of the same source, which flags how much lexical choice shifts the token sets. The code and data release as an extension of the Darshana Graph corpus is a clear positive.

The soft spots are straightforward. All the Zipf, MATTR, and overlap claims come from different translators (Sujato for Sutta, Brahmali for one Vinaya, Horner for another), and the paper's own 24.2% overlap result shows those choices move the statistics substantially. Without significance tests or error bars, the near-identical MATTR scores between Sutta and Vinaya (0.399 vs 0.400) are hard to interpret as anything beyond the translators' habits. The size-controlled subsampling helps a bit but does not fix the translation layer.

This is niche material for readers already working on Buddhist corpora or digital humanities applications to religious texts. It does not claim theoretical advances and stays within standard metrics.

I would send it to peer review. The data release and the translation diagnostic give it enough substance to justify referee time, even if the interpretation section needs tightening around what can actually be attributed to the Pali originals.

Referee Report

2 major / 2 minor

Summary. The manuscript conducts a stylometric analysis of the English translations of the Pali Canon, comparing Zipf rank-frequency distributions (OLS-fitted exponents with R² > 0.989), moving-average type-token ratio (MATTR-500), numeral-word density, and vocabulary overlap (Jaccard and Szymkiewicz-Simpson) across Sutta (Sujato), Vinaya (Brahmali and Horner), Abhidhammattha Sangaha (multiple translations), and cross-tradition Vinaya texts. Key claims include Zipf consistency with Vinaya closest to slope -1 and Sangaha deviating most (with 'consciousness' at rank 8), near-identical lexical diversity between Sutta and Vinaya (0.399 vs. 0.400) versus higher in Sangaha (0.560, confirmed by size-controlled subsampling), elevated numeral density in Sangaha (3.26%), 20.0% Jaccard overlap between Mulasarvastivada and Theravada Vinaya, and only 24.2% vocabulary overlap between two translations of the same Vinaya source.

Significance. If the differences can be shown to reflect properties of the Pali source texts rather than translator-specific choices, the work provides quantitative evidence of stylistic variation across the three Pitakas and demonstrates the applicability of standard corpus metrics (Zipf, MATTR, overlap coefficients) to Buddhist textual corpora. The open release of code and data as extensions to the Darshana Graph corpus is a clear strength for reproducibility.

major comments (2)

[Abstract] Abstract, finding (2): The claim that Sutta and Vinaya Theravada corpora have nearly identical lexical diversity (MATTR-500 of 0.399 and 0.400) reflecting underlying Pali structures is load-bearing for the central comparison yet directly contradicted by finding (5), which reports only 24.2% vocabulary overlap between two English translations of the identical Vinaya source text; this demonstrates that translator lexical preferences (e.g., 'musing' vs. 'absorption') substantially alter token sets and derived statistics.
[Abstract] Abstract, findings (1) and (3): Attribution of Zipf slope differences, the rank-8 displacement of 'consciousness', and the Sangaha's elevated numeral density (3.26%) and diversity (0.560) to properties of the original texts rather than translation artifacts lacks support, as the same translator-variation effect shown for Vinaya applies across all corpora translated by different individuals (Sujato, Brahmali, Horner, and Sangaha translators).

minor comments (2)

The manuscript should explicitly state in the methods or discussion section how segment boundaries and tokenization were handled across translations to allow readers to assess potential confounds.
[Abstract] All reported quantities are point estimates with no error bars or significance tests; while acknowledged, adding bootstrap intervals or permutation baselines would strengthen the presentation of differences such as the Sangaha deviation.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their detailed and insightful comments, which highlight a key interpretive challenge in attributing observed differences to source texts versus translation choices. We address each major comment below and will make targeted revisions to strengthen the manuscript's claims and limitations discussion.

read point-by-point responses

Referee: [Abstract] Abstract, finding (2): The claim that Sutta and Vinaya Theravada corpora have nearly identical lexical diversity (MATTR-500 of 0.399 and 0.400) reflecting underlying Pali structures is load-bearing for the central comparison yet directly contradicted by finding (5), which reports only 24.2% vocabulary overlap between two English translations of the identical Vinaya source text; this demonstrates that translator lexical preferences (e.g., 'musing' vs. 'absorption') substantially alter token sets and derived statistics.

Authors: We agree that finding (5) demonstrates substantial translator influence on vocabulary sets and that this creates interpretive tension with the attribution in finding (2). MATTR-500 is a windowed diversity metric that can remain stable under lexical substitutions, unlike Jaccard overlap, but we cannot rule out translator effects without same-translator controls. We will revise the abstract to remove the phrase 'reflecting underlying Pali structures' and replace it with 'as observed in the respective English translations,' add an explicit limitations paragraph discussing translator variation as a potential confound, and note that the Sutta-Vinaya similarity is an empirical observation in the translated corpora rather than a direct claim about the Pali originals. revision: partial
Referee: [Abstract] Abstract, findings (1) and (3): Attribution of Zipf slope differences, the rank-8 displacement of 'consciousness', and the Sangaha's elevated numeral density (3.26%) and diversity (0.560) to properties of the original texts rather than translation artifacts lacks support, as the same translator-variation effect shown for Vinaya applies across all corpora translated by different individuals (Sujato, Brahmali, Horner, and Sangaha translators).

Authors: The referee is correct that the manuscript compares corpora produced by different translators and that finding (5) shows this can affect derived statistics. The results are therefore properties of the specific English translations examined. For the Sangaha, consistency across its three translations provides some internal check, but no parallel same-translator versions exist for the full set of Pitakas. We will revise the abstract and discussion sections to frame all attributions as observations in the translated corpora, add a dedicated limitations subsection on translation artifacts, and qualify claims about 'original texts' to 'potentially reflecting properties of the source texts, subject to translator effects.' revision: partial

standing simulated objections not resolved

Absence of same-translator parallel translations for the Sutta, Vinaya, and Abhidhamma corpora prevents definitive isolation of source-text effects from translator-specific lexical choices.

Circularity Check

0 steps flagged

Minor self-citation to prior corpus work; all stylometric quantities computed directly from token counts with no reduction to fitted parameters or self-referential definitions

full rationale

The paper computes Zipf exponents via OLS, MATTR-500, Jaccard/Szymkiewicz-Simpson overlap, and numeral density directly from the segmented English token streams. These are standard descriptive statistics applied to the provided translations; none are defined in terms of each other or obtained by fitting a parameter whose value is then re-reported as a 'prediction'. The only self-reference is to the Darshana Graph corpus (arXiv:2606.18222) as the data-release vehicle and to an earlier Sutta-only analysis; neither supplies a load-bearing uniqueness theorem, ansatz, or fitted input that forces the headline comparisons. The 24.2% overlap between Vinaya translations is itself a direct Jaccard computation and does not create circularity in the reported diversity or Zipf results.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The analysis applies standard tokenization, frequency counting, and set-overlap formulas to a fixed corpus; no new entities are postulated and the only free parameters are the conventional window size of 500 for MATTR and the OLS fit for Zipf, both of which are method choices rather than data-fitted constants required by the claims.

axioms (2)

domain assumption English word tokens after standard preprocessing are a sufficient proxy for lexical properties of the source texts
Invoked when computing all frequency, diversity, and overlap statistics across translations
standard math Zipf's law is expected to hold for natural-language corpora of this size
Used to interpret the fitted exponents and R2 values as 'consistent'

pith-pipeline@v0.9.1-grok · 5980 in / 1706 out tokens · 26199 ms · 2026-06-25T21:20:17.485885+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

8 extracted references · 2 canonical work pages · 1 internal anchor

[1]

Bodhi, Bhikkhu (Ed.). (2007). A comprehensive manual of Abhidhamma: The Abhidhammattha Sangaha of Acariya Anuruddha. Buddhist Publication Society

2007
[2]

Bose, J. (2026a). Darshana graph: A parallel commentary corpus for comparative Indian philosophy, with stylometric and exploratory graph analyses. arXiv:2606.18222

work page internal anchor Pith review Pith/arXiv arXiv
[3]

Brahmali, Bhikkhu. (2026). Theravada Collection on Monastic Law (5th ed., 6 vols.). SuttaCentral. CC0. https://suttacentral.net/editions/pli-tv-vi/en/brahmali

2026
[4]

A., and McFall, J

Covington, M. A., and McFall, J. D. (2010). Cutting the Gordian knot: The moving -average type-token ratio (MATTR). Journal of Quantitative Linguistics, 17(2), 94 -100. https://doi.org/10.1080/09296171003643098

work page doi:10.1080/09296171003643098 2010
[5]

Horner, I. B. (1938). The book of the discipline (Vinaya -Pitaka), Vol. 1: Suttavibhanga. Pali Text Society

1938
[6]

Sukwitthayakul, C., and Thongrin, S. (2022). Understanding the Pali Canon through keyword analysis: A comparison between different reference corpora. Rangsit Journal of Social Sciences and Humanities, 9(2), 34-45

2022
[7]

Sujato, Bhikkhu. (2021). Sutta Pitaka translations. bilara -data. SuttaCentral. CC0. https://github.com/suttacentral/bilara-data

2021
[8]

Zigmond, D. (2020). Toward a computational analysis of the Pali Canon. R package tipitaka v0.1.1. https://CRAN.R-project.org/package=tipitaka Appendix A: Numeral Word List The following 32 terms were used to compute numeral density. Cardinal numbers: one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, fifteen, si...

2020

[1] [1]

Bodhi, Bhikkhu (Ed.). (2007). A comprehensive manual of Abhidhamma: The Abhidhammattha Sangaha of Acariya Anuruddha. Buddhist Publication Society

2007

[2] [2]

Bose, J. (2026a). Darshana graph: A parallel commentary corpus for comparative Indian philosophy, with stylometric and exploratory graph analyses. arXiv:2606.18222

work page internal anchor Pith review Pith/arXiv arXiv

[3] [3]

Brahmali, Bhikkhu. (2026). Theravada Collection on Monastic Law (5th ed., 6 vols.). SuttaCentral. CC0. https://suttacentral.net/editions/pli-tv-vi/en/brahmali

2026

[4] [4]

A., and McFall, J

Covington, M. A., and McFall, J. D. (2010). Cutting the Gordian knot: The moving -average type-token ratio (MATTR). Journal of Quantitative Linguistics, 17(2), 94 -100. https://doi.org/10.1080/09296171003643098

work page doi:10.1080/09296171003643098 2010

[5] [5]

Horner, I. B. (1938). The book of the discipline (Vinaya -Pitaka), Vol. 1: Suttavibhanga. Pali Text Society

1938

[6] [6]

Sukwitthayakul, C., and Thongrin, S. (2022). Understanding the Pali Canon through keyword analysis: A comparison between different reference corpora. Rangsit Journal of Social Sciences and Humanities, 9(2), 34-45

2022

[7] [7]

Sujato, Bhikkhu. (2021). Sutta Pitaka translations. bilara -data. SuttaCentral. CC0. https://github.com/suttacentral/bilara-data

2021

[8] [8]

Zigmond, D. (2020). Toward a computational analysis of the Pali Canon. R package tipitaka v0.1.1. https://CRAN.R-project.org/package=tipitaka Appendix A: Numeral Word List The following 32 terms were used to compute numeral density. Cardinal numbers: one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, fifteen, si...

2020