Sanskrit for computation: More nationalism than fact

Indian nationalists have long promoted the idea that Sanskrit is not merely an ancient language but a sophisticated computational system that anticipates modern Artificial Intelligence. This belief spread rapidly after a widely shared speech by the late External Affairs Minister Sushma Swaraj, whose praise for Sanskrit’s scientific framework generated substantial television ratings. It inspired a wave of social media posts linking Pāṇini’s grammar with machine learning, voice recognition and Natural Language Processing.

The narrative is attractive. It promises a civilisational edge, it flatters national pride, and it encourages the idea that modern innovation emerges naturally from ancient cultural wisdom. Yesterday, the assertions and rebuttals were all over X. Let's settle the debate once and for all.

A closer look reveals that these claims rest on a mixture of partial truth, selective interpretation and wishful thinking. Sanskrit is a remarkable language, Pāṇini’s Aṣṭādhyāyī is a masterpiece of formal analysis, and the Maheshwara Sutrani offers an elegant phonetic arrangement. None of this is in dispute. Yet the leap from intellectual brilliance to technological superiority reflects a misunderstanding of how modern computing works.

Artificial Intelligence today is driven by statistical learning, vast datasets and billions of parameters, not by human-designed grammatical rulebooks. The belief that Sanskrit is inherently better suited for computing than English or Hindi confuses social history with computational reality.

I bring together linguistic facts, computational principles, facts in information technology, and the evolution of speech technology to explain where the claims hold merit and where they drift into myth.

Sanskrit is a great ancient language that, despite disappearing from our speeches, was not allowed to die, as its words filled all descendants of Prakrit. Depending on the extent of tatsama usage in your mother tongue, you can figure out the meanings of even purely Sanskrit…

— Surajit Dasgupta (@surajitdasgupta) December 3, 2025

Rise of Sanskrit grammar claim

The central nationalist argument begins with Pāṇini. His Aṣṭādhyāyī, composed more than two millennia ago, is indeed the earliest known example of a comprehensive generative grammar. It uses meta-rules, transforms and compressed notation to produce a tightly organised description of Sanskrit. The Maheshwara Sutrani, fourteen short aphorisms describing phonetic classes, support this system by enabling concise abbreviations called Pratyāhāras. These allow large sets of sounds to be referenced by two-letter codes. Scholars often marvel at the resemblance between Pāṇini’s rule-based method and modern formal systems used in computer science.

This resemblance is real, yet the comparison requires context. Pāṇini’s grammar is algorithmic in the sense that it uses ordered rules which generate correct word forms. However, modern computing does not treat languages through such rulebooks. Early computational linguists attempted rule-based parsing in the twentieth century and found that human language, full of ambiguity and inconsistent usage, resisted complete formalisation. Today’s AI does not model English or Hindi as sets of rules. Instead, it uses machine learning to discover patterns automatically from immense quantities of text and speech.

The nationalist argument mistakes historical elegance for technological relevance. Pāṇini’s system is brilliant, but brilliance alone does not solve the challenges that contemporary AI faces.

Sound structure claim, Maheshwara Sutras' allure

Supporters of the Sanskrit–AI thesis often claim that the Maheshwara Sutrani (plural of sutra) create a perfect, physics-based arrangement of sounds that aligns with modern computational logic. The sutras indeed move from open vowels to semi-vowels, nasals, stops, sibilants and the aspirate. The categorisation is precise, and the notational scheme is both compact and powerful.

Yet the argument that this structure gives Sanskrit a computational advantage rests on an assumption that modern AI benefits from phonetic regularity. This assumption is weak. Current ASR systems do not decode text by reading the script or by consulting grammatical classifications. They treat language as a continuous acoustic stream, then map that stream to statistically learned text patterns. Whether a language is phonetic or irregularly spelt does not determine ASR accuracy. English is full of irregular spellings, yet English ASR is highly accurate in controlled conditions. The model does not decode the sound of “read” and consult a dictionary to decide between “reed” and “red”. It predicts the correct word by learning from immense amounts of data.

The strength of English ASR arises from decades of industrial investment, abundant speech datasets and an early culture of widespread technological adoption. This is a sociotechnical advantage, not a linguistic one.

If Sanskrit had millions of recorded hours across diverse accents, spoken environments and natural settings, ASR systems would eventually handle it with comparable accuracy. However, Sanskrit today lacks such data. Its phonetic script may offer a mild advantage when mapping sounds to written forms, yet this barely affects the real difficulties of speech recognition, which include background noise, microphone variation, speed, dialect and co-articulation.

Data abundance reality

Computing and modern speech technology developed predominantly in the United States, which ensured that English accumulated far more resources than any other language. The early internet, digital media archives, open datasets, university projects and commercial applications all generated vast English-language corpora. Speech recognition quality improved because millions of people used the technology, not because English is uniquely suited to AI.

The nationalist narrative often attributes English dominance to unfair Western privilege or cultural bias. A more accurate explanation is simpler. ASR systems improve through real-world usage. The more people speak into phones, video calls and smart assistants, the more the models evolve. English benefited from early access to infrastructure, research funding and a global user base. If Hindi, Tamil, or Bengali had received similar data investment at an early stage, their ASR systems would be equally advanced today.

This principle applies even more sharply to Sanskrit. Few people speak it natively, and most contemporary Sanskrit speech is formal, rehearsed or pedagogical. AI models require natural, spontaneous speech across large populations. Pride in tradition cannot substitute for data diversity.

Generative capacity claim

A common nationalist argument states that Sanskrit’s generative structure makes it ideal for creating new scientific and technical vocabulary. Pāṇini’s rules indeed allow roots and suffixes to form new words systematically. The language can coin neologisms elegantly, without borrowing from English. Saṅgaṇaka for a computer is a neat example.

The issue is not whether Sanskrit can generate new words. Many highly inflected languages, including Latin and Greek, possess similar generative capacity. The question is whether this capacity matters for modern NLP. Contemporary AI does not rely on rulebooks for word formation. Neural models learn new terms through exposure. They do not consult roots or suffixes. They do not follow classical grammar. They adapt statistically, based on contextual patterns.

Thus Sanskrit’s generative grammar is culturally rich but technologically irrelevant. AI does not need Pāṇini to coin new vocabulary. It can generate terms automatically through learned embeddings, just as it does in every other language.

Case marking claim

Sanskrit’s use of Vibhakti endings allows flexible word order. This has prompted some nationalists to argue that Sanskrit is more “logical” than English because sentence relationships are clearly marked on the words themselves. English relies on word order, which means that reversing subject and object reverses meaning. In Sanskrit, Rāmaḥ Rāvaṇam hanti conveys the same meaning as Rāvaṇam Rāmaḥ hanti because the case endings identify the roles.

Sanskrit’s inflectional structure indeed reduces ambiguity in syntactic relations. However, modern NLP models are not rule-based parsers. They do not identify grammatical roles by reading case markers the way a human student of Sanskrit does. Instead, they infer semantic relationships from probabilistic patterns. Models trained on English handle flexible syntax and varied constructions with ease because they learn from context, not from grammar.

Thus, Sanskrit’s case system may delight linguists, yet it offers no special computational advantage under current AI paradigms.

Sandhi claim, tokenisation myth

Another popular claim is that Sanskrit’s Sandhi rules function as perfect mathematical operations, which allow ideal tokenisation. Sandhi describes how sounds merge across word boundaries. The mergers follow predictable patterns, but they also produce long compound words that appear intimidating to modern readers.

Neural models do not require rule-based tokenisation. They break text into subword units, learn merges automatically and optimise the representation based on statistical patterns. Sandhi is therefore neither a barrier nor an advantage. Sanskrit compounds can be tokenised like German or Finnish compounds. Machine learning models do not lose accuracy because of Sandhi, nor do they gain computational efficiency by exploiting Sandhi.

The nationalist argument seeks to elevate Sandhi into a universal tokenisation mechanism. This misrepresents how modern language models work. Statistical tokenisation operates independently of classical grammar.

NLP in modern computing

To understand why Sanskrit’s rule-based beauty does not translate into AI utility, one must understand what NLP actually does. Natural Language Processing refers to the set of techniques that allow computers to read, write, interpret and generate human language. It powers chatbots, translation engines, predictive typing, search queries, sentiment detection, voice assistants and news summarisation.

These systems do not rely on hand-crafted grammatical rules. They rely on machine learning. They learn patterns from enormous quantities of real text and speech. Modern NLP models use neural networks with billions of parameters. They crunch statistical regularities rather than apply explicit syntactic logic. The model that understands English does so not because it knows the rules of English grammar, but because it has seen millions of examples of how words behave.

Therefore, linking Sanskrit’s traditional grammar to modern NLP misunderstands the basic architecture of contemporary AI.

Cultural pride, scientific clarity

None of this diminishes the sophistication of Sanskrit or the intellectual brilliance of Pāṇini. The Aṣṭādhyāyī is one of humanity’s outstanding analytical achievements. The Maheshwara Sutrani demonstrate extraordinary insight into phonetics. Sanskrit carries immense cultural value and remains central to Indian heritage.

What needs correction is the belief that these features grant Sanskrit an inherent technological superiority. AI does not privilege rule-based systems. It privileges scale, data richness, computational power and statistical optimisation. If India wishes to build world-class speech and language technology, it must invest in data collection, annotation, acoustic modelling and multilingual research. Celebrating civilisational heritage is useful for cultural confidence, yet scientific progress demands grounded understanding rather than mythologised claims.

Sanskrit will continue to inspire scholars, and its grammar will remain a treasure of linguistic science. Its future in modern AI will depend not on ancient sutras but on contemporary effort.

If a reader wishes to read the article above more structurally, he/she may refer to my Facebook post or X (formerly Twitter) thread.