Skip to content

Chapter 14: Encounters and Continuities: From Colonial Philology to Computational Grammar

The Sanscrit language, whatever be its antiquity, is of a wonderful structure; more perfect than the Greek, more copious than the Latin, and more exquisitely refined than either, yet bearing to both of them a stronger affinity, both in the roots of verbs and in the forms of grammar, than could possibly have been produced by accident; so strong indeed, that no philologer could examine them all three, without believing them to have sprung from some common source, which, perhaps, no longer exists.

— William Jones, Third Anniversary Discourse, 2 February 1786

14.1 Introduction

The Indian sciences of language did not end with the last Sanskrit commentaries. From the late 18th century European scholars, working with Indian pandits, made Sanskrit grammar known in Europe, where it helped to shape comparative and descriptive linguistics. In the 20th century Pāṇini became an object of study for theoretical linguists, and in the 21st Sanskrit has become a field for computational linguistics. The Persian scholarship of Chapter 13 is the immediate background to the colonial philology described here: Jones learned Persian before Sanskrit, the East India Company governed in Persian until 1837, and the first British grammars of Hindustani were built on the work of Indian munshis trained in Persian learning. This chapter surveys these encounters and corrects some popular myths that have grown up around them.

14.2 Indian Grammar beyond India

Indian grammatical learning travelled with Buddhism and with Sanskrit literary culture.

14.2.1 Tibet

Tibet received Indian grammar more fully than any other land beyond India. Tibetan tradition holds that the script was created in the 7th century by Thon mi Sambhoṭa, a minister of King Srong btsan sgam po who was sent to India to learn writing, on the model of an Indian script of the Gupta type (Section 12.7). To him are also attributed the two foundational treatises of Tibetan grammar, the Sum cu pa (“The Thirty”) and the rTags kyi ’jug pa (“The Application of Signs”), short verse texts on the letters of Tibetan, the particles that mark grammatical relations and the use of the prefixed letters. Modern scholars doubt the attribution and date the texts later, though how much later is disputed. The treatises apply Indian categories to a language of a quite different family: the Tibetan particles are arranged under the eight cases of Sanskrit grammar, and the letters are classed as “masculine”, “feminine” and “neuter”, terms borrowed from Sanskrit grammar and given a new, phonetic sense.

The translation of the Buddhist canon brought Sanskrit grammar itself to Tibet. The Cāndra and Kātantra grammars and their commentaries were translated into Tibetan and included in the collection of commentarial works (bsTan ’gyur), and they were studied in the monasteries for centuries, while Pāṇini’s own grammar was less studied (Verhagen 1994). The translators also standardised their vocabulary. The Mahāvyutpatti, a Sanskrit–Tibetan glossary compiled in the early 9th century under royal patronage, fixed the Tibetan equivalents of thousands of Buddhist terms and is one of the earliest bilingual dictionaries in Asia; a companion work, the sGra sbyor bam po gnyis pa, set out the principles of translation and explained many terms by analysing their Sanskrit derivation in the manner of the nirukta tradition. Tibetan grammar became a scholarly discipline of its own, and its later commentators, such as Si tu Paṇ chen Chos kyi ’byung gnas in the 18th century, continued to read the old treatises in the light of Sanskrit grammar.

14.2.2 China, Southeast Asia and Sri Lanka

In China, Buddhist translators wrestled with the differences between Sanskrit and Chinese; the pilgrim Yijing left a description of how Sanskrit grammar was taught in India in the 7th century, and Chinese scholars’ encounter with Sanskrit phonetics contributed to the development of Chinese phonological analysis. In Southeast Asia, Sanskrit inscriptions composed in correct classical style from the 5th century onwards, in Cambodia, Java and elsewhere, show that grammatical training of the Indian kind was available at royal courts far from India (Pollock 2006). In Sri Lanka and Burma, Pali grammar flourished in its own right (Chapter 9).

14.3 The European Encounter with Sanskrit

14.3.1 Missionaries and pandits

European missionaries in India studied Sanskrit from the 17th century, beside the vernaculars whose first grammars they wrote (Section 9.11), and the first Sanskrit grammar printed in Europe was published at Rome in 1790 by the Carmelite Paulinus a Sancto Bartholomaeo. In 1767 the French Jesuit Gaston-Laurent Coeurdoux sent to Paris a memoir pointing out resemblances between Sanskrit, Greek and Latin. The British scholars of the Asiatic Society of Bengal, founded in Calcutta in 1784, learned Sanskrit from Indian pandits, whose knowledge of the grammatical tradition was the basis of everything they achieved; the names of many of these teachers are poorly recorded (Trautmann 1997).

14.3.2 William Jones, 1786

In his “Third Anniversary Discourse” to the Asiatic Society on 2 February 1786, the judge and orientalist William Jones made the observation, quoted at the head of this chapter, that is usually taken as the starting point of comparative Indo-European linguistics (Jones 1788).

Jones did not claim that Greek and Latin came from Sanskrit; he claimed that all three came from a common source. That remains the view of linguistics, and it disposes of the popular belief that Sanskrit is the “mother of all languages”: the Dravidian, Munda and Tibeto-Burman languages of India, like most of the world’s languages, are not related to it at all (Section 1.8), and its Indo-European relatives are its sisters. The idea that Sanskrit is the parent of the European languages, held by some early 19th-century writers such as Friedrich Schlegel, was abandoned as comparative method developed. Sanskrit, Greek, Latin, the Germanic, Celtic, Slavic, Iranian and other Indo-European languages are sister branches descended from Proto-Indo-European, a language reconstructed by comparison but never written (Masica 1991).

The discovery also had a darker afterlife. The linguistic term “Aryan” was turned in 19th-century Europe and India into a racial category, with consequences ranging from colonial theories of caste to European racism. Friedrich Max Müller, who popularised the term, later insisted that it referred to language and not to race; the confusion proved hard to undo (Trautmann 1997).

The traffic in words also ran the other way. Through trade, administration and scholarship, English took in hundreds of words from Indian languages: from Sanskrit or its descendants pundit, guru, avatar, karma, jungle and loot; from Hindustani bungalow, shampoo and pyjamas; from Tamil and Malayalam curry and catamaran. Henry Yule and A. C. Burnell’s glossary Hobson-Jobson (1886) collected such words with their histories, and is itself a document of the colonial encounter.

14.3.3 Colebrooke, Böhtlingk and the reception of Pāṇini

H. T. Colebrooke’s Grammar of the Sanskrit Language (1805) and his essays introduced the Indian grammatical tradition to Europe; Otto Böhtlingk published the first European edition of the Aṣṭādhyāyī in 1839–40. European Sanskritists were divided in their attitude to Pāṇini. Some, like the American W. D. Whitney, criticised the native grammar as artificial and preferred to describe Sanskrit from the texts; others, increasingly, recognised in it a model of descriptive rigour.

14.4 Comparative Philology

Comparative Indo-European linguistics was built in the half-century after Jones, largely in Germany and Scandinavia, and Sanskrit was its touchstone (Table 14.1).

Table 14.1. Some founders of comparative Indo-European linguistics
Scholar Date Contribution
Friedrich Schlegel 1808 Über die Sprache und Weisheit der Indier; spoke of “comparative grammar”
Rasmus Rask 1818 Systematic sound correspondences between Germanic and other languages
Franz Bopp 1816; 1833–52 Comparative study of the verb; Vergleichende Grammatik
Jacob Grimm 1822 Statement of the Germanic consonant shift (“Grimm’s Law”)
Hermann Grassmann 1863 Explained apparent exceptions to Grimm’s Law by dissimilation of aspirates in Sanskrit and Greek
August Schleicher 1860s Family-tree model; reconstructed Proto-Indo-European forms
Karl Verner 1875 Explained further exceptions by the position of the Indo-European accent, preserved in Vedic
Ferdinand de Saussure 1878–79 Mémoire on the Indo-European vowel system; later the founder of structural linguistics

Sanskrit was central to this work because it is attested early and preserves many archaic features, including, in the Vedic language, the position of the Indo-European accent. But it is not always the most conservative branch: its vowel system, for example, merged the Indo-European e, o and a into a, so that Greek or Latin often preserve older vowels, and Hittite, deciphered in 1915–17, preserves consonants (the “laryngeals” predicted by Saussure) that Sanskrit has lost (Tables 14.2 and 14.3).

Table 14.2. Some Sanskrit words and their Indo-European cognates
Sanskrit Greek Latin Gothic Meaning
pitṛ patēr pater fadar father
bhrātṛ phrātēr (clansman) frāter brōþar brother
nāman onoma nōmen namō name
asti esti est ist is
trayaḥ treis trēs þreis three
nava ne(w)os novus niujis new
veda “I know” oida — wait I know
deva — deus, dīvus — god (Greek theos is not related)
Table 14.3. Grimm’s Law: Indo-European stops in Sanskrit, Latin and English
Sanskrit Latin English Change in Germanic
pad- ped- foot p \(\rightarrow\) f
trayaḥ trēs three t \(\rightarrow\) th
śvan- canis hound k \(\rightarrow\) h (Sanskrit ś from Indo-European palatal k)
daśa decem ten d \(\rightarrow\) t
jānu genu knee g \(\rightarrow\) k
bhrātṛ frāter brother bh \(\rightarrow\) b

Indian terms also entered the vocabulary of the new science: the European study of Indo-European vowel gradation adopted the words guṇa and vṛddhi from the Sanskrit grammarians, and the phonetic analysis of the Prātiśākhyas influenced European phoneticians (Chapter 2). It is not correct, however, to say that the Neogrammarians’ principle of exceptionless sound laws was derived from Pāṇini, whose grammar is synchronic and contains no theory of sound change.

Other Sanskrit grammatical terms passed into general linguistics and are still used for languages that have nothing to do with India. Sandhi is the standard name for sound changes at the junctions of words and morphemes in any language. The compound types of the Indian tradition supplied the terms dvandva (coordinative compounds, as in English bittersweet), tatpuruṣa (determinative compounds, as in bookcase) and bahuvrīhi (possessive or exocentric compounds, as in redcap for a porter, “one who has a red cap”). The terms were taken over because the Indian classification of compounds was more detailed than the one that European grammar had inherited from Greek and Latin.

14.5 Describing the Languages of India

The comparative method was applied to the languages of India themselves, first and most famously in the demonstration of the Dravidian family (Section 11.7). The Munda languages were recognised as a separate family, later shown to belong to Austroasiatic, and the languages of the north-east were related to Tibeto-Burman. George Grierson’s Linguistic Survey of India (1903–1928), a colonial undertaking that used thousands of local informants and specimen texts, recorded 179 languages and 544 dialects in British India (Grierson 1903–1928). Its classifications, though often revised, remain a starting point for the study of many Indian languages.

After independence, language became central to politics. The reorganisation of states on largely linguistic lines from 1956, the place of Hindi as the official language of the Union, the Tamil agitations against the imposition of Hindi, culminating in 1965, and the “three-language formula” for schools were all shaped by questions of language and identity (Ramaswamy 1997; King 1994). The Eighth Schedule of the Constitution now lists twenty-two languages. Many smaller languages, especially those of Adivasi communities and of the Andaman Islands, are endangered: the last speaker of Bo, a Great Andamanese language, died in 2010. Documenting and supporting these languages is one of the most urgent tasks of linguistics in India today (Hock and Bashir 2016).

The Constitution also gave the old science of word-formation a new task. Article 351 directs the Union to develop Hindi by drawing for its vocabulary “primarily on Sanskrit and secondarily on other languages”, and in 1961 the Government of India set up a Commission for Scientific and Technical Terminology to coin and standardise technical terms. Much of the resulting vocabulary of administration, science and law in Hindi and other languages was built from Sanskrit roots, prefixes and suffixes, using exactly the derivational resources that Pāṇini had described (Chapter 6). Critics have pointed out that many of these coinages remained unfamiliar to ordinary speakers, who often preferred English or Perso-Arabic words already in use; the debate between a Sanskritised and a more popular Hindi continues.

14.5.1 How many languages?

The question “how many languages are spoken in India?” has no single answer, and the different figures that circulate reflect different methods rather than errors. The Census of India records the “mother tongues” that people report, which run into the thousands, and then groups them under a much smaller number of “languages”; many varieties that their speakers regard as separate languages, such as Bhojpuri, Rajasthani or Chhattisgarhi, are classified under Hindi. Linguists, by contrast, try to group varieties by mutual intelligibility and structural similarity, which yields figures in the hundreds, and recent independent surveys have reached similar orders of magnitude. In the Census of 2011, for example, some 19,500 raw mother-tongue names were reduced to about 1,370 “rationalised” mother tongues, and these were grouped under 121 languages, 22 of them in the Eighth Schedule; the People’s Linguistic Survey of India, a volunteer project led by G. N. Devy in the 2010s, reported about 780 languages. The boundary between a “language” and a “dialect” is in any case partly political, as the old saying that a language is a dialect with an army and a navy suggests. What all the counts agree on is that India is among the most linguistically diverse countries in the world.

14.6 Indian Linguistics in the Twentieth Century

Indian scholars played a growing part in modern linguistics from the early 20th century. Suniti Kumar Chatterji’s The Origin and Development of the Bengali Language (Chatterji 1926) applied the methods of historical linguistics to a modern Indian language with a thoroughness that made it a model for later work; Sukumar Sen wrote histories of Bengali and grammars of Middle Indo-Aryan; S. M. Katre and the Deccan College in Pune became a centre of historical and Sanskrit linguistics, where a long-running project produced an encyclopaedic Sanskrit dictionary on historical principles. Bhadriraju Krishnamurti’s work on the Telugu verb and on comparative Dravidian, and M. B. Emeneau’s on the Dravidian languages of the Nilgiris and on South Asia as a linguistic area, transformed the study of the Dravidian family (Krishnamurti 2003; Emeneau 1956). The Central Institute of Indian Languages, founded at Mysore in 1969, has promoted the study and teaching of Indian languages, including minority and tribal languages.

Traditional Sanskrit scholarship also continued. Pandits in Varanasi, Pune, Chennai, Kerala and elsewhere maintained the oral teaching of Pāṇinian grammar through the Siddhāntakaumudī and the works of Nāgeśa, and their collaboration was essential to modern translations such as those of S. D. Joshi and his colleagues (Joshi and Roodbergen 1986). The meeting of these two traditions, the pandit’s mastery of the texts and the modern linguist’s comparative and theoretical questions, has been one of the most fruitful features of Indian linguistics.

14.7 Beyond Indo-Aryan and Dravidian

Surveys of Indian linguistics often give the impression that the subcontinent’s languages are either Indo-Aryan or Dravidian. The other families are smaller in numbers of speakers but no less important for the history of language in India, and their study has largely been the work of modern linguistics, since few of them had a pre-modern grammatical tradition.

14.7.1 Munda and Austroasiatic

The Munda languages of central and eastern India — among them Santali, Mundari, Ho, Kharia, Juang, Sora and Gutob — belong to the Austroasiatic family, whose other branches include Khasi in Meghalaya, the Nicobarese languages, and Khmer and Vietnamese in Southeast Asia (Anderson 2008). Munda languages have a very different structure from their Indo-Aryan and Dravidian neighbours: verbs can carry markers for subject and object, nouns distinguish singular, dual and plural, and there is a distinction between inclusive and exclusive “we”. Santali, with several million speakers in Jharkhand, West Bengal, Odisha, Bihar, Assam, Bangladesh and Nepal, was added to the Eighth Schedule in 2003 and has its own script, Ol Chiki (Chapter 12). Words of Munda or related origin have been identified in Sanskrit from the Vedic period onwards, a sign of very old contact (Kuiper 1991).

14.7.2 Tibeto-Burman

Tibeto-Burman languages, part of the Sino-Tibetan family, are spoken along the Himalaya and throughout the north-east (Thurgood and LaPolla 2003). Many are tonal. Meitei (Manipuri), the language of the Manipur valley, has a written history in its own script, Meetei Mayek, which was replaced by the Bengali script in the 18th century and revived in the late 20th; it was added to the Eighth Schedule in 1992. Bodo, written today in Devanāgarī, was added in 2003. Mizo and many Naga languages were first written in the Latin script by Christian missionaries in the late 19th and 20th centuries. In Sikkim and the Darjeeling hills the Lepcha people have their own script, and Ladakhi, Sikkimese and other Tibetic languages are written in Tibetan script and share the Tibetan grammatical tradition, which itself adapted Sanskrit grammatical models.

14.7.3 The Andamanese languages and isolates

The languages of the Andaman Islands form two groups that are not demonstrably related to each other or to any other family: Great Andamanese, once a group of about ten languages, now spoken by only a handful of people in a mixed form; and the Ongan languages, Onge and Jarawa. The language of the Sentinelese, who avoid contact, is unknown. The death in 2010 of Boa Sr, the last speaker of Bo, a Great Andamanese language, drew international attention to the danger facing these languages; Anvita Abbi and her colleagues documented Great Andamanese in its last years (Abbi 2006). Isolates are also found on the mainland: Nihali, spoken in a small area on the border of Maharashtra and Madhya Pradesh, and, beyond India’s present borders, Burushaski in the Karakoram and Kusunda in Nepal.

14.7.4 Sign languages

Deaf communities in India use sign languages that are full natural languages with their own grammar, not codes based on spoken Hindi or English. Indian Sign Language, related to the sign language of Pakistan, has been increasingly recognised in education in recent decades, and a national institute for research and training in it was set up in 2015.

14.8 Hindi, Urdu and the Modern Standard Languages

The modern standard languages of India took shape between the 16th and the 20th centuries, and their standardisation drew on several traditions at once. In northern India the speech of the Delhi region (Khaṛī Bolī) became the basis of a literary language whose names, glossaries and first grammars are described in Chapter 13; its earliest substantial literature was produced not in Delhi but at the Deccan courts of Bijapur and Golconda. The prose texts that Gilchrist’s Hindustani department at the College of Fort William commissioned from its Indian staff after 1800, in what came to be called Urdu and Hindi, began the institutional separation of the two registers, written in Perso-Arabic and Devanāgarī scripts respectively. In the course of the 19th century Hindi was increasingly identified with Hindus and Sanskrit vocabulary, Urdu with Muslims and Persian vocabulary, although at the level of everyday speech they remained largely one language (King 1994; Rai 2001). After 1947 Hindi in Devanāgarī became the official language of the Indian Union and Urdu the national language of Pakistan; Urdu is also one of the scheduled languages of India.

Grammarians played a part in all of this. Indian and European authors wrote grammars of Hindustani, Bengali, Marathi, Tamil and the other languages on European models in the 19th century, while Indian scholars such as Kamtaprasad Guru, whose Hindī Vyākaraṇ (1920) became the standard grammar of Hindi, combined European categories with terms from the Sanskrit tradition. Much of the modern technical vocabulary of the Indian languages was coined from Sanskrit roots and suffixes, using the derivational resources described in Chapter 6.

14.9 Pāṇini and Modern Linguistics

14.9.1 Bloomfield and the descriptivists

Leonard Bloomfield, the leading American linguist of the inter-war years, studied Pāṇini closely (his praise of the grammar is quoted in Section 5.18). For Bloomfield the Aṣṭādhyāyī was a model of the complete and explicit description of a language from its own forms, without appeal to the categories of Latin grammar, and his own analysis of morphology drew on it.

14.9.2 Generative grammar and phonology

From the 1960s, generative linguists saw in Pāṇini an ancestor of their own enterprise: a finite system of explicit rules that derives the well-formed expressions of a language. Paul Kiparsky in particular showed that Pāṇini’s principles of rule interaction correspond closely to concepts developed independently in generative phonology, such as the “Elsewhere Condition” (Section 5.8) and the ordering of rule blocks (Kiparsky 1979, 2009). J. F. Staal, George Cardona, S. D. Joshi and J. A. F. Roodbergen made the tradition accessible to modern linguists (Staal 1972; Cardona 1976, 1997; Joshi and Roodbergen 1986). See Table 14.4.

Table 14.4. Pāṇini and modern grammar: parallels and differences
Topic Pāṇini Comment
Explicit rules About 4,000 ordered rules derive the correct forms Shares with generative grammar the aim of explicitness
Rule interaction Exceptions override general rules; the tripādī applies in strict order Closely parallel to concepts of generative phonology
Semantic relations Kārakas mediate between meaning and case Compared with case grammar and thematic roles
Economy Extreme concision (lāghava) Economy of description, not a claim about the mind
Scope One language (Sanskrit), including Vedic and regional variation No theory of universal grammar or of language acquisition
Syntax Derives words and their relations; little phrase structure Unlike phrase-structure grammars, focused on morphology

14.9.3 Comparisons with other modern theories

Pāṇini’s kāraka theory has been compared with dependency grammar, and Bhartṛhari’s views with semantic holism and with theories that language shapes thought (Chapter 8). Such comparisons help modern readers to see what the Indian theorists were doing. They become misleading when they turn into claims that ancient authors “anticipated” modern theories in every respect, or that modern linguists merely rediscovered them. The value of the Indian traditions does not depend on such claims.

14.10 Pāṇini and Computer Science

14.10.1 The “Pāṇini–Backus form”

In 1967 Peter Zilahy Ingerman suggested, in a short letter to the Communications of the ACM, that the notation used to define the syntax of programming languages, known as Backus–Naur Form (BNF), should be called “Pāṇini–Backus Form” in recognition of Pāṇini’s use of a comparable rule notation (Ingerman 1967). The suggestion was not adopted. Backus said that he drew on Emil Post’s production systems, and the notation is closely related to Noam Chomsky’s contemporary work on formal grammars; there is no evidence that it was based on Pāṇini. The parallel is nonetheless real: both systems use a formal metalanguage of rewrite rules to define the well-formed strings of a language. Statements that Peter Naur or Backus acknowledged a debt to Pāṇini are not correct.

14.10.2 Pāṇini’s grammar as a formal system

Table 14.5 describes the Aṣṭādhyāyī in computational terms, as an analogy.

Table 14.5. The Aṣṭādhyāyī described in computational terms (an analogy)
Computational notion Pāṇinian counterpart
Input A meaning to be expressed; a root or stem
Data Śivasūtras, Dhātupāṭha, Gaṇapāṭha
Operations Substitution (ādeśa), deletion (lopa), augment (āgama), suffixation
Control Headings (adhikāra), recurrence (anuvṛtti), interpretive rules (paribhāṣā)
Conflict resolution apavāda, antaraṅga, nitya, para
Output A finished word (pada)

Researchers have built computer implementations of parts of the Aṣṭādhyāyī that apply its rules to generate forms, and in doing so they have had to make explicit many of the conventions that the tradition left implicit. The question of where the grammar stands in the hierarchy of formal languages has been discussed, but has no simple answer, since the grammar presupposes a human user who knows what is to be expressed.

14.11 The “Sanskrit is Ideal for Computers” Myth

A widespread popular claim holds that NASA, or “scientists”, have found Sanskrit to be the best language for computers or for artificial intelligence. The claim is usually traced to an article by Rick Briggs, a researcher at the Research Institute for Advanced Computer Science at NASA’s Ames Research Center, published in AI Magazine in 1985 (Briggs 1985).

What Briggs actually wrote

Briggs’s article, “Knowledge Representation in Sanskrit and Artificial Intelligence”, compared the semantic networks then used in artificial-intelligence research to represent the meaning of sentences with the analyses of sentence-meaning (śābdabodha) given by Sanskrit grammarians and logicians. He argued that these traditional analyses, which paraphrase a sentence as a precise structure of relations between an action and its participants, resemble the knowledge representations of AI, and that this showed that a natural language could in principle serve as a precise means of representation. He did not report any NASA programme to use Sanskrit, and he did not show that Sanskrit was better for computers than other languages.

The popular version of the claim is mistaken on several counts:

  1. NASA has never adopted Sanskrit for programming or for any computing purpose; Briggs’s article was the work of one researcher.

  2. What Briggs admired was not Sanskrit as such but the technical analyses of the grammarians and Naiyāyikas, written in a highly regimented scholastic Sanskrit.

  3. Ordinary Sanskrit is as ambiguous as any natural language. Its poets cultivated ambiguity deliberately (śleṣa, the “double meaning”), and its sandhi and compounds create notorious difficulties for computational analysis (Section 14.12).

  4. Programming languages are artificial formal languages; no natural language, Sanskrit or any other, is used to program computers.

The genuine points of interest — the explicitness of Pāṇini’s grammar, and the precision of the traditional analysis of sentence-meaning — are significant enough without exaggeration.

14.12 Computational Sanskrit Today

Since the 1990s computational linguistics for Sanskrit has become an active field, with regular international symposia from 2007 onwards, whose first proceedings appeared in 2009 (Huet et al. 2009). The work is spread across a small number of long-running projects in India, Europe and the United States (Table 14.6), most of which combine a digital lexicon with rules drawn from the Pāṇinian tradition.

Table 14.6. Some computational projects on Sanskrit
Project Researchers Focus
Sanskrit Heritage Platform Gérard Huet (INRIA, France) Lexicon, morphological generation, sandhi analysis and segmentation
Saṃsādhanī Amba Kulkarni (University of Hyderabad) Morphological analysis, kāraka-based parsing, machine translation
Sanskrit Library Peter Scharf and Malcolm Hyman Digital texts, encoding, grammatical tools
Paninian Grammar Framework Akshar Bharati, Vineet Chaitanya, Rajeev Sangal Kāraka-based analysis of Indian languages (Bharati et al. 1995)
Computational linguistics groups IIT Bombay (Malhar Kulkarni and others); JNU (Girish Nath Jha and others) Rule simulation, Sanskrit–Hindi translation, lexical resources
GRETIL University of Göttingen Archive of electronic texts in Indian languages

14.12.1 Texts, corpora and manuscripts

Computational work depends on digital texts. Archives such as GRETIL and SARIT provide large collections of Sanskrit texts in standard encodings, and the Digital Corpus of Sanskrit, built by Oliver Hellwig, offers a large corpus with morphological and lexical annotation that has been used to study the history of the language quantitatively. Behind the printed and digital texts lies an immense manuscript heritage, only partly catalogued; the National Mission for Manuscripts, set up by the Government of India in 2003, has been documenting manuscripts in public and private collections across the country. The digitisation of manuscripts, and the development of character recognition for the many scripts in which they are written, are among the most important practical tasks for the study of Indian languages.

14.12.2 The problem of sandhi

Because external sandhi is applied in written Sanskrit, a computer, like a human reader, must first split a continuous string into words. The task is highly ambiguous: a given string can often be divided in several grammatical ways. For example, devālayaṃ gacchati is naturally read as devālayam gacchati “(he) goes to the temple”; but devālaya- itself can be analysed either as deva + ālaya “abode of a god” or, less plausibly, as devā + ālaya. Computational tools run the sandhi rules in reverse to generate all possible divisions, then rank them using a lexicon, grammar and statistical information.

Current challenges

  1. Segmentation: sandhi at word boundaries.

  2. Compounds: long compounds with many possible analyses.

  3. Free word order: relations are marked by case endings rather than position.

  4. Ellipsis and context: literary and philosophical texts leave much to be supplied.

  5. Data: digitised and annotated corpora are small compared with those of major modern languages, and much of the manuscript heritage is still uncatalogued.

Large neural language models have recently been applied to Sanskrit, with some success in segmentation and translation; because the available data are limited, approaches that combine statistical methods with the explicit knowledge encoded in Pāṇinian grammar remain attractive.

14.13 Popular Claims and the Evidence

Several claims about the Indian sciences of language circulate widely in popular books, on the internet and occasionally in textbooks. Table 14.7 summarises the most common of them and what the evidence shows; the relevant chapters give the details.

Table 14.7. Common popular claims and the scholarly position
Claim What the evidence shows See
Sanskrit is the “mother of all languages” Sanskrit is one branch of the Indo-European family; most of the world’s languages, including the Dravidian, Munda and Tibeto-Burman languages of India, are unrelated to it This chapter
The European languages descend from Sanskrit Sanskrit, Greek, Latin and the Germanic languages are sisters, descended from unwritten Proto-Indo-European This chapter
NASA found Sanskrit to be the best language for computers One NASA researcher (Briggs, 1985) compared the grammarians’ analyses of meaning with AI knowledge representation; NASA has never used Sanskrit This chapter
BNF in computer science was based on Pāṇini Ingerman (1967) proposed the name “Pāṇini–Backus Form”; Backus did not draw on Pāṇini This chapter
The International Phonetic Alphabet was modelled on Sanskrit phonetics Indian phonetics influenced 19th-century European phonetics indirectly; the IPA is based on the Latin alphabet Ch. 2
Pascal’s triangle and the Fibonacci numbers were “discovered” by Piṅgala The triangle is explicit in Halāyudha (10th c.); the recurrence in Virahāṅka, Gopāla and Hemacandra; both predate Europe, neither is clearly in Piṅgala Ch. 3
The Indus script has been deciphered as Sanskrit (or as Tamil) No decipherment is accepted and the language is unknown; the Dravidian hypothesis (Parpola, Mahadevan) is a serious scholarly proposal, but not a proven reading Ch. 12
The Tolkāppiyam is several thousand years old It is layered, its core probably of the last centuries BCE or early centuries CE Ch. 10
Pāṇini dates to the 6th century BCE or earlier Most scholars place him in the 4th century BCE, some in the 5th Ch. 5
Sanskrit is free of ambiguity Like every natural language it is ambiguous; poets cultivated double meanings, and sandhi and compounds create further ambiguity This chapter

14.14 Sanskrit in the Modern World

Sanskrit today is a scheduled language of the Indian Constitution and is taught in schools, universities and traditional institutions across India and abroad. It is used in religious ritual, in the composition of new poetry and scholarly works, and, on a small scale, in journalism and broadcasting: a daily newspaper in Sanskrit, Sudharmā, has been published from Mysore since 1970, and All India Radio broadcasts news in Sanskrit. Movements to revive spoken Sanskrit have organised short courses for many thousands of people, and a few villages have been publicised as Sanskrit-speaking communities. Census returns show only a small number of people who report Sanskrit as their mother tongue.

These developments raise questions that the sciences of language described in this volume help to frame. A living tradition of Sanskrit learning depends on the grammatical knowledge built up over two millennia; at the same time, the history of the language shows that Sanskrit has always coexisted with, and drawn from, other languages, and that its prestige was bound up with social hierarchies. The most valuable legacy of Śabda Vidyā may lie less in any single language than in its methods: careful observation of speech, explicit description by rule, and sustained argument about meaning.

14.15 Conclusion

The Indian sciences of language — Vedic phonetics and recitation, Pāṇini’s grammar, Bhartṛhari’s philosophy, the theory of metre, the explanation of words, the grammars of Pali, Prakrit, Tamil and the other languages, the scripts and the lexicons, and the Persian scholarship of later centuries — form one of the richest bodies of linguistic thought produced by any civilisation. Their achievements are best appreciated on their own terms: as answers to the questions their authors asked, in the social and religious worlds in which they lived, including worlds in which learning was restricted by caste and gender. They need no inflated claims about priority or modern science to secure their place in the history of human thought. A verse from the Pañcatantra, in which a king’s counsellor reflects on the length of grammatical study, makes a fitting close:

Pañcatantra, introduction (kathāmukha 6 in R. Jha’s edition; wording varies between recensions)

anantapāraṃ kila śabdaśāstraṃ svalpaṃ tathāyur bahavaś ca vighnāḥ
sāraṃ tato grāhyam apāsya phalgu haṃsair yathā kṣīram ivāmbumadhyāt

“The science of words has no end, life is short and obstacles are many. So take the essence and leave the chaff, as swans take milk from the midst of water.”

Further Reading

  • Trautmann, Thomas R. 1997. Aryans and British India. University of California Press.. — On Orientalism, philology and the Aryan idea.

  • Trautmann, Thomas R. 2006. Languages and Nations: The Dravidian Proof in Colonial Madras. University of California Press.. — On the discovery of the Dravidian family.

  • Alam, Muzaffar. 2003. “The Culture and Politics of Persian in Precolonial Hindustan.” In Literary Cultures in History: Reconstructions from South Asia, edited by Sheldon Pollock. University of California Press.. — On Persian in precolonial India.

  • King, Christopher R. 1994. One Language, Two Scripts: The Hindi Movement in Nineteenth Century North India. Oxford University Press.. — On the Hindi–Urdu controversy.

  • Kiparsky, Paul. 2009. “On the Architecture of Pāṇini’s Grammar.” In Sanskrit Computational Linguistics: First and Second International Symposia, edited by Gérard Huet, Amba Kulkarni, and Peter Scharf. Lecture Notes in Computer Science 5402. Springer.. — On Pāṇini and modern linguistics.

  • Staal, J. F., ed. 1972. A Reader on the Sanskrit Grammarians. MIT Press.. — Western and Indian writings on the Sanskrit grammarians, with Staal’s introductions.

  • Briggs, Rick. 1985. “Knowledge Representation in Sanskrit and Artificial Intelligence.” AI Magazine 6 (1): 32–39.. — The article behind the “Sanskrit and computers” myth.

  • Verhagen, Pieter C. 1994. A History of Sanskrit Grammatical Literature in Tibet. Volume 1: Transmission of the Canonical Literature. Handbuch Der Orientalistik, 2.8. Brill.. — On the translation and study of Sanskrit grammar in Tibet.

  • Huet, Gérard, Amba Kulkarni, and Peter Scharf, eds. 2009. Sanskrit Computational Linguistics: First and Second International Symposia, Rocquencourt, France, October 29–31, 2007, Providence, RI, USA, May 15–17, 2008. Revised Selected and Invited Papers. Lecture Notes in Computer Science 5402. Springer.. — The proceedings of the first two symposia on Sanskrit computational linguistics.