1. The Wrong Question

The usual answer to “which is older, Tamil or Telugu?” compares two dates: Tamil-Brahmi in the late third or early second century BCE, and the Telugu record at Kalamalla around 575 CE. The comparison is accurate. The conclusion drawn from it is not, because both dates measure entry into writing, not birth.

The Tamil date is Iravatham Mahadevan’s conservative one, set on palaeography and the stratigraphy of inscribed pottery (Mahadevan 2003). Earlier radiometric dates from Porunthal and Kodumanal have been proposed and remain disputed. The Telugu date is the Kalamalla inscription of the Renati Chola king Dhanañjaya, the first long record written wholly in Telugu.

An inscription needs far more than a language. It needs a script, trained scribes, a patron, durable material, and an institutional judgement that this language is fit for this stone. A language can exist for centuries while failing that last test.

Four different dates are routinely collapsed into one:

The fallacy is to read T_E as T_G. A comparison of two epigraphic dates is then mistaken for a comparison of two lineages.

This paper is written from the Telugu side of the question, and does not pretend otherwise. Its claims are restricted to those that survived attack, recorded in section 14.

It does not claim that Telugu is older than Tamil. It claims that the stones cannot rank them, and that the evidence for a large, long-differentiated Telugu speech community hidden beneath a Prakrit archive is strong.

2. Genealogy First: Where Telugu Sits in the Dravidian Tree

The principal comparative and Bayesian models examined here separate the Tamil and Telugu lineages long before either is written. They differ on how long before.

In Bhadriraju Krishnamurti’s classification, Telugu belongs to South Dravidian II, with Gondi, Konda, Kui, Kuvi, Pengo and Manda (Krishnamurti 2003). Tamil belongs to South Dravidian I, with Kannada, Malayalam, Tulu, Kodagu, Toda and Kota. Telugu is therefore not a daughter of Tamil, and no Tamil text is an ancestral text of Telugu.

Krishnamurti defines South Dravidian II by innovations its members share and South Dravidian I lacks. They include initial apicals and consonant clusters produced by metathesis and vowel contraction, distinct oblique stems in the personal pronouns, and a different non-animate plural suffix. Shared innovations of this kind require a period when the ancestors of Telugu, Gondi and Kui formed one speech community, apart from the ancestors of Tamil and Kannada.

His relative chronology comes from borrowing. Some Indo-Aryan loans entered undivided South Dravidian I before Tamil and Kannada separated, and the branch “of which Telugu is an offshoot” had already left that community. The numeral for ‘thousand’ shows the pattern: Tamil āyiram and Kannada sāvira trace to Indo-Aryan sahasra, while Telugu keeps its native vēyi.

In calendar terms, Krishnamurti’s estimates, as summarised by Franklin Southworth, put the South Dravidian I/II split as complete by about 1100 BCE, and the Tamil–Kannada split by about 600 BCE (Southworth 2006). Southworth cautions that linguistic evidence alone says little about absolute chronology. The order of events, however, rests on shared innovations, not on dates.

The computational evidence pushes the split deeper, not shallower. Kolipakam and colleagues ran a Bayesian analysis of first-hand basic vocabulary from 20 Dravidian languages (Kolipakam et al. 2018). They recover the four main subgroups and date the family’s root to about 4,500 years ago, with a mean of 4,650 and a 95% interval of roughly 3,000 to 6,500 years.

Their best-supported tree does not join South I and South II. Its first split separates South I from all the other branches, at the root itself, and the unconstrained analysis recovers the same split. The traditional South I + South II grouping appears in only 42 of 18,000 sampled trees, and the traditional topology returns only when the analysis is forced to include it.

The authors stress that the higher-order structure is poorly resolved. For this question, though, the uncertainty runs in one direction. If South I and South II are not sisters, Tamil and Telugu meet only at Proto-Dravidian.

One methodological detail of the Bayesian study deserves emphasis. Its authors used first attestations only as lower bounds. South I could not be younger than 2,250 years because Tamil is recorded from the third century BCE, and Telugu’s divergence was given a floor of 1,400 years.

The model encodes this paper’s principle exactly. An inscription bounds a lineage’s age from below. It never dates its birth.

3. Sister Lineages Begin Together

In a family tree, the two daughters of a single split become distinct at the same instant. Any difference in their first inscriptions therefore measures something other than their age as lineages.

Here A and B are sister lineages arising at node N. Their epigraphic dates are later events, set by scribes, patrons and states. Nothing in descent fixes the gap between them.

The claim needs stating carefully. A phylogenetic split dates the beginning of two diverging descendant trajectories. It does not date either daughter from its first surviving text. The antiquity of ‘Tamil’ and ‘Telugu’ as historically recognisable languages depends on later diagnostic innovations, but in no model does epigraphic priority alone fix that chronology.

Applied to Tamil and Telugu, the symmetry at the node is exact. At the node where they part — Proto-South Dravidian in Krishnamurti’s tree, Proto-Dravidian in the best Bayesian tree — pre-Telugu became distinct from pre-Tamil at the same moment pre-Tamil became distinct from pre-Telugu.

Ranking languages by their first inscriptions produces absurdities. On that rule Sanskrit is younger than Tamil.

The question “which is older?” is ill-posed. What comparative linguistics can order is events, and on Krishnamurti’s reconstruction they run as follows:

1. The South Dravidian II branch, ancestral to Telugu, separates; complete by c. 1100 BCE.
2. Centuries later, the Tamil and Kannada lineages separate; complete by c. 600 BCE.
3. Tamil is inscribed, in the late third or early second century BCE.
4. Telugu is inscribed as a matrix language, c. 575 CE.

The stones put Tamil before Telugu. The genealogy, on this reconstruction, puts the separation of the Telugu branch before the Tamil–Kannada split. By the time a lineage distinct from Kannada existed to be called Tamil, the branch that carries Telugu had been on its own for centuries.

That is not a claim that Telugu is older than Tamil. It is a demonstration that the order of inscriptions and the order of genealogy can run in opposite directions, and on Krishnamurti’s reconstruction, for this pair, they do.

4. Controls: Kannada, Tulu, Gondi, Sanskrit — and Malayalam

The cleanest test of the inference is to apply it to languages whose genealogy is not in dispute. It fails every time, except where genealogy independently says the split was late.

Kannada is Tamil’s own sister inside South Dravidian I. Their lineages separated at one node, by about 600 BCE on Krishnamurti’s estimate. Yet the first full Kannada inscription, at Halmidi, is conventionally placed in the mid-fifth century CE, some six and a half centuries after Tamil-Brahmi (Mahadevan 2003); later datings proposed for Halmidi only widen the gap.

No linguist infers that Kannada is six centuries younger than Tamil. The lag is shared by Kannada and Telugu, which sit on different branches of the tree but in the same Upper-South region. A lag shared across branches but confined to one region points to a regional cause.

Tulu is the first language to split from South Dravidian I in Krishnamurti’s tree, and again, though less securely, in the best-supported Bayesian tree (Kolipakam et al. 2018). Its earliest inscriptions are medieval, from the twelfth to the fifteenth centuries (Tulu). Read epigraphically, the most divergent lineage of Tamil’s own branch would be among its youngest.

Gondi, Kui and Kuvi are Telugu’s sisters inside South Dravidian II. Southworth notes that most such non-literary Dravidian languages were not written at all before the twentieth century (Southworth 2006). Read epigraphically, they would be modern languages, which no one believes.

Sanskrit is the decisive control from outside the family. Its oldest texts were composed and transmitted orally from the late second millennium BCE, yet Sanskrit stays rare in inscriptions for centuries after Brahmi appears. Its first major, extensive literary inscription is Rudradāman’s eulogy at Junagadh, around 150 CE, though shorter Sanskrit records are earlier.

Mahadevan attributes that delay to a priestly tradition invested in oral transmission. Whatever the cause, the epigraphic rule would make Sanskrit some four centuries younger than Tamil. The rule is falsified by the most prestigious language of ancient India.

Malayalam is the counter-case, and it matters. Its first records date from the ninth century, and its separation from Tamil is genuinely late; Kolipakam et al. calibrate it at about 1,000 years before present. Here the late inscription and the late split coincide, because the comparative evidence independently says so.

The method follows directly. Genealogy is read from the comparative method, never from stones; only then is the late stone explained. For Malayalam the comparative evidence says late. For Telugu it says early.

5. Vēṅkaṭam and the Land Where the Language Changes

Classical Tamil literature testifies that the land north of Vēṅkaṭam spoke a different language in the very centuries when that land’s inscriptions were Prakrit.

The prefatory verse of the Tolkāppiyam defines the Tamil world by speech: vaṭavēṅkaṭam teṉkumari āyiṭait tamiḻ kūṟum nalluḻakattu, the good world where Tamil is spoken between northern Vēṅkaṭam and southern Kumari. The territory is drawn by language, not by dynasty.

Beyond that line, the poems speak of moḻipeyar tēem. In Kuṟuntokai 11 it lies beyond the good land of Kaṭṭi, at the frontier of the Vaṭukar (Kuṟuntokai). Akanāṉūṟu 67 uses the same phrase (Akanāṉūṟu 61–70).

The compound is precise. Moḻi is speech or language; peyar is to shift or change. The country is named for the change of speech itself, and modern Tamil still uses moḻipeyarppu for ‘translation’. This is not a loose word for foreign parts; it marks a language boundary.

Vaṭukar, literally ‘northerners’, needs more care. Later Tamil uses vaṭuku and vaṭukar for Telugu and its speakers, and one early twentieth-century account of the Pallava plates glosses Andhrapatha as the Tamil Vaḍukavaḻi (early Pallava charters). In Sangam usage, however, the word may also cover Kannada-speaking northerners.

The argument does not need the narrow identification. It needs only the safe claim: early Tamil poets heard a non-Tamil language north of Vēṅkaṭam. Set beside the inscriptions, that claim produces a contradiction, and the contradiction has one resolution.

The inscriptions north of the line are Prakrit. Prakrit was the language of the empire and the monasteries, and Mahadevan rules it out as the people’s speech. Comparative linguistics places South Dravidian II, and Kannada, in exactly that zone; the speech the poets heard was Dravidian, beneath a Prakrit record.

K. Rajan draws this juxtaposition explicitly, in a 2014 lecture for the Archaeological Survey of India. Because parts of Rajan’s early chronology are disputed (section 14), the argument here rests on the primary texts and on Ashoka’s political map, not on his dates.

The paradox this section leaves is the paper’s central question. If the speech north of the line was Dravidian, why is its archive Prakrit? Sections 6 to 9 answer it.

6. Empire Before Vernacular

Brahmi reached Andhra and Tamilakam in the same century. It reached Andhra inside an empire and Tamilakam outside one; that asymmetry is the strongest available explanation for why the script entered the two regions under different linguistic institutions.

Mahadevan dates the arrival of Brahmi in both regions to about the third century BCE, following the southward spread of Jainism and Buddhism (Mahadevan 2003). He then contrasts the outcomes. Tamil literacy began earlier, used the local language for every purpose, and reached all social strata; in the Andhra–Karnataka region, literacy remained an elite skill exercised in Prakrit.

Ashoka’s own inscriptions draw the political line. Rock Edict XIII names the Andhras among the peoples within the king’s domain. Rock Edict II names the Cholas, Pandyas, Satiyaputras and Keralaputras as peoples beyond his borders (Ashoka, Rock Edicts).

The empire was physically present in the Telugu country. Ashokan edicts stand at Yerragudi and Rajula-Mandagiri in Kurnool district. None stands in the Tamil country.

The first writing of the Upper South was carried by imperial personnel. At Brahmagiri, Siddapura and Jatinga-Rameshwara, the scribe Chapaḍa added his signature in Kharoṣṭhī, a script of the north-western frontier (The Caravan). Writing arrived as a package: script, language and institution together.

Mahadevan’s explanation is political. Absorbed into the Nanda–Mauryan domain, the region was governed in Prakrit, which then became the language of the local elite, of learning and of public discourse. The pattern persisted under the Satavahanas, Ikshvakus, Kadambas, Śālaṅkāyanas, Vishṇukuṇḍins and Pallavas.

He offers two parallels from Indian history itself: Persian under the Mughals, and English under British rule. In each, a governing elite held a prestige language as the medium of power while the population spoke something else.

The Tamil kingdoms faced no such package. Mahadevan observes that the Jain and Buddhist monks who entered Tamilakam learned Tamil in order to preach, much as later Christian missionaries learned Indian languages. The script was borrowed; the language was not.

The chancery habit outlived the empire. The earliest Pallava charters, the Mayidavolu and Hirahadagalli plates of the early fourth century CE, are in Prakrit (early Pallava charters). The Mayidavolu grant was issued from Kanchi to the governor at Dhaññakaḍa (Amaravati), for a village in Andhrapatha.

Later Pallava grants, about a dozen of them, are in Sanskrit. These are decisions of a chancery. They are not census returns.

7. Abundantly Literate, in Someone Else’s Language

Early Andhra was not a land without writing. It was among the most inscribed regions of early historic South India, and almost all of it was in Prakrit.

The stūpa sites of the Krishna valley — Bhaṭṭiprolu, Amarāvatī, Jaggayyapeṭa, Nāgārjunakoṇḍa — and Salihuṇḍam to the north yield a large corpus of Prakrit donative and royal records. Mahadevan notes that the early inscriptions of the Upper South on stone, copper plates, coins, seals and pottery are all in Prakrit, and that inscribed pottery there comes mostly from religious centres such as Amaravati and Salihundam (Mahadevan 2003).

That distribution is diagnostic. In Tamilakam, inscribed sherds turn up in hamlets as well as ports, scratched by ordinary people in Tamil. In Andhra, writing on pots clusters around monasteries, in the monasteries’ language.

The absence of Telugu is therefore not an absence of literacy. Writing was present, skilled and abundant. What was absent was sustained institutional adoption of the local vernacular as the documentary matrix.

Bhaṭṭiprolu sharpens the point. Its relic-casket inscriptions, of roughly the third to first centuries BCE, use a Brahmi variant that discards the inherent vowel: a bare consonant stands for the consonant alone, and a following short a receives its own mark (Bhaṭṭiprolu script, after Salomon 1998). Early Tamil-Brahmi made the same move.

The device suits Dravidian phonology, in which words commonly end in consonants. The device is hard to explain from Prakrit phonotactics alone and is more naturally motivated by a Dravidian phonological environment. On Salomon’s account, which must be checked against his text and page references before publication, the system was devised for a Dravidian language and then reapplied to Prakrit.

Which Dravidian language shaped it cannot yet be settled; contact with the Tamil-Brahmi tradition is one possibility. Either way, the scribes of the Krishna delta worked within a Dravidian phonological world while their caskets speak Prakrit. That is documentary underrepresentation caught at the moment of writing.

8. What Leaked Through

A vernacular under a prestige archive surfaces first in what cannot be translated away: names, places, local terms and stray morphology. That is exactly where early Telugu surfaces.

Mahadevan, the leading authority on Tamil-Brahmi, is categorical: “It cannot be that Prakrit was the spoken language of Upper South India…” (Mahadevan 2003). His proof is the large number of Telugu and Kannada personal names and place names inside the region’s early Prakrit inscriptions.

He adds that Dravidian linguistics had already established Telugu and Kannada as separate languages well before the early historic period. On his reading, the language of the documents and the language of the people were simply different things.

The Satavahana court’s own literature leaks. Hāla’s Gāthāsaptaśatī, a Prakrit anthology of around the first century CE, is reported by Mahadevan to contain about thirty Telugu words. That count is not used as evidence here. Separating Telugu from inherited Dravidian, Kannada and Indo-Aryan material needs item-by-item checking against Burrow and Emeneau, Turner and a critical edition. Appendix A lists the candidate forms with a first-pass grading.

At Amarāvatī a single label reads nāgabu. The DHARMA edition dates it to the second century CE and analyses it as Sanskrit nāga- ‘snake’ with the suffix -mbu, which it describes as characteristically Telugu on tatsama words (DHARMA). Rama Prasad Chanda, its first editor, took it as probably a stonemason’s name.

The two readings do not cancel each other. If it is a word, it is Telugu morphology in the Prakrit heart of the delta. If it is a mason’s name, it is an artisan’s name with Telugu-shaped morphology, which is precisely the category this paper predicts.

At Keesaragutta, near Hyderabad, a label of around the early fifth century reads tolucuvāṇḍru (Keesaragutta). Its ending -vāṇḍru is the Old Telugu human plural. Its stem is plausibly tolucu, ‘to bore, to hollow out’, which would make it a label for the people who cut the rock.

The Keesaragutta reading comes here from a secondary web source. Before it carries weight it needs the inscription’s publication, number, date, exact reading and the editor’s grammatical analysis. If it survives that check, morphology of this kind is far more diagnostic than any single vocabulary item.

Later epigraphy makes the same pattern explicit. Sheldon Pollock describes a strict division of linguistic labour, which he calls hyperglossia, in which Sanskrit carried the expressive and political voice of inscriptions while local language handled documentary business (Pollock 2006, reviewed). In bilingual charters, the vernacular enters through the operative clauses: the land, the boundaries, the conditions.

Telugu’s first footholds match that prediction by category: names, labels, a suffix, Dravidian words in a court anthology. None of these is a language learning to speak. Each is a spoken language showing through a written one.

9. Tamil Under the Same Filter

The clearest demonstration that a documentary language need not be its writers’ language comes from Tamil itself. Wherever Tamil speakers wrote inside a Prakrit or Sanskrit institution, Tamil too went underground.

At Anuradhapura, near the Abhayagiri stūpa, a rock inscription in Prakrit records a terrace of the Dameḍa gahapati, the Tamil householders, built by a Dameḍa samaṇa, a Tamil monk (Tamil Householders’ Terrace, after Paranavitana 1970). Paranavitana placed it before the Common Era. Other early Sri Lankan Prakrit records name a Tamil merchant, Visākha, and other Tamil donors.

These were Tamil speakers, contemporaries of the Tamil-Brahmi writers at home. Inside the island’s Buddhist documentary world they wrote Prakrit. Their Tamil survives in the record only as an ethnonym, Dameḍa: a name leaking through.

The Pallava court supplies a second test, inside the Tamil country. The earliest Pallava charters, of the fourth and fifth centuries, are in Prakrit and then Sanskrit. The Pallankoyil plates of Simhavarman, of about the sixth century, are reported to be the first Pallava document in which the royal order is in Tamil (Pallavan copper-plates). One discussion notes that these plates may have been re-issued in the eighth century, so their date and wording need checking against the published editions (early Pandya copper plates). Pandya charters such as the eighth-century Vēḷvikkuḍi grant are likewise bilingual, in Sanskrit and Tamil (Velvikudi inscription, after Tieken 2001).

By then Tamil had been written for some eight centuries. Its entry into royal charters through the grant itself is consistent with the leakage account, though that needs confirming edition by edition.

These cases are instructive because the spoken language is beyond dispute. No historian concludes that Kanchipuram stopped speaking Tamil when its kings wrote Sanskrit, or that Tamil traders at Anuradhapura spoke Prakrit at home. The same inference should not be denied to Andhra.

10. A Population, Not a Void

Documentary silence could, in principle, mean a small or scattered population. For early Andhra the external sources say the opposite.

Megasthenes, the Greek envoy at the Mauryan court around 300 BCE, is preserved through Pliny. He describes the Andarae as a powerful people with numerous villages and thirty walled towns, whose king could field 100,000 infantry, 2,000 cavalry and 1,000 elephants (Megasthenes, via Pliny). The same source credits the king of Kalinga with 60,000 infantry, 1,000 horse and 700 elephants.

Greek figures are secondhand and almost certainly rounded, so they are used here only for relative scale. On that scale an envoy at the Mauryan court ranked the Andhras above Kalinga, the kingdom whose conquest Ashoka himself recorded in Rock Edict XIII as a catastrophe of war. A people that a foreign observer could set beside the great powers of the subcontinent was no scatter of hamlets. Whatever language its chanceries wrote, its population was large enough to be counted in armies.

Ashoka’s Rock Edict XIII then names the Andhras as a people within his realm. The Satavahana and Ikshvaku states, the city of Dhaññakaḍa, and the monuments of Amarāvatī and Nāgārjunakoṇḍa complete the picture.

Mahadevan draws the conclusion directly. Telugu and Kannada, he writes in substance, were spoken by large and settled populations in well-organised states, and there is no reason to think their oral traditions were poorer than Tamil’s on the eve of Tamil literacy (Mahadevan 2003).

One further fact sits in the background. Southworth proposes that the most economical location for the last stage of Proto-Dravidian unity is the lower Godavari basin, because all three main subgroups meet in eastern central India (Southworth 2006). The proposal rests on present-day distribution and is not settled; Kolipakam et al. treat the homeland as unknown.

If Southworth is right, the Telugu country is not a late periphery of the family but close to its geographic centre. The argument of this paper does not depend on it.

11. External Controls: Roman Britain and Medieval England

The configuration proposed for early Andhra is documented in places where nobody disputes the spoken language. In both, the archive and the population spoke different languages for centuries.

Roman Britain is the closest match. The Roman Inscriptions of Britain project counts written evidence that is mainly Latin, with some Greek, one Palmyrene, one Coptic, one Punic, and four British Celtic texts (RIB, Language). British Celtic was the language of most of the province’s people.

Continuous Celtic survives only on a handful of curse tablets from Bath and Uley. Yet of the roughly 150 personal names on the Bath tablets, about half are Celtic, though not all necessarily British (Common Brittonic). Matrix language: Latin. Names: Celtic. That is the leakage hierarchy in a single assemblage.

Medieval England repeats the pattern with fuller documentation. Latin was the principal language of record throughout the medieval period, and vernacular words entered Latin documents either Latinised or whole, often flagged with the article le (DMLBS).

The Dictionary of Medieval Latin from British Sources records borrowings attested in Latin centuries before they appear in vernacular writing. Paul Brand notes that the English common-law courts kept their records in Latin while proceedings were spoken in Anglo-Norman French (DMLBS abstracts).

No historian dates English from the year it displaced Latin in the rolls. The same restraint is owed to Telugu. A Telugu word inside a Prakrit charter is evidence that Telugu was spoken, not that Telugu was being invented.

The controls establish three principles independently of South India. Spoken and written prevalence can diverge by orders of magnitude; the divergence can last for centuries; and the vernacular leaks first through names and local terms.

12. 575 CE Is a Threshold, Not a Birth

The Kalamalla inscription marks the year a court in the Telugu country let Telugu onto the stone. It does not mark the year Telugu began.

Kalamalla, in Kadapa district, records a royal gift of the Renati Chola king Dhanañjaya, styled Erikal Mutturāju, around 575 CE (Telugu Chodas). It is the first long record written wholly in Telugu. Telugu records of the same dynasty follow at several sites in the region.

Before it, Telugu appears as trace: nāgabu at Amarāvatī, tolucuvāṇḍru at Keesaragutta, names inside Prakrit and Sanskrit grants. After it, the language moves quickly into new domains. The Addanki inscription of about 848 CE carries the earliest surviving Telugu verse, in taruvōja metre, nearly two centuries before Nannaya’s Mahābhāratamu (Old Telugu).

Pollock distinguishes literization, a language’s first commitment to writing, from literarization, its use for literature. Both gaps are institutional, and Kannada shows the same profile: Halmidi in the fifth century, the Kavirājamārga in the ninth. Mahadevan adds that earlier literary works in both languages are known from references but do not survive.

There is an irony in the debut. The Renati Cholas claimed descent from Karikāla, the Sangam-age Tamil king. The first dynasty to put Telugu on stone claimed a Tamil ancestor.

The detail matters to the argument, not only to its colour. Telugu’s documentary debut was a political decision by a particular court, not a linguistic event. And the two histories this paper separates were never sealed from each other.

13. The Tolkāppiyam Moment

The inscriptional gap has a literary twin: Tamil codified its language in a grammar and a canon about a thousand years before Telugu did. The gap is real. Read with the discipline this paper applies to stones, it confirms the thesis rather than weakening it.

Tamil has phases too. Tamil’s record is not one ancient block but a sequence: Tamil-Brahmi inscriptions from about the second century BCE, the Sangam anthologies dated conservatively from around the turn of the era and compiled later, and the Tolkāppiyam, itself a layered text. Linguists divide the language into Old Tamil to about 700 CE, Middle Tamil to about 1600, and Modern Tamil after that (Tolkāppiyam).

The Tolkāppiyam has three books and about 1,610 sūtras: on letters and sounds, on words, and on poetic subject matter. Proposals for its date once reached back to 5320 BCE; scholarly debate now lies between roughly the third century BCE and the eighth century CE.

Within that range the text is read in layers. Kamil Zvelebil placed its nuclear portions in the second or first century BCE, hardly before 150 BCE, with a final redaction around the fifth century CE. Takanobu Takahashi puts the oldest layer in the first or second century CE and the last in the fifth or sixth; H. S. David argued in 1952 for distinct chronological strata within its first book.

Mahadevan’s epigraphic test is the sharpest. The Tolkāppiyam treats the puḷḷi, the dot that marks a pure consonant, as integral to the script, and that mark becomes common in Tamil inscriptions only after the second century CE. Tamil’s first grammar therefore describes a script stage some centuries younger than Tamil’s first writing.

Tamil shows, inside its own history, the decoupling this paper describes for Telugu. Its first writing, its first surviving grammar and the final redaction of that grammar are separated by centuries. Antiquity is a sequence of thresholds, for Tamil as for Telugu.

The codification was cosmopolitan in method. The prefatory verse praises its author as versed in the Aindra school of Sanskrit grammar; Zvelebil traced sūtras that render Patañjali’s Mahābhāṣya; Hartmut Scharfe found the imprint of the Vedic Prātiśākhyas and thought the author likely a Jain. Tamil grammatical thought engaged Sanskrit grammatical science, as Telugu grammar would again a millennium later.

Myth is not chronology, on either side. Tamil tradition tells of three academies lasting 4,440, 3,700 and 1,850 years, of Agastya’s primordial grammar, the Agattiyam, lost to the sea, and of Tolkāppiyar as his disciple. The first full account appears in Nakkīrar’s commentary on the Iṟaiyaṉār Akapporuḷ, around the eighth century, and scholars reject its flood chronology for want of any evidence (Legendary Tamil Sangams).

Telugu tradition has its own version. The Āndhra Śabda Cintāmaṇi, credited to Nannaya, was supposedly thrown into the Godavari by a jealous rival and recovered by Appakavi in the seventeenth century through a dream of Vishnu. Kētana, writing in the thirteenth century, shows no knowledge of it and presents himself as the first to frame rules for Telugu.

Velcheru Narayana Rao and David Shulman read Appakavi’s account as “a mythology of grammar itself,” projected back onto the first poet (Classical Telugu Poetry). The structure resembles the Tamil legend: a sacred founder, a lost book, a miraculous recovery. Neither is a date.

Strip both myths and the comparison becomes historical. Tamil’s grammar is older than Telugu’s by roughly eight to fourteen centuries, depending on which layer of the Tolkāppiyam is meant, not by the ten thousand years of the academies. Telugu’s first securely dated grammar is Kētana’s Āndhra-bhāṣā-bhūṣaṇamu of the thirteenth century (Mulaghatika Ketana).

Codification and the prestige language. Call the decisive event a Tolkāppiyam moment: the point at which a court-patronised poetic culture receives a theorised grammar and poetics in its own language, and becomes teachable and canonical. Zvelebil read the Tolkāppiyam as a manual for bardic poets, its first two books leading to the third, on composition; Mahadevan names the bardic tradition first among the causes of early Tamil literacy.

At roughly the period in which early Tamil poetic traditions become recoverable, the Satavahana-associated Deccan also produced an important lyric anthology, the Gāthāsaptaśatī, in Maharashtri Prakrit. George Hart argued that the themes and techniques of the southern poets were absorbed into Prakrit and Sanskrit poetry such as the Sattasaī and Kālidāsa (Hart 1975). The contrast raises the possibility that courtly literary patronage in the Deccan was channelled through a prestige language even where Dravidian vernaculars were locally spoken.

That is a hypothesis about patronage, not a claim that the anthology belongs to Telugu. Its language is Prakrit; what the hypothesis asks is why the Deccan’s earliest courtly canon was not in a Dravidian language, and whether the answer is the same institutional filter that shaped its inscriptions.

The vernacular moments of the Upper South came late, and together. Kannada’s came with the Kavirājamārga at the Rashtrakuta court in the ninth century; Telugu’s with Nannaya’s Mahābhāratamu under the Eastern Chalukya Rājarāja Narendra in the mid-eleventh, with Kētana’s grammar two centuries later. The pattern is regional again, not genealogical.

Was the missing moment the cause of the wait? In part. Codification follows literary practice; it does not create it. Telugu verse was inscribed at Addanki around 848, Nannaya wrote two centuries before Kētana’s grammar, and much Sangam poetry is older than the final layers of the grammar that describes it.

But codification works as a ratchet. Once a vernacular has a grammar, a poetics and a canon, it becomes teachable, prestigious and self-reproducing; Tamil’s commentary tradition, from Iḷampūraṇar onward, kept that loop turning for centuries. Telugu entered the same loop only in the eleventh century.

A causal order consistent with the evidence is therefore:

1. A political regime places a prestige language in the chancery and the monastery.
2. Documentary writing follows it, and the vernacular is filtered (sections 6–9).
3. The courtly literary slot follows it too: Prakrit under Hāla, then Sanskrit eulogy.
4. Without that slot there is no vernacular canon to codify, and so no Tolkāppiyam moment.
5. Without codification there is no ratchet, and the delay compounds.

The missing Tolkāppiyam moment may be the literary face of the same filter that kept Telugu off the stones. It measures when a court let the language of its people become a language of learning. It does not measure the age of the language.

14. Robustness Tests and Alternative Explanations

Each claim above was put against the strongest objection available. What follows records what survived, what was narrowed, and what was dropped.

1. Prakrit was the people’s language, and Telugu spread in later. Mahadevan rejects this outright, on the evidence of Telugu and Kannada names inside the Prakrit record. The Dravidian words reported in the Satavahana anthology (Appendix A) and Bhaṭṭiprolu’s orthography point the same way. Survives.
2. Rajan’s chronology is disputed. Harry Falk argues that some marks read as early writing are non-linguistic megalithic graffiti, and early radiometric dates for Tamil-Brahmi are contested (Tamil-Brahmi). The Vēṅkaṭam argument uses neither; it rests on Sangam texts and Ashoka’s edicts. Survives; Rajan is cited only as secondary.
3. Kolipakam et al. find weak support for Krishnamurti’s subgrouping. Correct, and it cuts toward this paper: without the South I + II clade, Tamil and Telugu meet only at Proto-Dravidian. Absolute dates remain approximate in every model discussed here. Survives.
4. Earlier Tamil-Brahmi dates would widen Tamil’s lead. The thesis is invariant to the start date. A longer Tamil lead also lengthens the gap between Tamil and its sister Kannada, which strengthens the decoupling of stones from genealogy. Survives.
5. Nāgabu is late and ambiguous. DHARMA dates it to the second century CE, and its reading is disputed. On either reading it is predicted material, but it is not used as a cornerstone. Narrowed.
6. Tamil is the more archaic language. Archaism measures rate of change, not age. Icelandic is more conservative than Norwegian; both descend from Old Norse, and neither is older. Telugu’s innovations are those of its branch, and innovation is not youth. Survives.
7. Vaṭukar may mean Kannada speakers. Conceded. The argument uses only the claim that non-Tamil speech was recognised north of Vēṅkaṭam. Narrowed.
8. Bhaṭṭiprolu’s Dravidian orthography may reflect Tamil contact. Conceded as open. Either way, the delta’s scribal milieu was Dravidian while its documents were Prakrit. Survives in weaker form.
9. Telugu literature begins late. So does Kannada’s, and Sanskrit’s inscriptional record begins later still relative to its age. Telugu verse is inscribed at Addanki nearly two centuries before Nannaya. Survives.
10. Prakrit also appears in the Tamil country. It does: a seal at Arikamedu, gold rings at Karur, a few sherds at Kanchipuram and Arikamedu, all urban or port settings, per Mahadevan. That confirms the filter is institutional everywhere, and strongest where institutions were imported. Survives.
11. The Tolkāppiyam moment is circular: a grammar presupposes a literature. Partly conceded. Codification follows practice, so section 13 keeps it as a ratchet downstream of the political filter, not as the root cause. Narrowed.
12. The ‘Telugu words’ of the Gāthāsaptaśatī are mostly pan-Dravidian. Partly conceded. Graded item by item in Appendix A, several are pan-Dravidian, tuppa is shared with Tamil and Kannada, and one may be Indo-Aryan; poṭṭa and words like it still place the court in Upper-South Dravidian speech and exclude Tamil. Narrowed.
13. Classical army figures are unreliable. Conceded. Greek numbers are secondhand and probably rounded, so they serve only as relative scale. The conclusion needs only that the Andhras were a large, organised people, which the walled towns, Ashoka’s edict and the stūpa sites also show. Narrowed.
14. The Keesaragutta reading and the Pallankoyil plates rest on thin sourcing. Conceded. Both are flagged for checking against primary editions and neither carries the argument; the Sri Lankan, Roman and English cases stand without them. Narrowed.

Dropped from this version, because they did not survive checking or do not bear on the question:

• That Telugu is older than Tamil. No evidence used here supports it, and the symmetry argument makes it ill-posed.
• Telugu words in an Ashokan edict, sometimes asserted in popular sources. Unverified.
• Pre-Ashokan dates for Telugu material at Bhaṭṭiprolu. Contested.
• Andronov’s glottochronological date for Telugu. Krishnamurti criticised the data behind it.
• The ‘Trilinga’ etymology of Telugu as evidence of antiquity. Speculative.
• A ‘descendant test’ from Prakrit to Indo-Aryan heirs in coastal Andhra. Too crude: Indo-Aryan language history is more complicated than the replacement of populations.

15. The Documentary Leakage Hypothesis: A Research Programme

The account above is testable. Its core can be stated as a single hypothesis that a corpus can confirm or falsify.

Here V is identifiable vernacular material. For early Andhra the hypothesis makes six predictions.

1. Category clustering. In Prakrit and Sanskrit records from Andhra, 300 BCE to 650 CE, Telugu-shaped material concentrates in names, places, water bodies, boundaries and local titles.
2. Continuity across regime change. The same toponymic and morphological traces persist through the switch from Prakrit to Sanskrit, because the population did not change.
3. Institutional timing. Telugu’s arrival as a matrix language tracks changes in courts and chanceries, such as the Renati Cholas, not any break in the language.
4. A mirror image in Tamilakam. Tamil-Brahmi shows Tamil as the matrix, with Prakrit confined to names, religious terms and loans.
5. The Sri Lankan control. Tamil traces in early Sri Lankan Prakrit records concentrate in ethnonyms and personal names.
6. The British control. Roman Britain shows the same hierarchy, with Celtic names common and Celtic text nearly absent.

Three findings would weaken the hypothesis. Telugu-shaped material spread evenly across formal prose rather than clustered in names would count against it, as would a break in local toponymy at the Prakrit-to-Sanskrit switch. So would widespread secular Prakrit graffiti by ordinary people at rural Andhra sites, comparable to Tamil’s inscribed pottery.

The corpus already exists in part. The DHARMA project’s editions of early Andhra and Telugu inscriptions can serve as its base. Each record would be coded for matrix language, date range, dynasty, religious setting and coordinates, with every name and local term graded for certainty of identification.

The decisive statistic is the rate of vernacular material by category and period. If the rate in names greatly exceeds the rate in formal prose, and holds steady across the Prakrit-to-Sanskrit transition, the documentary account of early Telugu moves from argument to measurement.

16. Conclusion

Tamil reached the stone first. That fact stands, and nothing here diminishes it. What does not stand is the inference built on it: that the order of the stones is the order of the languages.

Comparative linguistics separates the Tamil and Telugu lineages in prehistory, under the principal models examined here. The symmetry of descent means the two lineages begin diverging together, whatever their first inscriptions say. Kannada, Tulu, Gondi and Sanskrit show, one after another, that first inscriptions do not track genealogy.

The cause of Telugu’s late stone is visible in the record itself. Ashoka counted the Andhras inside his realm and the Tamil kings outside it. Writing came to Andhra with an empire’s language, and stayed with the monasteries and chanceries that followed.

The country beneath that archive was large, organised and Dravidian-speaking. Its names, its suffixes and its scribes’ orthography leak through the Prakrit, and Tamil poets of the same centuries heard its speech across Vēṅkaṭam. The same filter silenced Tamil itself wherever Tamil speakers wrote inside imported institutions.

The books tell the same story as the stones. Tamil codified its speech in a grammar within a few centuries of first writing it; the Deccan spent its anthology moment on Prakrit, and Telugu’s literary canon began with Nannaya nearly a millennium later. The legends on both sides — Agastya’s lost grammar, Nannaya’s drowned one — are myths of origin, not dates.

The Kalamalla inscription of c. 575 CE is therefore a threshold. It records the moment a court allowed the language of its people onto stone.

The question worth asking is no longer which language is older. It is what kept an old and populous speech community out of its own archive for nearly eight centuries, and how much of it can still be recovered from what leaked through.

Appendix A. Candidate Dravidian Forms in the Gāthāsaptaśatī

Not used in the argument. This is a first-pass grading drawn from secondary sources, checked against Burrow and Emeneau’s DEDR where noted. Each form must be re-verified against DEDR, Turner’s CDIAL and a critical edition before any of it is cited. No claim in the main text depends on this table.

Sources for the list: Telugu Wikipedia, Gāthā Saptaśatī; cognates from Wiktionary, poṭṭa and Wiktionary, tuppa. This grading is a first pass; the standard checks are Burrow and Emeneau’s Dravidian Etymological Dictionary and Turner’s Comparative Dictionary of the Indo-Aryan Languages.

Graded this way, the list still does real work. It places the Satavahana court’s Prakrit inside an Upper-South Dravidian speech world — poṭṭa above all — and it shows why an aggregate count like ‘about thirty Telugu words’ needs this sieve before it can carry weight.