What share of the Poona Dictionary's canon does DCS cover?

The Poona Dictionary (PD, An Encyclopaedic Dictionary of Sanskrit on Historical Principles) cites its sources by siglum — a de-facto declaration of the Sanskrit literary canon. The Digital Corpus of Sanskrit (DCS) is the largest lemmatised Sanskrit corpus. This page measures one against the other, for PD's published a- volumes. Full write-up: PD_DCS_CORPUS_COVERAGE_2026.md.

The two headline numbers diverge on purpose. DCS holds only a quarter of PD's citation practice (metric 1) yet its own bulk is 78 % PD-cited (metric 3): DCS is a deep sample of the archaic/classical core PD leans on, not of PD's encyclopedic breadth.

Where PD's citation mass goes

Top residue — PD works DCS does not hold

The purāṇas, the lexicographic tradition (kośa), classical kāvya, and the grammatical commentary layer. The single largest item (Padmapurāṇa, 3,506) outweighs all but three covered texts.

Covered DCS texts, by PD citation frequency

Trust Block

• Evidence. PD sigla harvested from pd.txt (107,630 entries, 398,359 citation occurrences); DCS inventory + token counts from VisualDCS Corpus-Delta 2021–2026 (276 texts). n and dates are on every artifact.
• Method. Anchored on DCS's bounded 276-text set; every siglum carries an adjudicated match_type (covered / residue / secondary / structural). Scripts: pd_extract_sigla.py, pd_dcs_crosswalk.py.
• Limitations. PD is published under letter a- only (6 of 37+ volumes), so these are PD's sources as exercised under a-, not its full canon. Siglum→title expansion is Opus-adjudicated, not sourced from PD's printed abbreviation list. See the report's §6.
• Owner repo. csl-atlas. Data: data/pd/.
• Download. Covered-text crosswalk (TSV) · Top residue (TSV) · Metrics (JSON).

Covered-text crosswalk (data table)

Conclusions — and the dictionary's own clock

The residue matters more because the dictionary that would define it will not finish for centuries. The Poona Dictionary began publishing in 1976; fifty years later it has issued ~6 bound volumes / 6,056 pages / 104,959 lemmas and reached only a- to ~apaca- — it has not finished the short vowel a, let alone ā. At one volume every 8.3 years against its ~37-volume plan, PD completes in roughly 250 years — around 2280.

A geometric corpus against an arithmetic dictionary. DCS grew 4.58 M → 5.69 M tokens in five years — +4.45 %/yr, doubling every ~16 years — while PD spent five decades in one letter. The two scale by different laws and do not converge: over one 30-year planning horizon DCS grows ~×3.7 (to ~21 M tokens, absorbing most of the purāṇas and kāvya it now lacks), while PD adds ~3–4 volumes and merely finishes a. The 75 % residue this page measures is therefore a moving front, not a permanent deficit — DCS's 2021→2026 Vedic surge is what closing it looks like.

Three yardsticks — AP, PWG, MW

Locate PD's frontier (apaca-) inside three finished Sanskrit dictionaries and the horizon sharpens. After 50 years PD has reached only 4–6 % of the alphabetical span a complete dictionary covers — yet its coverage of just a-…apaca- (104,959 lemmas) already holds as many entries as the entire Petersburger Wörterbuch, exceeds all of Apte, and is 54 % of all Monier-Williams. Per headword PD is 13–23× denser than any completed dictionary.

Two speeds, by what year does PD finish? Anchoring scope two ways — the ~37-volume plan vs holding PD's current density — and the rate at 121 pp/yr (its 50-year average) vs a post-2019 200 pp/yr:

Scope moves the finish by ~900 years; a 65 % speed-up buys only ~100–470. Completion stays centuries out either way — the constraint is PD's density, not its pace.

Milestones — clearing a, the vowels, the halfway mark

Turning the horizon into dates (at current density; both rates carried through):

PD will not clear even the short vowel a until ~2094 — the letter a alone is ~118 years of work. The vowel block (one-sixth of the alphabet) is not done until ~2179; the alphabetical mid-point sits at p, not k, because Sanskrit is so front-loaded with prefix families. (Vowel milestones are firm; consonant ones are upper bounds — the 37-volume plan compresses them.)

Is a the hardest letter?

A good question — is a hardest because it holds all the a-/an- privative compounds (a-dharma "non-dharma", an-artha "misfortune"), the negated samāsas buildable from almost any word? The headword lists answer precisely — and it depends on the dictionary:

In MW and PWG, a is only the 4th-largest letter — the sibilant s wins, then p and v. But in Apte (the compound-inclusive practical dictionary) a is #1 at 14.3 %, exactly because the privative and prefix families (a-, an-, ā-, adhi-, anu-…) import a negated or prefixed twin of much of the language under a. So the samāsa intuition is real — but s/p/v carry their own huge prefix families (sam-, pra-, vi-…) and rival it. The sting: PD's hardest letters are still ahead — s, p and v each hold more MW headwords than the a that has already cost fifty years.

And how many a-entries are actually compounds? — 83 %. MW's printed-form list marks every compound joint with a dash: of the 23,590 a/ā-entries, 19,601 (83 %) are dash-marked samāsas, only ~4,000 simple stems. a is overwhelmingly a letter of combinations, not roots — which is the deepest reason a maximalist dictionary drowns in it: it isn't defining ~4,000 words, it's documenting ~20,000 compounds, each with its own attestations.

The long-dictionary league table

Same language, done fast. Seven Sanskrit/Indo-Aryan dictionaries were completed in 10–27 years — Böhtlingk's kürzere Fassung logged 151,349 lemmas across the whole alphabet in ten years. Only the two that chose exhaustive historical detail stall:

The graveyard of a. Sanskrit is not the problem — exhaustive detail is. The proof is 170 years old: Goldstücker's 1856 remake of Wilson set out to improve the whole dictionary, sank into the letter a, published 6,761 exhaustive entries, and was abandoned there at his death — the one clean death-in-a. Two contemporaries are often lumped in but don't belong: Kossowicz (1854, Sanskrit→Russian) reached 13,488 entries but is ordered by the Russian, not Sanskrit, alphabet and left unfinished — a partial work of a different kind, not on the Sanskrit-letter axis at all; and Böhtlingk–Roth's PWG spent its whole first volume (1855) on a- yet completed the alphabet, because it refused to be exhaustive. PD is Goldstücker's project reborn at ~16× the density (104,959 vs 6,761 for a) — the same undertaking, the same letter, the same trap. Why a traps the exhaustive is anatomised in Letter anatomy: it heads five preverb families plus the privative, so 83 % of its entries are compounds, not roots.

The controlled experiment — two dictionaries, one city, one editor. Across town in the same Pune, the Bhandarkar Institute (BORI) Comprehensive & Critical Dictionary of the Prakrit Languages began in 1988 — twelve years younger than PD — and has already reached a through ujjhittu: all the vowels and into the consonants, ~33,600 lexemes against a bounded ~90,000-entry plan — still far from finished (~37 % after 38 years), but moving through the alphabet, which is the point. Its founding editor was A. M. Ghatage — the very scholar who edited PD's first volumes. Same man, same city, same decade; the difference is scope. BORI bounded its corpus and cleared the vowels in 38 years; PD chose the unbounded maximum and in 50 has not left a. A great Sanskrit-family dictionary finishes only if it refuses to let a become infinite.

PD's real peers are the century-long giants — and it is projected to outlast them all. No dictionary ever finished took more than ~180 years:

Even PD's fastest scenario (2182, 206 yr) exceeds the longest dictionary ever completed; its density-held scenario is ~9× the Dutch WNT. On present course PD becomes the longest single lexicographic project in history — a direct consequence of its 13–23× density, not of any lack of effort.

The other model — born-digital corpus lexicography

Every dictionary above belongs to one paradigm: the print historical dictionary, composed from slips and published alphabetically over decades. The sharpest contrast is the paradigm that retires it. TamiLex (Universität Hamburg), a digital lexicography project for Tamil, differs from PD in kind, not speed:

• Born-digital, not print-fascicule — a queryable database is never stuck at a letter; the death-in-a trap is structurally impossible.
• Whole-language-at-once, not alphabetical — partial-but-growing coverage everywhere at once, not a perfected before b is touched. PD's centuries-long horizon is an artifact of the alphabetical-completion constraint a database doesn't have.
• Corpus-integrated, not corpus-blind — lexicon and attestation are one system. For Sanskrit, dictionary (PD) and corpus (DCS) are separate — this whole study is the after-the-fact weld.
• Aggregative and open, not from-scratch and print.

TamiLex is, for Tamil, essentially what the next section recommends for Sanskrit. The drastic difference isn't that it's faster — it's that it abandons the model whose clock this page has been measuring. PD may be the last of the great print historical dictionaries; DCS welded to a lexical layer is the Sanskrit form of what comes after. (Exact TamiLex figures pending — the site returned a transient error at time of writing; the comparison turns on paradigm, not dates.)

Who finishes a dictionary — and why

Are Germans simply the best dictionary-makers, and Indians confined to the kośa? The record says no — it is about method, institution and scope, not nation.

• The real German trait is finishability, not maximalism. The German/European school dominated historical lexicography through its academies and a discipline Mayrhofer put as a rule: do not begin a dictionary you cannot finish. PD is not German — it is a Deccan-College institutional project on the slip-archive method (slips now merely scanned); its problem is unbounded scope, not any imported method.
• The "entries shrank over time" pattern is European, not Indian. The entry-size study found the decay in PWG (−0.19), PWK (−0.34), Grassmann (−0.20) (and it runs through KEWA/EWA and the big non-Sanskrit dictionaries) — while the two Indian encyclopaedias SKD and VCP show none (ρ≈0.00). The belief that funding-cut compression was an Indian failing is backwards.
• The finished dictionaries were near-solo; the unfinished one is a committee. V. S. Apte (Fergusson College, Pune) compiled a complete, still-standard Sanskrit–English dictionary almost alone and died at 34 (1858–92), two years after it appeared. Böhtlingk carried the kürzere Fassung single-handed and, by Delbrück's estimate (1904), wrote 9/10 of the "Böhtlingk–Roth" PWG (Roth did the Ṛgveda + Suśruta; Whitney the AV; Weber the Śatapatha; Stenzler Manu; Kern & Schiefner the rest) — one man, two complete Sanskrit dictionaries, in ~22 years. Against that, PD's committee: 74 years since its 1948 conception and chronically under-staffed — a severe shortage of Sanskrit lexicographers, ~a dozen-to-two-dozen active against millions of planned entries — a cause of its non-completion, not a footnote. A single vision on a bounded scope finishes; a large, under-staffed institution on an unbounded one (PD) may never finish at all.
• Indians did not do "only kośas." The kośa is a sophisticated genre, not a ceiling — and beyond it Apte, SKD and VCP are finished Indian encyclopaedias. The struggling projects are the modern institutional ones (BORI far from done; the Madras Tamil Lexicon now remade in Hamburg as TamiLex).

Verdict: the great dictionaries came from wherever bounded scope met a discipline of finishability — most often the German academy, and the lone scholar who scoped to one lifetime. PD is stuck because it is unbounded and institutional, not because it is Indian; the reef named a sinks the exhaustive of every nation alike.

What follows. DCS attests; PD analyses — a corpus cannot supply PD's historical sense-development, so for the 97 % of the lexicon PD has not reached it remains the only project attempting it. But the practical conclusion is unambiguous: don't wait for the dictionary — feed the corpus. The highest-leverage additions to DCS are exactly PD's high-frequency uncovered works, and PD's own citation frequency (the residue chart above) is a ready-made digitisation priority list. Full argument: reports/PD_DCS_CORPUS_COVERAGE_2026.md §8.