இலக்கணம் grammar

Tholkappiyam first

தொல்காப்பியம் is the primary authority for word classes, வேற்றுமை and புணர்ச்சி. நன்னூல் is the fallback, and the primary only where Tholkappiyam does not cover the ground. The clearest such place is the six-part பகுபத உறுப்பு scheme, which Tholkappiyam does not enumerate.

The order is a design rule rather than a preference, and it has been tested once already. Modern course books quote Nannūl far more often than Tholkappiyam, so a system built from teaching material drifts toward Nannūl without anyone deciding to. Ours records which authority each claim came from, so the drift shows up instead of settling in.

tamil-grammar.md assigns வேற்றுமை (the eight cases) to Tholkappiyam as primary, Nannūl as fallback. Tholkappiyam-first is design rule #1 — it is never demoted because a modern course happened to quote Nannūl instead.

Where the authorities disagree, both are recorded

The third வேற்றுமை is the standing example. Tholkappiyam names one உருபு. Nannūl names four, and splits the meanings differently.

Both are true, so the table carries both and the answer says which one it used. Collapsing them into one tidy list would be a small act of scholarship destroyed for the convenience of a database schema.

The rules ship as cited data, not as code

Five tables live in the server as JSON: இடைநிலை, விகுதி, சாரியை, வேற்றுமை உருபு and விகாரம். Each one carries a source_priority block naming its authority, and the test suite fails if a table is missing one. So the linguistics can be audited as data, by someone who never opens a Python file.

Here is one real entry, the seventh வேற்றுமை, which is the case மரத்தில் lands in. The உருபு list is open and கண்-headed in both authorities:

Verse text quoted from the pinned editions and copied here by scripts/sync-grammar.py, never retyped. நூற்பா quoted from the pinned Project Madurai etexts (data/classical/), reproduced with their headers intact — see LICENSING.md. TVA extracts are cited, not redistributed; the ePUBs stay out of git.

How a citation is written

தொல்காப்பியம் numbers restart at 1 in every இயல் and collide across இயல் and அதிகாரம். A bare number is unusable, so it is always qualified. நன்னூல் runs continuously from 1 to 462, so a bare number there is unambiguous.

தொல்காப்பியம், சொல்லதிகாரம், வேற்றுமையியல், நூற்பா 3
தொல்காப்பியம், எழுத்ததிகாரம், புணரியல், நூற்பா 7
நன்னூல், நூற்பா 244

And never take a verse number from a secondary source, however accredited. The Tamil Virtual Academy course books quote Nannūl selectively and renumber: their 336, 319 and 136 are 337, 320 and 137 in the pinned edition. All three were wrong in our tables before we pinned the full texts, which is why the texts are now read at runtime rather than trusted from a note.

What is not settled

A rule table records its own gaps, and they are published rather than smoothed over. Two kinds. First, places where an authority simply does not print what we need:

• நான்காம் வேற்றுமை has no Nannūl verse recorded — TVA A0211 does not quote it and no Nannūl edition is pinned. The Tholkappiyam primary (வேற்றுமையியல் 14) IS cited, so the case is grounded.
• The decoder currently picks ONE உருபு per case by longest surface suffix match. With multi-உருபு cases now encoded (inst ஒடு/ஆல்/ஆன், abl இன்/இல், gen அது/ஆது/அ, loc கண்-headed open list) the selector should match against every listed உருபு for the tagged case, and fall back to the FIRST listed form (Tholkappiyam's) only when nothing matches.
• ஏழாம் வேற்றுமை's உருபு list is declared open by both authorities (‘அன்ன பிறவும்’ / ‘…முதல்’). Treat non-matches as unclassified, not as errors.
• Not yet extracted: சொல்லதிகாரம், வேற்றுமைமயங்கியல் (case syncretism, 35 நூற்பா) and விளிமரபு (37 நூற்பா). Both are Tholkappiyam-primary material relevant to this table.

Second, claims the project has derived rather than found stated in a verse. Those are marked as inferred and are not treated as authority until a verse is found or Saran rules on them. One example, in its own words:

The four curated paradigm entries whose stems change rather than take a doubled tense marker — சொல்→சொன்ன் (ல்→ன்ன்), கல்→கற்ற் (ல்→ற்ற்), கேள்→கேட்ட் (ள்→ட்ட்), விற்→விற்ற் (ற் doubling) — are விகாரம் (திரிதல் / ஒற்று இரட்டல்) affecting the பகுதி, NOT சந்தி + இடைநிலை.

Three questions are open on the concept map right now:

• விகாரம் / paradigm entries: are சொல்→சொன்ன், கல்→கற்ற், கேள்→கேட்ட் விகாரம் (திரிதல்) and விற்→விற்ற் ஒற்று இரட்டல், rather than சந்தி + இடைநிலை? Our derivation is under `விகாரம்.inferred`. If accepted, four paradigm entries change and கொடு needs marker='த்த்'.
• எச்சவினை விகுதி: is there a நூற்பா that names these as a class, so the inventory can rise above lesson-level citation?
• சந்தி: is there a verse giving சந்தி a positional definition, as 141 does for இடைநிலை and 243 for சாரியை? Currently only Saran's ruling plus elimination.

Then how a word comes apart →