The Tamil noun as a grammar: structure, endings and joining rules

This page is the reference behind Parts 1 and 2. It sets out the order of the parts of a noun, the endings with the verse behind each, and the rules that join them, over TLP1, the one-byte-per-sound encoding. Where a verse says it, the row names the verse. Where none does, the row says so.

The word, slot by slot

A noun form is a stem followed, in this order, by an optional plural or oblique-stem marker, an optional case with an extension that only some cases take, and an optional particle. The slots are written below in EBNF. The names in quotes are the feature labels of the reference lexicon, ThamizhiMorph.

noun_form = "noun", [ stem_kind ], [ case_part ], [ particle ] ;   (* the order of the slots *)
stem_kind = "infInc" | "pl" ;   (* infInc: the oblique stem with the இன் சாரியை; pl: the plural கள் *)
case_part = "nom", [ "cmpr" ]
          | "acc", [ "vpart" | "vpartp" ]
          | "dat", [ "ben" | "n" | "pl" ]
          | "inst"
          | "loc"
          | "abl"
          | "gen"
          | "soc"
          | "so" ;   (* a case; only some cases take an extension *)
particle  = "foc" | "psp" | sandhi ;   (* foc தான், psp a postposition, or a joining consonant at a word boundary *)
sandhi    = "sandhik" | "sandhic" | "sandhip" | "sandhit" ;   (* க் ச் ப் த் *)

The optional slots are the point. Naṉṉūl 243 says a சாரியை may appear, drop out or be optional, and Naṉṉūl 253 does not say which. So the grammar licenses a set of forms for each slot, and the output is that set.

This is a grammar of order, not of which combinations a given noun takes. That depends on the noun's class. The reference grid the chip is measured against uses 81 feature bundles. The grammar accepts all 81 and licenses 336 combinations in all. The model derives 67 of the 81; the other 14 carry labels it has no ending for (psp, a postposition, and n, vpart, vpartp). python3 tools/check_noun_grammar.py reruns this check.

The endings

Twenty-three endings, one row per label. Some labels give more than one form. infInc (the oblique stem) and nom (the bare name) have no ending of their own.

label	ending	basis
`pl`	கள்	lexicon; no verse found. Six spellings in the lexicon are one ending once the joining is derived
`in`	இன்	Naṉṉūl 244 lists it as a சாரியை
`acc`	ஐ	Naṉṉūl 296
`inst`	ஆல்	Naṉṉūl 297
`dat`	கு, with the உ சாரியை (உக்கு) or the இன் சாரியை (இற்கு)	கு: Naṉṉūl 298 and Tolkāppiyam வேற்றுமையியல் 14. The உ and இன்: Naṉṉūl 244, optional by 243. Where each applies: Arden 110 (rule P12 below)
`ben`	ஆக	lexicon; no verse found
`loc`	இல், இடம்	Naṉṉūl 302 lists both among the place words. The கண் of 301 is not generated
`abl`	இல், இன்	Naṉṉūl 299
`abl`	இலிருந்து	lexicon; no verse found
`gen`	அது	Naṉṉūl 300
`gen`	இன்	lexicon; no verse found
`soc`	ஓடு	Naṉṉūl 297
`soc`, `so`	உடன்	lexicon; no verse found
`foc`	தான்	lexicon; a particle
`cmpr`	போல்	lexicon; a particle
`sandhik` `sandhic` `sandhip` `sandhit`	க் ச் ப் த்	lexicon; the joining consonants at a word boundary, see P3

The joining rules

Where two parts meet, the rules below apply. The first one that fits applies, in this order: the particle rule P10, then P9, then, at a word boundary, a filter that lets only ச், ட், த் and ப் trigger anything, then P5, P6, P3, P1, P2, P11 and P4. P8 and P12 act earlier, on the stem and on the dative ending.

In TLP1 the top two bits of a byte give its kind: 00 for a vowel and 01 for a consonant. For consonants the next two bits give the class (0100 வல்லினம், 0101 மெல்லினம்) and the low three bits give the place. Bytes are in hex.

rule	what it does	condition	action	verse
P1	a மெல்லினம் takes the place of a following வல்லினம்	the left ends in ம் (`54`), the right begins with a வல்லினம் (`b >> 4 = 4`), not at a word boundary	replace ம் by `(a & ~7) \| (b & 7)`: ம் + க் gives ங்	Tolkāppiyam பிறப்பியல் 7 to 9, the place of each pair. The model applies it to ம் only
P2	ல், ள், ன் harden before a வல்லினம்	the left ends in ல் (`65`), ள் (`62`) or ன் (`55`), the right begins with a வல்லினம், not at a word boundary	`(a & 7) \| 0x40`: ல் gives ற், ள் gives ட், ன் gives ற்	ல், ள்: Naṉṉūl 227. ன்: Tolkāppiyam புணரியல் 21 (the இன் before கு) and புள்ளிமயங்கியல் 37
P3	a nasal is lost before ச் ட் த் ப்	at a word boundary only, the left ends in a மெல்லினம், the right begins with ச், ட், த் or ப்	drop the nasal	lexicon; no verse found. The model follows ThamizhiMorph, which keeps the nasal before க் and ற்
P4	a வல்லினம் doubles after a vowel	the left ends in a vowel, the right begins with a வல்லினம், the noun's class says it doubles, not at a word boundary	insert a copy of the வல்லினம்	Tolkāppiyam உயிர்மயங்கியல் 1, 19, 47, 52, 62 and 87 (அ, ஆ, ஈ, உ, ஊ, ஓ), and 33 and 78 (இ and ஐ double only in a case relation); Naṉṉūl 165
P5	a குற்றியலுகரம் drops before a vowel	the left ends in உ (`02`) after a consonant, the ending is a குற்றியலுகரம், the right begins with a vowel	drop the உ	Naṉṉūl 164; Tolkāppiyam புணரியல் 3. Naṉṉūl 94 says which உகரம் is a குற்றியலுகரம்
P6	a glide fills a vowel-vowel junction	the left ends in a vowel and the right begins with one	insert ய் after இ, எ or ஐ, otherwise வ்	Naṉṉūl 162
P8	ன் becomes ர் before கள்	the noun is in the class of மாணவன் and ends in ன் (மாணவன் gives மாணவர்கள்); the model's condition is the class	replace the final ன் by ர்	lexicon; no verse found
P9	two equal vowels at the seam of a stem and its சாரியை become one	the left ends in a vowel, the right is a சாரியை beginning with the same vowel	drop the right's first sound: மர + அத்து gives மரத்து	Naṉṉūl 252; Tolkāppiyam புணரியல் 23. Naṉṉūl 244 lists அத்து
P10	a particle doubles its initial வல்லினம் after a vowel	the right is a particle marked to double, the left ends in a vowel, the right begins with a வல்லினம்	insert a copy of the வல்லினம்	lexicon; no verse found
P11	the final ஒற்று doubles before a vowel ending	the left ends in a consonant, the right begins with a vowel, the noun's class is marked for it	repeat the consonant: கண் + ஐ gives கண்ணை	Naṉṉūl 205 restricts it to a monosyllable with a short vowel. The model marks whole classes, which is wider
P12	the dative is கு alone after இ, ஈ or ஐ	the base ends in இ, ஈ or ஐ and the case is the dative	use கு, not உக்கு or இற்கு, and double the க்: குழந்தைக்கு	கு: Naṉṉūl 298 and Tolkāppiyam வேற்றுமையியல் 14. The உ is optional: Naṉṉūl 243 and 244, Tolkāppiyam உருபியல் 30. Where it is absent: Arden 110

What is checked and what is not

The EBNF accepts all 81 feature bundles of the grid, and its text above is produced by the same data the check runs on. The joining rules above are the eleven labelled rules in tools/phon_rules.py, and the hardware version of them agrees with the Python on 927,200 test vectors.

Two things are open. The table of which stems are cut and which take an increment, the class table of Part 2, is still fitted to ThamizhiMorph's lexicon; deriving it from the verses is the next job. And the readings of the verses above have not yet had outside review. If one is wrong, please say so.

Everything is at github.com/Ramaseshanr/tamil-grammar-silicon.