संस्कृतव्याकरणम् · 05 of 24 Home
Aṣṭādhyāyī · सुबन्तप्रकरणम्

One stem,
सप्त विभक्तयः

Every Sanskrit noun bends the same way: seven cases, three numbers, twenty-one slots to fill. What changes from word to word is only the shape of the stem — and there turn out to be a handful of shapes, not thousands. Learn the shapes, not the words, and every noun you will ever meet already declines.

देव
+
तृतीया · एकवचन
→
देवेन
अ-stem · m. god by, with
प्रातिपदिकम् — the stem a rule builds on विभक्तिः — the case-and-number ending added पदम् — the inflected word that results

What a case is — and is not

2.3.1

It is tempting to learn प्रथमा, द्वितीया … सप्तमी as "the subject case, the object case, the instrument case" and stop there. That shortcut breaks the first time you meet a passive sentence, because the same two participants can take different cases depending on what the verb has already said.

अनभिहितेA case marks a role only when nothing else already has

Six semantic roles — कारक — are defined in the first quarter of chapter 1.4, purely by what a participant does in the sentence, with no mention of any case ending: कर्ता is the independent one (, 1.4.54), कर्म is what the agent most wants to reach (, 1.4.49), करण is the most effective means (, 1.4.42), सम्प्रदान is who the action is aimed at (, 1.4.32), अपादान is the fixed point something moves away from (, 1.4.24), and अधिकरण is the locus (, 1.4.45).

Only afterwards, in 2.3, does Pāṇini attach a vibhakti to each role — and every one of those rules is conditioned by , 2.3.1: "[only] when [the role is] not already expressed." A verb's own personal ending always expresses exactly one kāraka in a sentence — the agent, in the ordinary active voice. That kāraka is therefore already "spoken for" and does not get its kāraka-vibhakti; instead the noun just states its own meaning, gender and number, which is what प्रथमा is defined to do (, 2.3.46).

Watch it flip

ददाति।active — the ending -ति already expresses कर्ता, so गुरुः is प्रथमा; कर्म (फलम्) is unexpressed, so it takes द्वितीया, 2.3.2 दीयते।passive — the ending -ते now expresses कर्म, so फलम् is प्रथमा; कर्ता (गुरुणा), now unexpressed, must take तृतीया, 2.3.18

Nothing about "the teacher" or "the fruit" changed. What changed is which role the verb already carries — and the case system fills in exactly the gap that leaves. That is the whole content of "the seven cases are not the six kārakas."

Two vibhaktis mark no kāraka at all. प्रथमा states the प्रातिपदिक's own meaning (2.3.46) — a role only when the verb hasn't taken it. षष्ठी is defined even more bluntly: , 2.3.50, "षष्ठी [applies] to what remains" — whatever relation between two nominals is left over once the six kārakas are accounted for, chiefly plain possession. Of seven vibhaktis, five mark kārakas (तृतीया doing double duty for कर्ता and करण); two do not. Page 8 works out all six roles properly — this page only needed you to know they are not the same list as the seven cases.
The vocative, briefly. Grammars often print an eighth row, सम्बोधन. It is not a seventh kāraka bolted on: 2.3.47 simply extends प्रथमा to cover addressing someone, and 2.3.48 just names the addressed word आमन्त्रित. Grammatically it is a special, usually shortened, प्रथमा — not an independent case — which is why it stays out of the seven-row tables on this page and appears here instead:

The stem families

Pattern, not memorisation

Twenty-one endings look like a lot to memorise per word — until you see that the endings barely change across the whole language. What changes is how the stem's own final sound reacts to them. Six shapes cover essentially everything.

अ · आThe biggest family — and two declensions inside it

Stems ending in अ split by gender: masculine and neuter share the stem (देव, फल) but decline differently from row 1; a feminine in this family does not stay अ at all — it takes the टाप् affix and becomes an आ-stem (, 4.1.4), declining on a completely different pattern, as रमा shows. Why a given noun ends up masculine, neuter or feminine is page 6's question; here, just watch what each shape does with the same seven cases:

Notice नपुंसक फलम्: its nominative and accusative are identical in every number. That is not a coincidence about फल specifically — it is true of every neuter noun in the language, and it is the first hint of this page's next big idea.

इ · ईTwo feminine patterns hiding under one vowel

हरि is the masculine इ-stem: unremarkable, the ऋ/इ/उ pattern this page keeps returning to. The two feminines are where it gets interesting. मति keeps its short इ and belongs to a class Pāṇini names घि — , 1.4.7, "the rest [of the short इ/उ feminines] — except सखि — is घि." नदी, with a long ई, is a different technical animal: , 1.4.3, names every long-ई or long-ऊ word denoting a female नदी, full stop. Two names, two rule-sets:

Compare मत्या (तृतीया) with नद्या: both look like "drop the vowel, add या", but मति also keeps a second, unreduced option in the dative/ablative/genitive/ locative (मतये-मत्यै, and so on) that नदी never offers. घि-stems hedge; नदी-stems commit.

उ · ऊThe same split, one vowel over

उ/ऊ mirrors इ/ई exactly, rule for rule: गुरु is the masculine उ-stem; धेनु, short उ, is घि again (1.4.7); वधू, long ऊ, is नदी again (1.4.3) — the same sūtra that caught नदी catches it too, because 1.4.3 explicitly names both long vowels.

One more thing इ/ई and उ/ऊ share, and अ-stems don't: गुरोः is both the ablative and the genitive singular of गुरु — identically. So is हरेः. This is not a typo in the corpus; it is a real, systematic gap in these two case-forms for every इ-, उ- and ऋ-stem, resolved only by context. अ-stems don't have this problem: देवात् (5th) is never confusable with देवस्य (6th).

ऋKinship terms, and a vowel that grows

Almost every ऋ-stem in the language is a relation — पितृ father, मातृ mother, भ्रातृ brother, कर्तृ "doer" (the agent-noun that every verbal root can form). Their nominative/accusative singular and dual, and the whole nominative plural, strengthen ऋ to the vṛddhi grade आर् — , 7.3.110, "[गुण, for a stem ending in ऋ,] before ङि (locative singular) and सर्वनामस्थान." Everywhere else the stem stays plain ऋ, softened only by ordinary sandhi.

कर्ता / कर्तारौ / कर्तारः vs कर्त्रा / कर्त्रे / कर्तुः — the same two-grade story as the consonant stems below, just carried by a vowel instead of a lost nasal. Once you see one ऋ-stem you have effectively seen the whole family: पितृ and मातृ differ from कर्तृ only in the last few sandhi-level details, never in the pattern.

हल्Consonant stems — endings that don't change, stems that do

A consonant-final stem meets every affix bare, so whatever the affix's first sound needs from ordinary sandhi (page 3), it gets. मरुत् (त्), वाच् (च्, → क्/ग् at a word boundary) and मनस् (स्, → ओ before भ्) show three different final sounds doing the same job with the same endings:

The affix भिः never changes; only the sandhi at the seam does. राजन् and आत्मन् look the odd ones out because their stem itself alternates — see the next section — but their instrumental plural obeys the exact same rule: strip the nasal before a consonant, by , 8.2.7, "the final न् of a nominal stem is elided [when a पद ends there]."

A sixth pattern: participles that gain a nasal

गच्छत् "going" (present participle, शतृ) and भगवत् "possessing भग" (मतुप्) both end bare, with no nasal at all — but सर्वनामस्थान inserts one that was never in the stem to begin with: , 6.4.14, "and of a stem ending in अत् or वस्, [नुम् is added, in सर्वनामस्थान]." गच्छत् + सु gives गच्छन्, not *गच्छत्.

This is राजन्'s story run backwards: राजन् loses a nasal it already has, in the weak grade; गच्छत् and भगवत् gain one they never had, only in the strong grade. Same trigger, opposite direction — which is exactly why "strong vs. weak," not "nasal vs. no nasal," is the pattern worth remembering.

अनियमाःThree irregulars worth meeting on purpose

Every language has a handful of nouns too common to regularise. Sanskrit's are genuinely irregular, but each is irregular by an explicit rule — not by exception to no rule at all.

गो — cow, ox

Monosyllabic, and the only word in this corpus attested in both genders with an identical paradigm — गौः/गावौ/गावः serves "cow" and "ox" alike. , 7.1.90, marks it for a special strengthening (गो + णित् affixes → गो becomes गव्/गौ depending on what follows) that no other word triggers by name.

पथिन् — road

Nominative singular पन्थाः shares almost nothing in shape with the oblique पथः, पथा… — an ablaut this extreme gets its own sūtra, , 7.1.85, naming पथिन् (and two rarer look-alikes) by name.

सखि — friend

Short इ, so it looks like it should be a घि-stem, मति's family — and 1.4.7 explicitly carves it out: "घि, except सखि." Its strong stem instead takes गुण straight through, ऋ-stem style: सखा / सखायौ / सखायः.

Strong and weak — the idea behind all of it

1.1.42 · 1.1.43

Every alternation above — फल's identical nom./acc., ऋ's वृद्धि, राजन्'s disappearing न् — is one phenomenon wearing different sandhi. Pāṇini names the triggering environment once, and it governs the whole noun system.

सर्वनामस्थानम्The "strong" cases, defined

(1.1.42) plus (1.1.43) name a specific set of case-affixes सर्वनामस्थान: सु, औ, जस् (nominative, all three numbers) and अम् — accusative singular only, not शस् (accusative plural) — for a non-neuter word; for a neuter, only शि (which covers accusative/nominative dual and plural) counts, and not सु/अम् singular at all. Wherever a सर्वनामस्थान affix attaches, the stem takes its strongest shape: वृद्धि for ऋ-stems, a nasal-plus-lengthening augment for अन्-stems.

Everywhere else divides again, by whether the affix starts with a vowel or a consonant. Before a vowel-initial affix that is not सर्वनामस्थान, the stem is called भ — , 1.4.18 — and takes its weakest shape. Before a consonant-initial one it is called पद — , 1.4.17 — and only ordinary word-final (पदान्त) sandhi applies, a middle shape. Three environments, not two, even though "strong vs. weak" is the usable shorthand.

Why accusative plural is the trap. Learners assume "nominative and accusative are the subject/object pair, so they behave alike." They don't, systematically: accusative singular and dual are सर्वनामस्थान (strong); accusative plural is not, and instead patterns with every other oblique case (weak). राजन्'s राजानम् (acc.sg, strong) and राज्ञः (acc.pl, weak) are the same kāraka, the same case name, two different stem-grades.
राजन् · आत्मन्Two न्-stems, one keeps its न्

राजन् loses the अ of अन् entirely in the weakest (भ) cases — , 6.4.134, "the अ of a stem ending in अन् is elided" — leaving just राज्न्, which further sandhi turns into राज्ञ्. आत्मन् simply does not undergo 6.4.134 and keeps अन् throughout its weak cases. Both still take the same strong-case strengthening (6.4.8, सर्वनामस्थाने चासम्बुद्धौ — "in सर्वनामस्थान, and not in the vocative"):

Same affixes, same सर्वनामस्थान rule turning on the same strong grade — and still two different weak stems, because 6.4.134 simply does not apply to every अन्-stem. This has to be learned per stem; there is no shortcut past it, only the pattern that makes it a two-way choice instead of twenty memorised words.

The cases at work

सप्त वाक्यानि

Every marked word below is exactly the corpus's own form for that stem, case and number — read straight out of the same tables above, not retyped. Words are shown at their citation (pre-sandhi) boundary, so the ending stays visible; page 3 covers what happens when they actually meet in speech.

One sentence per case
वदति।
प्रथमा — "The teacher speaks." कर्ता, already expressed by -ति; no other kāraka is in play.
पश्यति।
द्वितीया — "The student sees the tree." कर्म, unexpressed by an active verb, so it takes its own case (2.3.2).
शोभते।
तृतीया (करण) — "The student shines through learning." — the means, not the agent, this time.
ददाति।
चतुर्थी — "The teacher gives a fruit to the god." सम्प्रदान, the one aimed at (2.3.13).
पतति।
पञ्चमी — "The fruit falls from the tree." अपादान, the fixed point of separation (2.3.28).
वदति।
षष्ठी — "The teacher's son speaks." Not a kāraka — a relation between two nouns, not to the verb (2.3.50).
वसति।
सप्तमी — "The student lives in the village." अधिकरण, the locus (2.3.36).
Four at once.
ददाति।
"The teacher gives knowledge to the student in the village." कर्ता (प्रथमा) · कर्म (द्वितीया) · सम्प्रदान (चतुर्थी) · अधिकरण (सप्तमी) — one verb, four roles, four cases, and every one of them decided by 2.3, not by word order (Sanskrit has none to speak of — page 21 covers that).

Try a stem

Interactive

Type a prātipadika, or pick a chip, and its paradigm comes straight from the same corpus every table above did — cross-checked, not guessed. This page carries a worked set covering every pattern it teaches, not the whole corpus; type something outside that set and the page says so plainly, with a link to where the remaining 9,007 live. A beloved name like गणेश is a genuine gap even there — this is a grammatical corpus, not a general dictionary, and pretending otherwise would be worse than the gap.

Every stem-ending in the corpus

Coverage

EndingParadigmsFamilyExample
Where this comes from. Every row is a live count over the corpus's own 9,007 paradigm entries, grouped by how each stem actually ends — nothing here is an estimate. The "family" column is this page's teaching label for that ending, not a claim the corpus makes; a handful of rare finals (ऐ, औ, प्, व्, भ्, थ्, ल्…) are real but too small to build a section around, and are marked accordingly.

Check yourself

अभ्यासः

Six questions. Every answer names the rule that settles it.