Sanskrit Grammar Decoded
Not a textbook — a thinking tool. Sanskrit grammar is a machine: every piece connects to every other with ruthless logic. This guide shows you the machine, not the parts list.
English uses word order + prepositions. The boy gives a book to the girl. Move any word and it breaks.
Sanskrit uses endings (pratyaya) on every word. The ending tells you who does what, to whom, with what, where, when. Word order is free — rearrange a Sanskrit sentence however you like, meaning stays identical. The endings carry all the information.
Prose and titles are in English for readability. Sanskrit terms use IAST transliteration inline (gacchati, devam, vibhakti). Devanagari appears in tables, examples, and dedicated blocks — where you're actually looking at the language itself. 💡 = aha moments. 🧠 = memory tricks. ⚠️ = common traps.
The Sound System varṇamālā
Sanskrit's alphabet isn't random — it's a phonetic periodic table. Sounds are organized by where in your mouth they form (sthāna) and how your vocal cords behave (prayatna). Once you see the grid, the alphabet stops being a list and becomes a map of the human mouth.
Every vowel comes in a hrasva (short, 1 beat) and dīrgha (long, 2 beats) pair. Then come the diphthongs (two vowels blended).
| Type | Short | Long | Position | Sounds like |
|---|---|---|---|---|
| simple | अ a | आ ā | Throat (kaṇṭhya) | "u" in but / "a" in father |
| इ i | ई ī | Palate (tālavya) | "i" in pit / "ee" in feet | |
| उ u | ऊ ū | Lips (oṣṭhya) | "u" in put / "oo" in boot | |
| ऋ ṛ | ॠ ṝ | Retroflex (mūrdhanya) | "ri" in grip, tongue curled | |
| diphthong | ए e ऐ ai | Throat + Palate | "ay" in say / "ai" in aisle | |
| ओ o औ au | Throat + Lips | "o" in go / "ou" in loud | ||
All of Sanskrit morphology runs on vowel gradation — strengthening a vowel to signal grammar. Three grades:
Weak: i, u, ṛ → Guṇa (medium): e, o, ar → Vṛddhi (max): ai, au, ār
Think of it as turning up a dial: i → e → ai | u → o → au | ṛ → ar → ār
This one pattern explains verb conjugation, noun derivation, compound formation, sandhi — everything. If you remember nothing else, remember this table.
| Weak | Guṇa | Vṛddhi | Real example |
|---|---|---|---|
| a | a (no change) | ā | |
| i / ī | e | ai | √nī → netā (leader) → naitika (ethical) |
| u / ū | o | au | √śuc → śoka (grief) → śaucam (purity) |
| ṛ / ṝ | ar | ār | √kṛ → karma (action) → kāraka (agent) |
Guṇa = add "a" in front. a + i = e, a + u = o, a + ṛ = ar. That's literally what's happening — the "a" merges with the vowel. Vṛddhi = add "ā" in front. ā + i = ai, ā + u = au, ā + ṛ = ār. Same logic, stronger prefix. This is also a sandhi rule — guṇa and vṛddhi ARE sandhi.
The stop consonants form a 5×5 grid — 5 mouth positions (rows) × 5 voicing types (columns). This is a map of your vocal tract from back to front.
Semivowels: य ya, र ra, ल la, व va | Sibilants: श śa, ष ṣa, स sa | Aspirate: ह ha | Special: anusvāra ं (ṃ), visarga ः (ḥ)
Visarga (ḥ) before ca/cha → ś (palatal row). Before ṭa/ṭha → ṣ (retroflex row). Before ta/tha → s (dental row). The consonant assimilates to the same row as what follows. The grid isn't just for memorizing the alphabet — it predicts sound changes.
Sandhi when sounds meet
When two words sit next to each other, their edges change sound. English does this informally ("did you" → "didja"). Sanskrit does it formally, with strict rules. Sandhi is the #1 reason texts look intimidating — words melt together. But the rules are mechanical and predictable.
Sanskrit was oral first. When you speak fast, your mouth naturally smooths transitions. Try saying "na asti" quickly — it becomes "nāsti." Try "sūrya udaya" fast — "sūryodaya." Sandhi just writes down what the mouth already wants to do. Your guṇa/vṛddhi table from chapter 1? Those ARE the sandhi rules for a/ā meeting other vowels.
Core principle: two vowels cannot sit next to each other. They must merge.
| Rule | Pattern | Example | Result |
|---|---|---|---|
| Same vowel merges long | a+a → ā, i+i → ī, u+u → ū | न + अस्ति | नास्ति |
| Guṇa sandhi | a/ā + i/ī → e a/ā + u/ū → o | देव + इन्द्र | देवेन्द्र |
| Vṛddhi sandhi | a/ā + e/ai → ai a/ā + o/au → au | तव + एव | तवैव |
| Semivowel sub. | i/ī → y, u/ū → v before dissimilar vowel | सु + आगतम् | स्वागतम् |
Same vowels merge long. a/ā + different → guṇa or vṛddhi. i/u become their semivowel (y/v) before different vowels. That covers 90% of vowel sandhi.
The visarga (ḥ) at the end of a word is really an old -s or -r in disguise. What it becomes depends on what follows.
| Context | Rule | Example |
|---|---|---|
| aḥ + a | → o, second a drops (avagraha ऽ) | शिवः + अहम् → शिवोऽहम् |
| aḥ + voiced cons. | → o | रामः + वदति → रामो वदति |
| ḥ + ca/cha | → ś (palatal row) | रामः + च → रामश्च |
| ḥ + ta/tha | → s (dental row) | नरः + तत्र → नरस्तत्र |
| ḥ + ka/kha/pa/pha | ḥ stays | रामः करोति |
śivo'ham (शिवोऽहम्) looks like one word but it's THREE: śivaḥ + aham → śivo + 'ham. "I am Śiva." If you can split this, you can split any sandhi.
The 8 Cases vibhakti
English uses prepositions + word order. Sanskrit bakes all of that into the ending of the noun itself. There are 8 vibhakti, each encoding a different kāraka (semantic role).
| # | Vibhakti | Kāraka | Function | Hindi equiv. | deva- example |
|---|---|---|---|---|---|
| 1 | Prathamā | kartā | doer / subject | ने (or unmarked) | देवः |
| 2 | Dvitīyā | karma | object / destination | को | देवम् |
| 3 | Tṛtīyā | karaṇa | instrument / means | से | देवेन |
| 4 | Caturthī | sampradāna | recipient / purpose | को / के लिए | देवाय |
| 5 | Pañcamī | apādāna | source / separation | से (from) | देवात् |
| 6 | Ṣaṣṭhī | sambandha | possession / relation | का/की/के | देवस्य |
| 7 | Saptamī | adhikaraṇa | location / time | में / पर | देवे |
| 8 | Sambodhana | — | direct address | हे! | हे देव |
Hindi postpositions map directly: ने=prathamā, को=dvitīyā/caturthī, से=tṛtīyā/pañcamī, का=ṣaṣṭhī, में=saptamī. The difference? Hindi uses separate postpositions. Sanskrit fuses them into the noun ending.
Hindi uses से for both. Sanskrit splits them: Tṛtīyā (karaṇa) = "with/by means of" — hastena likhati (writes with hand). Pañcamī (apādāna) = "from/because of" — nagarāt āgacchati (comes from city). The test: does the noun participate in the action (tṛtīyā) or is something moving away from it (pañcamī)?
Declension Tables śabda-rūpa
A noun "declines" — changes its ending for each vibhakti × vacana combination. That's 8 × 3 = 24 slots per noun. The pattern depends on the stem ending and gender. Learn the a-stem masculine (deva) first — it's the foundation.
| Vibhakti | Sg | Du | Pl |
|---|---|---|---|
| Prathamā | देवः | देवौ | देवाः |
| Dvitīyā | देवम् | देवौ | देवान् |
| Tṛtīyā | देवेन | देवाभ्याम् | देवैः |
| Caturthī | देवाय | देवाभ्याम् | देवेभ्यः |
| Pañcamī | देवात् | देवाभ्याम् | देवेभ्यः |
| Ṣaṣṭhī | देवस्य | देवयोः | देवानाम् |
| Saptamī | देवे | देवयोः | देवेषु |
| Sambodhana | हे देव | हे देवौ | हे देवाः |
Dual has only 3 distinct forms: -au (pra/dvi/sam), -ābhyām (tṛ/catur/pañ), -yoḥ (ṣaṣ/sapt). Caturthī = Pañcamī in dual AND plural. Ṣaṣṭhī = Saptamī in dual. Once you see these collapses, the 24-cell table shrinks to ~15 unique forms.
| Vibhakti | Sg | Du | Pl |
|---|---|---|---|
| Prathamā | फलम् | फले | फलानि |
| Dvitīyā | फलम् | फले | फलानि |
| Tṛtīyā through Saptamī — identical to deva (masculine a-stem) | |||
Prathamā = Dvitīyā always in neuter. This holds across ALL declension types. Tṛtīyā onwards is identical to the masculine. So you only learn the first two rows separately.
| Vibhakti | Sg | Du | Pl |
|---|---|---|---|
| Prathamā | माला | माले | मालाः |
| Dvitīyā | मालाम् | माले | मालाः |
| Tṛtīyā | मालया | मालाभ्याम् | मालाभिः |
| Ṣaṣṭhī | मालायाः | मालयोः | मालानाम् |
| Saptamī | मालायाम् | मालयोः | मालासु |
| Vibhakti | Sg | Du | Pl |
|---|---|---|---|
| Prathamā | अग्निः | अग्नी | अग्नयः |
| Dvitīyā | अग्निम् | अग्नी | अग्नीन् |
| Tṛtīyā | अग्निना | अग्निभ्याम् | अग्निभिः |
| Saptamī | अग्नौ | अग्न्योः | अग्निषु |
| Vibhakti | Sg | Du | Pl |
|---|---|---|---|
| Prathamā | मधु | मधुनी | मधूनि |
| Dvitīyā | मधु | मधुनी | मधूनि |
| Tṛtīyā | मधुना | मधुभ्याम् | मधुभिः |
| Saptamī | मधुनि | मधुनोः | मधुषु |
Gender & Number liṅga & vacana
Grammatical gender is mostly arbitrary for non-living things. You learn it with the word. Key rule: adjectives must match the noun in gender, number, and case.
| Vacana | When | deva- |
|---|---|---|
| ekavacana (singular) | one | देवः |
| dvivacana (dual) | exactly two | देवौ |
| bahuvacana (plural) | three or more | देवाः |
In dual: pra = dvi = sam (one form), tṛ = catur = pañ (one form), ṣaṣ = sapt (one form). Only 3 distinct forms instead of 8.
Pronouns sarvanāman
Pronouns decline through all 8 vibhakti with irregular patterns. The most important: asmad (I/we), yuṣmad (you), tad (he/she/it/that).
| asmad (I / we) | yuṣmad (you) | |||||
|---|---|---|---|---|---|---|
| Sg | Du | Pl | Sg | Du | Pl | |
| Pra. | अहम् | आवाम् | वयम् | त्वम् | युवाम् | यूयम् |
| Dvi. | माम् | आवाम् | अस्मान् | त्वाम् | युवाम् | युष्मान् |
| Ṣaṣ. | मम | आवयोः | अस्माकम् | तव | युवयोः | युष्माकम् |
Hindi वह/वो descends from tad. उसने (= tena), उसको (= tam), उसका (= tasya). Forms changed but the system survives.
Roots & Classes dhātu & gaṇa
Every Sanskrit verb comes from a dhātu (root) — usually one syllable. ~2,000 roots exist. To conjugate: transform the root into a stem using the gaṇa's vikaraṇa-pratyaya, then add person/number endings.
√kṛ spawns karma, kartā, kṛta, saṃskṛta, kartavya, kārayati — and Hindi inherited nearly all of them. Learning one dhātu gives you 10-20 words for free.
| Gaṇa | Name | How it works | Root → Stem | 3rd sg |
|---|---|---|---|---|
| 1 | bhvādi | Guṇa the root vowel + a | √bhū → bhava- | bhavati |
| 2 | adādi | No suffix — endings on root directly | √ad → ad-/at- | atti |
| 3 | juhotyādi | Reduplicate the root | √hu → juho-/juhu- | juhoti |
| 4 | divādi | Add ya | √div → dīvya- | dīvyati |
| 5 | svādi | Add no (strong) / nu (weak) | √su → suno-/sunu- | sunoti |
| 6 | tudādi | Add a without vowel change | √tud → tuda- | tudati |
| 7 | rudhādi | Insert na/n before final consonant | √rudh → ruṇadh- | ruṇaddhi |
| 8 | tanādi | Add o (strong) / u (weak) | √tan → tano-/tanu- | tanoti |
| 9 | kryādi | Add nā (strong) / nī (weak) | √krī → krīṇā-/krīṇī- | krīṇāti |
| 10 | curādi | Strengthen vowel + aya | √cur → coraya- | corayati |
Gaṇa 1, 4, 6, 10 (highlighted) are thematic — the stem stays the same throughout. No strong/weak alternation. Gaṇa 1 alone covers the majority of verbs.
Conjugation tiṅanta
Verbs agree with the subject in puruṣa (person) and vacana (number). 3 × 3 = 9 forms per tense per voice.
| Sg | Du | Pl | |
|---|---|---|---|
| 3rd (prathama) | -ti | -taḥ | -anti |
| 2nd (madhyama) | -si | -thaḥ | -tha |
| 1st (uttama) | -mi | -vaḥ | -maḥ |
| Sg | Du | Pl | |
|---|---|---|---|
| 3rd | -te | -ete | -ante |
| 2nd | -se | -ethe | -dhve |
| 1st | -e | -vahe | -mahe |
-ti = parasmaipada. -te = ātmanepada. That one letter tells you the voice. Also: 1st person always has m (mi/maḥ/mahe). 2nd person has s or th. 3rd person has t.
| Sg | Du | Pl | |
|---|---|---|---|
| 3rd | पठति | पठतः | पठन्ति |
| 2nd | पठसि | पठथः | पठथ |
| 1st | पठामि | पठावः | पठामः |
Tenses lakāra
Pāṇini uses 10 lakāra to cover all tenses and moods. Each named with la + a letter. The 6 most important:
| Lakāra | Name | When | Formation | √gam (go) |
|---|---|---|---|---|
| laṭ | Present | now / habitual | stem + primary endings | gacchati |
| laṅ | Imperfect | narrative past | a- + stem + secondary | agacchat |
| liṭ | Perfect | remote past | reduplication + special | jagāma |
| luṅ | Aorist | simple past event | a- + root + various | agamat |
| lṛṭ | Future | will happen | root + -iṣya-/-sya- | gamiṣyati |
| loṭ | Imperative | commands | stem + special endings | gacchatu |
laṅ (imperfect) = narrative camera. "he went, she said." 90% of past tense in stories. Learn this first.
luṅ (aorist) = news headline. Bare fact. Morphologically complex, rare in classical prose.
liṭ (perfect) = completed/remote. Uses reduplication (jagāma). Largely interchangeable with laṅ in classical texts.
Master laṭ and laṅ first. That's 80% of what you'll encounter.
Moods & Voices prayoga
Signals self-interest. Not reflexive — more "for own benefit." Some roots ONLY exist in middle: manyate (thinks), āste (sits), labhate (obtains). These are ātmanepadī dhātavaḥ.
Participles kṛdanta
Participles = verbs acting as adjectives. "The sleeping child, the broken pot." The past passive participle (PPP) is probably the single most common verb form in all classical Sanskrit.
| Type | Meaning | Formation | Example |
|---|---|---|---|
| Present Active | doing (ongoing) | stem + -ant/-at | gacchant- "going" |
| PPP (kt-kṛdanta) | was done / done | root + -ta/-na | gata- "gone" · kṛta- "done" |
| Gerundive | should be done | root + -tavya/-anīya | kartavya- "duty" (to be done) |
rāmeṇa sītā dṛṣṭā = "Sītā was seen by Rāma" — but it really means "Rāma saw Sītā." Hindi does the exact same thing: राम ने सीता को देखा — verb agrees with object. Direct inheritance.
| Root | Meaning | PPP | You know this from |
|---|---|---|---|
| √gam | go | gata | Bhagavad Gītā |
| √kṛ | do | kṛta | saṃskṛta = refined |
| √dṛś | see | dṛṣṭa | dṛṣṭi = vision |
| √jñā | know | jñāta | jñāna = knowledge |
| √bhū | become | bhūta | bhūta = being, creature, past |
| √budh | wake/know | buddha | that Buddha = "the awakened one" |
| √muc | release | mukta | mukti, mokṣa = liberation |
| √sidh | accomplish | siddha | siddhi, siddhānta |
Compounds samāsa
Sanskrit smashes words into compounds. Only the last word gets case endings. There are 4 main types.
| Type | Structure | Vigraha (expanded) | Compound |
|---|---|---|---|
| Tatpuruṣa | A of/for/from B | rājñaḥ puruṣaḥ | rāja-puruṣaḥ (king's man) |
| Dvandva | A and B | rāmaḥ ca kṛṣṇaḥ ca | rāma-kṛṣṇau |
| Karmadhāraya | A which is B | nīlam utpalam | nīlotpalam (blue lotus) |
| Bahuvrīhi | having A-B (possessive) | pītam ambaram yasya saḥ | pītāmbaraḥ = Viṣṇu |
The compound describes something external. daśānana = "ten-faced" = Rāvaṇa. cakrapāṇi = "disc-in-hand" = Viṣṇu. Test: if you can insert "the one who has ___" it's bahuvrīhi.
Vigraha = breaking a compound back into words with full vibhakti. Every compound can be vigraha'd. This is THE skill for reading real texts. When you hit a long compound: stop, vigraha it, then read.
Derived Forms pratyayānta
Sanskrit derives new verb types from any root — verb transformers that modify meaning systematically.
| Form | Does what | Formation | Example |
|---|---|---|---|
| Causative (ṇijanta) | make someone do X | strengthen + -aya- | √paṭh → pāṭhayati = teaches Hindi: पढ़ना → पढ़ाना |
| Desiderative (sannanta) | want to do X | reduplicate + -sa- | √jīv → jijīviṣati = wants to live |
| Absolutive (tvānta) | having done X, then... | root + -tvā | snātvā pūjāṃ karoti = having bathed, worships Hindi: नहाकर पूजा करता है |
| Infinitive (tumanta) | to do X | root + -tum | draṣṭum icchati = wants to see |
Sentence Structure vākya-racanā
Subject — Object — Verb (like Hindi, Japanese, Korean). But since cases mark roles, word order is for style/emphasis, not meaning.
| Word | Meaning | Word | Meaning |
|---|---|---|---|
| ca | and (placed AFTER) | vā | or |
| na | not | eva | indeed, only |
| iti | thus, "end quote" | api | also, even |
| tatra | there | atra | here |
| yatra | where | kadā | when |
| sadā | always | katham | how |
| iva | like, as | tu | but |
| hi | for, because | yadi | if |
rāmaḥ kṛṣṇaḥ ca = "Rāma and Kṛṣṇa" — NOT ca rāmaḥ kṛṣṇaḥ. Same for eva, api, tu, hi, vā. iti marks the END of quoted speech.
Real Text putting it all together
Everything above was theory. Here's what it looks like in practice — real śloka broken down word by word.
मामकाः पाण्डवाश्चैव किमकुर्वत सञ्जय॥
Every concept from this guide in ONE verse: saptamī (location), PPP (samavetāḥ), desiderative (yuyutsavaḥ), sandhi (pāṇḍavāścaiva = 3 words fused), laṅ past tense (akurvata), sambodhana (sañjaya), compounds (dharmakṣetra, kurukṣetra).