Grammar

The rules that build Classical Sanskrit words: sandhi, declension, conjugation and compounding.

Sanskrit grammar (vyākaraṇa, one of the six traditional Vedāṅga disciplines) was studied and codified from the later Vedic period onward. Its high point is Pāṇini's Aṣṭādhyāyī ("eight chapters"), from around the 4th century BCE: a complete, rule-based description of the language so compact that it can be recited in about two hours. Kātyāyana and Patañjali added commentary a few centuries later. Together the three are called the Trimuni Vyākaraṇa, "grammar of the three sages". The tradition continued through Bhaṭṭoji Dīkṣita's 17th-century Siddhānta Kaumudī and into 19th-century European Sanskrit scholarship. This page covers Classical Sanskrit, the language Pāṇini standardized. The older Vedic language behind this site's Sanskrit chants differs in several ways, which the Vedic Grammar page covers.

Sandhi

Sanskrit words change shape at their edges depending on what they touch. You have already seen this in the numerals (ṣaṣ → ṣaḍviṃśati) and in the Characters page's grid of places and manners. There are two kinds:

External sandhi

between two separate words

Similar vowels merge into one long vowel (a + a → ā). Dissimilar ones glide or form a diphthong (i + a → ya, a + i → e). A voiced consonant makes a neighbouring voiceless one voiced to match (regressive assimilation), which makes the join easier to pronounce.

Internal sandhi

inside a single word, at a root/suffix seam

This kind is more tightly constrained by the root's own structure. It uses the same place-of-articulation and voicing/aspiration logic, but where a suffix meets a root instead of where two words meet.

Vedic external sandhi is more variable and less fixed than the Classical rules. The Vedic Grammar page covers this and other differences.

Vowel gradation: guṇa & vṛddhi

This pattern is inherited from Proto-Indo-European ablaut (the same mechanism behind English sing/sang/sung/song), and it is the most important one in Sanskrit word-building: a root's vowel strengthens or weakens depending on what is being built from it. There are three grades, tied to the same five vowel places listed on the Characters and Numerals pages:

	a	i / ī	u / ū	ṛ	ḷ
Zero grade	—	i / ī	u / ū	ṛ	ḷ
Guṇa (1st)	a	e	o	ar	al
Vṛddhi (2nd)	ā	ai	au	ār	āl

As a rule of thumb, an accented root syllable takes guṇa or vṛddhi, and an unaccented one reduces to zero grade. भू (bhū, "become") shows all three: bhū- (zero), then bho- (guṇa, seen in bhavati, "he becomes"), then bhāu--type vṛddhi forms in further derivatives. This one alternation underlies a large share of verb-stem formation and of noun and adjective derivation.

Sanskrit's pitch accent (udātta, anudātta, svarita; the same marks drawn on the Rudram pages and listed on the Characters page) decided which grade a root took. It was still alive in Pāṇini's time but had left everyday speech within a few centuries of the Classical standard being set. See Vedic Grammar.

Nominal declension

Every noun and adjective is declined along three axes: 3 numbers (singular, dual, plural), 3 genders (masculine, feminine, neuter) and 8 cases. Adjectives simply take the gender, number and case of the noun they describe. Each of the 8 cases marks a different grammatical role:

Nom.

subject

Acc.

direct object

Instr.

by / with

Dat.

indirect object

Abl.

away from

Gen.

possession

Loc.

in / on / at

Voc.

addressing

One model paradigm shows the pattern: देव (deva, "god"), the most common noun class, a masculine stem ending in short -a. Every other stem class (i- and u-stems, ā- and ī-stems, consonant-final roots) takes the same 8 cases × 3 numbers with different endings, by the same logic.

	Singular	Dual	Plural
Nominative	देवः devaḥ	देवौ devau	देवाः devāḥ
Accusative	देवम् devam		देवान् devān
Vocative	देव deva		देवाः devāḥ
Instrumental	देवेन devena	देवाभ्याम् devābhyām	देवैः devaiḥ
Dative	देवाय devāya		देवेभ्यः devebhyaḥ
Ablative	देवात् devāt		देवेभ्यः devebhyaḥ
Genitive	देवस्य devasya	देवयोः devayoḥ	देवानाम् devānām
Locative	देवे deve	देवयोः devayoḥ	देवेषु deveṣu

The dual and several plural cells repeat on purpose. This is syncretism: one form covers several case and number slots, so it is part of the paradigm, not a gap. There are 24 case-and-number slots but only 15 distinct forms.

Verb conjugation

A finite verb form combines five independent choices: 3 persons, 3 numbers, 3 voices (active, middle, passive), 4 moods (indicative, optative, imperative, conditional) and 6 tenses (present, imperfect, perfect, aorist, future, periphrastic future). It is the same grid idea you saw in the numerals and declension, with more axes.

Every root builds its present-tense stem in one of ten traditional ways, the "ten classes". No single principle decides which root gets which class, but the ten fall into two families:

Class	Root	Treatment	Stem	Sample (3rd sg.)
1st	√bhū-	root accent, guṇated	bháv-	bhávati
6th	√tud-	none (ending accented)	tud-	tudáti
4th	√div-	-ya- suffix	dī́vya-	dī́vyati
10th	√cur-	-aya- + root gradation	córaya-	córayati
2nd	√ad-	none (athematic)	ad-	átti
3rd	√hu-	reduplication	juhó-	juhóti
5th	√su-	-no- suffix	sunó-	sunóti
8th	√tan-	-o- suffix	tanó-	tanóti
9th	√krī-	-nā- suffix	krīṇā́-	krīṇā́ti
7th	√rudh-	nasal infix	ruṇádh-	ruṇáddhi

Once the stem exists, the same personal endings attach in every class. Here are the first-class active (parasmaipada) present-tense endings, shown with भू (bhū → bhav-, "become"). Nearly half the roots in the language belong to this class:

	Singular	Dual	Plural
1st person	bhavāmi	bhavāvaḥ	bhavāmaḥ
2nd person	bhavasi	bhavathaḥ	bhavatha
3rd person	bhavati	bhavataḥ	bhavanti

A parallel set of middle-voice (ātmanepada) endings is used when the fruit of the action comes back to the subject. Beyond the finite forms above, there are five "secondary conjugations" (passive, intensive, desiderative, causative, denominative), each building a new stem from the root with its own meaning. There are also three non-finite forms: participles, a single infinitive, and the gerund.

Compounds (samāsa)

Joining several words into one is a defining habit of Sanskrit. It is inherited from Proto-Indo-European but taken much further: later literary Sanskrit builds compounds of a dozen or more members. Four relationships cover almost every compound:

Dvandva: "and"

coordinative: X and Y

rāma-lakṣmaṇau — "Rama and Lakshmana"

Tatpuruṣa: "of/for/in X"

determinative: first member modifies second by an implied case relation

rāja-puruṣaḥ — "the king's man" (genitive)
A descriptive subtype, karmadhāraya, fuses an adjective straight onto its noun: nīla-utpalam → nīlotpalam, "blue lotus."

Bahuvrīhi: "having X"

possessive: the whole compound describes something outside itself

pīta-ambaraḥ — "yellow-garmented", i.e. Kṛṣṇa (not the garment itself)

Avyayībhāva: "in the manner of X"

adverbial: the whole compound becomes an indeclinable

yathā-śakti — "according to (one's) ability"

Indeclinables & syntax

Prepositions, adverbs, particles, conjunctions and interjections never decline, so there is no case, number or gender to track. That makes them easy to underestimate. A few particles (sma, a-/an-) carry real grammatical weight; sma, for example, turns a present-tense verb into a past one.

Case endings, not word order, carry the grammatical relationships, so Sanskrit word order is mostly free. Pāṇini left syntax unfixed in the Aṣṭādhyāyī. In practice the usual order is subject, object, verb.

Sanskrit's vocabulary and constructions run very deep. The Sanskritist Daniel Ingalls found roughly fifteen ways to say "you must fetch the horse": you can mix active and passive, imperative and optative, add an auxiliary verb, or use any of three gerundive forms, and each gives a different metrical shape.