The term 'ellipsis' in grammatical theory most generally refers to linguistic material being omitted, deleted, or simply left unpronounced. The omitted material is needed for the full interpretation of a sentence, but it is not expressed because it can be recovered from the linguistic or real-world context (Winkler 2006: 109). Modern studies of ellipsis concentrate on formalizing the licensing and recoverability conditions for such elided constituents.

All natural languages permit ellipsis, but they differ with respect to which constituents can be elided in which configurations. For example, in English the focus has been on elided material in the context of coordination, as shown in the examples in (1). These constructions differ in terms of what is elided (whether a word or a phrase) and whether the ellipsis is discourse-bound (e.g. VP-ellipsis) or sentence-bound (e.g. gapping), resulting in a range of different types.

In Sanskrit, arguments seem to be freely omissible. So NP ellipsis, at least, is not syntactically constrained as it is in English. In modern work on the Indian grammatical tradition, the term 'ellipsis' tends to have a broader sense than the phenomena in (1). 'Ellipsis' is used for all linguistic phenomena which involve some sort of incompleteness of the surface structure or expression (Deshpande 1985, 1989). Within this broader use of the term, a distinction can be drawn between what is assumed as elided but does not obtain any special provision in Pāṇini's grammar, i.e. 'natural ellipsis', and the ellipsis phenomena that are explicitly governed by rules, i.e. 'prescriptive ellipsis' (Deshpande 1985, 1989). Other Sanskrit studies have focused on actual occurrences of ellipsis in coordination in corpus data (such as Gillon 2010), but we will focus here on the treatment in the grammatical tradition.

1. General notion of elision: lopa

The term used in the grammar to refer to elision or deletion in a general sense is lopa. Pāṇini defines lopa in Aṣṭ. 1.1.60: adarśanaṃ lopaḥ '(the technical term) lopa is defined as non-perception (of a speech-element)'. 'Non-perception' is itself defined in the Kāśikā as non-hearing (aśravaṇa) and non-pronunciation (anuccāraṇa). Lopa can 1) function as a substitute for a perceptible element of grammar, that is, it can function as a zero-substitute for any stem, suffix or segment; 2) it can refer to a situation where the meaning of a word is understood but the word is not used. The first is the 'prescriptive ellipsis' of Deshpande; the second is his 'natural ellipsis'.

The term lopa is most commonly used in Pāṇini's grammar to refer to 'prescriptive ellipsis', i.e. deletion, of phonemes or suffixes in grammatical derivations (Filliozat 1991: 678). For example, Aṣṭ. 6.1.66 lopo vyor vali states that lopa of v or y occurs before any consonantal segment except y; this rule accounts for the elision/deletion of stem- or root-final y before consonantal suffixes, as e.g. in the formation of the past participle ūta- 'woven' to the root ūy- 'to weave': ūy- + ta- → ūy- + ta- → ūta- 'woven'.

Lopa is a type of substitute (ādeśa), and the item which is substituted is called the sthānin. In the case of 'natural ellipsis', the term lopa is not commonly found, but the term sthānin is, suggesting the occurrence of lopa. For example, Pāṇini uses the term sthānin to refer to the optional non-occurrence (pro-drop) of the second person pronoun tvam in sentences like (2), in Aṣṭ. 1.4.105 yuṣmady upapade samānādhikaraṇe sthāniny api madhyamaḥ 'second person is used when co-referential with an attendant word 'you', even if substituted'. According to Kiparsky (1982), without this statement the elision of the pronoun in (2) would precede and bleed the verbal agreement rule, resulting incorrectly in a third person ending on the verb. In order to prevent this from happening, it is necessary to state explicitly that agreement applies even when the pronouns are deleted. Kiparsky argues that this is evidence that 'free deletion' interacts with the rules of Pāṇini's grammar.

The term sthānin is also used in Aṣṭ. 2.3.14 kriyārthopapadasya ca karmaṇi sthāninaḥ to account for the use of dative case in a construction suchas (3a). The rule states that dative is used to denote the object of an infinitive verb (e.g. āhartum 'to fetch') which has been elided(sthānin). That is, the verb 'go' itself cannot govern a purposive dative argument, although it can govern a dative expressing the goal of motion. But the grammaticality of (3a) alongside (3b) is understood with reference to an (assumed) ellipsed infinitive.

Although Pāṇini does not use lopa to refer to the phenomenon illustrated in (3a), Kātyāyana explicitly uses the term to refer to a similar construction, in his Vārttika 1 on Aṣṭ. 2.3.28 pañcamī-vidhāne lyab-lope karmaṇy upasaṃkhyānam 'in the prescription for the fifth case, an addition should be made (to allow the fifth case) for the patient when there is lopa of a gerund'. In this case the sentence in (4a) is assumed to derive from an underlying structure as in (4b), with deletion of the gerund.

One rule in which the term lopa is explicitly used to refer to 'natural ellipsis' of words in Pāṇini is Aṣṭ. 8.1.63 cādilope vibhāṣā, which indicates that when ca, vā, ha, aha or eva are elided, the first verb optionally retains its accent. For example, (5) involves two clauses in what we might consider to be asyndetic coordination; this is analysed as involving lopa of the coordinating conjunction ca 'and', and the optional retention of accent on the finite verb (which would otherwise be lost in this position in a main clause) is explained with reference to this (assumed) lopa.

Given the use of the term lopa for both 'prescriptive ellipsis' within the system of grammatical derivation and for 'natural ellipsis', i.e. omission of syntactic material, Deshpande (1989: 122) observes that at the time of Pāṇini there was no clear distinction between ellipsis/deletion as such and simple 'non-usage' (aprayoga) of an otherwise expected word.

2. Ellipsis of the copula

In Sanskrit, purely nominal sentences, that is sentences consisting purely of a subject noun phrase and a predicate noun or adjective phrase, but no explicit finite verb, are common. As in modern Western linguistics, the question arose as to whether such sentences should be understood to contain an ellipsed copula verb. In defining the notion of 'sentence', Kātyāyana, Vārttika 11 on Aṣṭ. 2.3.1 claims that every sentence must have a finite verb, and that hence in nominal sentences a copula must be understood: […] astir bhavantīparaḥ prathamapuruṣo 'prayujyamāno 'py asti 'a third person form of the verb as 'to be' in the present tense (i.e. asti [sg.], staḥ [du.], or santi [pl.]), even though it may not have been used (in the surface sentence), still exists (in that sentence).' For Kātyāyana, then, a sentence such as (6a) must be an elliptical form of (6b). Deshpande (1987) shows that Kātyāyana's position is different from that assumed in the Aṣṭādhyāyī, where it appears to be the case that Pāṇini does not require every sentence to necessarily involve a finite verb.

The Samanvayadiś, a non-Pāṇinian Sanskrit grammar on syntax from Kashmir dating to possibly c. 1100 AD, offers an interesting view with regard to the meaning of the supplied copula. Quoting part of Vārttika 11 on Aṣṭ. 2.3.1 (see above), it states that a verb asti 'is', bhavati 'is, becomes', or vidyate 'is, is found' is to be supplied where no verb is explicitly uttered. Regarding the meaning of such a supplied asti, bhavati, or vidyate, it offers two alternative views. Under the first (conventional) view, such a verb expresses a general existence (sattāsāmānyavācin); that is, it retains its basic copular/identificatory sense. The second view is more interesting: such a verb expresses, rather, a general action and is for indicating any particular verb (kriyāsāmānyavāci dhātūpalakṣaṇaparam). In other words, the supplied asti, bhavati, or vidyate can stand for any verb and is like a verbal equivalent of the pronoun, a 'pro-verb' much like some uses of English do. Which lexical verb or meaning is referred to by this 'pro-verb' depends on the context.

3. Ekaśeṣa: A special type of deletion?

Another case of elision discussed in the Indian grammatical tradition, starting with Pāṇini, is the notion of ekaśeṣa. Ekaśeṣa is introduced by Pāṇini in Aṣṭ. 1.2.64 sarūpāṇām ekaśeṣa ekavibhaktau 'there is remaining of one (ekaśeṣa) in case of multiple identical forms with the same case ending'. This is Pāṇini's approach to dual and plural number: granted an assumption that any single occurrence of a stem should have a single referent, dual and plural reference is derived by assuming as many individual occurrences of a stem as required for the number of referents, and then assuming deletion of all but one of these. For example, if we want to refer to two trees, we begin with two instances of vṛkṣa 'tree', underlyingly, and then delete one, as shown in (7). This process is different from 'natural ellipsis' in that it is explicitly prescribed by the rules of the system.

In the case of simple dual and plural marking, ekaśeṣa is an obligatory process, applying to sequences which would otherwise undergo compounding. That is, a dvandva compound like vṛkṣa-vṛkṣa- 'two trees' is not possible, in Pāṇini's system. But some ekaśeṣa processes are optional, as specified in Aṣṭ. 1.2.69-71. For example, Aṣṭ. 1.2.71 generates the form in (8b) from the one in (8a), by specifying that, optionally, only the masculine form remains when these two words co-occur in the same case. As this is optional, the compound in (8c) is also possible.

Although not directly referred to as lopa, ekaśeṣa 'remaining of one' evidently involves the same notion of underlyingly present linguistic material which is eliminated in the derivation of the surface form.

4. Later traditions and the interpretation of ellipsis

In Indian schools of thought, there are three requirements for any syntactic construction: saṃnidhi 'proximity', yogyatā 'appropriateness' and ākāṅkṣā 'expectancy'. The syntactic unity of the sentence depends mainly on ākāṅkṣā, or the mutual expectancy of the words.

Sometimes, even when some of the words are missing, they could be easily understood as implied from the context, allowing verbal comprehension. This means that ellipsis is conditioned mainly by the two conditions saṃnidhi 'proximity' (at least in the sense of mental proximity of the missing item to the explicit material) and ākāṅkṣā 'expectancy' (that is the syntactic and/or semantic expectancy of the explicit material for the missing item). According to Filliozat (1991), words should be close to one another in actual use, and if a word is not used it should be present in the mind in order to be understood. Hence, in cases of ellipsis the elided term is present in the mind because there is a communicable intention of the speaker or because it has been mentioned in a previous sentence (or even in a subsequent sentence).

In the Indian philosophical traditions of Mīmāṃsā, Nyāya, and Vedānta (dealing with ritual hermeneutics, logic, epistemology, metaphysics, among other topics), there are debates on exactly what is supplied in the case of an elliptical sentence. The Bhāṭṭa school of Mīṃāmsā as well as Vedānta and Nyāya adopt the view of padādhyāhāra 'supplying of words', arguing that one must supply the missing words in order to provide the missing meaning. This depends on a conception of meaning which ties it directly to words: linguistic meaning cannot arise without an associated word supplying that meaning. In contrast, the Prābhākara school of Mīmāṃsā adopts the view of arthādhyāhara 'supplying of meanings', arguing that one can directly supply meanings without invoking the missing words (Deshpande 1989: 114 and Raja 1958: 28).

For example, when the elliptical sentence consisting of only the accusative noun dvāram 'door' is uttered, as in (9), to attain the intended verbal knowledge of 'shut the door', under the padādhyāhāra view, one would supply a word (here an imperative verb) such as pidhehi, which then provides the meaning of 'shut'; under the arthādhyāhāra view, however, one would directly supply the meaning 'shut' without supplying any additional word.

Mīmāṃsā theorists argue that it is through arthāpatti 'postulation of a fact' that we cognize the omitted word or idea in an elliptical sentence. This same process applies beyond language; for example, on seeing that Devadatta, who is known to be alive, is not at home, his presence outside is presumed.

According to Raja (1958), there are two kinds of incomplete sentences: the normal elliptical sentence where the syntactic expectancy is not fully satisfied, and the syntactically complete sentence where the psychological expectancy is not fully satisfied. These are referenced by the terms adhyāhāra and vākyaśeṣa, respectively, by Bhoja in his Śṛṅgāraprakāśa. The first is the type seen in (9); the second is more like pragmatic inference. For example, in the sentence 'the road is full of thieves' an inferred meaning 'do not go that way' can be derived through vākyaśeṣa.

Bhartṛhari argues that if what appears to be part of another sentence is capable of conveying a complete sense in the particular context in which it is used, that is also to be considered as a complete sentence. For example, if dvāram is used to mean the same as dvāraṃ pidhehi 'shut the door', the former should not be considered a reduced version of the latter, but the two should be considered distinct sentences (Raja 1958). This contrasts with the assumption of adhyāhāra, of either variety, in such cases.

5. Ellipsis in coordination. See Coordination.