The term coordination refers to the juxtaposition of two or more words or phrases under a single phrasal node, often linked by a conjunction such as and or or. Unlike subordination, in which one element is asymmetrically embedded within another, coordination gives the appearance of being symmetric in many ways; for example, the conjuncts are typically all constituents of the same category. All languages appear to possess coordination constructions of some kind, but there is considerable cross-linguistic variation in how coordinate phrases are structured and marked. For instance, coordinate constructions may lack an overt conjunction (asyndetic coordination) or have some overt linking device (syndetic coordination) (Haspelmath 2006).

Modern linguistic work on coordination has focused on accounting for its interaction with other syntactic phenomena. For example, extraction out of a coordinate structure fails if it targets only one conjunct (1a), but is acceptable if the extraction targets all conjuncts equally (1b). Another area of interest is ellipsis; for example, the fact that when two sentences are conjoint, some material (often the verb) is missing in the second conjunct (1c). Another is Right Node Raising, the phenomenon in which a shared argument surfaces at the right periphery of a coordinate structure (1d). Finally, agreement patterns (feature resolution) in coordination show a wide range of complex and sometimes asymmetric behaviour cross-linguistically.

In the Indian grammatical tradition, we find discussions on the different values of the particle ca 'and' in a sentence, as well as the coordinate value of dvandva compound formations. Principles of gender and person resolution in coordinated structures also emerge from their specification in grammars, although they are not the topic of explicit comment.

1. Coordination in Sanskrit

The most common conjunction in Sanskrit is the particle ca 'and'. The rule in Pāṇini's Aṣṭādhyāyī (ca. 400 B.C.) which is central to the later discussion is Aṣṭ. 2.2.29 cārthe dvandvaḥ '(when two or more words ending in a case suffix are semantically related to each other (samartha) and) stand in a relation expressible by ca 'and', they are (optionally) made into a compound; and the compound so formed is called dvandva.' Purely on the basis of the Aṣṭādhyāyī one could assume a variety of different approaches for interpreting the meaning of ca (cārtha) and the conditions for dvandva compound formation, and this is indeed what we find discussed in the later tradition.[1]

In the tradition following the Aṣṭādhyāyī, two main rephrasings of rule Aṣṭ. 2.2.29 can be found. First, in his vārttikas (possibly ca. 250 B.C.), Kātyāyana rephrases the rule as yugapadadhikaraṇavacane dvandvaḥ 'a dvandva is formed when the items meant are simultaneously referred to.' For Kātyāyana this reformulation is necessary to account for cases where the sense of ca is understood but a compound should not be formed. An example of this is given by Patañjali: see below (2) (the words in issue are marked in bold). In this way, the notion of simultaneous reference becomes a constraint for compounding.

Patañjali (ca. 150 B.C.) rejects Kātyāyana's rephrasing; his own solution to account for cases such as the one found in (2) is to understand cārthe as cena kṛte 'rthe 'when the sense (of ca) is conveyed by the word ca'. Patañjali further enumerates (Mahābhāṣya 2.2.29.15 (1: 434.9−13)) four different senses of the word ca listed in the table below.

Table 1: Meanings of ca according to Patañjali

Note from the table above that in the case of samuccaya and anvācaya the conjunction of items referred to seems to be based on nouns found in two distinct sentences. This is not explicitly stated by Patañjali. We can infer that compound formation is not allowed in either of these two cases where the semantic connection indicated by ca crosses over from one sentence to another, as shown in the context presented by Kaiyaṭa in (3). The difference in meaning between samuccaya and anvācaya is determined by the context and whether plakṣa is considered dependent.

The other two types, itaretarayoga and samāhāra, do allow compound formation. Based on the examples provided by Patañjali, one can infer further (as Nāgeśa does) that dvandva formation is tied to the occurrence or otherwise of ca after each constituent member within a single sentence and under one action. That is, compounding is only possible when all nouns to be compounded are each followed by ca. If one or more nouns are not followed by ca, compounding is not licit (= samuccaya or anvācaya).

Under Patañjali's reformulation of cārthe there is also no need for a special rule to distinguish between itaretarayoga and samāhāra, as was the case with Kātyāyana's reformulation, since the dual and plural is used 'naturally' in itaretarayoga, e.g. (4b), given the sense of 'each accompanied by the other'; and the singular in samāhāra, e.g. (5b), given the sense of 'a single whole (unitary meaning) consisting of both'.

The four functions of ca given above are the most widely found in the Indian tradition, and they are characterised by referring to item coordination. Beyond these four, the tradition teaches further meanings of ca. For instance, the Sanskrit thesaurus Amarakośa (ca. 500 A.D.) lists these four (terming the third itaretara instead of itaretarayoga) and adds a fifth, pādapūraṇa 'filling of a verse-quarter' (this function is also given to the particles tu, hi, sma, ha, and vai). This is a purely metrical function; that is, the particle ca 'and' is sometimes used in metrical texts merely to fill the metre, and with no functional load. In addition, Vardhamāna's Gaṇaratnamahodadhi (ca. 1150 A.D.), a commentary on a gaṇapāṭha, ascribes nine meanings to the word ca. These meanings are listed in (6):

2. Gender resolution

Gender resolution in the Aṣṭādhyāyī is dealt with under the purview of a process called ekaśeṣa 'remaining of one'. Ekaśeṣa reflects a strict view of the denotation of stems: the underlying assumption is that one occurrence of a stem necessarily has a single, and singular, reference. Dual and plural reference must then derive from simplification of a sequence of multiple occurrences of the same stem. For example, if we want to refer to two trees, we begin with two instances of the stem vṛkṣa 'tree', underlyingly, and then delete one, as shown in (7), before adding the dual ending.

Cases of gender resolution necessarily involve either dual or plural reference. For example, we might want to say that Rāma (masculine) and Sītā (feminine) are worthy of worship (pūjanīya). This can be expressed in Sanskrit as (8).

The word pūjanīyau here must be treated as derived by ekaśeṣa from a sequence of two singular forms, one masculine (referring to Rāma) and one feminine (referring to Sītā): pūjanīyaḥ pūjanīyā. This is reduced to pūjanīyau by the same process exemplified above for vṛkṣau in (7), but in this case we must also account for the gender. The question is, if only one of the words pūjanīyaḥ and pūjanīyā remain under ekaśeṣa, is it the masculine or the feminine? Aṣṭ. 1.2.67 pumān striyā teaches that when masculine (pumān) and feminine (strī) occur together in the case of ekaśeṣa, the masculine remains. That is, masculine and feminine resolve to masculine, as in (8).

In the same way, Aṣṭ. 1.2.69 napuṃsakam anapuṃsakenaikavac cāsyānyatarasyām teaches that in the case of ekaśeṣa of a neuter and a non-neuter, it is the neuter that remains. For example, given the following coordinated phrase in (9),[2]

to refer to all three as white one must assume the following, in (10).

Given that Aṣṭ. 1.2.67 specifies that masculine trumps feminine, and Aṣṭ. 1.2.69 that neuter trumps non-neuter, we can conclude that specifies the following hierarchy for gender resolution neuter > masculine > feminine.[3]

3. Person Resolution

In contrast to the treatment of gender resolution, there are no rules in the Aṣṭādhyāyī explicitly treating situations where different grammatical persons interact. However, the rules treating person occur in a section of the grammar which is governed by Aṣṭ. 1.4.2 vipratiṣedhe paraṃ kāryam 'in case of conflict the later rule applies'. In particular, Aṣṭ. 1.4.105 specifies second person endings in the case of an (explicit or implicit) second person reference, and Aṣṭ. 1.4.107 specifies first person endings in the case of an (explicit or implicit) first person reference. By virtue of being the later rule, only Aṣṭ. 1.4.107 will apply. For example, in (11), we find both first and second person reference, so that both Aṣṭ 1.4.105 and 107 can apply. By virtue of being the later rule, only Aṣṭ. 1.4.107 applies, so that the first person is chosen for the verb. Here, then, we have first person and second person resolving to first person.

The third person is taught in Aṣṭ. 1.4.108 śeṣe prathamaḥ, which specifies it in a context referred to by the word śeṣe 'in the remainder', namely where there is neither second person nor first person reference. It follows from this that the third person is the lowest on the person hierarchy: 1 > 2 > 3.

We take the account presented here to represent Pāṇini's original model for person resolution. Other interpretations have been proposed, such as the traditional solution, utilizing Aṣṭ. 1.2.72 tyadādīni sarvair nityam, and the recent proposal of Mocci and Pontillo (forthcoming). See the latter paper for details on both.[4]

While the Aṣṭādhyāyī treats person resolution only inferentially, the Kātantrasūtra (K) and the Saṃkṣiptasāra, two non-Pāṇinian grammars, have sūtras explicitly teaching person resolution in coordination. K 3.1.3 trīṇi trīṇi prathamamadhyamottamāḥ teaches the technical terms prathama 'third person', madhyama 'second person', and uttama 'first person' to the verb endings. The next sūtra, K 3.1.4 yugapadvacane paraḥ puruṣāṇām, teaches that when occurring simultaneously, the later term in K 3.1.3's prathama-madhyama-uttama prevails—in other words, it teaches the following person hierarchy: 1 > 2 > 3. Saṃkṣiptasāra 2.21 teṣāṃ paraḥ kriyākālaikatve (p. 138) teaches essentially the same thing: when two or more actions occur at the same time, the later term among prathama-madhyama-uttama prevails—in other words, the same person hierarchy 1 > 2 > 3.

4. Summary on resolutions

We see then, that there is no single method of specifying feature resolution in the Aṣṭādhyāyī or the Indian grammatical tradition more generally. While the hierarchy of gender resolution in the Aṣṭādhyāyī is dealt with by rules explicitly treating co-occurrence of different genders, person resolution is dealt with by precedence relations between rules, leading to an implicit hierarchy. This contrast is likely due to the differing status of these features in the grammar. While the person feature has to be defined and specified in the grammar in any case, gender is not a general part of the grammatical specification. So the rules that specify person are placed within a particular part of the grammar where their relative order can affect the desired hierarchy in cases where resolution is needed. But there are no such rules for gender, leading to the need for separate rules which serve only to control gender resolution where needed.

5. Ellipsis in coordination

See Ellipsis. See also Agreement.

[1] For the following discussion, cf. (in part) Roodbergen (1974: 138-194).

[2] Aṣṭ. 1.2.69 additionally states that when there is resolution of neuter and non-neuter to neuter, the resulting form may optionally take singular number, so śuklam is a possible alternative outcome in (10).

[3] This is the general pattern. Additional specifications relating to gender resolution in particular semantic fields are treated in Aṣṭ. 1.2.66 and Aṣṭ. 1.2.73. Hock (2012) compares the complex patterns of feature resolution found in Vedic Sanskrit with Pāṇini's prescriptions, and finds some differences.

[4] Compared with the Pāṇinian hierarchy, we find diverging practices in Sanskrit texts, e.g. (12) - (14).

Note that while the Sanskrit in both (12) the Mahābhārata and (13) the Śukasaptatī is not considered as of a 'high' register, the Uttararāmacarita in (14) is thought to be written in good, 'high' Classical Sanskrit. A more comprehensive study of the evidence from Sanskrit literature is needed to understand the variation found here.