In modern linguistics, the term control refers to a construction (or family of constructions) in which there is coreference between an argument of a matrix predicate and an argument of a(n often nonfinite) predicate embedded under that matrix predicate. For example, in (1), there is a single overt NP (Devadatta) which surfaces as a syntactic and semantic argument of the matrix predicate (try), but which also functions as the semantic argument of the embedded predicate (understand). In early Transformational Grammar, control phenomena were analysed in terms of Equi-NP deletion (i.e. deletion of one of the two identical NPs) (Landau 2013). Control is now understood as an anaphoric relation between the matrix argument and a null pronoun in the subject position of the embedded verb.
|Devadattai tried [ (proi) to understand Pāṇini's grammar. ]|
A related construction is that seen in (2), called raising, where the NP Devadatta is the syntactic subject of the matrix predicate (seem), but functions as a semantic argument (here experiencer) of the embedded predicate and is not a semantic argument of the matrix predicate. The transformational understanding of this construction is that the argument in question starts out in the embedded clause and “raises” to the subject position of the matrix.
|Devadattai seemed [ (ei) to understand Pāṇini's grammar.]|
The constructions in (1) and (2) are similar in the fact that there is a single NP which surfaces in the main clause but syntactically or semantically functions as the subject of the embedded verb; the differences lie in the selectional properties of the main verbs.
In the Indian grammatical tradition, there was a similar concern with understanding such shared arguments, with particular reference to embedded infinitival complements (with the infinitive -tum) and embedded adjunct absolutive clauses (primarily with the absolutive -tvā). The discussion centers on the 'identical agents condition', and the meaning of the nonfinite form as denoting either the agent (kartṛ) or the action (bhāva). However, the treatment of each of these forms (absolutive and infinitive) differs to some extent.
1. The Sanskrit Data
1.1. The absolutive
One of the two main control constructions found in Sanskrit involves an embedded absolutive adjunct clause (-tvā). For example, in (3), the main verb is pibati 's/he drinks' and the absolutive verb is gatvā 'having gone'. According to Aṣṭ. 3.4.21 samāna-kartṛkayoḥ pūrva-kāle, when two actions have the same agent (kartṛ), the affix -tvā comes after the verb referring to the earlier of the two. Hence, the tradition understands two conditions for the use of the absolutive in this rule: temporal precedence (pūrva-kāla) and both the finite and nonfinite forms share the same agent (samāna-kartṛ), which is rāma in (3).
|'Having gone to the village, Rama drinks water.'|
1.2. The infinitive
The other kind of control constructions found in Sanskrit involve an embedded infinitival complement clause (-tum). According to Pāṇini, the infinitive -tum has five main functions: (i) expressing purpose of another action, e.g. (4a); (ii) with words denoting 'enough', e.g. (4b); (iii) with words denoting 'time', e.g. (4c); (iv) with verbs of 'desire', e.g. (4d); and (v) with verbs of 'ability', e.g. (4e). Of these five constructions, the tradition has focused on the control constructions (4d) and (4e), and to some extent the purposive use in (4a).
|'Rama cooks rice to eat.'|
|b.||rāmo||gantum||alam||(with words 'enough')|
|'Rama is able enough to go.'|
|c.||gantuṃ||kālaḥ||(with words 'time')|
|'(It is) time to go.'|
|d.||rāma||icchaty||odanaṃ||bhoktum||(with verbs of 'desire')|
|'Ramai wants (ei) to eat rice.'|
|e.||rāmaḥ||kaṭaṃ||kartuṃ||śaknoti||(with verbs of 'ability')|
|'Ramai is able (ei) to make a mat.'|
Identical Agents Condition
As mentioned above, the condition of identical agents is stated for the use of the absolutive (-tvā) according to Aṣṭ. 3.4.21 samāna-kartṛkayoḥ pūrva-kāle. Example (5) illustrates an absolutive embedded in a main clause with an active finite verb: here, devadattaḥ is both the agent of the subordinate action of cooking denoted by the absolutive (paktvā), and the agent of the action of eating expressed by the main verb (bhuṅkte).
|'Having cooked (it), Devadatta eats the rice.'|
The crucial question is the case-marking on the nouns, in particular devadatta. As described in 'Argument structure', a kartṛ ('agent') by default receives an instrumental case ending, unless the kāraka role is denoted by the verb, in which case the nominative ending is used, by Aṣṭ. 2.3.46 prātipadikārtha-liṅga-parimāṇa-vacana-mātre prathamā. In (5), the finite verb ending denotes the agent role which devadatta bears with respect to the action expressed by the main verb. Following Aṣṭ. 3.4.67 (kartari kṛt 'the affixes called kṛt are used in the sense of a kartṛ 'agent''), the absolutive suffix -tvā, which is a kṛt affix, likewise denotes the agent role that devadatta bears with respect to the action expressed by the absolutive. Thus both agent roles associated with devadatta are denoted by verbal endings, and so the nominative case applies, correctly, to devadatta.
The infinitive -tum, like the absolutive -tvā, is a kṛt affix; hence, it should also express the agent. For infinitives, the condition of identity of agents is explicitly stated only for infinitival constructions with verbs of 'desire', according to Aṣṭ. 3.3.159 samāna-kartṛkeṣu tumUN 'the affix -tumUN is used after a verbal root used in conjunction with a verb meaning 'to desire', provided that the agent (kartṛ) of both actions is identical.' Hence, in (4d), repeated here as (6), icchati 's/he wants' and bhoktum 'to eat' must share the same agent, Rāma.
|'Ramai wants (proi) to eat rice.'|
Deshpande (1980) argues that the identical agent condition is not limited to infinitives with verbs of 'desire', but can be extended to cover uses of the infinitive as a complement to other verbs too, since there are hardly any usages of -tum found in Sanskrit literature which violate this condition (see von Böhtlingk 1887, 1888, Speyer 1896: 67, and Hook 1980 for exceptions). This is seen in example (7), in which rāma is the agent of both the infinitive (bhoktum) and the main verb (śaknoti). This construction is licensed by Aṣṭ. 3.4.65 śaka-dhṛṣa-jñā-glā-ghaṭa-rabha-labha-krama-sahārhāsty-artheṣu tumun which comes under the general rule Aṣṭ. 3.4.1 dhātu-sambandhe pratyayāḥ. This rule refers to the use of the affix -tumUN after a verbal root used in conjunction with verbs meaning 'to be able to', etc. when there is some relation between the two actions (dhātu-sambandha).
|'Ramai can (proi) eat rice.'|
As with the absolutive example in (5), the two examples with infinitives in (6) and (7) work perfectly well on the assumption that the non-finite verb ending denotes the agent: the subject, rāma in both examples, has two agent roles, but both are denoted by the respective verb endings, and so rāma gets nominative case, correctly.
When the absolutive appears in a clause with a passive main verb, as in (8), the situation is more problematic. In (8), the absolutive -tvā should still denote the agent (kartṛ), but the passive verb ending -te in bhujyate denotes the object (karman), according to Aṣṭ. 3.4.69 laḥ karmaṇi ca bhāve cākarmakebhyaḥ and Aṣṭ. 1.3.13 bhāva-karmaṇoḥ. Hence, there is a conflict. Devadatta is the agent of both actions, but only one of its agent roles is expressed by a verbal ending (its role with respect to the action of the absolutive, expressed by the ending of the absolutive). In the same way, the rice in (8) is the karman of both actions; one of these roles is expressed by a verbal ending (its role with respect to the action of the finite verb, expressed by the finite verb ending), while the other role (its role with respect to the action of the absolutive) is not. The question is then how to determine the case marking. Odana 'rice' should appear in the nominative, and devadatta in the instrumental, as shown in (8), but as stated both words have one of their roles expressed and the other unexpressed by a verbal ending, so the question is how to resolve this conflict in one direction in the case of odana, but in the other direction in the case of devadatta.
|'Rice, having been cooked, is eaten by Devadatta.'|
Pāṇini does not address this conflict, but Kātyāyana proposes that one should consider the absolutive -tvā (and the infinitive tum) as denoting the same thing (adhikaraṇa) which is denoted by the main finite verb in the same sentence. On the rule Aṣṭ 3.4.26 he adds the specification ā ca tumUNaḥ samānādhikaraṇe 'suffixes up to tumUN (i.e. including the absolutive and infinitive) denote the same thing (i.e. kāraka) denoted by the main verbal affix' (Deshpande 1981). In other words, the absolutive would be interpreted as passive when the main verb is passive, and hence denote the object (karman) and not the agent (kartṛ).
Kātyāyana's proposal works as long as the main verb and embedded verb do not have different objects. So, in (8), if we take the absolutive as denoting the karman role, then we can say that both karman roles of odana are expressed by verbal endings, so it should get nominative case, while neither kartṛ roles of devadatta are expressed by verbal endings, meaning that it should get instrumental case.
With infinitival complements, if the main verb and infinitive both take an object, that object must be shared; see the example in (9), which is the passive of (6). But even in infinitival complements we can have some complications. For example, in (10), passives of (7), the main verb is a form of śak, which is inherently intransitive (akarmaka). But in (10a), odana appears in the nominative case, apparently as the object (karman) of the passive form of śak, which is what we would expect for transitive main verbs (as in (8)). Alternatively, odana may appear in the accusative case, as in (10b). Deshpande (1980, 1981) and Kiparsky (2009) address these constructions, and the question of how the argument structure of śak + infinitive can be understood so as to license these structures.
|'Rice is wanted to be eaten by Rama.'|
|'Rice is wanted by Rama to eat.'|
|'Rice is wanted by Rama to eat.'|
Turning back to absolutives, it is not necessary for absolutive and finite verb to share the same karman; see example (11).
|'Having gone to the village, Rama drinks water.'|
|'Having gone to the village, the water is drunk by Rama.'|
Kātyāyana's proposal that the ending of the absolutive denotes the same kāraka as the finite verb fails here: it cannot account for the accusative in (11b). The finite verb in (11b) is passive, so must denote its karman, the water (jala). But if the absolutive follows the finite verb, then it would have to denote its karman, the village (grāma), and then grāma should occur in the nominative case, which it does not.
Patañjali points out the problems in Kātyāyana's proposal and states that the absolutive (and also the infinitive) neither denotes agent (kartṛ) nor object (karman), but invariably denotes only bhāva 'action' (avyayakṛto bhāve bhavanti). So, according to Patañjali's analysis, in example (11a) the affix -ti on the main verb denotes the agent (kartṛ), while the absolutive affix -tvā denotes action or bhāva, while in (11b) the suffix on the main verb denotes the karman, and the absolutive affix again denotes bhāva. This however causes a problem in (11a): the agenthood of rāma with respect to the finite verb is denoted by the finite verbal ending, but its agenthood with respect to the action of the absolutive is not denoted by a verbal affix. We once again have a conflict, and Patañjali does not explain how we resolve this in such a way that rāma gets the expected nominative case.
Bhartṛhari proposes that the case endings of nouns which have multiple kāraka roles in relation to different verbs are determined by their kāraka relation to the main verb (Deshpande 1981). So, in example (11a), since the agenthood of rāma with respect to the main action of drinking is expressed by the verbal affix -ti in the main verb pibati, his agenthood with respect to the subordinate action of going (gatvā) 'appears', i.e. is treated, as if it has been expressed (abhihitavat prakāśate), even though it has not been expressed by the absolutive -tvā (or by any other element). The accusative grāmam in (11b) and odanam in (10b) depend on the nonfinite forms and not the main verbs. By recognizing a hierarchy between main (pradhānakriyā) and subordinate (guṇakriyā) actions, all case-marking problems are resolved.
Kauṇḍabhaṭṭa also states, in agreement with Patañjali, that the absolutive -tvā (as well as the infinitive -tum) does not express the agent (kartṛ), but rather action (bhāva). Following Bhartṛhari, he maintains that in examples like (5) and (8), odana 'rice' is the object of both actions, and devadatta is the agent of both actions. When the sense of agent (in the active) or object (in the passive) is expressed by the main verb, the agent/object appears in the nominative. The question of whether a kāraka role is abhihita 'expressed' or anabhihita 'non-expressed' by the ending of the main verb determines which case endings are to be added to a nominal stem, in preference to any question of a kāraka relation to a subordinate verb. For Kauṇḍabhaṭṭa, the notion that main and subordinate verb necessarily share the same kartṛ, as well as notions such as the temporal precedence of the absolutive, derive contextually.
 Aṣṭ. 3.3.10 tumun-ṇvulau kriyāyāṁ kriyārthāyām.
 Aṣṭ. 3.4.66 paryāpti-vacaneṣv alam-artheṣu.
 Aṣṭ. 3.3.167 kāla-samaya-velāsu tumun.
 Aṣṭ. 3.3.159 samāna-kartṛkeṣu tumun.
 Aṣṭ. 3.4.65 śaka-dhṛṣa-jñā-glā-ghaṭa-rabha-labha-krama-sahārhāsty-artheṣu tumun.
 This rule states that if a kāraka has been already denoted by another affix (Aṣṭ. 2.3.1 anabhihite), then the nominative case ending is added after the relevant nominal stem, denoting merely the meaning of the stem, gender, and number.
 The Samanvaya tradition also discusses constructions with shared karmans (e.g. Samanvayapradīpasaṃketa 55 [Hahn 2008: 200 ll. 6-12 and 202 ll. 1-2]).