1. Introduction

Argument structure concerns the processes and relations involved in the syntactic realization of the semantic dependents of a predicate. Crucial notions in many treatments of argument structure in modern Western linguistics are 1. the notion of grammatical functions or grammatical relations, such as subject and object (including where structurally defined); and 2. the notion of semantic or thematic roles, such as agent, patient, etc. Argument structure concerns the interrelation of these two. Consider the following sentences:

The first argument in each of the sentences is grammatically the subject, but the semantic roles of each argument in relation to its verb are different: experiencer in (1a), agent in (1b), and theme in (1c). While the theme is the subject in (1c), the theme in (1a) is the object; the object in (1b) has a different semantic role again, namely patient.

These differences depend on the specific syntactic and semantic properties of the respective verbs. But any model of argument structure must also account for phenomena like passivization and causativization, processes which systematically alter the links between semantic roles and grammatical relations for some or all verbs or predicates in a language. For example, in the passive of (1a) the experiencer is no longer the subject, but either an (optional) oblique argument or an adjunct, while the theme is not an object but the subject.

Modern approaches to argument structure depend in one way or another on the seminal work of Fillmore (1968), who first proposed a notion of “Case” relations, effectively the modern notion of semantic roles. Case relations depend on the semantic entailments of a predicate, but are (deep structure) grammatical categories which relate to surface grammatical relations such as subject and object. For example, Fillmore (1968: 60) proposes the following generalization for the 'unmarked' subject choice in English:

Perlmutter and Postal (1977) rejected the early structural accounts of argument alternations like passivization, and sought to account for these processes rather in terms of operations on grammatical relations. They generalize the passive operation thus:

From these early, and in hindsight very preliminary, steps, much more sophisticated and descriptively adequate models have been developed. For example, it is vital not merely to acknowledge sets of different grammatical functions and semantic roles, but also to recognize that these roles stand in hierarchies, so that reference can be made to concepts like higher or lower grammatical functions, or the highest semantic role of a predicate. Many modern treatments of passivization, for example, follow Ostler (1979: 108) in treating it as an operation to “delete the highest [semantic] role” in the argument structure of a predicate.

In these terms we can revisit the alternation between (1a) and (2) above. If, for the sake of illustration, we assume a principle that 'the highest semantic role becomes the subject', then (1a) is immediately accounted for, granted that experiencer is 'higher' on the hierarchy of semantic roles than theme. Given this principle, the passive in (2) must involve the deletion of the highest semantic role, following which the highest remaining semantic role is that of theme, which can now become the subject. If the experiencer was deleted from the argument structure of the verb, it can appear only as an adjunct phrase, hence its appearance in the optional 'by' phrase in (2).[1]

2. Case and grammatical functions in Sanskrit

For the most part, Sanskrit is a relatively standard old Indo-European language in relation to argument structural phenomena. It is a synthetic language distinguishing case, number and gender on nouns, and marking person and number of the subject on finite verbs. There are eight cases: nominative, accusative, instrumental, dative, ablative, genitive, locative and vocative. At its core, Sanskrit has nominative-accusative alignment: the sole argument of intransitive verbs (S) is treated in the same way as the agentive argument of transitive verbs (A), in distinction from the patientive argument of transitive verbs (O). Subjects of finite verbs (A/S) are nominative, while objects (O) are usually accusative. The instrumental, dative, ablative and locative are primary semantic rather than structural cases. As a nominative-accusative language, Sanskrit has a passive construction, in many ways parallel to the passive familiar from languages like English.

Alongside this nominative-accusative core, however, Sanskrit also shows some evidence of ergative alignment. The past participle, which becomes increasingly frequent and one of the most common ways of expressing a simple past tense, agrees with the sole argument S when formed to intransitive verbs, but with the patientive argument O when formed to transitives. When a past participle is used as the head of a finite clause, then, S and O arguments pattern together, appearing in the nominative case, while A is treated differently, appearing in the instrumental; this is the origin of the ergative perfective in modern Indo-Aryan languages.

There are additional complications and details here which go beyond the scope of the present discussion, and the construction with the past participle will not come into the discussion below. But the presence of this ergative construction in Sanskrit is worth bearing in mind, particularly given the apparent lack of a notion of 'subject' in the Indian tradition, a notion which is entirely natural and unproblematic in exclusively nominative-accusative languages, but which is problematized precisely by ergative phenomena.

3 Argument structure in the Aṣṭādhyāyī

Pāṇini does not use any concept akin to grammatical functions in his model of argument structure, but is crucially concerned with accounting for the relations between the semantic dependents of a verb and the surface case marking of those dependents. Just as in modern theories of argument structure, Pāṇini's system involves an intermediate level between the two, somewhat comparable to the 'Case' relations of Fillmore (1968), but more syntactically and less semantically determined. The relations of this intermediate level are called kārakas. The kārakas are grammatical categories defined in terms of the semantic relation between a verbal action and its arguments. The kārakas are defined in the following rules:

These rules are in a section headed by Aṣṭ. 1.4.1–2, according to which whenever two or more technical terms can be applied to a single item, only a single term, the later one, can be applied. This creates a kind of hierarchy among the kārakas, in that the kārakas defined in later rules trump those defined in earlier rules, where both could apply. For example, if one is talking about cutting with an axe, one might ordinarily say:

An axe is an instrument of cutting, in Pāṇinian terms the 'most effective means', designated karaṇa, while the independent agent, the kartṛ, which in the active voice will get nominative case, is the human, Devadatta. But it is also possible to say:

Here, since it receives nominative case and the verb is active, the axe must be the kartṛ. The tradition explains this partly by recourse to the perspective or scope of the expression, and partly to speaker intention: it is possible to restrict one's perspective, or the scope of an expression, to a sub-part of an event; in this case, if we exclude the human agent from consideration, a speaker may wish to speak of the axe as the independent actor. The point is that given such an understanding, the axe can qualify for both the label karaṇa and the label kartṛ, but since kartṛ is defined later, only that label will apply. For more detailed discussion, see Cardona (1974: 233ff.); see also 'Subjecthood'.

The kārakas mediate between the semantic relations of arguments with their verbs and the morphosyntactic case marking of those arguments. There is a default relation between kārakas and case endings, summarized in the following table (specified in Aṣṭ. 2.3.2, 2.3.13, 2.3.18, 2.3.28, and 2.3.36):

The kārakas are not directly equivalent to cases, however, since these default relations can be overridden by more specific rules. For example, Aṣṭ. 2.3.3 specifies the instrumental to designate the karman with the verb hū in Vedic, overriding the default association of karman with the accusative (Aṣṭ. 2.3.2):

More importantly, there is no default relation between a kāraka and the nominative case, but most clauses in Sanskrit have a nominative subject. By Aṣṭ. 2.3.1 (anabhihite), which governs the rules containing the default kāraka-case relations summarized in Table 1, these defaults apply only if the relations are not denoted by some other element. By Aṣṭ. 3.4.69 (laḥ karmaṇi ca bhāve cākarmakebhyaḥ), the abstract tense markers are introduced to denote the kartṛ or (in the passive) karman (or the action of the verb itself, with intransitives). Thus the abstract tense markers, which are ultimately replaced by the person/number suffixes on the verb, effectively cancel the default association of the kartṛ with instrumental case, or – in the passive – of the karman with accusative case. In the lack of other case specification, the nominative case applies (by Aṣṭ. 2.3.46). This neatly obtains the nominative case marking of the kartṛ in the active and of the karman in the passive.

As discussed by Kiparsky (2009), this system depends on three principles: every kāraka must be expressed by some morphological element (e.g. verb ending or case ending); no kāraka may be expressed by more than one morphological element; every morphological element must express something.

Argument alternations

For instance, consider the following active sentence:

Since the verb is active, the verb ending denotes the kartṛ relation, so Devadatta appears in the nominative; the nominative expresses features like number, but no argument relation. The rice appears in the accusative, since it is the karman. But the same rules which licence this active sentence also permit a passive derivation. If instead of denoting the kartṛ with the verbal suffix, we choose to denote the karman with the verbal suffix, then the verb will appear in passive form (Aṣṭ. 1.3.13, 3.1.67), and the karman will appear in the nominative, while the kartṛ will appear in the instrumental:

In this model, there is a free choice between active and passive: unlike in modern approaches to the passive, neither is primary, and the passive is not derived from the active. See also 'Passive'.

The causative can also be derived mostly via rules already introduced above. There are two possibilities for the causative of a transitive verb in Sanskrit: in the active, the causer is always the nominative subject argument, and the object of the non-causative construction retains its object role (and case, usually accusative) in the causative, but what was the original subject of the non-causative surfaces either in the instrumental or the accusative.

The causer is designated both hetu and kartṛ by Aṣṭ. 1.4.55 above. The verb ending denotes this kartṛ, and hence the causer appears in the nominative. The original karman of the corresponding active sentences remains unchanged, and so appears in the accusative. By the rules of default case assignment given above, the original kartṛ of the active sentence, which remains a kartṛ in the causative (but is distinct from the hetu-kartṛ), receives instrumental case. This accounts for (11a).

(11b) requires a specific rule to license it. Aṣṭ. 1.4.52, given above, states that in the causative of certain verb classes the non-causative kartṛ is given a new designation of karman. This means it will get accusative case by the default case assignment rules. Interestingly, this rule implies a conception of derivation, or at least alternation, between non-causative and causative: a causative (at least in the specified verb classes) necessarily implies the existence of a non-causative, and in some sense the latter is the more basic or primary formation.

4 Argument structure in the non-Pāṇinian tradition

The Pāṇinian model of argument structure was never bettered in other strands of the Indian tradition. In some non-Pāṇinian grammars, the kāraka model is essentially adopted without change, or with certain apparent simplifications. For example, in the Kātantra exactly the same set of kārakas is assumed, but they are defined in somewhat circular terms:

Here the kārakas apādāna, sampradāna, and karaṇa are defined using a relative construction with the default case specification for that kāraka. That is, for example, apādāna, which by the later case specifications is linked with ablative case, is itself defined in terms of the ablative case. (Case specifications are given in K 2.4.17–42.) The karman is effectively defined as the subject (or more precisely, nominative argument) of a passive verb, and similarly the kartṛ and hetu are respectively defined as the nominative arguments of an active non-causative and causative.

Other non-Pāṇinian grammars avoid the kārakas altogether, attempting to specify a more direct connection between semantic relations and case. For example, Candragomin's Cāndravyākaraṇa adopts two approaches. Some rules directly connect case endings with meanings, for example:

In other cases, names corresponding to the Pāṇinian kārakas are used, but these are not given any semantic definition, and are directly linked with case endings. For example, Candragomin directly links the accusative case endings with āpya 'that which is reached' (his term corresponding to Pāṇini's karman), while the genitive case is defined as expressing 'relation':

Now consider the following sentences:

The verb smṛ 'remember' can take an object in either the accusative or genitive case.[3] For Pāṇini, in both cases the object is the karman; the realization of the karman in the accusative follows from the default specifications introduced above; the option of realising the karman with the genitive case with verbs of remembering is specified by Aṣṭ. 2.3.52. For Candragomin, however, it is not possible to treat this, or comparable alternations, in terms of variable realization of the same underlying argument type. Rather, he is forced to have recourse to vivakṣā 'the intention of the speaker'. Since in (15a) the mother is given accusative case, the intention of the speaker of such a sentence is to indicate that the mother is the kriyāpya, the entity reached through the action of remembering. On the other hand, since in (15b) the mother appears in the genitive, the indication is merely that the mother is related to the action of remembering. As shown by Joshi and Roodbergen (1975: xvi–xix), this reliance on speaker intention is unable to properly constrain the grammar; essentially, we must assume that the only vivakṣā a speaker will use is one that will give a result in accordance with acceptable usage. In contrast, Pāṇini's kāraka system constrains the grammar by explicitly restricting its possible outputs.

[1] This is an intentionally simplified illustration, merely aiming to show the basic aims and mechanics of argument structure models.

[2] This is the traditional interpretation, but Kiparsky (2009) argues that it is better to interpret this as referring to a karman which is ellipsed.

[3] It is not clear from the Aṣṭādhyāyī itself whether Pāṇini intended for the construction with the genitive case to entirely block the construction with the accusative case, or whether he intended both constructions to exist alongside one another. We follow the traditional interpretation, as found e.g. in the Kāśikāvṛtti, in assuming that Pāṇini intended the latter. Both constructions are attested in ordinary Sanskrit usage, in any case.