Language change and the actuation problem: grammaticalization in Vedic Sanskrit

3.1 The Early Vedic future

The Early Vedic system differed slightly from that of later stages of the language. Most importantly, Early Vedic had a fully productive subjunctive with a prospective use (Tichy 2006). As the subjunctive could be formed from all aspect stems, aspectual differences could be expressed in rendering future events. Besides the inherited subjunctive, Early Vedic also had the more recent synthetic -sya-future mentioned above. It is based on an old desiderative and still quite rare in the Rigveda, where it is attested with no more than 15 verbal roots. In the Atharvaveda, it is already more frequent (>30 roots). A third way of expressing the future was the use of the so-called gerundive, a future passive participle, in nominal or copula sentences.

3.2 Early Vedic agent nouns

Vedic agent nouns are formed from the full grade root with an ablauting suffix -tar-/-tr̥- (Kim 2005; Kiparsky 2016; Lühr 2005; Tichy 1995). They come in two types, hysterotones with accent on the affix and akrotones with accent on the root.

The akrotone agent noun is rare from the onset. It shows verbal phrase syntax: with nouns derived from transitive verbs, the theme stands in the accusative. There is a general consensus as to its core temporal semantics. Akrotone agent nouns denote a “présent général, duratif” [general present, durative] (Renou 1938: 108), a “generelle Funktion” [general function] (Tichy 1995: 223), or, in the words of Kiparsky (2016: 173–174), a “habitual/generic meaning”, “restricted to actions performed vartamāne, ‘at the current time’.” From the Atharvaveda on, the type merges into the hysterotone, and the VP-syntax vanishes.

Hysterotone agent nouns always show noun phrase syntax. With nouns derived from transitive roots, the theme surfaces as a modifier in the genitive. Their semantics are slightly disputed. Renou (1938: 111) argues that they denote an “aspect ponctuel” [punctual aspect]. In a similar vein, Tichy (1995: 105) claims that they are “situationsgebunden” [situative], meaning that they do not denote a general property, but rather that a referent participates in a specific event. Kiparsky (2016: 174), on the other hand, states bluntly that “-tár- has no additional meanings, only the general meaning of agency.” Kiparsky’s assessment, however, is a bit simplistic. He is right that -tŕ̥-stems are not categorically situative. The affix can, for example, be used in occupational titles like jaritár- ‘singer of praise’ or stotár- ‘id.’^[4] Cf. the following example:

Here, a situative reading is excluded as it would lead to a tautology since netā́ and nayati would refer to the same event.

But in the vast majority of cases, nouns in -tŕ̥- do have a special semantics: they typically denote participation in an event which is situated relative to the event depicted by the matrix verb. In the terminology of Levin and Rappaport (1988), they are used as event-agent nouns.^[5] A clear example is (5), the last two half-verses from a spell against snakebites (Renou 1938: 111):

The spell is spoken on the occasion of a treatment against a specific snake bite and the last verse summarizes the successful therapy. Thus, it is exactly the serpent who participated in this specific biting event that the last halfverse addresses. This analysis is corroborated by the use of the aorist indicating an immediate past. The temporal semantics of situative daṃṣṭár- in (5) is given in the following DRS. Again, the formalization is necessary as the situative reading will be taken as the starting point of the grammaticalization of the -tā́-future in Section 4 below.

As in (3), t₁ is the topic time. e₁ is defined as the biting event by the first condition. The second condition again denotes neutral aspect. The last condition states that there is some temporal relation between the topic time and another event e which serves as the reference time. This event is introduced by Q, a variable over a property. In the case of (5), e is the ánu gā ‘returned’ event of the matrix sentence, which is merged with the daṃṣṭár ‘who bit’ DRS to form a complex DRS. This is indicated by ρ(t₁, τ(e)). In (5), the relation ρ between e₁ and the ánu gā event is one of anteriority. However, as the following examples show, simultaneity or posteriority are likewise possible, depending on the context.

The following two examples illustrate simultaneity, both with converbal (7) and adnominal (8) use of the agent noun:

A second example for an anterior reading is (9), while the posterior reading is illustrated by (10):

Note that in (10) táyor is a theme and thus evinces the event-reading.

But what happens if no reference time is given in the sentence? This is the case of predicative -tŕ̥-nouns in copula sentences. Cf. (11):

Copula sentences lack a reference point given in the sentence. This problem, however, is resolved in that situative -tŕ̥- triggers the presupposition that there is some event e available in the discourse which is the reference point to the -tŕ̥-event. Presupposition resolution depends on world knowledge and context. In (11), the reference point is the speech act introduced by the preceding sentence. The relation between the two events is one of anteriority. This anterior reading is possible despite the use of a present tense copula, since depending on the context the present tense may be interpreted as a future or as a past (Delbrück 1888: 278–279).^[6]

We are now able to give a general account of the temporal semantics of situative -tŕ̥-nouns:

In (12), P is a variable over all possible -tŕ̥-nouns. They introduce an event e₁, the topic time t₁ and again neutral aspect. On top of that, they trigger a presupposition, which is marked by ∂ as in (3) above. The presupposition introduces a reference time t₂ and the condition that there is some temporal relation between t₁ and t₂.

4.1 The source

There are three possible sources for the grammaticalization process turning copula sentences with agent nouns into a future tense. The starting point may be the akrotone agent noun, the hysterotone agent noun, or both. The akrotone is attractive since like the prospective new future it shows VP-syntax. On the downside, it has accent on the root, while the future is almost always accented on the affix – if the accent is transmitted at all. On rare cases of akrotone futures see Lowe (2017: 135). Another argument against an akrotone source is its semantics, since according to Pāṇini (and most recently Kiparsky 2016: 173–174) it is restricted to vartamāne, i.e. the present tense. But the most important reason to dismiss the akrotone as the source of the periphrastic future is much more straightforward: already in the times of the Atharvaveda, the akrotone, and with it its verbal syntax, had vanished. The emergence of the periphrastic future, on the other hand, is clearly a post-atharvavedic development. Thus, the only possible scenario starts with the hysterotone. VP-syntax, then, is not a case of persistence, but rather a consequence of the integration of the new future into the verbal paradigm.

4.2 Speaker–hearer interaction

Grammaticalization is often attributed to the inventiveness of speakers (see Eckardt 2012 for a critical assessment). However, as Eckardt (2012) and others point out, it is much more likely that grammaticalization processes are a mere side-effect of the interaction between speakers and hearers and thus unintentional. The scenario developed here, which has the additional advantage of aligning grammaticalization to phonological change (see Section 4.3), follows this line of thought.

Essential to reanalyses emerging from speaker-hearer interaction is the existence of so-called bridging contexts (Giomi 2017; Heine 2002: 86). A bridging context is a situation in which the hearer may attribute the utterance she parses a structure different from the one intended by the speaker. For such a reanalysis to work it is crucial that the “misparsing” does not impair the discourse: If it would not remain felicitous, the communication would break down or the hearer would be corrected.

In the concrete case of the periphrastic Sanskrit future, a possible bridging context must fulfill the following requirements:

1. As the prospective future is a copula construction, the same should hold true for its source. Since use in copula constructions is the most frequent anyway (Renou 1938: 112, 123), this is unproblematic.

2. It would probably facilitate the reanalysis if the agent noun had achievement or accomplishment semantics, because “la considération d’un terme peut donner l’illusion d’ un futur” [the consideration of an end point can give the illusion of a future] (Renou 1938: 108).

3. Finally, the bridging context should be able to account for the fact that the typologically isolated grammaticalization of an agent noun into a future marker occurred in the first place. A feature of the hysterotone agent noun which is unique in a manner similar to the grammaticalization path from agent noun to future tense is its situative primary reading. Therefore it is highly probable that it was this feature which triggered the change. Thus, it seems that a necessary prerequisite for the reanalysis is a situative reading or the possibility thereof.

Contexts fulfilling these requirements can actually be found in Early Vedic. An example for a possible bridging context in the Rigveda is (13):

As indicated by the alternative translations given, two different readings intended by the speaker are conceivable for (13). Under the first reading, vibhaktár- is a generic state-level predicate similar to an occupational title. Vibhaktā́si, then, is to be understood as a proper present tense. Under the second reading, the agent noun has situative semantics. This is indeed feasible, since the context of the sacrifice allows for an event-related interpretation: The donation is actually part of the ritual. This reading may further be promoted by the fact that the addressee, Agni, the holy fire, is permanent only on the abstract level of general types. The fire kindled during the ritual is a token of this type, but since the ability to donate is a direct consequence of the kindling, it seems likely that the actual instantiation of Agni is invoked as such. In this interpretation, the situative reading becomes unavoidable. The content of the main sentence invites a futurate reading since the worshipper receives gifts only after Agni comes to him. As mentioned above, this is unproblematic as the present tense allows for future and past readings.

The frame adverb sadyáḥ further strengthens the argument for a situative reading. In most attestations in the Rigveda, this adverb presupposes an event which precedes the event the adverb scopes over. Thus, Witzel (in Witzel and Gotō 2007), following Geldner (1951), translates it as ‘sogleich.’^[7] In the framework developed above, this means that t₂ equals n or follows n immediately, thus forcing an interpretation of t₁ as related to n. See Figure 1.

Figure 1:

Temporal semantics of vibhaktā́ ‘give away’ with situative reading in (13).

The timeline given in Figure 1 builds on the semantics developed in (12). n is the time of the speech act. The grey section marked τ(e₂) is the time the event denoted by dāśúṣe kṣarasi ‘you flow to the worshipper’ takes up. This event is an accomplishment. t₂ is the point in time denoted by sadyáḥ ‘instantly’. This, the reference time, is analyzed as the end point of e₂. This is the most likely interpretation. In principle, t₂ could also denote the starting point. But as this is the 6th verse and as the hymn is sung during the ritual, an inchoative reading is unlikely, especially since the part of e₂ leading up to the presence of Agni is not what the singer is interested in. As before, t₁ is the topic time. It lies somewhere within the extension of e₁, the event depicted by vibhaktā́si. This event, too, is an accomplishment.^[8] ρ(t₁, t₂) is interpreted as t₂ ≺ t₁, i.e. as a precedence relation, by presupposition resolution. Since dāśúṣe kṣarasi is immediate, it necessarily implies a futurate reading for vibhaktā́si.

In the scenario developed here, this situative reading is the one intended by the speaker. Due to the context in which the utterance is made, the hearer undoubtedly grasps both the situative semantics and its consequence, the futurate reading of vibhaktā́si. However, in her attempt to derive this interpretation from the speaker’s utterance, she is confronted with a challenge. The utterance itself is just a sequence of sounds and does not contain any cues regarding its structure and its semantics. To make sense of it, she has to infer a structure herself resorting to abductive reasoning. With Eckardt (2012) I assume that this reasoning is guided by the premise that in essence semantics is compositional.^[9]

Given these assumptions, the hearer is confronted with an ambiguity when computing (13): while the posteriority of e₁ is beyond doubt, it can be derived in two different ways. The hearer may infer a presupposed relation t₂ ≺ t₁ in line with the speaker’s intention and the structure outlined in (12). Alternatively, she may attribute the futurate reading of vibhaktā́si to a relation n ≺ t₁ directly encoded in the sentence. Both interpretations are indistinguishable in the interaction of speaker and hearer as they both keep the discourse felicitous. The second analysis, however, leads to change. While all else, including the neutral aspect, remains unchanged, the presupposition in (12) is turned into a novel conventional meaning. The result is a new semantic structure, the one introduced above in (3). Following the premise of compositionality, this new conventional meaning is assigned to -tā́ + copula, which thus turns into a new future marker. Note that the present tense of the copula fits this new interpretation as it contributes to the compositional semantics by accounting for the speech act n.

A similar case can be made for (14):

As the second translation indicates, a situative reading of the agent noun is again conceivable. The addressee of the hymn is Soma, which, being a plant pressed in the ritual, only becomes operative (e₁) – and comparable to gleaming beestings – after the purification process (e₂). In this case, t₂ is the endpoint of the purification event, while t₁ is a point in time within the extension of the support event. As in the case of (13), presupposition resolution forces a futurate reading upon dhartā́si ‘are the supporter > will support’ (t₂ ≺ t₁). And since e₂ is supposed to take place immediately after the speech act, the conditions are met for a hearer-induced reanalysis of the temporal relation as n ≺ t₁.

But why would the reanalysis happen in the first place? A hearer, especially a learner, is likely to opt for the least costly analysis feasible in a given context. This implies that she tries to reduce the number of pragmatic computations to the unavoidable minimum (Schwarz et al. 2014; Schwenter and Waltereit 2010). This urge is even more pressing in a case like (13) and (14), as the kataphoric relation between the two events adds even more computational load: As shown by Schwarz (2007) and Tiemann et al. (2011), the lack of an antecedent significantly delays presupposition resolution.

It is important to stress that under the reanalysis stipulated here the discourse remains felicitous. As a consequence, the change goes unnoticed and there is no need for the hearer to revise her interpretation. However, turned speaker, the listener will then use the new item with its compositional semantics freely outside of bridging contexts. This will eventually lead to the integration of the new future as an inflectional form into the verbal paradigm. Full grammaticalization then manifests itself in lack of agreement and VP-syntax (but see Lowe 2017: 135, who ponders the idea that semantic reanalysis and syntactic transitivization need not go hand in hand).

This scenario raises an immediate question: Why are predicative agent nouns with situative readings not reanalyzed as a past tense? The ingredients to the reanalysis would be essentially the same. Likewise, possible bridging contexts are conceivable and would probably be found. As shown above, anterior relations between the matrix event and the agent noun are indeed attested, and the general semantics of situative agent nouns as outlined in (12) leaves the nature of the relation unspecified. One might want to resort to frequency as an answer to this rather vexed problem. However, as we know nothing about frequencies in utterances in Vedic discourse, such an approach would be futile.

But there is a difference between the past and the future in Early Vedic which may have played a crucial role. Grammaticalizations like the one discussed here are based on reanalysis. Reanalysis, however, seems to be guided by the existence of comparable patterns. Thus, postpositions are only reanalyzed into case endings in languages (like e.g. Lithuanian) which already have case endings. Similarly, the reanalysis of English going to or German nicht brauchen as auxiliaries is facilitated by the existence of other auxiliaries in the same realm (future tense and modality respectively). In Vedic, the situation is similar for the --tā́future as periphrasis is well attested in futurate contexts from early on: the gerundive/future passive participle in -ya-, -tavyà- occurs regularly in nominal sentences since the Rigveda (Delbrück 1888: 397). We may thus hypothesize that the existence of the analytic pattern in the passive facilitated its spread to non-passive future contexts. In the past, on the other hand, the situation was strikingly different: past tense periphrasis with -tá-, which might have served as a foil for the development of a potential --tā́past, did not yet exist.^[10]

4.3 A parallel in phonology

Above I already made the argument that grammaticalization has no ontological status as a process distinct from other types of language change. In this section, I seek to strengthen this point by giving an example from phonological change which relies on the same mechanisms as the development of the analytic future in Sanskrit sketched in this paper.

The scenario developed here for the -tā́-future rests on the assumption that a presupposition turns into part of the compositionally derived meaning. More generally, a systematic property of a linguistic structure which can be cancelled without compromising the interpretability of the structure itself turns into one of its core features. Phonemicization of an already phonologized but peripheral feature (Hyman 1976) is a parallel from phonology. In such cases, non-distinctive phonological features turn distinctive, resulting in a reduction of computational load. A case in point is the feature [+noise] as e.g. in the Germanic case of /p^h/ or /pᶲ/ > /f/ (see Boersma 2003 for a possible scenario). In this and comparable cases, the signal is accurately perceived by the hearer, but it is ambiguous: the perceived noise may be an accessory feature, but it may also be distinctive. If the hearer opts for the latter, change takes place (this is a case of “chance” in the terminology of Blevins 2004).

Similarly, a hearer of (13) accurately perceives the utterance. But she, too, has to resolve an ambiguity: either the posterior reading is a presupposition, or it is part of the core semantics. Both in phonology and semantics, an abduction has to be made, which is guided by an analytic or cognitive bias towards a reduction of computational load (see Moreton 2008 for phonology and Schwarz and Tiemann 2017 for semantics). Note that this type of grammaticalization parallels phonemicization in that the presupposition is part of the message. Other instances of grammaticalization compare to phonologization based on coarticulation effects (Ohala 1981, 2012). For an example see Keydana (2017).