The Vedic Root Variants of the Type CaC // C(C)ā:
Formal Patterns and Transitivity Types

This paper has a long (pre)history. It has come out of my Moscow Candidate degree dissertation (1989), and the very first draft of the paper where I summarized my ideas on the syntactic features of the verbal pairs of the type CaR (CaC) // CRā (CCā) was written in the same year. Two years later, I presented a short summary of my results at the conference "Slavic, Indo-European and Nostratic studies", held on the occasion of V.A. Dybo's 60th birthday (see Kulikov 1991; a somewhat extended English version of the brief summary Kulikov 1991 was published as Kulikov 2011). Yet several aspects of the phenomena in question remained unclear to me, which was one of the reasons to repeatedly revisit this issue. In November 1992 I gave a draft version of my paper to Prof. Raimo Anttila when meeting him during Ann Arbor Workshop on Reconstruction at University of Michigan, and in December 1992 I had the pleasure to discuss some ideas formulated in my paper with Raimo during my short stay in Los Angeles. Both reading Raimo's monograph (Anttila 1969), which remained the most comprehensive treatment of the Indo-European Schwebeablaut till the appearance of Kaspars Ozoliņš' PhD dissertation (2015), and our 1992's discussion have strongly encouraged me to continue my research. Needless to say, Raimo's memorial volume seems to me the best occasion to update my paper (the complete version of which was never published) and to present here the results of my study.

Besides Raimo Anttila, I am thankful to Tatjana Elizarenkova, Frits Kortlandt, Alexander Lubotsky, Werner Knobl, Sergej Starostin and Ilya Yakubovich for their comments on earlier drafts of this paper. I also would like to express my thanks to the audience of the conference "Slavic, Indo-European and Nostratic studies" (Moscow, May 1991) and of the XIII. Fachtagung der Indogermanischen Gesellschaft "Historischer Sprachvergleich und linguistische Theorie: Indogermanistik und allgemeine Sprachwissenschaft im Dialog" (Salzburg, September 2008), where some parts of the present paper were discussed, for remarks and criticism. I acknowledge the support of the Marie Skłodowska-Curie grant no 665778, the NCN POLONEZ grant no. 2015/19/P/HS2/02028 for a research stay at Adam Mickiewicz University in Poznań, and the FWO research grant no. G004121N, which allowed me to finalize this paper.

1. The C//ā-Alternation: A Short Survey

1.1. ā-roots: synchronic patterns and diachronic sources

The Vedic verbal lexicon contains some twenty root pairs of the type ay (i) 'go' // yā 'drive', gam 'go' // gā 'tread', tari (tr̥̄) 'pass' // trā 'protect, rescue', dhami // dhmā 'blow', pari (pr̥̄) // prā 'fill', bhas 'devour' // psā 'chew', man 'think' // mnā 'mention', etc. In all such pairs, the second member ends in ā and can be derived, in formal terms, by adding ā to a certain modification (most often, the zero grade) of the first member (i-ā, psā [= bhs-ā], mn-ā, etc.). Schematically, the formal relationship between the members of such pairs can be represented as CaC // C(C)ā, where the final consonant is, most often, a sonant (i = ay, tr̥̄ = tari, etc.), thus: CaRi (/CR̥̄) // CRā. Accordingly, I will hereafter refer to the second members of such pairs as ā-roots (ā-verbs), while the first members, the 'base roots', will be called, for lack of a better term, C-roots (C-verbs). The alternation of this type will be referred to as 'C/ā-alternation'.

The formal relations between the members of such pairs are quite diverse, and so are their origins. Some of them can be treated in terms of the pattern CaC // CCā, which suggests that the second member of the pair is derived by means of a laryngeal root extension (*-eH- > -a-), cf. i – yā, man – mnā. Some others follow the pattern CR̥ (CaRi) // CRā (where R stands for a sonant), and thus, at the level of the Indo-European reconstruction, instantiate Schwebeablaut CeRH- // CReH-, a phenomenon discussed in detail by Raimo Anttila (1969). The members of the schwebeablauting pairs, CaRi and CRā, are often called, according to the Indo-Europeanist tradition, 'full grade I' (Vollstufe I) and 'full grade II' (Vollstufe II), respectively (see, for instance, Gotō 1987: 45f.). Finally, a few pairs exemplify the type CaC (CaR) // Cā, as in the case of gam 'go' // gā 'tread' and dru (drav) // drā 'run'; some of these pairs may be formed by etymologically unrelated roots as a result of their semantic and phonological convergence.

No less variety is found in the treatment of the relationship between the members of such pairs in the traditional Sanskrit (Vedic) scholarship. Some of them are taken as root variants distributed between the formations of one single paradigm (individual verbal system, in terms of Jamison 1983), as in the case of dhami // dhmā 'blow' (see below, Section 3, sub voce). The emergence of two different full grades is mostly explained by several secondary developments and paradigmatic reanalyses. In some other cases, the members of such pairs are more commonly treated as distinct lexical units (resp. roots), which, nevertheless, historically belong together as etymologically related roots (see below on pari (pr̥̄) // prā 'fill'). Finally, the members of some pairs are never treated as representing one synchronic unit, while their historical (etymological) connections vary from clear and undoubted (cf. bhas 'devour' // psā 'chew') to questionable or implausible (cf. below on gam 'go' // gā 'tread' and kan // kā 'be pleased, enjoy').

Altogether, the synchronic status of the C//ā-alternation within the Vedic verbal system is unclear. On the one hand, it is beyond any doubt that there must be some sort of relationship between such roots as i and yā or bhas and psā. No Sanskritist or Indo-Europeanist will deny that the members of such pairs are related rather than result from accidental formal coincidence (convergence). There are good reasons to believe that, synchronically, the members of such pairs as pari (pr̥̄) // prā, dham // dhmā or pay(i) (pi [pī?]) // pyā were considered as belonging together, too, even in spite of the somewhat unclear and non-productive character of relation between them, thus resembling the English type foot : feet, tooth : teeth, rather than near : next (cf. Old English nēah : nēarra : nēahsta) or old : elder.

On the other hand, no Sanskrit (Vedic) grammar deals with the pairs of the type tari (tr̥̄) // trā in the chapter on verbal derivation, treating the second members (of the type trā etc.) as separate lexical units.

There are several reasons for this state of affairs. On the one hand, there are cases such as dhami // dhmā 'blow', where the genetic and synchronic relationship between the members of a pair is beyond any doubt, and the two members appear to be (nearly) exact synonyms, so that it is unclear if any functional value whatsoever might be ascribed to the C//ā-alternation. On the other hand, in some other pairs, such as tari (tr̥̄) 'pass, overcome, carry across' // trā 'protect, rescue' or mari (mr̥̄) 'crush' // mlā 'wither, relax', the formal and/or semantic relation between the members is blurred, which prevents us from considering them as forming a synchronic unity; such pairs rather belong to the type near : next. Thus, ironically enough, the members of the pairs like dhami // dhmā or pari (pr̥̄) // prā are (semantically) too similar to be treated as representing a morphological derivation, whereas the members of the pairs like tari (tr̥̄) // trā or mari (mr̥̄) // mlā are too different to be regarded as synchronically (derivationally) related.

Last but not least, there is yet another reason which has contributed to the 'bad reputation' of the C//ā-alternation. It seems that pairs of the type pari (pr̥̄) // prā and tari (tr̥̄) // trā, however transparent their synchronic relations might appear, bear a heavy burden of diachronic "sins". Specifically, most of these pairs are associated with two quite ill-famed phenomena of the proto-language. One is the Schwebeablaut, the alternation of the type CeRC- // CReC-, observed in such examples as *perḱ- (cf. OHG. fergôn 'to ask') ~ *preḱ- (Got. fraihnan id.); see Anttila 1969: 150–151 and Ozoliņš 2015: 29. Here must belong, in particular, such pairs as tr̥̄ // tari (cf. class I pres. tárati < *terH-e-ti) // trā (cf. class IV pres. trā́yate < *treH-). Anttila's 1969 monograph removes the Schwebeablaut from the proto-language, explaining such pairs as due to several secondary developments. Yet, much remains unclear about this morphological phenomenon, and, in any case, we are hardly able to ascribe any functional value to this alternation.

Another – and even more vexed – problem directly related to our C//ā-pairs is the highly controversial issue of the Proto-Indo-European laryngeal root extension and/or suffix *-ē-. As is well known, the final consonants in such Indo-European roots as *trep- (OCS trepetъ 'trembling'), *tres- (Skt. trasati 'trembles') and *trem- (Gr. τρέμω 'tremble'), or *dreu- (Skt. drávati 'runs'), *dreH- (cf. Skt. impv. drā́tu) and *drem- (Gr. δραμεῖν 'run'), might be treated, at least from the formal point of view, as suffixes. However, given that the meaning or function of these final elements is unclear, Indo-European scholarship sticks to more neutral terms, such as 'root enlargements', 'root extensions', 'Wurzeldeterminativa' or 'Wurzelerweiterungen'. Attempts to determine the function of these elements have largely failed. Since the monograph Persson (1912), which remains the most fundamental treatment of the issue till now, no special study has been devoted to this phenomenon.

Thus, pairs of the type ay (i) // yā or dah // kṣā can only be taken as related if the second members are treated as comprising the morphological element (suffix? submorph? root extension?) -ā- (< PIE *-ē- or *-eH-). Such a suffix, presumably with an intransitivizing function and/or stative meaning, is posited in many Indo-European handbooks, but Vedic roots such as yā or kṣā are (usually) treated separately from the -ē-verbs in Indo-European scholarship of the last century. Accordingly, we are forced to posit a laryngeal extension (*-(e)H-) at the end of these roots. This analysis is adopted, in particular, by Anttila (1969: 59-63) and Mayrhofer (EWAia, sub voce) for kṣā (EWAia I, 430), psā (EWAia II, 198) and yā (EWAia II, 407).

Obviously, the analysis of such roots as kṣā, psā and yā as containing the reflexes of the laryngeal root extensions (= suffix *-eH-?) leaves open the question on their function of this element and, more generally, on the semantic difference between them and the corresponding non-extended roots, i.e. dah, bhas and ay (i).

Likewise, the functional value of the Old Indo-Aryan C//ā-alternation altogether remains an enigma, and the very phenomenon is largely disregarded by Sanskritists. Yet, there is a feature that makes a research of the C//ā-alternation in Vedic a particularly interesting and important task. This alternation appears to be better represented in Indo-Iranian (and especially in Indo-Aryan) than in most other Indo-European branches. This may point to the fact that the origin and expansion of this phenomenon must represent an Indo-Iranian (or even an Indo-Aryan) innovation.

Accordingly, in what follows, I will make no attempt to investigate or reconstruct at full scale the Proto-Indo-European origins of the C//ā-alternation. Rather I will concentrate on the systematic treatment of the features of the members of the Vedic C//ā-pairs, above all in a synchronic perspective.

1.2. Remarks on the morphophonological features of ā-roots

Before I proceed to the analysis of the C- and ā-verbs, a few remarks on their morphophonological features are in order.

Obviously, for our purposes we need to identify any relevant formation as belonging to the system of the C- or ā-root. Generally, this task poses no problem, cf. infinitives étave and tar(i)tum (built on the C-roots ay and tari), as opposed to yā́tave and trātum (ā-roots yā and trā). Problems are only posed by the zero grade forms of the schwebeablauting roots, such as tari (tr̥̄) // trā or pari (pr̥̄) // prā. From the formal point of view, such formations as verbal adjectives tīrṇá- and pūrṇá- might belong to either of the two variants, i.e. either to the C-root CaRi / CR̥̄ (< *CR̥H- / *CeRH-), or to the ā-root CaRi (CR̥̄) / CRā (< *CaRH- (*CR̥H-) / *CReH-), that is, in our case, either to pari/pr̥̄, tari/tr̥̄ or to prā, trā. This problem does not actually arise in the case of pairs such as tari/tr̥̄ 'pass, carry across' // trā 'protect, rescue', where the C- and ā-roots clearly differ in meaning, cf. tīrṇá- 'passed, crossed' (≠ 'protected, rescued'). The zero grade formations made from members of C//ā-pairs with (nearly) no semantic difference may pose difficulties, however: for the correct identification/attribution of such forms we need additional criteria. There are some indications that all zero grade forms should be grouped with the C-roots. Specifically, many ā-roots such as trā and prā tend to generalize full grade (i.e. ā), using it also in those formations where we expect zero grade. Thus, we find ā-grade in verbal adjectives (past perfect participles) in -ta/-na, cf. trāta-, dhmātá-, prātá-, as opposed to the adjectives sthitá- and dhītá-, made from the 'independent' (= non-schwebeablauting) roots sthā 'stand' and dhā 'suck'. In other words, the ā-roots such as trā, dhmā and prā belong to the 'non-alternating' morphophonological type in terms of Zaliznjak (1975: 68ff.). This morphophonological peculiarity of the ā-roots has not of course remained unnoticed. Sanskrit grammars and dictionaries usually do not connect formations such as pūrṇá-, tīrṇá-, pres. pr̥ṇā́ti, tiráti etc. with ā-roots. I will basically follow this tradition, including zero grade formations into the verbal systems of the corresponding C-roots, unless there are clear semantic indications for the opposite analysis (as in the case of yā 'drive' – pres. ī́yate).

2. Syntactic Features of the C//ā-Roots: A Preliminary Hypothesis

To begin with, let us have a closer look at the features of two C//ā-pairs.

(i) pari (pr̥̄) // prā 'fill'

The roots pari (pr̥̄) and prā 'fill' are synonymous and occur in similar constructions, as, for instance, in (1a, b):

There is, however, a remarkable difference between their properties that seems to have escaped scholarly attention. The verbal system of pari (pr̥̄) contains both intransitive and transitive formations; both groups are well-attested from early Vedic (= the language of the R̥gveda and Atharvaveda) onwards, cf. intransitive constructions as in (2–3) and transitive-causative usages as in (1a):

By contrast, prā is mostly employed in transitive constructions, as in (1b) and (4):

The intransitive class IV present pū́ryate is likely to belong to the C-root pari (pr̥̄), and there is no present passive **prāyáte. The only attestation of an intransitive (passive?) form built on this root, the medio-passive i-aorist -aprāyi (with the preverb ā́), appears at the end of the early Vedic period, in a late stanza (5), for which see Kümmel 1996: 72f.; Griffiths 2009: 213f.; Kulikov 2012 [2014]: 124–125:

(ii) ay (i) 'go; send, set in motion' // yā 'drive, speed'

As in the case of pari (pr̥̄) // prā, the ā-root yā neatly differs from its C-counterpart ay (i) in syntactic features. For the root ay (i), both intransitive and transitive formations are well-attested from the early Vedic period onwards. Intransitive derivatives, meaning 'go', are represented, in particular, by the class II present (= athematic root present) éti, as in (6). The transitive-causative counterpart of éti is the class V present inóti and its thematicization ínvati, meaning 'send, impel, set in motion', as in (7):

By contrast, the ā-root yā is basically intransitive (cf. pres. yā́ti, ī́yate 'drives, speeds', etc.). The -aya-causative yāpayati first appears in the Brāhmaṇas.

In spite of the difference between the syntactic types of the ā-roots in the two above-discussed cases, there is a remarkable syntactic feature shared by the pairs pari (pr̥̄) // prā and ay (i) // yā. While the C-verbs are well-attested in both intransitive and transitive (transitive-causative) usages, usually from early Vedic onwards, their ā-counterparts show a noteworthy limitation of their 'syntactic flexibility', restricting their usages either to intransitive or to transitive only. The former, more flexible, type of syntactic behaviour, exemplified by such C-verbs as pari (pr̥̄) and ay (i), will hereafter be called 'diffuse', for the lack of a better term. The most typical representatives of the diffuse type are those verbs some forms of which can be employed both intransitively and transitively, thus showing the labile syntax. Thus, for instance, 3pl.pf.act. vāvr̥dhúḥ of the verb vardh (vr̥dh) 'grow, increase' occurs 6 times in intransitive usages (as in (8a)) and 14 times in transitive-causative usages (as in (8b)) in the R̥gveda (see Kümmel 2000: 469-473 for details):

Apparently, both ā-verbs under discussion, yā and prā, belong to the non-diffuse syntactic type: their forms can only be employed intransitively or transitively, while the opposite type of usage (transitive or intransitive, respectively) is either exceptional and/or only attested in late texts, or does not occur entirely.

Thus, the clue to the functional value of the C//ā-alternation is likely to be found in the domain of syntactic features and transitivity of the verbs in question.

In what follows, I will scrutinize the C//ā-pairs for their syntax, checking my assumption against the evidence available from Vedic.

3. The Vedic C//ā-Verbs and Their Syntax

In what follows, I will briefly discuss almost twenty root pairs exemplifying the C//ā-alternation.

av (u) // vā 'weave'

The ā-root vā is of clearly secondary nature, being created alongside the aniṭ-root av (u) (see Hoffmann 1974: 23 [= Hoffmann, Aufs. 335], fn. 17; Mayrhofer EWAia I, 275f.; II, 538). The derivatives of both root variants are attested in the same type of transitive constructions. There are only two forms that can be ascribed to vā: 1) the infinitive vā́tave at AVŚ 10.7.44 (corresponding to the earlier form ótave in the parallel passage RV 10.130.2) and 2) the future participle (RV 7.33.12), cf.

The infinitive made from the main root variant av (u) attested in the parallel R̥gvedic verse (10.130.2cd) clearly points to the secondary character of the Atharvic ā-infinitive:

kani // kā 'be pleased, enjoy'

The verb kani, attested in the perfect (1sg.act. cākana, 2-3sg.inj.act. cākán, etc.) as well as in a few sigmatic aorists, is construed either with the accusative (as in (11)), or, more often, with oblique cases (locative or genitive) (as in (12)), thus being 'intransitive/transitive' in terms of Jamison (1983: 31-39); see Kümmel 2000: 130-133 for details.

The only attested perfect middle form, 3pl.subj. cākánanta, appears in intransitive constructions, meaning 'be pleasant', as in (13):

By contrast, the forms of the verb kā (middle perfect cake and the R̥gvedic hapax pres.part. kā́yamāna-) are employed in transitive constructions, meaning 'yearn, enjoy', as in (14–15):

In formal terms, the class IV pres. stem kā́ya- is ambiguous. It may be based on the root kā < *keh₂, thus, belonging with the middle perfect cake (thus Joachim 1978: 67f.; Mayrhofer, EWAia I, 334; Kümmel 2000: 142f.; LIV 343). Alternatively, kā́ya- might be connected with the seṭ root kani 'rejoice', as its class IV present (cf. jani - jā́yate). However, as Narten (1964: 94f.) points out, the individual verbal systems of these two roots, as well as their syntactic and semantics can be neatly distinguished. In contrast to kani, which only builds active forms (perfect cākán- and sigmatic aorist akāniṣ-) and is employed intransitively, kā forms the middle perfect (ā́) cake, cakāná-, which is mostly construed with an accusative. This favours the analysis of the hapax kā́yamāna- as a middle participle of the -ya-present built on the root kā, connected with the accusative vanā́. Accordingly, the construction in the pāda should be interpreted transitively ('longing for wood').

These two roots are usually considered as genetically unrelated in the literature (see Mayrhofer, EWAia I, 296f., 334 and LIV 343, 352 on the roots kani 'Gefallen an etwas finden, sich freuen' and kā 'begehren, gern haben'). However, in view of the semantic affinity of these two meanings, possible (secondary) connections between them should not be ruled out.

kāś 'become visible, appear (?); see' // kśā (khyā) 'see, consider, reckon'

The root kāś probably goes back to PIE *ku̯eḱ- (cf. Gr. τέκμωρ, τέκμαρ 'sign, feature'); the vowel length is likely to be secondary (see Gotō 1987: 115, with fn. 102; Mayrhofer, EWAia I, 344f.; LIV 383ff.). In early Vedic, this verb is only attested in the active intensive -cākaśīti etc. (RV+) 'consider, see, look at' (see Schaefer 1994: 102-104; Roesler 1997: 199-204) and causative sáṃ kāśayāmi (AV 14.2.12), which probably means 'make visible' (see Jamison 1983: 125; for a detailed discussion of the relevant passage, see also Schaefer 1994: 103f., fn. 264). The intransitive class I present -kāśa-te 'become visible, appear' occurs from the Brāhmaṇas onwards (ŚB, Jaiminīya-Upaniṣad-Brāhmaṇa). In spite of the rather late attestation of the intransitive usages, there are some indirect reasons for determining the original meaning of the root as 'become visible, appear; consider, see, look at'; as Jamison (1983: 125) suggests, kāś may belong to the same syntactic type as darś (dr̥ś), cf. med. dadr̥śé 'appears' ~ act. darśáyati 'reveals'. Correspondingly, intransitive usages can be tentatively reconstructed for early Vedic (cf. also the historically related root cakṣ 'look at, appear', on which see Roesler 1997: 205-209; LIV 383-385).

The ā-verb kśā (khyā) 'consider, count' is fundamentally transitive. The present -yá-passives appear from the Brāhmaṇas onwards, but in early Vedic (RV) we find the middle thematic aorist (3sg. -akhyata) attested in a passive construction (for a detailed analysis of the passage, see Kulikov 2008: 251-252):

The Iranian evidence (Cheung 2007: 245-246) supports the secondary character of the ā-root as well as the antiquity of the intransitive usage.

gam 'go' // gā 'go, tread'

In spite of the semantic and phonological similarity of these roots, their historical relationship is far from clear. In the early scholarship, gam and gā are often connected (cf., e.g., Benfey 1837: Sp. 927 [= Kl. Schr. I/2, 29]; Reichelt 1904: 40; Persson 1912: 572ff.), and this view is adopted by Mayrhofer (EWAia I, 466: "Mit GAM vermutlich wurzelverwandt … ist GĀ1"; see also EWAia I, 482). *gu̯e- could not be a possible root structure in Proto-Indo-European, and thus PIE *gu̯em- (> Ved. gam) and *gu̯eh₂- (> Ved. gā) cannot be directly related in terms of root extensions. Nevertheless, the semantic affinity between the members of this pair could be supported by the model of the semantically similar pair dram // drā 'run' (see below, s.v.), which may be associated with gam // gā as 'rime-words' ('Reimbildungen').

Whatever the historical relations of these two roots in Proto-Indo-European, their syntactic behaviour is amazingly similar to that of most other C//ā-pairs. The C-verb gam is fundamentally intransitive, but its present causative gā̆máyati is well-attested from early Vedic onwards (3x in the RV); the causative aorist ajīgamat first appears in the Atharvaveda (see Jamison 1983: 172). By contrast, causative of the intransitive gā (*gāpáyati) is lacking.

jani 'be born; beget, generate' // jñā 'know'

The attempts to connect these two roots semantically and historically proved unsuccessful (see Anttila 1969: 130). Yet, this pair is worthy of mention in our discussion, foremost because of the fact that the syntactic behaviour of its members perfectly fits the pattern of the type pari (pr̥̄) // prā. The verb jani, well attested both in intransitive (pres. jā́ya-te, pf. jajñé, medio-pass. aorist ájani, sigmatic aorist ájaniṣṭa) and transitive-causative (pres. jána-ti, janáya-ti, pf. jajā́na, etc.) usages can serve as a parade example of the diffuse syntactic type. By contrast, jñā is fundamentally transitive; passive usages are only attested for the present passive jñāyá-te 'be known' (RV 4.51.6 +; see Kulikov 2012: 92-95).

tan // tā 'stretch, extend'

The syntactic type of tan 'stretch' can be determined as diffuse. There is a remarkable correlation between tenses and syntactic patterns attested for this verb (for details, see Kulikov 1999: 26ff.). On the one hand, forms of the present system most often occur in transitive-causative usages, as in (17–18):

On the other hand, perfect forms are more common in intransitive constructions, as in (19), although transitive-causative usages are possible, too, cf. (20):

The root variant tā is never treated as a separate root. The origin of this secondary root is unclear. It only appears in the R̥gvedic hapax 3sg.pf.med. tate (RV 1.83.5) 'has extended' (transitive, cf. (21)) and present passive tāyáte (RV+) 'is stretched, extended'.

In late Vedic, we also find the medio-passive -i-aorist prātāyi (hapax, attested in the AĀ). It must be conditioned by the adjacent -ya-present (probably passive) prātāyata. Both forms are employed for an "etymological explanation" of prātar 'early in the morning':

Thus, with the exception of one isolated perfect form (transitive), the secondary root tā only appears in two intransitive (passive) formations.

tari (tr̥̄) 'pass, carry across' // trā 'protect, rescue'

The C-verb tari (tr̥̄) is well-attested both in intransitive constructions (e.g. class I pres. tára-ti 'pass', cf. (23)), and in transitive-causative constructions (e.g. class VI pres. tirá-ti 'make pass', with the preverb prá typically meaning 'make someone's life(time) safely pass over [obstacles and dangers] and reach its natural end', cf. (24)), thus, being a typical example of a diffuse verb:

By contrast, the ā-root trā 'protect, rescue', probably based on the transitive-causative usage of the C-root tari (tr̥̄) ('protect, rescue' ≈ 'carry across' = 'make pass'), is only attested in transitive constructions, for instance, in the class IV present formation trā́yate 'protects, rescues', cf.:

dah // kṣā 'burn'

The verb dah 'burn' is fundamentally transitive. However, the intransitive present dahya-te, attested both with root (non-passive) and suffix (passive) accentuation, becomes quite common at the end of the early Vedic period, from the Atharvaveda onwards. Importantly, for many of its occurrences both passive ('X is burned') and non-passive intransitive, or anticausative ('X burns, is on fire') interpretations are possible, as in (26–27) (for details, see Kulikov 2012: 390-396):

Thus, by the end of the early Vedic period, dah behaves as a diffuse, rather than as a predominantly transitive, verb. See also rich evidence for the intransitive (diffuse or even labile) syntax of the Iranian cognates (Proto-Ir. *daǰ- 'burn': Av. daž- id. etc.) in Cheung 2007: 53-54.

By contrast, the ā-root kṣā (< *dhgu̯h-eh₁-; see Mayrhofer EWAia I, 430; LIV 133f.), attested, for instance, in the class IV present kṣā́ya-ti (AVP+; see Kulikov 2012: 532-533), is fundamentally intransitive. Causative formations of kṣā appear from the late R̥gveda onwards (injunctive of the causative aorist cikṣipas RV 10.16.1; pres.caus. kṣāpáya-ti AV+; see Jamison 1983: 140).

drav (dru) // drā 'run'

The synonymous roots drav (dru) and drā are clearly related and, as some scholars suggested, could even form one suppletive paradigm, with the class I present of dru (dráva-ti), on the one hand, and the root and sigmatic aorist of drā (3sg.subj.act. drāsat etc.), on the other hand; see Gotō 1987: 178; Kümmel 2000: 254; LIV 129, but see serious criticism against this assumption in Casaretto 2002: 45–49. The syntactic relationships between the formations derived from the roots of this pair essentially reproduce the pattern of gam // gā. The C-root dru is mostly employed intransitively, but its causative drāváyati is twice attested in the RV (on its antiquity, see Jamison 1983: 114). The causative of drā, drāpayati, first appears in middle Vedic, in the Śatapatha-Brāhmaṇa (9.1.1.24).

dhami // dhmā 'blow, inflate'

The members of this pair are usually taken as root variants, not as separate roots. The majority of formations built on the C-variant (foremost, the class I present dháma-ti, for which see Gotō 1987: 180f.) are employed transitively, except for the present passive dhamyate, a Vedic hapax, attested in the late RV:

The ā-variant is scarcely attested in early Vedic. Its syntactic type can tentatively be determined as (predominantly) transitive. The only attested finite form is the sigmatic aorist -adhmāsam found in the Paippalāda recension of the Atharvaveda (see LIV 153):

Apart from this form, we only find non-finite derivatives: 1) verbal adjective dhmātá- 'blown' (RV 7.89.2); 2) agent noun dhmā́tar- 'the one who blows; wind player'; and 3) action noun dhmātár- 'fan; blower' (both in stanza RV 5.9.5). While the passive built on the C-variant, dhamyate, is a R̥gvedic hapax, present passive dhmāyá-te first appears in late Vedic (JB, Up.; see Kulikov 2012: 125-130).

dhayi (dhī) // dhyā 'consider, think (about), reflect'

The verb dhayi (dhī) mainly appears in the perfect dīdhaya (also pluperfect ádīdhet and reduplicated present created on the basis of the perfect subjunctive), well-attested in early Vedic and employed in transitive usages; for a detailed discussion of the attested formations and their semantics, see Kümmel 2000: 257-261. The ā-root dhyā appears, in particular, in the class IV present dhyā́ya-ti 'think of, meditate, contemplate' (construed with the accusative), which first occurs in late early Vedic (AVP 9.21.1-12), but becomes common only in middle Vedic (YVp+); for its attestations and genesis, see Kulikov 2012: 565-568.

pay(i) (pī̆) // pyā 'swell'

The C-verb pay(i) (pī̆) [pay (pi)?] is well-attested both in intransitive and transitive-causative usages; see, in particular, Thieme 1929: 40–41, 43, 49. The syntax of the thematic nasal present pínva-ti/te depends on the diathesis: middle forms typically occur in intransitive constructions ('swell'), as in (30), while active forms are transitive-causative ('make swell'), as in (31):

Cf. also the labile (albeit predominantly intransitive) syntax of the active perfect (see Kümmel 2000: 298-304):

By contrast, the ā-verb pyā 'swell' (-ya-present -pyā́ya-te etc.) only occurs in intransitive constructions in the RV, as in (34); the -áya-causative pyāyáyati first appears in the Atharvaveda (see Jamison 1983: 149):

bhan 'speak' // bhā 'shine'

The roots bhan 'speak' and bhā 'shine' (cf. also bhāṣ 'speak' and bhās 'shine') are usually taken as etymologically related in the Indo-European scholarship, in spite of a considerable semantic distance between their meanings. The syntactic features of the ā-verb bhā resemble much those of gā, the ā-member of the formally similar pair gam // gā (see above). The only early Vedic (RV+) formation of bhā, root present bhā́ti, is employed intransitively, as in (35); causatives of this root are lacking in Sanskrit.

By contrast, the verb bhan, in spite of its rather scant attestation in Vedic (which amounts to four occurrences of the class I present bhána-ti/te in the R̥gveda), exhibits a much greater variety of syntactic patterns. The active forms (3sg.act. bhánati at RV 6.11.13 and 3pl.act. bhananti at RV 4.18.6) are employed transitively, as in (36); the middle form bhananta appears in the reflexive (RV 7.18.7; cf. (37)) and reciprocal (RV 4.18.7) usages; see Gotō 1987: 222f., with fn. 472-473.

bhas 'devour' // psā 'chew'

Both the C-verb bhas 'devour' (RV+) and the etymologically related ā-verb psā (< *bhs-ā-; see Mayrhofer EWAia II, 198, 257 and LIV 82, 98) 'chew' (AV+) are fundamentally transitive; passives are unattested.

man 'think, believe; respect' // mnā 'mention'

The verb man is attested in transitive usages of two types:

(i) The class IV present mánya-te and the sigmatic aorist (ámaṃsta, maṃsi etc.) mostly occur connected with the direct speech construction, meaning 'X [nom.] thinks (that) P' or with two accusatives ('X [NOM] considers/believes Y [ACC] to be Z [ACC]'), as in (38):

(ii) The present manuté (class VIII in the traditional classification, originally, a -nu-present: *mn̥-nu-tai) is typically construed with the accusative or genitive of the deity or his/her aspects, meaning 'respect, remember with respect'. The root aorist (ámata, ámanmahi etc.) is most commonly employed in this latter usage ('respect' etc.), although type (i) usages are possible as well.

The -ya-present mánya-te is also common in intransitive (reflexive) usages, meaning 'X [NOM] considers/believes him-/herself to be Z [NOM]', as in (39):

The secondary ā-verb mnā, traditionally regarded as an extension of man (see, e.g., Mayrhofer, EWAia II, 385; LIV 447), is attested from the middle/late Vedic period onwards (ā-mnāta- Br., ā-mnāyá- Ār.+, etc.; see Gotō 1987: 239; 1997: 1025). This verb is fundamentally transitive; its present passive first appears in the (post-Vedic) Bhāradvāja-Śrautasūtra (3pl. ā-mnāyante).

mari (mr̥̄) 'crush' // mlā 'wither, wilt'

The verb mari (mr̥̄) 'crush' is fundamentally transitive (the rare passive present -mūryá-te occurs only in ŚB 1.7.3.21 ≈ 1.7.4.12). The historically related ā-root mlā 'wither, wilt', attested, in particular, in the class IV present mlāya-ti (AVP, ŚB; see Kulikov 2012: 599-600), is fundamentally intransitive; its causative mlāpáya-ti first appears in the Atharvaveda (see Jamison 1983: 143).

śar (śr̥) // śrā 'become ready; cook'

The only early Vedic occurrence of a present form (part. śrā́yant-) of the ā-root śrā appears in a rather difficult construction (40), which can be tentatively interpreted as intransitive, adopting Karl Hoffmann's translation of the passage:

The -áya-causative śrapáyati 'cooks, prepares' first appears in the Atharvaveda (attested, in particular, at AVŚ 11.1.4; see Jamison 1983: 145). The reduplicated causative aorist first occurs in the Brāhmaṇas (ŚB-Mādhyandina 3.8.2.28 = ŚB-Kāṇva 4.8.2.21 áśiśrapāma 'we have cooked').

Evidence for the syntactic type of the C-root śar/śr̥ (aniṭ type) is scant. It is only attested in the verbal adjective śr̥tá- 'cooked; ready' (RV+), which might be based either on a transitive ('cook') or on an intransitive ('become ready') verb.

havi (hū) // hvā 'call'

The verb havi (hū) is fundamentally transitive (presents hávate and hváyati 'calls', pf. juhā́va 'has called', etc.; see Gotō 1987: 347ff.; Kümmel 2000: 606ff.), but its passive (pres. hūyá-te RV+; passive aorist participle huvāná- RV; cf. (41–42)) is well-attested from early Vedic onwards (see Kulikov 2012: 306-310):

The root variant hvā (= full grade II), has probably arisen on the model of some ā-roots which form -áya-presents, such as dhā – dháyati 'sucks' and dā – -dáyate 'distributes' (i.e. dhā : dháyati = X : hváyati). All formations built on the root variant hvā, viz. agent noun hvātar- JB, fut. -hvāsya-ti/te, caus. -hvāpayati ŚrSū., etc., first appear in late Vedic texts, thus being of little comparative value (though cf. Late Avestan zbātar-); see Gotō 1987: 350, fn. 863; Kümmel 2000: 608; LIV 180-181 for a discussion. All these formations attest the transitive syntax.

4. Syntactic Features of the ā-Verbs: A Recapitulation

The results of the present study are summarized in Table 1 below. The members of the above-discussed verbal pairs are distributed across five syntactic classes in accordance with their transitivity features. Two non-diffuse classes include (1) intransitive verbs causatives of which are unattested or exceptional in early Vedic texts (i.e. in the RV and AV); and (5) transitive verbs passives of which are unattested or exceptional in early Vedic. Three diffuse classes consist of (2) basically intransitive verbs -áya-causatives of which are attested from early Vedic onwards (weak-diffuse intransitives); (3) verbs which are well-attested in both intransitive and transitive (causative) usages; and (4) fundamentally transitive verbs intransitive (passive) derivatives of which are well-attested from early Vedic onwards (weak-diffuse transitives). Verbs of these five syntactic classes can be arranged in accordance with their degree of diffuseness/non-diffuseness in terms of the following Diffuseness Hierarchy (43):

Note: ā-verbs are shown in boldface.

However variegated the syntax of the C- and ā-verbs might appear, there is at least one remarkable feature (tentatively formulated in Section 2) which is shared by nearly all ā-verbs and makes this distribution non-random. The ā-verbs (shown in boldface in the table) generally attest much less syntactic flexibility, being employed either mostly/exclusively in intransitive usages, or mostly/exclusively in transitive usages, and thus belong to the non-diffuse syntactic type. The corresponding base verbs (C-verbs) show a great variety in syntax, but typically are more diffuse (= more flexible in transitivity), cf. especially yā (intransitive) // ay (i) (intransitive and transitive), trā (transitive) // tari (tr̥̄) (intransitive and transitive), drā (intransitive) // drav (dru) (intransitive and transitive-causative). Within the pair pyā // pay/pi (payi/pī?), pay(i) (pī̆) is well-attested in both intransitive and transitive usages already in the RV, while pyā is predominantly intransitive; -áya-causatives first occur in the AV (4x)); thus, the ā-verb is clearly less diffuse (weak-diffuse) than the base verb, at least in the language of the RV. There are also a few pairs where both members belong to the same syntactic class, cf. dhyā // dhayi (dhī) (both transitive) and psā // bhas (both transitive). The only pair where the ā-verb can be considered (somewhat) more diffuse than the corresponding C-verb is mlā // mari (mr̥̄). mlā is fundamentally intransitive, whilst mari (mr̥̄) is transitive, but the causative of the former, mlāpáya-ti, is a bit older (AV+) than the passive of the latter, -mūryá-te (ŚB). In fact, this seems to be an exception that proves the rule: due to the difference in final sonants (l/r) (probably a dialectal feature), the historical relations between mlā and mari (mr̥̄) are more blurred than those between the members of any other root pair, and synchronically they clearly do not belong together.

As far as more specific correlations between the syntactic characteristics of the verbs and the type of formal relationship between C- and ā-roots are concerned, the following regularities can be observed.

(i) Within the pairs which follow the schwebeablauting pattern CaRi (CR̥̄) // CRā (i.e., in diachronic terms, *CaRH- // *CRaH-), the ā-member is often transitive, as opposed to the (more) diffuse C-verb; cf. especially tari (tr̥̄) 'pass, carry across' // trā 'protect, rescue' and pari (pr̥̄) // prā 'fill'. It is interesting to note that present passives with the suffix -ya- and passive aorists (i-aorists) are unattested in Vedic for most of these ā-roots. Thus, aprāyi is a hapax, which only appears in the RVKh. and Atharvaveda; dhmāyate first appears in late Vedic; pass. trāyate 'is (being) protected, is (being) rescued' does not occur before Classical Sanskrit; for other ā-roots -ya-passives and i-aorists are unattested.

(ii) By contrast, many ā-verbs which follow the pattern CaC // C(C)ā, i.e., in diachronic terms, contain the root enlargement (suffix) -ā- (< PIE *-eH-), are (predominantly) intransitive, as opposed to the (more) diffuse C-verbs. Note, in particular, that present causatives with the suffix -(p)áya- (well-attested in early Vedic for some roots in -ā such as sthā 'stand' and dhā 'suck') are (relatively) late or entirely lacking for the ā-verbs (ā-roots) of the CaC // C(C)ā-pairs. Thus, causatives of yā and drā first appear in the Brāhmaṇas; causative of gā is unattested. The intransitivizing effect of -ā- is also fairly obvious in the pair dah // kṣā 'burn': unlike dah, which is basically transitive but later is drifting into the diffuse type, kṣā is a predominantly intransitive verb, which forms an -áya-causative.

5. Concluding Remarks: Historical Sources of the C//ā-Alternation and Its Synchronic Status

Much remains unclear about the origins of the above-formulated correlations between the formal patterns of the C//ā-alternation and the syntactic features of the verbs in question. In general, evidence from Indo-European languages outside Indo-Iranian furnishes few parallels to the syntactic patterns described in Section 4. Moreover, many of the ā-verbs have no reliable cognates outside Indo-Iranian, and, thus, we have to look for the origins of this syntactic patterning on Indo-Iranian (or even Indo-Aryan) ground.

There may have been several sources of the correlations between the attested formal patterns and syntactic features.

(i) In the case of the CaC//CCā-type, the (predominantly) intransitive character of some ā-verbs may be a vestige of the intransitive/stative function of the hypothetical Proto-Indo-European suffix *-ē- (*-eH-). In fact, as mentioned above, evidence for reconstructing this 'stative' suffix in ā-verbs is scant: while in the 'stative' suffix *-ē- we have to reconstruct h₁ (*-eh₁-; see Beekes 1995: 230), in most of the above-discussed ā-roots we are probably dealing with the reflex of another laryngeal, h₂. The full evidence can be summarized as follows (the reconstruction mostly follows Mayrhofer's EWAia and LIV):

Apparently, there are as few as one or two root pairs where the intransitivity of the ā-verb can be explained as a direct reflex of the intransitive function of the PIE suffix *-eh₁-. Note, however, that the development of the syntactic features ('non-diffuseness') of the ā-verbs should probably be dated to Proto-Indo-Iranian, where the three PIE laryngeals have fallen together. Accordingly, it cannot be ruled out that just a few (derived) roots with the reflex of the PIE 'stative-intransitive' suffix *-eh₁- > PIIr. *-aH- (*dhgu̯h-eh₁-?, *ḱl-eh₁-?) could trigger and/or support the development of similar syntactic properties of the verbal forms derived from all *CC-aH- roots, irrespective of the quality of the PIE laryngeal.

(ii) In some cases, the syntactic features of the formations built on different grades of one verb/root (cf. the transitive aorist aprāt as opposed to the intransitive present pū́ryate and transitive-causative present pr̥ṇā́ti) could be associated with the corresponding (C- vs. ā-) root variants. Subsequently, one paradigm could split in two sub-paradigms, and, accordingly, one lexical unit (verb) gave rise to two different (albeit etymologically and derivationally related) verbs. Thus, the transitive syntax of the root aorist áprās could be generalized for all formations built on the full grade (II) of the root pr̥̄ // prā 'fill', as opposed to formations derived from the zero grade (pres. pū́rya-te, pūryá-te, pr̥ṇā́ti, pr̥ṇá-te), which, eventually, has led to the split of one single lexical unit in two, tari (pr̥̄, pūr) 'become full; fill' and prā 'fill' (see Albino 1999; Kümmel 2000: 325-328), differing in syntactic features: diffuse vs. (predominantly) transitive. This difference in syntax could be expanded to another root pair following the same pattern (CaRi // CRā), tari (tr̥̄) 'pass, carry across' // trā 'protect, rescue'. In some cases, this syntactic difference could be supplemented with idiomatic shifts (cf. tari (tr̥̄) 'pass, carry across' // trā 'protect, rescue'; man 'think, believe; respect' // mnā 'mention'), but their character is of a less regular nature than the above-discussed syntactic oppositions.

(iii) Finally, it cannot be ruled out that the difference in syntactic properties between some historically (and semantically) unrelated but formally similar roots has contributed to the development of the functional (syntactic) value of the C//ā-alternation. Particularly instructive is the case of jani 'be born; generate' // jñā 'know'. In spite of the lack of semantic and historical connections between these two roots (see above), their formal similarity and the remarkable difference in syntactic behaviour (jani is diffuse; jñā is fundamentally transitive) could have supported the syntactic model of the etymological CaRi // CRā pairs such as pari (pr̥̄) // prā.

To conclude, a few remarks on the status of the C//ā-alternation within the system of verbal categories and transitivity oppositions will be in order. Although its connection with such syntactic features of the verb as transitivity or lability does not raise any doubt, at least for a considerable number of root pairs, it would be incorrect to consider this morpho(phono)logical phenomenon as a valency-changing category. The idiomatic character of changes observed for several verbal roots does not allow to characterize this morphological operation as one of (in)transitivizing derivations known from typology. Rather, we are confronted here with quite a complex phenomenon relevant both for the semantics of the verbal roots and for its paradigmatic properties, and the function of this operation should be qualified in terms of quite vague tendencies, rather than in terms of strict rules. Such synchronic 'vagueness' of the C//ā-alternation must be due to the heterogeneity of its origin (see above) as well as to its diachronic instability. Before it had rooted in the Old Indian verbal system, it started to lose its functional features as early as by the end of the early Vedic period and, especially, in middle Vedic – most probably, due to drastic changes in the system of valency-changing categories and, foremost, because of the increasing productivity of morphological causatives (with the suffix -áya-) and passives (with the suffix -yá-; see, in particular, Kulikov 2006a: 75ff.). Nevertheless, the importance of this phenomenon for establishing the Old Indo-Aryan system of transitivity oppositions, especially in the early Vedic period, is quite obvious. Although the C//ā-alternation was operating for a relatively small part of the verbal dictionary, and its status should be qualified as submorphic, rather than as morphological, it played an important role for establishing the syntactic potential of the vast fragments of verbal paradigms, influencing some basic trends within the Vedic verbal system. The status of such submorphic phenomena and their diachronic typology, both in Indo-European and beyond, is poorly studied and represents one of the most interesting domains of research in historical linguistics in general, and in Indo-Iranian and Indo-European linguistics, in particular.