1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References

I discuss the most controversial and widely studied part of the Ṛgvedic Sanskrit clause, the clause-initial string. I focus in particular on a few sets of words — relative pronouns, demonstrative pronouns and preverbs — arguing that their treatment in previous analyses requires revision. This is because they can be, in some contexts, clitics. I argue that clisis must be defined in primarily syntactic rather than prosodic terms: Ṛgvedic clitics all share certain syntactic properties, but they do not all necessarily share any single prosodic property (such as lack of lexical accent). I subsequently examine the consequences of this reanalysis of pronouns and preverbs in the clause-initial string for syntactic accounts of Ṛgvedic word order.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References

The Ṛgveda is the earliest surviving text in Sanskrit, composed perhaps c. 1500–1200 bc; its language is considerably more archaic than later forms of Sanskrit, and is syntactically more similar to other ancient Indo-European languages such as Old Avestan and Ancient Greek.1 As such it has long been of considerable importance for the comparative study of old Indo-European languages and for the reconstruction of Proto-Indo-European. Ṛgvedic word order is often characterized as ‘free’, which means merely that, unlike English, Ṛgvedic word order is not based on the grammatical function of the constituents involved, but rather on the discourse functions, or information structure, of those constituents. So, for example, the first XP in the sentence is not necessarily the subject, as usually in English, but the most topical or highly focused element; Viti (2010) has recently shown that all possible orderings of S(ubject), O(bject) and V(erb) are found in the Ṛgveda, and that these orderings are based on the respective topicality or focus assigned to each element of the clause by the speaker, i.e. information structure. This is true of word-order variation in most if not all ‘free word-order’ languages; it is well known from, for example, Australian languages such as Warlpiri (Austin & Bresnan 1996; Laughren 2002; Simpson 2007; Simpson & Mushin 2008), and is found in other Indo-European languages such as the closely related Ancient Greek (Dik 1995; 2007), and modern Russian (King 1995).

Due, perhaps, to its apparently ‘free’ word order, it was only in the 1980s that work on Ṛgvedic word order took off, primarily focusing on the start of the clause, where the greatest degree of regularity (or the smallest degree of irregularity) could be observed. Various approaches to what is often called the ‘initial string’ have been proposed, by Hock (e.g. 1982; 1989; 1996; 1997a), Hale (e.g 1987a,b; 1996; 2007), Krisch (1990, 1997, 2002) and most recently Keydana (2011); these will be discussed in more detail below. In this paper I revisit the Ṛgvedic clause-initial string from a primarily descriptive perspective; by examining the evidence of certain groups of words more carefully I aim to show that certain assumptions commonly held about the syntactic structure of the start of the Ṛgvedic clause should be revised, and that certain groups of words, in contrast with traditional analyses, should be treated as clitics.

In section 2 I briefly introduce second-position clitics and some of the problems associated with them; then in section 3 I survey previous approaches to the Ṛgvedic initial string; in section 4 I argue for the clitic status of some relative and demonstrative pronouns; in section 5 I argue the same for clause-initial preverbs; finally in section 6 I examine the consequences for syntactic analyses of the Ṛgvedic clause-initial string.

2. Second position

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References

Wackernagel's Law (Wackernagel 1892) can be approximated by the statement that unaccented elements occur in second position in the clause. But for the Ṛgveda at least, this is an oversimplification. Example (1) fits the pattern, but in examples (2)–(4) unaccented elements appear, at least, to be later than second position.

(1)dyaúś ca tvā pṛinline image yajñíyāso/hótāraṃsādayantedámāya
Heaven and you Earth worship_worthy down priest set for_house
‘And Heaven and Earth, (and) the worship-worthy (gods), establish you as priest for the house.’ (3.6.3ab)
(2)eṣáíndroarhatipītím asya
this Indra deserves drink of_it
‘This Indra deserves a drink of it.’(2.14.2d)
(3)utá no marcáyādánāgasaḥ
also or who us would_harm innocent
‘or also who would harm us innocent.’(2.23.7a)
(4)divyinline image inline imagepoabhíyád enam inline imageyan
divine waters to when him came
‘when the divine waters came upon him.’(7.103.2a)

In (2) the clitic pronoun asya ‘of it’ (the intoxicant Soma, mentioned in the previous line) appears cliticized to its governing noun pītím ‘drink’; this is an uncommon but nevertheless entirely possible position for enclitic pronouns, and they are likewise found cliticized to governing verbs within VPs (cf. example 10) and prepositions within PPs. Similarly clitic particles which modify individual words or phrases, as opposed to whole clauses, appear enclitic on or within the constituents they modify (cf. example 24 with fn. 33). In all these cases the clitics appear in their expected or ‘base’ syntactic position, as if they were ordinary, non-clitic words. It is not these instances which will interest us here, and they will be considered no further.

In (3) and (4), on the other hand, the clitic pronouns no and enam are positioned in the ‘initial string’ of their clauses according to their morphosyntactic status as clitics rather than their syntactic status as pronouns. That is, they are examples of Wackernagel clitics, only they do not appear in ‘absolute’ second position (i.e. following the first non-clitic word, ignoring other clitics). In relative terms, this is uncommon. To take one example, the 2sg.acc pronoun tvā ‘you’ is attested 589 times in the Ṛgveda, of which it appears in ‘absolute’ second position 532 times, that is 90.3 per cent. At least half of the remaining 10 per cent appear inside VP or PP constituents, as discussed in the previous paragraph, leaving at most 5 per cent of instances in which the pronoun is positioned as a clitic in the initial string but does not appear in absolute second position.

However, it is possible to define second position in a number of ways. In syntactic terms, absolute second position directly following the first word might mean relatively little. If the first word in such instances were always a separate syntactic constituent, there would be no syntactic difference between clitics following those words and clitics in other contexts appearing after clause-initial constituents made up of two or more words, as long as all the words preceding the clitic formed a single constituent. In other words the clitics would be in second syntactic position, which, when the first constituent was a single word, would happen to be the same as ‘absolute’ second position. Alternatively, it could be that some or all ‘second position’ clitics are not so much second position in the clause, but second position within their constituent. If clitics in absolute second position could be construed in the same constituent as the preceding (or following) word, then syntactically their position might be explicable not in terms of the clause as a whole, but in terms of the constituent of which they were a part.

The uncertainties and ambiguities of Ṛgvedic syntax mean that it is difficult to be certain, in a large number of instances, precisely what we are dealing with. To take tvā once again, in 402 (75.6 per cent) of its 532 appearances following the first word of a clause it is possible but not necessary to construe the pronoun as part of the same constituent as the immediately preceding or following word. This is the case, for example, when the preceding or following word is an accusative referring to the same element in the clause as the pronoun (5), or when the preceding or following word is a preverb or verb functionally governing the accusative (6).

(5)trātinline imageraṃ tvā taninline imagenāṃhavāmahe
protector you of_selves we_call
‘We call on you (as) the protector of our selves.’ (2.23.8a)
(6) inline image tvā viśantvíndavaḥ
into you let_enter drops
‘Let the drops enter you.’ (1.15.1b)

It seems statistically unlikely, however, that in 68.3 per cent (402/589) of its occurrences tvā simply happens to be in absolute second position in the clause due to its being the second element in the first constituent in that clause. All things being equal, and assuming an average of somewhere over two constituents per clause, it would be surprising to find tvā in the first constituent that often, let alone second position within that constituent.

Furthermore, in 35 instances tvā appears following the first word, but between two words with which it cannot itself be construed, but which can themselves be taken as a single constituent. This is the case in (1), where tvā appears between the two nominative nouns, dyaús ‘heaven’ and pṛthiv inline image ‘earth’, which together form part of the subject of the clause, and which could be interpreted as a single constituent asyndetically conjoined.2 In these instances the pronoun really does appear to be positioned following the first word, blind to the fact that it interrupts a separate syntactic constituent. Ṛgvedic Sanskrit does permit discontinuity of constituents, i.e. all elements of any XP need not appear adjacent or contiguous in the clause; it would, therefore, be possible to assume that the first element in all these instances is, despite appearances, syntactically separated from the word following the clitic. However, given the strong tendency for tvā to appear following the first (non-clitic) word, forcing such an interpretation onto the data in every such instance seems unnecessary. Table 1 summarizes my count of the relative positions of tvā in the Ṛgveda.3

Table 1. Position in clause of tvā, 2sg.acc pronoun
#inline image # [inline image] inline image]…402
# [inline image] inline image95
# [inline image(])inline image# [inline image]inline image5
# [inline image] inline image2
# [inline image[inline image(])inline image# [inline image[inline image]inline image40
# [inline image [inline image]inline image4
# [inline image][inline image yá-/s(y)á-] inline image 6

In Table 1 Σ represents a single non-clitic word, and other clitics are ignored. Of the 532 instances of tvā following the first word of the clause, 402 could, but need not, be taken as part of the same constituent as that first word, as discussed above and shown in (7). Thirty-five appear to interrupt a constituent made up of the first word and one or more words following tvā (cf. example 1); I have represented this by placing tvā outside the constituent with an arrow pointing to its actual position (cf. section 6.2). The final 95 instances of tvā follow a word which must be interpreted as a separate constituent, i.e. the clitic pronoun follows the whole of the first constituent of the clause, which happens to consist of a single word, and is therefore the second prosodic and syntactic element in the clause (8). There are only seven instances in which tvā follows two or more words which together form the first constituent of the clause. In five of these it would be possible, but again not necessary, to interpret the pronoun as a part of that first constituent. In the other two instances (4.16.19a, = example (9), and 8.6.20a) the clitic cannot be construed with the preceding constituent but, like the 95 instances in which it follows a single-word constituent, must be analysed as the second syntactic constituent in the clause.4 The same pronoun can also occur even further from the start of the clause. In 40 instances tvā can, but again need not, be interpreted as part of a larger syntactic constituent which is not the first constituent of the clause. This figure can actually be broken down further: ten of these follow (and may form part of) the second constituent and therefore may, if not taken as part of the second constituent, be parallel to the patterns in the rows below. The other 30 appear in a constituent, VP or PP, which is neither the first nor the second constituent in the clause, as in (10), parallel to the position of asya in (2). In four instances tvā unambiguously follows the second constituent, and so must be interpreted as the third syntactic element in the clause (11).5 The six instances in the final row of the table can be analysed as identical to these four, the second element merely happening to be a form of a relative demonstrative pronoun, as in (12), parallel to no in (3) or enam in (4). I have noted them separately, however, in anticipation of the following discussion.

(7)prá tvā yajñinline imagesaiméaśnuvantu
forth you offerings these let_them_reach
‘let these offerings reach forth to you.’ (6.23.8b)
(8)vayámu tvā dívāsuté/vayáṃnáktaṃhavāmahe
we also you by_day in_Soma we by_night call
‘We also call on you in the Soma by day, we call (on you) by night.’(8.64.6ab)
(9)ebhírninline imagebhirindratvāyúbhiṣ ṭvā /maghávadbhirmaghavanvíśvaājaú
with_these men Indra desiring_you you with_generous generous all contest
‘with these men who desire you, O Indra, with the generous (patrons I call on) you in every contest, O generous one.’ (4.16.19ab)
(10)tváminódāśúṣovarūt'/étthinline imagedhīrabhínákṣati tvā
you mighty of_worshipper defender so_thinking to who approaches you
‘you are the mighty defender of the worshipper who, thinking thus, approaches you.’ (2.20.2cd)
(11)yádadya tvā prayatíyajñéasmín/hótaścikitvo'vinline imageṇīmahīhá
because today you beginning sacrifice this priest perceptive chose=here
‘because we chose you here today, O perceptive priest, as this sacrifice was beginning.’ (3.29.16ab)
(12)jātáṃyát tvā páridevinline image ábhūṣan/mahébhárāya puruhūta víśve
born when you around gods strengthened for_great for_battle much-invoked all
‘when all the gods strengthened you, (just) born, for the great battle, O much-invoked one.’ (3.51.8cd)

Clitic pronouns are not the only clitics in Ṛgvedic Sanskrit. There is also a set of clitic particles which show broadly the same positional distribution as these clitic pronouns. In addition there is a class of clitic conjunctions, which only ever appear following the first word of their domain (specifically, following the first non-enclitic word); these include, for example, the common conjunctions ca ‘and’, seen in (1), and ‘or’ in (3). The contrast between the clitic pronouns (and particles) and the conjunctions is particularly clear in this latter example, where the conjunction follows the first phonological word, while the clitic pronoun appears later.

All attempts to explain the ‘initial string’ of the Ṛgvedic sentence are faced with the challenge of accounting for this variety of clitic positioning at the start of the clause.

3. Previous approaches

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References

In this section I survey four previous theoretical approaches to the Ṛgvedic clause-initial string. All have their advantages and weaknesses, and although it will become clear that my own views are closest to those of Hale's, the central claims of this paper, as developed in sections 4 and 5, would be valid also in other theoretical frameworks.

3.1. Hale: movement and ‘prosodic inversion’

Hale (1987a, b; 1995; 1996; 2007) was the first to study Wackernagel clitics, and hence the start of the Ṛgvedic clause, from a generative syntactic perspective. His earlier approach differs from his later analysis in certain important respects, and it is the latter that I focus on here.6 Working in a Government & Binding (GB) framework, Hale assumes movement out of the IP into various phrase-structure positions to account for clitic positioning in the clause-initial string. So for (3), above, Hale (2007: 208, example (9.31)) gives the following syntactic tree:image

In Hale's framework, wh words move into SpecCP, clitics from the IP (pronouns or sentence particles) move into inline image, and a topicalized element can appear in SpecTopP; this accounts for the potentially ‘late’ second position of clitic pronouns and sentence particles discussed above. A further rule is needed, however: any enclitic which ends up in the syntax at the start of the clause moves to the right of the first phonological word in the output by means of Halpern's (1995) ‘prosodic inversion’, a ‘late-level rule of sentence phonology’ which ensures no clitic appears in an unpronounceable position. This is the case with in the above tree, which as a sentence conjunction is generated to the left of the rest of the clause, but in the output (3) appears after utá. This rule applies not only to enclitic conjunctions, but to any clitic, so that if inline image were the first filled position in the tree, and were filled by a clitic, that clitic pronoun or particle would undergo prosodic inversion such that it appeared following the first phonological word of the IP.

Hale's approach has been criticized by Hock (1996) and Keydana (2011), both of whom seek to explain clitic positioning in purely prosodic terms. Keydana (2011) in particular criticizes Hale's approach for its stipulative nature regarding clitic positioning. Insofar as clitics are a unique and problematic phenomenon, however, a degree of stipulation is required in any explanation (even Keydana's own), and in purely formal terms Hale's approach can account for all or nearly all the data within his chosen framework. Nevertheless it will become clear that some of his assumptions, in particular regarding wh words and preverbs, would have to be revised to take account of the analyses presented in the following sections.

3.2. Hock's ‘phonological template’

The ‘phonological template’ approach to the Vedic initial string was developed by Hock (1982; 1989; 1992; 1996; 1997b) and followed by Schäufele (1996). According to Hock, the initial string of a Vedic sentence consists of a series of positions, or slots, into which the various elements such as particles, conjunctions and pronouns fit. The ‘phonological’ element to the theory refers to the idea that (in principle) these slots alternate between accented and unaccented words. Hock's maximal template is shown in (14); all positions except 1 are optional; all except 1 permit doubling.

inline image P inline image E inline image
inline image inline image

The first position in the sentence, ‘NEXUS’, contains connectives which are in some way extra-sentential as they do not ‘count’ as the first element in the clause, parallel to e.g. German und, aber, denn in contrast to dass; however Hock (1997b) rejects this position, and it need not be considered further.7 The first slot proper contains an accented word/phrase (inline image) or an accented pronominal (inline image); the second contains unaccented sentence particles and/or conjunctions (P); the third contains accented sentence particles (inline image) or accented pronominals (inline image, if these do not appear in position 1); the fourth position contains enclitic pronouns (E); the fifth slot was originally suggested as a possible position for accented pronominals, but there is little evidence for such a position and Hock later stated that it may not exist. Example (15) shows how some of the passages above would be analysed under Hock's schema:

dyaús ca tvā (1)
utá no (3)
divy inline image po abhí yád enam (4)
jātáṃ yát tvā (12)

Hock's phonological template has been criticized by Hale (1996) and most recently Keydana (2011). In particular they argue that Hock does not make clear his understanding of the interface between syntax and phonology, such that there is little theoretical support for his claims. In Hock's account phonology is the driving force behind word order in the initial string, but the supposed phonological basis, the alternation of accented and unaccented slots, is rarely reflected in the phonological output (as seen in the examples above), since most slots can be omitted or doubled. In fact this potential for omission and doubling, combined with the two or three possible positions for accented pronominals and the fact that inline image could in principle host a word of any kind, means that practically any word order can be accounted for, and indeed many unattested orders are predicted. As noted also by Keydana (2011), Hock's template is descriptively powerful, but so powerful that it lacks explanatory adequacy.

3.3. Krisch's ‘Schemata’

Krish (1990; 1997; 2002) treats clitic positioning in Sanskrit alongside parallel phenomena in Greek, Hittite and other related languages in an attempt to account for Indo-European word-order patterns. He proposes two basic patterns of clitic positioning, called Schemata or Satzbaupläne, as shown in (16).

(16)a.i.# X(E)… #
ii.# C(E)… #
b.# X(-inline image)… #

The first pattern consists of two sub-types: in the first, a clitic follows a topical or focused word in clause-initial position; in the second, the clitic follows an initial element appearing in inline image. The second pattern attempts to account for clitics appearing later than the second word, by licensing a ‘complementizer’ clitic slot after the first word and a second clitic slot following a complementizer in inline image which is not itself the first word in the clause. The first, complementizer clitic slot is filled by conjunctions such as ca ‘and’ or ‘or’, discussed above; whether these words can really be treated as complementizers is of little importance, since the slot could easily be re-labelled to cover co-ordinating conjunctions. So examples (1) and (5)–(7) are clear examples of Krisch's first pattern, and (3) and (12) clear examples of his second.

This can account for many of the ‘initial-string’ sequences found in the Ṛgveda, but it cannot account for all. This has recently been demonstrated by Keydana (2011), who also points out that by extending Krisch's model with a third schema, as in (17), the model would be descriptively adequate for all Ṛgvedic patterns attested.

(17)# X inline image… #

The deficiencies of Krisch's model have recently been discussed by Keydana (2011); he raises a similar objection to Krisch's account as to Hock's ‘phonological template’: it is not clear theoretically what these patterns are, nor how they are licensed. The model is explicitly influenced by Hale (1987a), but by disregarding his theoretical framework the model is reduced to a relatively arbitrary set of orderings, the choice between which is not clearly determined by anything except the particular clitics appearing in the clause.

3.4. Keydana: PF ordering

The most recent approach to the clause-initial string is that of Keydana (2011), who argues that clitic positioning is entirely phonological (taking place at PF). He claims that clitic conjunctions are positioned following the first phonological word of an intonational phrase, whereas clitic pronouns and sentence particles are positioned following the first phonological phrase of an intonational phrase; it is this latter claim which accounts for ‘late’ clitics such as the pronouns in (3) and (4). As Keydana admits, however, there is no evidence as to what is or is not a phonological phrase in the Ṛgveda.

There are further difficult assumptions specific to Keydana's account which he himself admits, such as an obligatorily filled left periphery. Given these problems, the value of Keydana's approach stands or falls on his fundamental assumption that clitic positioning is a purely prosodic, i.e. PF, phenomenon. If this were clearly the case, or if it were even a more reasonable likelihood than any other, it would be clearly preferable to syntactic accounts such as those of Hale and Krisch, and perhaps preferable also to the only other purely prosodic account, that of Hock, by virtue of its clear theoretical basis.

Keydana does not discuss the appearance of clitics outside the initial string, apparently in ‘base-generated’ syntactic position, as in (2) and (10). In these positions the clitics are no less prosodically deficient than they are in the initial string, yet they are not, it seems, positioned by the phonology following either the first phonological word or phrase of the clause. Phonology alone cannot account for this variation; in conjunction with syntax it could, if, say, it is only clitics that are moved out of the IP which are positioned by phonology. But Keydana's main criticism of Hale's approach is that a combined syntactic and phonological approach is problematic, and introducing any syntactic positioning into Keydana's theory would fundamentally undermine its theoretical foundation.

As a purely prosodic account, Keydana's theory assumes that clisis is a fundamentally phonological property of words. As discussed in more detail below, however, it is by no means clear that all Ṛgvedic clitics were prosodically deficient in the same way or even at all, although many shared, for example, a lack of lexical tone. An abstract prosodic deficiency of some kind or another could be attributed to all clitics only by stipulation. And we could only know which words to attribute such a property to on the basis of their syntax. For this reason, as discussed in more detail below, I would maintain that the property shared by all clitics is in fact syntactic in nature. Nevertheless if such a phonological property were admitted, and the other problems overcome, Keydana's approach, like that of Hale, could deal with at least the majority of the data in a theoretically principled manner.

A further theoretical point can be raised here. Keydana claims that the ‘null hypothesis’ regarding clitic placement is that it is prosodic (‘a PF phenomenon’). Why this should be the null hypothesis is not made clear.8 Clitics are, undoubtedly, constrained by phonology. However, the order and arrangement of words is, by definition, the field of syntax.9 At the same time, there is no clear cross-linguistic evidence for even the possibility of purely phonological or PF movement or positioning of words.10 The null hypothesis, in my opinion, ought to be that all word-order phenomena can be accounted for in the syntactic component of the grammar; at the very least as much as possible of the order and arrangement of words should be accounted for in the syntax. Given the entirely different theoretical standpoint taken by Keydana, perhaps all that can be said is that his approach is simply incompatible with the theoretical position taken here.

3.5. The initial string: a descriptive summary

Hock's analysis of the Ṛgvedic initial string as a template is based on the observation that whatever elements appear in the initial string in a particular clause, the relative order of those elements is almost invariably the same as in any other initial string. Descriptively, then, the initial string can be analysed as an ordered series of optional elements.

By combining the evidence of different Ṛgvedic clauses, such as those given above, it is possible to build a template, as we have seen Hock did. Hock's template is actually an abstraction of the template directly recoverable from the data, since his aim was to arrange syntactic categories in phonological terms. A less abstract syntactic template for the initial string is the following: image

All elements in the initial string are in principle optional. Ignoring the first column for the moment, the first possible element is a non-enclitic conjunction such as utá, which cannot be preceded by any other element of the clause (assuming it is a clausal, not phrasal, conjunction). Then follow one or perhaps two positions which can be filled by any XP from the clause; these are usually considered to be topicalization/focus positions (e.g. by Hale 1987a; 1996; Krisch 1998; Keydana 2011), and will be discussed in more detail below; I refer to these positions as DF (Discourse Function) positions. Following this a preverb can occur, after which we find the regular position of the relative and demonstrative pronouns yá-, sá-, syá-, etc.11 Then come enclitic sentence particles and finally enclitic pronouns. Enclitic sentential conjunctions such as ca ‘and’, ‘or’, u ‘and’, separated from the other categories in the diagram above by the vertical line, are the only elements of the initial string with variable position relative to the other elements: they obligatorily follow the first word of the first non-enclitic element of the clause, which may be the non-enclitic conjunction (as in (3)), the first word of an XP in a DF position, a preverb, pronoun or, if no element of the initial string is filled, the first word of the rest of the clause.

Apart from Hock, who deals not in clitics and non-clitics but in accented and unaccented words (for the difference see below), all previous approaches assume that the only enclitics appearing in the initial string are enclitic conjunctions, enclitic pronouns, and enclitic sentence particles (those positions given in italics in (18)). In the following sections I discuss evidence which brings into question that assumption, and suggests that our view of the Ṛgvedic initial string requires some alteration.

4. Relative and demonstrative pronouns

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References

Whatever else may appear at the start of a Ṛgvedic clause, if a relative pronoun appears in the initial string (which is their regular location) it will be directly followed by any clitic sentence particles and personal pronouns which happen also to appear in the initial string; this rule is (almost) invariable, and can be seen in the template given above. Explanations for this vary: for Hale, wh words move into SpecCP while clitics move into inline image, which in terms of linear order directly follows. For Keydana, wh words move into SpecCP or inline image, positions which happen to be at the end of the first phonological phrase of the intonational phrase corresponding to the clause and within which clitics are positioned; since clitic pronouns and particles are positioned following the first phonological phrase, they appear directly following any relative pronoun.

4.1. Interrogatives and relatives

All previous approaches assume that relative and interrogative pronouns are positioned in the same way (as ‘wh words’ for Hale and Keydana).12 However close examination reveals that this is not, in fact, the case. Interrogative pronouns are rarely preceded by another constituent, while this is relatively common in the case of relative pronouns; moreover while the latter are occasionally preceded by more than one syntactic constituent, no interrogative pronoun in the Ṛgveda is ever preceded by more than one constituent. Additionally, the regular position of relative pronouns is following any initial-string preverb, whereas interrogative pronouns are never preceded by preverbs. Taking this evidence together, it appears that interrogative pronouns surface nearer to the start of the clause than relative pronouns.

Table 2 shows the relative positions of the nominative singular masculine of the relative pronoun yá- and the interrogative pronoun ká- respectively.13

Table 2. Relative positions of yáḥ and káḥ
Position yáḥ káḥ
# yá-/ká-72269.6%11088%
# [inline image] yá-/ká-30929.8%1512%
# [inline image] [inline image] yá-/ká-60.6%00%

The Table shows, as stated, that the interrogative never appears more than a single constituent from the start of the clause, unlike the relative (cf. (4) and (10)); occasionally an interrogative appears late in a clause, but in such cases it always follows a single constituent, as in (19) where it follows the VP. Due to the relative infrequency of the interrogative pronoun compared with the relative, this gap may be accidental. But in addition, the interrogative is proportionately much more frequent in absolute initial position than the relative (for which see (21) and (22)); this difference at least is not likely to be due to chance. Both can appear following the first word or constituent (cf. (12), (20) and (48)), but this is more than twice as frequent with the relative. The same statistics hold for other case forms, although most are considerably rarer: so alongside fifteen examples of genitive singular interrogative kásya in first position, there is only one example of it following the first constituent (8.84.5a, = (21)).14

Nor do these pronouns appear in phrase-structure positions that are necessarily linearly adjacent (such as SpecCP and inline image). This is because of their contrasting positioning in relation to preverbs: interrogative pronouns are never preceded by preverbs, while relative pronouns regularly are (contrast (21) with (4); approximately one quarter of the 309 instances of ‘second position’ yá- directly follow a preverb). There must therefore be the potential for at least one clear phrase-structure position between interrogative and relative pronouns.15

(19)hárīíndrasyacikāya káḥ svit
horses of_Indra prv perceived who q
‘Who do you think has perceived Indra's two horses?’ (10.114.9d)
(20)dinline imageśema kásya mánasā/yajñásyasahasoyaho
we_should_worship of_what with_mind of_sacrifice of_power son
‘With thought on which sacrifice should we worship (you), O son of power?’ (8.84.5ab)
(21) kám áchāyuñjātherátham
whom toward yoke chariot
‘toward whom do you yoke your chariot (to go).’(5.74.3b)

Looking at the template for the Ṛgvedic initial string given above, it should be clear that interrogative pronouns are generated in the second of the two XP slots, since they can only be preceded by a single XP and are never preceded by a preverb. This corresponds to the fact that this is a DF position, and interrogative pronouns are naturally topical. Relative pronouns are also often topical, and indeed we cannot rule out the possibility that they can appear in this same DF position. In the Ṛgveda there are four exceptional examples of relative pronouns followed by initial-string preverbs, contrasting with the usual rule.16 There is then nothing to distinguish these instances of the relative pronoun from the usual pattern of the interrogative, and of course in the vast majority of instances both interrogatives and relatives appear in initial position with no preverb present, such that there is no reason in these cases not to assume that clause-initial relative pronouns appear in the same position as interrogatives.

It should be clear, however, that this does not account for all instances of relative pronouns in the Ṛgveda. Their ability to surface one constituent further from the start of the clause than interrogatives and their tendency to appear more frequently in second position suggest that another phrase-structure position must be available to relative pronouns in at least some instances.

4.2. Demonstrative pronouns

In what position, then, do such relative pronouns appear? Interestingly, relative pronouns do not share their syntactic positioning with interrogative pronouns, but with certain demonstrative pronouns. The intermediate/weak demonstratives sá-/tá- and syá-/tyá- in particular show broadly the same patterns of distribution as the relative pronoun.17 Table 3 shows the relative position of tám, the accusative singular masculine of the demonstrative sá-/tá-, in all its Ṛgvedic occurrences.18

Table 3. Position in clause of tám, pronoun in RV
# tám41877%
# [inline image] tám10519.4%
# [inline image] [inline image] tám50.9%
# …[inline image tám inline image]…152.7%

I give figures for the accusative singular masculine tám since it is the third most frequent form, while the two more frequent forms, the nominative singular masculine and the nominative/accusative singular neuter tát, can also be used as sentence connectives in initial position, thus skewing their statistics.19 There is in fact a degree of variation in the relative positioning of different case forms of the demonstrative pronouns, but this seems to be due to their occurrence in set phrases. For example tyám, the accusative singular masculine of the pronoun syá-/tyá-, the most common form of that pronoun in the Ṛgveda, appears in first position in 25 of its 46 occurrences (54 per cent), while the nominative singular masculine syá never appears in first position. But this pronoun, particularly in the nominative, is largely confined to set sequences where it follows another word, such as eṣá syá… ‘this very’, utá syá… ‘and this’, etc. Similar variety is also found with the strong demonstrative pronouns, proximal eṣá-/etá- ‘this’ and distal asaú ‘that, yonder.’ Metrical factors also play a part: for example, of the two forms of the nominative singular masculine of eṣá- ‘this’, the preconsonantal variant eṣá appears in absolute first position in 76 of its 85 occurrences (90 per cent), while the pre-vocalic/pre-pausa variant eṣáḥ does so in only six of 23 occurrences (26 per cent). A heavy second syllable is overwhelmingly favoured in all verse types, and since this is regularly possible only with eṣá and not eṣáḥ in line-initial position, the clear dispreference for the latter must be metrical. Nevertheless for all these demonstratives, the same broad patterns are found as for the relative pronoun: they can be and usually are preceded by any initial-string preverb; they can appear following the second constituent of the clause; overall (averaging out differences between case-forms) they are considerably less frequent in absolute initial position than interrogative pronouns.

What positions, then, can these demonstrative and relative pronouns appear in? As stated above, the similarity with the positioning of interrogative pronouns is not entirely superficial. Relative and demonstrative pronouns can often be topical elements. When they appear in first position in a clause, therefore, they can be treated as positioned syntactically in exactly the same way as interrogative pronouns. The following verse illustrates highly topical first-position relative pronouns.20

(22) hatvinline imagehimariṇātsaptásíndhūn/ ginline image udinline imagejadapadhinline image
who having_slain=snake released seven rivers who cows drove by_taking_off
valásya/ áśmanorantáragníṃjajinline imagena/ … sájanāsaíndraḥ
of_Vala who stones between fire produced he O_men Indra
‘He who after slaying the snake released the seven rivers, who drove out the cows after removing Vala, who produced fire between the stones … he, O men, is Indra.’ (2.12.3)

Relative and demonstrative pronouns are not identical in their positioning: there is at least one syntactic feature that groups relatives with interrogatives against demonstratives, namely that clitic pronouns occasionally appear before demonstratives, but never (or only extremely rarely) before relatives and interrogatives.21 Nevertheless, in terms of their positioning relative to other clausal elements, relatives and demonstratives clearly pattern together against interrogatives. To assume, then, that the interrogative and relative pronouns are syntactically identical on the basis of some syntactic similarities, and the fact that they can both be classified as ‘wh words’, is too simplistic. Supporting evidence given by Hale (1995: 79–80) in favour of the claim that relative and interrogative pronouns pattern together (as wh words) in contrast to demonstratives, which entirely differ in their syntactic behaviour, is contrasting sandhi effects with a following word: the former do not show retention of a final -s before a following p or k, while the demonstrative sá- does, this retention (a kind of ‘close’ sandhi, see further below) being largely dependent on the degree of syntactic cohesion between the two words concerned. Hale therefore assumes that this confirms that interrogative and relative pronouns move higher up the tree than demonstrative pronouns, to a position which is sufficiently far removed in syntactic terms from the following element that such ‘close’ sandhi does not occur. In fact there is only one example of sá- showing such close sandhi, and as Hale notes it is not clear whether in that instance the pronoun appears in the initial string (is subject to deictic fronting) or not. Hale's explanation of the few (and therefore more common) instances in which yás, nominative singular masculine of the relative pronoun, shows ‘close’ sandhi before páti- ‘lord’, as an ‘overgeneralization on the part of late poets of the very frequent sequence -s páti-, where the s represents the auslaut of a preceding genitive, to this structurally inappropriate context’, is not entirely convincing, but even accepting this for the sake of argument there is simply no clear evidence that sá- would regularly show ‘close’ sandhi, in contrast to yá- and ká-. In any case I do not suggest that interrogatives and relatives cannot appear in the same syntactic contexts, but that the relatives have also another syntactic possibility which they do not share with the interrogatives, but rather with demonstratives.

As already stated, however, relative and demonstrative pronouns are not syntactically identical to interrogative pronouns. In fact there is evidence that in contrast to the latter, the former can, when non-initial, be enclitic.22

4.3. Clitic positioning

Before examining this evidence, it is worth discussing precisely what we mean by enclitic in Ṛgvedic Sanskrit. Most Sanskrit words have a single accented syllable, marked in transliteration with an acute accent.23 Enclitics, however, generally lack such an accent: the enclitics appearing in the examples given above all exemplify this. And this makes sense insofar as prosodic deficiency is commonly taken to be a central defining feature of clitics.24

However in Ṛgvedic Sanskrit certain accented sentence particles pattern syntactically in precisely the same way as non-accented sentence particles. So (23) looks like (3), with a clitic conjunction following the first accented word and an enclitic pronoun following the second non-clitic word. However, it is in fact parallel to (1), where all the enclitics follow the first non-clitic word; the only difference is that is accented, despite patterning syntactically just like unaccented enclitic sentence particles.25

(23)minline image u ṣú ṇaḥ somamṛtyávepárādāḥ
not and indeed us Soma to_death away give
‘And do not hand us over to death, Soma.’ (10.59.4a)

Several accented particles pattern with the clitic particles, including ‘for’ ((25) and (58)), a subordinating conjunction, and the sentence particles ‘indeed’ ((23), (24) and (60)), and ‘now.’ Accented words that pattern syntactically with clitics are not unique to Ṛgvedic Sanskrit, but are paralleled for example in Ancient Greek.26 On the other hand one major category of word in the Ṛgveda is unaccented but not syntactically ‘clitic’, namely non-initial finite verbs in main clauses.27

Rather than assume two distinct word types, one clitic the other non-clitic, both appearing in the same syntactic position in the initial string (indeed, in the clitic cluster of the initial string) of the Ṛgvedic sentence, the most parsimonious explanation is that both accented and unaccented sentence particles are ‘enclitics’, appearing in the same position in the clitic cluster. The consequence of this is that accentuation is not a defining feature of clitic-hood, at least in Ṛgvedic Sanskrit (and, e.g., Ancient Greek). But if enclitics cannot be defined by lack of accent, how can they be defined? What the enclitics of the Ṛgveda all share are syntactic peculiarities. There are two such ‘peculiarities’ which, although different, share the feature of being roughly associated with ‘second position.’ For the majority of enclitics, this involves the syntactic possibility of appearing in what is, at least roughly, Wackernagel's position near the start of the clause, a position in which they ‘cluster’ with other such clitics, syntactically discontinuous from any larger constituent of which they are a part. For enclitic conjunctions, discussed in more detail below, it involves appearing directly following the first non-clitic word of their domain. The differences between the two broadly correspond to the differences between ‘simple’ and ‘special’ clitics respectively in Zwicky's (1977) terminology. Nevertheless, in both cases, enclisis can be defined in purely syntactic terms (at least for our purposes, if not also cross-linguistically).28 In the Ṛgveda this property applies to a subset of words in the lexicon, which subset largely but not entirely overlaps with the subset of phonologically unaccented words.

Nevertheless this overlap is significant. Clisis is a syntactic phenomenon, but its indirect relation to prosodic deficiency, of one kind or another, is undeniable. Many Ṛgvedic clitics lack inherent accent; even for those that do not, however, there may be other ways to be prosodically deficient. The relative and demonstrative pronouns under discussion always appear accented (at least as transmitted in our texts), whatever position they occur in. As should be clear, however, this says nothing necessary about whether they may or may not be clitics. However, there is evidence that these pronouns are prosodically deficient in other ways. Specifically, when enclitic they do not form an independent prosodic word, but attach to the prosodic word immediately to their left. Evidence for this comes from previously unexplained internal-sandhi phenomena affecting the demonstratives sá-/tá- and syá-/tyá-.

Sanskrit has both internal- and external-sandhi rules: the latter govern the phonological interactions between independent phonological words which appear next to one another in the clause (including between two words appearing together in compound, even where the resulting compound is a single ‘word’ insofar as it has only one accent). Internal sandhi applies within phonological words, and for these purposes clitics do not count as independent phonological words, but as part of the preceding or following word.29 In principle this provides a clear criterion for determining whether a given word is clitic or not in a given context. However there are several complications: internal and external sandhi differ in only a few details, so that we often cannot be sure which we are dealing with; in addition not all sequences of segments are affected by sandhi rules; furthermore internal sandhi is also found in some contexts where we are clearly not dealing with enclisis, such as contexts of close syntactic cohesion between two words.30

To take the first of these two complications, the relative pronoun yá-, for example, begins with the segment y; this segment is not affected by either internal or external sandhi and itself causes no distinct internal sandhi phenomena. It is therefore impossible to use this feature to determine whether or not yá- (or any other word beginning with y) is ever a clitic. The first segments of the demonstrative pronouns sá-/tá- and syá-/tyá-, on the other hand, are affected by internal but not external sandhi rules. Specifically an s can be retroflexed according to the so-called ‘ruki’ rule, when one of the four segments r, u, k, i directly precedes it in the same phonological word.31 In addition a t assimilates to a preceding retroflex ṣ in the same phonological word.

There are many examples in the Ṛgveda of these demonstrative pronouns undergoing internal sandhi under the influence of a preceding word, which can only be explained on the assumption that such pronouns are enclitic in these positions.32

(24)prá ṣá víbhyomarutovírastu
before well that from_birds O_Maruts bird let_be
‘Let that bird be entirely before (all other) birds, O Maruts.’33 (4.26.4a)
(25)nákiḥ ṣó astyáraṇo,jahúrtám
no-one this is enemy left for him
‘This one is no enemy, for they left him.’(2.24.7d)
(26)níṣ ṭ;áj jabhāracamasáṃvṛkṣinline imaged/bṛhaspátirviravéṇāvikinline imagetya
out that brought ladle like from_wood Bṛhaspati with_roar having_opened
‘Bṛhaspati brought that out like a ladle from wood, having opened (the mountain) with a roar.’ (10.68.8cd)
(27)utásvásyāárātyāarír ṣá /utinline imagenyásyaárātyāvinline imageko ṣáḥ
and from_own enemy rival for he and=from_other enemy wolf for he
‘(Protect us) both from the enemy in our midst, for he is a rival, and from the enemy outside, for he is a wolf!’ (9.79.3ab)

In none of these contexts can such internal sandhi be attributed to close syntactic cohesion betwen the two words concerned. However Hale (1995: 77–80) discusses one other context in which this ‘close’ sandhi is found: monosyllables ‘are regularly incorporated into the prosodic domain of the element which follows them,’ resulting in ‘close’ sandhi. Hale includes the sequences m inline image kis and nákis in this set, since they also show this close sandhi (25) and can be analysed as the negative m inline image/ followed by a separate monosyllable kis. All the passages given above therefore could be analysed on such terms, in which case they would provide no evidence for internal sandhi caused by enclisis. However in (24) is itself enclitic, which makes it somewhat unlikely that it is also prosodically proclitic (which, in one sense or another, must be what Hale has in mind); the same is true of ‘for’ in the sequence hí ṣáḥ, which appears in (27) and a number of other passages at the end of a line, the functioning as a clausal conjunction. In total internal sandhi of a demonstrative occurs after an enclitic in 17 of the 47 instances of the sequence. Even more significantly such internal sandhi occurs after several disyllabic words, and even after one trisyllabic word; nor in any of these cases can ‘close syntactic cohesion’ be invoked as an explanation.

(28)pári ṣyá suvānóakṣā/índurávyemádacyutaḥ
around that pressed flows drop in_sheep's_wool moving_with_intoxication
‘That pressed drop flows around in the sheep's wool (sieve), moving with intoxication.’ (9.98.3ab)
(29)urú ṣá saráthaṃsinline imagerathayekar/índraḥkútsāyasinline imageryasya
wide that on_same_chariot for_charioteer made Indra for_Kutsa of_the_sun
‘That Indra, sharing his chariot, made for the charioteer Kutsa a wide (path) in his winning of the sun.’ (6.20.5cd)
(30)yástesinline imagedhiṣṭhó'vasa/índrakrátuṣ ṭám inline image bhara
which of_you most_effective for_aid Indra wise that here bring
‘Bring here, O Indra, as the wise one, that (intent) which is most effective for aid.’ (5.35.1ab)
(31)agníṣ inline image víśvābhúvanāniveda
Agni those all worlds knows
‘Agni knows all those worlds.’ (3.55.10c)
(32)purinline image śáṃsenavāvṛdhuṣ ṭá índram
much with_praise strengthen they Indra
‘They strengthen Indra greatly with praise.’(10.73.2b)

Example (28) shows a disyllabic preposition preceding a form of syá-, the first segment of which undergoes internal sandhi due to the final segment of the preposition. However, prepositions are themselves of variable prosodic independence (as discussed in more detail below), and so it may be possible here to treat the sandhi as due rather to a particularity of the preposition rather than the pronoun. The same argument affects also (26), since the element preceding the pronoun is not only monosyllabic but also a preverb. But even discounting disyllabic prepositions we are still left with twelve instances of pronouns showing internal sandhi following an unambiguously prosodically independent word, as in (29)–(32).34 In all these passages the pronoun is to be analysed as following the first syntactic constituent of the clause, which in all but one instance is a single non-clitic word. In (30) krátuḥ is the first word of the correlative clause, separated from the relative clause as it is by not only the interjected vocative índra, which is not syntactically integrated into the rest of the clause, but more importantly by a metrical line break, which directly precedes the vocative.35 Example (32) is more complicated, but can be interpreted as precisely parallel. The first three words are all part of the VP, and can be taken therefore as a single constituent; it is therefore parallel to (19), in which the interrogative pronoun follows the entire VP. In this example the clitic pronoun follows the VP in second syntactic position and shows internal sandhi. The object of the verb is right-dislocated at the end as background information, since in the context of the hymn (which is to Indra) it is clear who is being strengthened.36 The internal sandhi found in all these passages is problematic for previous accounts, but finds a natural explanation under the proposal made here that the pronouns concerned are enclitic.

There are a few instances in which these non-initial demonstratives do not display internal sandhi. It is not found with syá following an u at 9.3.10a, 9.38.1a (33), 9.89.1a (34); nor with at 10.31.7a, s inline image at 5.8.5d, tásmād at 10.90.9d (35), sásmin at 1.186.4d and 10.95.11c, at 1.33.5a,8c, t inline image n at 1.164.16a, nor tábhir at 8.22.12d (36).

(33)eṣáu syá vinline imageṣārátho/'vyòvinline imagerebhirarṣati
this very this manly chariot of_sheep through_wool flows
‘This very one, this manly chariot flows through the sheep wool (sieve).’ (9.38.1ab)
(34)pró syá váhiḥpathyinline imagebhirasyān
forth=and this conveyor on_paths flowed
‘This conveyor flowed forth on his paths.’ (9.89.1a)
(35)chándāṃsijajñiretásmād/yájus tásmād ajāyata
metres were_born from_this formula from_this was_born
‘The metres were born from this (sacrifice), the (sacrificial) formula was (likewise) born from this.’ (10.90.9cd)
(36)yinline imagebhiḥkríviṃvāvṛdhús t inline image bhir inline image gatam
with_which Krivi they_strengthened with_these to come
‘With which (help) they strengthened Krivi, with that come here!’(8.22.12d)

Of course we should not necessarily expect these pronouns to show internal sandhi in every possible instance, because as argued above they have both clitic and non-clitic variants. In all these ‘exceptions’, there is no need to treat the pronoun as enclitic. In (34) the pronoun can be taken not as a syntactically separate second-position clitic, but simply as part of the constituent ‘this conveyor’, i.e. in a constituent with the following word. In (35) the pronoun is the second word in the correlative clause, but there is no need therefore to take it as clitic; in the same way that a single word or constituent can sometimes precede the non-clitic interrogative pronoun, so the same can be expected to occur with the non-clitic forms of the relative and demonstrative pronouns; there is in fact an emphatic repetition of tásmād in this and the adjacent verses which argues against a clitic interpretation. The form t inline image bhiḥ in (36) is even more clearly distinct from the non-clitic forms of the pronoun: it is first in the correlative clause, and so is clearly non-clitic. The contrast is clear between the pronoun in this passage and the second-position pronoun in (32), which happens to follow the same word as the pronoun here, or between this and the pronoun in second position in a correlative clause in (30).37 The explanation of (33) could well be parallel to (34), but here (as also at 9.3.10a) there is another explanation: internal sandhi of s is avoided when it would result in the appearance of two retroflex sibilants, , in the same phonological word. For example forms of the verb √stu ‘praise’ regularly show internal sandhi of the initial cluster in the appropriate contexts ((37), also at 1.173.5a, 4.21.4b, 5.42.11a, 6.45.16a, 8.95.6a, 8.96.6a), except where the verb already contains a retroflex sibilant ((38), also 2.20.4a, 5.36.3d, 8.3.14a,d).

(37)támu ṣṭuhi yó…
him indeed praise who
‘Praise him indeed who…’(6.18.1a)
(38)támu stuṣa índraṃyó…
him indeed I_praise Indra who
‘I praise this very Indra who…’ (6.21.2a)

Given this phonological constraint on internal sandhi, it would be possible to take the syá in (33) as clitic. The combined evidence presented here suggests that clitic variants of the relative and demonstrative pronouns existed alongside the non-clitic forms of the words in Ṛgvedic Sanskrit. But how should these be treated syntactically? Looking back to our descriptive ‘template’ of the clause-initial string, it can be seen that such pronouns directly precede the two categories of traditionally recognized clitics in the initial string: the clitic sentence particles, and the clitic personal pronouns. Therefore from a syntactic perspective the simplest analysis is to treat the clitic relative and demonstrative pronouns in precisely the same way as those clitics. That is, the ‘clitic cluster’ includes not just the clitic sentence particles and personal pronouns, but also the clitic relative and demonstrative pronouns. So in Hale's terms, for example, these forms of the relative and demonstrative pronouns will move not into SpecCP, where for him all wh words go, but rather with other clitics from the IP into inline image.

As noted it is equally possible for these pronouns to appear clause initially, in which case they can be treated as non-clitic XPs in the DF position. The lack of internal sandhi with some forms of the demonstratives makes the existence of two distinct variants clear. There is a partial parallel for this contrast between enclitic and non-enclitic demonstrative pronouns in the oblique case-forms of the near-deictic ayám ‘this’: both accented and unaccented oblique case-forms exist, the former deictics proper, e.g. genitive singular asyá ‘of this one’, the latter unemphatic enclitic anaphoric pronouns, e.g. asya ‘his.’ The obvious difference is that the latter are distinguished not only by clisis or lack thereof, but also by accent or lack thereof (not to mention the semantic differences), whereas there is no recorded accentual difference between the two variants of the relative and demonstrative pronouns. But as discussed above the sets of unaccented and enclitic words only partially overlap, so that the existence of two phonologically identical words, both accented but one enclitic, is perfectly conceivable.38

Although there can be no sandhi evidence for the clitic status of the relative pronoun, as shown above there is no significant difference between the syntactic position of the relative pronoun and the demonstrative pronouns. Moreover both Watkins (1963: 29–) and Hettrich (1988: 758–93) consider the relative pronoun yá- to be optionally ‘enclitic’ in the Ṛgveda, on the basis of its common ‘second position’ syntax.39 Neither Watkins nor Hettrich claims enclisis for the Ṛgvedic relative pronoun for its own sake: Watkins uses the frequent second position of the relative pronoun to support his analysis of the Celtic reflex of an enclitic relative *-inline image o, while Hettrich's aim is likewise primarily comparative. Neither provide any synchronic syntactic analysis of the phenomenon in Sanskrit, and both focus primarily on the common appearance of the relative in ‘second position’ in the broadest sense, which as discussed above does not necessarily correlate with enclisis. The best evidence provided by Hettrich (1988: 760–2) for a truly enclitic positioning of the relative pronoun is a few passages in which the pronoun apparently interrupts either a constituent (examples (39) and (40)) or two co-ordinated constituents (examples (41) and (42)).

(39)uraú antárikṣemádanti/divórocanésántidevinline image
in_wide or who in_sky rejoice of_heaven or who in_bright are gods
‘(gods) who rejoice in the wide sky, or the gods who are in the bright realm of heaven.’ (3.6.8ab)
(40)váruṇo yásya darśató/mitróvánategíraḥ
Varuṇa whose beautiful Mitra or wins songs
‘whose songs Varuṇa, the beautiful, or Mitra, wins.’(5.65.1cd)
(41)merājanyújyosákhāvā /svápnebhayámbhīrávemáhyam
who my king worship-worthy or friend or in_dream fear to_timid to_me
inline imageha/ stenó dípsatinovinline imageko/ tváṃtásmādvaruṇapāhy
speaks robber or who deceives us wolf or you from_that Varuṇa protect
asminline imagen
‘Whichever person, O king, worthy of my worship, or a friend, speaks fear to timid me in a dream, or whichever robber deceives us, or wolf: protect us from him, O Varuṇa.’ (2.28.10)(2.28.10)
(42)índrā agninline image sáhurīsaparyinline imaget
Indra who Agni mighty honours
‘He who honours mighty Indra and Agni.’(6.60.1b)

Example (39) is the clearest attested example of the relative pronoun interrupting the first XP of the clause. In this and (40) it could, however, be argued that the two parts of the NP separated by the relative pronoun are discontinuous, i.e. in (39) we should translate ‘those who in the wide one, (i.e.) in the sky, rejoice…’, parallel to the translation given for (40); but clitic positioning parallel to, for example, tvā in (1), is no less reasonable an analysis. Example (40) is also listed by Hettrich as an instance of the relative pronoun interrupting conjoined constituents, as are examples (41) and (42). Hettrich's main argument here is that the position before the relative pronoun is supposed to be ‘marked’ (i.e. for topic or focus), but that in these instances it makes no sense to interpret the part of the conjunct preceding the relative as marked but the part following as unmarked. Example (41) is actually not a good example in this respect: Hettrich quotes only the third line, from stenó to v inline image ko vā, translating ‘welcher Dieb oder Wolf uns schädigen will’ (‘which robber or wolf desires to harm us’); when the whole line is considered, however, it can be seen that the following stenó is not conjoining that noun with v inline image ko, but rather conjoining the whole clause with the preceding parallel relative clause. v inline image ko vā is then simply added as an afterthought, and there is no difficulty in taking stenó as the marked, topical element in the phrase. The other examples Hettrich discusses are better, however.40 Example (42) in particular shows the relative pronouns interrupting a devatā-dvandva, a loose compound type found in early Vedic with divine names, in which the two names involved are both inflected in the dual to signal their close cohesion, despite the fact that there is only one of each. The relative is therefore not just interrupting a constituent, but something slightly closer to a word than a phrase. On the other hand, however, índrā alone, being dual, is sufficient to indicate both Indra and Agni in the context of the hymn, such that the following dual agn inline image could, perhaps, be taken as a semantically superfluous and syntactically parenthetical addition. Nevertheless this, and the other passages quoted by Hettrich, are highly suggestive of clitic positioning for the relative pronoun.

Hettrich's (1988: 758ff.) main goal is to compare the apparent clitic positioning of the relative pronoun with comparable positioning of cognate forms in other Indo-European languages. Relative pronouns can appear to a limited extent in syntactic second position in Avestan, Old Persian, Latin and Ancient Greek, although as discussed above this is only suggestive, not probative, of clisis. In Balto-Slavic the definite adjective declension developed from a postposed relative construction (Vaillant 1942; Bratishenko 2003: 93) involving, at some stage at least, enclisis of the pronoun (followed by grammaticalization as an affix). An enclitic relative marker, inline image , almost certainly related to the Sanskrit relative yá-, is clearly attested in Celtic (Watkins 1963: 28–30; Sims-Williams 1984; Ziegler 1992; McCone 2006: esp. 247–76); the form of the relative pronoun in Old Persian, hạya-/tạya-, may derive from an enclitic form of the relative pronoun fused onto the inherited demonstrative *sá-/tá- (Hettrich 1988: 765, fn 342).41 Hettrich brings all this data together in an attempt to prove that in Proto-Indo-European there were two distinct relative pronouns: a restrictive relative inline image-, which was not enclitic, and an appositive relative pronoun inline image, which was enclitic.

My aim is not specifically to make claims about Proto-Indo-European. Although the evidence for the existence of an enclitic form of the relative pronoun inline image, whether alongside a non-enclitic form or not, and regardless of its original function, seems relatively robust, nevertheless it is possible that in all these languages the enclitic form of the relative is a later development. For our purposes we can simply note that the existence of an enclitic variant of the relative pronoun in the Ṛgveda is a reasonable proposition not only on language-internal grounds, but also on comparative grounds (whether those comparative grounds support inheritance or parallel development).

5. Preverbs

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References

Further support for the clitic status of some relative and demonstrative pronouns comes from a reconsideration of preverbs in the clause-initial string. Preverbs are directional adverbs which can occur either near the front of the clause, or else within the VP, usually directly preceding the verb.42 Like their cognates in other Indo-European languages, preverbs gradually underwent a process of integration with the verb forms they modified, with the result that ultimately, in Classical Sanskrit, they were obligatorily prefixed to the verb; in the Ṛgveda, however, they are still relatively free in relation to the verb.

Preverbs in main clauses are most commonly initial, ignoring any preceding ‘extra-clausal’ conjunctions; this is the case with at least 60 per cent of main-clause preverbs. With roughly two-thirds of these clause-initial preverbs, the verb appears later in the clause, separated from the preverb by at least one constituent (and often final in the clause), as in (44). In the other third the verb directly follows (45). Parallel to the discussion of clitic pronouns above, it would be possible in these instances either to analyse the preverb and verb as a single constituent (inline image), or to analyse the preverb separately from the verb, a discontinuous part of the same constituent which happens to appear adjacent to it. Again for the sake of argument the former analysis is preferred. Around one-third of main-clause preverbs are found within, i.e. not initial in, the main clause, and within the VP. Most commonly in these cases the preverb directly precedes the verb (46), but it can appear separated from the verb by, for example, the object, and can even follow the verb (47). Finally in a minority of cases preverbs appear near the start of the clause, syntactically separated from their verb by at least one constituent, but preceded by a separate constituent which, following the discussion above, can be assumed to be topicalized or focused and to appear in the DF position (as in (4)).

The same patterns can appear also in subordinate clauses, but the distribution is slightly different. Preverbs can appear initially in a subordinate clause, with the verb syntactically separated from it by at least one constituent (48); in a minority of cases the only thing separating verb from preverb in a subordinate clause is the subordinating relative pronoun which, if we take it as enclitic in these instances, can be ignored (43). More frequently, the preverb appears later in the subordinate clause within the VP, again most commonly, but not necessarily, directly preceding the verb (49).43

(43) prá śúmbhantejánayosáptayo/yinline imagemanrudrásyasūnávaḥ
forth who adorn women like horses course of_Rudra sons
‘Those who adorn themselves like women, horses on the racecourse, the well-accomplishing sons of Rudra …’ (1.85.1ab)
(44) úd uṣyádeváḥsavitinline image yayāma/hiraṇyáyīmamátiṃ…
up indeed this god Savitinline image has_held golden form
‘This god Savitṛ has held up his golden form …’ (7.38.1ab)
(45) uc chrayasvamahatésaúbhagāya
up rest for_great for_blessing
‘Rest upright for great blessing!’ (3.8.2d)
(46)ninline image mebráhmāṇyagna úc chaśādhi
now me sacred_prayers Agni up instruct
‘Now instruct me in the sacred prayers, Agni.’ (7.1.20a)
(47)yáyāsvépinline imagetresiñcása út
with_which own cup pour out
‘… [the ladle] with which you pour out into your own cup.’(10.105.10c)
(48) úd yátsáhaḥsáhasa inline imagejaniṣṭa
up when power from_power was_born
‘when (your) power was born from power.’ (5.31.3a)
(49)sahásraśṛṅgovṛṣabhó/yáḥsamudrinline imaged ud inline imagecarat
thousand_horned bull who from_ocean up.out.came
‘The thousand-horned bull who came up out of the ocean.’(7.55.7ab)

Tabulating the precise frequency of preverbs in their possible positions is somewhat problematic, due to the usual ambiguities of interpretation. The majority of preverbs can also function as prepositions, which adds a further layer of complication to their analysis; for such preverbs, for example, it may be unclear in any one instance whether they are directly modifying a transitive verb or heading a PP which itself functions as complement to the verb. A subset of preverbs, however, are never used prepositionally, and so their syntactic analysis is somewhat simpler; they are ápa ‘away’, áva ‘down’, úd ‘up’, ‘down’, níṣ ‘out’, párā ‘far’, prá ‘forth’, ‘apart.’ Even with these, it is sometimes unclear whether they should be taken with the main verb or with a non-finite verb form such as a participle; moreover, finite verbs are sometimes ellipsed, and the repetition of preverbs without finite verbs can function as a kind of asyndeton.44 Table 4 summarizes my count of the relative frequencies of position of the preverb prá, the most common non-prepositional preverb in the Ṛgveda ignoring passages in which the verb is ellipsed (of which there are around 40), or in which the preverb modifies not the finite verb but a subordinate verb (of which there are c. 240).45

Table 4. Position in clause of prá
Main clausesSubordinate clauses
# prá…V (…)46044.8%5536%
# prá (-) V …20820.3%2113.7%
# XP prá … V (…)181.8%31.9%
# inline image prá (…)]34033.1%7448.4%
Totals1026 153 

As can be seen in the table, the preverb prá appears in clause-initial position in over 65 per cent of its main-clause occurrences. In about one-third of these the main verb directly follows, so what we may be dealing with is in fact a clause-initial VP rather than a clause-initial preverb. Even so we still have to account for 44.8 per cent of the main-clause occurrences of prá in initial position. In subordinate clauses the figures differ slightly: the preverb appears clause initially in just over 50 per cent of instances, of which again about one-third can, at least in principle, be analysed as examples of clause-initial VPs (if, by treating it as enclitic, we can ignore the appearance of the relative pronoun inside the preverb-verb complex). As stated above, prá is the most common preverb in the Ṛgveda (of those which cannot also be used as prepositions). Other preverbs differ slightly, but not significantly, in their distribution. For example the preverb úd, as seen in (44)–(49), appears clause initially in 204 (71.1 per cent) of 287 main-clause occurrences, of which roughly one-third (71) are directly followed by the main verb; the remaining 83 (28.9 per cent) appear within the VP later in the clause. This is very similar to the distribution of prá. The figures for úd diverge more noticeably in subordinate clauses: only 4 (12.9 per cent) of 31 occurrences of úd in subordinate clauses are initial, while the remaining 27 (87.1 per cent) are clause and VP internal. The fact that relative clauses are less frequent than main clauses means that the numbers involved are distinctly smaller, and the percentages correspondingly less reliable. The two preverbs prá and úd in fact show contrasting extremes in terms of placement in subordinate clauses; on average, in relative clauses where the verb is modified by a preverb, that preverb precedes the relative in around 25 per cent of cases. The differences between preverbs in this respect are largely due to lexical differences between the particular verbal roots with which they appear.

When preverbs appear within the VP, whether clause internally or clause initially, there is no problem, syntactically speaking, with their analysis; their base position is directly modifying the verb, forming a inline image with it. What we must account for is the 40–50 per cent of occurrences in which this is not the case, where in contrast the preverb appears at or very near the start of the clause.

As we have discussed above, there are two main reasons why words appear at or near the start of the Ṛgvedic clause: either they have a specific discourse function such as topicalization or focus and hence appear in (one of) the initial DF position(s), or they are in the string of clitics appearing in second position. Recall that in the descriptive ‘template’ given above, preverbs appear after the DF position and directly before the relative and demonstrative pronouns which I have now argued to be part of the clitic sequence. What, then, does this descriptive position mean in syntactic terms?

Keydana (2011) does not specifically discuss the positioning of preverbs. Hock (1982; 1989; 1992; 1996; 1997a) treats preverbs as part of his category inline image, i.e. along with relative and demonstrative pronouns, assuming doubling when both occur; in his terms this is justified prosodically (since all words in both categories are accented), but it cannot easily be correlated with a syntactic account, since preverbs (as adverbs) and pronouns are from different syntactic categories.

For Krisch (1997, 2002) preverbs are no different from any other element appearing at the start of the clause before the clitics; transposing this into our terms, preverbs must then be taken in the initial DF position. Similarly, Hale (1987b) accounts for the position of preverbs by moving them into his inline image (topicalization/focus) position, which is equivalent to the DF position in our descriptive template above. This assumption of topicalization or focus for clause-initial (or near clause-initial) preverbs has relatively wide currency. For example Renou (1933: 54) says that:

Le transfert en tête de phrase du préverbe séparé n'a pas seulement pour effet de mettre en évidence un mot important; il sert aussi à porter en avant une part de la notion verbale tout en maintenant le verbe à sa place normale, c'est-à-dire à la fin de la proposition. [The movement of a separable preverb to the start of a phrase is not only for highlighting a significant word; it serves also to bring to the front part of the verbal meaning while allowing the verb to keep its regular position, that is at the end of the clause].

That is, a clause-initial preverb can either itself be marked for discourse function (‘un mot important’), or it can serve to mark the verb itself in this way, an alternative (available only, of course, when a preverb is present) to positioning the verb at or near the clause start. This view goes back at least as far as Delbrück (1888: 45) and has been accepted as valid not only for Vedic Sanskrit, but also for the parent language Proto-Indo-European (see e.g. McCone 1997).

It is undoubtedly true that in some instances we can treat preverbs near the start of their clause, separated from their verb, as marked for topicalization/focus or as marking the whole verb in this way. However, it is questionable whether this can account for all instances of clause-initial preverbs in the Ṛgveda. The inherent ambiguity and interpretative difficulties of Ṛgvedic Sanskrit means that it is always possible to assume either focus or topicalization of either the preverbal element or the main verb, whenever a preverb appears near the start of a clause. Although in some instances such discourse functions are clear, e.g., when preverbs serve to focus the verb in imperatival clauses (example (50), similarly (7)), in others it is distinctly unclear (example (51), similarly (4)).

(50) inline image norayíṃvahatamótávīrinline imagen
towards us wealth bring towards=and heroes
Bring here to us riches and (bring) here heroes.’ (5.76.5c)
(51) gavyatinline image mánasāsedurarkaíḥ
down with_cow_desiring with_mind sat with_praises
‘With cow-desiring mind and with praises they sat down.’ (3.31.9a)

The preverb abhí, which appears in (4), is semantically weak, often serving merely to mark an accusative of goal or a direct object. In many passages, including (4), it could be omitted without changing the meaning, since bare verbs of motion can take bare accusatives of goal. Thus we cannot suppose that the preverb itself is focused in (4); nor is there any reason to suppose that the verb of motion itself is specifically focused or topicalized. The same is true of (51): the fact that the priests sat down can hardly be particularly noteworthy, nor merely the fact that they sat; in fact the most important part of the clause is the instrumental phrase gavyat inline image mánasā ‘with cow-desiring mind’, i.e. their manner in sitting.

Although it could be argued that a more restrictive approach to Ṛgvedic syntax, here the assumption either that clause-initial preverbs are always specifically DF-marked, or that they are specifically not DF-marked, is in principle preferable to admitting both possibilities, such an approach can only be maintained if it is consistent with the evidence of the text and enforces coherent and reasonable interpretations. Interpretation is of course difficult, insofar as we have no native speakers to confirm our judgements, and insofar as the Ṛgveda is sufficiently ambiguous to be interpreted in multiple ways (particularly as regards discourse structure). But the fact is that there are many passages like (4) and (51) where it would be forcing the text to read focus or topicalization on the verb or preverb, yet also many passages like (50) where the discourse function of the preverb appears incontestable; we must conclude, then, that the more restrictive approach is too restrictive in this case, and that clause-initial preverbs are not all necessarily DF-marked.

If DF marking cannot account for all clause-initial preverbs, we must assume that the clause-initial position of preverbs is simply an unmarked syntactic possibility. This alternative viewpoint is, in fact, widely held; for example Oldenberg (1907) and Hettrich (1988: 759–60) assume that the initial position of preverbs is the regular, unmarked position. Somewhat parallel, Hale (1996) moved away from his former analysis of clause-initial preverbs as having moved to the head of his Topic Phrase, arguing rather that they are adjoined to CP, below his topicalization position.

Moreover, there are a few examples of relative pronouns preceding preverbs in the Ṛgveda, as in (52) and (53).46 These are unproblematic under the schema suggested here, since we can analyse the preverb as clitic and the relative pronoun as DF-marked.

(52)ayáṃvoyajñáṛbhavo'kāri/yám inline image manuṣvátpradívodadhidhvé
this your sacrifice Ṛbhus was_made which prv like_Manu of_old you_founded
‘This your sacrifice, O Ṛbhus, has been made, (the sacrifice) which you founded of old, like Manu.’ (4.34.3ab)
(53)rathirinline imagesohárayo/…/yébhir dásyummánuṣonighóṣayaḥ
chariot-horses bays with_which prv Dasyu to_Manu made_obedient
‘Your bay chariot-horses …with which you made the Dasyu well obedient to Manu.’ (8.50.8)

To summarize, then, preverbs at or near the start of the clause can sometimes be analysed in a DF position, marked themselves, or marking the verb, for topic or focus. In many passages, however, there is no justification for such a discourse structure. Preverbs never precede interrogative pronouns, but usually, though not always, precede relative pronouns. As discussed above, interrogative pronouns and sometimes relative pronouns appear in what I have analysed as one of the DF positions at the start of the clause, which in Hale's terms would be SpecCP. Enclitic relatives, however, appear in the clitic cluster which immediately follows this position (inline image for Hale). Given the evidence presented above, it seems that preverbs appear somewhere between these two positions.

Hale's (1996) solution was to assume two CP nodes, one directly dominating the other, so that a preverb can appear in the specifier position of the higher CP and a wh word can appear in the specifier of the lower CP simultaneously. Yet there would seem to be little justification for such a double CP construction beyond the desire to find some sort of phrase-structure position for preverbs which precedes relative pronouns but follows the DF position (Hale's inline image). There is, however, another possibility.

Besides their initial-string positioning, preverbs can also appear within the VP; when they do, they most commonly directly precede the verb. This is parallel to enclitic pronouns and some enclitic particles, which can appear either within the initial string or enclitic within the VP. Preverbs are not enclitic, but they can be proclitic. Besides their appearance in the clause-initial string, their second most frequent position is directly preceding the verb. In subordinate clauses, where the verb is accented, any directly preceding preverb usually appears unaccented and proclitic on the verb.47 In main clauses this does not happen; rather, the preverb retains its accent and appears to be a separate word, but this could be an epiphenomenon of the fact that main-clause verbs are themselves unaccented and hence cannot ‘support’ a proclitic.

It is, then, at least possible that preverbs appearing in the clause-initial string can likewise be clitic. Just like relative and demonstrative pronouns, there are many cases where they cannot be so analysed, but must be taken as ‘full words.’ But in those cases where preverbs cannot easily be accounted for from a syntactic point of view, except by some particular stipulation such as a double CP, clisis provides a simple solution. There is no need to search for a special position for preverbs somewhere between the DF positions (or SpecCP) and the clitic cluster (or inline image), because such preverbs are, in fact, in the clitic cluster.

There is a difference between the clitics we have seen so far and the preverbs. As stated, there is no evidence that preverbs could be enclitic, only that they could be proclitic. There is, however, no difficulty in assuming that whatever node hosts the clitic sequence can host both proclitics and enclitics.48 Enclitics are prohibited from appearing in absolute clause-initial position since, whether syntactically, phonologically, or both, they require something to their left on which to ‘lean.’ This means that if nothing appears in any of the optional (DF) positions preceding the clitic cluster, the cluster cannot surface in first position, assuming it contains only enclitics, but actually appears following the first non-clitic word that is, syntactically, to its right.49 Proclitics, on the other hand, are not so restricted. The internal ordering of the clitic sequence is such that any proclitic preverb will always precede any and all enclitics. Therefore it is possible that the clitic cluster can surface as the first element in the clause if nothing appears to its left in the syntax and if its first element is a preverb.

One final point can be made concerning the prosodic features of the clitic cluster. When it contains only enclitics, the clitic cluster requires a word to its left on which it can ‘lean’, and the entire cluster is integrated into the prosodic word on which it leans.50 But what about a sequence of proclitic plus enclitic? In Ancient Greek some sequences of proclitic plus enclitic form single phonological words, e.g. ei ‘if’ (proclitic) plus ge ‘at least’ (enclitic) results in [inline image eí ge]inline image; some such sequences even become lexicalized as independent, non-clitic, words, e.g. eíte ‘whether’ from ei ‘if’ (proclitic) + te ‘and’ (enclitic).51 There is nothing to prevent us assuming that the same is true here, i.e. that a sequence of proclitic preverb plus one or more enclitics forms a single, separate phonological word.

Analysing preverbs within the Ṛgvedic clitic cluster confirms the argument made in the previous section regarding the clitic status of relative and demonstrative pronouns, which can now be seen, in examples such as (4), to surface between other clitic cluster elements. In this and the preceding section I have shown that certain categories of word which regularly appear near the start of the Ṛgvedic clause have two distinct positions, one clitic, the other non-clitic. In the following section I make clear how this affects our overall syntactic analysis of the start of the Ṛgvedic clause.

6. Ṛgvedic syntax at the left edge

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References

6.1. The clitic cluster

As I have argued above, clitic positioning is a fundamentally syntactic phenomenon. That is, we should be able to describe the positioning of clitics by reference to syntactic categories and structure. Undeniably, prosodic constraints also have a part to play, but their role is just that: of constraining syntactic structures rather than fundamentally driving word order.

Syntactic accounts of clitic sequences vary considerably according to the theory in which they are based, but many accounts treating clitic sequences or ‘clusters’ in different languages recognize such sequences as single syntactic groupings or constituents within the clause (see e.g. Halpern 1995: 191–222 with references). The internal structure of such groupings is determined not hierarchically, but linearly (Perlmutter 1970). Looking back to our descriptive ‘template’, we can see that the clitic cluster accounts for a large part of the clause-initial string. Every slot from the preverb position to the clitic personal-pronoun position is a part of the clitic cluster, and the internal order of that cluster is determined linearly, i.e. any preverb precedes any relative or demonstrative pronoun, which precedes any sentence particle, which precedes any personal pronoun.52 In terms of a syntactic analysis, then, although all these slots are themselves separate syntactic elements, together they constitute one superordinate syntactic constituent for which we must account.

As discussed above, there is one DF position preceding the clitic cluster in which interrogative pronouns, some relative and demonstrative pronouns, and some preverbs can appear. However, it is also possible for a constituent to precede an interrogative pronoun, meaning there must be a second DF position to the left of the first. Constituents can also precede relative pronouns, demonstrative pronouns, and preverbs, but this is less probative since those elements can also be clitics. In fact this is all relatively uncontroversial; for example, it can easily be transposed into the framework of Hale (1987a, b, 1995; 1996. The clitic cluster appears in inline image; the first DF position is SpecCP which, as in English, hosts interrogative and some relative pronouns; the DF position to the left is equivalent to Hale's inline image position. What is significantly different from Hale's account is that preverbs and relative/demonstrative pronouns can also appear in the clitic cluster, i.e. in inline image.

In a different syntactic framework the same positions can be analysed in slightly different ways. In the non-transformational framework of Lexical-Functional Grammar (LFG) a syntactic category CCL (clausally-scoped clitic cluster), dominating one or more terminal CL(itic) nodes, has been utilized to provide a syntactic grouping for the clitic cluster (Bögel et al. 2010; Lowe 2011). Moreover LFG can utilize the ‘headless’ exocentric category S, which can directly dominate any number of XP nodes, to account for languages in which the order of words in a clause is non-configurational. In an approach such as that of Lowe (2011), where Ṛgvedic word order is assumed to be entirely non-configurational, a (partial) phrase structure rule such as the following can account for the clause-initial string.53

inline image inline image [UPWARDS ARROW]=[DOWNWARDS ARROW]
((inline image df)=topic)((inline image df)=focus)

The annotations under the first two XPs indicate the discourse functions of the two DF positions.54 I have assumed these respectively to be topic and focus simply for clarity, the precise values being notoriously hard to determine and not vitally important here.55

The point is that once we recognize the potential (syntactic) clitic status of relative and demonstrative pronouns and preverbs, it is relatively straightforward to provide a coherent syntactic analysis of the Ṛgvedic clause-initial string in theoretical frameworks of such diversity as GB (Hale) or LFG (Lowe 2011). For exemplification I give the syntactic analyses of (4), repeated as (55), in Hale's transformational terms (56) and in non-transformational LFG terms (57).56

(55)divyinline image inline imagepoabhíyád enam inline imageyan
divine waters to when him came
‘when the divine waters came upon him.’ (7.103.2a)
(56)GB syntactic structure for RV 7.103.2a (4) (= (55))
(57)LFG syntactic structure for RV 7.103.2a (4) (= (55))

The clause consists of an initial, topicalized NP, a sequence of three clitics, preverb, subordinating conjunction and personal pronoun, and the verb. Example (56) differs from the way in which Hale himself would analyse the line in the appearance of the preverb and subordinating conjunction under inline image as clitics; in Hale's analysis yád would appear in SpecCP and a further CP dominating the first would be required, the specifier of which would host abhí.

6.2. Prosodic constraints

There is one further point still to be dealt with. As discussed above, enclitics cannot stand at the start of a clause. In relation to the clitic cluster, this is not a problem when either one of the DF positions to the left of the cluster is filled, or when a proclitic preverb begins the cluster, i.e. when non-enclitic material precedes the enclitics. Both occur in (4), only the former in (58), the non-transformational syntactic analysis for which is given in (59).57 The clitic cluster, generated in its expected syntactic position (whether C, or under the CCL node), causes no problem in the prosodic output. But there is no obligation for either DF position to be filled, nor for a preverb to appear in the clitic cluster. So what happens when nothing precedes a clitic cluster containing only enclitics?

(58)mahékṣatrinline imageyaśávase jajñé
to_great to_dominion to_might for was_born
‘For he was born to great dominion (and) might.’(7.28.3c)
(59)LFG syntactic structure for RV 7.28.3c (58)

In such cases, what we find in the output is that the clitic cluster appears within the first constituent of the clause, as in (60). Or to put it another way, when the clitic cluster appears inside the first constituent of the clause, we cannot analyse that constituent as occupying either of the DF positions to the left of the clitic cluster, since in that case the cluster would follow the entire constituent; the only place we can analyse the constituent is in the first XP slot to the right of the clitic cluster (whether this is the first constituent in the IP, as in Hale's model, or the immediate rightward sister of the CCL, in Lowe's 2011 model). But this means that in the syntactic analysis, in the tree, the clitic cluster is wholly to the left of the constituent within which it appears in the output. A partial syntactic tree (assuming a flat structure, but easily transposed into a hierarchical structure as in (56)) for (60) is given as (61).

(60)víśvā no vithurinline image pibdaninline image vaso/'mitrinline imagensuṣáhānkṛdhi
all indeed for_us unstable firm good enemies easy_to_conquer make
‘Indeed, make everything which is unstable firm for us, O good one, (and make) our enemies easy to conquer.’ (6.46.6cd)
(61)Partial LFG syntactic structure for RV 6.46.6cd (60)

The same problem also affects enclitic conjunctions. This is seen in (1), repeated as (62), where the clitic tvā ‘you’ is the only member of the clitic cluster, while the clitic ca ‘and’ functions as a clausal conjunction.

(62)dyaúś ca tvā pṛthivinline image yajñíyāsohótāraṃsādayantedámāya
Heaven and you Earth worship_worthy down priest set for_house
‘And Heaven and Earth, (and) the worship-worthy (gods), establish you as priest for the house.’ (3.6.3ab)

Non-clitic clausal conjunctions always directly precede the clause they conjoin. They are sisters to the mother node of the clause which they conjoin; this can be represented using a specific phrasal node dominating the highest node of the clause, as in Hale's ConjP (Conjunction Phrase) and DisjP (Disjunction Phrase), the latter seen in (13); an alternative is to assume a flat structure between conjuncts, as in (63).


Either way, enclitic clausal conjunctions pattern in parallel to non-clitic clausal conjunctions, with the significant exception that the enclitic forms always follow the first non-enclitic word of the clause they conjoin (i.e. the clause they head or directly precede, respectively according to the analyses of conjunction presented above). This is the case even when, as in example (1) (= (62)), this involves appearing inside another syntactic constituent.

It is here that the prosodic constraints on clitic positioning come into play. Firstly, it is all but impossible to account in regular syntactic (X-bar theoretic) terms for the appearance of a word (clitic or not) inside a syntactic constituent of which it is not a part. Secondly, the syntactic evidence itself (i.e. the evidence for the specific clitic-cluster node, and the evidence of non-clitic clausal conjunctions) suggests that the clitics which appear inside constituents should or could rather be analysed directly to the left of the word which they follow in the output. It naturally follows, therefore, that in this situation there is a difference between the position of elements in the syntactic tree and the final (prosodic) output position.

As discussed above, Hale achieves this by recourse to the widely-used ‘prosodic inversion’ of Halpern (1995). Under this theory, enclitics which appear at the left edge of a clause in the syntactic component can be moved to the right of the nearest non-enclitic word at PF. In relation to the non-transformational LFG account also considered in this section, the same principle of prosodic inversion has been utilized to account for the same phenomenon, in one formalization by Bögel et al. (2010), in another by Lowe (2011). Lowe (forthcoming) presents an alternative account, which enables the apparent ‘movement’ to be accounted for in the syntactic component, despite its clear prosodic origin, making use of the separate syntactic level of the s-string (a level of syntactic representation separate from the syntactic tree, which encodes only linear order), as originally suggested by Dalrymple & Mycock (2011).

The details of the various possible formalisms are beyond the scope of this paper; all we need to recognize is that some mechanism can be used to account for the prosodic constraints on enclitics, which prevents them from surfacing at the left edge of a clause, even if this is where they appear in the hierarchical syntactic representation (the syntactic tree), and which causes them to appear following the first non-enclitic word.58

This mechanism applies both to enclitic conjunctions, which, since they appear not in a special clitic node but in the expected syntactic position for their word class, are ‘simple’ clitics in Zwicky's (1977) terminology, and to the various enclitics which appear in the clitic cluster and which are therefore ‘special’ clitics in Zwicky's terms. The transformational and non-transformational trees for (1) (= (62)) are given as (64) and (65) respectively.

(64)GB syntactic structure for RV 3.6.3ab (1) (= (62))
(65)LFG syntactic structure for RV 3.6.3ab (1) (= 62)

Although in these syntactic structures the clitics ca ‘and’ and tvā ‘you’ appear at the left edge of the clause, the phonological constraints on their positioning will result in their surfacing to the right of the first non-clitic word, dyaús. The respective order of the two clitics in the output is determined by their ‘underlying’ order; in other words the underlying order of elements in the clause is retained as far as possible, and since ca precedes tvā in the underlying syntactic analysis, it precedes also in the output, even though both words are repositioned relative to dyaús.

7. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References

In this paper I have reconsidered one of the most problematic areas of Ṛgvedic syntax, the clause-initial string. I have presented evidence to support the existence of enclitic relative and demonstrative pronouns, phonologically identical to the well-known non-clitic variants but syntactically distinct, even though it is not always easy or possible to distinguish them purely in terms of surface word order. That yá-, at least, could be analysed as reflecting an inherited enclitic syntactic positioning was suggested by Watkins (1963) and assumed by Hettrich (1988), but investigated thoroughly from a syntactic perspective by neither. I have argued that the relative pronoun yá- patterns more closely with the demonstrative pronouns than with the interrogative pronoun ká-, with which it is usually paired in syntactic analyses of the Ṛgvedic clause-initial string.59 Phonological evidence in the form of internal sandhi phenomena, combined with syntactic evidence, strongly suggests that both the relative and demonstrative pronouns could be enclitic, appearing in the ‘second position’ clitic cluster.

I have also shown that preverbs near the clause start should in many instances be analysed as clitics, positioned likewise in the clitic cluster on the basis of their morphosyntactic clitic status, rather than always being analysed as focused or topicalized. I have therefore shown that the class of ‘Wackernagel’ second-position clitics in the Ṛgveda is somewhat larger than previously recognized, including not only the unaccented pronouns and accented or unaccented sentence particles, but also accented demonstrative/relative pronouns and preverbs.

Moreover this means that the clitic cluster is more regularly ‘second position’ than could otherwise be assumed: for example enam in (4) appears to follow the fourth phonological word and third syntactic constituent under traditional analyses, whereas I have shown that it is part of a cluster of clitics which is here generated in the second syntactic position. The appearance of clitic clusters within constituents, following the first phonological word of the clause, is actually due to the generation of the clitic cluster in first syntactic position combined with prosodic constraints on the position of enclitics. On the other hand, the clitic cluster can now be analysed as sometimes surfacing in the first syntactic and prosodic position in the clause, in cases where the clitic cluster begins with a proclitic preverb.

A further point that emerges from this paper is that, at least when dealing with corpus-based languages such as Ṛgvedic Sanskrit, we should beware of making assumptions as to whether a particular word is or is not a clitic. I have defined clisis in primarily syntactic terms, and although it is clear that prosodic features (or rather deficiency of certain prosodic features) are usually present alongside the syntactic properties of clitics, such prosodic features are not obligatory. So for Ṛgvedic Sanskrit, lack of accent is commonly taken to be a defining feature of clitics, but we have seen that, in addition to well-known exceptions such as ‘for’ and ‘indeed’, it does not apply to the clitic relative and demonstrative pronouns, nor to the proclitic preverbs. The inability to constitute an independent prosodic word may be common to all Ṛgvedic clitics, but unfortunately evidence appears only in the case of certain clitics. Ultimately in the lack of clear prosodic evidence one way or another, any definition of clisis must be founded on primarily syntactic grounds.

  1. 1

    I am extremely grateful to Andreas Willi, Elizabeth Tucker, Mary Dalrymple, and Louise Mycock for their invaluable comments on and criticisms of earlier versions of this work, and also to three anonymous reviewers, whose detailed comments and criticisms benefitted this paper enormously; also the audiences at LFG11, the SE-LFG meeting at SOAS and the Oxford Syntax Working Group who heard and commented on a related paper (Lowe 2011). No responsibility is borne by anyone but myself, of course, for the opinions and errors that lie herein. I gratefully acknowledge the financial support of the AHRC, who supported my doctoral study during which this work was undertaken, the Leverhulme Trust, the Faculty of Linguistics, Philology and Phonetics at the University of Oxford, and the Lorne Thyssen Research Fund for Ancient World Topics at Wolfson College.

  2. 2

    The conjunction ca ‘and’ is another clitic, and can be ignored in this context; such clausal conjunctions will be discussed in more detail below. The word following pṛthiv inline image, yajñíyāsaḥ ‘worship-worthy ones’, is likewise asyndetically conjoined with ‘heaven and earth.’

  3. 3

    A degree of subjectivity is unavoidable in analysing so many Ṛgvedic passages. I have tried to overcome this difficulty as far as possible by explicitly noting formal ambiguities in Table 1 and by opting, when in doubt, for the otherwise most common syntactic possibility.

  4. 4

    In (9) the vocative indra is interjected inside the first constituent, but does not play a part in the syntax of the main clause.

  5. 5

    Again the vocative indra in (11) plays no part in the syntax of the clause, meaning that tvā follows the second rather than, as might appear, the third syntactic constituent of the clause.

  6. 6

    Under Hale's earlier approach, clitics are placed in second position before (sometimes after) the movement of a topicalized element to the front of the clause; this is intended to account for clitics appearing later than second position. Keydana (2011) has recently demonstrated that this cannot account for all the facts, and since Hale himself altered his approach, his earlier view is not considered further.

  7. 7

    The ‘NEXUS’ position came originally from Klein (1991: 141–2 fn .6)

  8. 8

    The purely phonological approach to clitic positioning must be seen as part of a wider trend within Minimalism to attribute all linear ordering to PF, see e.g. Chomsky (1995: 334–40, Erteschik-Shir (2005a, b), Åfarli (2010), Agbayani & Golston (2010), Erteschik-Shir & Rochman (2010).

  9. 9

    Greek súntaxis, ‘putting together in order, arranging… grammatical putting together of words’ (Liddell et al. 1940: 1724).

  10. 10

    For example, there are no instances (known to me) of clitic positioning in purely phonological terms, such as appearing without exception following the first syllable of a clause; syntactic units are always relevant.

  11. 11

    Other authors treat the interrogative pronoun ká- in the same position(s) as the relative and demonstrative pronouns, but see section 4.

  12. 12

    In general on the position of the relative pronoun in the Ṛgveda, see Hettrich (1988: 546–8).

  13. 13

    As in other tables, other clitics and sentential conjunctions have been ignored.

  14. 14

    It is notable that a disproportionate number of the non-initial interrogative pronouns occur in the later books 8 and 10. While only containing around one-third of the total attestations of káḥ in the Ṛgveda, of the fifteen examples of non-initial káḥ books 8 and 10 attest ten, i.e. two-thirds (non-initial káḥ appears at 1.80.15b, 1.164.18c, 4.23.2b, 4.25.2d, 5.53.2b, 8.3.14b, 8.7.20c, 8.45.37c, 8.64.7c, 8c, 9c, 10.114.7c, 9d, 10.130.3b, 10.135.5b). There is no such clear tendency found with the relative. This could represent a genuine linguistic development in the later layers of the Ṛgveda, whereby the syntactic employment of ká- became increasingly reanalysed as identical to that of yá-, or perhaps an archaizing hypercorrection at a period where the non-initial position of relatives was losing ground.

  15. 15

    Admittedly, due to the usual uncertainties of Ṛgvedic syntax, there are no absolutely unambiguous examples of an initial-string preverb following an interrogative pronoun in the Ṛgveda. In (21) it is possible to take the preverb in the same constituent as the following verb, in which case it would not be in the initial string, or, indeed, as a postposition governing the interrogative pronoun. The first option is likewise possible in two of the three other potential examples (5.74.2c and 10.135.5b); in the third (6.21.4b), the preverb could be analysed as a preposition governing the following noun. Nevertheless the weight of evidence seems to point in this direction; given the relative rarity of the interrogative pronoun in comparison with the relative and demonstrative pronouns, four examples which can be analysed as attesting the order interrogative–preverb against none suggesting the order preverb–interrogative is relatively good. The preverb-preposition áchā, which appears in (21), is one of three (the others are tirás ‘across’ and purás ‘before’) which are highly restricted in preverbal use and function more as sentential adverbs (they do not, for example, compound with a finite verb in a subordinate clause); áchā forms verbal complexes only with verbs of motion, and it is not otherwise found with the verb √yuj ‘yoke’ in the Ṛgveda, making it unlikely to be part of the verbal complex in this passage. Moreover it is regularly a preposition when governing nouns, not a postposition, so is unlikely to be governing the interrogative. The most likely analysis, therefore, is as an initial-string preverb. Alternative analyses are always possible, however, such that we are left with only ambiguous evidence for the claim that preverbs follow interrogatives.

  16. 16

    These examples are discussed further below.

  17. 17

    These relatively weak demonstratives are variously translated ‘this’ or ‘that’ according to the context, but they have neither the clear proximal sense of ‘this’, for which Sanskrit uses eṣá-/etá-, nor the clear distal deixis of ‘that’, for which Sanskrit uses asaú. On syá-/tyá-, see in particular Klein (1998).

  18. 18

    Note that the final row in this table covers a syntactic possibility excluded for the nominative pronouns treated in Table 2, hence there are only three rows in that table.

  19. 19

    On see, for example, Jamison (1992b).

  20. 20

    The hymn from which this example is taken is very clear in this respect, since it is so clearly concentrated on Indra as its topic and almost every line begins with a relative pronoun referring to him.

  21. 21

    As pointed out by an anonymous reviewer. Nearly all the examples with demonstratives involve a more ‘article-like’ demonstrative, modifying and part of the same syntactic constituent as a following noun that is not itself part of the initial string.

  22. 22

    As already suggested for the relative, but not systematically addressed, by Watkins (1963) and Hettrich (1988).

  23. 23

    Phonetically this was a pitch accent; on the Vedic accent see, e.g., Cardona (1993).

  24. 24

    There is a vast literature on clitics, their definition, classification and status in phonology, morphology and syntax, such that even a representative selection cannot easily be made. A brief overview of many of the issues surrounding clitics can be found in Gerlach & Grijzenhout (1992); for older literature, see Nevis et al. (1994). Etymologically the name clitic implies a word that ‘leans’ on another, i.e. that is incapable of standing on its own, a fact which is usually correlated with some kind of prosodic deficiency, such as lack of stress or lack of a syllable nucleus. I follow the traditional assumption that clitics are independent words in the lexicon which, like any other word, have their own phrase-structure node in the syntactic representation. They are therefore clearly distinct from affixes (on the basis of criteria such as those proposed by by Zwicky & Pullum 1983). There is no need to assume, at least for Ṛgvedic Sanskrit, that any clitics fall into any kind of third category between words and affixes such as ‘phrasal affixes’, as assumed by, e.g. Klavans (1982, 1985) and Anderson (2005); for a criticism of the concept of phrasal affixes, see Bermúdez-Otero & Payne (2011).

  25. 25

    In (23) I have undone the sandhi of the first words for the sake of clarity; in the actual text, m inline image ‘not’ and u ‘and’ surface as .

  26. 26

    For example, ‘and/but’, oûn ‘so’, syntactically parallel to clitics such as ge ‘at least’, te ‘and.’ See also Klavans (1982: 91ff.) on the existence of accented clitics.

  27. 27

    On this see Klein (1992, 1997) and most recently Keydana (2011); unaccented finite verbs appear in examples (1), (2), (7) and (23) among others.

  28. 28

    I therefore consider the term ‘clitic’ to cover both the clitics and ‘particles’ of Anderson (1992: 204.

  29. 29

    Following my definition of clisis in fundamentally syntactic terms, it is conceivable that clitics could exist which do form independent phonological words. As discussed in the main text the evidence, which comes from internal sandhi, appears only with certain segments in certain phonological contexts, so for many Ṛgvedic clitics there is simply no evidence, and indeed there is no clear evidence that any Ṛgvedic clitic could form an independent phonological word. At the very least then, we can say that, as with lack of accent, inability to form independent phonological words appears to largely overlap with syntactic clisis; but cf. fn 37 below.

  30. 30

    Hale (1990; 1995: 30–81).

  31. 31

    The vowels e and o as well as ai and au are treated as diphthongs ending in i/u for the purposes of this rule, which they clearly still were at this stage of the language. On the ‘ruki’ rule in Sanskrit, see e.g. Kobayashi (2004: 148ff.).

  32. 32

    The pronoun sá- undergoes retroflexion at 1.54.3d, 2.24.7d, 3.13.3b, 4.26.4a, 5.2.4c,7b, 6.2.4c, 6.14.1c, 6.20.5c, 6.51.14d, 7.104.10d, 8.13.1d, 8.18.13c, 8.20.16c, 8.33.16a, 8.97.3c, 9.73.8b, 9.79.3a,b, 9.97.38b; syá- undergoes retroflexion at 2.38.1a, 4.52.1a, 6.71.1a,4a, 7.8.2a, 7.38.1a, 8.25.19a, 8.27.12a, 9.98.2a,3a, 10.176.3a; the context for retroflexion of tá- is more restricted (and tyá- never appears in the appropriate context), such retroflexion occurring at 1.113.11a, 1.117.15c, 2.27.13c, 3.55.10c, 4.30.23c, 5.35.1b, 6.66.3c, 7.50.2c, 8.31.17a = 8.70.3a, 8.88.3d, 10.2.4c, 5c, 10.16.6c, 10.64.15a, 10.68.8c, 10.73.2b, 10.162.2c.

  33. 33

    Like other particles, can modify both clauses and specific constituents; in the latter function it usually appears enclitic to that constituent. In this example the appearance of before , in violation of the template provided above, is due to its appearance not as an independent element in the initial string, but as a clitic modifying the preverb prá: as one might say in colloquial English, ‘let that bird be well before (all other) birds.’

  34. 34

    The other instances are at 6.2.4c where the noun ūt inline image ‘with aid’ precedes, at 7.50.2c, 10.2.4c,5c, 10.16.6c and 10.162.2c where the noun agníḥ ‘Agni’ precedes (as in (31)), at 1.113.11a where the accented verb form īyúḥ precedes, and at 6.66.3c where the noun mah inline image ‘great’ precedes.

  35. 35

    My interpretation of this verse depends on taking krátu- here with an adjectival sense, rather than the more common substantive sense ‘intent, power’; Grassmann (1873), s.v., lists several other passages where this adjectival sense (notably parallel to Gr. kratús) is seen, though he (and Geldner) interpreted this instance differently. As can be seen from the translation there is a play on both uses of krátu- in this line.

  36. 36

    The verse could therefore be more literally translated ‘they strengthen (him) greatly with praise, (i.e. they strengthen) Indra.’

  37. 37

    Internal sandhi is never found with disyllabic forms of these pronouns, such as tásmād and tásmābhiḥ discussed here. Such disyllabic forms are relatively rare, in comparison with monosyllabic forms, such that the lack of disyllabic forms displaying internal sandhi could be merely an accidental gap; the same gap appears also, however, in later Vedic. It is possible either that the clitic variants of these pronouns existed only for the monosyllabic forms, or indeed that the syntactically clitic forms of the disyllabic cases formed their own prosodic words (while being no less clitics, cf. fn 29).

  38. 38

    It might alternatively be possible to assume that the transmitted accent on these enclitic demonstratives reflects a later restoration at a stage when the unaccented variants of these words were no longer in existence; there is no evidence to support or refute such an assumption.

  39. 39

    It should be stressed that the definition of ‘enclitic’ adopted here is distinct from the definition assumed by either Watkins or Hettrich; nevertheless, it is sufficiently similar for the comparison to have value, insofar as their definitions are, like mine, fundamentally syntactic (based on appearance in ‘second position’), rather than prosodic or phonological.

  40. 40

    Besides examples (40) and (42), he gives also 5.54.7d, 7.40.6b and 8.25.13c.

  41. 41

    As argued by Kent (1944), Risch (1954), Hoffmann (1956), Strunk (1969), Schmitt (1976), Adiego Lajara (2000) contra e.g. Meillet & Benveniste (1931), Szemerényi (1975). It is tempting also to connect the development of the ezafe construction from a relative pronoun in Iranian languages (Haider & Zwanziger 1984; Bubeník 2009; Haig 2011); this could only work, however, if it derives from a head internal relative clause construction rather than from an appositive relative clause following its noun, since in the latter case the pronoun would originally not have been enclitic but the first word in a separate clause.

  42. 42

    The meanings and functions of the preverbs and prepositions of the Ṛgveda, though not specifically their positioning in the clause-initial string, are treated in an ongoing series of publications initiated by Hettrich: Hettrich (1993, 2002), Hettrich et al. (2004), Casaretto (2010a, b, 2011a–d, forthcoming a, b), Schneider (2009, 2010a, b, 2011a, b, forthcoming a, b).

  43. 43

    In this case, the preverb usually loses its accent and apparently becomes enclitic on the following verb; this is the only context in which a preverb appears before an accented verb (verbs being accented only initially and in subordinate clauses), and the lost accent of the preverb can be seen as an epiphenomenon of this.

  44. 44

    The syntax of preverbs with participles is briefly discussed by Lowe (2012: 97–101); on preverb repetition see Dunkel (1979), Klein (2007).

  45. 45

    As in previous tables, clitics have been ignored in this table, e.g. the second row includes those instances in which a clitic appears between the preverb and verb.

  46. 46

    Similarly 7.38.3 and 10.126.2. This is, interestingly, identical to the regular order of relative and preverb found in Avestan (cf. Hale 1993), and also becomes the regular order in the Brāhmaṇas. The loss of the usual Ṛgvedic order of preverb–relative can result from the loss of the clitic variant of the relative pronoun. A parallel development is seen in Hittite (though I do not necessarily imply that it results from the same processes): Hewson & Bubeník (2006: esp. ch. 4) show that at the earliest attested stage of the language Hittite preverbs appear in the equivalent position to that seen in the RV, whereas in a slightly later stage of the language preverbs appear in the same position as in Avestan, while in a third stage they stand directly before the verb (which is the position reached also in later Avestan and Sanskrit). On Anatolian clitic sequences and preverbs see also Garrett (1992). The position of preverbs in Ancient Greek may parallel the Avestan treatment, cf. Horrocks (1981: esp. 90–171).

  47. 47

    At a later stage of the language this proclisis becomes regular, and it results in univerbation of preverbs with the verbs they modify.

  48. 48

    Bulgarian attests a partially parallel phenomenon, insofar as the Bulgarian clitic cluster hosts at least two types of clitic with differing properties. Three of the large collection of clitics which appear together in the Bulgarian clitic cluster, which is usually in second position in the clause, can stand clause initially; they are šte, the future auxiliary, da, a modal particle, and ne, the negative particle. Whether or not these can all be analysed as proclitic when clause initial, they are at least not enclitic in the same way as all the other Bulgarian clitics; see e.g. Anderson (2005: 151–2, Spencer (2000), Legendre (2000), Franks (2008). Similarly in Pashto some of the large collection of prosodically deficient clitics in the language are syntactically enclitic, in that they cannot appear before the first word bearing lexical stress, while others, though prosodically equally ‘deficient’, have no such restriction (see e.g. Bögel 2010: 89).

  49. 49

    This is discussed in more detail below.

  50. 50

    This is a well-known phenomenon, but its prosodic analysis is controversial, since it seems to require violation of the Strict Layer Hypothesis of prosodic organization (Selkirk 1984: 26; Nespor & Vogel 1986: 7). Assuming that the Strict Layer Hypothesis is in fact a violable set of constraints (as per Selkirk 1995), however, there are still various possible analyses for prosodic-word-plus-clitic sequences. Nespor & Vogel (1986) assume a prosodic category of ‘clitic group’, below the phonological phrase but above the phonological word. An alternative is to assume that phonological words can be recursively formed, following, e.g. Peperkamp (1997) and like Selkirk's (1995) ‘affixal clitic’ structure. The precise prosodic analysis is not important for the discussion here.

  51. 51

    Likewise in Bulgarian proclitic ne and the first following enclitic form a phonological word, as shown by the positioning of the question particle clitic li directly following such a sequence (Franks 2008).

  52. 52

    And indeed there are regular (though not inviolable) orders when more than one element from any one of those categories occurs (e.g. when two sentence particles appear adjacent to one another in the same clause). It is beyond the scope of this paper to account for those patterns.

  53. 53

    In Lowe (2011) the first DF position was assumed to be left-dislocated and adjoined to the E(xpression) Node (Aissen 1992); a further option would be to assume that all such clauses involve a CP, and that the first DF position is the specifier (cf. the syntactic structure for Ancient Greek suggested by Haug 2008); precisely how this is done makes no difference to the current argument.

  54. 54

    For information structure in LFG and the annotations used in (54), see Dalrymple & Nikolaeva (2011).

  55. 55

    Topic preceding focus does at least have the support of the syntactically similar and closely related Ancient Greek (Dik 1995; 2007; Haug 2008), and is assumed for the parent language Proto-Indo-European by, for example, Kiparsky (1995: 153).

  56. 56

    In (57) I have marked the initial NP as a topic, although by rule (54) it could equally be analysed as in focus position. As stated above, the precise nature and distribution patterns of discourse functions in the Ṛgvedic clause are highly uncertain, which may well be for the very same reason that most clause-initial XPs are ambiguous in precisely this way. It can be assumed that DF differences were prosodically marked in some way, e.g. by stress, but that this was not passed down in the oral tradition. In this particular passage, which comes from the famous ‘frogs’ hymn (on which Jamison 1992a), the ‘divine waters’ are topical insofar as the context of the hymn is the start of the rainy season.

  57. 57

    The same qualification applies to (59) as to (57) in regard to the specification of the first NP's DF function as topic; cf. the previous footnote.

  58. 58

    Non-enclitic, not non-clitic: if the first word in a clause following a clausal conjunction is a (proclitic) preverb, the conjunction directly follows the preverb. Enclitics in Ṛgvedic Sanskrit therefore follow not the first non-clitic word, but the first word which is not itself prevented from appearing in first position.

  59. 59

    In this respect my view is more similar to that of Hock, for whom yá- and sá-, etc., appeared in the same ‘inline image’ slot.


  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Second position
  5. 3. Previous approaches
  6. 4. Relative and demonstrative pronouns
  7. 5. Preverbs
  8. 6. Ṛgvedic syntax at the left edge
  9. 7. Conclusion
  10. References
  • Adiego Lajara, Ignasi-Xavier , 2000. ‘Sobre el relativo del antiguo persa’, in Bernhard Forssman & Robert Plath (eds.), Indoarisch, Iranisch und die Indogermanistik: Arbeitstagung der indogermanischen Gesellschaft vom 2. bis 5. Oktober 1997 in Erlangen, Wiesbaden: Reichert, 113.
  • Åfarli, Tor A. , 2010. ‘Adjunction and 3D phrase structure: a study of Norwegian adverbials’, in Nomi Erteschik-Shir & Lisa Rochman (eds.), The Sound Patterns of Syntax, Oxford: Oxford University Press, 932.
  • Agbayani, Brian & Golston, Chris , 2010. ‘Phonological movement in Classical Greek’, Language 86, 13367.
  • Aissen, Judith L. , 1992. ‘Topic and focus in Mayan’, Language 68, 4380.
  • Anderson, Stephen R. , 1992. A-morphous Morphology, Cambridge: Cambridge University Press.
  • Anderson, Stephen R. , 2005. Aspects of the Theory of Clitics, Oxford: Oxford University Press.
  • Austin, Peter & Bresnan, Joan , 1996. ‘Non-configurationality in Australian Aboriginal languages’, Natural Language and Linguistic Theory 14, 21568.
  • Bermúdez-Otero, Ricardo & Payne, John , 2011. ‘There are no special clitics’, in Alexandra Galani , Glyn Hicks , & George Tsoulas (eds.), Morphology and its Interfaces, Amsterdam: John Benjamins, 5796.
  • Bögel, Tina , 2010. ‘Pashto (endo-)clitics in a parallel architecture’, in Miriam Butt & Tracy Holloway King (eds.), Proceedings of the LFG10 Conference, Stanford, CA: CSLI Publications, 85105.
  • Bögel, Tina , Butt, Miriam , Kaplan, Ronald M. , King, Tracy Holloway , & Maxwell III, John T. , 2010. ‘Second position and the prosody-syntax interface’, in Miriam Butt & Tracy Holloway King (eds.), Proceedings of the LFG10 Conference, Stanford, CA: CSLI Publications, 106126.
  • Bratishenko, Elena , 2003. ‘Genitive-accusative and possessive adjective in Old East Slavic’, Scando-Slavica 49, 83103.
  • Bubeník, Vít , 2009. ‘The rise and development of the possessive construction in Middle Iranian with parallels in Albanian’, in Vít Bubeník , John Hewson , & Sarah Rose (eds.), Grammatical Change in Indo-European Languages: Papers Presented at the Workshop on Indo-European Linguistics at the XVIIIth International Conference on Historical Linguistics, Montreal, 2007, Amsterdam: Benjamins, 89101.
  • Cardona, George , 1993. ‘The bhāṣika accentuation system’, Studien zur Indologie und Iranistik 18, 140.
  • Casaretto Antje , 2010a. ‘Syntax und wortarten der lokalpartikeln des Ṛgveda. VI: práti ’, International Journal of Diachronic Linguistics and Linguistic Reconstruction 7, 153.
  • Casaretto, Antje , 2010b. ‘Syntax und inline imageortarten der Lokalpartikeln des Ṛgveda. VIII: abhí ’, Historische Sprachforschung 123, 97156.
  • Casaretto Antje , 2011a. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. IX: ánu ’, Münchener Studien zur Sprachwissenschaft 65, 764.
  • Casaretto, Antje , 2011b. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. XII: ’, Historische Sprachforschung 124, 10851.
  • Casaretto, Antje , 2011c. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. XVI: áti und tirás ’, International Journal of Diachronic Linguistics and Linguistic Reconstruction 8, 173216.
  • Casaretto, Antje , 2011d. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. XIV: ácchā ’, Studien zur Indologie und Iranistik 28, 125.
  • Casaretto, Antje , forthcoming a. ‘Zum Verhältnis von Kasusfunktion und Lokalpartikel im Ṛgveda am Beispiel des Akkusativs’, to appear in Historische Sprachforschung.
  • Casaretto, Antje , forthcoming b. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. XXI: purás, purástāt und purinline image ’, to appear in Münchener Studien zur Sprachwissenschaft 66.
  • Chomsky, Noam , 1995. The Minimalist Program, Cambridge, MA: MIT Press.
  • Dalrymple, Mary & Mycock, Louise , 2011. ‘The prosody-semantics interface’, in Miriam Butt & Tracy Holloway King (eds.), Proceedings of the LFG11 Conference, Stanford, CA: CSLI Publications, 17393.
  • Dalrymple, Mary & Nikolaeva, Irina , 2011. Objects and Information Structure, Cambridge: Cambridge University Press.
  • Delbrück, Berthold , 1888. Altindische Syntax, Halle: Verlag der Buchhandlung des Waisenhauses.
  • Dik, Helma , 1995. Word Order in Ancient Greek: A Pragmatic Account of Word Order Variation in Herodotus, Amsterdam: J. C. Gieben.
  • Dik, Helma , 2007. Word Order in Greek Tragic Dialogue, Oxford: Oxford University Press.
  • Dunkel, George E. , 1979. ‘Preverb repetition’, Münchener Studien zur Sprachwissenschaft 38, 4182.
  • Erteschik-Shir, Nomi , 2005a. ‘Sound patterns of syntax: object shift’, Theoretical Linguistics 31, 4793.
  • Erteschik-Shir, Nomi , 2005b. ‘What is syntax?’, Theoretical Linguistics 31, 26374.
  • Erteschik-Shir, Nomi & Rochman, Lisa (eds.), 2010. The Sound Patterns of Syntax, Oxford: Oxford University Press.
  • Franks, Steven , 2008. ‘Clitic placement, prosody, and the Bulgarian verbal complex’, Journal of Slavic Linguistics 16, 91137.
  • Garrett, Andrew , 1990. The Syntax of Anatolian Pronominal Clitics, Ph.D. thesis, Harvard University.
  • Garrett, Andrew , 1992. ‘Topics in Lycian syntax’, Historische Sprachforschung 105, 20012.
  • Gerlach, Birgit & Grijzenhout, Janet , 2000. ‘Clitics from different perspectives’, in Birgit Gerlach & Janet Grijzenhout (eds.), Clitics in Phonology, Morphology and Syntax, Amsterdam: J. Benjamins, 129.
  • Grassmann, Hermann , 1873. Wörterbuch zum Rig-Veda, 6th edn 1996, edited by Maria Kozianka . Wiesbaden: Harrassowitz.
  • Haider, Hubert & Zwanziger, Ronald , 1984. ‘Relatively attributive: the ‘ezāfe’-construction from Old Iranian to Modern Persian’, in Jacek Fisiak (ed.), Historical Syntax, Berlin: Mouton de Gruyter, 137172.
  • Haig, Geoffrey , 2011. ‘Linker, relativizer, nominalizer, tense-particle: on the ezafe in West Iranian’, in Foong Ha Yap , Karen Grunow-Hårsta , & Janick Wrona (eds.), Nominalization in Asian Languages: Diachronic and Typological Perspectives, Amsterdam: John Benjamins, 363390.
  • Hale, Mark R. , 1987a. Studies in the Comparative Syntax of the Oldest Indo-Iranian Languages, Ph.D. thesis, Harvard University.
  • Hale, Mark R. , 1987b. ‘Notes on Wackernagel's Law in the language of the Rigveda’, in Calvert Watkins (ed.), Studies in Memory of Warren Cowgill, 1929–1985: papers from the Fourth East Coast Indo-European Conference, Cornell University, June 6–9, 1985, Berlin: Walter de Gruyter, 3850.
  • Hale, Mark R. , 1990. ‘Preliminaries to the study of the relationship between sandhi and syntax in the language of the Rigveda’, Münchener Studien zur Sprachwissenschaft 51, 7796.
  • Hale, Mark R. , 1993. ‘Tmesis and movement in Avestan’, Indo-Iranian Journal 36, 2943.
  • Hale, Mark R. , 1995. ‘Wackernagel's Law’, Draft MS, available online at
  • Hale, Mark R. , 1996. ‘Deriving Wackernagel's Law: prosodic and syntactic factors determining clitic placement in the language of the Rigveda’, in Aaron L. Halpern & Arnold M. Zwicky (eds.), Approaching Second: Second Position Clitics and Related Phenomena, Stanford, CA: CSLI Publications, 16597.
  • Hale, Mark R. , 2007. Historical Linguistics: Theory and Method, Oxford: Blackwell Publishing.
  • Halpern, Aaron L. , 1995. On the Placement and Morphology of Clitics, Stanford, CA: CSLI Publications.
  • Haug, Dag T. T. , 2008. ‘Detecting word order freezing in old Indo-European languages’, Paper presented at NRG 2008 (New Reflections on Grammaticalization 4), Leuven, 16 July 2008.
  • Hettrich, Heinrich , 1988. Untersuchungen zur Hypotaxe im Vedischen, Berlin: Mouton de Gruyter.
  • Hettrich, Heinrich , 1991. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. I: ádhi ’, Münchener Studien zur Sprachwissenschaft 52, 2776.
  • Hettrich, Heinrich , 1993. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. II: antár ’, Münchener Studien zur Sprachwissenschaft 54, 147176.
  • Hettrich, Heinrich , 2002. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. III: pári ’, in Matthias Fritz & Susanne Zeilfelder (eds.), Novalis Indogermanica. Festschrift für Günter Neumann zum 80. Geburtstag, Graz: Leykam, 215242.
  • Hettrich, Heinrich , Casaretto, Antje , & Schneider, Carolin , 2004. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. IV: I. Allgemeines, II. úpa, III. áva ’, Münchener Studien zur Sprachwissenschaft 64, 17130.
  • Hewson, John & Bubeník, Vít , 2006. From Case to Adposition: the Development of Configurational Syntax in Indo-European Languages, Amsterdam: Benjamins.
  • Hock, Hans Henrich , 1982. ‘Clitic verbs in PIE or discourse-based verb fronting?’, Studies in the Linguistic Sciences 12, 138.
  • Hock, Hans Henrich , 1989. ‘Conjoined we stand: theoretical implications of Sanskrit relative structures’, Studies in the Linguistic Sciences 19, 93126.
  • Hock, Hans Henrich , 1992. ‘What's a nice word like you doing in a place like this? Syntax vs. phonological form’, Studies in the Linguistic Sciences 22, 3987.
  • Hock, Hans Henrich , 1996. ‘Who's on first? Toward a prosodic account of P2 clitics’, in Aaron L. Halpern & Arnold M. Zwicky (eds.), Approaching Second: Second Position Clitics and Related Phenomena, Stanford, CA: CSLI Publications, 199270.
  • Hock, Hans Henrich , 1997a. ‘Chronology or genre? Problems in Vedic syntax’, in Michael Witzel (ed.), Inside the Texts, Beyond the Texts: New Approaches to the Study of the Vedas: Proceedings of the International Vedic Workshop, Harvard University, June 1989, Cambridge, MA: Department of Sanskrit and Indian Studies, Harvard University, 10326.
  • Hock, Hans Henrich , 1997b. ‘Nexus and ‘extraclausality’ in Vedic, or ‘sa-figé’ all over again: a historical (re)examination’, in Hans Henrich Hock (ed.), Historical, Indo-European, and Lexicographical Studies : a Festschrift for Ladislav Zgusta on the Occasion of his 70th Birthday, Berlin: Mouton de Gruyter, 4978.
  • Hoffmann, Karl , 1956. ‘Zur parenthese im Altpersischen’, Münchener Studien zur Sprachwissenschaft 9, 7986.
  • Horrocks, Geoffrey C. , 1981. Space and Time in Homer: Prepositional and Adverbial Particles in the Greek Epic, New York: Arno Press.
  • Jamison, Stephanie W. , 1992a. ‘Natural history notes on the Rigvedic ‘frog’ hymn’, Annals of the Bhandarkar Oriental Research Institute 73, 13744.
  • Jamison, Stephanie W. , 1992b. ‘Vedic “sá figé”: An inherited sentence connective?’, Historische Sprachforschung 105, 21339.
  • Kent, Roland G. , 1944. ‘The Old Persian relative and article’, Language 20, 110.
  • Keydana, Götz , 2011. ‘Wackernagel in the language of the Rigveda’, Historische Sprachforschung 124, 80107.
  • King, Tracy Holloway , 1995. Configuring Topic and Focus in Russian, Stanford, CA: CSLI Publications.
  • Kiparsky, Paul , 1995. ‘Indo-European Origins of Germanic syntax’, in Adrian Battye & Ian G. Roberts (eds.), Clause Structure and Language Change, Oxford: Oxford University Press, 14069.
  • Klavans, Judith L. , 1982. Some Problems in a Theory of Clitics, Bloomington, IN: Indiana University Linguistics Club.
  • Klavans, Judith L. , 1985. ‘The independence of syntax and phonology in cliticization’, Language 61, 95120.
  • Klein, Jared S. , 1991. ‘Syntactic and discourse correlates of verb-initial sentences in the Rigveda’, in Hans Henrich Hock (ed.), Studies in Sanskrit Syntax : a Volume in Honor of the Centennial of Speijer's Sanskrit Syntax, 1886–1986, Delhi: Motilal Banarsidass, 12343.
  • Klein, Jared S. , 1992. On verbal accentuation in the Rigveda, New Haven, CT: American Oriental Society.
  • Klein, Jared S. , 1997. ‘On verbal accentuation in the Rigveda’, in Michael Witzel (ed.), Inside the Texts, Beyond the Texts: New Approaches to the Study of the Vedas: Proceedings of the International Vedic Workshop, Harvard University, June 1989, Cambridge, MA: Dept. of Sanskrit and Indian Studies, Harvard University, 13966.
  • Klein, Jared S. , 1998. ‘Rigvedic syá/tyá-’, in Jay H. Jasanoff , H. Craig Melchert , & Lisi Oliver (eds.), Mír Curad, Innsbruck: Institut für Sprachwissenschaft der Universität Innsbruck, 36172.
  • Klein, Jared S. , 2007. ‘On the nature and function of preverb repetition in the Ṛgveda’, Studien zur Indologie und Iranistik 24, 6982.
  • Kobayashi, Masato , 2004. Historical Phonology of Old Indo-Aryan Consonants, Tokyo: University of Foreign Studies, ICLAA.
  • Krisch, Thomas , 1990. ‘Das wackernagelsche Gesetz aus heutiger Sicht’, in Heiner Eichner & Helmut Rix (eds.), Sprachwissenschaft und Philologie: Jacob Wackernagel und die Indogermanistik heute. Kolloquium der Indogermanischen Gesellschaft vom 13. bis 15. Oktober 1988 in Basel, Wiesbaden: Reichert, 6481.
  • Krisch, Thomas , 1997. ‘Delbrücks Arbeiten zur Wortstellung aus heutiger Sicht’, in Emilio Crespo & José-Luis García Ramón (eds.), Berthold Delbrück y la Sintaxis Indoeuropean Hoy. Actas del Coloquio de la Indogermanische Gesellschaft, Madrid, 21–24 de septiembre de 1994, Wiesbaden: Dr. Ludwig Reichert Verlag, 283309.
  • Krisch, Thomas , 1998. ‘Zum Hyperbaton in altindogermanischen Sprachen’, in Wolfgang Meid (ed.), Sprache und Kultur der Indogermanen: Akten der X. Fachtagung der indogermanischen Gesellschaft, Innsbruck, 22–28. September 1996, Innsbruck: Institut für Sprachwissenschaft der Universität Innsbruck, 35184.
  • Krisch, Thomas , 2002. ‘Indogermanische Wortstellung’, in Heinrich Hettrich (ed.), Indogermanische Syntax: Fragen und Perspektiven, Wiesbaden: Reichert, 24961.
  • Laughren, Mary , 2002. ‘Syntactic constraints in a “free word order” language’, in Mengistu Amberber & Peter Collins (eds.), Language Universals and Variation, Westport, CT: Praeger, 83130.
  • Legendre, Géraldine , 1996. ‘Clitics, verb (non)-movement, and optimality in Bulgarian’, Technical Report JHU-CogSci-96-5, Department of Cognitive Science, Johns Hopkins University, Baltimore, MD.
  • Legendre, Géraldine , 2000. ‘Morphological and prosodic alignment of Bulgarian clitics’, in Joost Dekkers , Frank van der Leeuw , & Jeroen van de Weijer (eds.), Optimality Theory: Phonology, Syntax and Acquisition, Oxford: Oxford University Press, 42362.
  • Liddell, Henry George , Scott, Robert , Jones, Henry Stuart & McKenzie, Roderick , 1940. A Greek-English Lexicon 9th edn, Oxford: Clarendon Press.
  • Lowe, John J. , 2011. ‘Ṛgvedic clitics and ‘prosodic movement”, in Miriam Butt & Tracy Holloway King (eds.), Proceedings of the LFG11 Conference, Stanford, CA: CSLI Publications, 36080.
  • Lowe, John J. , 2012. The Syntax and Semantics of Tense-Aspect Stem Participles in Early Ṛgvedic Sanskrit, D.Phil. thesis, University of Oxford.
  • Lowe, John J. , forthcoming. ‘Clitics: separating syntax and prosody’, To appear.
  • McCone, Kim , 1997. ‘Delbrück's model of PIE word order and the Celtic evidence’, in Emilio Crespo & José-Luis García Ramón (eds.), Berthold Delbrück y la Sintaxis Indoeuropean Hoy. Actas del Coloquio de la Indogermanische Gesellschaft, Madrid, 21–24 de septiembre de 1994, Wiesbaden: Dr. Ludwig Reichert Verlag, 36396.
  • McCone, Kim , 2006. The Origins and Development of the Insular Celtic Verbal Complex, Maynooth: Department of Old Irish, National University of Ireland.
  • Meillet, Antoine & Benveniste, Émile , 1931. Grammaire du vieux-perse, Paris: H. Champion, 2nd edn. Original edition by A. Meillet, Paris 1915. Second edition “entiérement corr. et augm. par É. Benveniste”.
  • Nespor, Marina & Vogel, Irene , 1986. Prosodic Phonology, Dordrecht: Foris.
  • Nevis, Joel A. , Joseph, Brian D. , Wanner, Dieter , & Zwicky, Arnold M. , 1994. Clitics: A Comprehensive Bibliography, 1892–1991, Amsterdam: J. Benjamins.
  • Oldenberg, Hermann , 1907. ‘Vedische Untersuchungen. 18. Zu den Verbalpräfixen’, Zeitschrift der deutschen morgenländischen Gesellschaft 61, 803815.
  • Peperkamp, Sharon , 1996. ‘On the prosodic representation of clitics’, in Ursula Kleinhenz (ed.), Interfaces in Phonology, Berlin: Akademie, 10227.
  • Peperkamp, Sharon , 1997. Prosodic Words, The Hague: Holland Academic Graphics.
  • Perlmutter, David M. , 1970. ‘Surface structure constraints in syntax’, Linguistic Inquiry 1, 187255.
  • Renou, Louis , 1933. ‘La séparation du préverbe et du verbe en védique’, Bulletin de la Société de Linguistique de Paris 34, 4997.
  • Risch, Ernst , 1954. ‘Zur lautlichen Struktur des Altpersischen’, Asiatische Studien 8, 14954.
  • Schäufele, Steven , 1996. ‘Now that we're all here, where do we sit? Phonological ordering in the Vedic clause-initial string’, in Aaron L. Halpern & Arnold M. Zwicky (eds.), Approaching Second; Second Position Clitics and Related Phenomena, Stanford, CA: CSLI Publications, 44775.
  • Schmitt, Rudiger , 1976. ‘Epitaphios auf altpersisch hya ’, Studia Iranica 5, 14954.
  • Schneider, Carolin , 2009. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. V: ninline image ’, Historische Sprachforschung 122, 118169.
  • Schneider, Carolin , 2010a. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. VII: níṣ ’, International Journal of Diachronic Linguistics and Linguistic Reconstruction 7, 149193.
  • Schneider, Carolin , 2010b. ‘Lokalpartikel im Ṛgveda (Folge 13): ápa ’, Studien zur Indologie und Iranistik 27, 135.
  • Schneider, Carolin , 2011a. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. X: úd ’, Münchener Studien zur Sprachwissenschaft 65, 185241.
  • Schneider, Carolin , 2011b. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. Teil XV.1: sác inline image, sahá ’, International Journal of Diachronic Linguistics and Linguistic Reconstruction 8, 117160.
  • Schneider, Carolin , forthcoming a. ‘Syntax und Wortarten der Lokalpartikeln des Ṛgveda. XI: sám’, to appear in Historische Sprachforschung.
  • Schneider, Carolin , forthcoming b. ‘Zusammen ist nicht immer gemeinsam. Zu einigen Partikeln mit soziativer Bedeutung im Ṛgveda’, to appear in Historische Sprachforschung.
  • Selkirk, Elisabeth O. , 1984. Phonology and Syntax: the Relation between Sound and Structure, Cambridge, MA: MIT Press.
  • Selkirk, Elisabeth O. , 1995. ‘The prosodic structure of function words’, in Jill N. Beckman , Laura Walsh Dickey , & Suzanne Urbanczyk (eds.), Papers in Optimality Theory, Amherst, MA: GLSA, 43969.
  • Simpson, Jane , 2007. ‘Expressing pragmatic constraints on word order in Warlpiri’, in Annie Zaenen , Jane Simpson , Tracy Holloway King , Jane Grimshaw , Joan Maling & Chris Manning (eds.), Architectures, Rules and Preferences: Variations on Themes by Joan Bresnan, Stanford, CA: CSLI Publications, 40327.
  • Simpson, Jane & Mushin, Ilana , 2008. ‘Clause-initial position in four Australian languages’, in Ilana Mushin & Brett Baker (eds.), Discourse and Grammar in Australian Languages, Amsterdam: John Benjamins, 2557.
  • Sims-Williams, Patrick , 1984. ‘The double system of verbal inflexion in Old Irish’, Transactions of the Philological Society 82, 138201.
  • Spencer, Andrew , 2000. ‘Verbal clitics in Bulgarian: a paradigm function approach’, in Birgit Gerlach & Janet Grijzenhout (eds.), Clitics in Phonology, Morphology and Syntax, Amsterdam: J. Benjamins, 35586.
  • Strunk, Klaus , 1967. ‘Wortstruktur und Pronomen im Altpersischen’, Zeitschrift für vergleichende Sprachforschung 81, 26575.
  • Strunk, Klaus , 1969. ‘Wortstellung und Univerbierung altpersischer Korrelativverbindungen’, Zeitschrift für vergleichende Sprachforschung 83, 4958.
  • Szemerényi, Oswald John Louis , 1975. ‘Iranica V (nos. 59–70)’, in Monumentum H. S. Nyberg II , Leiden: Brill, 31394.
  • Vaillant, André , 1942. ‘L'article en vieux slave’, Revue des Études Slaves 20, 512.
  • Viti, Carlotta , 2010. ‘The information structure of OVS in Vedic’, in Gisella Ferraresi & Rosemarie Lühr (eds.), Diachronic Studies on Information Structure: Language Acquisition and Change, Berlin: De Gruyter, 3762.
  • Wackernagel, Jacob , 1892. ‘Über ein Gesetz der indogermanischen Wortstellung’, Indogermanische Forschungen 1, 333436.
  • Watkins Calvert , 1963. ‘Preliminaries to a historical and comparative analysis of the syntax of the Old Irish verb’, Celtica 6, 149.
  • Ziegler, Sabine , 1992. ‘Zur Entwicklung der Relativsätze mit dem Relativpronomen inline image o- in den keltischen Sprachen’, Münchener Studien zur Sprachwissenschaft 53, 25170.
  • Zwicky, Arnold M. , 1977. On Clitics, Bloomington, IN: Indiana University Linguistics Club.
  • Zwicky, Arnold M. & Pullum, Geoffrey K. , 1983. ‘Cliticization vs. inflection: English n't ’, Language 59, 50213.