CC BY
23
A. Keidan
Word boundaries in Panini and Avesta: a linguistic view
I. Introduction
When the theories of the grammatical traditions of the Antiquity (such as Panini's grammar in the Ancient India) and those of contemporary linguistic studies seem to match up, in most cases it is only a coincidence, due to the fact that a science of language has to answer the same questions in every period of time, so that some solutions could be found on the same logical ground, both in the traditional linguistics and in the modern studies. Indeed, some theoretic gaps between the ancient approach to the language and modern linguistics always remain strikingly evident, although, at first sight, the results may appear as similar. The most important difference between the two methods consists in a difference of their main purpose: Panini wants to teach correct Sanskrit (but the question has been largely debated, cf. Cardona 1997, 543ff.), whereas modern linguists usually try to establish the laws governing the growth and the functioning of the Language in general or, as in the case of the IE studies, that of a highly abstract and fictitious linguistic entity, which could not have any practical use.
Thus, for example, as shown by Cardona (1992) in his investigation on Panini's treatment of Lex Bartholomae, what Indian Grammar says about the underlying form of the double aspirate stems (like bhudh- 'to wake') is only apparently matching with the nowadays diachronic reconstruction of such stems. In fact, in Panini's view, this is a logically deduced, abstract phonological form, which was no way conceived as a really existing form, but served as a starting point for further morphonological transformations bringing ultimately to the correct phonetic form found in the actual language. In modern historical linguistics, instead, the stems with two aspirates are reconstructed as belonging to a more or less "real" linguistic form ascribed to a reconstructed PIE stage.
In this paper I will investigate some aspects of Panini's theory of the word boundaries (in terms of external sandhi), in comparison with what the modern Indo-European studies say on this topic and, furthermore, with the state of affairs in the orthography of the sacred books of Avesta. As I will try to show, the two ancient grammatical
traditions, the Paninian one and that of native Avestan philologists, gave two very similar conceptions concerning this problem, in both cases sharply distinguishing from what the modern historical linguistic studies say about it.
II. Case endings in -bh- and -su
As acknowledged by many scholars, in the history of the Indo-Iranian languages (and, more in general, in the PIE) there was a period in which the paradigm of the singular number was more developed than that of the plural (not to speak about the dual). For instance, the plural was lacking of several oblique case-endings. For instance, the case endings in *-bh- and the locative suffix *-su, which appears in "standard" Vedic (and Avestan) as almost completely univerbated declensional terminations, in this earlier period may have been a kind of unbound morphemes, namely adpositions (i.e., pre- or postpositions). Furthermore, a few examples of the unbound use of such case-endings are to be found in the Veda as well. Thus, RV VIII. 11.1 reads deve märtyesu 'among the gods [and] in the men': what we find here are two nouns terminating with -e followed by a unique loc.pl. ending -su (see Bartholomae 1888, 583; however, his reading of this passage of Rgveda is debatable, cf. Oldenberg 1912, 87). This ending seems to act just like a postposition that governs an entire NP, similarly to what is found in RV VII.1.19 däme [...] väne a 'in the houses [and] the woods' (for the seek of clarity, some of the sandhi are undone), with postposition a governing a conjunction of two locatives (still a Bartholomae's reading). Instead, in Classical Sanskrit the oblique endings in -bh- and the loc.pl. ending -su are no longer detachable from the noun and are to be considered as true case terminations.
However, even in Classical Sanskrit, in the neuter stems in sibilant and some of those in palatal the -bh- and -su endings were clearly attached to a nom.sg. form instead of the regular declensional stem. E.g., samraj- 'king' followed by the ending -bhyas gives the form samradbhyas, with rad- virtually identical to the nom.sg. case-form of this word in a voiced context (e.g. samrad bhavati '[he] becomes a king'). Instead, if the ending were attached directly to the declensional stem as in the rest of the paradigm, the word form would have been *samragbhyas. Furthermore, manas- 'thought' gives manobhyas apparently from nom.sg. mano (so, with the external sandhi of -as in voiced context), instead of the regularly expected form *manadbhyas (stated by Brugmann and Delbrück
1897-1916, v. I, §830a; such a sandhi is attested e.g. in RV I.6.3 usädbhir from usas 'dawn'; however, Debrunner and Wackernagel 1929, §150 regard -dbh- < -sbh- as analogical and innovated). In Avestan we can observe some other instances of nom.sg. functioning as stem, e.g. vayzbis 'with words' from nom.sg. vaxs instead of the stem form vac-, see Hoffmann and Forssman (1996, 116).
In the other declensional types the phenomenon is more opaque, or completely non-existing. Thus, the feminine stems in -a are ambiguous, being their nom.sg. the same as the bare stem: devabhyas from deva- 'goddess', nom.sg. deva. In many other declensional stems the -bh- and -su endings are regularly attached to the bare stem: patibhyas from pati- 'lord, master' (cf. nom.sg. patih, patir in voiced context). Lastly, in the thematic declension we can observe an innovated stem in -e- influenced by the plural forms of the pronouns: devebhyas from deva- 'god', nom.sg. devah (devo in voiced context), cf. tebhyas 'to/from them' from sa- 'he', cf. Brugmann and Delbrück (1897-1916, v. II, §277).
Some other facts from Sanskrit and other IE languages corroborate the hypothesis of a later origin of these case-endings from ancient unbound morphemes, as well.
1. To begin with, no such endings are found in Hittite, that is the most early attested of the IE languages. This led some scholars (see a summary in Szemerenyi 1972, 150) to conclude that the endings in -bh- and -su entered the nominal paradigm during the so-called "fourth period" in the history of the PIE, when the Hittite branch was already separated from the rest of the Stammbaum.
2. Neither in the remaining IE languages these morphemes became always true case-endings. The endings in -bh- are found in: Indo-Aryan, Iranian, Armenian, Greek, Italic, and Celtic (besides some other minor languages). Balto-Slavic and Germanic languages show almost regularly endings beginning with -m- in the corresponding case forms (in Germanic, however, a preposition tracing back to the morpheme in *bh- is preserved), cf. Brugmann and Delbrück (1897-1916, v. II, 119).
3. In the Greek language the case-endings in -bh- never entered the nominal paradigm completely. The only form of the IE morpheme attested in Greek is -q>i(v), which is usually employed in what appears as a polysemic oblique case-form of the plural, e.g. Mycenaean po-pi 'with/on/by feet' (read nonqii). Furthermore, the termination -91 occurs in Mycenaean exclusively with consonantal stems and feminine astems (see Lejeune 1958, 167). In the Homeric Greek -q>i(v) ending has been expanded also to the thematic declension, and to the singular
(cf. Homer 9 295 'H1Ô91 'from Ilium', and so on). However, the singular interpretation is debatable: an adverbial reading is almost always admissible since in the post-Mycenaean period this formations were semi-adverbial, and not inflectional, in nature.
4. Moreover, in some declensional types the case-endings in -bh-and -su could bring no accent in Vedic, even if stressed flections were required, cf. Sanskrit mahâdbhis but mahata. The Greek case-ending -91 does not bring the accent at all, cf. vavqn but v^ôç So, they could be considered as having entered the paradigm in a later period, when the accentual pattern of the noun declension underwent a complete reorganization, cf. Kurylowicz (1935, 209).
5. In addition, such an accentual feature is also shared by many derivatives: Greek ^sukôç 'white' but ^svkoç 'white fish', Vedic krsnâ 'black' but krsna 'Krishna', which leads us to the conclusion that the forms in -91 were perceived as derivatives, «[...] simplement des adverbes dérivés de thèmes nominaux, et non pas des cas au sens propre du mot» as says Kurylowicz (1935, 167).
6. In Avestan, the initial -b- of these case-endings is almost always preserved from being lenited into -fi-, as if it stood at the beginning of a separate word, and not in internal position, see De Vaan (2003, 351fn435).
7. No other case termination begins with a bilabial stop: this puts the -bh- terminations in a particular position, cf. Meillet (1937, 298).
8. There are other such semi-adverbial particles in the IE languages, which never became case-endings (as those found in Greek oùpavôBev 'from heaven', ovpavôBi 'in heaven'), or entered the inflectional paradigm only in a later period (such as Sanskrit -tas in mukhatâs 'from the mouth', cf. also the next point).
9. The history of the Indo-Aryan is a continuing rebirth of case-endings: the phonetic erosion of the end of the word was contrasted by creating new case inflections from former adverbial morphemes, cf. Pischel (1981, §369) for Prakrit abl.pl. ending -himto from -bhi(s) plus the already mentioned particle -tas (a very similar agglutination process had taken place also in Avestan: cf. the loc.pl. case ending -huua going up to the original suffix plus the postposition a). Eventually, this process generated the present day system of three different "layers" of case-endings clearly visible in the Modern Indo-Aryan languages, cf. Masica (1991, §8.4).
To sum up, the case forms in -bh- (and, to a certain degree, also those in -su) represented, in the earliest IE languages, not exactly flectional, but rather adverbial forms, very polysemic from the
functional point of view, and with survival of a free use. To say it with Bartholomae (1888, 581), «Es ist zweifellos, dass hier bh is, [bhyas], su, su vom sprechenden deutlich als bildungselemente empfunden wurden». Such originally adverbial morphemes entered the inflectional paradigm properly only in a later period, not in the whole IE domain, and not always to the same degree. In particular, in Vedic these terminations were definitively univerbated when the category of declensional stem was no longer alive, and thus were attached to the nom.sg. form (in its voiced sandhi variant) functioning as a new stem.
III. Panini's point of view
Let us see now what Panini's Grammar says about the -bh- and -su case endings, and which differences exist between the Ancient Indian point of view and that of modern linguistics. In the sutra 1.4.14 Panini gives the definition of the category of pada 'word', or, better, 'basic word form'. The rule says: suptinantam padam 'what ends in sUP or in tiNis pada\ This means that those elements which terminate with one of the declensional endings belonging to one of the two sets: sUP 'nominal endings' (listed in the sutra 4.1.2) and tiN 'verbal endings' (listed in the sutra 3.4.78) are considered as pada. The terms sUP and tiN are pratyahara 'artificial technical terms', i.e. they are abbreviations obtained by compounding the first and the last member (in their abstract shape) of the list itself.
What is unusual with Panini's concept of pada is that he considers as complete a word-form not necessarily provided with the declensional ending. This is stated in the subsequent sutra 1.4.17 svadisv asarvanamasthane 'what precedes su etc., excluding the sarvanamathana set, is also considered pada\ Here, by the term svadi 'beginning with su', Panini defines a group of nominal endings, among which there are those beginning with -bh-, such as -bhis (instr.pl.), -bhyas (dat.pl. and abl.pl.), -bhyam (instr.du, dat.du and abl.du), and the loc.pl. ending -su. Therefore, Panini considers as pada also the word-forms deprived of such case-endings. On the boundary between certain types of nominal stems and these terminations some apparently external sandhi phenomena (i.e. euphonic transformations depending on the position of the word in the utterance) can, indeed, be observed.
At first sight, such a twofold definition of 'word' would be perfectly fitting with the modern Proto-Indo-European reconstruction: a stem being considered as a word even without the case endings attached seems to imply that, in Panini's view, the
terminations in -bh- and -su were free morphemes. However, the conception of pada does not entail any consequence on Panini's theory of lexical semantics and syntax (pace Cardona 1997, 42): he does not in fact consider forms like radbhyas as a merging of two different semantic items, or two different syntactical units, two "words". For nowhere in his Grammar the morphemes in -bh- and -su are stated to be padas themselves.
The concept of pada is purely phonological—if not completely abstract and procedural—in nature. It is only a "descriptive trick" used by Panini in order to account, in a wholly synchronic manner, for the—real or apparent—external sandhi phenomena involving some consonantal stems and the following case-endings.
IV. Evidence from Avesta
In Avestan the loc.pl. ending -huua is preceded by a period in writing, as if it were a compound; the compounds, indeed, were written with a period splitting the members thereof. This features of Avestan seem to prove the late origin of this case ending, which would be still a semi-autonomous lexical item in documented period, perceived by the speakers as a compounded stem, and not a grammatical morpheme. However, this interpretation is false.
In fact, the text of Avesta had undergone what was defined as a redactional compound split, which was often made on an etymologically wrong ground, cf. Narten (1986, 275). The ancient Iranian philologist, or the diasceuasts, as Hoffmann (1958, 8) calls them, reshaped the text of Avesta splitting graphically the members of the compound words. The split was marked by writing a period, but also by restoring artificially the nom.sg. case form in the first member of compound (druxs.manah- 'liar' instead of *druj.manah-), or by introducing the "composition-vowel" (Bartholomae 1894-95, 150) -o at the end of the first member, by analogy with the nom.sg. of stems in sibilant and those of the thematic declension, cf. De Vaan (2003, 434).
The actual criterion used by Avestan diasceuasts in order to individuate the compounded items was simply that of the possible autonomous occurrence, without any consideration of meaning. As a consequence, since loc.pl. -huua coincided with pronoun huua- 'this', the decision was made to mark this case ending with a period and a preceding -o as if it was a second member of a compound (e.g. tsmo.huua < tsmah- 'darkness', xsapo.huua < xsapan- 'night'). That this was no way based on a true etymological analysis is proven by the fact that many other nominal and verbal terminations (e.g. aoso.tara-comparative of aosa- 'harmful'), or even word fragments wrongly
considered as morphemes (e.g. instr.pl. drsguuo.dsbis < draguuant-'deceitful', noticeably not °.bis) were marked in the same way, see De Vaan (2003, §22.5).
V. Conclusion
We can see now that two different ancient philological traditions, that of Vedas and that of Avesta, reached the same results independently. They both managed with a sacred text that had to be preserved from the corruption of the time, being no more natively spoken the languages in which the two books were written. In both cases, this process brought to an etymologically wrong analysis, which however was necessary in order to account for the orthoepy of the ritual recitation.
The main difference is that Panini limited himself in endorsing the results of the padapatha analysis (which predated him), when he introduced his merely artificial and descriptive concept of pada. The Avestan philologist, instead, had gone further in reshaping analogically the members of compounds, and in individuating wrong compounds. In any case, the two ancient tradition in consideration are completely different, as far as the general approach is concerned, from the modern diachronic analysis of the Indo-Iranian languages. Any coincidence is only fortuitous.
Reference
Bartholomae, Ch. 1888: "Die arische flexion der adjektiva und partizipia
auf nt-", KZ 29: 487-586 Bartholomae, Ch.1894-1895: "Vorgeschichte der Iranischen Sprachen", in Geiger, W. and Kuhn, E. (eds.), Grundriss der Iranischen Philologie. I.1, Strassburg
Brugmann, K. and Delbrück, B. 1897-19162: Grundriss der vergleichende
Grammatik der indogermanischen Sprachen, Strassburg Cardona, G. 1992: "Indian grammatical traditions and historical linguistics", in Polome, E.C. and Winter, W. (eds.), Reconstructing Languages and Cultures, Berlin, New York, 239-259 Cardona, G.19972. Panini. His Work and its Traditions. 1. Background and
Introduction, Delhi De Vaan, M. 2003: The Avestan Vowels, Amsterdam, New York Debrunner, A. and Wackernagel J. 1929: Altindische Grammatik. III.1
Deklination der Nomina, Zahlwörter—Pronomina, Göttingen Hoffmann, K. 1958: "Altiranisch", in Handbuch der Orientalistik. I.IV Iranistik, Leiden — Köln
Hoffmann K. and Forssman B. 1996: Avestische Laut- und Flexionslehre, Innsbruck
Kurylowicz J. 1935: Etudes indo-européennes. 1, Cracow
Lejeune, M. 1958: "La désinence -91 en mycénien", in Mémoires de
Philologie Mycénienne, 159-184 Narten, J. 1986: "Zum Vokalismus in der Gatha-Überlieferung", in Schmitt R. and Skjœr0 P.O. (eds.), Studia grammatica iranica. Festschrift für Helmut Humbach, München, 257-278 Masica, C. P. 1991: The Indo-Aryan languages, Cambridge Meillet, A. 1937: Introduction à l'étude comparative des langues indoeuropéennes, Paris
Oldenberg H. 1912: Rgveda. Textkritische und exegetische Noten. Siebentes
bis zehntes Buch, Berlin Pischel, R. 19812: A Grammar of the Prakrit Languages, transl. from
German by S. Jhä, Delhi Szemerényi, O. 1972: "Comparative Linguistics", in Haugen, E. and Winter, W. (eds.), Linguistics in Western Europe (CTL 9), The Hague, Paris, 119-195
