You are on page 1of 34

R O O M AT T H E T O P IN SANSKRIT Ancient and modern descriptions of nominal composition by J. F.

STAAL

Amsterdam

INTRODUCTION If John Braine's book Room at the top were translated into Sanskrit, regardless of the question whether it would attract any readers, its title would consist of one single noun: it could be literally translated as dirog.rham (from diras "top" and g.rham "house, room"), kut.dgdra (from kfqa "peak", agdra "house"), or rendered more freely as candraddld or candraprdsdda.h (from candra.h "moon", ddld "hut", prdsdda.h "upper storey"). 1 The aim of this study is to describe some of the processes which underly the formation of such compound nouns. At the same time this paper attempts to make a contribution towards narrowing the gap between Sanskrit and linguistics. These two fields of study, formerly closely related, have drifted so far apart that publications in the one field have become quite unintelligible to most students of the other. This was different a century ago, a state of affairs which at first sight may seem to be due to the fact that the prevailing preoccupation with language at the time was philological rather than linguistic. But the causes of this divergence lie elsewhere. Linguistics originated more than two millennia ago in India, when rules for the correct description of Sanskrit were composed in Sanskrit. In India it always remained a living tradition. Modern linguists have accordingly acknowledged their indebtedness to P~.nini. However, Western Sanskritists grew up in a tradition of philology; their trade and customary accuracy made them suspicious of the many schools and disputes of modern linguists. On the other hand linguists, proud to have overthrown the tyranny of Latin grammar, became interested in non-Indo-European structures and must have assumed that Sanskrit had little to offer from 1 Quoted from Raghuva.mda 13.40 in R. Harweg, Kompositum und Katalysationstext vornehmlich im spliten Sanskrit (The Hague, 1964), 86.

166

J.F. STAAL

a linguistic point of view. Moreover, to test new ideas and methods, it seemed preferable to turn either to one's own language, or to languages which appeared exotic and well - though not too abundantly - documented. But to scholars with such experiments in view, the Sanskrit "corpus" of continuous documentation over more than 3000 years looked forbidding. Anyone who considers a rapprochement between Sanskrit and linguistics desirable, can find flaws and misinterpretations underlying the attitudes of scholars in both fields. The Indian tradition is characterized by linguistic methods to an extent that must impress, and cannot be neglected by any unbiassed student of Sanskrit ;3 actually, Indian texts from the Veda onwards bristle with linguistic observation and speculation. In addition Sanskritists have derived no small benefit from the data collected by Sanskrit grammarians of various schools, Pan.inean as well as non-P~n.inean. 8 Contemporary linguists on the other hand are beginning to dispense with the idea of a corpus upon which to base linguistic description. So Sanskrit need no longer be a deterrent. But what is more important, there are structures in the Sanskrit language which no student of Latin could have dreamt of, and which are anything but characteristically Indo-European. These features, which are of special linguistic interest, appear predominantly in later so-called scholarly Sanskrit ("wissenschaftliches Sanskrit"4), as illustrated by the commentator's style of the bhds.ya literature. Renou refers to these features, when he states: "En drfinitive, le bhd.sya ~t son apogre - surtout le bhd.sya philosophique - a 6t6 une rrussite achevre dans la voie de l'abstraction, de la condensation, et cela dans des conditions d'autant plus extraordinaires que rien, dans le substrat linguistique 616mentaire, n'y prrparait. ''5 Foremost among the features of this language is nominal composition. It has been maintained that Sanskrit differs in this respect from other 2 Cf. D. H. H. IngaUs, "The comparison of Indian and Western philosophy", Journal of Oriental Research (Madras), 22 (1952-3), 4; J. F. Staal, "Euclid and P~.nini", Philosophy East and West (to appear). Cf. M. B. Emeneau, "India and Linguistics", JAOS, 75 (1955), 145-53; K. Kunjunni Raja, "The Indian influence on linguistics", Journal of the Madras University, 30 (1958), 93-111. s See e.g.J. Wackernagel/L. Renou, Altindische Grammatik: Introduction g~n~rale ~GSttingen, 1957), 34-42, 112-25; W. S. Allen, Phonetics in ancient India (London, 1953), 1-4. 4 H. Jacobi, "~ber den norninalen Stil des wissenschaftlichen Sanskrits", Indogermanische Forschungen, 14 (1903), 236-51; P. Hartmann, Nominate Ausdrucksformen im wissenschaftlichen Sanskrit (Heidelberg, 1955). L. Renou, Histoire de la langue sanskrite (Lyon/Paris, 1956), 144.

ROOM AT THE TOP IN SANSKRIT

167

I n d o - E u r o p e a n languages only quantitatively, i.e., "in the enormous development which the system has undergone, which is unparalleled elsewhere". 6 But as we shall see, these structures lead to expressions for which it would be difficult to find any close parallels in other I n d o E u r o p e a n languages. 7 T o bridge the gap between Sanskrit and linguistics it is necessary first to take such facts into account. F r o m here one could proceed in two directions. Either it can be attempted to present the methods and techniques o f Indian grammarians in such a way that a linguist can derive benefit f r o m it ;s or it can be attempted to describe Sanskrit with the help o f m o d e r n linguistic tools. 9 The purpose o f this paper is to contribute to both aims by studying nominal composition as described in three different systems: (1) P~t.nini's system; 1~ (2) the system recently used for this purpose by H a r w e g ; n (3) a system o f generative syntax, lz It m a y be added here that Pfi.nini's description o f nominal composition does not treat the bhd.sya Sanskrit, which developed m u c h later. But as we shall see, his descriptions deal already with most o f these later features. To Sanskritists even the study o f those three systems o f linguistic descriptions m a y seem suspect. But then it seems very unreasonable, at least to the present author, to study Sanskrit f r o m a linguistic point o f view without considering what Pg.nini said. As regards the remaining two systems, Sanskritists must be provided with an opportunity to choose between competing methods describing material with which they are 6 T. Burrow, The Sanskrit language (London, 1955), 208. 7 Hartmann discusses Chinese parallels: op. tit., 207 sq. 8 See B. Shefts, Grammatical method in Pd.nini: his treatment of Sanskrit present stems (New Haven, 1961), cf. Language, 39 (1963), 483-8; J. F. Staal, "A method of linguistic description: the order of consonants according to Pfin.ini", Language, 38 (1962), 1-10. 9 See Harweg, op. cir., of. below II; J. F. Staal, "Reification, quotation and nominalization", Logic and Philosophy: Essays in honour of l. M. Bochet~ski (Amsterdam, 1965), 151-87. 10 Use has been made mainly of the following sources: P~n.ini, As.t.ddhydyis~trapdtha, ed. C. Sankara Rama Sastri (Madras, 1937); Boehtlingk's edition, translation and indices (reprint Hildesheim, 1964); Renou's translation (Paris, 1948-54; cf. Thieme in JAOS, 76 (1956), 1-23 which is unduly critical); L. Renou, Terminologiegrammaticale du sanskrit (Paris, 1957); Jayhditya, Kagikd, ed. ~obhita MiCra (Banaras, 1952). For a fuller discussion the views of many more commentators should have been taken into account - in the first place of Patafijali; in the present context the practical and simplified notions of the Kdgika should suffice. n See above note 1. 12 See the article quoted above, note 9; other particulars, discussed below, were presented in a paper "Indian logic and Sanskrit syntax" for the International Congress for Logic, Methodology and Philosophy of Science, Jerusalem 1964, of which an abstract was published (pp. 126-7).

168

J . F . STAAL

fully familiar. Also linguists, ultimately aiming at adequacy and simplicity, can only judge the relative value of theories when these are measured against the same data.
1. P.3.ibIINI O N N O M I N A L COMPOSITION

In line with his customary desire for economy of description, thereby juxtaposing what to Western students seems at first sight disconnected, Pan.ini has provided information on nominal composition scattered over different parts of his grammar. Omitting many details and the long account of the accentuation of compounds (6.1.223-2.199), we shall mainly consider the most important and longest connected exposition of the structure of compounds, viz. that occurring in the second adhydya (2.1.3-2.38). This seems in full accordance with Pan.ini's own intention, for this section is headed by the title sam~sa "compound" (2.1.3). The samdsa section has been briefly analysed by Faddegon. 13 In the following account only a few of the general problems dealt with by P~n.ini will be touched upon. In order to understand the samdsa-section some basic notions of Pa.nini's treatment must be first introduced. According to Pa.nini, compounds consist of a main part (pradhdna) and a subordinate part (upasarjana). How in each particular case the pradhana and the upasarjana 14 convey the meaning, is "not a matter to be taught by grammar" (agi.sya), since it follows from "common usage" (Ioka) 15 alone (1.256-7). In the section on compounds, the subordinate part is always the main subject of the rule and will therefore be marked in each rule by the nominative case ending (1.2.43). The compound itself, e.g. rdjapuru.sah., is considered on a par with nominal stems and is referred to by the same term, i.e., prdtipadika (1.2.46). But the analysed compound, e.g. rdj~a.h puru.sah., is not a prdtipadika. ~ Being a pratipadika, the compound loses its case ending (2.4.71), after which it is fit to be declined like other nouns. We are now sufficiently prepared to study the samasa-section. We have seen: (1) in this section rules will in the first place lay down what is the subordinate part of a compound; (2) each compound behaves
is B. Faddegon, Studies in Panini's grammar (Amsterdam, 1936), 61-3. 14 Actually, in this s~tra, P~mini speaks about pradh~na and pratyaya. However, the rule applies to the pair pradhana-upasarjana as well as to the pair prakrti-pratyaya (see K~ik8 ad lot.), from each of which Pfi.nini has taken one tenn. 1~ The term loka is not mentioned in this context, but occurs in similar contexts (e.g., gender) and is given by the author of the K~ika commenting on 1.2.56. z6 Erroneously given by Boehtlingk ad 2.4.71; the Kagik~ mentions only the unanalysed form rajapuru.sab.

ROOM AT THE TOP IN SANSKRIT

169

like a noun. The question immediately presents itself, how the main part of the compound is to be characterised. P~.nini answers this in the rule immediately following upon the chapter heading 2.1.3, as follows: saha supd (2.1.4). This rule has anuv.rtti of sup, indicating the subordinate part; it seems to suggest that the main part of a compound will in each case be marked by the instrumental ending. Boehtlingk, 17 Faddegon 18 and Renou 19 don't seem to have noticed that the referent of supd must be the main part of the compound (not necessarily its second member, as Faddegon says). The convention suggested by this rule is actually adopted by Pfi.nini. The Kdiikd, commenting upon 2.1.4, illustrates it with a later rule, i.e. dvitTyd ~rita ... (2.1.24). Here, what ends in a second case or accusative termination (i.e. the subordinate part of the compound, subject of the rule) is compounded with the words beginning with grita, terminating in nominal endings (dviffydntam. gritddibhi.h saha samasyate). An example is ka.st.agrita.h "undergoing hardship", the result of compounding kas.t.am "hardship (with accusative ending)" with grita.h "undergoing". After these preliminaries Pan.ini starts the description of compounds with the avyay~bhdva or adverbial compounds (2.1.5-21), defined as compounds consisting of an indeclinable part (avyaya) and a declined noun (2.1.6). Examples are adhikumdri "concerning girls", upakumbham "near the pot", etc. That these compounds have neuter endings, is mentioned much later (2.4.18). According to P~t.nini, some occur only as compounds, i.e., they do not appear in an analysed form as well. These analysed forms, later called vdkya, are sentences which generally function in the same way as the unanalysed compounds, are identical in meaning and consist of the same elements. It is true that adhikumdri does not occur in any such analysed form, as for example rdj~a.hpuru.sa.h occurs beside rdjapuru.sa.h. The Kddikd commentary gives as a paraphrase of adhikumdri: kumdr~.sv adhikrtya, which is formed in a different way and from different elements. But there are also avyay~bhdva compounds that appear in an analysed form as well, e.g., prdggrdmam beside prdg grdmdt "east of the village". According to Phn.ini this situation prevails and is applicable to all compounds apart from some of the avyay~bhdva compounds. In accordance with his custom 2~ P~n.ini 17 "Hierbei ist 'mit einem flectirten Nomen' hinzuzudenken".

is Op. cit., 8.

19 "(Les 61~mentssuivants de l'6nonc6:)1 (un mot termin6 par une d6sinenceeasuelle, autrement dit un nora fl6chi,z forme un compos~) avec (un autre mot termin6 par)3 une d6sinence easuelle (autrement dit, avec un autre nom fl6ehi),- (sont trois 616ments de gouvernement, valables depuis le sfi. suivant jusqu'~t II.2,21)". 2o Cf. the article quoted above, note 8.

170

S.F. STAAL

therefore juxtaposes at the beginning of the section the exceptions, i.e. compounds that do not occur in an analysed form. After these have been dealt with, he inserts the rule: vibhd.sd "(from now on there is) option" (2.1.11), thereby indicating that all following compounds (up to 2.2.38) can occur beside their analysed forms. Since Pa.nini, as is shown by this description, always tries to describe compounds by relating them to their analysed forms, it is clear how the question arose to what extent such analysed forms are grammatical. The next type of compound, the tatpuru.sa, dealt with in 2.1.22-2.22, lend themselves much more easily to a regular treatment. P~t.niniclassifies these by indicating to which case of the analysed form a subordinate member of a tatpuru.sa corresponds. In this way the significance of the convention that the subordinate member of a compound occurs as the subject of the rules becomes clear: the subject of each rule is the subordinate member as it occurs in the analysed form of the compound. Pan.ini arrives at the following classification: the subordinate member of the analysed compound can be an accusative (dvitfyd, 2.1.24), an instrumental (trtfyd, 2.1.30), a dative (caturthf, 2.1.36), an ablative (pa~camf, 2.1.37), a locative (saptarM, 2.1.40) or a genitive (.sa.st.hf, 2.2.8) (the separate position of the genitive reflects its so-called quasi-adjectival, i.e. noun-noun relating character~l). In many of these rules and in the intervening rules as well, P~n.ini specifies with what kind of words the subordinate part can be compounded. These kinds of words are sometimes listed and sometimes described semantically, but this is not relevant in the present context. The Kdgikd commentary provides in each case the appropriate examples. The following may be mentioned as illustrations:
CASE COMPOUND ANALYSED FORM

2 3 4 5 6 7

ka.st.adrita.h rn6sapftrvah, yftpaddru v.rkabhayam rdjapuru.sa.h girog.rham

ka.sfam. ~rita.h m~ena pf~rva.h yftp6ya ddru v.rkebhyobhayam rdj~a.hpuru.sa.h girasi g.rham

"undergoing hardship" "earlier by a month" "wood for the sacrificial post" "fear of wolves" "king's servant" "room at the top".

Among the special cases discussed by Pgn.ini in the intervening rules, mention may be made of those compounds where the subordinate is compounded with a noun in the same case (2.1.49-70), e.g.: ~1 Cf. the article quoted above, note 9: note 57.

ROOM AT THE TOP IN SANSKRIT CASE COMPOUND ANALYSED FORM

171

nflotpalam

nflam utpalam

"blue lotus".

These are generally called karmadhdraya, also by Pft.nini who introduced these compounds much earlier (1.2.42) as a subclass of the tatpuru.sa compounds. In the samdsa-section, P~m.ini avoids the term karmadhdraya. This was wrongly explained by Faddegon 22 and is not dealt with here since it will be treated separately in a study by E. R. Sreekrishna Sarma. 23 Among special cases of tatpuru.sa compounds treated separately, mention may further be made of the type of compound illustrated by kumbhak~ra "maker of pots" (2.2.19). For an analysed, but not synonymous form (perhaps for that reason not given by the Kdgikd) one could select: kumbham, karoti "he makes the pot". The element -kdra is here treated as a verbal form, but not a finite verbal form. In the case of nominal forms such compounds are excluded, since the objective genitive cannot be compounded (2.2.14): odanasya bhojanam, but not *odanabhojanam "eating of cooked rice". The next type of compound studied by Pfi.nini is bahuvrihi (2.2.23-8), first defined as those compounds that remain (2.2.23), i.e., those not described earlier. This is not entirely satisfactory, since the dvandva compounds are treated next, but it does not lead to ambiguities. The description given is semantic: such compounds refer to something different from the references of the constituent parts (2.2.24). Though P~.nini does not mention the customary analysed form of bahuvrihi compounds, he may well have known it. Patafijali, at any rate, uses it frequently. This analysed form, e.g. for bahuvrfhi itself, is bahur vrfhir yasya sa.h "he whose rice is much". Finally the dvandva compounds are treated. Their analysed form contains ca "and" (2.2.29). For example, plak.sanyagrodhau beside plak.sag ca nyagrodha~ ca (the names of two kinds of trees). Pg.nini concludes this section with several rules (2.2.30-8) concerning the order of members of the compound. It is interesting to observe that this topic has so far been postponed; for one might have expected P~t.nini to have started with the physical shape of compounds, in each case analysable into a first member (pflrvapada) and a last member (uttarapada). In that case, instead of speaking about main and subordinate members, Pfi.nini could have immediately introduced first

~ Op. cit., 29. 2~ This study will deal with the effects of the sfttra 1.4.1 (ekd sa.mjSa).

172

J.F. STAAL

and last members; thereby the final section 2.2.30-8 would have become superfluous. The reason why Pfi.nini has not done this is not far to seek; in fact, it is implicitly given in the section under discussion. This section starts with the rule, that the subordinate member is the first member (2.2.30). The remaining rules (2.2.31-8) are exceptions. Had there been no exceptions, however, the subordinate members could have been called first members from the start and the term upasarjana would have been superfluous. But since there are exceptions, there are relationships between the members of compounds which do not depend exclusively on the manifest order of the members in the compound, and which therefore cannot be described in terms of order alone. Examples, provided by the remaining rules, include rdjadanta.h beside dantdndm, rdjd "principal tooth", i.e. "king among teeth" but not "tooth among kings". Similarly we have sapatn~mdtd, not "mother of the co-wife", but "cowife with the mother", i.e. co-wife of the husband of the mother i.e. co-wife of the father. Thus, most of P~n.ini's rules establish relationships between the members of compounds, which may or may not be manifest in their physical shape. In other words, if a compound is described as a string of two elements, this string possesses a structure which need not be manifest in and derivable from the order of these elements. This structure is provided, in general, by the analysed form. A final remark may be made with regard to the distinction between (unanalysed) compounds and the corresponding analysed forms. We have seen that a nominal stem loses its case ending according to 2.4.71. As a rule, the nominal stems occurring as first members of compounds lose the case ending as well In the grammar this would already follow, if not from anything else, from the exceptions which are explicitly enumerated in another section (6.3.1-24). An example of such an exception is durdddgata.h "arrived from far". The regular cases (e.g., rdjapuru.sa.h) are described in terms of the analysed forms (e.g., rdjga.h purus.a.h), but could not be described in terms of the manifest elements, since a deletion has been applied which need not be recoverable from the given string alone. The logicians therefore argued that the lost case ending is not expressed but implied (through lat~.san, d). ~4 Compounds are distinguished from analysed forms in that case relationships are implicit. But this does not imply that compounds can only be understood by those who are consciously aware of the implied but lost case ending. This is not only 24 See e.g.K. Kunjurmi Raja, Indian theories of meaning (Madras, 1963), 267-9.

ROOM AT THE TOP IN SANSKRIT

173

explicitly stated by later logicians, but also proved by referring to the fact that compounds are also understood by people who do not remember the case endings. ~5 To speak of deletions of case endings, then, is a linguistic device, not a description of what happens inside the speaker or hearer.

2. HARWEG ON NOMINAL COMPOSITION Harweg treats Sanskrit compounds, especially in the later bhd.sya literature, by making use of modern linguistic tools derived from various sources, in particular from Hjelmslev's glossematics and Pike's tagmemics with a dash of Hockett, Hartmann and others. The resulting study aims at a treatment of the linguistic data from a synchronistic and formal point of view; at least this appears to be implied in the author's statement that his analysis is "1) nicht historisch, 2) aber auch in keiner tiber die Linguistik im engeren Sinne hinausgehenden Weise interpretativ" (12-3). 26 The author stresses in this connexion the fact that his approach "keine Geschichte hat" (11), meaning thereby that his analysis of Sanskrit compounds is purely descriptive and in this respect new: for presumably, previous authors (e.g., Wackernagel, Jacobi, Renou) were primarily philologists and adopted a diachronistic or at the most comparative point of view, whereas e.g. Hartmann, whose outlook was not diachronistic, made use of semantic criteria ("ist auch schon interpretativ"27). The surprising fact is that neither in this connexion, nor elsewhere in the book, is any reference made to Pan.ini or to any other linguist in the Indian tradition. Still, one need hardly doubt that it could be known in Germany since Kielhorn, and wellknown everywhere by 1964 that Pfi.nini and his colleagues adopted a descriptive point of view which is strictly synchronistic and largely formal. Of course, the motivation for adopting these principles of linguistic research in India are different from those underlying modern developments in Western linguistics. But the fact remains and a linguistand-Sanskritist can hardly be excused for what is not only negligence,

Siddhanta-muktavali,ed. Haft Rhma ~ukla (Banaras, 1951), 294 quoted in Staal, "Correlations between language and logic in Indian thought", BSOAS, 23 (1960), 111 note 1. 26 In this section, Roman numerals in parentheses refer to pages of Harweg's work. 27 In paraphrasing the author's German expressions, use is sometimes made of English terms that are customary in English (American or British) linguistic terminology. This is of course avoided where misinterpretations are possible.
25

174

J.F. STAAL

but also, in view of the statement that this analysis has no historical background, a plain error. At the very outset there is another puzzling statement which betrays lack of familiarity with elementary facts regarding the Sanskrit language. In elucidating the subtitle, "vornehmlich im sp/iten Sanskrit", Harweg says that this approximately refers to the period posterior to Dan..din and B~.na. Da.n.din's date has been a matter of much dispute, but Bfin.acan be safely assigned to the seventh century. Harweg then continues: "Die Beschr~nkung auf diese Epoche in der Untersuchung hat ihren Grund aber nicht ntLr in der Tatsache der reichen Bildung exzessiver Komposita in dieser Epoche, sondern mehr noch in dem fiir sic charakteristischen Fehlen bzw. der Nichtbezeichnung des (Wort-)Akzeuts" (13). The first reason refers to a fact, but the second rests on an error. Accents disappeared already in later Vedic (although more brdhman, as may have been accentuated than is apparent from oral tradition) ~g and their absence is characteristic of all classical Sanskrit (i.e., from more than a millenium before Ban.a), not of late classical Sanskrit alone. In the first part of this book, which consists of three parts, the author embarks upon a critical review of several definitions of "word" offered by linguists, followed by a proposal for a definition of his own, for which the way is carefully paved. Though this topic need not be considered relevant to the subject of nominal composition, the reverse clearly holds: for word-definitions, to be adequate for Sanskrit or to be universally adequate, have to cover nominal compounds as well; and that many definitions, current in contemporary linguistics, fail to meet this requirement, is obvious at a glance. It is true that this conclusion is based upon the assumption that compounds are words, and not sentences or entities intermediate between words and sentences. But this can easily be justified. Harweg's motivation is the following: "eine Hierarchie wie hier ist ... ein sowohl Text(e) als auch System einer Sprache konstituierendes Ph/inomen. Insofern nun, als sie ersteren, also den Text bzw. Texte konstituieren soll bzw. als diese(n) konstituierend gelten soll, muss es als dem Hierarchiebegriffinad/iquat erscheinen, wenn nicht die Einheiten einer jeden hierarchischen Stufe f'fir sich den Text liickenlos und identifizierbar erstellen, indem sic kontinuierliche Strings (Reihen) bilden. Wollte man jetzt fiir das Kompositum eine besondere hierarchische Stufe etablieren, so wiirden sowohl die Einheiten dieser Stufe als auch die Einheiten der Wortstufe diskontinuier2s See e.g., Staal, Nambudiri Veda Recitation (The Hague, 1961), 53.

ROOM AT THE TOP IN SANSKRIT

175

liche Strings bilden und grosso modo miteinander alternieren. Deshalb ordnen wit Wort und Kompositum auf einer hierarchischen Stufe an, deren Einheiten wir Wgrter nennen, wobei diese in zwei Varianten, n~imlich als Simplizia und als Komposita, vorkommen" (18). In somewhat simpler terms this may be expressed as follows. Linguistic descriptions are made at various levels, each consisting of strings of elements. Compounds cannot be elements of an intermediate level between word and sentence levels, since strings at this intermediate level would not consist of compounds only, but also of other elements (i.e., words). This would lead to a mixed representation. This argument is sound within the context of a linguistic theory of levels worked out consistently and in sufficient detail, as attempted not only by Longacre (whom Harweg quotes 18 note 11), but by many other, mostly American, linguists. For Harweg himself, who picks linguistic notions from various sources, such an argument could at the most have curiosity value. Instead, a simple reason for treating nominal compounds in Sanskrit (and also in German) not as sentences or intermediate entities and not merely as words, but in particular as nouns, is not only their intersubstitutability with nouns in most contexts, but their nominal terminations. This simple argument was used by Pan.ini, who applied as we have seen the term pratipddika to the stems of nouns and compounds alike. That compounds are words follows from P~n.ini's definition of word (padam) as suptihantam "ending in a case or personal termination" (1.4.14), since compounds end in sup "case terminations". Therefore, if compounds are nouns, and afortiori words, they have to be included in word-definitions. Harweg argues convincingly that so far, and for various reasons, neither universal word definitions (21-8), nor word definitions for a particular language, e.g. Sanskrit (29-35), have been successful. As in many similar cases, the criticism in this book of theories propounded by predecessors is more effective than the positive theory finally offered to take their place. The new definition proposed by the author is not without pretentiousness: "Unsere Definition ist konzipiert als eine universell gi~ltige" (39). In fact, counterexamples are easily found and this definition conflicts not only with the requirement of unmixed representation mentioned above, but with more basic requirements of theoretical consistency. The proposed definition (36-9) defines words as the minimal answers that can be given to certain questions: "Minimum-Wortfrage-Antwort-Korrelate". For example, an answer to "What?" might be "a large house"; a minimal answer to the same question might be "a house". The latter form is

176

J.F. STAAL

therefore a word. Parts of what are intuitively considered words can only be asked about metalinguistically, e.g., "What ending did you say?" Compounds on the other hand are given as answers even to questions that ask only about their members: e.g., in blackbird, the answer to "What kind of bird?" can only be "a blackbird"; whereas in black bird the answer to "what kind of bird?" could be "a black one". The first difficultyin connexion with this definition is the determination of the kind of questions that are intended. Asking about words (as dangerously suggested by the term Wortfrage, but cf. 36 note 45) in order to obtain words would beg the question. Harweg refers especially to questions which contain such interrogatives as wer, wessen, wann, wo, wie (called wh-questions in the field of generative grammar) and excludes yes-no questions (in recent literature reduced to wh-questions). ~9 Even from a purely theoretical point of view, asking such unspecified questions in order to obtain answers is a notoriously difficult problem. What statements, for instance, will count as adequate answers? Not only are such methods suspect in methodology and practical fieldwork, but they have been seriously misleading in philosophy and in a handful of sciences dependent upon it. Since Aristotle based his list of categories upon question words of this type, Western thought has remained under the spell of a particular type of grammar, a~ On Harweg's theory, what words are would entirely depend on the way interrogatives function in a given language. That they function everywhere in the same manner is a mere assumption of a fact that would have to be established first. Interrogatives themselves could on this criterion not be words, since statements containing the same or even other interrogatives would surely be excluded as proper answers to questions. Harweg admits that his definition prevents articles, propositions, conjunctions and auxiliary verbs from being words (37). This admission, however, would lead to the kind of mixed representation discussed before, since these forms do occur in sentences beside "words". But there are other difficulties too. On Harweg's theory, adjectives, at least in combination with articles, can be words in German but apparently not always in English. For a German could answer "Was fiir ein Haus?" by saying "ein grosses", but an Englishman can only answer "What kind of house?" by saying "a large one". We have to take "one" into the bargain. That this would also lead to a theory where elements consist of elements, since "one" itself is a
39 J . J . Katz & P. M. Postal, An integrated theory of linguistic descriptions (Cambridge, Mass., 19fur), 85 sq. zo Cf. Staal, Euelides en Pa.nini (Amsterdam, 1963), 20-3.

ROOM AT THE TOP IN SANSKRIT

177

word (when answering such questions as "how many?") might also be considered inconvenient. Actually, this problem arises more specifically in connexion with compounds, which may consist of compounds and which may be considered nouns consisting of nouns (cf. 18). However, this need not worry us here. It is in fact, as we shall see, a specific characteristic, accounting for one aspect of the infinity of language. In further attempts to test the proposed definition we are faced with personal pronouns. On Harweg's theory, in German "ich" and "reich" are words, but in French "je" can never answer the question "qui?", so that "moi" is a word whereas "je" is not. In English the answer to "who?" can be taken as "me" or "I"; but if it be denied that the latter is a grammatical answer, ' T ' also ceases to be a word. These counterexamples invalidate the universal validity of the definition, and also its validity for English or German. What happens when less familiar languages are studied, can only be imagined. In Sanskrit, the definition leads to most inflected forms of the noun being included as different nouns (e.g., a~va.h, a~vam, 51-2), which is perhaps unwieldy but need not be intolerable. But one case is excluded, namely the vocative, which cannot answer a question (3831); this consequence is surely undesirable. In Sanskrit it is of course crucial to observe as well, as the author does (40) without however taking it fully into account, that we cannot meaningfully ask questions when we are concerned with past stages of the language. These limitations continue to hamper the discussion (e.g., 92). The assumptions of the author with regard to the putting of questions to informants seem particularly naive (e.g., 46; 49: "Ein Test wtirde schnell ergeben...'). As illustrated before, compounds are defined as words, whose parts can be asked about, but only the whole answered. In a characteristic endeavour to overwhelm his readers with heavy definitions, the author inserts the definition of word, just obtained, again in the definition of compound, thereby obtaining: "Komposita sind Minimum-WortfrageAntwort-Korrelate (der Antwortseite), welche einen bestimmten Teil (oder bestimmte Teile) enthalten, der (die) zwar eigens erfragt, aber nieht allein, sondern nur zusammen mit dem Hauptglied des Korrelates geantwortet werden kann (k6nnen)" (40). One curious conclusion drawn from this definition is, that there are no verbal compounds "in German". This is disconcerting, as the author seems to realize, since the definition is intended as universal so that the conclusion cannot
31 The author argues, at least on these grounds quite unconvincingly, that "das alte Problem des sog. Satzwortes ... 16st sich damit als eine Fiktion auf".

178

J . F . STAAL

possibly be confined to German alone. This clearly contradicts the facts (if not in Sanskrit or German, where these could be otherwise described, at least elsewhere, for instance in modern Indo-Aryan). Since these elaborate discussions are only distantly related to the classification of compounds which the author gives later, their general linguistic import need not be discussed in greater detail. Harweg proceeds to introduce two notions adopted from Hockett, which are very important for the description of compounds: Kontentiv (contentive) and Funktor (functor). The first are defined as "Minimum-WortfrageAntwort-Korrelat-Kerne" (45). Again the author could have said "Wort-Kerne" had he attempted to use definitions in order to simplify the description. But for him, a definition seems to be an end in itself. Of course, in neither form would these definitions clarify what contentive means. Nor is this done by further elucidations, one of which may be quoted in order to illustrate the author's style and apparent assumption that all statements are adequate, provided they are sufficiently involved: "'Kern' soil dabei, auf der Seite der (Wort-)FragewSrter (sic), diejenige Resteinheit d.h. dasjenige Restmorphem bzw. - bei Tagmemfusionen innerhalb eines Morphems - dasjenige Resttagmem heissen, das yon einem Fragewort nach Abzug derjenigen Morpheme bzw. Tagmeme, die keinen Fragecharakter haben, iibrig bleibt. 'Kerne" auf der Seite der Antwort-W6rter sind demgegeniiber die Korrelate der Fragekerne, d.h. die Antwort-W6rter vermindert um die den auf der Frageseite abgezogenen Morphemen bzw. Tagmemen entsprechenden Morpheme bzw. Tagmeme" (45; author's italics). Here not only is the wording needlessly complicated (as could, throughout the book, have been avoided only by frequently rewriting passages), but the meaning is obscure. For the nucleus ("Kern") of the question is considered as the interrogative part of the question, and the nucleus of the answer accordingly as the answering part of the answer. Such distinctions, if feasible at all, can only be based upon semantic decisions, which are excluded by the author in advance. Speaking about compounds it would have been much more natural to avoid prior definitions of "word" and to call contentives lexical elements which are invariant parts of compounds. In Sanskrit these can easily be found even if the linguist had to confine himself to compounds alone; for these parts are not declined and undergo only sandhi modifications. In his glossary the author says in fact that a contentive is "Der 'lexikalische' Teil einer Worteinheit" (158), further referring to the explanations under discussion. This could have been better left unexplained. Functors are defined as the remaining parts of words, once the con-

ROOM AT THE TOP IN SANSKRIT

179

tentives have been extracted; or in the author's terminology: "MinimumWortfrage-Antwort-Korrelat-Nichtkerne" (47). This is again somewhat unmanageable. Leaving theory for what it is, let us note that in connexion with compounds, important functors are the Positionsmorpheme (49, referring to Harris). These indicate the order in which contentives (lexical items) occur in the compound. And so we have finally reached the level where P~n.ini started, i.e., the description of compounds as stems which consist in sequences of stems in a definite order. So far, the result yielded by a herd of complicated investigations, is rather pedestrian. On this basis P~ .nini quickly arrived at vastly more interesting conclusions: e.g., the relationship of the parts of a compound is not entirely given by the parts and their manifest order, but presupposes other underlying structures. Let us first see whether Harweg faces this problem. With the help of the above analysis, he observes that in kr~.dd-parvata-nitambe "on the ridge of the mountain of amusement", the contentive nitamba- is complemented by a functor and occurs in a thereby easily identified position "als Gipfel der Konstruktion'. "Als solcher hat es eine absolute Zuordnungs-(Anordnungs-)Stelle im Text: im Sanskrit wie im Deutschen und vielen anderen Sprachen die Schlussgliedstelle" (57). P~n.ini avoided this generalization by first speaking about pradhdna and introducing the concept of uttarapada only later. That Harweg knows that there are exceptions follows from his subsequent reference to "umgekehrte Bahuvrihis" (76-7). For the regular cases he proceeds as follows. Having singled out, in the above illustration, nitamba-, and having placed it at the end, given the lexical items there are two remaining possibilities: kr~.dd-parvata-., and parvata-kr~.dd-... In order to specify either of these, we require additional information, e.g., a Positions- or Stellungsmorphem. Though this is quite obvious, the author states it at length and extends it to compounds with more members (3, 4,..., n), specifying in each case how many functors are needed to determine the compound: i.e., n--1 if there are n contentives. It is seen that these functors are only used to state the relation of concatenation between successive members of the compound. In other words, if a compound is written in an unambiguous terminology as an ordered string (al ... an) consisting of elements (members) {al . . . . , an), 8~ the functors are merely required to state that a2 follows al, aa follows as . . . . . an follows an-1 (or, equivalently, an-1 precedes an, an-~ precedes an-1. . . . . al precedes a2). The really interz2 The customary notation is adopted, by which (al ... an) or (al,...,an) refers to an ordered sequence, and {al ..... an} to a n (unordered) s e t o f elements.

180

J.F. STAAL

esting cases, however, go beyond this simple pattern of order in at least two respects: (1) by constituent structure which can be displayed by bracketing (labelled, if necessary), so that e.g. the uniquely ordered sequence ala~a3 can be interpreted in two ways, as al(a2as) and (ala~)aa (for instance, Sanskrit kr~.dd-parvata-nitambah, interpreted as "the mountain ridge of amusement" and "the ridge of the amusement mountain", respectively); (2) by another underlying structure, which can be made explicit by specifying how the string has been derived, but not by bracketing, e.g. when a~a~ can be derived from either "the a~ of a2" or from "the a2 of al". As we saw before, both these cases were studied by P~n.ini. A special variant was treated much earlier in connexion with the well-known case of indragatru, which is ambiguous when unaccented? z Harweg does not discuss these interesting features systematically. The description of the elementary order mentioned above concludes the first of the three parts of this book, which is the hardest to read and contains the least information and analysis. Faced with the cumbersome and heterogeneous terminology of this part, benevolent readers might consider it the theoretical part and foundation for the rest. But the definitions are haphazard and do not constitute a theory since they lack what is essential to a theory in addition to adequacy, viz., connectedness, consistency and simplicity. Although the terminology derived from the definitions of the first part persists through most of the book, the second part is relatively independent of the definitions themselves and of the underlying theory of questions and answers. This part, entitled "Katalysationstexte", is based upon the notion of Katalyse ("catalysis"), adapted from Hjelmslev. This term gives expression to the kind of analysis of compounds which produces what was called before "analysed forms". Though the author starts by devoting a number of pages (61-6) to the Klarifizierung und Definition of this concept, it is easily seen what is meant. We shall use a terminology which comes close to what Harweg adopts in a few section headings (67, 71, 84, 91) and which will be developed more fully in the next section. Let A, B, ... be nominal or other stems and let n~, n~, na. . . . be nominal terminations. There are compounds, e.g., A B nl which may be analysed in an analysed form: A n ~ - B n3 (2) (1)

8a Quoted in J. E. B. Gray, "An analysis of Nambudiri l.~gvedicrecitation and the nature of the Vedic accent", BSOAS, 22 (1959), 512 note 1.

ROOMAT THE TOP IN SANSKRIT

181

where "-" denotes word boundary. If necessary, the compound A B n itself may be written A q- B q- n, where "q-" denotes concatenation. The expression (2) corresponds to what Harweg calls Katalysationstext. For example, rdj~a.h purus.a.h is the Katalysationstext of rdjapuru.sa.h. For other types of compounds, other forms may have to be selected. Let us briefly comment upon the significance and use of such symbols and devices, since these are of basic importance to what follows and Harweg himself does not seem to be very clear about what theories are for. In fact, he seems to consider Katalysationstexte as a kind of variants, interesting to study and enumerate, and in principle indefinite in number. They seem to be paraphrases somehow connected with the unanalysed forms (61). He says for instance, with reference to such cases as (1) and (2): "Wo eine der beiden Stufen realisiert ist, ist die jeweils andere potentiell existent" (60). What could this mean? We rarely know what is realized and what is potentially existent, and it surely seems preferable not to bother linguistic analysis with such philosophical puzzles. Linguistic theory aims at a description, and if possible an explanation, 84 of what is grammatical This aim may be reached by introducing certain forms, e.g. the analysed forms mentioned before. Such forms may or may not be grammatical themselves. What counts is that they describe, and perhaps explain, the grammatical facts which were the object of the investigation. Compounds may for example be conveniently described on the assumption that much of the syntax of a language has already been described in an earlier part of the grammar. Let us assume that we possess already a description of simple phrases containing (noncomposite) nouns and verbs. Then different types of compounds could for example be distinguished by describing them in terms of analysed forms based upon these simple phrases of the prior description. If a proposed analysed form happens to be grammatical, Harweg's requirement that it should offer a paraphrase of the unanalysed form seems meaningful. It might be required that they can be substituted for each other in certain contexts and under certain conditions. But Harweg does in fact give for the same type of compound more than one Katalysationstext, not just by mistake, but apparently as a matter of fact of which he is fully aware. By this the entire construction becomes pointless and the author's motivation transcends comprehension. Perhaps he thinks of Katalysationstexte as a kind of interpretations. Actually they are sometimes inspired by the interpretations of Indian commentators - in 34 See N. Chomsky, Current issues in linguistic theory (The Hague, 1964), oh. 2.

182

J.F. STAAL

itself an interesting feature to which Harweg rightly draws attention. Only, this is not linguistic description of any kind. When looking at the classification which is actually proposed, the reader is faced with a similar puzzle. Harweg first treats compounds of the type A q- B - e, and subsequently those of the type A -[- B - x. In the first case e, which strictly speaking remains unexplained, is seen to denote the locative case termination. In the second case x, explained as "beliebiger Externfunktor" (84, note 28) and used as a variable, is seen to denote an arbitrary case termination. It seems obvious that the first type constitutes a special case of the second; but most surprisingly, this is nowhere taken into account. This state of affairs may be related to another curious procedure. Harweg postulates that there are several types of compounds, whose catalysis does not give any indication as to the external functor of the unanalysed form. These compounds are given with a locative termination, whereas the catalysis appears undetermined, e.g. in the following illustrations (72-3):
COMPOUND CATALYSIS

k.rta-ak.rte mauna- vrate dfrgha-bdhau

k.rtam ak.rtam, ca yat mauna.m vratam, yasya d~rghau bdhf~ yasya.

A reasonable reaction to this would be to seek other forms of catalysis. But Harweg explicitly states this lack of determination as a characteristic feature of the catalysis, which therefore contains as its final clause an expression of the type "minus {Kasustagmen des Externfunktors des Kompositums}". This addition appears extremely puzzling to the present author. If in each of the cases illustrated above a catalysis of a similar kind had to be given, the obvious alternatives would be the following:

k.rtam ak.rtam, ca yat T A S M I N maunam, vratam, yasya T A S M I N dirghau bdhfl yasya T A S M I N


This would easily restore the equivalence and avoid a state of affairs introduced by wholly artificial and arbitrary means. It would moreover conform to well known Indian practice as far as the bahuvrfhi is concerned (see above p. 171). At the same time it could show most clearly that the locative is just one case among others, in no way particular or basically different. In Harweg's analysis these two mistakes cause much

ROOM AT THE TOP IN SANSKRIT

183

confusion. Moreover the one cannot even be taken to explain the other:
ayam aparo ga.n.dasyopari sphot,a.h.35

The complete description proposed by Harweg consists of three "Hauptparadigmas" (two of which are distinguished by the use made of the locative ending, and of any ending, respectively) the first with two "Subparadigmas". Since space does not permit a detailed analysis of this system in its entirety, we shall only quote in full the first subparadigm of the first main paradigm, referred to by the author as "das reine Funktorinterpolationsparadigma von A q - B - e". The author's terminology will not be fully explained, since the reader may deduce the meaning of unfamiliar terms from the illustrations and examples, and also from the traditional explanations sometimes given in parentheses: "1) Eigenstgndige Flexive (interpoliert zwischen A und B): (Tatpuru.sas) Akk. sg. Instr. Dat. Ablat. Gen. Lok. Akk. pl. Instr. Dat. Ablat. Gen. Lok. Akk. dual. Instr. Dat. Ablat. Gen. Lok.

Diese Flexive k6nnen durch iva 'wie' komplementiert werden. Dabei erscheint auch das Interpolativ "Nominativ - Flexiv q- ira" in Karmadharayas aus Substantiv-5 Adjektiv. Beispiele:
svarga-patite : svargdt patite ('in einem yore Himmel gefallenen'); deva-datte : devena datte ('in einem yon Gott gegebenen');
USW.

megha-~ydme : megha ira ~ydme ('in einem wolkenschwarzen: in einem wie eine Wolke schwarzen').

"2)

Kongruenz-Flexive: a) Kasus + Numerus - Kongruenz - Flexive ( + eva) (Karmadharayas aus Substantiv + Substantiv) b) Kasus + Numerus + G e n u s - Kongruenz- Flexive (Karmadharayas aus Adjektiv + Substantiv)

a5 "this is another swelling on top of a boil".

184

J.F. STAAL

e) Kongruenz - Flexiv ( + ca oder va)


(Komposita aus Adjektiv + Adjektiv) Beispiele: a) kanya-ratne: b) cira-kale: c) ~ubha-agubhe: "3)

kanyayam (eva) ratne cite kale gubhe agubhe ca (va)

Komplexere Reihen: a) (Nominativ-Flexiv} q- {Stamm des Relativpronomens qNominativ- Tagmem + Numerus d- G e n u s - KongruenzTagmeme} b) {Nominativ - Tagmem q- Numerus q- Genus - Kongruenz Tagmeme} q- {Stamm des Relativpronomens q- Nominativ Tagmem q- Numerus q- Genus - Kongruenz - Tagmeme}. Beispiele: a) kanya-ratne: b) cira-kale:

kanya yad ratne ciro ya~ kale

"3a) ist eine Interpolations-Parallelformel ftir 2a), 3b) eine InterpolationsParallelformel fiir 2b). Die Parallelformel f'tir 2c) findet sich im zweiten Subparadigma". The last clause, introducing a "Parallelformel", reveals where, here and throughout the system, the shoe pinches. The author halts between two opinions. Either the system is intended as a description of types of composition, in which case a "Parallelformel" is redundant; or it is intended as a catalogue of commentators' or possible interpretations, in which case it would be without linguistic value and anyhow incomplete. Some further analysis will be provided here to bring out more clearly the limitations of this system and to lead up to the terminology adopted in the next section. The relation between a(n unanalysed) compound form and its analysed form will be expressed by "--*", to the left of which the former, and to the right of which the latter is written. The various congruences mentioned are expressed most easily by identical symbols to the left and the right of " ~ " . The cases will be referred to in P~n.ini's order, partly adopted (unwittingly?) by Harweg, with the Nominative in the first and the Vocative in the eighth place. Since number and case generally go together, we shall introduce subscripts referring to 8 cases in 3 numbers, written 1,..., 8, where a numeral x denotes x, xx, xxx (e.g., 333 denotes the Instrumental Plural, 2 denotes the Accusative Singular, 7 denotes the Locative Singular, Dual or Plural, etc.). The

ROOM AT THE TOP IN SANSKRIT

185

gender will be written, when necessary, as a superscript, bracketed if optional. Harweg's system quoted above can now be written as follows: 1) A + B A+B 7 ~ Aj - ( i r a ) 7 ~ A1 - (iva)B7 ( j = 2 , . . . , 7 )
B7

2) a) A + B7 ~ A 7 - B7 b) A+B g ~ A~ - B 7

c) A + B ~ - ~ A ~
3) a) A + B ~ b) A + B ~

-B~

- ca (vd)
- B~ - B~

~ A1 - y a [ -~ A[ - y a ~

It is easily observed that the analysed forms under 3) are mere variants which do not add to the classification of compounds dealt with. It is also clear that the two cases under 1) can be combined by deleting the second and extending the range o f j by writing j = 1..... 7. It is similarly obvious that this system can be easily turned into a much more general system (corresponding to parts of Harweg's second and third main paradigms as well) by replacing final subscripts 7 by variables i ranging over 1 ..... 8. Using further the bracketing of superscripts to denote option, we obtain the following system which incorporates the former but is much more simple and powerful: 1) A + Bi 2) A + B ( ~ )
A + B~ ~ ~ Aj - (iva) - Bi

0 = 1.... ,7; j = 1,...,8)

--* A(~) - B i
AT - B~ - ca (va').

Once this terminology is adopted, it becomes fairly easy to translate Harweg's system into a new more simple and adequate language. For example, the second subparadigm of the first main paradigm (71-3) which consists of six clauses ("Formeln") can be expressed as follows:
I. A + B7 ~ A~ - BTca

II.

A+Blt

- C g + D~I

---* A ~ - C~ - c a -

B~

- D~

- c a se

36 Harweg places ca between B and D.

186 III.
IV.

J.F. STAAL A+B~


A -]- B 7

~ A] - B~] - c a - y a ]
--~ A777 - B777

V. VI.

A + B 7 ~ A1 - B 1 - ya 6 A + B 7 ~ An - Bn - ya 6.

As noted before, some of these clauses should be supplemented, viz., III, V, VI, with tasmin or tar Next the system can again be generalized by replacing the subscript 7 by variables ranging over 1,...,8. By means of these modifications almost all cases described by Harweg with the help of later paradigms, are included in advance. Exceptions are the avyayfbhdva, compounds whose first members are numerals or whose last members are verbal roots, and what Harweg calls "iJberlagerte" compounds (see below). Harweg introduces an unnecessary complication by requiring an additional noun in the catalysis of such forms as"

an-ahgaO sa-udarah jala-jaO


for which the following analysed forms are given, respectively (94):

devo na vidyate aftga.m yasya bandhu.h samdnam udaram, yasya jale jdtam, kusumam.
But in such cases, deva.h "deity", bandhu.h "relative" and kusumam "flower" need not be mentioned provided the general rule is adopted that sa is supplied in all analysed forms of bahuvdhi compounds. The syntactic analysis of an-a~ga.h is merely "he whose limbs do not exist", "he who has no limbs", or "limbless". That this form is used to refer to a deity is a matter of semantics and usage. Similarly, the syntactic analysis of sa-udara.h is "he who has a womb in common with", i.e. "uterine"; that this is used to refer to relatives is again a matter of semantics. This is perhaps most obvious with regard to jalaja.h which must be analysed as "(what is) born in (or perhaps: from) water", and happens to be used of certain flowers. However, this information belongs

ROOM AT THE TOP IN SANSKRIT

187

to the dictionary, just as the facts that ndga.h refers to snakes as well as elephants, vars.a.h means "rain" but also "a year", etc. etc. belong to the dictionary. It may safely be concluded that these complications are artificial and there is no need to add specific syntactic rules to deal with such forms as an-a~ga.h, etc. In the last chapter of the second part (100-9: ch. 19) Harweg deals with "kombinierte (En)katalysationsreihen" or longer compounds which are subdivided into various classes with the help of two characteristics that may be present or absent, i.e., "unterbrochen" and "iiberlagert". Harweg speaks of unterbrochen in cases where a compound AB consists of members of which at least one is again composite, e.g. B ---- CD, in such a way that in the resulting compound two members are separated by one of the sub-members of the composite member; e.g., in the above illustration, in such a way that in ACD, where A and D could also be directly compounded as AD, A and D are separated by C. This can obviously occur on account of the semantic fact that compounds may denote their "second member". In the counterexample AB where A = CD there is no "Unterbrechung" in the resulting CDB, since only DB is a compound. In the example AB, where B = CD, Harweg stresses the semantic fact that A and D "belong together", where it would have been sufficient to observe that ACD can be said to be "unterbrochen" because of the possible occurrence or grammaticality of AD. An illustration of this state of affairs is baddha-t~k.sn.a-kauks.eyaka.h (102) "girded with a sharp sword", as compared with baddha-kauk.seyakah. "girded with a sword". The most obvious method for expressing such relationships in general is by means of brackets, e.g., baddha (tik.sn.akauks.eyaka.h) or A(CD); the "Unterbrechung" consists merely in the possible occurrence of AD. But this method is not adopted by Harweg. It might have saved him about half of this additional chapter. The other half is provided by the "tiberlagerte" compounds. These are interesting and cannot be simply dealt with by bracketing, however complicated. Examples are (105): k.r~a-ak.r~a-udara-uras-sthala.h "whose stomach is thin and whose chest is broad"; or: gakti-siddhi-pu.spa-phala.h "whose flower is gakti.h and whose fruit is siddhi.h". These cases correspond to what can be expressed in English with the help of such words as "respectively". Symbolically, if we have ABCD, this could be formulated semantically by saying that C corresponds to A and D corresponds to B. For a syntactic formulation the possible occurrence or grammaticality of compounds AC and BD is all that has to be taken into account. It is clear that the analysis of the resulting compound

188

J.F. STAAL

cannot be effected with the help of brackets, e.g. as (A)B(C)D, for each expression of this kind can be interpreted in many ways. Subscripts could be used, provided their use is defined in an appropriate manner, e.g. writing the above compound as A1B2C1D2. A rule describing these forms will be given later. The third part of Harweg's book, dealing with "Kontextselektion", will not be dealt with here. This part treats a topic which is much discussed amongst contemporary linguists. It may well be doubted whether any question regarding context belongs to either syntax or semantics. The main problem of linguistic description rather appears to be the characterization and explanation of what makes grammatical expressions grammatical - i.e., in the case under discussion, what compounds are grammatical and what compounds do or can occur in the language. Once these are given (by means of grammatical rules - because of the " c a n " - rather than by enumeration) the question may arise, when and under what circumstances specific expressions are used in the language. The solution of this problem depends on many factors quite different in nature from syntactic and/or semantic factors. However, to expound these views in full and to adduce the arguments needed for their foundation would require more space than is available here. 37 Harweg's treatment of Sanskrit compounds has been analysed in some detail, not only because mistakes are very instructive, but also because an attempt should be made to prevent unbiassed philologists from concluding that any treatment of a feature of Sanskrit with the help of modern linguistic methods must necessarily be confused and obscure.

3. A GENERATIVE GRAMMAR OF NOMINAL COMPOSITION The following system attempts to deal in a systematic way with the various kinds of compounds described so far, by making use of ideas on generative syntax in general and transformational syntax in particular developed in recent years by Chomsky and his followers.3s For this kind of treatment the way has already been paved in the previous sections of this paper; in fact such treatment would naturally result from further 8~ Cf. e.g., Staal, "Generative syntax and semantics", Foundations o f Language, 1 (1965), 133-54. 88 See e.g., A. N. Chomsky,Syntactic Structures (The Hague, 1957); Current Issues in Linguistic Theory (The Hague, 1964).

ROOM AT THE TOY IN SANSKRIT

189

developments of Pfinini's methods and further criticisms of Harweg's methods. A discussion of the reasons which have led to these new developments in linguistics in general cannot be undertaken here; but they are not so very different from the arguments used so far in the present context. It is hoped that those who are familiar with the material described will be in a position to appreciate the considerable simplification and systematization which can be reached within the framework of these new methods. It will be interesting to observe in what respects they constitute a return to P~.nini. s9 However this parallelism should not be overstated since there are indubitable innovations, in particular in the realm of formalization as well as linguistic theory, which go beyond what could be easily dealt with within a purely P~n.inean framework. The resulting simplification is partly achieved by formalization. This formalization, which may not be immediately perspicuous to philologists though it is certainly in tune with the methods of Indian grammarians, 4~ is basically similar to some of the formalizations used in mathematics and other mathematical sciences. For example, formalization enables the mathematician to replace numerous expressions of arithmetic by algebraic expressions in a simple and systematic manner. A simple illustration is provided bythe arithmetical expressions 4.2 ----(3 + 1).(3 -- 1) =321 = 8; 5.7 ----(6 --1).(6 + 1 ) = 62 - 1 = 35; 635.681= ( 6 5 8 - 23).(658 + 23) ---- 6582 -- 232 = 432435, etc., in contrast with the algebraic expression (a + b).(a -- b) ---- a S -- b 2. The question whether such formalization could ever be useful in linguistics may give rise to endless discussion; its general treatment can only be begun in an efficient manner when specific formalizations are suggested in order to deal with specific linguistic problems. Here a specific formalization will be given for nominal composition in Sanskrit. Nominal compounds will be described by applying transformations to sentences of a certain simple type. These sentences, belonging to what has recently been called the base component of syntax, 41 will be described first (A). Next the various types of compounds will be defined on these sentences by means of transformation rules (B). Lastly further nominalizing transformations will be referred to (C). 39 cf. also Staal, "Euclid and P~.rfini",Philosophy East and West (to appear). 4o See Wackernagel-Renou, op. cit. (above, note 3), note 614; Staal, "Contextsensitive rules in Pfi~ni", Foundations of Language, 1 (1965), 63-72. 41 See N. Chomsky, "Topics in the Theory of Generative Grammar" (Preliminary Version) to appear in Current Trends in Linguistics, 30; el. Foundations of Language, 1 (1965), 154.

190

J. F. STAAL

A. The base component


A large group of sentences belonging to the base component can be described in terms of the simple sentence structure noun - verb - noun
or
N 1 - V - N 2.

(1)

That such sentences together constitute a special case within a larger group of sentences may be seen from their generation in generative syntax, which could be effectuated by the rules:
S -o N P - VP VP -o V - N P N P ~ N,

where N P and VP stand for "noun phrase" and "verb phrase", respectively, and where -o is a rewriting symbol used in such a way that "X -o Y" is read as "rewrite X as Y". As in Pan.ini's grammar, nouns and verbs m a y be defined in terms of nominal and verbal terminations, respectively. Nominal terminations are written as nl, where 0 < i < 24; i = 0 denotes zero-occurrence of the noun. The subscripts i of n refer to 8 cases in 3 numbers, written 1 . . . . . 8 where a numeral x denotes x, xx, xxx. The cases are enumerated in P~m.ini's order, i.e.: 1 nominative 2 accusative 3 instrumental 4 dative 5 ablative 6 genitive 7 locative 8 vocative. The numbers are singular, dual and plural and are written as single, double and triple numerals, respectively. In other words, the subscripts of the nominal terminations refer to case and number as follows:

ROOM AT THE TOP IN SANSKRIT SINGULAR 1 2 3 DUAL 11 22 33

191

Nominative Accusative Instrumental

PLURAL ANY NUMBER 111 1 222 2 333 3, etc.

Verbal terminations are written as vt, where t < T, T is a constant and t = 0 denotes zero-occurrence of the verb. These terminations could be further specified as in the case of nominal terminations, but this is not required in the present context. If roots are denoted by capital letters, concatenation by ' + ' and word boundaries or 'q- @ q-' by '-', (1) can now be written as: A + n l - B + v t - C + nj.

(2)

From this structure some simple sentence types can be produced with the help of the following rules: (I) (II) (III) (IV) (V) I f i @ 0 , t = 0 , j = 0 , t h e n i = 8. ffi=0, tv s0,j=0,thent=f(T). I f i @ 0 , t = 0 , j =~0, t h e n i = 1 , 4 a n d j = 1. Ifi@0, t@0, j=0, theni@6. I f i :/= 0, t ~ 0, j @ 0, then: i f j @ 4 , 6 , t h e n i = 1; ifj=4, theni= 1,2; i f j = 6, then B does not occur between A and C.

Sentences illustrating these rules were already treated in a somewhat similar order by B. Delbrtick) z Since the base component will be utilised here only in order to describe nominalizations, examples may be confined to the following: (III): for t ---=0, i = j = 1 : ghat.o nila.h "the pot is blue"; (IV) : for i = t = 1, j ---=0: ghato bhavati "(it) is a pot". Such examples are easily multiplied, producing e.g. one-word sentences consisting of a vocative or imperative. One disadvantage of this kind of description is the following. In

~2 VergleichendeSyntax der indogermanischen Sprachen, III (Strassburg, 1900), especially chapters 35-7.

192

s.r. STAAL

highly inflexional languages such as Sanskrit the word order in such sentences as (1) is optional in the sense that (1) and any of the following are generally considered synonymous: N1 - N~ - V N2-VN1 N2 - N 1 - V V - N 1 - N2 V - N~ - N1 (cf. Latin puer amat puellam, puer puellam amat, etc.). 43 Now only in so far as these sentences are going to be considered synonymous from the point of view of the semantics attached to the syntax of the grammar, 4* will it be necessary to express their equivalence by special rules such as: N-V--r V-N N 1 - Nz ~ N 2 - N1. If such rules are adopted, however, they will not only greatly detract from the simplicity of the description, but also give preferential treatment to one of six permutations which seems hard to justify on purely syntactic grounds. This suggests a description in which the introduction of order is postponed until a later stage; i.e., a description in other terms than strings. However, we shall see that these objections do not apply if a similar description is given to nominal compounds.

B. Nominal composition The main types of nominal composition can now be described with the help of the following transformation rules: (i) (ij) dvandva A -}- ni - B q- nl - ca -~ A -b B -b nil karmadh~raya A q- nl - B q- nl --, B q- A q- nl for i = 1

(3) (4)

48 Cf. H. B. Curry in Structure of language and its mathematicalaspects (Providence, 1961), 66, who raises similar objections. 4, Cf. the paper quoted above note 37.

R O O M A T T H E T O P IN SANSKRIT

193 (5) (6)

tatpuru~;a

A + nt - B + n l - . A + B + nl for i :~ 2, 8 A + n l - B + vt ~ A + N(B) + nl f o r i = 2, 3, 5, where N(B) is a nominalization of B. (iij)


bahuvrihi A + n1 - B + n x -ya

+ ns - ta + nl ~ B + A + n 1.

(7)

In each rule, what occurs to the left or right of the re-writing symbol can be further specified when use is made of brackets, e.g. as follows: (A -t- nO - (B "-k nl) ~ (A q-- B) -{- nl. In this case bracketing on the left, but not on the right, was already implicit in the notation. Such bracketing, which may be labelled when necessary (i.e., when brackets occur within each other), can also be written by means of a phrase-marker ("P-marker") or "tree", e.g., in the above case:

_ .> . A ~ ~
A nl

nl

nl

The tree to the left of the symbol --* may be called the underlying Pm a r k e r ; that to the right, the derived P - m a r k e r . It is easily verified that in each of the cases (3)-(7) the derived P-marker has the form:

The "usefulness" of nominalization is illustrated by the simple form of this tree as compared with the more complicated trees of the underlying P-markers. By each nominalizing transformation the amount of structure, a quantity which could be measured by an index, is reduced. Actually, as we shall see, it can be reduced so far that ambiguity results, in which case one could speak of "nominalizational homonymy". It may be added here that the transformations which map an underlying Pmarker into a derived P-marker express adequately the underlying structure mentioned before that was illustrated by means of "analysed forms". From the shape of the rules (3)-(7), where some of the elements ocur-

194

J.F. STAAL

ring on the right can be written on the left in the same or another rule, it is seen that nominalizing transformations are "recurrent" in the sense that nominalization can be iterated indefinitely. For example, the right of (3) may be substituted for A in (4), where B must then be replaced by another symbol, e.g., C, with the following result: A+B+nl-C+nx~CWA+B+nv When these transformations are iterated indefinitely, the derived Pmarkers branch indefinitely far to the left. This can be expressed by saying that nominal composition in Sanskrit is left-recursive.45 It may be added here that this recursiveness expresses adequately what was referred to before as "one aspect of the infinity of language" (above p. 177).

EXAMPLES. (i) (4) f o r i = 6:

rdj~ah, putra.h -* rdjaputra.h


(ij) (iij) (5) for i = 5:

svargdt patita.h --* svargapatita.h


(6):

kaet.ho n~loyasya sa.h ~ n~lakan,t.ha.h left-recurrence: x ~ S / ~y~paka ddi "etc." jola


"water"

vat~ "possessing"

S'p.rthiv~tva"earthness" "pratiyogi "counterposifive" A J l a b h d v a " n e g afion'' "p.rthiv~tva


"pervading"

4~ See N. Chomsky, "On the notion 'rule of grammar'", Structureof languageand its mathematicalaspects(Providence, 1961), 10 (properly speaking, it is the recurrent symbols which are called left-recursive). The Sanskrit example given here and the fact that branches with as many as 12 nodes can be found in Sanskrit (see next note) appear to refute Yngve's hypothesis expounded in the same volume (pp. 130-8).

ROOM AT THE TOP IN SANSKRIT

195

labelled bracketing: ( ( ( ( ( ( jala-ddi) vydpaka)p.rthiv~tva)abhdva)pratiyogi)prthiv~tva)vaff


654321 1 2 3 4 5 6

jaladivydpakap.rthivftvdbhdvapratiyogip.rthivitvavat~ "possessing earthness,


which is the counterpositive of the negation of earthness pervading water etc. ''46 The cases of compounds corresponding to what in English can be expressed with the help of "respectively", e.g., A1B2C1D2, may be defined from two underlying P-markers as follows: A+C+nl --* A + B -t- C -t- D + nv B+D+nl Strings of this kind, when of arbitrary length, are of considerable interest for linguistic theory. Chomsky has shown that languages consisting of such strings cannot be generated by a very common type of grammar, i.e., a context-free grammar. That rules such as (4)-(6) can lead to "nominalizational homonymy" is clear from the fact that the variable i which occurs to the left of " ~ " in these rules has disappeared on the right. This reflects the well-known fact that e.g., such compounds as devaddnam (from deva.h "deity" and ddnam "gift") can be analysed as devasya ddnam "gift of the deity", devdndm, ddnam "gift of the gods", devdya ddnam "gift to the deity", devebhyo ddnam "gift to the gods", etc. Moreover, the case termination translated here by English "of" is itself as ambiguous as English "of" (eft Latin amor dei). It may not be superfluous to stress the fact that order is all-important in the derived P-markers of nominalizing transformations; in general, if the order of elements within a compound is changed, the compound itself changes in function, meaning and analysis. The exceptions quoted above from Pfm.ini47 underline this fact. Therefore a description of nominal composition in terms of generative syntax and based upon strings is particularly appropriate. ~6 This is taken from Ma.nika.na. A navya-nyaya manual, ed. and transl. E. R. Sreekrishna Sarma (Adyar, 1960), 42. See also J. S. Speyer, Vedische und Sanskrit-Syntax (Strassburg, 1896), 34. ~7 See above p. 172.

196

J . F . STAAL

C. Nominalizing transformations
There are other kinds of nominalizing transformations which may combine with the above transformations to form larger composite expressions. Mention may be made in this connexion of the abstract formative suffixes such as -tva "-ness" which can be attached to any nominal stem, and also of the nominalization of verbal stems, e.g., of bhdva from bhava-, or of kalpand from kalpa-, etc. Such nominalized forms have been referred to as N(B), where B stands for a verbal stem. Some transformations of this type, defined on strings of the form (2), are the following:

A -t- ni - B -k nl --* A -k ns - N(B) -k nl --. A -{- N(B) -k nl (cf. (4)) A + n l - B q- nl - C + vt --* A + ns - B + N(C) + nl -* B -t- nl - A q- N(C) -t- nl -* B -t- A q- N(C) -t- nl (cf. (4)) for i v~ 1, 6, 8.

(8) (9) (10) (11) (12)

EXAMPLES:

ghat.o n~laO --* ghatasya n~latvam ghatanilatvam


"the pot (is) blue"

bhaktir j~dndya kalpate --* bhakter jgdnakalpand j~dndya bhaktikalpand j~dnabhaktikalpand


"love is conducive to knowledge". For a fuller treatment of these transformations the reader is referred to "Reification, quotation and nominalization". 4s However, one remark may be added to the statements there made. In these transformations there are again recurrent elements which are characterized by leftrecursiveness. Actually, the derived P-markers of those transformation results which are nominal compounds, i.e., of (9) and (12), are respectively:

4s See above note 9.

ROOM AT THE TOP IN SANSKRIT

197

and

A final observation may be made with regard to the interpretation of such transformations and of rewriting rules in general; this may prevent misunderstandings which have frequently arisen in the past, and may also lead us back again to P~mini's system. Rules of grammar such as these rewriting rules are laid down by linguists in accordance with requirements of empirical adequacy and general simplicity. Such rules may often be replaced by more adequate rules or simpler ones. But they need not form part of a model of the speaker or embody a kind of psychological reality. 49 Such rules need not correspond to parts of a speaker's brain, nor is there any need for them to be consciously applied by a speaker before he produces the utterance for the description of which these rules are used. Though it is obvious that the establishment of such a thesis would require tests of a psychological and not of a linguistic nature, it is repeatedly thought that this is the linguistic hypothesis presupposed in generative grammar. A recent (1964) example is provided by Dixon, who says that there is no point in comparing linguistic rules with traffic rules: "In the matter of driving, New Yorkers do have an explicit set of rules that they learn and that they consciously try to follow. But speakers of a language do not have explicit rules that are consciously followed in this way ...-50 Long before there was any traffic in New York, Indian scholars may have made the same mistake, for it was repudiated by the later Indian logicians. Pg.nini laid it down that a case ending is lost when a word is nitroduced into a compound. This rule was recognized as a linguistic 49 Models of the speaker have different properties: see e.g.G.A. Miller & N. Chomsky, "Finitary Models of Language Users", Handbook of Mathematical Psychology, II (New York, 1963), 419-91. For the relationship between description with the help of rules and the postulation of laws in historical linguistics, see especially: M. Halle, "On the role of simplicity in linguistic descriptions", Structure of Language and its Mathematical Aspects (Providence, 1961), 89-94. 50 R.M.W. Dixon, "'A trend in semantics': Rejoinder", Linguistics, 4 (1964), 16-7.

198

J.F. STAAL

device. Its adoption does not imply that compounds can only be understood by those who are consciously aware of the lost case ending, for people who do not remember the lost case ending can still understand compounds. 51 The generative syntax of compounds in Sanskrit, of which an outline has been sketched, covers in a consistent and fairly simple manner all the facts dealt with by Harweg in his monograph. Many of the special features not mentioned by Harweg but dealt with by Pan.iN can be accounted for by small extensions of the above system. A fully adequate treatment of the facts of Sanskrit composition might require more sophisticated methods. 52 But the few conclusions reached here may already be sufficient to show that the most recent methods of linguistic description known as "generative grammar" are in some respects reminiscent of Pfi.nini's methods and system. This concurrence may enable us to judge P~.nini in the light of contemporary linguistic standards. Moreover, it may improve our Sanskrit grammars. But it may also pave the way for further developments in linguistics, prompted and inspired by Pgn. ini and the Indian grammarians. 5s

51 See above note 25. 52 [Added in the proof.] Cf. R. B. Lees, "Turkish nominalizations and a problem of ellipsis", Foundations of Language, 1 (1965), 112-21. s3 I am very grateful to Professors J. C. Heesterman and A. L. Vos and to Dr. K. Kunjunni Raja for helpful criticism of contents and presentation of an earlier version of this paper.

You might also like