THE AUTONOMY OF WORD FORMATION: EVIDENCE FROM CLASSICAL SANSKRIT

Brendan S. Gillon McGill University

Introduction

Linguists such as Selkirk (1982) and Di Sciullo and Williams (1987) are of the view that lexical syntax (that is, morphology) and phrasal syntax are distinct sub-theories of grammar, each with its own primitives and principles governing well-formedness. This view, I believe, is especially enlightening with respect to Classical Sanskrit. The division of syntax into a syntax for word formation, on the one hand, and a syntax for phrase and clause formation, on the other, results in an elegant and compelling analysis of each. Here I show that compounding and derivation in Classical Sanskrit yield to a simple analysis by context free rules. Moreover, the simplicity of this analysis is not gained at the expense of complexity in the analysis of Classical Sanskrits phrase structure, for it too yields to a simple analysis, but in terms of syntactic devices distinct from those used here to analyze Classical Sanskrits lexical structure. 1 The undertaking of this paper is organized as follows. Presented at the very outset is the prima facie evidence for the application of context free rules to derivational morphology and compound formation in Classical Sanskrit. Immediately thereafter, I set about a detailed substantiation of the claim. To do this, I rst give an abridged presentation of the analysis and classication of Classical Sanskrit compounds found in the As adhy ay , the earliest extant grammar of .t . Sanskrit, written or compiled by the great Indian grammarian, P an . ini. Then I show how the details captured in P an inis analysis all fall within the ambit of . the application of context free rules of the sort proposed by both Selkirk (1982) and Di Sciullo and Williams (1987). From this, a remarkable fact emerges: the Sanskrit and English compound structures are seen to be nearly identical. Finally, I show that the very same rules accurately characterize Sanskrit derivational morphology. In short, the context free rules which Selkirk (1982) showed to characterize both compound formation and derivational morphology in English characterize both compound formation and derivational morphology in Sanskrit, a result to be hoped for by those who see grammars as variations upon a universal theme. In spite of the fact that the lexical structures found in the two languages are virtually identical, many compounds of Classical Sanskrit cannot be rendered
1 For

substantiation of this claim, see Gillon 1991 and 1992.

into English by simple calque translation. In many cases, no doubt, lexical idiosyncracies bar the way. But, there are deeper grammatical reasons as well, one of which is brought to light by the approach to lexical syntax adopted here. It turns out that a number of generalizations which hold for English compounds do not hold for Classical Sanskrit compounds; and their dierence over these generalizations can be traced to a dierence in conditions on the transmission of argument structure from a compounds immediate constituents to the compound itself. Pinning down this dierence comprises the nal portion of this paper.

Separating Lexical and Phrasal Syntax: The Prima Facie Case

The prima facie case for the separation of syntax into lexical and phrasal is based on the following regularities evinced by word formation in Sanskrit. 1. The accentuation of a compound is that of a simple word, not that of a phrase (A 6.1.158; MBh to A 2.1.1, i.e., Kielhorn (ed) 1880, v. I, pp. 362.8-9).2 2. Compounds are subject to the inectional and derivational morphological forms of simple words (A 2.4.71; A 6.3.1; MBh to A 2.1.1, i.e., Kielhorn (ed) 1880, v. I, p. 362.5; and Cardona 1988, pp. 264-265). In particular, inection occurs at the end of compounds, not within them; derivational suxes can be added as easily to compounds as they can be to words. 3. Compounds are of unbounded complexity. 4. Constituents of a compound, unlike constituents of phrases, have a xed linear order (A 2.2.30; MBh to A 2.1.1, i.e., Kielhorn (ed) 1880, v. I, p. 362.8; Cardona 1988, pp. 261-264). In general, whereas no two immediate constituents of a compound can be transposed and the sense of the compound retained for its members; any two immediate constituents of a phrase can be transposed and the sense of the phrase retained. 5. A compound is usually analyzable into two immediate constituents (A 2.1.4); and if there is a head, it is the second immediate constituent (A 1.2.43; A 2.2.30; Cardona 1988, pp. 261-263). 6. Inected words, which are external to a compound, are not construed with uninected constituents subordinate within it (MBh to A 2.1.1). 7. A compound has a typical, and for P an . ini, a canonical, phrasal paraphrase (vigraha-v akya) such that, if a compound has the form (C D)i then
2A

key to the abbreviated references is furnished at the end of the article.

its phrasal paraphrase has the form C j Di (where i and j denote one of the seven Sanskrit cases). Moreover, the head of a canonical phrasal paraphrase is the head of the compound being paraphrased.3 In short, the lexical structures are hierarchical, binary, right headed, of unbounded length, integrating compounding and derivational axation without inectional axation occurring on subordinate constituents within the structure: that is, these are precisely the kind of structures which one would expect the context free rules of the sort given by Selkirk (1982) for English would generate. To be sure, there are exceptions to these regularities, but happily they are not productive. For example, some compounds, such as conjunctive compounds (dvandva compounds) formed from the names of gods, do not have the accent of simple words. Moreover, there are cases where an inectional ax occurs on a subordinate constituent within a compound (A 6.3.7-8), so-called aluk compounds, or within lexical derivation (A 6.3.17). However, these cases are not considered productive by any Sanskritist and they are best treated as items to be listed in the lexicon. (For examples and discussion, see Cardona 1988, pp. 264265 and Whitney 1881, 1250.) Finally, while the linear order for compounds is xed, there are a few cases where a compounds constituents are transposable without disturbing the meaning of the compound. Closer inspection shows that these are cases where two distinct derivations happen to yield two compounds with the same meaning.

P an . inis Analysis of Sanskrit Compound Formation

These regularities are captured in the As adhy ay , the rst complete extant .t . grammar of Sanskrit, composed by P an . ini, the great Indian grammarian, in the sixth century B. C. His approach to the grammar of Sanskrit compounds anticipates by 2,500 years, that used by Robert Lees in his well-known work, The Grammar of English Nominalizations (Lees 1960). The approach is to pair a compound with a canonical phrasal paraphrase (vigraha-v akya) with which the compound shares a common derivational ancestor. P an . ini recognized four principal classes of compounds: adverbial (avyay bh ava), conjunctive (dvandva), determinative (tatpurus a ), and exocentric ( bahu. vr hi) compounds.4 This classication is grounded, to a large extent, in the
3 Special notation used here and later on is explained in a second key at the end of the article. This key includes an enumeration of the Sanskrit cases. 4 The English terms used here are intended, not as translations of the Sanskrit terms, but as suggestive of the classes of compounds picked out by the Sanskrit terms. I have chosen standard renditions of the Sanskrit terms, except either where there is none, as in the case

phrasal syntax of the canonical phrasal paraphrases of compounds. An abbreviated presentation of this classication is given next. Subclassication within these four classes and idiosyncracies associated with them will be introduced as needed in the discussion to follow. Adverbial (avyay bh ava) compounds are ones which appear with an adverbial ax typically, identical to the inectional ax for neuter, singular of the second, or accusative, case (Speijer 1886, 55). Moreover, their function is adverbial, much as prepositional phrases in English may function adverbially. The rst constituent in such a compound is, for the most part5 , a preposition and the second a noun. Its canonical phrasal paraphrase comprises the preposition, appearing in the same form as it has in the compound (prepositions have no inectional axation), and the second word, appearing in the case governed by the preposition, often dierent from the second, or accusative, case. In other words, an adverbial compound has the form (Adv (P C ) (N D ) ) and its canonical phrasal paraphrase has the form [P P C [N Pi D ] ] (where i denotes the case governed by C).6 (8) Compound: Analysis: Paraphrase: Adverbial (avyay bh ava) Compound: uparibh umi (above ground) (MW: sv) umi)) (Adv (P upari )(N bh (Adv (P above )(N ground )) [P P upari [N P6 bh umeh . ]] above ground above the ground

Conjunctive (dvandva) compounds are ones comprising two or more nouns whose canonical phrasal paraphrase comprises a conjunction of just the nouns in the compound itself, each with its appropriate inectional ax and all in the same case. Thus, if the conjunctive compound has the form (N (N A ) + (Ni B ) ), then its canonical phrasal paraphrase has the form
of adverbial (avyay bh ava) compounds, or where the rendition is misleading, as in the case of conjunctive (dvandva) compounds, misleadingly called copulative compounds. 5 A few adverbial compounds are formed, not with prepositions, but with the bound prexes su-, sa-, dur-, and nis-. These, as well as those formed with the adverb yath a, require another analysis. 6 In the absence of a larger context, i defaults to the rst, or nominative, case.

[N P [N Pi A ca ][N Pi B ca ]. (9) Compound: Analysis: Paraphrase: Conjunctive (dvandva) Compound r amakr .s .n . au (*Rama+Krishna; cf., Alsace+Lorraine) ama) + (N kr (N (N r .s .n . au)) (N (N Rama) + (N Krishna)) [ [N P1 r amah ca ] ] . ca ] [N P1 kr .s .n . ah . Rama and Krishna and Rama and Krishna and Rama and Krishna

Determinative (tatpurus . a) compounds are ones comprising two constituents, whose canonical phrasal paraphrase consists in a phrase, whose head is the second constituent of the compound and whose case is that of the compound, and whose sole complement is the rst constituent appearing in a case appropriate for a subordinate constituent. In other words, if a determinative compound has the form ( (X C ) (Yi D ) ), then its canonical phrasal paraphrase has the form [Y Pi [XPj C ] D ]. (where i and j denote one of the seven Sanskrit cases and X and Y are either N or A). Examples of such compounds are provided below: (10) Compound: Analysis: Paraphrase: Determinative (tatpurus . a) Compound asikalahah . (sword-ght) ((N asi)(N1 kalahah . )) ((N sword)(N ght)) [N P1 [N P3 asin a ] kalahah . ] with sword ght a ght with a sword Determinative (tatpurus . a) Compound khudam a sitah (rat-bitten) . ((N akhu)(A1 dam sitah . )) ((N rat)(A bitten)) [AP 1 [N P3 akhun a ] dam sitah . ] by a rat bitten bitten by a rat

(11) Compound: Analysis: Paraphrase:

There is a special class of determinative compound called descriptive (karmadh araya) compounds. A descriptive compound is distinguished by the fact 5

that, in its canonical phrasal paraphrase, both constituents appear in the same case. In light of this, one would expect that the paraphrase would contain no supplemental constituents; that is, one would expect what is displayed below. (12) Compound: Analysis: Paraphrase: Descriptive (karmadh araya) Compound d rghakan .t . hah . (long-neck; cf., blackbird) ((A d rgha)(N1 kan .t . hah . )) ((N long)(N neck)) [N P1 [AP 1 d rghah . ] kan .t . hah . ] long neck a long neck Descriptive (karmadh araya) Compound d rghakan t hah (long-neck; cf., blackbird) .. . ((A d rgha)(N1 kan t hah )) .. . ((N long)(N neck)) [ [AP 1 d rghah . ca ] sah . [N P1 kan .t . hah . ca ] ] long and neck and something being both long and a neck.

However, the tradition adopts another paraphrase, illustrated as follows. (13) Compound: Analysis: Paraphrase:

Here, I shall adopt the former, instead of the latter, as the canonical phrasal paraphrase. This departure from the indigenous tradition is justied, since it provides a better characterization of the data. First, the proposed replacement paraphrase is more nearly synonymous with that of which it is a paraphrase than the traditional one. Specically, the traditional paraphrase is inaccurate for any descriptive compound where the rst constituent is a subsective adjective. Thus, something which is both small and an elephant is not generally the same as something which is a small elephant, since what is small for things in general is not what is small for elephants in particular. (See Chierchia and McConnellGinet 1990, ch. 8.3 for discussion.) Second, the predominant constituent of the proposed replacement paraphrase corresponds to the predominant constituent of the compound paraphrased, whereas the traditional Indian canonical phrasal paraphrase has no predominant constituent. Specically, the predominant constituent of the compound in the example above is kan tha (neck), but the neither .. kan t ha ( neck ) nor d rgha ( long ) can be said to be predominant in the traditional .. paraphrase, since no conjunct is more predominant than any other in a conjunction. Neither of these aws is attributable to the canonical phrasal paraphrase of the descriptive compound found in (12), and so such a paraphrase is adopted here as the canonical one. An exocentric (bahuvr hi) compound is one which is construed as modifying a constituent external to itself in the same way that an adjective modies a noun. Internally, one constituent is subordinate to the other. Its canonical phrasal paraphrase is a relative clause. So, if an exocentric compound has the form 6

( (X C ) (N D ) ), then its canonical phrasal paraphrase has the form [RC [V P V [XP1 C ] ][N P1 D ] yati (where X is either A or N; V is a suitable form of one of the verbs to be; and yati is a relative pronoun in some case other than the nominative, that is, where i is not 1.) (14) Compound: Analysis: Paraphrase: Exocentric (bahuvr hi) Compound: samacittah . (even-minded) ((A sama)(N cittah . ))1 ((N even)(N mind )) [RC [V P (asti) [AP 1 samam ] [N P1 cittam ] yasya ] (is) even mind whose whose mind is even

Context Free Rule Analysis of Sanskrit Compound Formation

With this sketch of the fundamental traditional Indian classication of Sanskrit compounds, I rst show how determinative (tatpurus . a) compounds are characterized by context free rules which expand one of the four major lexical categories (A, N, P, V) into a pair of such categories.7 (15.1) A A A (vi ses araya): . an . a-ubhaya-pada-karmadh sn at anulipta: bathed and oiled (A (A sn ata )+(A anulipta ) ) bathed oiled A A A (vi ses araya): . an . a-ubhaya-pada-karmadh tulya sveta: equally white (A (A tulya )(A sveta )) same white N A N (vi ses urva-pada-karmadh araya): . an . a-p n lotpala: blue lotus la )(N utpala ) ) (N (A n blue lotus

(15.2)

(16)

7 Provided in parentheses, here and below, are the traditional Indian grammatical terms for smallest class of compound to which the example in question belongs.

(17)

V A V (cvi-gati): laghukr . : make light )) (V (A laghu )(V kr . light make A N A (upam ana-p urva-pada-karmadh araya): analos n a: re-hot (cp. ice-cold) .. (A (N anala )(A us .n .a ) ) re hot A N A (vibhakti-tatpurus . a): gohita: benecial to cows hita )) (A (N go )(A cow benecial N N N (vibhakti-tatpurus . a): dadhyodana: rice with curds (N (N dadhi )(N odana ) ) curd rice N N N (upapada-tatpurus . a): kumbhak ara: pot maker ara ) ) (N (N kumbha )(N -k pot -maker N N N (avadh arana-p urva-pada-karmadh araya): r ajars . i: sage-king (N (N r aja )(N r .s .i ) ) king sage N N N (upam ana-uttara-pada-karmadh araya): purus avy a ghra: tiger of a man . aghra ) ) (N (N purus . a )(N vy man tiger V N V (cvi-gati): sren u: to line up . ibh (V (N sren u )) . i )(V bh line be A P A (pr adi-tatpurus . a): atimadhura: very sweet )(A madhura ) ) (A (P ati beyond sweet

(18.1)

(18.2)

(19.1)

(19.2)

(19.3)

(19.4)

(20)

(21)

(22)

N P N (pr adi-tatpurus . a): adhir aja: over-lord aja ) ) (N (P adhi )(N r over king V P V: adhi s : to lie upon (V (P adhi )(V s )) upon lie

(23)

There are two subclasses of determinative (tatpurus . a) compound Indian grammarians single out as special. They are upapada-tatpurus adi. a and pr tatpurus a compounds. The former comprises determinative compounds whose . nal constituents are bound forms derived from verbal roots, such as -j na (knower), -k ara (doer or maker), and -dhara (bearer), to mention a few. The latter comprises determinative compounds whose initial constituents are forms taken from a special list which begins with the preposition pra. This list, from the point of view being pursued here, is heterogeneous: it includes, on the one hand, various prepositions, and it includes, on the other, various initially occurring bound forms such as su- (good/well) and ku- (bad/badly). On the approach here, nothing special needs to be said concerning either upapadatatpurus adi-tatpurus . a compounds or those pr . a compounds with initially occurring bound forms, beyond what must be stipulated for any lexical entry which is a bound form, namely, that it has a certain subcategorization frame. Thus, for example, -j na (knower) includes in its lexical entry, among other things, the subcategorization frame [ N ], while su (good/well) includes the subcategorization frame [ N/A ]. The pr adi-tatpurus . a compounds whose initial constituents are prepositions are identied by the Indian grammatical tradition as ones with a verbal root deleted (dhatu-lopa). They are so identied because deletion is an operation invoked in the derivation of the compound. The compound is analyzed as comprising two constituents: the rst is a perfect passive participle of a verb to which the preposition is prexed, and the second is the second constituent of the compound itself. The compound results from the deletion of the verbal root. Consider, now, the context free rules for English compound formation found given by Selkirk (1982, p. 16): A A A *A AA NA PA VA N N N N AN NN PN VN *P *P *P *P AP NP PP VP *V *V V *V AV NV PV VV

Table 1

These dier from those for Sanskrit compound formation found above in the following way: on the one hand, Sanskrit has the rules V A V and V N V, which Selkirk says are absent from English; on the other hand, English is said by Selkirk to have the rule N V N, which is absent from Sanskrit. Are these dierences merely apparent? Consider rst compounds of the form N V N. They are extremely rare in Sanskrit. And though they are common in English, the class itself is heterogeneous and productivity diers from subclass to subclass. There are three subclasses in English, identiable by their phrasal paraphrases. The rst subclass of English compound of the form N V N comprises ones paraphrasable either as that which Vs an N or as one who Vs an N. Examples of this kind of compound include tug-boat, turn-key, pick-pocket, break-water, and spoil-sport (Marchand 1969, ch. 2, sec. 15). Though the form of this compound has numerous instances, it does not appear to characterize spontaneously produced compounds (Bauer 1983, ch. 7.2.1.2). While compounds of this subclass occurred in Vedic Sanskrit8 , I know of only one such instance in Sanskrit.9 The second subclass comprises those which are paraphrasable as an N which or who Vs. They include not only ones mentioned by Selkirk (1982, p. 14), such as rattle-snake and scrub-woman, but also many others such as glow-worm, and stop-watch. The third subclass comprises compounds paraphrasable as n N for V-ing. They include swearword, whetstone, think tank, rap session, and washroom. These last two subclasses appear to be completely absent from Sanskrit.10 Let us now consider compounds of the form V A V and V N V. There is no doubt that these rules are extremely productive in Sanskrit, though only three verbs can participate in such formations, namely, the verb kr . (to do), the verb bh u (to be), and the verb as (to be). There is also no doubt that compounds of this form obtain in English, as Selkirk herself points out. Thus, for example, dry-clean has the form V A V and hand-carry has the form V N V. Now, Selkirk (1982, p. 17), following Marchand (1969, ch. 2, sec. 37-38.), regards such compounds as backformations from compounds of the form A A A (dry8 Whitney (1881, 1309) reports that the Vedic literature, and primarily the early Vedic literature, contains around thirty examples of compounds whose rst member is a present participle in its stem form and whose second member is a noun functioning as the participles complement, typically, its direct object. 9 The term jahat-sv artha, found in Pata njalis Mah abh a. sya to A 2.1.1 (Kielhorn (ed) 1880, v. I, p. 364.6 ), is of such a form. There, Pata njali distinguishes two views on the status of the denotation of subordinate constituents in compounds: one view holds that compound formation (vr . tti) is such that the subordinate constituent loses its own denotation (i.e., jahat-sv arth a vr . ttih . ); and the other holds that compound formation (vr . tti) is such that the subordinate constituent does not loses its own denotation (i.e., a-jahat-sv arth a). 10 In fact, neither subclass may exist in English either, for as Bauer (1983, ch. 7.2.1.2) points out, they might be analyzable as N N N.

10

cleaned) and of the form A N A (hand-carried) respectively, and hence as not truly satisfying the context free rules of V A V and V N V. Even if these compounds came into English as backformations, the fact remains that new compounds of this form can be created spontaneously: for example, to screen preview, to tax reduce, to peer review, to mention but a few. Indeed, they seem to abound in technical and journalistic writing. Brought within the purview of context free rules are two of the four major classes of Sanskrit compounds: determinative (tatpurus . a) compounds and conjunctive (dvandva) compounds. The question arises: Is the near coincidence of Sanskrit and English compound formation rules seen above spoiled by the two other kinds of Sanskrit compounds not yet considered, namely exocentric (bahuvr hi) and adverbial (avyay bh ava) compounds? I believe not, as I intend to show below.

Exocentric (Bahuvr hi) Compounds

The classical Indian grammatical tradition identies a number of dierent kinds of exocentric (bahuvr hi) compounds. They include: privative exocentric (na nbahuvr hi) compounds, comitative exocentric (saha-bahuvr hi compounds, prepositional exocentric (pr adi-bahuvr hi) compounds, homo-denotative exocentric (sam an adhikaran hi) compounds, and hetero-denotative exocentric (vy. a-bahuvr adhikaran hi) compounds. . a-bahuvr Homo-denotative exocentric compounds, illustrated in (14) above, are the most important subclass among the exocentric compounds. Below, I shall show that, under the approach to word formation adopted here, these compounds receive a simple analysis. Next, I shall show that this simple analysis extends straightforwardly to the other productive subclasses. At the same time, I shall show that English too has productive forms of exocentric compounds, each of which corresponds to one of the subclassses of exocentric compounds found in Sanskrit, and that the English forms of exocentric compounds yield to the same analysis as their counterparts in Sanskrit, thereby upholding the nearly perfect co-incidence between English and Sanskrit compound formation.

11

5.1

Homo-denotative Exocentric Compounds

Homo-denotative exocentric (sam an adhikaran hi)11 compounds are so. a-bahuvr called because, in their canonical phrasal paraphrase, the two principal constituents comprising the compound are placed into a relative clause in which the rst constituent of the compound is predicated of the second, and they thereby share the rst, or nominative, case and if the rst constituent is an adjective, they agree in number and gender as well. Notice that the relationship of agreement in case, and where possible number and gender as well, is true of the canonical phrasal paraphrase not only of exocentric compounds, but also of descriptive compounds. Finally, it was observed earlier that exocentric compounds are so-called because they modify other constituents much in the way adjectives modify nouns. Putting these observations together, one arrives at the natural hypothesis that exocentric compounds are derived from descriptive compounds by zero axation which converts a descriptive compound into an adjective. The evidence that homo-denotative exocentric (sam an adhikaran hi) . a-bahuvr compounds are adjectives is that they have all the properties adjectives in Sanskrit have. First, adjectives in Sanskrit, like those in Latin, agree with the nouns they modify in case, number, and gender. Consider the adjective t ks .n .a (sharp). If it modies a noun in the nominative, singular, masculine, say asih . (sword), then it has the form t ks .n . ah . ; and if it modies a noun in the nominative, singular, feminine, say chur (knife), then it has the form t ks a; and .n . if it modies a noun in the nominative, singular, neuter, say patram (blade), then it has the form t ks rgha-kan tha. If .n . am. Now consider the compound d .. it is to be construed with a masculine, singular noun in the nominative case, say purus . ah . (man), to yield the sense long-necked man, then the compound must have the nominative, masculine, singular form, namely, d rgha-kan thah .. .. If it is to be construed with a feminine, nominative, singular noun, say str (woman), to yield the sense long-necked woman, then the compound must have the feminine, nominative, singular form, d rgha-kan th a. And nally, if it is to .. be construed with a neuter, nominative, singular noun, say mitram (friend), to yield the sense long-necked friend, then the compound must have the neuter, nominative, singular form, d rgha-kan tham. .. Next, adjectives in Sanskrit can be turned into abstract nouns by the afxation of the sux -tva (-ness): for example, the adjective kr sa (thin) may .
(s am an adhikaran . ya) is the counterpart in the Indian grammatical tradition of the Western technical notion of concord or agreement. As will be elaborated below, adjectives which modify nouns in Sanskrit agree with the nouns in case, number and gender. The concord is seen by the traditional Indian grammarian as accruing to the fact that a noun and an adjective modifying it have the same denotation (sam an adhikaran . a). A homodenotative exocentric compound is one in which the two principal overt constituents denote the same thing, that is, they appear in the same case, and if the rst principal overt constituent is an adjective, they agree in case, number, and gender.
11 Homo-denotation

12

be converted into the abstract noun, kr sa-tva (thin-ness). Exocentric bahuvr hi . compounds are susceptible of the same conversion: for example, d rgha-kan tha .. (long-neck-ed; cf., level-head-ed) be turned into d rgha-kan tha-tva (long-neck-ed.. ness; cf., level-head-ed-ness). Moreover, just as an adjective such as kr sah . . (thin) can function, as its English translation can, as a common noun, meaning the same thing as its English nominal counterpart, the thin, so too should an exocentric (bahuvr hi) compound be liable to function as a common noun. And this too is true, as observed by Speijer (1886, 222, fn. 1) and as exemplied by following compound and its commentarial gloss. (24.1) (24.2) NBT 48.4 (vyutpannasamketasya) NBTP 49.1-2 n atah yena sah vyutpannah . j . samketah . . arisen known convention by whom he One by whom the conventions of language are known (j n ata (known) glosses vyutpanna (arisen).)

There is independent conrmation that homo-denotative exocentric (sam anaadhikaran hi) compounds are best treated as descriptive (karmadh araya) . a-bahuvr compounds to which a phonetically null, possessive, adjectival sux (symbolized hence forth with B) is axed. Sanskrit has a phonetically overt, possessive, adjectival sux -ka which is virtually synonymous with the phonetically null one just hypothesized. Though their distributions are somewhat dierent (A 5.4.151 .), nonetheless, they overlap to such an extent that commentators to a text in which an exocentric (bahuvr hi) compound occurs frequently repeat the compound, adding the -ka sux to signal the fact that the compound in question is to be construed, not as a descriptive (karmadh araya) compound, but as an exocentric (bahuvr hi) compound (Boose and Tubb 1981, ch. 5, sec. 15). English too has homo-denotative exocentric compounds. By and large, they are marked by the adjectival, possessive sux, -ed. These English compounds, exemplied by such compounds as longlegged, literal-minded, and two-footed, have a distribution narrower than that of its counterpart in Sanskrit a fact which will be dilated on below. Not every English homo-denotative exocentric compound has the -ed sufx. In particular, English homo-denotative exocentric compounds which serve as proper names or epithets seem to require a phonetically null counterpart to the -ed sux.12 Examples of proper names are particularly common in childrens stories. For example, in the childrens movie, Land Before Time, the two
12 It is interesting to note in this connection that the Sanskrit sux -ka, used to mark phonetically a bahuvr hi compound, is said by P an . ini (A 5.4.155) to be prohibited from axation to bahuvr hi compounds which serve as names.

13

dinosaurs which are the main characters are named, big foot and long neck, instead of big-footed and long-necked. Examples of epithets are such compounds as red-head, dim-wit, hard-back, etc., to which there correspond red-headed, dimwitted, hard-backed, and so forth. (See Marchand 1969, ch. 2, sec. 18 for other examples.) Moreover, it seems that the -ed sux and its phonetically null counterpart are in free variation in exocentric compounds which are initial constituents in larger compounds: long-necked bottle plant and long neck bottle plant both denote plants for bottles whose necks are long. Another parallel between English and Sanskrit homo-denotative exocentric (sam ana-adhikaran hi) compounds is the predication relation in the . a-bahuvr canonical paraphrase may be metaphorical, instead of literal. Thus, in the compounds candra-mukha (moon-faced), sth ula-caran ti . a (club-footed), and ayo-mus .. iron-sted, a face (mukha) is likened unto a moon (candra), a foot (caran a ) unto . a club (sth ula), and a st (mus ti) unto iron (ayas). ..

5.2

Hetero-denotative Exocentric Compounds

It was mentioned above that Sanskrit also has hetero-denotative exocentric (vyadhikaran hi) compounds. They are so called because the two overt . a-bahuvr constituents of the compound appear in dierent cases in their canonical phrasal paraphrases. These hetero-denotative exocentric compounds are of two kinds, depending on the case of one of the constituents in the canonical phrasal paraphrase. In one case, the compound, whose form is (N C)-(N D), yields the paraphrase one who has a C in his D: for example, sastra-pan . i (who has a sword in his hand), a sru-kan tha (who has tears in his throat), etc. Here, the sec.. ond constituent, D, appears in the seventh, or locative, case in the compounds canonical phrasal paraphrase. Whitney (1881, 1303) reports that these compounds are conned to ones in which D denotes a part of the body; indeed, they seem to be conned to a few xed expressions. In the other case, the compound, whose form is also (N C)-(N D), yields the paraphrase one whose D is like that of a C: for example, us tra-mukha (camel-faced, i.e., having a face like a .. camels). Here, the rst constituent, C, appears in the sixth or genitive case in its canonical phrasal paraphrase. These latter, all of which express metaphors, are, in fact, quite productive. While English does not have the rst kind of hetero-denotative exocentric compound, it has no shortage of the second. Examples include camel-faced, duck-billed, and pot-bellied. Moreover, these compounds, whether in English or in Sanskrit, express metaphors. Though the canonical phrasal paraphrases of productive hetero-denotative exocentric compounds, be they English ones or Sanskrit ones, do not result in the pair of lexical constituents of the compound sharing the rst, or nominative, case, nonetheless, the constituents in the paraphrase are in a modicational or

14

appositional relation. Consider the example us tra-mukhah .. . (camel-faced). Its paraphrase is us trasya iva mukham yasya (one whose face is like a camels). .. mukham (face) is in the nominative case while us trasya (camels) is in the sixth .. or genitive. Notice that both English and Sanskrit tolerate the omission of the head noun in a predicated noun phrase immediately containing a genitive noun phrase, as shown below. (25) tasya mukham us .t . rasya (mukham) iva (vartate) . his face camels (face) like (is) his face is like a camels (face).

And it is this form of predication which surfaces in paraphrases of heterodenotative exocentric compounds. But it is also predication which surfaces in canonical paraphrases of homo-denotative exocentric compounds: for example, d rgha-kan tah rghah thah .. . (long-necked) is paraphrased as d . kan .. . yasya (one whose neck is long); and ayo-mus t i ( iron-sted ) is paraphrased as ayah tih .. . iva mus .. . yasya (one whose st is like iron).

5.3

Other Exocentric Compounds

Three classes of exocentric bahuvr hi compounds remain to be discussed: prepositional exocentric (pr adi-bahuvr hi) compounds, comitative exocentric (sahabahuvr hi) compounds, and privative exocentric (na n-bahuvr hi) compounds. Prepositional exocentric (pr adi-bahuvr hi) compounds are exocentric compounds whose rst constituent is either a preposition or one of several initially occurring bound forms. In their canonical phrasal paraphrase, the initial constituents, which as such cannot appear as predicates, appear in a derivationally related word which can appear as a predicate. (26.1) Compound: Analysis: Paraphrase: Prepositional Exocentric Compound: dur atm a (evil-natured) atm a ))1 ((A dur)(N ((N evil)(N nature)) [RC [V P (asti) [AP 1 dus atm a ] yasya ] .t . ah . ] [N P1 (is) evil nature whose whose nature is evil

15

(26.2) Compound: Analysis: Paraphrase:

Prepositional Exocentric Compound: adhomukhah . (?down-faced; cf., down-hearted) ((A adhas)(N mukhah . )) ((N down )(N faced)) [RC [V P (asti) [AP 1 adhogatam ] (is) down-turned [N P1 mukham ] yasya ] face whose whose face is down-turned

In the rst example above, the bound adjectival form dur- (evil) is rendered in the paraphrase by the synonymous free form dus ta (evil), while in the sec.. ond example, the preposition adhas (down) is rendered by the perfect passive participle of the verb adhogam (to turn downward). Interestingly, similar exocentric compounds appear in English. Thus, for example, English has the bound adjectival form multi-, and it is used to form the exocentric compound multi-valued, synonymous with many-valued. And English may have prepositions occurring as initial constituents in such exocentric compounds as o-colored (Aklujkar 1992, v. 1A, 34.8) and down-hearted. However, on the basis of their canonical phrasal paraphrases, one whose heart is down and that whose color is o, the compounds initial constituents might be better analyzed as adjectives. Indeed, precisely such an analysis was proposed by the lost P an agas. (See Deshpande 1985, pp. . inian commentarial tradition of the Saun 52-53 for discussion.) Be that as it may, the pattern here is the same as the one found in homo-denotative exocentric compounds: the rst constituent bears a modicational relation to the second. Still another subclass of exocentric compound is the privative bahuvr hi exocentric compound, formed by prexing the privative prex a- to a noun. This prex has two functions. One is to create a word whose denotation is the complement of the denotation of the word to which it is prexed. Thus, for example, the adjective s adhu means good while the privative adjective a-s adhu, derived from it, means not good, or even bad. Similarly, the noun br ahman . a means brahmin, the noun a-br ahman . a, derived from it, means non-brahmin. In this function, the privative prex is analogous to the English privative prexes of non- and un-. The other function of a- is to create a word which denotes someone or something deprived of what the word to which it is applied denotes. For example, the word for a wife, bh ary a, when prexed with a, may mean, not a non-wife, but rather someone who is without a wife. In this function, a- is analogous to the English privative sux -less. The last subclass of exocentric compounds are comitative exocentric (sahabahuvr hi) compounds. They result when the initial constituent of the exocentric compound is the prex sa-. Thus, sa- can be prexed to the neuter noun sma sru (beard) to yield an adjective, which, when used to modify a masculine noun such

16

as pum an (man), will agree with it in case, number, and gender, sa- sma sruh . pum an (a bearded man). Unlike other exocentric compounds, its traditional canonical phrasal paraphrase does not comprise a relative clause. Instead, it is paraphrased by removing the prex sa-, putting the noun which is the second constituent of the compound in the third (instrumental) case and supplying the preposition saha, which requires a noun in the third (instrumental) case, to govern the noun. Thus, for example, sa- sma sruh an can be paraphrased as . pum (saha sma srun a pum an). English has the equivalent of these Sanskrit comitative exocentric compounds. Besides the English rendition of the Sanskrit example just discussed, there are these: armed, crested, feathered, pointed, principled, and winged. However, this formation has a dierent distribution from its Sanskrit counterpart: sa-putr a str (*sonned woman: woman with a son) is acceptable in Sanskrit, but its English calque sonned woman is not. It must be said that the use of the -ed sux in the formation of these words is not as free as its use in the formation of homo-denotative exocentric compounds.13 K aty ayana, the rst great commentator on P an adhy ay , thought . inis As .t . the last two subclasses of exocentric compounds should be assimilated to homodenotative exocentric compounds, treating as special bound adjectives the prexes sa- and a-. (See v arttikas 12 through 15 to A 2.2.24; see Deshpande 1985 pp. 49-51 for discussion.) In sum, exocentric (bahuvr hi) compound formation can be seen as a case of suxation. In Sanskrit, the phonetically covert sux applies to a stem Z if and only if the stem Z is of the form C-D where either (i) C modies, or is in apposition with, D or (ii) C is the privative prex a- or the prex of accompaniment sa-. Of course, this condition on the sux does not account for seventh case hetero-denotative bahuvr hi compounds. But, these are probably best treated as listed in the lexicon. In English, the adjectival sux -ed applies to a stem Z if and only if the stem Z is simple or it is of the form C-D where C modies D. Thus, productive exocentric compound formation for each language nearly coincides, consisting primarily in the addition of a sux to compounds satisfying the same semantic relation.

Adverbial (Avyay bh ava) Compounds

Adverbial (avyay bh ava) compounds present their own challenge to the analysis of compound formation undertaken here and to the claim made earlier that there is a near coincidence between the rules needed for Sanskrit and those for
(1975) and Ljung (1976) have suggested that what I have called comitative exocentric compounds are sensitive to inalienable possession in English. Thus, an armed man could not be a man with upper body limbs, which are presumably inalienable as far as human beings are concerned, but must be a man with a weapon, which is presumably alienable.
13 Hudson

17

English. Recall that adverbial compounds function as adverbs and so, prima facie at least, have the form (Adv (P C)-(N D)), which requires a formation rule of the form Adv P N. But, no formation rule discussed so far uses the lexical category of adverb. This raises the more general question of whether or not the context free rules needed to analyze compound formation in Sanskrit require, beyond the four basic lexical categories, the category of adverb. This more general question is best pursued in two steps. The rst step is to determine whether or not the rst constituent of a compound is an adverb; and the second is to determine whether or not compound formation itself creates adverbs (without the further assistance of a derivational rule). Here are some compounds whose initial constituents are susceptible of an adverbial interpretation: (27.1) PSED: sv nisarganipun .a nature-gloss naturally clever PSED: sv m asadeya month-payable payable in a month Coulson 1976, p. 189 prayatnapreks ya . an eort-discernible discernible with eort.

(27.2)

(27.3)

In spite of what the English translations of these compounds might suggest, there are no adverbs here. This is evident from two facts: rst, the initial constituent of each compound is a noun; and second, the compounds themselves have canonical phrasal paraphrases in which the subordinate member appears in an oblique case. These paraphrases are: [N P3 nisargena ] nipun . ah . (clever by nature), [N P3 m asena ] deyah ( payable in a month ), and [ prayatnena ] N P . 3 preks an yah ( to be discerned with eort ). . . More problematic are compounds whose initial constituents fall into the category of adjective and yet have an adverbial intepretation. Thus, for example, tulya (same) is an adjective and serves as an adjectival modier in the compound tulyadharma (same dharma). Yet, it also serves as an adverbial modier in such compounds as tulya sveta (equally white). The same holds for the compounds udagraraman ya ( intensely lovely) (Coulson 1976, pp. 92-3) . and madhuraukta (sweet spoken) (Coulson 1976, pp. 188-9). Compounds which have the same constituency and whose initial constituent is an adjective having an adverbial interpretation occur in English as well: for example, fresh-cut-potatoes and soft-spoken-guy. The dierence between San18

skrit and English is that Sanskrit seems to tolerate any class of adjective as the second constituent in compounds whose rst constituent is liable to an adverbial interpretation, whereas English tolerates only present and past participles in such compounds. Thus, while the adjective equal, like its Sanskrit translation tulya, may occur in a compound, for example, equal-opportunity-employer; it may not do so felicitously, if the sister to its right is not a participle, which is borne out by the unacceptability of English compounds such as equal-white. The fact that an initial adjectival constituent in a compound is liable to an adverbial interpretation does not require that this element be an adverb: after all, one might simply retain the syntactic description of a compound of the above sort as (A A A) and consign the treatment of the adverbial interpretation of the rst constituent to that part of the compounds analysis which provides for its interpretation. Such a treatment receives some independent grounding in compounds of the very same form, wherein the rst adjective is construed as a secondary predicate introduced by the second: sr anta- agata (*tired-arrived: arrived tired) (Coulson 1976, p. 93). It is worth observing that this very same option arises in English as well: adjectives appearing as initial constituents in compounds such as nice-sounding and sweet-smelling retain an adjectival interpretation, whereas those appearing as initial constituents in compounds such as soft-spoken man and fresh-cut potatoes acquire an adverbial interpretation. This solution might be baulked at, since there are words which are adverbs, when taken in isolation, and which occur as the initial constituent of a compound, retaining their adverbial interpretation. Such words are those which end in -th a, such as tath a (thus) and anyath a (otherwise), as well as mithy a (in vain), punar (again), evam (so), etc. Compounds of this ilk are these: (28.1) Coulson 1976, p. 93 anyath asambh avana otherwise-supposing supposing otherwise NBTP 49.16; PSED: sv anyath aanupapatti otherwise-impossible being impossible otherwise PSED: sv punarukta again-spoken spoken again PSED: sv tath agata thus-happened happened thus 19

(28.2)

(28.3)

(28.4)

But even these facts are not decisive, since these very same words, while having only an adverbial interpretation while outside of a compound, are also susceptible of an adjectival interpretation within some compounds: (29.1) NBTP 29.11 tath apravr . tti such-adversion such adversion NBTP 29.13 tath avr . tti such-activity such activity NBTP 32.11 tath avr . tti such-meaning such meaning PSED: sv mithy aabhidh ana false-assertion false assertion PSED: sv evampr aya such-kind such kind NBTP 67.5 a-punar avr . tti non-again-return no further return

(29.2)

(29.3)

(29.4)

(29.5)

(29.6)

There would be no need to expand the categories used in the context free rules, at least not on account of these compounds, should one hypothesize that these adverbs have in fact two forms, one free and one bound and that the bound forms are adjectives while the free forms are adverbs. As a result, the bound forms, like all adjectives, are liable to an adverbial interpretation, when they occur in compounds. Finally, it should be remarked that prepositions in determinative prepositional (pr adi-tatpurus . a) compounds behave in a similar way. On the one hand, when the second constituent is an adjective, as in ati-madhura (overly or excessively sweet) or utcan .d . a (*high-violent: highly violent) (Coulson 1976, p. 135), the initially occurring constituent, a preposition, has an adverbial interpretation. The same is attested in English. Compare over-dose (i.e., excessive dose) with over-abundant (i.e., excessively abundant). On the other hand, when the

20

second constituent is a noun, as in atibh ara (over-load: i.e., excessive load), the preposition has an adjectival interpretation. That prepositions in such congurations have an adjectival interpretation is borne out by the fact that these very same compounds can serve as exocentric (bahuvr hi) compounds, where the initial constituent bears a predicational relation to the second element, a relation borne by adjectives and nouns and not by prepositions. It seems, then, that initial constituents in compounds do not require the use of the lexical category of adverb as the leftmost label to the right of the arrow of context free rules for lexical structure. Adverbial (avyay bh ava) compounds, however, have the form and function of adverbs. Perhaps these compounds require the lexical category of adverb to the left of the arrow for lexical context free rules. To explore this possibility, I turn rst to English adverbial compounds. They include: over-head, over-land, over-seas, up-town, up-stairs, up-stream, downtown, down-stairs, down-stream, in-doors, in-land, etc.14 Each is clearly a compound: Common nouns in English require an overt determiner to form a noun phrase, unless they are either plural or mass; but in several instances, the common noun is singular and not mass, and yet no determiner intervenes between it and the preceding preposition. Moreover, each of them may function as an adverb, as shown in the following sentences: (30.1) (30.2) (30.3) The airplane ew overhead. The visitors went uptown to see the movie. The explorers paddled downstream for two kilometers.

Moreover, they should not be confused with English determinative prepositional (pr adi-tatpurus . a) compounds, such as over-dose, which function as nouns and never as adverbs. If such are the facts of adverbial compounds, what is their analysis? A relevant fact here is this: the very English expressions just identied as adverbial compounds can also occur as left sisters in compounds: over-head-compartment, up-stairs-apartment, in-door-pool, etc. This is true of adverbial (avyay bh ava) compounds in Sanskrit as well (Aklujkar 1992, v. 1A, 31.10). Adverbial compounds, then, have the same chameleon character as other adverbs discussed above (see (28) and (29) above): when occurring initially within a compound, they are susceptible of an adjectival interpretation; and when occurring on their own, they are adverbs. The hypothesis I shall adopt here is that adverbial (avyay bh ava) compounds are bound adjectives, of the form (A (P C)-(N D))-, which are turned into adverbs. I should add that each of the components which go into this hypothesis has been previously encountered, namely, the existence
14 English adverbial (avyay bh ava) compounds seem more restricted in their productivity than their counterparts in Sanskrit: not every combination of preposition and noun yields a felicitous English adverbial compound: over-head is acceptable, but over-shoulder is not; up-stairs is acceptable, but up-ladder is not.

21

of bound adjectives and the existence of derivationally complex derived forms. Thus, bound adjectives such ku- (bad/badly) and su- (good/well) are included by the Indian tradition among determinative prepositional pr adi-tatpurus . a compounds; and these bound adjectives are subject to both an adjectival and an adverbial interpretation, depending on whether the second constituent is a noun or an adjective respectively. Moreover, bound forms which are derivationally complex, such as -k ara (doer or maker) and -dhara (bearer), are found as the nal constituents of the special subclass of determinative compounds known as upapada-tatpurus . a compounds. Adverbial (avyay bh ava) compounds do not spoil the near co-incidence of the rules for compound formation in English and in Sanskrit, for English and Sanskrit not only each have such compounds but also yield to the same analysis in terms of context free rules. However, their analysis does require the introduction of a new rule of the form A P N, not envisioned by Selkirk (1982). (See Table 1 above.)

Context Free Rules and Sanskrit Derivational Morphology

An even more striking case for the utility of context free rules in the analysis of word formation and for the claim of the structural anity between word formation in Sanskrit and word formation in English is provided by derivationally complex words. The applicability of lexical context free rules to the analysis of Sanskrit derivation is shown by the examples below. (31) A A A (comparative and superlative adjectives): sucitara: purer (A (A suci )(A -tara ) ) pure -er N A N (abstract nouns): gurutva: heaviness (N (A guru )(N -tva ) ) heavy -ness N A N (adjectival nouns): kr sa: a weak one . sa ) (N ) ) (N (A kr . weak one V A V: (denominative verbs): lohitaya: to redden (V (A lohita ) (V -aya ) ) red -en 22

(32.1)

(32.2)

(33)

(34.1)

A N A (adjectives with possessive sux): dh man: thoughtful ) (A -man ) ) (A (N dh thought -ful A N A (adjectives of appurtenance): saiva: belonging to s va (A (N s va ) (A -a ) ) Shiva N N N (abstract nouns): vr . ks . atva: treeness (N (N vr . ks . a ) (N -tva ) ) tree -ness V N V (denominative verbs): putraya: to treat like a son (V (N putra ) (V -ya ) ) son A V A (participles): kr . ta: made ) (A -ta ) ) (A (V kr . make -en N V N (deverbal nouns): dar sana: sight (N (V dr s ) (N -ana ) ) . see -ing V V V (causative): dar saya: show s ) (V -aya ) ) (V (V dr . see

(34.2)

(35)

(36)

(37)

(38)

(39)

What is remarkable is that these are exactly the rules needed for the analysis of English derivation. (See Selkirk 1982, p. 82.) A A *A A AA NA PA VA N N *N N AN NN PN VN *P *P *P *P AP NP PP VP V V *V V AV NV PV VV

Table 2 Derivational axes have two salient characteristics: rst, that they are bound morphemes; and second, that they frequently change the lexical category 23

of the stems to which they are applied. The second is captured by assigning them lexical categories (Williams 1981), as a result of which derivationally complex words yield, in principle, to an analysis in terms of context free rules; the rst is captured by assigning subcategorization frames to derivational axes lexical entries. Thus, for example, the comparative adjectival sux -tara would ]. (For details, see have included in its lexical entry the following frame: [ A Selkirk 1982, Ch. 3.1.) To conclude, the foregoing examination of the details of derivational morphology and compound formation in Sanskrit shows that they yield to a simple analysis in terms of context free rules and that these rules are virtually the same as the ones used in similar analysis of derivational morphology and compound formation in English.

Dierences between English and Sanskrit Compound Formation

There are four generalizations which I think are considered true of compound formation in general and of English compound formation in particular: rst, proper names do not occur productively within compounds (Di Sciullo and Williams 1987, p. 50); second, pronouns do not occur within compounds; third, proper constituents of compounds do not enter into anaphoric relations; fourth, proper subordinate constituents of compounds are not construed with constituents external to them. Below, I shall show that none of these generalizations is true of Sanskrit, though the last three are certainly true of English. I shall also show that the dierence between English and Sanskrit over the last three can be accounted for in terms of conditions on percolation of argument structure. The evidence for the rst generalization, namely, that proper names are not constituents in English compounds, is the unacceptability of compounds such as *Bill admirer. To begin with, this generalization does not hold for Sanskrit. While the specic generalization is not discussed in the traditional grammatical literature, every Sanskritist knows that proper names do occur freely in compounds and are subject to derivational suxation. Here are some examples gleaned from the classical literature at random. (40.1) V 1.13 (= SG 1.1.1) [N P5 (Kuberabhavan at) ] pratinirvart an a from Kubera-house returning returning from Kuberas house

24

(40.2)

U 4.1.7 (= SG 3.1.11) [GC [V P [N P2 [AP2 (Arundhat puraskr an) . t Arundati-headed [N P6 (mah ar ajaDa sarathasya) d ar an ] adhis aya ]] .t . h king-Da saratha wives having escorted having escorted the wives of King Da saratha, headed by Arundhat Mu 1.19.12 (= SG 21.1.11) [S [V P upavarn antam) ] ] ] . aya [N P2 (Kusumapuravr . tt describe Kusumapura-situation Describe the situation in Kusumapura. Mu 1.13.1 (= SG 21.2.5) [S [P rt kim ] [N P1 utkh aditam [N P6 (Nandavan sasya) ] ] either extirpation Nanda-clan ] [V P ] ] [S [P rt kim v a ] [N P1 sthairyam is or rmness [V P utp aditam] ] [N P6 (Candraguptalaks ah . my .) ] ] produced Candragupta-glory Either has there been the extirpation of Nandas clan or has Candraguptas glory been made rm?

(40.3)

(40.4)

5.3.1 (= SG 23.1.9) S [IC [N P2 ((Kan sis agamanam) ] [N P4 asmai ] . va . ya) Kan va-student-arrival to him . niveditum ] to inform to inform him of the arrival of Kan . vas student. Indeed, this generalization does not hold for English either. After all, English has such compounds as: a New York Times article, a Ford truck, the Berlin airlift, the Boston Harbor, the Smith family, the Kennedy clan, the Quebec crisis, an Ontario lawyer, and a Nixon admirer.15 Moreover, proper names, including family names, names of countries, and names of cities, can be turned into adjectives through derivational axation: Hume becomes Humean, Canada becomes Canadian and Paris becomes Parisian. Analogous derivations obtain for Sanskrit: the name for the god Siva forms the adjective Saiva and the name for the river Yamun a forms the adjective Y amuna. (40.5) Di Sciullo and Williams have suggested that, in such cases in English, the proper names are not truly referential, and hence not really proper names. They point to the fact that only the second sentence below is self-contradictory, as support for their claim of the non-referentiality of proper names in compounds.
15 As pointed out in Kiparsky 1983, English proper names are also subject to derivational suxation.

25

(41.1) (41.2)

John is a Nixon admirer in every sense except that he does not admire Nixon. John admires Nixon in every sense except that he does not admire Nixon.

The sentences certainly do show a contrast. But this contrast appears, even when a proper name in a compound is replaced by a common noun. (42.1) (42.2) John is a truck driver in every sense except that he does not drive a truck. John drives a truck in every sense except that he does not drive a truck.

Indeed, the culprit seems to be the -er sux, for a similar contrast can be eected, even when there is no compound: (43.1) (43.1) John is a writer in every sense except that he does not write. John writes in every sense except that he does not write.

What seems to be the correct generalization is that English strongly resists compounds with personal names, but not ones with other kinds of proper names.16 It is undeniable that the usual pronouns the interrogative, relative, and personal pronouns do not occur within English compounds.17 In Sanskrit, however, all pronouns - personal (Whitney 1881, 494 and 497; Boose and Tubb 1981, ch 5.16), interrogative (Whitney 1881, 505; Boose and Tubb 1981 ch. 5.16), and relative (Boose and Tubb 1981, ch. 5.16) occur within the compounds of Sanskrit. Below are sentences in which appear: the relative pronoun (in its stem of form yat), (44) PVS 18.12 antah [S [RC [N P1 [AP1 (yat1 arthah . )-B ] dr .s .t . . ] which1 -sake-ed example [V P ucyate ] ] [N P1 [AP1 sah . 1 ] arthah . ] is said that1 sake [V P [AP1 siddhah ]]] . is established The thing for the sake of which the example is stated is established.

16 Other factors are also at work. For example, if a proper name has a corresponding adjective, a compound formed with the proper name is far less acceptable than a phrase formed with the corresponding adjective. Thus, no adjective is formed from the proper name Ontario, hence Ontario lawyer is acceptable; but the adjective Canadian is formed from Canada, and Canadian lawyer is acceptable but Canada lawyer is not. 17 Of course, he and she do appear in compounds such as he-goat and she-goat; but the expressions he and she are not functioning as pronouns, rather they are expressing the sex of the animal. See Marchand 1969, ch. 2, sec. 9 for other examples.

26

the interrogative pronoun (in its stem form of kim), (45) M 4.5.7 (= SG 14.1.2) [S [N P6 [AP6 tayoh . ] baddhayoh . ] [V P the two prisoners is moks [AP1 (kimnimittah )-B ] ] [ [ ayam ] N P AP . ah . ]] . 1 1 release what-basis-ed the What basis does the release of the two prisoners have?

and the third person personal pronoun (in its stem form of tat), (46.1) PVS 10.3 [S [AC [N P7 [N P6 r agasya1 ] (an-upayoge) ] ] of passion1 non-use saktih [ADV katham [N P1 (tat1 . ) ] [V P upayujyate ] ] how it1 -potentiality use If there is no causal role for passion in the production of speech in what way does its potentiality have a causal role? anvayasya ] ni scayena ] concomitance ascertainment [N P6 (((viruddha1 )+(tat1 paks an am)) ] nir asah . y . . ]] contrary1 +it1 -similar rejection There is the rejection of contraries and ones like them by the ascertainment of concomitance. (47.1) PVS 22.7 aryasya1 ] [P P [N P2 (sva1 k aran [AC [N P7 [N P6 k . am) ] eect1 self1 -cause antaren ave ]] . a ] bh without existence if an eect were to exist without its cause PVS 19.25 [S [N P1 tau1 ] ] [V P [N P2 (sva1 pratibaddham) ] the two1 self1 -connected nivartayatah . ]] exclude The two exclude what is connected with them. PVS 11.1 [S [N P1 [N P3 [N P6

(46.2)

(47.2)

The third generalization, namely, that proper constituents of compounds do not enter into anaphoric relations, is clearly untrue of Sanskrit. As the foregoing examples show, subordinate constituents within a Sanskrit compound may be a term in an anaphoric relation. This is true not only of the relative pronoun (44), the third person personal pronoun (46), and the reexive pronoun (47), but also of non-pronominal constituents, as shown by the fact that the antecedent of the

27

third person personal pronoun in (46.2) is a subordinate constituent within a compound. The facts are less clear in English. Postal (1969), Corum (1973), and others, have observed that sentences such as the following appear to exhibit referential dependence: (48.1) (48.2) John became a guitarist because he likes the way it sounds. (it is construed with guitar.) Texans think that it is the greatest State in the Union. (it is construed with Texas.)

However, it has been plausibly suggested by some, for example Bosch (1983, ch. 5.2.3), from whom the rst example above is taken, that the co-reference above does not result from referential dependence, but from pragmatic inference. This explanation receives support, I believe, from the fact that the potentiality for coreference deterioritates as the context becomes more neutral. (49.1) The guitarist thinks that it is in the car. (Dicult to construe it with guitar.) (49.2) Texans think that it is a million square miles. (Dicult to construe it with Texas.) The last generalization, put forth by Di Sciullo and Williams (1987, p. 30), is that constituents outside of a compound cannot be construed with constituents subordinate within an compound. This generalization is undoubtedly true of English compound formation, as illustrated by the contrast in the interpretability of the expressions in (50). (50.1) (50.2) ((maneating)shark) *(eatingshark) of men

Interestingly, this generalization was regarded as true of Sanskrit by P an . ini. His treatment of compounds is to pair them with canonical phrasal paraphrases with which they share a common derivational ancestor; in addition to their semantic relation, they bear the syntactic relations of having the same heads and of having the same constituency. Hence, the constituency of compounds mirrors that of their canonical phrasal paraphrases. A condition on compound formation is that two elements cannot undergo compounding, the deletion of morphology from the subordinate element, unless the two elements form a constituent (A 2.1.4). A consequence of this is that inected lexical items exterior to a compound are not construable with subordinate constituents within it. The applicablity of this rule is illustrated both by Pata njali, in his Mah abh as . ya, or Great Commentary, on P an inis As t a dhy a y (at A 2.1.1), and by Bhartr hari, in . .. . his work on the semantics of Sanskrit (VP 3.14.46), with the following example:

28

(51.1)

(51.2)

[N P1 ((r aja )purus . ddha r . ah . )] rich king man servant of a rich king *[N P1 [AP6 r ajapurus . ddhasya ] (r . ah .) ] of rich king-man servant of a rich king

Though the generalization holds of English compounds, it does not of Sanskrit compounds. Counter-examples are furnished both by Pata njali (MBh on A 2.1.1) and by Bhartr . hari (VP 3.14.47): (52.1) (52.2) (52.3) [N P6 [N P6 Devadattasya ] guroh ] kulam . of Devadatta of teacher family (Devadattaguru)kulam Devadatta-teacher-family [N P6 Devadattasya ] (gurukulam) of Devadatta teacher-family Devadattas teachers family

Indeed, compounds appearing in congurations such as that in (52.3) are given a special name by Sanskrit grammarians: they call them asamartha compounds (i.e., non-constituent compounds). Moreover, these compounds are well attested in the classical literature. A study of over three-hundred sentences, chosen essentially at random from the Sanskrit corpus, reveals thirteen clear cases of non-constituent (asamartha) compounds. (See Appendix I in Gillon 1993.) And a study of the rst approximately ve-hundred sentences of a single text reveals forty-three clear cases. (See Appendix II in Gillon 1993.)18 Thus, for example, in the best known play by the nest dramatist of Sanskrit literature, Kalid asas Sakuntal a, one nds precisely these congurations. 3.9.16 (= SG 3.1.6) (53) S [N P1 [N P3 [N P7 tasy am ] (snigdhadr a) ] .s .t . y on her xed-gaze (s ucitaabhil as . ah . )-B ] indicated-aection-ed .. whose aection was indicated by his gaze being xed on her Here, the perfect passive participle, s ucita (indicated), which is a subordinate constituent within the exocentric (bahuvr hi) compound s ucitaabhil a. sah . (*indicated-aectioned: whose aection was indicated), is construed with the third, or instrumental, case noun phrase tasy am snigdhadr s. ty a (by his gaze .. being xed on her). Moreover, this noun phrase itself exhibits a non-constituent (asamartha) compound, for the perfect passive participle, snigdha (xed) is
put these frequencies in perspective, I should point out that non-constituent (asamartha) compounds occurred more frequently in each corpus taken separately or jointly than either indirect questions or relative clauses.
18 To

29

found as a subordinate constituent in the compound snigdha-dr s. ty a (by xed.. gaze: by his gaze being xed), yet it is construed with tasy am (on her), a seventh, or locative, case noun phrase, for which the verb snih (to x) subcategorizes. One person to attempt to meet the challenge presented by these compounds to P an . inis grammar of Sanskrit was Bhartr . hari, who suggested that nonconstituent (asamartha) compounds are limited to cases where the subordinate constituent in the compound expresses a relation. It is tempting, following Bhartr . haris lead, to conjecture that the subordinate constituent in the compound has a thematic role which is assigned to the inected lexical item external to the compound. A survey of the cases mentioned above, as culled from the classical literature, shows that, for the most part, such is the case: the subordinate constituent is a deverbal noun or adjective, either subcategorizing for an NP complement or having associated with its verbal root a thematic argument position, which the external lexical item satises. One way to reconcile this fact about Sanskrit with the view of Di Sciullo and Williams of the inaccessibility of phrasal rules to proper constituents of lexical items is to reject the claim by Di Sciullo and Williams (1987, p. 30) that the arguments of the nonhead are no part of the argument structure of the compound. While their claim holds of English, it does not hold of Sanskrit. Further evidence of this dierence between Sanskrit and English comes from exocentric (bahuvr hi) compounds. To begin with, consider the following exocentric (bahuvr hi) compounds in Sanskrit: (54.1) Compound: Analysis: Paraphrase SK 830 pr apt atithih amah . gr . (pr aptaatithih amah . -B) gr . reached-guest-ed village [RC atithayah apt ah . pr . yam ] guest reached which apt atithih amah sah . pr . gr . that village the village which guests have reached SK 830 u d an . harathah . anad . v anad an ( ud ha rathah )-B . . . v drawn-cart-ed bull [RC rathah d yena ] . u . hah . cart bull by which d an sah . u . harathah . anad . v that bull the bull by which a cart is drawn

(54.2) Compound: Analysis: Paraphrase:

30

(54.3) Compound: Analysis: Paraphrase:

SK 830 upahr suh sah . tapa . puru . sah (upahr suh . tapa . )-B puru . oered-cattle-ed man [RC pa suh . upahr . tah . yasmai ] cattle oered to whom suh sah sah . upahr . ta-pa . puru . that man the man to whom cattle is oered SK 830 uddhr a sthal . taudanan (uddhr a)-B sthal . taodanan removed-rice vessel [RC odanah yasy ah ] . uddhr . tah . . rice removed from which s a uddhr a sthal . taudanan that vessel the vessel from which rice has been removed SK 830 p t ambarah . purus . ah . (p taambarah )-B purus . . ah . yellow-garment-ed man tam ambaram yasya ] [RC p yellow garment whose sah t ambarah sah . p . puru . that man the man whose garments are yellow

(54.4) Compound: Analysis: Paraphrase:

(54.5) Compound: Analysis: Paraphrase:

In the paraphrase and translation of Sanskrit exocentric (bahuvr hi) compounds, the relative pronoun of the paraphrasing relative clause may be construed with either the subject (54.5) or the predicate (all other examples). Thus, the relative pronoun is construed with the predicate in such a way as to express the goal in the rst example, the agent in the second, the beneciary in the third, the source in the fourth, and the location in the last. Indeed, as noted by Coulson (1976, p. 121), Sanskrit exocentric (bahuvr hi) compounds are ambiguous between two readings: on one, the denotation of the lexical item modied by the exocentric compound is interpreted as the possessor of what is denoted by the nal constituent of the compound; and on the other, it is interpreted as bearing a thematic role of any unsaturated argument position associated with the initial constituent of the compound.

31

(55) Compound: Analysis: Reading 1: Reading 2:

Coulson 1976, p. 121 dr a str .s .t . akas .t . str (dr a)-B .s .t . a-kas .t . witnessed-misfortune-ed woman a woman whose misfortune has been witnessed (i.e., a woman whose misfortune people have witnessed) a woman by whom misfortune has been witnessed (i.e., a woman who has witnessed misfortune)

Moreover, an exocentric compound has available a reading corresponding to each of the unsaturated argument positions associated with its initial constituent. (56) Compound: Analysis: Reading 1: Reading 2: Coulson 1976, p. 121 datt adar a raj n (datta- adar a)-B raj n given-respect-ed queen a queen by whom respect is given (i.e, a respectful queen) a queen to whom respect is given (i.e., a respected queen)

Here emerges an important dierence between English and Sanskrit exocentric compounds. Notice that, of the six examples, only the fth allows an acceptable English calque: *reached-guested, *drawn-carted, *oered-cattled, *removed-vesseled, and hero-manned19 , but yellow-garmented. At the same time, while an English exocentric compound is paraphrasable with a relative clause, yet the relative pronoun of the paraphrase, whose, is construed only with the subject of the relative clause, which corresponds to the nal constituent of the compound paraphrased. Thus, mean-spirited is paraphrasable as one whose spirit is mean, level-headed as one whose head is level, and long-legged as one whose legs are long. English and Sanskrit exocentric (bahuvr hi) compounds dier as follows: the English adjectival sux -ed does not permit the transmission of unsaturated argument positions of an exocentric compounds initial constituent; whereas the Sanskrit adjectival sux B does permit the transmission of such argument positions. The foregoing dierences between compounds in English and Sanskrit suggests the following hypothesis: information about the argument structure of initial constituents in lexical structure, in particular, in compound, percolate in Sanskrit but does not in English. This hypothesis accounts for two facts: rst, that, in Sanskrit, unsaturated argument positions associated with the initial constituent of an exocentric compound can be assigned to the lexical item the compound modies, whereas in English they cannot be; second, that Sanskrit
19 This is acceptable under the interpretation which construes manned as the past participle of the verb to man.

32

productively forms non-constituent (asamartha) compounds whereas English does not. Let us see how this account works. Each adjective has at least one argument position which is saturated either by the noun it modies or by the subject noun phrase of which it is predicated. This is illustrated below, for modication both within phrasal structure and within compound structure. (57.1) AP A yellow garment (57.2) @ @ N A yellow

NP

N @ @ N garment

As it was shown earlier, both the -ed sux in English and the -B sux in Sanskrit create adjectives from nouns. This means that they create an argument position. Associated with the resulting argument position is the thematic value possessor. When the English sux is applied to a simple noun like beard, one obtains the following: (58)

A PS @ @ A PS -ed PS

N beard

And when the resulting form modies a word such as man, the resulting interpretation is man who possesses a beard. Combining what has been said so far, one obtains an analysis for both the Sanskrit compound in (54.5) and its English claque translation. Moreover, the foregoing analysis shows precisely where Sanskrit and English dier. A morphologically complex English word accepts unsaturated argument positions associated only with its head. Whereas, a morphologically complex Sanskrit word accepts the unsaturated argument positions either of its head or of its heads sister. When an exocentric compound has no unsaturated argument 33

position other than the one associated with its possessive sux, then its English and Sanskrit versions are equally acceptable. (59) A PS  Q  Q Q Q N ambara garment -B PS -ed PS

N   A

A PS

p ta yellow

In this example, the argument position associated with yellow (p ta) is saturated by garment (ambara), and so the complex word yellow garment (p t ambara) has no unsaturated argument positions. The suxation of -ed (-B) to yellow garment (p t ambara) creates an unsaturated argument position with an associated valence, namely that of possessor (PS). The situation is otherwise when the left-hand constituent of an exocentric compound has an unsaturated argument position. Sanskrit permits unsaturated argument positions associated with either a head and a non-head to be transmitted to the mother node; and, depending on which unsaturated argument position is transmitted, the compound receives one or another interpretation. Thus, in the compound in (55), the unsaturated argument position associated with the entire compound may have associated with it either the value agent or the value possessor (annotated below as AG\PS ).

(60)

A AG\PS  Q  Q N AG A PS  Q  Q A AG,PT N kas .t .a -B PS

dr .s .t . a AG,PT

In constrast, English prohibits any unsaturated argument positions from being associated with a non-head, with the consequence that the English counterparts to (55) are ungrammatical (annotated below as ).

34

(61)

 A AG,PT

N AG Q Q

A AG  Q  Q

A PS

witnessed AG,PT misfortune -ed PS

The situation with non-constitutive compounds is precisely the same as that with exocentric compounds. Sanskrit permits unsaturated argument positions associated with a non-head to be transmitted to the mother node, while English prohibits non-heads from having unsaturated argument positions. Thus, for example, an expression such as (50.2) is prohibited in English. The reason is that, although one of the arguments positions associated with eating, namely the one whose valence is agent (annotated below as ag) is saturated by the noun shark, the other argument position associated with eating, namely the one whose valence is patient (annotated below as pt), is not.20 relevant (62)    N    A AG,PT Q Q Q N NP Q Q Q PP   P S S NP N eating AG,PT shark of men

In contrast, an expression in Sanskrit such as tasy am snigdha-dr s. ty a tolerates .. and transmits unsaturated argument positions associated with non-heads, as shown below.
20 Evidence that the argument position with the valence of PATIENT is relevant comes from the acceptability of (50.1), namely man-eating shark.

35

(62) "

NP3 " "

b b b " " " N3 LC b b b N3 AG,PT dr a .s .t . y gaze

AG,PT

NP7 N7

A AG,LC snigdha xed

AG,PT

tasy am her

Conclusion

The foregoing examination of the details of derivational morphology and compound formation in Sanskrit shows that they yield to a simple analysis in terms of context free rules, which have a status independent from those required for phrasal syntax. It also shows that virtually the same context free rules characterize the lexical structures both for English and Sanskrit. It is observed that a number of generalizations which hold for English compounds do not hold for Classical Sanskrit compounds. The dierence between English and Sanskrit in this matter is shown to result from a systematic dierence between the two languages: namely, the dierence in an entire compounds access to the argument structure of its non-head constituents. While these empirical results are interesting on their own, they also, I believe, support the view that syntax (that is, phrasal syntax) and morphology (that is, lexical syntax) are distinct sub-theories of grammar, each with its own primitives and principles governing well-formedness, for the specic analysis developed above is grounded in precisely such a separation of syntax and morphology.

36

ABBREVIATIONS FOR TEXTS A K M MBh Mu MW NBT NBTP PSED PVS PVST S SG SK U V VP = = = = = = = = = = = = = = = = = As adhy ay . See MBh. .t . Reference: adhy aya, p ada, s utra K adambar ; Peterson (ed) 1885. Reference: page, line M alavik agnimitra; Iyyar (ed) 1896. Reference: act, verse, and line following verse Mah abh as . ya; Kielhorn (ed) 1880. Mudr ar aks . asam; Telang (ed) 1935. Reference: act, verse, and line after verse Monier-Williams 1899 Ny ayabindut ka; Malvania (ed) 1955. . Reference: page, line Ny ayabindut kaprad pa; Malvania (ed) 1955. . Reference: page, line Apte 1890. Pram an arttika-svavr . av . tti; Gnoli (ed) 1960. Reference: page, line Pram an arttika-svavr k a; Sankr ayana (ed) 1935. . av . tti-t . . ty Reference: page, line Sakuntal a; Godabole (ed) 1933. Reference: act, verse, and line after verse Student Guide to Sanskrit Composition. Apte 1885. Reference: chapter, exercise set, example sentence Siddh anta-Kaumud ; Vasu (ed) (tr) 1906. Reference: s utra Uttarar amacarita; Belvalkar (ed) 19??. Reference: act, verse, and line following verse Vikramorva s yam; Pandit (ed) 1889. References: act, verse, and line after verse V akyapad ya; Iyer (ed) 1973 Reference: k an sa, k arik a .d . a, samudde

37

NOTATION

1. Lexical Constituent Structure (parentheses): (a) xy means that x is subordinate to y. (b) x+y means that x and y form a conjunctive (dvandva) compound. (c) x-y means that x and y form a compound. (d) -B indicates that what it is appended to is an exocentric (bahuvrihi) compounds. (e) encloses a list, possibly empty, of argument positions. Argument positions with which a valence is associated are indicated by the valence: ag for agent, lc for location, ps for possessor, and pt for patient. Argument positions with which no valence is associated are indicated by . 2. Phrasal Constituent Structure (square brackets): (a) The usual bracket labels are augmented by an indication of case marking: they accord with the P an . inian numbering (1 nominative; 2 accusative; 3 instrumental; 4 dative; 5 ablative; 6 genitive; 7 locative). (b) In addition, there are these labels: AC: absolutive clause Adv: adverb GC: gerundial clause IC: innitival clause LC: valence of location Prt: particle RC: relative clause 3. Miscellaneous: (a) Anaphoric Co-indexation: index is to the immediate right of the relevant constituent. (b) indicates a phonetically null element.

38

ACKNOWLEDGEMENTS The initial research for this paper was partially funded by a grant from the Social Sciences and Research Council of Canada. The basic research for it was done in 1986 while I was a senior fellow of the American Institute of Indian Studies, aliated with the Centre for the Advanced Study of Sanskrit at Poona University. While there I had the opportunity to learn from and work with a number of scholars. I am especially grateful to Dr. Siniruddha Dash, who devoted so much time to helping me through the intricacies of the P an . inian treatment of Sanskrit compounds. I was also privileged to receive advice and T. S. Sr niv astr Venugopalan of Deccan guidance from Pan asa S and Sr .d . it Sr College. Portions of earlier versions of this paper were presented 10 February, 1990, at the Workshop on Lexical-Syntactic Relations held at the University of Toronto; 9 June, 1990, at the Twelfth South Asian Languages Analysis Round Table, held at the University of California at Berkeley; and 12 November, 1992 to the D epartement de Linguistique of the Universit e de Qu ebec ` a Montr eal. I wish to thank the audiences in attendance at these presentations for helpful comments and criticism. An earlier version appeared in Toronto Working Papers in Linguistics, v. 11, n. 2. I also wish to thank Richard Hayes for his extensive comments on the rst and later versions of this paper as well as my colleague Glynne Piggot for his suggestions and criticisms pertaining to this papers most recent version.

39

REFERENCES Aklujkar, Ashok. 1992. Sanskrit: An Easy Introduction to an Enchanting Language. Richmond, British Columbia (Canada): Sv adhy aya Publications. Apte, V aman Shivar am. 1885. The Students Guide to Sanskrit Composition. A Treatise on Sanskrit Syntax for Use of Schools and Colleges. Poona, India: Lokasamgraha Press, 24th edition (1960). Apte, V aman Shivar am. 1890. The Practical Sanskrit-English Dictionary. Poona, India: Prasad Prakashan, revised and enlarged e dition (1957). Bauer, Laurie. 1983. English Word Formation. Cambridge: Cambridge University Press (Cambridge Textbooks in Linguistics). Belvalkar, S. K. (ed) 19??. Uttarar amacarita. Cambridge, Massachusetts: Harvard University Press (Harvard Oriental Series: v. 22), 19??). Bhat ks anta-Kaumud . Sanskrit Edition and English Transla.t . oji D . ita. Siddh tion: Vasu, Sr sa Chandra (ed) (tr) 1906. Boose, Emery and Gary Tubb. 1981. Rough Draft of Portions of a Handbook Designed to Aid Students in the Use of Sanskrit Commentaries. Unpublished mss., Harvard University. Bosch, Peter. 1983. Agreement and Anaphora: A Study of the Role of Pronouns in Syntax and Discourse. New York, New York: Academic Press (Cognitive Science Series). Cardona, George. 1988. P an . ini: His Work and Its Traditions. Background and Introduction. New Delhi: Motilal Banarsidass. Chierchia, Gennaro and Sally McConnell-Ginet. 1990. Meaning and Grammar: An Introduction to Semantics. Cambridge, Massachusetts: The MIT Press Corum, Claudia. 1973. Anaphoric Peninsulas. In: Papers from the Ninth Regional Meeting of the Chicago Linguistic Society , pp. 89-97. Coulson, Michael. 1976. Sanskrit: An Introduction to the Classical Language. London, England: Hodder and Stoughton (Teach Yourself Books). Deshpande, Madhav M. 1985. Ellipsis and Syntactic Overlapping: Current Issues in P an . inian Syntactic Theory. Poona, India: Bhandarkar Oriental Research Institute (Post-graduate and Research Department Series no. 24).

40

Di Sciullo, Anna Maria and Edwin Williams. 1987. On the Denition of Word. Cambridge, Massachusetts: The MIT Press. Gillon, Brendan S. 1991. Word Order in the Sv arth anum ana of Dharmak rtis Pram an av a rttika . In: Steinkellner (ed) 1991, pp. 59-65. . Gillon, Brendan S. 1992. Word Order in Classical Sanskrit. McGill University, unpublished mss. Gillon, Brendan S. 1993. Bhartr . haris Solution to the Problem of Asamartha Compounds. Etudes Asiatiques/Asiatische Studien: v. 47, n. 1, pp. 117-133. Gnoli, Raniero (ed). 1960. The Pram an arttikam of Dharmak rti: The First . av Chapter with the Autocommentary. Text and Critical Notes. Rome, Italy: Istituto Italiano per Il Medio ed Estremo Oriente: (Serie Orientale Roma: 23). Godabole, N. B. (ed). 1933. Sakuntal a. Bombay: Nirn agara Press (revised . aya S by W. L. Pan s kar). . Hudson, R. A. 1975. Problems in the analysis of ed-adjectives. Journal of Linguistics: v. 11, pp. 69-72. Iyer, K. A. Subramania (ed). 1973. V akyapad ya of Bhartr rn a sa . hari with the Prak . akaprak of Helar aja. Poona: Deccan College. Iyyar, Sesh adri (ed). 1896. M alavik agnimitra. Poona, India: publisher unknown. Kielhorn, F. (ed). 1880. The Vy akaran abh a. sya of Pata njali. Poona, . a Mah India: Bhandarkar Oriental Research Institute (4th edition revised by R. N. Dandekar 1985). Kiparsky, Paul. 1983. Word Formation and the Lexicon. In: Ingemann, Frances (ed) 1983 pp. ??-??. Lees, Robert. 1960. The Grammar of English Nominalizations. The Hague, The Netherlands: Mouton. Ljung, Magnus. 1976. -ed adjectives revisited. Journal of Linguistics: v. 12, pp. 159-168. Malvania, Dalsukhabhai (ed). 1955. D urveka Mi sras Dharmottaraprad pa. Patna, India: Kashiprasad Jayaswal Research Institute (Tibetan Sanskrit Works Series: v. 2). 2nd edition revised, 1971. Marchand, H. 1969. The Categories and Types of Present-Day English Word-

41

Formation. A Synchronic-Diachronic Approach. Munich: C. H. Becksche Verlagsbuchhandlung. (reprint) Monier-Williams, Monier 1899 A Sanskrit-English Dictionary. Oxford, England: Oxford University Press. Pandit, S. P. (ed). 1889. Vikramorva s yam. Bombay: Government Central Book Depot (Bombay Sanskrit Series: v. 16), 2nd edition. Peterson, P. (ed). 1885. K adambar . Bombay: Government Central Book Depot (Bombay Sanskrit Series: v. 24). Postal, Paul. 1969. Anaphoric Islands. Papers from the Fifth Regional Meeting of the Chicago Linguistic Society: pp. 205-237. Selkirk, Elizabeth O. 1982. The Syntax of Words. Cambridge, Massachusetts: The MIT Press. Speijer, J. S. 1886. Sanskrit Syntax. Leiden, The Netherlands: E. J. Brill. Steinkellner, Ernst (ed). 1991. Studies in the Buddhist Epistemological Tradition. Proceedings of the Second International Dharmak rti Conference, Vienna 11 - 16 June, 1989. Vienna: Verlag der Osterreichischen Akademie der Wissenschaften. Telang, K. T. (ed) 1935. Mudr ar aks . asam. Bombay, India: Nirn . aya Sagara Press (Bombay Sanskrit Series: v. 27). Vasu, Sr sa Chandra (ed) (tr). 1906. The Siddh anta Kaumud of Bhat toji .. Diks ta . Allahabad, India: The P a n ini Oce. Reprint: Delhi, India: Moti. . lal Banarsidass, 1962. Whitney, William Dwight. 1881. Sanskrit Grammar: Including both the Classical language, and the older Dialects, of Veda and Brahmana. Cambridge, Massachusetts: Harvard University Press, 2nd edition (1889), 11th reprint (1967). Williams, Edwin. 1981. On the Notions Lexically Related and Head of a Word . Linguistic Inquiry: v. 12, pp. 245-274.

42

ABSTRACT This paper shows that context free rules of the sort used by Elizabeth Selkirk and by Anna-Maria Di Sciullo and Edwin Williams to analyze English derivational morphology and compound formation prove to be enlightening when applied to the derivational morphology and compound formation of Classical Sanskrit. It turns out that virtually the same class of context free rules can accurately characterize the derivational morphology and compound formation found in both languages. In spite of the near co-incidence of the class of lexical structures available for the two languages, calque translations are not equally acceptable in both languages. This dierence is attributable to a dierence in an entire compounds access to the argument structure of its non-head constituents: Classical Sanskrit tolerates such access, while English prohibits it.

43