The treatment of upapada-tatpurusa compounds in Grimal, Venkataraja Sarma, and Lakshminarasimham's (2007) Book of Compound Words and the treatment of the starting point in Paninian derivation in several recent papers by Houben (2003, 2009a, 2009b, 2010) occasion a rearticulation of initial phases and particular points of Paninian derivational procedure. Grimal et al. (2007) omit early steps from their derivations and, as a result, show nominal terminations present at their first step in the derivation of upapada tatpurusa compounds. 1 Even though their annotations reveal the correct understanding of Panini' s derivational procedure, omitting early steps gives the incorrect impression that Pacini's derivational procedure begins with these speech forms present rather than with the semantic and syntactic conditions that occasion them. Their exposition closely follows that of Bhattojidiksita; yet the latter himself diminishes the role of semantic and syntactic conditions in derivational procedure in departure from his predecessors. Houben (2003, 2009a, 2009b, forthcoming) deliberately argues that Paninian derivation begins with speech forms and does not begin with the early steps in question at all. He asserts that the derivation begins with a sentence or phrase that the speaker uses the grammar to check for correctness. He argues that semantic and syntactic conditions are incapable of determining speech forms without the guidance of user decisions, and that the grammar is used merely to reconstitute a preliminary sentence that the user of the grammar has in view in order to validate its correctness. Given these challenges to the view that Paninian derivation begins with semantics, the occasion is ripe for an investigation of just what speech forms are in view at the start of a Paninian derivation and what semantic conditions are required. The pivotal issue arises in the derivation of the upapada-tatpurusa compound kumbhakara 'potter'.

1. WHAT THE POTTER HAS TO DO WITH SEMANTICS

1.1 Basic assumptions in linguistics

The clarification of what speech forms and what semantic conditions are in view at the start of a Paninian derivation requires first a clarification of some basic assumptions about the nature of linguistic science as it was conceived by the ancient Indians. Ancient Indian linguists begin from the conception of speakers and end with speech. While Indian grammatical works presuppose an analysis of speech and early modern Indian semantic works are concerned with cognition from the perspective of a listener, none of the extant Sanskrit grammars begins with actual speech. They all, from the ancient phonetic treatises proper to particular Vedic traditions (Pratisakhyas) to medieval non-Paninian grammars and early modern reworkings of Paninian grammars, derive actual speech from basic elements previously abstracted in accordance with an assumed prior analysis. The rules produce speech; they themselves, formulated to take the prior analysis into account, do not analyze it. In that sense Indian grammar is generative. It is constructed from the point of view of the speaker, not of the listener. Pacinian grammar in particular uniformly instructs which speech forms are to be used under various conditions, including some 735 semantics conditions described in Scharf 2009a (101-9); conversely, the grammar never instructs what meaning is to be understood from a speech form. Paninian grammar is therefore a generative grammar beginning from basic linguistic units and semantics and ending with actual speech forms. While Paninian grammar is generative, it is not fully transformational; that is, it does not transform one actual utterance into another. While it is transformational to the extent that certain morphemes are posited as basic and variations are produced by replacements, it does not give preference, for instance, to the active voice over the passive voice in the basic speech forms posited (as some forms of modern transformational grammar do). Instead, alternate syntactic constructions that express some common meaning are derived from abstract non-phonetic categories. Identical conditions stated in various rules account for the common meaning while variant conditions or unconditioned alternation account for the differences in the alternate speech forms. Paninian grammar therefore does not have a sentence as its starting point. It has as its starting point a conception in the mind of a speaker embodied to a limited extent, before the application of any rules, already in certain basic phonetic elements, namely roots and underived nominal stems.

The question of what, if any, speech forms are in view as the starting point for Paninian derivation is determinable from an examination of the set of rules and its supplementary lists. The only speech forms permissible at the start of a derivation are those (roots and stems) listed as basic elements, those inferrable as being of the same kind in lists of exemplary elements (akrtigana), and those included by specific semantic criteria. The supplementary lists consist in particular of the Dhatupatha and ganas to which rules of the Astadhyayi refer. Numerous rules provide operations on some 282 lists (gana) mentioned in those rules, beginning with A. 1.1.27 sarvadini sarvanamani by which speech forms in the list beginning with sarva 'all' are termed sarvanaman 'pronoun'. Roots listed in the Dhatupatha are termed dhatu by 1.3.1 bhüvadayo dhatavah. Finally an open class of additional speech forms is included as basic elements under the sole specification that they be meaningful. By A. 1.2.45 arthavad adhatur apratyayah pratipadikam, meaningful speech forms (arthavat), other than roots, affixes, and speech forms that end with them, are termed pratipadika 'nominal base'. By A. 1.2.46 krttaddhitasamasas ca, complex speech forms derived by the grammar, including dérivâtes from roots, dérivâtes from nominal stems, and compounds, are also termed pratipadika. Other basic elements (affixes and augments) are explicitly introduced by rules. Nominal bases and roots are then generally referred to as preceding contexts in rules that provide affixes after them (e.g., dhatoh in 3.1.91 and pratipadikat in 4.1.1). These are the only speech forms present at the start of Paninian derivation; there are no others. Semantic conditions serve as the remainder of the initial conditions for the operation of rules of the Astadhyayi.

1.2 Reconstitution rather than synthesis?

Houben accepts that there is a synthetic part to a grammar user's use of Paninian grammar. What he denies is that semantics lie at the foundation of sentence generation. He (2009b: 13) rightly points out that certain basic units of speech are included at the start of a Paninian derivation when he writes, for instance, "the selection of a suitable root is normally the starting point of the synthetic part of his consultation cycle." He indicates (p. 14) the complementary absence of pure semantics while elaborating on the presence of basic units of speech - writing, "the concrete starting point for a derivation in the synthetic phase of the consultation cycle of a user of grammar in Pacini's time will then never be 'pure' meaning or an autonomous level of semantic representations but the selection of a root - for instance, bhu 'to be' - or a form from lists of underived stems, pronominal forms, etc. in which form and meaning are inseparably integrated." He reiterates (p. 13) criticism formulated in Houben 1999 of the views of Kiparsky and Staal (1969), Bronkhorst (1979), Joshi and Roodbergen (1975), and Kiparsky (1982) "according to which 'semantics' or 'meanings' form the starting point of the derivation," and directs that criticism against Kiparsky (2009), who postulates a level of semantic information that forms the starting point of the derivation of a complete sentence in which "karakas are assigned on the basis of 'semantic information'." While accepting "at least two distinct levels of derivation ... a level of morphological representations (where we find roots, stems, suffixes) and a level of phonological representations (with words in their final form after the application of all substitution rules including those of sandhi)" (p. 15), Houben writes, "no additional level of representation is needed to account for Pacini's system." He regards syntax and semantics "as domains of consultation, which allow the user of the grammar to label the linguistic forms of his preliminary sentence according to the syntactically relevant categories of meaning or according to semantically relevant generalizations of form (suffixes)" (p. 15), stating, "As I argued extensively in 1999[: 26-27], the view that Pacini's grammar is a device 'to encode a given meaning and to produce an expression' is untenable" (p. 13).

Rather than accepting a semantic foundation for Paninian derivation, Houben asserts instead that the starting point is a preliminary statement. Houben asserts that "the starting point" of a Paninian derivation "is a preliminary sentence that needs to be checked or that needs some little extra refinement" (2009a: 524). He writes (2009b: 14),

The system of Panini' s grammar "clearly requires a user who wants to check and possibly improve a preliminary statement" (Houben 2003: 161). The system implies the presence of a knowledgeable user, a preliminary statement, and the application of first analytic and next synthetic procedures to the words in it, with the user keeping in mind the preliminary statement and its purport, and aiming at the best possible, sam-skrta form of his preliminary statement.

Houben writes (2009b: 19), "no-one has ever produced a correct form through Pacini's system that was not already his starting point, or among his starting options. Usually the correct form is put at the beginning after which it is derived through the system." He continues, "the derivation of a word in a preliminary statement by any potential user of Pacini's grammar will normally start with the selection of a root in the Dhatu-patha corresponding to a selected problematic word in his statement." In conclusion, he considers it "more comprehensive and realistic" to view "Pacini's grammar as 'reconstitutive' rather than one-sidedly 'synthetic'" (p. 19). Houben reiterates these views in his most recent work (forthcoming: 3-4), disputing that "the starting point is in semantics (meaning elements, meaning conditions, etc.)" and asserting instead that it is "in a preliminary statement."

1.3 Karman: conceptual object rather than speech unit

Although much of Houben's concern is with the sociological question of the practical use of the grammar rather than with its formal features, his description betrays a fundamental misconception of Pacini's linguistic system: he views speech forms rather than meanings as the fundamental conditions for syntactic organization. In Houben's view, speech forms rather than meanings are designated by karaka terms, and speech forms rather than meanings are the conditions for abstract tense. He would assign karaka terms and abstract tenses (lat, etc.) "to the words of the preliminary utterance" rather than "to the semantic representations of level one" (2009b: 16). Although his critical analysis of Vakyapadiya 1.46 (2003: 148-55) is perspicuous in other respects, he is confused himself when he calls "confused and confounding" understanding that the term bruvikarman refers to an object of saying (p. 151 n. 32). He insists there that the karman 'object' of saying cannot refer "to an extralinguistic object," that it must refer to "a grammatical object" and hence, "requires bruvi to refer to the verb, not to its meaning." Here Houben asserts that a karman is a speech form rather than a semantic object denoted by a speech form, and that it has relation to a speech form, i.e., a verb, rather than to the object denoted by a verb, namely an action. Such an assertion is erroneous. Semantic objects, not speech forms, are classed as karman under conditions stated in A. 1.4.49-53 kartur ipsitatamam karma, etc. Semantic objects so classed are then the conditions for speech forms, namely, for nominal terminations, as provided by 2.3.2 karmani dvitiya, etc. The karman is not a speech form; it is an object, viewed as a participant in an action, that is desired by the agent of the action. It is objects, not speech forms, that participate in action, and it is participants in action, not speech forms, that are designated by karaka terms.

It is precisely the issue of the status of what is termed karman as the condition for the occurrence of krt- affixes and nominal terminations that is the crux of a problem in the derivation of the upapada tatpurusa compound kumbhakaraby Bhattojidiksita and hence by Grimal et al. The fact that the derivation of the compound does not begin with a corresponding phrase is significant for Houben's contention that the derivation must begin with a "preliminary statement." The sequence in which speech elements in the derivation are introduced and the conditions for them reveal the extent to which Paninian derivation begins with abstract semantic entities. Examination of Paninian discussions concerning the derivation of the compound kumbha-kara 'pot-maker' demonstrates that nominal terminations are not present at the stage of the provision of krt-affixes, that kit-affixes are conditioned by speech forms denoting semantic items designated by specific karaka terms, which in turn are conditioned predominantly by semantics.

2. THE UPAPADA-TATPURUSA COMPOUND KUMBHAKaRA

A reader seeing the compound kumbhakara would easily recognize that it consists of the element kumbha 'pot' compounded with kara 'maker' and that the latter term governs the former. The first assumption concerning its Paninian derivation might be that it is a sasthitatpurusa compound equivalent to the corresponding phrase (vigraha vakya 'analytic phrase'), *kumbhasya karah, as provided for by A. 2.2.8. A. 2.2.8 sasthi provides that a word (pada) terminating in a sixth-triplet nominal termination is optionally compounded with another word ending in a nominal termination and that the resulting compound is termed tatpurusa. Such compounds are merely optional because A. 2.2.8 occurs under the heading A. 2.1.11 vibhasa, which allows the corresponding phrases to occur usually. The sutra accounts for compounds such as raja-purusa that have corresponding phrases such as rajnah purusah.

The possibility that kumbhakara is a sasthi-tatpurusa compound is indeed raised by Patañjali, who mentions the example as falling within the scope of 2.2.8 as well as 2.2.19 under 2.2.19 vt. 3. He later rejects this position, however, with linguistic justification. The phrase *kumbhasya karah never occurs in Sanskrit, and kara in the meaning 'maker' never occurs as an independent word, only as the final element of a compound. Hence, commentators on the Astadhyayi cite kumbha-kara as an example of an upapada-tatpurusa compound provided by A. 2.2.19 upapadam atiñ. For instance, Patañjali cites the example kumbhakarah in the Mahabhasya on this sutra as does Jayaditya in the Kasika.

Panini accounts for dérivâtes that occur only as compound-final elements in composition with the terms they govern by stating the governed words (upapada) as conditions in rules that provide an affix after a root, and by having syntactically subordinate speech forms serve as conditions for the morphological derivation of the final compound elements. He proceeds as follows. The governed terms are stated in the locative in rules under the heading 3.1.91 dhatoh, valid through the end of the third adhyaya. A. 3.1.92 tatropapadam saptamistham states the principle that an item taught in the locative in a sutra under that heading is termed upapada. The obligatory compounding of a governed word with the word that governs it is accounted for by A. 2.2.19-20. The term nityam 'obligatorily' recurs in A. 2.2.19-20 from A. 2.2.17. These rules occur under the heading A. 2.1.1 samarthah padavidhih, which requires that potential compound elements be syntactically connected with each other. The order of elements in the compound is determined by two additional metarules. A. 1.2.43 prathamanirdistam samasa upasarjanam provides that an item taught in the nominative in a sütra in the compound section is termed upasarjana, and A. 2.2.30 upasarjanam pürvam provides that an item termed upasarjana occurs first in the compound. The sütra A. 2.2.19 provides that a word termed upapada, excluding one that terminates in a finite verbal affix (tin), is obligatorily (nityam) compounded with a second item. Because the term upapada is taught in the nominative in 2.2.19, the governed words under the heading 3.1.91, termed upapada by 3.1.92, are termed upasarjanaby 1.2.43 and therefore occur first in the compound.

The compound kumbha-kara is derived as an upapada-tatpurusa compound with the vigraha vakya kumbham karoti, instead of as a sasthi-tatpurusa compound formed in accordance with A. 2.2.8 with the vigraha vakya *kumbhasya karah. The full derivation of the compound (excluding accent) is shown in Table I.2 The entry under kumbhakara in Grimal et al.' s (2007: 266) The Book of Compound Words clearly lays out the steps of the derivation immediately relevant to compound formation. The steps in their derivation in order are steps 7, 9, 17, 20, 21, 18, 19, 22, 24, 25 of the derivation shown in Table 1. The first line of their derivation cites A. 3.2.1 karmany an (cf. Table 1, step 7), which occurs under the heading A. 3.1.91 dhatoh. The term karman in A. 3.2.1 is taught in the locative in a sütra under the heading A. 3.1.91 and so is termed upapada by A. 3.1.92 (cf. Table 1, step 5). The affix an, termed krt by A. 3. 1 .93 krd atiñ (cf. Table 1 , step 6), occurs under the condition that an agent is to be denoted in accordance with A. 3.4.67 kartari krt (cf. Table 1, step 7a). Grimal et al. explain, in their brief comment on their first derivational step, that the affix an occurs after the root kr (marked with ñ) on the condition that the agent (kartr) is to be denoted if a direct object (karman) is the subordinate term (upapada) connected with it (karmany upapade krñ-dhatoh kartari an-pratyayah). After accounting for strengthening (vrddhi) of the root kr in the second step (cf. Table 1, step 9), Grimal et al. cite A. 2.2.19 upapadam atiñ in the third step (cf. Table 1, step 17) and explain that it accounts for the compound of the upapada kumbha with kara which ends in a krt-affix. In the sixth step they cite A. 1.2.43 prathamanirdistam samasa upasarjanam (cf. Table 1 , step 1 8) and explain that it accounts for the speech form kumbha being termed upasarjana (kumbha-sabdasya upasarjana- samjna). In the seventh step they cite A. 2.2.30 upasarjanam pürvam (cf. Table 1, step 19) and explain that it accounts for the upasarjana kumbha being placed first (upasarjanasya kumbha-sabdasya pürva-nipatah). Grimal et al. explain the formation of the compound in their notes (tippani): the affix an occurs after the root kr 'make' in the meaning of the agent where the speech form kumbha 'pot', denoting the direct object (karman), is the governed item (upapada) (karma-vacini kumbhasabde upapade krñah kartr-arthe an-pratyayah).

Although the sixth-triplet nominal termination arises after the nominal base kumbha in syntactic connection with kara, it is not the case that a nominal termination arises after kara (Table 1, step *). A sixth-triplet nominal termination is provided after a base, such as kumbha in syntactic connection with an item ending in a krt-affix, kara, by A. 2.3.65 kartr-karmanoh krti (Table 1, step 12b). The condition for the nominal termination in A. 2.3.65 is that it be an agent (kartr) or direct object (karman) in syntactic connection with an item ending in a krt-affix. These conditions are satisfied. The form kara ends in the krt-affix (an), and kumbha denotes the karman of the action of making denoted by the root kr. After step 16 the step marked with an asterisk would provide the nominal base kara with a nominal termination, which steps *a and *b would restrict to a singular first-triplet nominal termination. However, the steps never occur because the obligatory compounding between the prior element and the subsequent element that ends in the krt-affix in step 17 preempts it. The arising of a nominal termination after the separate speech form kara is prevented because the tatpurusa compound of the upapada kumbha with the speech form kara is brought about by A. 2.2.19 upapadam atiñ before nominal terminations have the opportunity to arise.

The issue of the non-occurrence of nominal terminations after upapada-tatpurusa compound constituents is discussed in Patañjali's Mahabhasya under A. 2.2.19 (Kielhorn vol. 1, p. 418, lines 1-13), which Grimal et al. aptly summarize in their notes. The principle (paribhasa) 75 gati-karakopapadanam krdbhih saha samasa-vacanam prak sub-utpatteh states that the provision of a compound of a gati, karaka, or an upapada with an item ending in a krt-affix occurs prior to the arising of nominal terminations. Since a nominal termination has not yet arisen, there is not even a chance for the formation of a sasthi-tatpurusa compound in accordance with A. 2.2.8, which requires that an item terminating in a sixth-triplet nominal termination compound with another item ending in a nominal termination. As Grimal et al. write, an-utpanne supi sasthi-samasa-prasaktir èva nasti. Even if one could somehow form a sasthi-tatpurusa in accordance with A. 2.2.8 before nominal terminations arose, such a compound is optional (vibhasa recurs in A. 2.2.8 from A. 2.1.11) while in contrast A. 2.2.19 is obligatory (nityam recurs in A. 2.2.19 from A. 2.2.17). The obligatory upapadatatpurusa compound would occur, leaving no scope for the optional compound. The result is that rule A. 2.2.8 never even comes into conflict (vipratisedha) with A. 2.2.19, so that even the vigraha vakya *kumbhasya karah has no opportunity to occur.

The derivation provided in the entry under kumbhakara by Grimal et al. (2007: 266) is almost entirely correct. Yet despite the practical utility of the kumbhakara entry and the penetrating analysis of subtle issues by the authors in the notes, there appears to be a problem with the derivation, which the authors have overlooked. Although Grimal et al. in their notes clearly recognize that the sixth-triplet nominal termination cannot arise prior to the provision of the krt-affix an, the first step of derivation shows the sixth-triplet termination ñas (as marked with n) already present when the krt-affix an (a marked with n) is provided. They silently include the sixth-triplet nominal termination ñas after the nominal stem kumbha in the first step of their derivation at the step in which A. 3.2.1 karmany an provides the krt-affix an (cf. Table 1, step 7). Their step 1 first presents the string kumbha-as + kr-a. However, such a string is impossible. The nominal termination cannot be present already in step one of the derivation where the krt-affix is provided, as it is presented, because the krt-affix must be provided first in order to serve as a condition for the provision of the sixth-triplet nominal termination.

Grimal et al. recognize that the krt-affix is a condition for the sixth-triplet nominal termination in their notes, which state, "the sixth-triplet nominal termination arises after the nominal base kumbha on condition that the latter occurs in syntactic connection with an item ending in a krt-affix (krd-yoge kumbha- sabdat sasthï)" Yet they apparently overlook the implication for the first step of their derivation. The sixth-triplet nominal termination after a base in syntactic connection with an item ending in a krt-affix is provided by A. 2.3.65 kartrkarmanoh krti (cf. Table 1, step 12b). The condition for the nominal termination in A. 2.3.65 is that it be in syntactic connection with an item ending in a krt-affix. In the case of kumbhakara, the krt-affix an is provided by A. 3.2.1 (cf. Table 1, step 7). The application of A. 2.3.65 requires A. 3.2.1 to have already applied; A. 2.3.65 has no scope prior to the application of A. 3.2.1. Hence the nominal termination cannot be present already in step 1 of the derivation.

Moreover, a nominal termination never has the opportunity to arise after the speech form kara by itself (Table 1, steps *, *a, *b) since compounding occurs obligatorily (Table 1, step 17) and takes precedence over the provision of the nominal termination there. In contrast to an upapada, which is subject to obligatory compounding with an element ending in a krtaffix by A. 2.2.19, words ending in sixth-triplet nominal terminations provided by A. 2.3.65 are subject to optional compounding with another element ending in a nominal termination (supa) by A. 2.2.8 sasthi. The terms sup and supa recur in A. 2.2.8 from A. 2.1.2 and A. 2.1.4 respectively so that the compounding takes place between elements termed pada by A. 1.4.14 suptiñantam padam. In particular, the varttika stated under A. 2.2.8., krdyoga ca, allows such compounds with syntactically connected words whose nominal bases end with krt-affixes. Only where there is such optional compounding is there the possibility for a nominal termination to arise after the krt-affix and then for compounding to take place between the two elements both of which end in nominal terminations. That there is no equivalent corresponding phrase *kumbhasya karah in Sanskrit usage for the compound kumbha-kara is therefore critical: it is for this very reason that Panini forms the compound with the subsequent element without the nominal termination by A. 2.2.19 rather than with one by A. 2.2.8.

Is is crucial to note that there is no equivalent corresponding phrase *kumbhasya karah in Sanskrit usage for the compound kumbha-kara with which to begin a Paninian derivation, nor does Pacini's derivational procedure begin with the string kumbha-as + kr-a since the krt-affix an does not arise until step 7, and the sixth-triplet nominal termination ñas does not arise until step 12. The only speech forms available for a "preliminary statement" are kumbha and kr. A preliminary statement consisting of these speech forms would be incomplete and incapable of determining the derivation of the desired compound. The derivation would still depend upon pure semantics - disembodied meanings still unencumbered by corresponding speech forms - to condition the proper affixes and compound formation.

3. COMPOUND ELEMENTS WITHOUT NOMINAL TERMINATIONS

3.1 Upapada-tatpurusa compounds

A close examination of the commentaries demonstrates that Panini' s derivation of upapada-tatpurusa compounds does not begin with a corresponding phrase (vigraha-vakya) nor with nominal terminations present. Such an examination also reveals complex linguistic issues in the syntax and morphology of compounds and the techniques adopted by various commentators to account for the complexities within the Paninian linguistic system. Some of the techniques employed by certain commentators to solve certain difficulties create undesirable side effects which are then dealt with by subsequent commentators. The presence of a sixth-triplet or second-triplet nominal termination on the initial compound element in upapada-tatpurusa compounds before the application of A. 3.2.1 karmany an is such an undesirable side effect produced by medieval commentators. That the presence of a nominal termination at this stage of derivation is a problem has apparently remained unnoticed. Its solution requires revision of the conclusions of the commentators in question as well as of the scholars who relied upon them.

In the Astadhyayi, compounds are generally formed from words ending in nominal terminations and alternate with corresponding phrases. To ensure that compounds be formed from elements ending in nominal terminations, the technical term for nominal terminations sup recurs throughout most of the compound section, which extends from A. 2.1.3 prak kadarat samasah to A. 2.2.38 kadarah karmadharaye at the end of the second pada of the second adhyaya. Interpreted in accordance with A. 1.1.72 yena vidhis tadantasya, sup refers to a speech form that ends in a nominal termination. The term recurs in two inflected forms, in the nominative from A. 2.1.2 sub amantrite parangavat svare and in the instrumental from A. 2.1.4 saha supa. Together with other headings, these terms indicate that a speech form ending in a nominal termination compounds with a semantically and syntactically connected speech form that ends in a nominal termination. Likewise, the term vibhasa 'optionally' is stated as a heading in A. 2.1.1 1 and recurs throughout most of the compound section to allow compounds to alternate with corresponding phrases.

There are, however, compounds that cannot properly be formed from constituent elements that end in nominal terminations. These include compounds in which the prior element must compound with a subsequent element that has not yet been supplied with a feminine affix. The feminine affix must in turn occur prior to the provision of a nominal termination. Because the selection of the appropriate feminine affix depends upon the specific semantic, syntactic, and co-occurrence conditions of the compound, the correct feminine affix can only be provided subsequent to compound formation, and the nominal termination only subsequent to that. Notable examples include compounds such as dhanakriti '(a female) bought with wealth' formed from A. 2.1.32 kartrkarane krta bahulam, and kacchapi 'a female tortoise', an upapada-tatpurusa compound formed from A. 2.2.19. The derivation of the example kacchapi is presented in Table 2. If the compounds were required to be formed from constituent speech forms terminating in nominal terminations, erroneously only the form dhanakrita would result from A. 2.1.32, and the incorrect form kacchapa would result from A. 2.2.19 (Table 2, step 16). The feminine affix tap would occur after the final constituents prior to compound formation in accordance with A. 4.1.4 ajadyatas tap (Table 2, step * after 15). Instead, in the derivation of the correct form, the feminine affix ñip occurs after the compound stem subsequent to compound formation in accordance with A. 4.1.48 kritat karanapürvat or A. 4.1.63 jater astrivisayad ayopadhat (Table 2, step 19).

In exception to the general pattern of forming compounds from words already equipped with nominal terminations, nominal terminations are avoided on the final compound element prior to compound formation in these examples. In the derivation of dhanakriti, the term krta in A. 2.1.32 specifies that the initial compound element combine with a subsequent element that is a nominal base ending in a krt-affix rather than with a word ending in a nominal termination. (The term bahulam 'variously' in A. 2.1.32 is interpreted as allowing dhanakrita as well.) Likewise, to form the upapada-tatpurusa compound kacchapi correctly, A. 2.2.19 upapadam atin must be made to apply in the absence of nominal terminations on the final compound element. The term a-tiñ, referring to a speech form that does not end in a verbal termination, indicates that the restriction to speech forms that end in nominal terminations is no longer valid. Commentators and modern translators differ in their characterization of the criteria specified by the rule and the interpretation of the significance of the term a-tiñ. They do agree that the term nityam 'obligatorily' in A. 2.2.17 nityam kridajivikayoh, which recurs through A. 2.2.20, stops the recurrence of vibhasa in the rule, which thereby forms compounds obligatorily and does not permit corresponding phrases.

3.2 Katyayana and Patañjali

Examination of the statements of commentators concerning the prevention of nominal terminations on final compound constitutents prior to compounding begins with Katyayana (fourth or third century b.c.e.). In A. 4.1.48 varttika 4 gatikarakopapadanam krdbhih saha samasavacanam, Katyayana requires that certain initial compound elements be compounded with a final compound element that is a nominal base terminating in a krt-affix. The initial compound constituents to which the requirement applies include preverbs and other preverbal elements termed gati, speech forms denoting participants in action (karakas), and upapadas. The second category includes compounds such as dhanakriti provided by A. 2.1.32; the third includes compounds such as kacchapi provided by A. 2.2.19.

Among the reasons for stating the varttika, Patañjali (c. 150 b.c.e.) mentions the provision of the feminine affix ñis after a generic term (jater nisvidhane prayojanam) and supplies vyaghri 'tigress' and kacchapi 'female tortoise' as examples. Patañjali explains the motivation for the varttika with respect to the first example;3 his explanation is adapted here to apply to the latter, since kaccha-pa is an upapada-tatpurusa compound, so that reference may be made to the derivation in Table 2. Patañjali cites kacchapah as an example to which the first portion of A. 3.2.4 divided into two rules is applicable (Table 2, step 12). If compound constituents ended in nominal terminations, the feminine affix (ap would occur after the nominal stem of the final constituent prior to compounding by A. 4.1.4 ajadyatas tap (Table 2, step * after 15), and the final constituent pa terminating in long a would be compounded. The feminine affix ñis would then not occur by A. 4. 1 .63 jater astrivisayad ayopadhat (Table 2, step 19) since it is provided only after a nominal base ending in a short a. (As the Kasika observes, the term atah 'after a short a' recurs from A. 4.1.4.) The statement of the varttika solves the problem.

Under A. 2.2.19, Patañjali argues that it is not necessary to state varttika 4 under A. 4.1.48 because the mention a-tiñ in A. 2.2.19 upapadam atiñ achieves its purpose. The recurrence of sup and supa in A. 2.2.19 would disallow the rule from applying to finite verbs anyway, even without mentioning that it does not apply to speech forms ending in verbal terminations (a-tiñ). Patañjali writes,

Therefore, since it is successful in this way, the fact that the teacher (Panini) mentions the negation, "not a speech form ending in a verbal termination," serves to make known that the terms sup and supa do not recur in these two rules (A. 2.2.18-19). What is the reason for making this known? The principle (paribhasa) that a gati, karaka, or upapada is compounded with a nominal base ending in a kit-affix need not be stated, (evam tarhi siddhe sati yad atiñ iti pratisedham scisti taj jnapayaty acaryo 'nayor yogayor nivrttam sup supaiti. kirn etasya jnapane prayojanam. gatikarakopapadanam krdbhih saha samaso bhavatity esa paribhasa na kartavya bhavati. MBh. 1.417.1 8-20)4

Finally, Patañjali clarifies that the final compound constituent with which elements termed gati, speech forms denoting participants in action (karakas), and upapadas are compounded is simply a semantically and syntactically related speech form. The Mahabhasya passage continues, "If this is made known, then with what are they compounded? With a semantically and syntactically related speech form" (yady etaj jnapyate kenedanim samaso bhavisyati. samarthena). The final compound constituent can be any speech form; it need not be one that ends in a nominal termination.

As mentioned in section 2, Katyayana's varttikas 3-4 under A. 2.2.19 and Patañjali's commentary thereon conclude that the formation of an upapada-tatpurusa compound by A. 2.2.19 takes precedence over the formation of a sasthi-tatpurusa compound by A. 2.2.8. Varttika 3, "an upapada-tatpurusa compound takes precedence over a sasthi-tatpurusa compound by vipratisedha" (sasthisamasad upapadasamaso vipratisedhena. MBh. 1.418) suggests that A. 2.2.19 takes precedence after the two rules come into conflict (vipratisedha) since each rule has its own scope while they both have scope in the formation of compounds such as kumbhakara. Presumably, the latter rule would apply in accordance with the principle stated in A. 1.4.2 vipratisedhe param karyam that the latter rule applies in cases of such conflict. Katyayana in varttika 4 and Patañjali in his comments thereon reject varttika 3' s suggestion that A. 2.2.19 takes precedence over A. 2.2.8 by vipratisedha. Varttika 4 states, "no, an upapada-tatpurusa compound occurs because there is no sasthi-tatpurusa compound" (na va sasthlsamasasyabhavad upapadasamasah. MBh. 1.418). A. 2.2.8 has no scope to form compounds such as kumbhakara, Patañjali points out, because of the statement of the principle that a gati, karaka, or upapada is compounded with a nominal base ending in a krt-affix prior to the arising of nominal terminations (gatikarakopapadanam krdbhih saha samasavacanam prak subutpatter iti vacanat. MBh. 1.418.7-8). Since he has just argued that the statement of this principle is not necessary, he offers a second reason: A. 2.2.19 is obligatory while A. 2.2.8 is optional (upapadasamaso nityasamasah sasthisamaso vibhasa. MBh. 1.418.10). An obligatory rule takes precedence over one that is not obligatory.

The fact that Katyayana and Patañjali consider the possibility that the compound be formed by A. 2.2.8 sasthl implies that they consider that a sixth-triplet nominal termination is present in the initial compound constituent at the stage of compounding (Table 1, step 17). The statement of the principle that a gati, karaka, or upapada is compounded with a nominal base ending in a krt-affix prior to the arising of nominal terminations preempts the occurrence of a nominal termination only in the final compound constituent. The inclusion of a-tiñ in A. 2.2.19 that makes the statement of this principle unnecessary likewise preempts the occurrence of a nominal termination only in the final compound constituent.

Although Katyayana and Patañjali accept that the initial compound constituent in an upapada-tatpurusa compound terminates in a nominal termination prior to compounding, Patañjali nowhere insists that the term upapada itself implies the presence of nominal terminations. Hence there is no reason to assume the presence of a nominal termination in kumbha at the time of application of A. 3.2.1 karmany an (Table 1, step 7) just because that which denotes the direct object (karman) is termed upapada by A. 3.1.92 tatropapadam saptamistham (Table 1, step 5). The question arises as to whether the term upapada implies the technical sense of the term pada provided by A. 1.4.14 suptiñantam padam, namely, that it terminate in a nominal termination because the term upapada includes the string pada. A similar question arises with regard to the terms for compound constituents pürvapada and uttarapada. The answer is that the terms do not imply the technical sense of the term pada provided by A. 1.4.14; they do not necessarily have to terminate in nominal terminations.

Under A. 3.1.92, Patañjali accepts that the reason for stating the long technical term upapada is that it be understood as a term in accordance with its conventional meaning (mahatyah sanjnayah karana etat prayojanam anvarthasañjña yatha vijnayeta. MBh. 2.76.7-8). The term upapada is a long term (upapadam iti mahatîyam sañjña kriyate. MBh. 2.76.6). The conventional meaning to be understood from it is the adjacent word uttered (upoccari padam upapadam. MBh. 2.76.8). The hint of the word pada in the term upapada serves to induce the principle in rules in which the term is mentioned that the rule concerns syntactically related speech forms (yavata cedanim padagandho 'sti padavidhir ayam bhavati. padavidhis ca samarthanam bhavati. MBh. 2.76.9-10). The term thereby prevents rules from applying to syntactically unrelated speech forms. The point is that the speech forms must be syntactically related, not that they terminate in nominal terminations.

In this context, Patañjali debates the application of A. 3.2.1 karmany an to cases where the verb has an independent connection with two verbal complements not directly connected with each other. He considers the case in which the vigrahaväkya contains two accusatives, mahäntam kumbham karoti. If the sentence means "he makes a great pot," there is a direct syntactic connection between the two accusatives, and these have a uniform connection with the verb. In that meaning Patañjali permits the rule to apply (bhavitavyam yadaitad väkyam bhavati: mahän kumbho mahäkumbhah, mahäkumbham !carotiti mahäkumbhakärah. MBh. 2.75.22-76.1). However, if the sentence means "he makes the pot large," there is no direct unmediated syntactic connection between kumbham and mahäntam, so the rule does not apply (yadä tv etad väkyam bhavati: mahäntam kumbham karotiti tadä na bhavitavyam. MBh. 2.76.2). In that case Patañjali disallows the rule to apply because of the lack of syntactic connection (tatra asämärthyän na bhavisyati. MBh. 2.76.10). He does, however, make an exception in the similar case of at least one compound formed with the affix evi (istam evaitad gonardiyasya). In the sentence "I want a maker of wild sugarcane grass into mats" (icchämy aham käsakatikäram), A. 3.2.1 does provide the affix an after kr with two complements käsa 'wild sugarcane grass' and kata 'mat' (MBh. 2.76.13-14).

While the debate concludes by broadening the scope of rules that include an upapada as a condition so that they include cases of slightly looser syntactic connection, it illustrates well what Patañjali means the purpose of stating the long term upapada to be: it indicates that rules apply to syntactically related speech forms, not to speech forms that are not syntactically related. Patañjali makes no mention of a requirement that the hint of the word pada (pada-gandha) in the term upapada implies that an upapada in a rule such as A. 3.2.1 must terminate in a nominal termination in accordance with the formal requirements of A. 1.4.1 suptiñantam padam. Hence there is no need for a nominal termination in kumbha at the time of application of A. 3.2.1 (Table 1, step 7). In contrast, the reason a nominal termination is required in kacchena at the time of application of A. 3.2.4a supi (Table 2, step 12) is that the rule specifically refers to a speech form ending in a nominal termination sup.

3.3 Jinendrabuddhi and Bhoja

According to Jinendrabuddhi (c. 750 ce.) in his Nyäsa on the Käsikä (seventh century) under A. 2.2.19 upapadam atiñ, nominal terminations are generally present neither in the initial nor in the final compound element in upapada-tatpurusa compounds. He considers that the mention of the term a-tiñ serves as an indication that neither sup (from A. 2.1.2) nor supä (from A. 2.1.4) recurs in A. 2.2.19. Jinendrabuddhi considers it appropriate that neither term recurs (yuktä dvayor api nivrttih) because the indication applies generally to interrupt the nominal termination heading (sämänyena sub-adhikära-nivrtty-upalaksanärthatvät). He considers that the term upapada does not necessarily mean a speech form ending in a nominal termination (subantam) in accordance with the technical sense of pada in A. 1.4.14 suptiñantam padam. First, in accordance with Patañjali' s statement under A. 3.1.92, he accepts that the term upapada includes not only what is taught in the locative under the heading A. 3.1.91 in accordance with A. 3. 1 .92 tatropapadam saptamistham (na hi dvitiyadhätvadhikäre yat saptamyä nirdistam tad evopapadasamjñam bhavati) but also that which is enunciated nearby (api tu yad apy upoccäritam padam tad apy upapadam bhavaty eva). Moreover, he takes the term pada in upapada to mean "that by means of which a meaning is understood" (padatvam punas tasyapadyate gamy ate 'nenärtha iti krtvä), not "that which ends in a nominal termination" (na tu subantatvät). The reason he interprets pada in this way is that it is impossible that a nominal termination occur after the final compound element in the derivation of forms such as asvakrltl 'a female bought with a horse' (iha supo 'sambhavät). In asvakrïtï as in kacchapi, nominal terminations occur after the feminine affix (Table 2, steps 23, 23a, 23b). The feminine affix I occurs in accordance with A. 4.1.50 krltät karanapürvät only after the compound asva-krlta is formed (cf. Table 2, step 19); before compound formation, the affix ä would occur after the final compound element krlta in accordance with A. 4.1.4 ajädyatas tap (Table 2, step *). Therefore, nominal terminations do not arise in upapada-tatpurusa compounds prior to compounding (subanutpatteh präk samäsät).

Jinendrabuddhi is aware that there are cases in which nominal terminations are required after the first compound element. To account for these, he asserts that the indication that neither sup nor supä recurs does not apply universally (asarvavisayatväd asya jnäpakasya). He asserts that the indicated principle (paribhäsä) that compounding occurs prior to the provision of nominal terminations for certain speech forms including upapadas does not apply universally (na hy anena sarvatra ' gatikärakopapadänäm krdbhih präk subutpatteh samäso bhavatV iti jnäpyate). Rather (kith tarhi) it applies only in certain desired instances (kva cid evesta-visaye). It is known that Panini permitted nominal terminations to occur at the end of the initial compound element before a final element ending in a krt-affix because he allows nominal terminations not to be deleted in such compounds. A. 6.3.14 provides non-deletion (a-luk) of a seventh-triplet nominal termination before a final compound element that ends in a krt-affix (tatpuruse krti bahulam iti krdanta uttarapade saptamyä alug-vidhänät). If nominal terminations never arose at the end of initial compound elements before final compound elements that end in krt-affixes, it would make no sense to provide for the non-deletion of seventh-triplet nominal terminations because they would not have arisen in the first place. Moreover forms such as bilesaya 'lying in a cave', where the singular seventh-triplet termination occurs, would not be accounted for. Therefore, Jinendrabuddhi concludes that a compound occurs prior to the arising of nominal terminations only in certain instances (tasmät kvacid eva präk subutpatteh samäsah), not universally (na sarvatra). In this way one can account for dhanakrltä, where the feminine affix fäp does occur after the final compound element prior to compounding, as well as dhanakrltl, where it doesn't. In the latter, the final compound element is left ending in a short a so that instead the feminine affix ñls occurs after compound formation (cf. Table 2, step 19).

In his Ergäraprakäsa (1005-1062 ce.), Bhoja agrees with Jinendrabuddhi on the one hand that neither sup nor supä, which specify that compound constituents end in nominal terminations, recurs in A. 2.2.18-19, and on the other that the principle that a gati, käraka, or upapada is compounded with a nominal base ending in a krt-affix prior to the arising of nominal terminations does not apply absolutely. He argues that the term sup does not recur because the mention of a-tiñ, which is explained as a separate sütra divided from A. 2.2.19 that completes both A. 2.2.18 and A. 2.2.19, stops it ('kugatiprädayah' ', "upapadam atiñ' ity atra atingrahanenobhayasütrasesatayä vyäkhyäyamänena sub ity etasya nivrttih kriyate. ÉPr., p. 46). Likewise the term supä does not recur because in A. 2.1.32 it is understood that compounds form at the stage where the final constituent ends in a krt-affix. One gets that the final constituent ends in a krt-affix already just by the fact that the compound is provided for initial constituents that denote an agent (kartr) or an instrument (karana). (Agents and instruments are participants in action. Action is denoted by roots, and krt-affixes are provided after roots. Hence the only speech forms that denote participants in action that take nominal terminations are krt-derivates.) Because krt is specifically mentioned even though one already understands this, its mention particularly indicates a krt-final nominal base without a nominal termination, ('kartrkarane krtä bahulam' ity atra ca kartrkaranayoh samäsavidhänäd uttarapadasya krdantatäyäm labdhäyäm krdgrahanäd atiriktät tadantävasthäyäm eva samäsäbhyanujnäne supety etad api nivartate. EPr., p. 46.) Bhoja concludes that the nonrecurrence of these terms justifies the formation of gati, käraka, and upapada compounds from speech forms that don't end in nominal terminations.

Conversely, Bhoja concludes that the inclusion of the term bahulam in A. 2.1.32 allows such terminations where desired.

Therefore, the principle that a gati, käraka, or upapada is compounded with a nominal base ending in a krt-affix prior to the arising of nominal terminations is made known. And it is determined that compounding occurs in some instances between two nominal bases, in some instances between two speech forms ending in nominal terminations, and in some instances between a speech form ending in a nominal termination and a nominal base because the mention of 'variously' (bahulam) in A. 2.1.32 serves the purpose of achieving whatever is desired. (tatas ca gatikärakopapadänäm krdbhih saha samäsavacanarh subutpatteh präg èva bhavatîty äkhyätam. bahulagrahanasya cestasiddhyarthatvät sa kvacin nämabhyäm kvacit subantäbhyäm kvacin nämasubantäbhyäm nisciyate. (EPr., p. 46 with correction of sentence and paragraph segmentation.)

Bhoja cites and justifies examples of compounds that require nominal terminations on initial constituents (carmakärah) and on final constituents (dadhisek, dhanakritä). He also cites and justifies examples of compounds that require the absence of nominal terminations on initial constituents (asüryampasyä) and on final constituents (dhanakriti). It is necessary to allow the initial or final element in käraka and upapada compounds to end in a nominal termination to account for operations on the initial or final element that can only occur under the condition that it is termed pada. A. 1.4.14 suptiñantam padam provides that a speech form that ends in a nominal or verbal termination is termed pada. A number of rules in the eighth adhyäya of the Astädhyäyi provide operations that take place at word boundaries. For example, A. 8.3.109 sätpadädyoh negates retroflexion of the initial s of a pada where retroflexion would otherwise occur after a simple vowel other than a or ä located in a prior compound element by A. 8.3.104 pürvapadät. Many rules provide replacements to sounds that occur final in a pada. Thus A. 8.2.7 nalopah prätipadikäntasya occurs in the section headed by A. 8.1.16 padasya. Thereby the term padasya is understood to recur in A. 8.2.7. This rule then provides the deletion of the final n in a nominal stem (prätipadika) that is termed pada. The rule accounts for the deletion of the n of räjan 'king' in the masculine nominative singular räjä, and in oblique forms beginning with a stop or spirant such as the instrumental plural räjabhih and locative plural räjasu. A. 8.2.7 likewise accounts for the deletion of the final n when the word occurs as the initial element in compounds such as räja-purusa.

Bhoja gives dadhi-sek 'yogurt-sprinkler' as an example of a compound requiring its final constituent to end in a nominal termination (SPr., p. 46). A. 8.3.109 sätpadädyoh negates retroflexion of the initial s of sek if it is a pada. If the final compound constituent sec did not end in a nominal termination prior to compounding, it would not be termed pada by A. 1.4.14, and the initial s of sec would be subject to retroflexion by A. 8.3.104.

As an example of a compound formed from an initial constituent ending in a nominal termination and a final constituent consisting of a nominal base, Bhoja gives carmakärah 'leather- worker' (EPr., p. 46). The compound carma-kära is an upapada-tatpurusa compound accounted for by A. 2.2.19 just as kumbha-kära is (Table 1, step 17). Prior to compound formation, A. 3.2. 1 provides the affix an after the root kr when carman occurs as an upapada in relation to the root kr, just as it does when kumbha occurs as an upapada in relation to the same root (Table 1, step 7). The deletion of the final n of carman 'leather' is required when it occurs as the prior member in the compound carma-kära. Now if the prior element did not end in a nominal termination, it would not be termed pada by A. 1.4.14, and the deletion of the final n would not occur by A. 8.2.7.

Bhoja writes, "the final subsequent compound constituent in dadhisek is made to end in a nominal termination to achieve the negation of replacement by retroflex s initial in a pada by A. 8.3.109 sätpadädyoh, and the prior compound constituent in carmakära is made to end in a nominal termination for the purpose of deletion of pada-ñnai ?." ('dadhisek'' ity atra uttarapadasya, 'sätpadädyoh' itipadädi-nibandhana-satva-pratisedha-siddhaye 'carmakära' ity atra tu pürvapadasya padänta-laksana-nalopärtham subantatä kriyate. EPr., p. 46.)

Why dhanakrîtî requires the absence of nominal terminations on the final constituent to condition the feminine affix ñis, and dhanakrltä requires their presence to condition the feminine affix fäp has been explained above. Finally, Bhoja cites asüryampasyäh. The initial constituent a-sürya, he asserts, is a compound formed from the nominal bases nan (the negative particle with the final marker ñ) and sürya 'sun' without nominal terminations.

There are no nominal terminations on nañ and sürya in asüryampasya because nan and sürya are not in direct syntactic connection. The negative particle nan and sürya 'sun' are mutually unconnected because negation denoted by nan and the sun denoted by sürya are both connected with the action of seeing denoted by the root drs (and by the present stem pasya which replaces it by A. 7.3.78 päghrädhmä . . .). For here, in the corresponding phrase, "They don't see even the sun" (süryam api na pasyanti), the negative particle nañ expects the action of seeing which has the sun as its direct object; it does not expect the entity the sun directly. The compound is formed just of the two nominal bases (ñaman), nañ and sürya (devoid of nominal terminations), even though they are not syntactically connected, because of the explicit mention of a-sürya in A. 3.2.36 asürya-lalätayor drsi-tapoh. ('asüryampasya' ity atra nañsüryayor drsikriyayä sambandhät parasparam asambandhe sämarthyäbhäväd vibhaktyabhävah, atra hi süryam api na pasyantîti nañ süryakarmikäm drsikriyäm apeksate, na süryasattäm, ' asüryalalätayor drsitapoh' (A. 3.2.36) iti vacanäd asämarthye 'pi nämnor eva samäso bhavati. ÉPr., pp. 46-47.)

Jinendrabuddhi and Bhoja understand Patañjali (see section 3.2) to mean that neither sup nor supä recurs in A. 2.2.18-19 and that the mention of a-tiñ allows both initial and final compound elements in upapada-tatpurusa compounds to lack nominal terminations at the time of compound provision. They account for the required presence of nominal terminations on these elements at the time of compounding in numerous examples by broadening the scope of indeterminate variation indicated by the term bahulam in A. 2.1.32 kartrkarmanoh krti bahulam. Rules of indeterminate variation carry a cost to the robustness of linguistic description. Linguistic science in general and Päninian grammar in particular engage in the systematic explanation of language. Rules of indeterminate variation should be appealed to as little as possible to preserve the robustness of the scientific explanation. As I wrote (2008: 16), paraphrasing Thieme (1935: 61), "it is likely that Panini formulated such rules to account for such unusual occurrences after he had exhausted all attempts at systematic explanation." I therefore concluded (p. 15), "the new school account of the subjunctive is more convincing than the old school account because it provides a more precise systematic account of a larger scope of data than the old school and relies on rules of indeterminate variation for a smaller scope of data." It is the undesirability of broad rules of indeterminate variation that prompts Kaiyata, Haradattamisra, and later grammarians to frame the rules regarding terminations on compound constituents more precisely. 5

3.4 Kaiyata and Haradattamisra

In disagreement with Jinendrabhuddhi and Bhoja, Haradattamisra (c. 1100 c.e.) in his Padamañjarl on A. 2.2. 1 9 upapadam atiñ understands that the paribhäsä gatikärakopapadänäm krdbhih präk subutpatteh samäso bhavati concerns the occurrence of nominal terminations only after the final compound element. He writes that the principle means, "the compound formed from a gati, käraka, or upapada as initial element compounded with a krt-derivate as final element is to be formed before the occurence of nominal terminations on the final element, but the initial element does indeed terminate in a nominal termination when it compounds" (gatlnäm kärakänäm upapadänäm ca krdbhih saha yah samäsas tena tena laksanena sa uttarapadät subutpatteh präg èva käryah, pürvapadam tu subantam eva samasyate). With Bhoja's remarks regarding asüryampasya in view, he apparently mocks his predecessors who allow terminations to occur at random and apparently pays respect to Kaiyata's Pradlpa commentary on Patañjali's Mahäbhäsya. For he concludes with the verse

tad etat pratipadyantäm bhäsye krtaparisramäh.

nänye sahasram apy andhäh süryam pasyanti nänjasä.

Let those who have exerted effort in the Mahäbhäsya understand this;

Even a thousand other blind people do not see the sun without ointment.

In his Pradlpa commentary on the Mahäbhäsya on A. 2.2.19, Kaiyata (c. eleventh century ce.) remarks on Patañjali's statement that a-tiñ serves to make known that the terms sup and supä do not recur in the two rules A. 2.2.18-19. He writes that the inclusion of the term a-tiñ indicates that only the term supä ceases to recur, but the term sup does indeed recur in order to allow operations that take place on a pada to occur on the initial constituent (tena supety asyaiva nivrttir jnäpyate. subgrahanam tu pürvasya padasya padakäryärtham anuvartata eva). The term supä in the instrumental indicates that the final compound element ends in nominal terminations; its cessation allows the final element not to have nominal terminations. Kaiyata takes the term a-tiñ in apposition to the heading samäsah in A. 2.1.3 präk kadärät samäsah. That the resulting compound is a-tiñ amounts to making the final compound element, which ends in a krt-affix, a-tiñ.6

Kaiyata initially rejects the example dadhi-sek 'yogurt sprinkler' adduced by Bhoja as evidence of an upapada-tatpurusa compound whose final compound constituent has nominal terminations prior to compound formation. Kaiyata suggests that the compound is not an upapada-tatpurusa compound formed by A. 2.2.19 upapadam atiñ at all; rather it is a sasthitatpurusa compound formed by A. 2.2.8 sasthî. The final element is not a derivate formed on condition that an upapada occurs in syntactic connection with a root; rather, it is a derivate formed by provision of the affix kvip after the causative of the root sic without dependence upon an upapada by A. 3.2.178 anyebhyo 'pi drsyate (kvip 177). 7 Since there is no doubt that A. 2.2.8 requires nominal terminations on both constituents, it is clear that the final constituent sec is termed pada and is subject to the negation of retroflexion stated in A. 8.3.109. Kaiyata represents the rejected view as follows:

But if a nominal termination does not arise after the final compound constituent, then in compounds such as dadhisecau (masculine or feminine nominative or accusative dual of dadhisec), the negation of replacement by retroflex s by A. 8.3.109 sätpadädyoh would not occur because the dental s does not occur at the beginning of a pada. And because it is not termed pada, it cannot be designated a final compound constituent uttara-pada and therefore the accent that depends upon it being so termed would not succeed, (yadi tarhy uttarapadät sub notpadyate tadä dadhisecäv ity adau padäditväbhävät satvapratisedho na präpnoti. padatväbhäväd uttarapadavyapadesas ca na, tatas ca tannibandhanasvaro na sidhyati.)

A. 6.2.139 gatikärakopapadät krt (uttarapada 6.2.111) provides that the original accent of the final constituent following a gati, käraka, or upapada is retained in a tatpurusa compound. Kaiyata rejects the objection:

This is not a problem. The negation of replacement by retroflex s will occur because a sasthitatpurusa compound will be formed after having provided the affix kvip following the causative of the root sic without an upapada. (naisah dosah. nirupapadät secayateh kvipi krte sasthisamäsah kriyata iti satvanisedho bhavisyati.)8

However, Kaiyata subsequently withdraws his suggestion for reinterpreting the compound dadhi-sec as a sasthi-tatpurusa instead of an upapada-tatpurusa because he recognizes that it is necessary to accept indeterminate variation regarding the presence or absence of nominal terminations on the final compound constituent anyway in order to account for dhanakritä, which requires nominal terminations on the final compound constituent, as well as dhanakriti which requires the absence of nominal terminations on the final compound constituent (see section 3.3).

Or rather, since the term supä ceases to recur, in order to achieve operations as desired, a compound is formed in some instances after a nominal termination has arisen and in some instance before a nominal termination arises. In this way, because of the mention of bahulam 'variously', a käraka compound too occurs in some instances after a nominal termination has arisen. Thus the usage "For she is his woman bought with money (dhanakritä)" is accounted for.9 (yad vä supety asya nivrttau satyärh yathestam käryasiddhaye kvacid utpanne supi samäsah kvacit präk subutpatteh. evam kärakasamäso 'pi bahulagrahanät kvacit subutpatter bhavatiti sä hi tasya dhanakriteti prayoga upapannah.)

Now if the final compound constituent in dadhi-sec is accepted as being a nominal base without nominal terminations at the time of compound formation, it remains to justify the accent in accordance with A. 6.2.139, which requires that the final compound element be termed uttarapada. Kaiyata concludes that the term uttarapada conventionally refers to any speech form that occurs as a subsequent compound constituent. It does not refer to what is termed pada in the technical sense of the term; that is, its scope is not limited to what ends in nominal or verbal terminations as required by A. 1.4.14 suptiñantam padam. Kaiyata therefore concludes, "there is no problem there either because the term uttarapada conventionally refers to a particular part of a compound" (uttarapadasabde samäsävayavavisesasya rüdhir iti taträpy adosah). Regarding the accentual rule A. 6.2.139, which provides that the original accent of the final constituent following a gati, käraka, or upapada is retained in a tatpurusa compound, Joshi and Roodbergen (1973: 223) clearly state, "in these rules the term uttarapada does not mean a case-inflected final cp.-member, that is, a pada in the technical sense of the word, but it only means the final part of a cp."

According to Kaiyata, the recurrence of sup in the nominative in A. 2.2.19 requires that the initial compound constituent terminate in a nominal termination, not the final compound constituent. The initial constituent is then termed pada in the technical sense of the term by A. 1.4.14. Since the initial compound constituent is termed pada, the principle stated in A. 2.1.1 samarthah padavidhih, which is relevant to rules concerning a pada, applies. The principle restricts compound formation to semantically and syntactically connected speech forms. Where Patañjali writes under A. 2.2.19 that an element termed gati, a speech form denoting a participant in action (käraka), or an upapada is compounded with a semantically and syntactically related speech form (see section 3.2), Kaiyata states that this is due to the fact that the principle of semantic and syntactic connection presents itself because compound formation is a rule concerning a pada by virtue of the fact that the term sup recurs (sub ity asyänuvrttau satyäm samäsasya padavidhitvät samarthaparibhäsopasthänät). The final compound constituent, according to Kaiyata then, can be any semantically and syntactically related speech form.

3.5 Bhattojidiksita, Nägesa, and their commentators

Bhattojidiksita (early seventeenth century) adopts the view propounded by Kaiyata and Haradattamisra that in upapada-tatpurusa compounds the initial compound constituent terminates in a nominal termination and that just the final compound constituent does not. He goes further in stating that it is a pada that is termed upapada by A. 3.1.92 tatropapadam saptamistham in rules under the heading A. 3.1.91 dhätoh. He thereby departs from Jinendrabuddhi's conclusion that the term upapada does not include the technical sense oí pada as that which ends in a nominal or verbal termination (see section 3.3). He makes clear that the nominal termination present at the time of compounding by A. 2.2.19 upapadam atiñ in the derivation of kumbha-kära is a sixth-triplet termination (a genitive ending), not a secondtriplet one (an accusative ending). Nägesa (eighteenth century) concurs.

In the Siddhäntakaumudi, Bhattojidiksita makes clear that a nominal termination occurs at the end of the upapada, which occurs as the initial member of the compound, but not on the derivate formed from the root, which occurs as the final member. The term sup, designating the subordinate compound element that ends in a nominal termination, recurs in A. 2.2.19 from A. 2.1.2 sub ämantrite parängavat svare, but the term supä in A. 2.1.4 saha supä, designating a principal compound element that ends in a nominal termination, does not. He writes under A. 2.2.19 upapadam atiñ, "an upapada that ends in a nominal termination is obligatorily compounded with a syntactically connected item" (upapadam sub-antam samarthena nityam samasyate). In contrast, he states that the term supä in the instrumental does not recur from A. 2.1.4 (supä iti ca nivrttam). It is the absence of the nominal termination on the subsequent compound element, the derivate -kära, at the time of compound formation by A. 2.2.19 that satisfies the principle (paribhäsä) that the compounding of an upapada with a krt-derivate occur prior to the arising of a nominal termination (tathä ca 'gatikärakopapadänäm krdbhih saha samäsavacanam präk subutpatteh' iti siddham).

Väsudevadiksita provides the example of carmakära in the Bälamanorama to demonstrate the necessity of understanding that a nominal termination occurs generally after the prior element in upapada-tatpurusa compounds. The deletion of the ? in carman is required if it occurs as an upapada in an upapada-tatpurusa compound. He writes, "nor can one argue that there is no reason for the term sup to recur (in A. 2.2.19 from A. 2.1.2) because it serves the purpose of the deletion of n in carma-kära" (na caivam sub ity anuvrtteh prayojanäbhäva iti väcyam, carmakära ity atra nalopärthakatvät).

In order to demonstrate that no nominal termination occurs after the subsequent compound element that is a krt-derivate in an upapada-tatpurusa compound, Bhattojidiksita cites the form kacchapi 'female tortoise'. Väsudevadiksita indicates that the parallel sentential usage that illustrates the meaning of the compound is either kacchena pibati ". . . drinks by means of the edge," or kacche pibati ". . . drinks at the edge." He writes kacchah tiram, tena tasmin vä pibatiti kacchapi. The derivation of the form shown in Table 2 assumes the first meaning.

In a departure from the views of Kaiyata and Bhoja, Bhattojidiksita introduces an innovation in stating that it is a pada that is termed upapada in a sutra of the section headed by A. 3.1.91 dhätoh. He writes under A. 3.1.92 tatropapadam saptamistham,

When there is a word ending in a seventh-triplet nominal termination, such as karmani, a pada such as kumbha that denotes a direct object (karman), present as that which is to be denoted by the term karmani, is termed upapada. And only when that is present does the affix that will be provided occur, (saptamy ante pade karmani ity ädau väcyatvena sthitam kumbhädi tadväcakam padam upapadasamjñam syät. tasmims ca saty eva vaksyamänah pratyayah syät.)

In the Bälamanorama, Väsudevadiksita writes thereon,

The affix an occurs after the root in the meaning of an agent, but the pada that denotes the direct object (karman), such as kumbha, is to be understood as termed upapada. The result is that only when the upapada is present does the affix an occur, (dhätor an syät kartary arthe, karmaväcakarh tu kumbhädipadam upapadasañjñam pratyetavyam. tasminn upapade saty evän syäd iti phalati.)

Commenting on A. 3.1.92 in his Laghusabdendusekhara, Nägesa too insists that the upapada terminates in a nominal termination in A. 3.2.1 karmany an as well as in sütras in which the term supi occurs. In commenting on Bhattojidiksita's use of the term pada in the phrase "the pada that denotes that (karman)" (tadväcakam padam), he states, "a pada here ends in a nominal termination" (padam atra vibhaktyantam). Bhairavamisra, in his commentary Cadrakalä on the Laghusabdendusekhara, summarizes Nägesa's conclusion, "the term upapada applies only to a pada" (padasyaivopapadasamjña).

Nägesa confirms that the prior element in upapada-tatpurusa compounds ends in a nominal termination, commenting on A. 2.2.19 in the Laghusabdendusekhara. He writes that Bhattojidiksita's qualification of the term upapada with the term 'ending in a nominal termination' (subanta) is gotten by force of the fact that it is a long term (subantam iti mahäsanjnäbalalabdham). The Candrakalä glosses Nägesa's use of the term mahäsanjnä under A. 3.1.92 stating that a long term is used for the purpose of indicating a sense in accordance with its meaning. In this case that meaning is the word (pada) enunciated nearby (sä cänvarthatväya krtä - samïpa uccäritam padam iti). Nägesa considers that any use of the term upapada refers to a word that ends in a nominal termination. Nägesa makes very clear, in sharp contrast to Jinendrabuddhi, that he considers that the term upapada includes the term pada in its technical sense, even in sütras headed by A. 3.1.91. He interprets the principle stated in A. 3.1.92 tatropapadam saptamlstham in application to A. 3.2.1 karmany an to mean that the word ending in a nominal termination that denotes the direct object (karman) is termed upapada.

Likewise in his Paribhäsendusekhara, Nägesa writes that the reason for stating paribhäsä 76 gatikärakopapadänäm krdbhih saha samäsavacanam präk subutpatteh is to prevent the incorrect feminine affix ä (fäp) from occurring on the final compound element asvakrltl, vyäghrl, and kacchapl. The correct affix I (ñls) occurs after the compound stem and requires that the compound be formed prior to the occurrence of nominal terminations. That the paribhäsä is not obligatory (nitya) allows ä where it occasionally occurs, as in asvakrltä. Alternatively, such words are included in the list ajädi, allowing fäp to occur in exception to ñls by A. 4.1.4 ajädyatas tap, and the paribhäsä is obligatory, including in cases such as kumbhakära. The paribhäsä does not prevent terminations from occurring after the initial compound element. Quite the contrary. Not only does Nägesa want the termination after the initial compound element prior to compound formation, he wants it prior to provision of the krt-affix at} that forms the final compound constituent.

The termination that occurs on the initial compound element is a sixth-triplet nominal termination, not a second-triplet nominal termination. The sentence with an accusative ending is provided just as an actual usage in parallel meaning, not as a prior step in the derivation of the compound. Immediately after he gives the example kumbhakära and shows its meaning with a parallel sentential usage that contains the word kumbha in the accusative (i.e., with a singular second-triplet nominal termination) (kumbham karotiti kumbhakärah), Bhattojidiksita shows a grammatical formula at a step in the derivation prior to the formation of the compound. The derivational formula contains a singular sixth-triplet nominal termination (iha kumbha as kära ity alaukikam prakriyäväkyam). Väsudevadiksita makes clear in the Bälamanorama that the grammatical formula, not the parallel sentential usage, is the precondition for the derivation of the compound: "the essence is that only the grammatical formula is the basis for the occurrence of the compound; the sentence kumbham karoti is merely for showing its meaning" (alaukika-vigraha-väkya èva samäsa-pravrttih. kumbham karotiti tadartha-pradarsana-mätram iti bhävah). He goes on to emphasize that a sixthtriplet nominal termination, not a second-triplet one, occurs in the derivational formula. He states that kumbha-am kära is an erroneous reading because the sixth triplet is provided in conjunction with a krt-derivate (kumbha am kära ity apapäthah, krdyoge sasthyä vidhänät). The sixth triplet occurs in accordance with A. 2.3.65, as explained above and shown in Table 1, step 12b.

Nägesa likewise affirms that it is a sixth-triplet nominal termination and not a secondtriplet nominal termination that occurs at the end of the word kumbha in the derivation of the upapada-tapurusa compound kumbha-kära. The sixth triplet provided by A. 2.3.65 kartrkarmanoh krti occurs in exception to A. 2.3.2 karmani dvitiyä. It is not the case that the latter takes precedence over the former by virtue of the principle of being more internally conditioned (antarañga). He entertains the supposition that A. 2.3.2 would take precedence over A. 2.3.65 because A. 2.3.65 depends on the direct object having connection with action denoted by a krt-affix because the sutra states krti. He rightly dismisses this suppostion because A. 2.3.2 equally depends upon the direct object having connection with action, even without mentioning a term referring to action, just by virtue of a direct object (karman) being a participant in action (käraka): "and here a sixth triplet occurs conditioned by connection with a krt-affix - nor is the second triplet more internally conditioned (antarañga) - because a general rule applies considering the domain of its exceptions" (krdyogalaksanä cätra sasthi. na cäntarahgä dvitiyä. prakalpyäpavädavisayam utsargapravrtteh). At the same time Nägesa denies that kumbha-kära is a sasthi-tatpurusa compound formed by A. 2.2.8. The reason he denies this is that an upapada-tatpurusa compound formed by A. 2.2.19 is more internally conditioned because it is provided prior to the arising of nominal terminations on the final compound constituent (atra sasthisamäso na. uttarapade vibhaktyutpatteh pürvam eväsya pravrttyäntarangatvät). 10

3.7 Joshi and Roodbergen

Joshi and Roodbergen (1973: 42) accept that sup recurs in A. 2.2.18-19 and just supä is discontinued, in disagreement with Jinendrabhuddhi and Bhoja, and in agreement with Kaiyata, Haradattamisra, Bhattojidiksita, and Nägesa. The result is that in an upapada-tatpurusa compound an initial compound constituent that ends in a nominal termination is compounded with a final constituent that is any syntactically related speech form. They comment, "Tradition rejects the continuation of the condition sup supä as a whole. . . . Our assumption is that supä is discontinued on the basis of sämarthya. . . . The point is that the cp.-constitutents are joined together before a case ending (or a fem. suffix) is added to the second cp.-constituent." Likewise they write (p. 203), "But the fact is that in all desired upapada cps the upapada is always a case-inflected word. . . . What we want is the discontinuation of sup with reference to the word with which the upapada is to be compounded. That is to say, we want the discontinuation of the condition supä." They correctly observe (p. 220), regarding the derivation of kacchapl, "in order to derive the desired form, the upapada is compounded with a krdanta stem, before the case-ending has been added." The purpose of discontinuing supä is to prevent the feminine affix tap from arising after the stem pa in kacchapl, after the stem ghra in vyäghrl, and after the stem krlta in dhanakrltl, asvakrltl, vastrakrltl, etc. They provide derivations of several of these forms.11

Working out the details of the derivation of the forms under discussion in the commentaries brings problems to light that went unnoticed previously. One such problem is determining exactly which nominal termination is present on the upapada prior to compounding. A second is determining the sequence of the provision of the nominal termination on the upapada and the provision of the krt-affix. In the derivation of kacchapl, Joshi and Roodbergen show the first step as (kaccha-am + pä-ka) with the nominal termination after kaccha already present at the time of the provision of the krt-affix after the root pä. There they make the provision of the krt-affix simultaneous with the compounding of the upapada with the krt-derivate kära by A. 2.2.19. 12 They argue that the upapada denoting the karman in kumbha-kära is accusative rather than genitive. In their translation of Patañjali's Mahäbhäsya on A. 2.2.19, they comment (p. 203), "we can derive kumbhakärah: 'pot-maker' from (kumbha + am) + kära," and show a singular second-triplet nominal termination on the upapada kumbha. In their translation of the Astädhyäyl (1997: 45), they analyze the compound differently to show that the upapada denoting the karman is a condition for the affix an: ((kumbha-am + kr)-an)su. Their braces indicate that the second-triplet termination is present before the affix an is provided by A. 3.2.1 karmany an. They rule out a genitive because the rule that provides a sixth-triplet nominal termination requires the presence of the krt-affix in advance. In their translation of A. 2.3.65 kartrkarmanoh krti, they write (1998: 112), "(the sixth case endings are added after a prätipadika) in the sense of kartr 'agent' or karman '(direct) object', given (construction with a word ending in a) fcrr(-suffix) (unless the kartr or karman has already been expressed otherwise)." The phrase, "given construction with a word ending in a krtsuffix," implies that the affix an is already present before the rule applies. They recognize (1973: 232) that a problem of mutual dependence would arise if the sixth-triplet nominal termination provided by A. 2.3.65 were required to be present prior to the provision of the krt-affix an by A. 3.2.1: "R 2.3.65 only applies when the word representing the object is connected with a krdanta form. That is to say, unless kära has been derived we cannot apply R 2.3.65. But in order to derive kära from the root kr- we must show that kr- is accompanied by a karma-upapada." They propose to solve the mutual dependency by having the karman be denoted by a second-triplet nominal termination provided by A. 2.3.2 karmani dvitlyä instead: "the only rule by which we can show that kumbha is a karma-upapada is P. 2.3.2. Therefore the technical analysis should read [(kumbha + am) + kr-] + an (1973: 232)."

Joshi and Roodbergen must be credited with recognizing that the problem of mutual dependency would arise if the sixth-triplet nominal termination denoting a karman were required prior to the provision of the krt-affix an. As pointed out towards the end of section 2 above, the sixth-triplet nominal termination denoting a karman is provided after a nominal base by A. 2.3.65 under the condition that it occurs in connection with a speech form ending in a krt-affix, but the krt-affix an is provided by A. 3.2.1 after a root on the condition that an upapada denoting a karman occurs. It is impossible for the upapada to get the sixth-triplet termination before the root gets the krt-affix that is a condition for getting the sixth-triplet termination. The only solution they see is to provide a second-triplet termination to denote the karman rather than a sixth-triplet termination. They therefore conclude (1973: 238), "For the derivation of the form kära we require an upapada which is a karman: 'object'. In order to assign the sense of karman to the upapada, we can only apply the general rule P. 2.3.2, which prescribes the accusative case and not the genitive. Therefore the analysis of kumbhakärah can only be [(kumbha + am) + kr-] + an."

Historically, their proposal has some merit. Many similar compounds, such as janamejaya, priyamvada, and väcamyama preserve what appears to be an accusative termination on the prior compound constituent. Yet Panini does not analyze them thus. He does not provide non-deletion (aluk) of a second-triplet nominal termination before a final compound constituent (uttarapada), although he provides such non-deletion in some twenty-four rules headed by A. 6.3.1 alug uttarapade for third- through seventh-triplet nominal terminations. Instead he provides the augment mum at the end of the initial compound constituent by 6.3.67-72. The first of these, A. 6.3.67 arurdvisadajantasya mum, provides the augment where the final compound element is formed by adding affixes marked with kh after roots. The affixes khas and khac are provided by A. 3.2.28-47 after roots under the condition that there is an upapada. A. 3.2.28 ejeh khas' provides the affix khas after the rooty/ in the example janamejaya, and A. 3.2.38 priyavase vadah khac and väci yamo vrate provide the affix khac after the roots vad and yam in the examples priyamvada and väcamyama respectively. 2.2.19 then forms upapada-tatpurusa compounds. If Panini had provided non-deletion (aluk) of the second-triplet nominal termination before a final compound constituent (uttarapada), then there would be the possibility that the wrong termination, namely the sixthtriplet nominal termination rather than the second-triplet termination, would enter into usage in examples such as janamejaya, etc. However, since Panini derives such examples with the augment mum instead, there is no such possibility. What looks like an accusative singular in these examples is not, according to Panini; hence it cannot serve as evidence of the provision of a second-triplet nominal termination rather than a sixth-triplet in upapada-tatpurusa compounds like kumbha-kära in Päninian derivation.

In spite of its historical merit, and in addition to the linguistic evidence adduced in its favor in the preceding paragraph being irrelevant, Joshi and Roodbergen's conclusion is untenable. In Pacini's derivational system a second-triplet nominal termination does not have the opportunity to arise. The second-triplet nominal termination is provided after nominal bases by A. 2.3.2 karmani dvitiyä on condition that a karman is to be denoted and under the additional condition that it has not already been denoted. A. 2.3.2 comes under the heading 2.3.1 anabhihite 'not already denoted'. In answer to the question, "Not already denoted by what?" (kenänabhihite?), Jayäditya replies in the Käsikä "by a verbal termination, a krt-affix, a taddhita affix, or a compound" (tinkrttaddhitasamäsaih), citing Kätyäyana' s värttika 5 and Patañjali's comment theron (tinkrttaddhitasamäsaih parisamkhyänam. MBh. 1.441.20-22). Nominal terminations are not provided after nominal bases denoting participants in action until after verbal terminations and krt-affixes are provided after the roots denoting the action to which the participants are subordinate. Even in the equivalent sentence kumbham karoti, derived from kumbha-am kr-u-tip, the verbal termination tip is provided by A. 3.2.78 tiptasetc., prior to the provision of the nominal termination am by A. 4.1.2 svaujas, etc. This is necessarily so, because it is only by virtue of being undenoted by the verbal termination tip that the direct object (karman) is denoted by the second-triplet nominal termination by 2.3.2 karmany an. If the karman were denoted by the verbal termination te (< ta), A. 2.3.2 would not apply. The first-triplet nominal termination would occur instead by A. 2.3.46 prätipadikärtha-linga-parimäna-vacana-mätreprathamä, and the passive sentence kumbhah kriyate would result instead. Just as selection of the nominal termination depends upon the selection of the verbal termination in the derivation of the equivalent sentence, it depends upon the selection of the krt-affix in the derivation of the upapada-tatpurusa compound. No nominal termination can arise on the upapada denoting a direct object (karman) until an affix arises after the verbal root denoting the action in which the direct object participates. Therefore, the rule that provides the krt-affix an after the root occurs prior to either of the rules that provide a nominal termination after the upapada come into play. In particular, A. 3.2.1 karmany an applies prior to either A. 2.3.2 karmani dvitlyä or A. 2.3.65 kartrkarmanoh krti coming into play.

Joshi and Roodbergen are correct to state (1973: 238), "for the derivation of the form kära we require an upapada which is a karman: 'object'." However, it is incorrect for them to suggest that A. 2.3.2 has anything to do with assigning the sense of karman to the upapada. They argue (1973: 238), "in order to assign the sense of karman to the upapada, we can only apply the general rule R 2.3.2, which prescribes the accusative case and not the genitive." Moreover, their statement (1973: 232) that A. 2.3.2 is "the only rule by which we can show that kumbha is a karma-upapada" is irrelevant. Such statements confuse the relation between semantics and phonetics in Päninian grammar. Panini does not "assign sense." He does not teach meanings on the ground of phonetic conditions; he teaches speech forms on the ground of semantic conditions. The sense of karman does not depend upon the accusative case or the genitive case; rather second-triplet or sixth-triplet nominal terminations are provided in various contexts under the condition that a karman is to be denoted. That an object is termed karman does not necessarily require any speech form at all; an object may be termed karman under purely semantic conditions without reference to any speech forms whatsoever. Although certainly some rules do take co-occurrence conditions into account, it is essential to note that general käraka rules do not. A. 1.4.49 kartur lpsitatamam karma, for example, terms a pot karman in the derivation of kumbhakära (Table 1 , step 4) under the sole condition that it is most desired by the agent. The pot is termed karman regardless of the speech form used to denote it, and, patently, regardless of the nominal termination (second triplet or sixth triplet) used to denote that it is a karman.

Moreover, the accusative case is not necessary to condition the affix an by A. 3.2.1; only that an object has been termed karman is. It is irrelevant whether or how the presence of such an object can be shown by speech forms. The upapada that serves as a condition for the affix af in A. 3.2.1 must therefore be any semantically and syntactically related speech form that denotes an object termed karman; it need not be a pada, in the technical sense of the term, ending in a nominal termination.

Therefore, Kaiyata is correct in his suggestion that the prätipadika denotes the karman. Commenting on kumbhakärah under A. 2.2.19, värttika 3, Kaiyata suggests that the prätipadika itself, possessed of five meanings (a generic property, an individual object, its gender, its number, and its participation in the action), denotes the karman: "if the group of five is the meaning of a nominal base, then because the nominal base itself denotes the direct object, the affix an must be provided on the condition that just the nominal base is the upapada" (pancake prätipadikärthe prätipadikenaiva karmana uktatvät tatraivopapade 'nä bhävyam). A. 2.3.2 or A. 2.3.65 would still apply to provide a nominal termination after the nominal base, even though its being the direct object in relation to the action was denoted by the nominal base, since the nominal base is not among the speech forms denoted by which a participant in action would not condition a nominal termination. Hence A. 3.2.1 applies when just the nominal base (prätipadika) is upapada. Moreover, this works even if participation in action is not accepted as being denoted by a nominal base. A. 3.2.1 requires that a speech form that denotes a direct object (karman) be upapada; it does not require that the speech form denote the relation of being a direct object (karmatva). The nominal base denotes the direct object already, even without a second-triplet or sixth-triplet termination conditioned by its being termed karman. Therefore, mutual dependency is avoided in the derivation of kumbha-kära even if a sixth-triplet nominal termination is provided after the nominal base kumbha; A. 2.3.65 will apply after the affix an has been provided by A. 3.2.1 but before A. 2.2.19 where a nominal termination is required. Although not required in the derivation of kumbha-kära, a nominal termination is required in the derivation of like compounds such as carma-kära to allow operations that depend upon its being termed pada, in the technical sense of the term, to apply to the initial compound constituent.

3.8 Grimal et al.

The derivation of kumbha-kära presented in Grimal et al. follows the views expressed by Bhattojidiksita, Väsudevadiksita, Nägesa, and Bhairavamisra. The initial compound constituent, the upapada, in an upapada-tatpurusa compound terminates in a nominal termination; the final compound constituent terminating in a krt-affix does not. The sixth-triplet nominal termination provided by A. 2.3.65 is accepted as the termination on the upapada. The presence of the sixth-triplet nominal termination on the upapada at the time of application of A. 3.2.1 indicates that they accept that the term upapada implies the technical sense of the term pada; that is, to be termed upapada, it must end in a nominal termination just as it must to be termed pada. Unfortunately, Grimal et al. did not notice the mutual dependency that these views entail.

3.9 Mutual dependency

In a few instances 13 Patañjali escapes from the mutual dependence of the provision of an affix upon the presence of a preceding speech form and vice versa by stating that the affix in the locative is a locative of domain (visaya-saptami) rather than a right-context locative (para-saptami). For example, he escapes from the mutual dependence of the provision of an ärdhadhätuka-affix conditioned by a preceding root and a root replacement conditioned by a following affix in this way. At the conclusion of his commentary on A. 2.4.35 ärdhadhätuke he proposes that the term ärdhadhätuke is a visaya-saptami. The replacement thereby occurs in the intended domain of an ärdhadhätuka-affix rather than when followed in sequence by the speech form (osati paurväparye visayasaptami vijnäsyate. ärdhadhätukavisaya iti). Jayäditya in the Käsikä on A. 2.4.35 states that thereby the replacements are made under the intention to use an ärdhadhätuka-affix; once the replacements have been made, the affixes occur as provided afterwards (visayasaptami ceyam, na parasaptami. tenärdhadhätuka-vivaksäyäm ädesesu krtesu pascad yathäpräptam pratyayä bhavanti). For example, A. 2.4.52 aster bhü provides that the root as is replaced by the root bhü in the domain of an ärdhadhätukaaffix. A. 3.1.97 aco yat provides that the affix y at occurs after a vowel-final root. In order to obtain the form bhávyam the affix yat must occur after the root bhü. However, the affix yat cannot occur until the root as is replaced by bhü since it only occurs after vowel-final roots; it doesn't occur after the root as, which ends in a consonant. If ärdhadhätuke were a para-saptami, the replacement of the root as by the root bhü could only occur after the ärdhadhätuka-affix had been provided.

Similarly, the question of the mutual dependence of a secondary-root-forming affix and a following ärdhadhätuka-affix arises under A. 3.1.31 äyädaya ärdhadhätuke vä. There Patañjali writes,

This is not a problem, ärdhadhätuke is not a para-saptami; rather it is a visaya-saptami meaning 'in the domain of an ärdhadhätuka-affix'. In that case, once the secondary-root- forming affixes beginning with äya (provided in A. 3.1.28-30) have been provided in the domain of an ärdhadhätuka-affix, the affix that would occur after the secondary root occurs, (naisa dosah. ärdhadhätuka iti naisä parasaptamï. kä tarhi. visayasaptamï. ärdhadhätukavisaya iti. taira ärdhadhätukavisaya äyädiprakrter äyädisu krtesu yah yatah pratyayah präpnoti sah tato bhavisyati. MBh. 2.41.17-19)

The Käsikä states, "the secondary-root-forming affixes beginning with äya (provided in A. 3.1.28-30) optionally occur in the domain of an ärdhadhätuka-affix, i.e., when there is the intention to articulate an ärdhadhätuka-affix" (ärdhadhätukavisaye ärdhadhätukavivaksäyäm äyädayah pratyayä vä bhavanti).

The third and final situation in which Patañjali solves the question of mutual dependence by resorting to a locative of domain is under A. 4.1.90. A. 4.1.90 yüni luk (aci 89) provides deletion (luk) of the affix previously provided in the sense of a yuvan-descendant. The deletion occurs if a vowel-initial affix in the section headed by A. 4.1.83 is to follow. The vowel-initial affix provided after the nominal base denoting the yuvan-descendant occurs after the form of the stem once the yuvan-affix has been deleted, but the yuvan-affix is deleted on condition that the vowel-initial affix is provided. If the locative in the term aci were a parasaptamï, the rule would provide deletion before a vowel-initial affix that had already been provided after the form of the nominal base terminating in the yuvan-affix. Thus wrong affixes would result, (yüni lug aclti cet pratyayasyayathestaprasañgah. A. 4.1.90, värttika 1. MBh. 2.242.15.) To get the correct form, provision of the vowel-initial affix has to occur once the deletion has been done. To avoid mutual dependence, Patañjali states that the term aci in A. 4.1.90 is a visaya-saptami meaning "in the domain of a vowel-initial affix." In that case, the affix that occurs after the nominal base is the affix that would occur once deletion has been done in the domain of the vowel-initial affix, (naisa dosah. aclti naisä parasaptamï. kä tarhi. visayasaptamï. ajädau visaya iti. taträci visaye luki krte yah yatah pratyayah präpnoti sah tato bhavisyati. MBh. 2.242.21-23.) The Käsikä states, "deletion (luk) occurs in place of the yuvan-affix when the vowel-initial affix provided under the heading A. 4.1.83 is intended to be articulated, still in mind, not yet arisen. Once the yuvan-affix has been deleted, the affix that would occur after the nominal base in that form occurs" (prägdlvyatlye ajädau pratyaye vivaksite buddhisthe 'nutpanna eva yuvapratyayasya lug bhavati. tasmin nivrtte sati y o yatah präpnoti sa tato bhavati).

The visaya-saptami is only resorted to under duress. It is preferable to find another means to achieve derivation. Panini avoids similar situations of the mutual dependence of stem and affix by stating the relevant rules in the asiddhavat section headed by A. 6.4.22 asiddhavad aträbhät. For example, the verbal stem säs is replaced by sä before the second person singular active imperative termination hi by A. 6.4.35 sä hau. At the same time, the second person singular active imperative termination hi is replaced by dhi after the root hu and roots ending in a non-nasal stop or spirant by A. 6.4.101 hujhalbhyo her dhih. Neither rule would apply if subject to the conditions produced by the other having applied first. The derivation works by applying rules in the section headed by A. 6.4.22 asiddhavad aträbhät as if operations provided by other rules in that section had not taken place.

If nominal terminations were required on upapadas prior to the provision of krt-affixes, the result would be mutual dependence between the rules that provide the nominal terminations and the rules that provide the krt-affixes. The fact that the Mahäbhäsya does not raise the issue of mutual dependence between rules that provide krt-affixes and rules that provide nominal terminations on upapadas serves as evidence, though it be evidence of silence, that Patañjali did not consider nominal terminations to be required.

The conditions under which the krt-affix an occurs require that there be a speech form termed upapada denoting an object termed karman. For the object to be termed karman, a nominal termination is not required. Quite the opposite; it is the condition for the occurrence of the nominal termination. The only circumstance that suggests that there is a nominal termination present at the time of A. 3.2.1 coming into play is the interpretation oí pada in the term upapada in the technical sense that it is provided by A. 1.4.14 suptiñantam padam. A. 1.4.14 terms a speech form pada if it ends in a nominal or verbal termination. This interpretation is questionable. It requires accepting that the purpose of using a long term (upapada) is that the term carry its meaning and that the specific meaning it carry be the technical sense requred by A. 1.4.14. That the purpose of using a long term (upapada) is that the term carry its meaning is not objectionable. But that pada therein carries the technical meaning of a speech form ending in a nominal termination is objectionable. The latter is not accepted by Jinendrabhuddhi or Bhoja. Jinendrabuddhi provides a conventional meaning for the term pada instead: that by means of which a meaning is understood. Although Haradatta is quite right to point out that a nominal termination must be permitted to occur on the upapada prior to compounding so that it does get termed pada according to A. 1.4.14 and become subject to operations that require the technical term (such as deletion of pada-final ? by A. 8.2.7 in which the term padasya recurs), there is no need for the term upapada to carry that technical sense of the term pada. After the occurrence of the krt-affix an (Table 1 , step 7), conditions are satisfied to allow a sixth-triplet nominal termination to arise in accordance with A. 2.3.65 (Table 1, step 12b) and for the upapada, which now does end in a nominal termination, to be termed pada by A. 1.4.14 (Table 1, step 14). Not before. Conditions are simply not present to allow a nominal termination to arise on the upapada prior to the provision of an affix after the root. The morphology of the governing word determines that of the governed.

4. SEMANTICS DRIVE PÄNINIAN DERIVATION

Grimal et al. did not include early steps in the derivation. They did not work out the steps by which a nominal termination would arise on the upapada prior to the provision of the krt-affix an after the root. The result is that they reproduced the views of Bhattojidiksita, Nägesa, and their commentators, and steps critical to demonstrate both the sense of the compound and Pänini's methodology are absent. One is allowed to get the impression that the derivation begins with speech forms already in mind, either in the form of a vigraha väkya such as *kumbhasya kärah or in the form of a string such as kumbha-as + kr-a. The latter string could result directly from the provision of the affix an by A. 3.2.1 karmany an only if the term krti in A. 2.3.65 kartrkarmanoh krti were a visaya-saptami. In that case the speaker's anticipation of a certain speech form would serve as the condition for the provision of another speech form. This anticipation of speech forms in Päninian derivation is resorted to only rarely, only three times under duress in the case of mutual dependence of speech forms on each other. It is not necessary in the derivation of upapada-tatpurusa compounds, nor is it the general procedure adopted in Päninian grammar.

That Pacinian derivation begins with speech forms already in mind is the assertion Houben erroneously makes. He asserts that the derivation begins with some sort of sentence that the speaker uses the grammar to check for correctness (see section 1.3). Yet, as explained in section 1.1, the only speech forms available at the start of a derivation are roots and nominal bases. Semantic conditions are required in the grammar to determine their relation, the proper affixes used to denote those relations, and compound formation. The only speech forms available at the start of the derivation of kumbha-kära are the nominal base kumbha and the root kr. The derivation of kumbha-kära does not require any other speech form at all until the affix an is introduced after kr in step 7 of Table 1. As sections 2-3 have argued, the condition for the affix an is a semantic object termed karman, not a sixth-triplet nominal termination as stated by Bhattojidiksita, Nägesa, and their commentators, and reiterated by Grimal et al., nor a second-triplet nominal termination as argued by Joshi and Roodbergen. The condition for the affix an is not a speech form at all; it is a semantic object devoid of any speech form whatsover; it is a disembodied meaning intended by the intellect of a speaker.

Although a user of the grammar may have sentences in mind he wants to check, the procedure of the grammar he uses to check them derives such sentences by relying on semantic conditions. It is not the case that semantics are resorted to just to "label the linguistic forms of his preliminary sentence according to the syntactically relevant categories of meaning," as Houben asserts. It is not the linguistic form that gets labeled; it is a meaning, accompanied or not by any linguistic form. In the derivation of kumbha-kära, it is not the speech form kumbha that is termed karman, it is the pot, regardless of the word used for it or the language. The pot is termed karman solely by that object's relation to an action, without regard to any speech form. Even the presentation of the derivation in Table 1 is susceptible to the misinterpretation that the speech forms such as kumbha are kärakas. They are not. The semantic objects denoted by these speech forms are kärakas. Kärakas are participants in the action. It is the objects that participate in action, not the words that denote those objects. The words that denote objects are introduced in the derivational steps in the tables only because in some derivations, though not in the ones presented, co-occurrence conditions are taken into account even at the stage in which objects are designated by käraka terms.

Panini derives speech forms from the point of view of the speaker. He begins with semantics, with what the speaker wants to express. Objects in the conception of the speaker are the starting point. Specifying semantic objects and co-occurring speech items as conditions, he designates items by käraka terms. Items designated by specific käraka terms condition verbal terminations, krt-affixes, and compounding. Only semantic conditions that remain undenoted after verbal terminations, krt-affixes, taddhita-affixes, or compounds have been provided condition the occurrence of nominal terminations. Therefore, nominal terminations would not arise after kumbha in the derivation of kumbha-kära until steps 12-12b, after the provision of the krt-affix an in step 7.

Patañjali explicitly states in several places that semantics drives the derivation of speech forms and not vice versa, and details the sequence of derivational steps from verbal semantics, to the semantics of participation in the action of the verb, to the provision of käraka terms for those participants, and finally to the arising of nominal terminations. He does so, for example, under 2.3.50 vt. 5 uktam pürvena. The context concerns an explanation of why a sixth-triplet nominal termination arises after the stem räjan and not after the stem purusa in the phrase rajñah purusah. While his remarks concerning the derivation of the particular phrase in question there are not relevant for the derivation of kumbhakära, his general remarks are.

na hi sabdakrtena nämärthena bhavitavyam. arthakrtena näma sabdena bhavitavyam. For it is not the case that meaning occurs caused by speech forms; speech forms occur caused by meanings. (MBh. 1.464.15-16; also at 1.362.17-18)

Patañjali proceeds to provide details of the sequence of derivational steps as follows:

Particular relations of the objects denoted by nominal stems originate caused by the action. And the terms karman, karanam, apadanam, sarhpradanam, adhikaranam arise caused by those particular relations. And those in turn are sometimes adopted as conditions for the arising of triplets of nominal terminations, sometimes not. And when are they adopted as conditions for the arising of triplets of nominal terminations? When they differ from the meaning of a nominal base. For when they don't differ from the meaning of the nominal base, then the explicit terms themselves, karman, karanam, apadanam, sarhpradanam, adhikaranam, occur. (pratipadikarthanam kriyakrta visesa upajayante tatkrtas cakhyah pradurbhavanti karma karanam apadanam sarhpradanam adhikaranam iti. tas ca punar vibhaktinam utpattau kadacin nimittatvenopadiyante kadacin na. kada ca vibhaktinam utpattau nimittatvenopadiyante? yada vyabhicaranti pratipadikartham. yada hi na vyabhicaranty akhyabhuta èva tada bhavanti karma karanam apadanam sarhpradanam adhikaranam iti. MBh. 1.464.18-23)

Kaiyata provides the example "he cuts with a knife" (datrena lunati) to show what happens when the object denoted by the nominal base participates in an action. The relation the knife (datra) has with the action differs from the meaning of the nominal base; the relation is not denoted by the base, but instead conditions a third-triplet nominal termination. He provides the example "the knife is the instrument" (datram karanam) to show what happens when the relation is explicitly stated by the nominal base. The relation does not differ from the meaning of the nominal base and does not condition a third-triplet nominal termination; it occurs in the nominative.

The crucial point is that semantics drive Paninian derivational procedure. Semantics condition the naming of certain intentional objects by karaka terms. These karaka terms then condition speech forms. It is worth reiterating my explanation (Scharf 2009a: 101) that the Astadhyayi is composed in a manner that selects certain speech forms for use on the basis of certain semantic conditions. Subrahmanyam (1983) demonstrates the significant role semantics plays in the Astadhyayi, and I describe the role of some 735 semantic conditions stated therein (2009a: 101-1 1). The procedure of the grammar models the fact that a speaker selects speech forms to use on the basis of the meaning he wishes to convey. As I explained (1995), Katyayana himself says so in his very first varttika (MBh. 1.6.8): "since the usage of speech is prompted by meanings in accordance with ordinary usage, the science (of grammar) restricts (usage to correct speech forms) for the sake of dharma." The restriction set forth by the grammar limits speech forms on the basis of semantic conditions in the same manner speakers select speech forms on the basis of the meanings they wish to convey.

1. See the review by Scharf 2009b.

2. Accent is not shown since it would needlessly complicate the issue at hand to which it is not relevant. For a discussion of accentuation replete with the derivation of examples see Scharf 2008.

3. subantanam samasah. tatrantarangatvat tap. tapy utpanne samasah. ghras'abdah samasyeta. tatra jater astrivisayad ayopadhad akarantad iti ñis naprapnoti. MBh. vol. 2, p. 218, line 26 - p. 219, line 2.

4. Cf. Joshi and Roodbergen's (1973: 214-15) translation.

5. Regarding the account of the subjunctive examined by me in Scharf 2008, Haradattamisra and Nägesa, in contrast, opt for broad coverage of rules of indeterminate variation over a more precise systematic account.

6. See Joshi and Roodbergen 1973: 218 for detail.

7. Joshi and Roodbergen (1973: 223) suggest alternatively that the final constituent sec is derived from the root sic + vie by A. 3.2.75 anyebhyo 'pi drsyate without causative meaning.

8. See Joshi and Roodbergen 1973: 216-17.

9. See Joshi and Roodbergen 1973: 217.

10. Bhairavamisra's Candrakalä glosses asya here as upapadasamäsasya.

11. vyäghri on pp. 218-19, *vyäghrä on pp. 219-20; kacchapl on p. 220, and *kacchapâ on pp. 221-22; vastrakrltl and vastrakrltä on p. 222.

12. It is apparently an oversight that they provide (p. 220) a second-triplet termination rather than the third or seventh indicated as possibilities by Vâsudevadîksita and give the sütra number for the affix an (A. 3.2.2) rather than ka (A. 3.2.4). See section 3.5 above and Table 2, steps 6b and 12.

13. ärdhadhätuke in A. 2.4.35 ärdhadhätuke; ärdhadhätuke in A. 3.1.31 äyädaya ärdhadhätuke vä; and aci in A. 4.1.90 yüni luk(aci 89).

References:

Bronkhorst, J. 1979. The Role of Meanings in Pänini's Grammar. Indian Linguistics 40: 146-57.

Cardona, George. 1967-68. Anvaya and vyatireka in Indian Grammar. In Dr. V. Raghavan Felicitation Volume. Adyar Library Bulletin 31-32: 313-52.

Grimal, F.; V. Venkataraja Sarma; and S. Lakshminarasimham. 2007. Päniniyavyäkaranodäharanakosah; La Grammaire paninéenne par ses exemples; Päninian Grammar through Its Examples, vol. ?: Samäsaprakaranam, Le Livre des mots composés; The Book of Compound Words. Rashtriya Sanskrit Vidyapeetha Series, no. 150; Collection indologie 93.2. Tirupati: Rashtriya Sanskrit Vidyapeetha; Pondichéry: École Française d 'Extreme-orient; Institute Française de Pondichéry.

Houben, Jan E. M. 1999 [2001]. Meaning Statements in Pänini's Grammar: On the Purpose and Context of the Astädhyäyi. Stuthen zur Indologie und Iranistik 22: 23-54.

________ 2003. Three Myths in Modern Päninian Studies. (Review of Recent Research in Päninian Studies, by George Cardona.) Asiatische Stuthen I Études Asiatiques 57: 121-79.

________ 2009a. Bhartrhari as a "Cognitive Linguist." In Bhartrhari: Language, Thought, Reality: Proceedings of the International Seminar on Bhartrhari, New Delhi, December 2003, ed. Mithilesh Chaturvedi. Pp. 523-43. Delhi: Motilal Banarsidass.

________ 2009b. Pacini's Grammar and Its Computerization: A Construction Grammar Approach. In Sanskrit Computational Linguistics: Third International Symposium, Hyderabad, India, January 15-17, 2009, Proceedings, ed. Amba Kulkarni and Gérard Huet. Pp. 6-25. Lecture Notes in Artificial Intelligence, no. 5406. Berlin: Springer- Verlag.

______. Forthcoming. Studies in India's Vedic Grammarians, 1 : Näräyanabhatta's Prakriyäsarvasva and Panini' s Ee (ms. provided by author, a preliminary version of which was presented at the 14th World Sanskrit Conference, 1-5 September 2009, Kyoto University, Kyoto). In Proceedings of the 14th World Sanskrit Conference, 1-5 September 2009, Kyoto University, Kyoto: Vyäkarana, ed. George Cardona and Hideyo Ogawa.

Joshi, S. D., and J. A. F. Roodbergen. 1973. Patañjali's Vyäkarana-Mahäbhäsya tatpurusähnika (P. 2.2.2-2.2.23): Edited with Translation and Explanatory Notes. Pune: University of Poona.

________ 1975. Patañjali's Vyäkarana-Mahäbhäsya kärakähnika (P. 1.4.23-1.2.55): Introduction, Translation, and Notes. Publications of the Centre of Advanced Study in Sanskrit, Class C, no. 10. Pune: University of Poona.

________ 1997. The Astädhyäyi of Panini: with Translation and Explanatory Notes, vol. VI: 2.2.12.2.38. New Delhi: Sahitya Akademi.

______ 1998. The Astädhyäyi of Panini: with Translation and Explanatory Notes, vol. VU: 2.3.12.3.73. New Delhi: Sahitya Akademi.

Kiparsky, P. 1982. Some Theoretical Problems in Panini' s Grammar. Post-graduate and Research Department Series, no. 16. Pune: Bhandarkar Oriental Research Institute.

________ 2009. On the Architecture of Pänini's Grammar. In Sanskrit Computational Linguistics: First and Second International Symposia, Rocquencourt, France, October 2007; Providence, RI, USA, May 2008; Revised Selected and Invited Papers, ed. Gerard Huet, Amba Kulkarni, and Peter Scharf. Lecture Notes in Artificial Intelligence, no. 5402. Berlin: Springer- Verlag, 2009. http:// sanskrit.inria.fr/Symposium/Program.html.

Kiparsky, P., and F. Staal. 1968. Syntactic and Semantic Relations in Panini. Foundations of Language 5: 83-117.

MBh.: The Vyäkarana-mahäbhäsya of Patañjali. Ed. Lorenz Franz Kielhorn. 3 vols. BSPS 18-22, 28-30. Bombay: Government Central Press: Vol. I, 1880; Vol. 2, 1883; Vol. 3, 1885. Second edition: 1892, 1906, 1909. Third edition revised and furnished with additional readings references and select critical notes by K. V Abhyankar. Pune: BORI, 1962, 1965, 1972. Reprint: 1985. (references: volume, page, line)

Scharf, Peter. 1995. Early Indian Grammarians on a Speaker's Intention. JAOS 1 15: 66-76.

________ 2008. Päninian Accounts of the Vedic Subjunctive: let krnvatte. Indo-lranian Journal 51.1: 1-21.

________ 2009a. "Modeling Päninian Grammar." In Sanskrit Computational Linguistics: First and Second International Symposia, Rocquencourt, France, October 2007; Providence, RI, USA, May 2008; Revised Selected and Invited Papers, ed. Gérard Huet, Amba Kulkarni, and Peter Scharf. Pp. 95-126. Lecture Notes in Artificial Intelligence 5402. Berlin: Springer- Verlag, 2009.

_____ . 2009b. Review of Grimal et al. 2007. JAOS 129.4: 715-19.

Subrahmanyam, P. S. 1983. Pänini's Use of Semantics. In Proceedings of the International Seminar on Studies in the Astädhyäyi of Panini (Held in July 1981), ed. S. D Joshi and S. D Laddu. Pp. 127-36. Pune: University of Poona.

EPr.: Srgäraprakäsa [Sähityaprakäsa], vol. 1 [1-14 Prakäaa], critically edited by Rewäprasäda Dwivedi and Sadäaivakumära Dwivedï. New Delhi: Indira Gandhi National Centre for the Arts; Varanasi: Kälidäsasamsthäna, 2007.

Thieme, Paul. 1935. Panini and the Veda: Studies in the Early History of Linguistic Science in India. Allahabad: Globe Press.

Author affiliation:

PETER SCHARF

BROWN UNIVERSITY