1. SIDDHA AND ASIDDHA

Pāṇini's theory of ordering rests on the fundamental ordering relations between rules which Sanskrit grammar calls siddha, literally 'effected', and asiddha 'not effected'. As the traditional explication puts it, "rule A is (a)siddha with respect to rule B” means “rule A is to be regarded as (not) having taken effect when rule B is to take effect”.

There are several equivalent ways to define siddha and asiddha precisely. We will here choose the one which may be most congenial to the linguist’s way of thinking about ordering. Let B(A(φ)) denote the result of applying, to a given input φ, rule A and rule B in that order; and let B,A(φ) denote the result of applying, to a given input φ, rules A and B simultaneously. Then:

(la) A is siddha w.r.t. B = For all φ such that B(A(φ)) ≠ B,A(φ), A is applied before B to φ.

(lb) A is asiddha w.r.t. B = For all φ such that B(A(φ)) ≠ B,A(φ), A is not applied before B to φ.

The definitions are deliberately framed so as to allow the relations to be specified vacuously even for rules which never actually interact (i.e., where there is no φ s.t. B(A(φ)) ≠ B,A(φ)), as is done in Pāṇini’s grammar.

Siddha and asiddha can be used to specify the ordinary sequential ordering relations in the obvious way, as will be clear from the following example.

Ex. 1: Consider the contraction rule which replaces vowel sequences of the form a+i by the single vowel e.³ Vowel sequences of this type arise by the operation of several rules. The contraction rule must apply to the a+i sequences produced by some of these rules, but not to the a+i sequences produced by others. For example, the operation of a vocalization rule (samprasāraṇa) changing y to i in the reduplication of certain verbs⁴ produces such sequences, e.g. atra yāj+yāj+a → atra iāj+yāj+a (→ atra iyāja) ‘he sacrificed here’, to which contraction must apply: atra iyāja → atreyāja. Vocalization, therefore, is siddha w.r.t. contraction. Another rule deletes word-final glides (y and v) after a and certain other vowels,⁵ turning e.g., ayajay indram ‘I sacrificed to Indra’ into ayaja indram. Vowel sequences resulting from glide-deletion must on no account undergo contraction: ayaja indram ↛ *ayajendram. Glide-deletion is therefore asiddha w.r.t. contraction. This corresponds to the ordering:

1. vocalization (. . . ay. . . → . . . a i. . .)

2. contraction (. . . ai . . . → . . . e . . .)

3. glide deletion (. . . ay i. . . → . . . a i. . .)

This ordering, however, is specified not directly but in terms of how the rules interact. Thus, when A feeds B, as vocalization does contraction here, A is siddha w.r.t. B, and when A counterjeeds (is blocked from feeding) B, as glide deletion does contraction here, A is asiddha w.r.t. B.⁶ Similarly, when A bleeds B, A is siddha w.r.t. B, and when A is blocked from bleeding B, A is asiddha w.r.t. B. When A changes the way B applies to a form, A is siddha w.r.t. B (e.g. let A be a rule deleting final syllables and B a rule assigning penultimate stress), and when A is blocked from doing so, A is asiddha w.r.t. B. When A could have no effect on the application of B, it can equally well be considered siddha or asiddha w.r.t. B.

The interest of the (a)siddha relation lies in the fact that it provides comparable definitions of various other types of ordering relations which figure both in Pāṇini’s grammar and in modern linguistics. In the tradition of generative grammar these have been characterized in quite heterogeneous, unrelated ways, so that no unified framework for talking about rule interaction has been able to emerge. By taking (a)siddha as the basic relation, all ordering can be located on a point between the two extremes of simultaneous ordering (no rule interaction) and transparent ordering (maximal rule interaction), definable as follows:

(a) Simultaneous ordering: all rules are asiddha w.r.t. all other rules.

(b) Transparent ordering: all rules are siddha w.r.t. all other rules.

(Note that intrinsic ordering is a term used for transparent ordering in rule systems with the special property that only the feeding relation can hold between rules.)

The other types of ordering can be defined by imposing various restrictions on transparent ordering. Thus, (b) can be restricted in one or both of the following ways:

(c) Linear ordering: any rule is asiddha w.r.t. any rule that precedes it in a listing.

(d) Strict cyclical ordering: any rule A is asiddha w.r.t. any rule B in cases where A applies in an outer cycle (viz. where A and B are applicable minimally within constituents P and Q, respectively, and P dominates Q).

The interaction of a rule with itself can be characterized in the same terms as the interaction of rules with each other, e.g.

(e) Simultaneous application: a rule is asiddha w.r.t. itself.

(f) Iterative application: a rule is siddha w.r.t. itself.

There are several points worth emphasizing here:

(1) In modern treatments (e.g. Kiparsky 1973a) transparent rule ordering comes across as a principle of a rather different sort from linear or simultaneous ordering. It is sometimes even said to be a “functional” principle as opposed to the other, “formal” principles (whatever that distinction may exactly mean, in this case). It becomes clear from the Sanskrit grammarians’ treatment that transparent ordering is entirely on the same footing as the other familiar kinds of ordering.

(2) In modern treatments (e.g. Kean 1974, Mascaro 1976) the condition of strict cyclicity, though empirically preferable to the older nonstrict version of cyclicity, is in no way more natural from a formal point of view. On the contrary, it appears as an added complication, a special restrictive clause put on top of the condition that rules apply cyclically. If (a)siddha is taken as the basic ordering relation, cyclic application will automatically be defined as strict cyclicity (see section 3 below), and it is rather any other (non-strict) version of cyclicity which requires additional restrictions. It is fair to conclude that if modern linguists had thought about the interaction of rules in terms of the (a)siddha relation, the correct interpretation of cyclic rule application would have been found from the beginning.

(3) The modern treatment of ordering provides no a priori reason to expect that rules should relate to themselves either in the simultaneous manner or in the iterative manner, viz. (e) or (f) above. In terms of a Pāṇinian framework, however, (e) would be naturally associated with (a) and (f) would be naturally associated with (b). In fact, the most general formulations of (a) and (b) (obtained by dropping other in them) would subsume (e) and (f) as special cases.

2. TRANSPARENCY AS THE BASIC ORDERING PRINCIPLE

We shall now proceed to outline how the sequencing of rules in derivations is determined in Pāṇini’s grammar. Most of the operative principles are not formulated in the grammar itself. Not all of them are formulated even in the traditional commentaries, and some of the principles which the commentaries do put forward must be eliminated or amended, in our opinion.

A preliminary remark on the character of Pāṇini’s grammar is necessary. Pāṇini’s basic procedure is to abstract everywhere as general statements as possible (sāmānya) and to restrict them appropriately by special statements (viśeṣa), each of which may in turn be restricted by even more special statements, and so forth. Moreover, the most general principles do not have to be specially stated if they are implicit in the special principles which restrict them. Patañjali explains this with an analogy from the religious laws. “Five five-clawed animals may be eaten” implies: “the others may not be eaten.” The latter, general case does not have to be specially stated, but can be inferred from the statement of the special case, which can be thought of as having an implicit “only,” viz. “only five five-clawed animals may be eaten.”

The most general principle governing the order in which rules are to be applied is that of transparent ordering:

(2) All rules are siddha w.r.t. all rules (sarvatra siddham).

Most rules are correctly sequenced by (2). But (2) must be curbed for cases where it gives the wrong ordering (e.g., for cases of counterfeeding or counterbleeding order), and supplemented for cases where the potential interaction of rules is symmetrical (e.g., mutual bleeding). We shall be concerned with the former kinds of cases here. A number of special conditions, of varying generality, serve to restrict the validity of (2) by implementing non-transparent ordering in certain cases. These special conditions accordingly state that (contrary to [2]) such-and-such rules are asiddha with respect to such-and-such other rules of the grammar. Where not overridden by one of these conditions, (2) fixes the sequencing of rules as transparent.

Like the law cited by Patañjali, (2) does not have to be specially stated in Pāṇini’s grammar. It is implicit in the various special restrictions which have been stated in the grammar, which say that certain rules are asiddha w.r.t. certain other rules. By them we know: the others are siddha.

In order to appy the rules of the grammar to a form φ, we must find a derivation which satisfies the ordering principles. In practice we might proceed as follows:

1. Find all which are applicable to φ.

2. Determine the partial ordering imposed on the rules by such special ordering principles as are applicable, and otherwise by the general ordering principle (2).

3. Apply rule R to φ, where R is any rule not preceded by another rule in the partial ordering established by step 2.

4. Taking the result of step 3 as the new φ, repeat the procedure beginning with step 1.

Pāṇini’s system of rules seems intended to enable a unique output (up to free variation) to be derived from any given input. An ordering which yields a wrong form should violate (2) or one of the special principles.

It is evident that a procedure of this sort could, in principle, lead to enormously complicated operations at stage 2. Given n applicable rules, it is necessary to check nx (n — 1) ordered pairs of rules for whether the siddha or asiddha relation holds between their members. But the actual forms which arise in derivations normally involve a choice between at most a few rules. Indeed, perhaps the most frequent situation is that only one rule is even applicable. It is a remarkable fact about Pāṇini’s grammar (and perhaps about language) that so few rules tend to compete for a form. It makes the practical application of the grammar rather straightforward in most cases.

Ex. 2: As a typical illustration, we choose the whole derivation of the expression atreyāja whose last stage was already discussed in Ex.1. The first word, atra ‘here’, is derived from the pronominal stem etad ‘this’. A semantically conditioned rule designates a “locus” of an action as being an adhikarana, a “deep structure” locative (kāraka).⁷ This feeds a rule which inserts suffixes of the seventh (locative) case (in this case i) after stems which are adhikaranas.⁸ No other rules are applicable:

etad + i

This in turn feeds a rule which adds the ending tra after pronominal stems which end in a locative case suffix.⁹ No other rules are applicable. (We disregard here accent rules, whose effect will be wiped out by the last rule in this derivation anyway. The rule adding tra does not reapply a second time, cf. section 6 below.)

etad + i + tra

This feeds deletion of the case suffix i by a rule which deletes case suffixes inside nominal stems.¹⁰

etad + tra

Since etad thereby comes to stand directly before tra, this rule feeds a rule replacing etad, when anaphoric, by unaccented a before tra (and tas), and making the suffixes unaccented at the same time.¹¹ It also potentially could feed a rule which devoices obstruents before voiceless obstruents (8.4.55 khari ca). Which rule must apply? Both devoicing and etad → a potentially bleed each other, so that (2) cannot decide the preference between them. But devoicing (8.4.55) is designated by Pāṇini as asiddha w.r.t. all rules that precede it, including in particular the etad → a rule (2.4.33), so that devoicing cannot bleed etad → a. (The etad → a rule would as it happens win over devoicing anyway by another basic principle of the grammar, which we will not discuss here.) We therefore derive the final form:

a + tra

The second word iyāja ‘he sacrificed’ (perfect tense) is derived from the root yaj. The abstract tense marker liṭ is added to the root when past time anterior to the present day and not witnessed by the speaker is to be denoted:¹²

yaj + liṭ

This liṭ is replaced by the basic active endings, in this case ti:¹³

yaj + ti

In the active perfect, these suffixes are replaced by another set of suffixes:¹⁴

yaj + a

The suffix a which thereby comes to stand after the root triggers the substitution of ā for a in it:¹⁵

yāj + a

The root is reduplicated in the perfect:¹⁶

yāj + yāj + a

The y of the reduplicating syllable is vocalized (fn.4):

iāj + yāj + a

The vowel after the vocalized i is deleted:¹⁷

ij + yāj + a

Non-initial consonants in reduplicating syllables are deleted:¹⁸

i + yāj +a

The syllable preceding the suffix -a is accented:¹⁹

i + yā́j + a

It will be seen that all crucial ordering relations in the derivation of both words involve feeding. The interaction of rules here is in its entirety predicted by (2).

There are other types of ordering, not involving feeding, which also fall under (2), if the siddha relation is defined in the general manner of (1), as we have suggested. Numerically the next most important set of cases involve the bleeding relation. Here principle (2) says that rules are to be applied in bleeding order. Examples 3-5 will illustrate how this is the case.

Ex. 3: śiṣṭāt (imper, of śās ‘instruct’). The underlying śās + hi is subject to a rule which replaces śās by śā before the suffix hi.²⁰ This process is bled by the optional replacement of hi by tāt²¹ The other, non-transparent ordering would give the incorrect output (śās + hi → śā + hi → *sā + tāt).

Ex. 4: rudihi (2 sg. imper. of roditi ‘cries’). In underlying rud + hi, there is a chance of changing the suffix hi to dhi by a rule whose context specifies a root ending in a consonant.²² But this is bled by a rule which adds the augment i (iṬ) to a consonantal suffix, changing it to ihi, which is no longer subject to the former rule.²³ Thus we derive rud +hi → rudihi and not the incorrect rud + hi → rud + dhi → *rud + idhi.

Since affixation rules (as opposed to augment rules) are rarely phonologically conditioned, but themselves feed and bleed the operation of morphophonemic rules, it follows that affixation rules will normally apply before morphophonemic rules. This does not have to be stated as a separate principle, but follows from the condition that rules apply in transparent order:

Ex. 5: tudati ‘hits’. In tud + ti, the root vowel might be replaced by guna (o) because it is the penultimate segment of a stem followed by a sārvadhātuka suffix.²⁴ But this rule is bled, and therefore preceded, by the morphological rule inserting the vikarana (Śa) between the root and the suffix. We thus get tud + ti → tud + a + ti, a form to which guṇa is no longer applicable. The other, non-transparent ordering would have given us tud + ti → tod + ti → tod + a + ti.

Cases like Examples 3-5 are common. But the later tradition, interestingly enough, does not account for them by (2). It instead supposes a separate principle to the effect that a nitya ‘constant’ rule takes precedence over a non-nitya rule. Given two rules A and B which are applicable to a given form, rule A is nitya (and rule B non-nitya) when A is still applicable if B applies first, but B is no longer applicable if A applies first. Saying that nitya rules precede is equivalent to saying that rules are applied in bleeding order.

In Kātyāyana’s vārttikas, the nitya relation is very rarely utilized (e.g., on 6.4.88). It may be a later development necessitated by the illegitimate expansion of the antaranga/bahiranga relation to the word-internal domain (see below).

Aside from ordinary bleeding and feeding, (2) covers rule interactions of the following sort.

Ex. 6: We take the derivation of susyūṣati ‘wants to sew’ from the stage siv + sa + ti. The desiderative suffix sa triggers two rules here: reduplication of the root (siv → sivsiv)²⁵ and replacement of the root-final v by ū.²⁶ Which wins? According to (2), v → ū is applied first, since that makes it siddha with respect to reduplication. The siddha order, then, is siv → siū → sisyū, whereas the asiddha order would be siv →sivsiv →sivsiū, equivalent to simultaneous application. After applying the first rule, reduplication is still applicable, but v → ū has fed a new rule, which replaces i by the glide y before a vowel.²⁷ The siddha order is again obtained if this glide formation rule is applied before reduplication (siu → syū → syūsyū, not siū → sisiū → sisyū, which would be the asiddha order, equivalent to simultaneous application). We thus derive ultimately susyūṣati (by deleting the second consonant in the reduplication and shortening its vowel) rather than *sisyūṣati. Whatever processes are applicable to the root must be applied before the root itself is copied: a root-changing rule precedes reduplication in the siddha order.

The relationship between the root-changing rules and reduplication in this example is not feeding or bleeding in the ordinary sense, since reduplication will apply regardless of whether the root-changing rules are applied or not. The root-changing rules affect the way reduplication will apply. This kind of relationship can arise when the second rule contains any sort of variable expression. From a formal point of view it may be convenient to extend the notions of feeding and bleeding so that they will apply here too. This can be done by regarding variables as abbreviations for all specific cases they cover. Thus reduplication would be an abbreviation for a large set of specific copying rules, one for each specific root shape. Then glide-formation, for example, can be regarded as bleeding one of these specific subcases (namely the rule siū → sisiū) and as feeding another (namely the rule syū → syūsyū).

The common principle which underlies all cases we have dealt with up to now can be simply put as follows: a rule which potentially affects the environment of another rule takes priority over it. In other words, environment-changing rules apply first.

3. ANTARANGA AND BAHIRANGA

The first restriction on (2) which we shall discuss involves the relation antaranga/bahiranga ‘(applying to) an internal/external constituent.’ Some questions about the precise characterization of this relation are controversial (and we shall add to the controversy below), but the general idea is clear. Suppose we have two constituents as indicated by the brackets, and three rules A₁, A₂, B with domains of application as marked off underneath:

Here both A₁ and A₂ are antaranga in relation to B, the corresponding bahiranga rule. In particular, any rule applicable within a word is antaranga in relation to a rule which applies across word boundaries. The restriction is then stated as follows:

(3) Bahiranga rules are asiddha w.r.t. antaranga rules.

Principle (3) is not stated in Pāṇini’s grammar, but it is formulated by the earliest commentator (Kātyāyana) and shown by Pāṇini’s formulations of rules to have been assumed in the construction of his grammar. As the tradition correctly notes, (3) means in particular that:

(3a) Bahiranga rules cannot bleed antaranga rules.

(3b) Bahiranga rules cannot feed antaranga rules.

Here (3a) corresponds to the cycle, and (3b) corresponds to the strict cycle.

Ex. 7: In a + yaj + a + i indram (cf. Ex. 1), two rules are applicable: A rule which contracts cognate vowels, e.g. i + i → ī,²⁸ and the rule which we have seen must actually apply there, the contraction a + i → e (fn. 3). The precedence of the latter contraction follows from the fact that it is antaranga:

(3) says that B is asiddha w.r.t. A, i.e. that A must not be applied to the output of B. Hence A must be applied before B, after which B is no longer applicable. The effect is that of cyclic application.

The relation between the rules is reversed in the following example.

Ex. 8: atra i + ij + us ‘they sacrificed here.’ Here we have i + i + ī applying first because it is antaranga:²⁹

Thus we have the derivation:

The following examples show how (3) induces an analog to the strict cycle, in cases where the antaranga and bahiranga rules are not applicable at the same time, but only the bahiranga rule is initially applicable, in such a way as to feed the antaranga rule. The effect of (3) is to block this feeding.

Ex. 9: pacāvedam ‘let us two cook this.’ The representation pacāva idam undergoes contraction of the sequence a + i to a single vowel e.³⁰ The single substitute e counts as the final segment of the imperative ending.³¹ Therefore it becomes subject to the substitution of ai for e in first person imperative endings.³² This unwanted consequence is prevented by (3). According to (3), the bahiranga operation a + i → e (involving two words) is asiddha w.r.t. the antaranga operation of replacing e by ai in imperative endings.

Ex. 10: dadhy atra ‘yoghurt here.’ In dadhi atra, i is replaced by y in the context of the following vowel.³³ As a result of this process, the rule deleting the last consonant of a word-final cluster becomes applicable.³⁴ This is prevented by (3), which directs that the operation i → y, bahiranga by virtue of being conditioned by the initial vowel of the next word, is asiddha with respect to the cluster simplification rule, which takes place within the first word and is thereby antaranga.³⁵

4. THE NON-EXISTENCE OF WORD-INTERNAL BAHIRANGA RULES

It is an open question in generative phonology whether rules must cycle on word-internal constituent structure. It is therefore especially interesting to inquire how Pāṇini’s grammar deals with this question. According to tradition, the antaranga/bahiranga relation also holds word-internally, and (3) is valid in such cases too. In our opinion this is certainly false. While operations across word boundaries are reliably asiddha w.r.t. word-internal operations in practically all the numerous cases which arise, the situation is altogether different when both rules are word-internal. As often as not, it is the “bahiranga” rule which then must apply first, contrary to what (3) would predict. The following two examples are representative of the two main types of cases which arise. The first example shows how a word-internal, supposedly “bahiranga” rule can bleed an “antaranga” rule.

Ex. 11: In prati + ac + as ‘turned toward (gen. sg.),’ the rule deleting the vowel of the derivational suffix ac before certain case suffixes, such as gen. sg. as (6.4.138 acah) bleeds the replacement of i by y before vowels (6.1.77 iko yan aci). Thus:prati + ac + as → (6.4.138) prati + c + as, and then → (6.3.138) pratīcas. Otherwise we would derive *pratycas. But 6.4.138 is supposedly bahiranga w.r.t. 6.1.77.

In the next example, a word-internal “bahiranga” rule must be allowed to feed an “antaranga” rule (contrast Ex. 9-10, illustrating that a bahiranga rule operating across word boundaries must not be allowed to feed an antaranga rule).

Ex. 12: In akṣa + div + sU³⁶ ‘gambler’, v is replaced by ū³⁷ in the context of the deleted suffix v (KvIP), and this ū must then be allowed to trigger the replacement of i by y before vowels (6.1.77 iko yan aci): akṣa + div + (KvIP +) sU → (6.4.19) akṣa + diū + s → (6.1.77) akṣa + dyū + s.

By tradition, the process v → ū is bahiranga relative to the process i → y, and nevertheless supplies the environment for the latter, in violation of (3):

(3) would prevent glide formation from applying when, as in this case, it is fed by a bahiranga rule, with the result that the wrong form *akṣadiūs would be derived.

A systematic review of the counterexamples to the antaranga-principle within words, and of the examples which supposedly require it, suggests the following conclusion:

(4) The antaranga/bahiranga relation does not hold between word-internal processes.

The ordering among word-internal processes is, rather, strictly determined by the general ordering principle (2), i.e. by transparency. It will be seen that this gives the correct ordering in Examples 11 and 12. In prati + ac + as (Ex. 11), the deletion of a in the suffix ac bleeds glide-formation and therefore applies before it. In akṣa + div + sU (Ex. 12) v → ū feeds glide-formation and therefore applies before it. This is consistency the Pāṇinian order of application within the domain of word phonology, regardless of which rule might be considered “antaranga” or “bahiranga” according to tradition.

We will review here some additional examples which differentiate between (3) and (2) in favor of the latter, in the domain of word-internal processes. First, some cases showing that word-internal supposed “bahiranga” rules must be allowed to bleed the corresponding “antaranga” rules.

Ex. 13: seduṣas (acc. pi. of pp. sedivas ‘having sat’). In sed + vas + as, two rules are applicable: the insertion of the augment i (iṬ) before consonantal ārdhadhātuka suffixes (vas),³⁸ and the samprasārana replacement of v by u conditioned by the ending as.³⁹ The insertion of i is antaranga, on the traditional theory, but if it is applied first, we get after saṃprasāraṇa and glide formation the wrong form *sedyusas. Principle (2), that rules are applied in transparent order (in this case, bleeding order), correctly predicts that saṃprasāraṇa is applied first. Then the suffix, now uas (˃ us), is no longer consonantal, so that i cannot afterward be inserted. The result is the correct form seduṣas.

Ex. 14: praśna ‘question.’ To the base form prach + na, two rules are applicable: insertion of the augment t (tUK) on a short vowel before ch,⁴⁰ and replacement of ch by ś before a nasal suffix.⁴¹ The insertion of t, being root-internal, is antaranga, on the traditional theory. But applying it first gives the wrong form (*pratśna). If the replacement of ch by ś is applied first, t can then no longer be inserted, and the correct form praśna is derived. The desired order (bleeding, i.e. the environment-changing rule applied first) is predicted by (2).

Ex. 15: prasthāya ‘having departed.’ The base form pra + sthā + Ktvā can undergo either replacement of root-final ā by i before the following K-it suffix in t,⁴² or replacement of the suffix Ktvā by LyaP in compounds, as here after the prefix pra. If the supposedly “antaranga” process ā → i is applied first, the wrong form is derived (pra + sthā + tvā → pra + sthi + tvā pra + sthi + ya → * prasthitya),⁴³ Transparency predicts that Ktvā LyaP should apply first, since it bleeds ā → i. This yields the desired form prasthāya.

The tradition introduces a special ad hoc condition to take care of the above cases with Ktvā → LyaP, to the effect that a replacement by LyaP takes precedence even over antaranga rules.⁴⁴ This condition is not necessary now, as its effect follows directly from (2).

We now give some additional examples of the feeding type, i.e., cases like Ex. 12, where a word-internal supposedly “bahiranga” rule must feed a supposedly “antaranga” rule.

Ex. 16: papusas (act. pp. of pā ‘protect’, acc. pl.).

pa + pā + vas + as⁴⁵

pa + pā + uas + as (v is replaced by the saṃprāsaraṇa vowel u)⁴⁶

pa + pā + us + as (u and a replaced by u)⁴⁷

pa + p + us + as (ā is dropped before the following K-it ārdhadhātuka u).⁴⁸

The “bahiranga” process of saṃprasāraṇa must feed the “antaranga” process of a-deletion. We cannot justify this derivation if we assume that (3) is applicable. But principle (2), that rules are to be applied in transparent order, predicts the correct form.

Ex. 17: yūnas (acc. pi. of yuvan ‘young’).

yuvan + as (4.1.2, see fn. 45)

yuuan + as (replacement of v by saṃprasāraṇa u conditioned by the following bha-forming suffix as)⁴⁹

yuun + as (see fn. 47)

yūn + as (replacement of uu by a long vowel)⁵⁰

Here the process of saṃprasāraṇa, which is supposedly bahiranga relative to contraction because it is conditioned by the case suffix, must nevertheless feed it. Principle (3) cannot justify this derivation. The correct result is obtained by the principle of transparency (2).

Ex. 18: maghonas (acc. pl. of maghavan ‘bountiful’).

maghavan + as

maghauan + as

maghaun + as

maghon + as (the vowel sequence au is replaced by the single guṇa vowel o)⁵¹

Except for the last rule, this example is parallel to the previous one. The derivation again requires the transparent (feeding) order (2), against the antaranga-principle (3).

Ex. 19: dvau ‘two’ (nom. du.).

dvi + au

dva + au (the final segment is replaced by a)⁵²

dva + TāP + au (the feminine suffix ṬāP is added because the nominal stem ends in short a)⁵³

The replacement by a, which is conditioned by the case termination, would traditionally be considered bahiranga relative to the addition of the suffix ṬāP, which depends on the stem only. But the replacement by a feeds the addition of ṬāP and therefore applies before it, in violation of the antaranga principle. The correct result is predicted by (2).

Similar examples are provided by other pronominal stems, e.g., etad ‘this’: nom. sg. fem. etad + sU → (7.2.102) etaa + sU → (4.1.4) etaa + ā + sU (→ esā).

Ex. 20: rājñas (acc. pi. of rājan ‘king’).

rājan + as

rājn + as (the vowel in the ending an is replaced by zero before the suffix as)⁵⁴

rājñ + as (assimilation of palatality)⁵⁵

The deletion of the vowel a, being conditioned by the case suffix, is according to the traditional view bahiranga relative to the replacement of n by ñ, and yet must feed it, contrary to the antaranga-principle (3) but in accord with transparency.

Ex. 21: pratidīvnas (acc. pl. of pratidivan).

pratidivan + as

pratidivn + as (see fn. 54).

pratidīvn + as (i of the root div is lengthened due to the following consonant)⁵⁶

The deletion of a (the “bahiranga” rule) must feed the lengthening of i (the “antaranga” rule) against the antaranga-principle, but in accord with transparency.

Ex. 22: jakṣatus (3. du. perf. of ad ‘eat’).

ad + atus⁵⁷

ghas + atus (optional suppletion of ad by ghas in the perfect)⁵⁸

ghas + ghas + atus (reduplication in the perfect)⁵⁹

ja + ghas + atus (phonological changes in the reduplicating syllable)⁶⁰

ja + ghs + atus (a is dropped before the following vocalic K-it suffix)⁶¹

ja + ghs + atus (s becomes ṣ after a velar)⁶²

ja + kṣ + atus (gh → k before a voiceless sound)⁶³

If we extend the antaranga/bahiranga relation to word-internal processes, then the deletion of a, which requires the person ending atus, has to be bahiranga relative to the root-internally conditioned process gh → k but it nevertheless feeds it, against (3). Once again, the correct order is in accord with (2).

Why then does the tradition maintain the validity of the antaranga-principle in word phonology? The case appears to rest mainly on a single family of examples, including such forms as dhiyati, asusruvat, adiīdipat. They are all of the same type, so that it will be enough for us to look at one.

Ex. 23: dhiyati (3 sg.pres. of dhi ‘think’). At the stage dhi + tiP (2) requires us to insert the vikaraṇa Śa (dhi + a + ti)⁶⁴ rather than changing the root vowel to guna⁶⁵ (cf. the discussion of tudati, Ex. 5). At that stage there are again two rules to consider: guṇa might still apply (by a different rule)⁶⁶ and the i of the root, being prevocalic, could be replaced by iy.⁶⁷ It is i → iy which actually applies, viz. dhi + a + ti → dhiy + a + ti. The tradition attributes this to the fact that i → iy is antaranga relative to guna:

But in point of fact, i → iy is the only rule which can apply to dhi + a + ti. Guṇa is blocked there by the prohibition formulated explicitly by Pāṇini that there is no guṇa before K-it and Ṅ-it suffixes.⁶⁸ And Śa is a Ṅ-it suffix in Pāṇini’s system by virtue of the rule that any sārvadhātuka suffix is Ṅ-it unless it is P-it,⁶⁹ Thus, the situation is as follows: in dhi + ti, Śa bleeds guṇa and is therefore inserted (transparency). Then in dhi + a + ti, guṇa can no longer apply and i → iy gives dhiyati. There is no question here of invoking the antaranga-principle at all.⁷⁰

5. SPECIAL RESTRICTIONS NECESSITATED BY (2)

The case for our interpretation can be made still stronger. In certain cases Pāṇini complicates his description in ways which would have been pointless if (2) were not the basic principle which determines the sequencing of rules in derivations, but which are necessary when that principle is assumed, because they serve to block its unwanted consequences. From such cases, then, we can be certain that we are not merely reading (2) into the system in order to make the rules work, but that Pāṇini, as shown by his own formulations of rules, wittingly based his grammar on (2).

The restrictions required by (2) are in some cases completely ad hoc, and in other cases rather general in nature. We will proceed from the former to the latter. The general point which we wish to illustrate is that all special conditions or rules for determining rule interaction which are found in the Aṣṭādhyāyī are limitations of transparent ordering, i.e., cases in which the implicit principle (2) would be insufficient.

Ex.24: Recall the rule 7.1.37 replacing the gerund suffix Ktvā by LyaP in compounds. In example 15 above we saw how Pāṇini tacitly lets this rule bleed processes which are conditioned by Ktvā. Now let us see how he deals with a contrary example where the Ktvā → LyaP replacement fails to bleed a rule. The root ad ‘eat’ is replaced suppletively by jagdh before suffixes beginning with t and carrying diacritic K, e.g. ad + Ktvā → jagdh + Ktvā (→ jagdhvā). We also happen to get jagdh before Ktvā in the cases where Ktvā is replaced by LyaP, viz. pra + jagdh + Ktvā → prajagdhya. This form could be derived by applying Ktvā → LyaP (7.1.37) after ad → jagdh. But that would be the non-bleeding, opaque order of application. For this reason, Pāṇini has to complicate the environment of the ad → jagdh rule by specially mentioning in it the context LyaP in addition to the general case: 2.4.36 ado jagdhir lyap ti kiti. This complements our earlier argument based on Pānini’s derivation of prasthāya (Ex. 15). Together, these two examples provide a clear contrast illustrating the role of transparency in the Astādhyāyī.

Ex. 25: In the derivation of adhītya ‘having learned,’ the representation adhi + i + LyaP⁷¹ is subject to both vowel contraction (i → ī)⁷² and the addition of the augment t (tUK) to the P-it krt suffix LyaP, because it is preceded by a short vowel.⁷³ Contraction will of course bleed t-augmentation, for it creates a long vowel, after which t-augmentation is no longer applicable. Transparency predicts, then, that contraction should apply first. But this yields the wrong form *adhīya. To get the correct adhītya, Pāṇini introduces a special ordering statement⁷⁴ to the effect that contraction is asiddha ‘not effected’ for purposes of tuk- augmentation.

Ex. 26: In nom. sg. asau ‘that’, underlying adas + sU fits the structural analysis of a rule deleting suffixal consonants after a consonantal stem. But this is bled by the change of the stem-final -s to au.⁷⁵ Transparency therefore would predict adau + sU, whence *asauh by other rules, whereas the correct form is asau. Since transparency forbids Pāṇini from making use of the underlying consonant stem to trigger the deletion of the suffix -sU, that deletion (sulopa) had to be specially mentioned for this pronoun in connection with the s → au change: 7.2.107 adasa au sulopaś ca.

Where transparency does not conflict with the antaranga- principle, the tradition itself cites such examples in the case of the bleeding (nitya) relation. Pāṇini cites the perfect participle upeyivān as a ready-made (nipātana) word.⁷⁶ Why was it necessary to cite the whole form? At the stage

upa + i + vas

two rules are applicable: reduplication⁷⁷ and the addition of the augment i (iT)⁷⁸ to the suffix vas. The augment rule requires a monosyllabic (ekāc) root. Hence it is applicable before reduplication but not after reduplication, while reduplication is applicable whether or not i has been added. Transparency (and its special case, the nitya relation) would require reduplication to be applied first, so as to bleed i-insertion. But the correct form has the augment i. Since the general ordering convention in this case gives the wrong result, Pāṇini is compelled to make a special provision to secure the augment i. He does this by citing the entire form upeyivān in 3.2.109. This citation would be pointless if Pāṇini did not suppose that the augment would otherwise fail to be inserted because of transparent ordering.

In two cases Pāṇini has designated the rules in a certain section of the grammar as being asiddha with respect to other rules in a more general way. The first case is the section headed by 6.4.22 asiddhavad atrā bhāt, which states that any rule in this section is to be considered asiddha w.r.t. any other rule in this section. That is, rules of this section do not interact at all, but apply in simultaneous fashion. The placement of rules under 6.4.22 is always motivated by the need to avoid undesirable feeding and bleeding order which would be imposed by (2).

Ex. 27: In the derivation of 3.pl. impf, āyan ‘went’, the representation i+an is subject to glide formation,⁷⁹ and the verb is also liable to receive a tense augment. This augment is ā before vowels and a before consonants.⁸⁰ Transparency would dictate that we first apply glide formation (the environment-changing rule) and then add the short augment, as the verb now begins with a consonant. But the actual form is āyan, not *ayan. Noticing the problem, Pānini has put both 6.4.81 and 6.4.71-2 into the special subsection headed by 6.4.22 in which every rule is asiddha with respect to every other rule (i.e. where all rules are to be applied simultaneously). Thus, as far as the augment rule is concerned y + an is still treated as beginning with a vowel (i+an) so that the desired āyan is obtained. There is no other reason for including 6.4.81 in the special subsection of simultaneous rules.

A parallel argument is furnished by āsan ‘were’, from as+an. The root as loses its vowel here⁸¹ but must be treated as vocalic to get the long augment. As this again goes against transparency, 6.4.111 also had to be put into the section of unordered rules.

Ex. 28: In the derivation of 3.pl.mid.perf. dadhre ‘gave’, da+dhā+ire is subject to a Vedic rule replacing the suffix ire by re,⁸² which would bleed, and therefore by transparency have to precede, the deletion of the root-final ā before a vocalic ārdhadhātuka suffix.⁸³ This would yield the form *dadhā+re (→*dadhire by 6.4.66 ghumāsthāgāpājahātisām hali). In order to derive dadhre, Pāṇini has put these rules into the simultaneously ordered section (6.4.22 ff.).⁸⁴

The second general section of asiddha rules, which constitutes the most important restriction on transparent ordering in the whole grammar, works in a somewhat different way. According to 8.2.1 pūrvatrāsiddham, any rule from 8.2.1 on to the end of the grammar is asiddha w.r.t. any preceding rule, regardless of whether this preceding rule is before or after 8.2.1 in the grammar. In other words, beginning with 8.2.1 Pāṇini’s grammar shifts into the familiar mode of linear ordering. The section 8.2.1 ff. is known as the Tripādī, “the Three Chapters.”⁸⁵

Ex. 30: Recall that glide deletion⁸⁶ (ayajay indram → ayaja indram, see Ex. 1) must be prevented from feeding contraction⁸⁷ (ayaja indram → *ayajendram) . This is done in Pāṇini’s grammar by putting glide deletion into the Tripādī section. By virtue of 8.2.1, then, it is asiddha w.r.t. contraction, which precedes it in the listing of the grammar.

Ex. 31: An n which is stem-final and also word-final is deleted.⁸⁸ This deletion rule is put into the Tripādī section in order to stop it from feeding a rule replacing the instr. pi. ending bhis by ais after stems ending in short a:⁸⁹ vṛkṣa+bhis vṛkṣa+ais, but rājan+bhis → rāja+bhis → *raja+ais⁹⁰ It is also put into the Tripādī section in order to stop it from bleeding a rule lengthening a vowel before stem-final n in the strong cases:⁹¹rājan → rājān → rājā. Thus the (a)siddha concept enables Pāṇini to block both feeding and bleeding by means of a single restriction.

When a rule A must be fed or bled by a rule B which is for some other reason in the Tripādī, A must also be put in the Tripādī, and it must follow B there. In such cases, A is assigned to the Tripādī not in order to be asiddha w.r.t. some rule preceding A, but in order that another Tripādī rule (viz. B) should be siddha w.r.t. A.⁹²

We have discussed some of the numerous devices by which the basic ordering principle (2) is kept in check where it overapplies. As might be expected, principle (3), being of more limited import to begin with, requires less such policing. Nevertheless, there are cases where Pāṇini was forced to complicate his rules in order to overcome an undesirable consequence of (3).

Ex. 32: The standard example is śivehi ‘come, Śiva,’ from śiva ā+ihi. The connection between the preverb ā and the root being closer than the connection between the vocative śiva and the preverb ā, by (3) we should first apply the antaranga contraction ā+i → e (fn. 30) in the second word and then contract the resulting e with the a of śiva,⁹³ yielding the wrong form *śivaihi. It is simply a fact about the preverb ā that we get the exceptional treatment śivehi, and similarly in any such three-vowel sequence. The right form could of course be derived by contracting the first two vowels first:śiva ā+ ihi → (fn. 28) śivā+ihi → śivehi. But this is impossible in Pāṇini’s system because it goes against (3). In order to account for this case, Pāṇini therefore specially states that a + V contracts to V when V contains the preverb ā,⁹⁴ not by the usual rule (fn. 93) to ai, au. The particular way in which Pāṇini handles the exception shows quite clearly that he thought of underlying śiva ā+ihi as first undergoing contraction of the second pair of vowels. The exceptional nature of these cases is then seen as consisting of a special treatment of a followed by an e or o which morphologically contains the preverb ā.

Internal analysis of the system shows, then, that Pāṇini wrote his grammar on the basis of the unstated principles (2) and (3). They are not only necessary in order to derive the right output (sections 3 and 4), but also presupposed by Pāṇini’s formulations of rules (this section). We find that a number of special complications of rules are motivated by the existence of (2) and (3), and that the assignment of rules to the asiddha sections headed by 6.4.22 and 8.2.1 is apparently wholly explicable as a consequence of these principles.

6. THE APPLICATION OF RULES TO THEIR OWN OUTPUT

From (2) it follows, as we already remarked, that a rule is not asiddha w.r.t. itself. This means that a rule will feed itself (and, in principle, also bleed itself, although there appear to be no examples of this case).

Ex. 33: 3.sg. prec. bhū + yās + st⁹⁵ is subject to a rule deleting s as the first member of word-final or pre-obstruent clusters.⁹⁶ The output bhū + yā + st again undergoes the same rule, becoming bhū yāt.

Ex. 34: In madhulid sīdati,⁹⁷ ‘the bee is sitting’, an augment dh is placed before the s: madhulid dhsīdati.⁹⁸ A devoicing and deaspiration rule, conditioned by a following voiceless sound,⁹⁹ then applies twice in a row: ḍ dhs → ḍ ts→ ṭ ts (madhulit tsīdati) .¹⁰⁰

The reapplication of a rule to its own output is, however, limited by another unstated constraint, which appears to be best formulated as follows:¹⁰¹

(4) A rule cannot be conditioned twice in a derivation by the same context.

The way in which (4) restricts self-feeding is illustrated by the following case:

Ex. 35: In atra ‘here’ (Ex. 2), the t may optionally be geminated by a rule which applies in the context between a vowel and a non-vowel.¹⁰² But the output attra may not again be subjected to this gemination (*attra, *attttra etc.), since the left-hand environment “after a vowel” would again be satisfied by the same vowel a.

Constraint (4) is not introduced merely for this purpose. It applies equally when other rules intervene in the derivation.

Ex. 36: In the perfect of āp ‘obtain’, the vowel of the reduplicated syllable is shortened,¹⁰³ and after losing its consonant¹⁰⁴ contracts with the root vowel:¹⁰⁵āp + āp + a → ap + āp + a → a + āp + a → āpa. The contracted ā counts as the reduplication (as well as the stem),¹⁰⁶ but it must nevertheless not undergo shortening once more by the reapplication of 7.4.59 (*apa) . This is prevented by (4).

The same principle (4) holds even when the first application is vacuous. In general, vacuous application must count as application in Pāṇini’s grammar, just as it must in generative phonology. The tradition expresses this by saying that the rules of grammar are like the rain, in that they fall equally on the empty and on the full.¹⁰⁷ Thus, precisely the same situation as in Ex. 36 holds in aṭ + aṭ + a → āṭa. Here too, shortening applies to the reduplication, albeit vacuously, so that it cannot then reapply after contraction.

That Pāṇini must have wittingly operated with (4) or something close to it can also be concluded from the fact that he formulates explicit restrictions designed to stop the application of rules to their own output only when (4) proves insufficient to do the job.

Ex. 37: Consider the derivation of vivyādha ‘pierced’ (perfect).

vyadh + vyādh + a

viadh + vyādh +a (saṃprasāraṇa by 6.1.17, fn. 4)

vidh + vyādh + a (fn. 7)

vi + vyādh + a (fn. 18)

What is crucial here is that the saṃprasāraṇa rule 6.1.17, which vocalizes a glide in the reduplication, must be prevented from reapplying once more to vivyādha, where it would give *uivyādha *uvyādha. Since this reapplication is not blocked by (4), Pāṇini needed an ad hoc restriction to block it. This is 6.1.37 na samprasārane saṃprasāraṇam, which simply states that a glide is not to be vocalized before a vocalized glide. Here again the existence of an implicit general principle is revealed through the explicit special measures adopted to deal with its undesirable effects.

7. CONCLUDING REMARKS

There is much more to determining the right interaction of rules in Pāṇini’s grammar than what we have brought up here. Most importantly, we have not touched at all on the interesting problem of how the vipratiṣedha (‘mutual contradiction,’ roughly ‘mutual bleeding’) situation is resolved. The disjunctive ordering of special (apavāda) and general (utsarga) rules,¹⁰⁸ and the precedence of rules which otherwise would have no chance to apply at all (anavakāśatva) require careful analysis, as does the establishing of the equivalent to opaque order by the devices of assigning the outputs of rules the contextual value of their inputs in certain specific types of cases (sthānivadbhāva and pratyayalakṣanatva). There are in addition a number of rules in which ad hoc cross-referencing is used. It could be shown that these devices, too, all serve to countermand the general transparent order imposed by (2), in the specific cases where it gives the wrong result. Some of them we hope to analyze elsewhere.

The purpose of this sketch has been to show how the main types of ordering relations between rules which have been envisaged in generative grammar also figure in Pāṇini’s grammar, but were thought of there in terms of a single, unified framework. The fundamental relation siddha and its contrary asiddha together serve to define the various types of ordering in a way which makes their formal kinship evident. As in many traditional grammars (Kenstowicz 1976), transparent ordering has a privileged status in Pāṇini, being the order in which rules apply unless some provision to the contrary is made in the grammar. Indeed, with the understandable exception of Bloomfield (1939), which is modeled on the Tripādī, the theory of Chomsky and Halle (1968) is the only phonological theory countenancing rule ordering at all which does not recognize the distinction between transparent and opaque ordering. It appears that this distinction, far from being an abstruse afterthought, actually lies at the very heart of the way we mentally organize the interaction of rules.

NOTES

This paper is supported in part by the National Institute of Mental Health MH 13390-11. Our thanks to Nicholas Oster for catching some slips.

1. For a vivid vindication of Pāṇini against some attempted re-analyses, see Sag (1974, 1976). Cf. also Bedell (1974), Allen (1962).

2. This is not to deny that fragments of linguistic description concentrating on selected phenomena in English and some other languages have achieved a greater depth than Pāṇini’s grammar. The comparison we are making concerns fully worked out grammars or phonological components of grammars.

3. Itself a special case of a more general rule which replaces a followed by i, u, ṛ, ḷ (long or short) by e, o, ar, al, respectively (6.1.87 ād guṇaḥ, to be taken with 1.1.51 ur an raparah, and delimited by other rules).

4. 6.1.17 liṫy abhyāsasyobhayeṣām.

5. 8.3.19 lopaḥ śākalyasya.

6. The terms feeding and bleeding are explained, e.g., in Kiparsky (1968b), Anderson (1974:145), Hyman (1975:129). Schane (1973) has them wrong.

7. 1.4.45 ādhāro ‘dhikaraṇam.

8. 2.3.36 saptamy adhikaraṇe ca.

9. 5.3.10 saptamyās tral.

10. 2.4.71 supo dhātuprātipadikayoḥ.

11. 2.4.33 etadas tratasos tratasau cānudāttau.

12. 3.2.115 paro’kṣe liṭ.

13. 3.4.78 tiptasjhi . . . The number is determined by the “reference” of the verb (1.4.21-22) and the person by the (present or deleted) “coreferent” (i.e. subject) of the verb (1.4.105-108).

14. 3.4.82 paras maipadānām ṇalatususthalatusaṇalvamāḥ.

15. 7.2.116 ata upadhāyāḥ.

16. 6.1.8 liṭi dhātor anabhyāsasya.

17. 6.1.108 saṃprasāraṇāc ca.

18. 7.4.60 halādiḥ śeṣaḥ.

19. 6.1.193 liti.

20. 6.4.35 śā hau.

21. 7.1.35 tuhyos tātaṅ āśiṣy anyatarasyām.

22. 6.4.101 hujhalbhyo her dhiḥ.

23. 7.2.76 rudādibhyaḥ sārvadhātuke.

24. 7.3.86 pugantalaghūpadhasya ca.

25. 6.1.9 sanyaṅoḥ.

26. ūṬH, by 6.4.19 chvoḥ sūḍ anunāsike ca.

27. 6.1.77 iko yaṇ aci.

28. 6.1.101 akaḥ savarṇe dirghaḥ.

29. Actually, this rule would apply here first anyway for another reason.

30. 6.1.87 ād guṇaḥ.

31. 6.1.85 antādivac ca.

32. 3.4.93 eta ai.

33. 6.1;77 iko yaṇ aci.

34. 8.2.23 saṃyogāntasya lopaḥ.

35. The interesting feature of Ex. 10 is that it shows how apparently quite “low-level” rules can become opaque through the effects of principle (3). Another striking example is nārpatya, where the final r of the first member of the compound does not become ḥ (contrast e.g. punaḥprasaṅga) because it is derived by the operation of vrddhi triggered by the suffix ya in the outer cycle. However, there are also several exceptions to (3) which involve antaranga rules in the last two sections of the grammar (8.3 and 8.4). It is sometimes held that (3) is not valid when the antaranga rule is in the Tripādī section (8.2.1 ff., see below). On that proposal Ex. 10 and similar cases would be the exceptions.

36. From akṣa + Śas + div + KvIP + sU (3.2.76 kvip ca), with the internal case suffix as deleted by 2.4.71 supo dhātuprātipadikayoḥ, and the derivational suffix v (KvIP) deleted by 6.1.67 ver apṛktasya.

37. ūTH, by 6.4.19 chvoḥ śūd anunāsike ca.

38. 7.2.35 ārdhadhātukasyeḍ vāladeḥ.

39. 6.4.131 vasoḥ saṃprasāraṇam.

40. 6.1.73 che ca.

41. 6.4.19 chvoḥ sūḍ anunāsike ca.

42. 7.4.40 dyatisyatimāsthām it ti kiti.

43. With t inserted by 6.1.71 hrasvasya piti kṛti tuk.

44. Pbh. 55 antarahgān api vidhln bahirango lyab bādhate.

45. 3.2.115 paro’kṣe liṭ, 3.2.107 kvasuś ca, 6.1.8 liṭi dhātor anabhyāsasya, 7.4.59 hrasvaḥ, 1.2.46 kṛttaddhitasāmasāś ca, 4.1.2 sv au jas am auṭ chaṣ ṭā bhyām bhis ṅe bhyām bhyas ṅasi bhyām bhyas ṅas os ām ṅy os sup.

46. 6.4.131 vasoḥ saṃprasāraṇam.

47. 6.1.108 samprasāranāc ca.

48. 6.4.64 āto lopa iṭi ca.

49. 6.4.133 śvayuvamaghonām aṭaddhite.

50. 6.1.101 akaḥ savarṅe dīrghaḥ.

51. 6.1.87 ād guṇaḥ.

52. 7.2.102 tyadādīnām aḥ.

53. 4.1.4 ajādyataṣ ṭāp.

54. 6.4.134 allopo’naḥ.

55. 8.4.40 stoḥ ścunā scuḥ.

56. 8.2.77 hali ca.

57. 3.4.82 parasmaipadānām ṇalatususthalathusanalvamāḥ.

58. 2.4.40 liṭy anyatarasyām.

59. 6.1.8 liṭi dhātor anabhyāsasya.

60. 7.4.60 halādiḥ śeṣaḥ, 7.4.62 kuhoś cuḥ, 8.4.54 abhyāse car ca.

61. 6.4.98 gamahanajanakhanaghasāṃ lopaḥ kṅity anaṅi.

62. 8.3.60 śāsivasighasīnāṃ ca.

63. S.4.55 khari ca.

64. 3.1.77 tudādibhyaḥ śah.

65. 7.3.84 sārvadhātūkārdhadhātukayoḥ.

66. 7.3.86 pugantalaghūpadhasya ca.

67. 6.4.77 aci śnudhātubhruvāṃ yvor iyaṅuvaṅau.

68. That is, before suffixes with diacritic K or N (1.2.5 kṅiti ca) .

69. 1.2.4 sārvadhātukam apit.

70. Of course the tradition has seen the possibility that 1.1.5 kṅiti ca blocks guṇa here. It rejects it on insufficient grounds (Mbh. on 1.1.5).

71. From adhi + i + Ktvā by 7.1.37, cf. above.

72. 6.1.101 akaḥ savarne dirghah.

73. 6.1.71 hrasvasya piti kṛti tuk.

74. 6.1.86 ṣatvatukor asiddhaḥ. Were it not for this type of example, the mention of tuk in 6.1.86 would be unnecessary.

75. 7.2.107 adasa au sulopaś ca.

76. 3.2.109 upeyivān anāśvān anūcānas ca.

77. 6.1.8 liṭi dhātor anabhyāsasya.

78. 7.2.67 vasv ekājādghasām.

79. A special rule for this root, 6.4.81 iṇo yaṇ. The general rule 6.1.77 iko yaṇ aci (Ex. 6, fn. 27) would not be applicable.

80. 6.4.71 luṅlaṅlṛṅkṣv aḍ udāttaḥ, 6.4.72 āḍ ajādīnām.

81. 6.4.111 śnasor allopaḥ.

82. 6.4.76 irayo re.

83. 6.4.64 āto lopa iṭi ca.

84. In principle, 6.4.22 allows Pāṇini to have simultaneous ordering of rules in cases which could not be fit into a framework making exclusive use of linear ordering. Though Pāṇini does employ this capability a few times, it is interesting that the phonology of Sanskrit never clearly requires it. The standard case is the derivation śās+hi + śā+dhi ‘instruct!’ by rules replacing the root śās by śā before the suffix hi (6.4.35 śā hau, cf. fn. 20) and a rule replacing the suffix hi by dhi after a root ending a consonant (6.4.101 hujhalbhyo her dhiḥ) . Either of the possible ordered applications would give the wrong result: *śāhi if the śās ➞ śā rule is applied first, and *śāsdhi if the hi ➞ dhi rule is applied first. But in a framework of linear ordering the problem can be gotten around easily by simply changing the environment of śās ➞ śā from hi to dhi.

85. This section is studied closely in Buiskool (1939).

86. 8.3.19 lopaḥ śākalyasya.

87. 6.1.87 ād guṇaḥ.

88. 8.2.7 nalopaḥ prātipadikāntasya.

89. 7.1.9 atobhisaais.

90. The reason why n-deletion applies here is that the stem is technically also a word (pada) before bhis (and other consonantal endings) by virtue of a special rule.

91. 6.4.8 sarvanāmasthāne cāsaṃbuddhau.

92. See Buiskool (1939:65), who refers to this type of situation as “secondary asiddhatva.”

93. 6.1.88 vṛddhir eci.

94. 6.1.95 omāṅoś ca. This rule also gives another case in which such exceptional contraction takes place, viz. the word om, e.g., śivāya om ➞ śivāyom ‘om to Śiva’.

95. From bhū + yās + t by 3.4.107 suṭ tithoḥ.

96. 8.2.29 skoḥ saṃyogādyor ante ca.

97. From madhu + lih sad + a + ti (and ultimately madhu + am + lih + KvIP + sU + sU sad + Laṭ) .

98. 8.3.29 ḍaḥsidhuṭ.

99. 8.4.55 khari ca.

100. Note that this is one of the exceptions to (3) which are found in the Tripādī section, as mentioned above in fn.35.

101. Cf. lakṣye lakṣanam sakṛd eva pravartate (see on Pbh. 111).

102. 8.4.47 anacica.

103. 7.4.59 hrasvaḥ. Cf. jagāha ‘plunged’, from gāh.

104. Note 18.

105. Note 28.

106. 6.1.85 antādivacca.

107. Pbh.l 11: parj any aval lakṣaṇapravṛttiḥ.

108. Cf. Kiparsky (1973b) on the need for adopting this principle in generative grammar too.