siddhaprinciple - PDF document

1 extended siddha-principle Paul Kiparsky
extended siddha-principle Paul Kiparsky 1 The siddha P ¯ . ini' s grammar several types of metarules which determine how rules apply . Among them are “trafc rules” which how rules interact with other grammatical deriva- tions. These are typically formulated as designating a rule or class of rules asiddha “not effected” (or asiddhavat “as effected”) with rule or class of rules. economy , the rules so designated are grouped into several sections, whose headings collectively declare them asiddha(vat). The such section, under the famous heading 8.2.1 p ¯ urvatr ¯ asiddham extends from 8.2.1 through the end of the grammar (8.4.68), called the T rip ¯ ad ¯  `Three Sections'. [1] 8.2.1 p ¯ urvatr ¯ asiddham (Any section is) asiddha with any previous the grammar). The [1] makes rule that falls invisible rules that precede This equivalent that the rules T rip ¯ ad ¯  apply in order which they are enumerated, after which derivation terminates. The rst commentator, K ¯ aty ¯ ayana, states that for rule A asiddha with B means two things. [2] a. ¯ Ade´ . an . apratis . edha `prohibition of operations conditioned by the output'. The of A invisible A cannot “feed” Utsargalaks . an . abh ¯ ava `allowing operations conditioned by the input'. The of A A cannot “bleed” anti-feeding function [2a] of the asiddha relation, a standard example .Pl. r ¯ . `by kings'. The following two rules are relevant its derivation: [3] a. 8.2.7 nalopah . pr ¯ atipadik ¯ antasya Stem-nal n at the end of a word. E.g. r ¯ ! r ¯ ajasu (there internal word boundary before -su 7.1.9 ato bhisa After a ending a, Instr .Pl. by ais. E.g. . a-ais ! vrks . a-bhis (! vrks . aih . , by other rules) `by trees'. These rules potentially interact, for the result of applying 8.2.7 n-stem r ¯ ajan-bhis r ¯ , a form which 7.1.9 is but must be allowed to . it apply , cause -bhis r ¯ replaced by -ais, just as stems (such as vr . ks . a-bhis ! vr . ks . a-ais In words, the following derivation must prevented: 1 r ¯ ajan-bhis r ¯ aja-bhis 8.2.7 nalopah . pr ¯ atipadik ¯ antasya *r ¯ aja-ais 7.1.9 ato bhisa *r ¯ ajaih . (other rules) The grammar achieves this by putting 8.2.7 nalopah . pr ¯ atipadik ¯ antasya into the T rip ¯ ad ¯  sectio

2 n and putting rule 7.1.9 ato bhisa earli
n and putting rule 7.1.9 ato bhisa earlier . The [1] then blocks 8.2.7 from feeding 7.1.9 (from supplying new inputs it). The function [2b] of the asiddha-relation can be illustrated with derivation of bhugna `bent', from bhujO-Kta (Here below the capitalized for markers, which not part of the phonological representation but a variety idiosyncratic grammatical information.) Again are two rules at stake: [5] a. 8.2.45 odita ´ s ca t . t . h ¯ a sufxes (such as the past participle ending -Kta is by n after roots with marker O. E.g. Opy ¯ ¯ I-Kta ! p ¯ `swollen'. 8.2.30 coh . . A by a velar an obstruent and at the end of a word. E.g. vac-tumUN ! vaktum bhujO-Kta both rules are applicable: the replacement of -na after the root bhujO, and the substitution of the root-nal an obstruent. -ta ! were take effect rst, bleed ! -g, the wrong form *bhujna (! *bhuj ˜ na by 8.4.41 ´ scun ¯ a ´ scuh . In that [5b] `does not count' with is after in T rip ¯ ad ¯ . is asiddha with it, fails to commentators' analyses make clear, imposing the asiddha relation on operations equivalent the relative which rules that enjoin those operations take effect. seems that P ¯ an . ini fact thinks of the asiddha relation as a restriction on rule ordering. Furthermore, P ¯ an . ini seems that rules take effect one after the other (and not simultaneously). 1 The “ A (not effected) with has the same import as the restriction “B and A take effect order”. Therefore, instead of saying that 7.1.9 stems because the operation of n-deletion enjoined by 8.2.7 with it [1], will that [1] prohibits derivations where 8.2.7 takes effect before 7.1.9 (such as derivation allows derivations where 7.1.9 takes effect before 8.2.7 (such as derivation r ¯ ajan-bhis — 7.1.9 ato bhisa (inapplicable) r ¯ aja-bhis 8.2.7 nalopah . pr ¯ atipadik ¯ antasya Similarly , of saying that 8.2.30 because the operation enjoined by 8.2.45 with it [1], will that [1] prohibits derivations where 8.2.45 takes effect before 8.2.30, and allows derivations where 8.2.30 takes effect before 8.2.45. Trafc such as [1] function as restrictions on more principles that govern how P ¯ an . ini' s rules interact. Although these background principles are not stated . t . ¯ ¯ ay ¯  infer them 1 noted by

3 Bronkhorst 1980, this concluded from .
Bronkhorst 1980, this concluded from . ini's asiddhavat rather than asiddha 6.4.22, where the effect simultaneous application desired. 2 the derivations of the system far these can be ascertained from such independent knowledge of the grammar' s outputs as of course), from overt stated grammar, from other internal clues wording of the text. These strands of evidence converge show , particular, that the converse of the asiddha relation, namely the siddha relation, holds between rules of the grammar some restriction stated grammar otherwise. Like asiddha relation, the siddha relation has two aspects. [7] a. ¯ Ade´ . an . abh ¯ ava `allowing operations conditioned by the output': The of A that, A new inputs B applies B). Utsargalaks . an . apratis . edha `prohibition of an operation conditioned by the input: the input of A invisible that, A removes B does not apply B). P ¯ . ini uses the `not realized, not effected' and asiddhavat “as realized, as effected”, functions of [7], ¯ ade´ salaks . an . apratis . edha and utsargalaks . an . abh ¯ ava. Although he does not use positive counterpart siddha “realized, effected” as a technical term, knows the concept, for his asiddha- restrictions only make as limitations on an implicit that rules are siddha with other far possible, so that that when a a form, the relevant effects of other rules are taken into account. The we the the siddha-principle (for discussion and various formulations of Joshi and Kiparsky Kiparsky Joshi and Roodbergen 1987, and Joshi and Kiparsky Maximize relations. The maximize relations to rules whatever order yields a result that different from the result of applying them simultaneously . words, rules apply order which they interact as much possible (which maximizes and bleeding). There be no doubt that the siddha-principle of the design of the grammar . Not does give the right result overwhelming majority derivations, and exploited the simplest possible wording of each rule, but where fails to the right output, P ¯ an . ini takes measures example, the placement of a rule into the Trip ¯ (the section headed by [1]) invariably motivated directly by the need prevent feeding and/or bleeding an earlier rule, or indirectly by a relation that a rule (Buiskool 1939). Our that the siddha-principle d

4 enes the normal, default modes of ru
enes the normal, default modes of rule interaction . t . ¯ ¯ ay ¯  (the “unmarked order”) uncontroversial as far siddha relations of type [7a] are concerned. The does not give this part of the siddha-principle a special name, but clearly takes granted. the part of the siddha-principle that relates relations of type [7b], the situation more complex. is similar to grammarians' nitya-principle: [9] A a `constant' rule with B if A is whether not B applies, but conversely . A nitya rule has precedence over a nitya rule. A typical instance of the nitya-principle which subsumed under the siddha-principle derivation of tad `that' (Neuter Sg.): tad-sU tad 7.1.23 svamor napum . ¯ at 3 The nominative singular case ending by the following morphologically conditioned rule. [11] 7.1.23 svamor napum . ¯ at Nom.Sg. and Acc.Sg. -am are deleted after neuter stems. until the case ending the conditions of rule [12] are met: 7.2.102 tyad ¯ ad ¯ ¯ ah . Before a case ending, (the nal segment) of pronouns of the tyad-class by a. rule were derivation [10], produce a vowel that would then be contracted by [13], [13] 6.1.97 ato gun . e a of the form: short non-word-nal a + a, o . a), the is with end result *ta: [14] tad-sU taa-sU 7.2.102 tyad ¯ ad ¯ ¯ ah . taa 7.1.23 svamor napum . ¯ at *ta 6.1.97 ato gun . e The this derivation wrong the nitya-principle (and a fortiori the siddha-principle, which a generalized form of the nitya-principle). Rule 7.2.23 with 7.1.102 because 7.2.23 whether 7.1.102 takes effect or not, while 7.2.23 takes effect, 7.1.102 longer applicable. The rule 7.2.23 has priority; hence, -sU rst after which ! -a is But import of the siddha-principle goes beyond that of the nitya-principle several respects. First, tradition assigns the nitya-principle a minor role because to antara ¯ -principle. cited references, have presented evidence that the antara ¯ -principle does not apply within but merely gives word-internal operations priority over that apply across word boundaries. there competition between the nitya- and antara ¯ -principles within and the nitya- principle form [7b]) comes a much larger role system than traditionally recognized. Secondly , the traditional form of the nitya-principle, holds only for simultaneously applicable rules: [15]

5 nity ¯ antara ¯ ngayor balavattvam ap
nity ¯ antara ¯ ngayor balavattvam api yaugapady ¯ asam . bhava “ A and an antara ¯ nga (rule) likewise then possesses greater force (or, words, only then supersedes, or takes effect before, another rule), when take effect simultaneously (with rule). ” ¯ age´ sa, on Pbh. 49, . Kielhorn). The ¯ asam . bhava to simultaneous applicability of rules. What says, then, the nitya relation (as well the antara ¯ nga relation) are locally assessed. following sections of our attempt the contrary position, that the siddha-principle, unlike the the nitya-principle, has a crucially global “lookahead” character . The effect of adopting the global version of the nitya-principle to even more, beyond the expansion already created by the elimination of the antara ¯ nga- principle from word-internal domains. 4 The global nature of the siddha The formulation still vague. There several ways maximize relations derivations. Let explore two of them more the RESTRICTED -principle and the EXTENDED - principle. The -principle applies at each point a derivation determine which the rules applicable at that point should take effect. The extended siddha-principle scans entire candidate derivations and chooses the one which -relations are maximized. a. Restricted siddha-principle: a local condition which rule application. Extended siddha-principle: a global (transderivational) “lookahead” condition on derivations. After explaining how the two versions of the siddha-principle work will internal evidence from the wording of the . t . ¯ ¯ ay ¯  that P ¯ an . ini assumed the . T o siddha-relation what exactly mean that a rule or operation A “effected”, or asiddha “not effected”, with rule or operation The A or invisible B in a derivation. Kiparsky proposed that A with B B applies crucially as A not taken effect, sense that B would produce a different result it as A taken effect; otherwise A with T o facilitate formal statement of this and other denitions of the siddha relation, let us introduce a bit of notation. Let C () for the result of applying C . Then B(A( is result of rst rule A , then applying rule B result. And A,B() result of applying A and B simultaneously ). Using notation, the idea that asiddha means “

6 crucially non-interacting” by the f
crucially non-interacting” by the following denition: [17] Denition 1 B(A( A with B if B(A( = ) A(B( 6  otherwise A w .t. denes the asiddha relation positively as crucial non-interaction, and makes anything which asid- dha siddha. Now us consider a somewhat different idea: dening asiddha as crucial non-interaction, let us redene siddha as crucial interaction leaving the relations undened case of mutually non- interacting rules: [18] Denition 2 B(A( A with B if B(A( = ) A(B( 6  A w .t. B B(A( 6 ) otherwise the siddha and asiddha relations are undened. The denitions [17] and [18] converge for rules that can interact, but they differ for rules that can't interact, namely where B(A( = ) = A(B( In case, A and B are vacuously siddha with other B(A( A(B( [17], but siddha relation undened by [18]. 5 Under the former denition, a vacuous siddha relation formally a one, under the latter, the crucial one counts. The -principle will differently depending on which denition rst might seem that the distinction between [17] and [18] Why matter whether a or not with a that interact with? Indeed, examples have discussed so far, makes difference. Let' s how the new interpretation of the siddha relation as crucial interaction works derivation of tad Applying ! -a -sU-deletion simultaneously results , as does applying them order (see [10]). The -principle accordingly dictates that -sU-deletion should be applied before ! -a. of -sU deletion, makes difference whether ! -a taken effect or not (we say that the siddha relation undened cases). So, effect of maximizing siddha relations nitya rules get precedence. Y et, surprisingly , the two siddha-relations lead different versions of the siddha-principle: [17] commits us principle [16a], which locally . dening siddha more narrowly , [18] makes siddha-principle more Only allows formulation of the extended siddha-principle [16b], a “look ahead” condition that maximizes siddha-relation across derivations. next sections will that this version one that operates P ¯ . ini' s gramnar . 3 Evidence for the extended siddha The derivation of adh ¯ `having approached' from adhi-i-Ktv ¯ a involves processes: the replacement of the absolutive sufx -Ktv ¯ a -LyaP

7 after prexed roots, insertion of the
after prexed roots, insertion of the augment a short root vowel -LyaP and vowel here i i ! ¯ . [19] a. 6.1.101 akah . savarn . e d ¯ rghah . contact, (a, r . , l . ) and a following vowel the same color are (together) replaced by a long vowel. 7.1.37 sam ¯ ase 'na ˜ ¯ urve ktvo a compound that does not begin with a ˜ , -Ktv ¯ a is by -LyaP c. 6.1.71 hrasvasya piti . tuk t is after a short vowel a kr . t sufx marked with P. If further ordering restrictions are placed on the three rules, the outcome ¯ ya. This determined as follows. The adhi-i-Ktv ¯ a and [19b]. takes effect rst, [19b] still and the derivation terminates ¯ ya. takes effect rst, both [19a] and [19c] are applicable output, and the siddha-principle selects [19a], because is , after which derivation again terminates ¯ ya. The grammar achieves the right output adh ¯ stipulating that vowel with of the augment tUK : 2 6.1.86 s . atvatukor asiddhah . The following rules (up 6.1.110) are not effected with -retroexion and insertion of the augment tUK [19a] invisible so that t of the surface length of the root' s vowel, adh ¯ . The derivations can be represented by a as follows (I omit the silent diacritics): 2 Again, 6.1.86 general than that, but the other asiddha relations that play a role here. 6 adhi-i-tv ¯ a adh¯ -Ktv ¯ a adh¯ ya adhi-it-ya adh¯ t-ya The through the boxed forms correct derivation. The remaining two paths, which converge on unwanted *adh ¯ ya, are excluded by [20] 6.1.86 s . atvatukor asiddhah . . How exactly stipulating that contraction w .t. t accomplish that, on the understanding that restricts ordering? There problem after ¯ a is by -ya give adhi-i-ya point, either version of the siddha-principle requires contraction, being the nitya rule, rst. But the initial stage of the derivation, the -principle does not choose between [19a] and [19b]. words, how do exclude the leftmost path (*adh ¯ ya via adh ¯ -Ktv ¯ a)? The answer is excluded because it (contraction) asiddha not siddha (t-insertion), while in actual (boxed) derivation is . we think of [20] as a lter which excludes all derivations which asiddha This example the global “lookahead” character of P ¯ an . inian derivations would

8 have worked with old denition [17].
have worked with old denition [17]. The for [18] demonstrated examples where siddha relations must chosen over what would have been vacuous siddha-relations. The active perfect participle of sad is , where replaces the reduplicated stem sa-sad- by rules which not be detailed here. stems (i.e. oblique vocalic sufxes such as Gen.Sg. sufx -vas vocalized by rule 6.4.131, given e.g. Gen.Sg. . ah . . Otherwise, when does not undergo vocalization, receives an initial augment i under by rule 7.2.67, given as Nom.Sg. ¯ an (with -sU a. 6.4.131 vasoh . . pras ¯ . am The semivowel v of the sufx -vasU vocalized stems. 7.2.67 vasv ¯ aj ¯ adghas ¯ am The augment . is before -vas after a monosyllabic root. c. 6.1.8 . i ¯ anabhy ¯ asasya Before . sufxes, an unreduplicated root The derivations of Nom.Sg ¯ ¯ shows the insertion of the augment i before -vas by [22b]. [23] sad sad-lIT . 3.2.115 paroks . e . sad-KvasU 3.2.108 ¯ . ¯ ay ¯ am . sadavasa ´ sruvah . sasad-vas [22c] 6.1.8 . i ¯ anabhy ¯ asasya, etc. sed-vas 6.4.120 ata ekahalmadhye ¯ ade ´ s ¯ ader . i sed-vas-sU 4.1.2 svau. . . sed-ivas-sU [22b] 7.2.67 vasv ¯ aj ¯ adghas ¯ am sediv ¯ an (other rules) 7 derivation of sedus . ah . , vocalization of -vas by [22a] bleeds i-insertion (the rst of the derivation same as and repeated [24] sed-vas sed-vas- ¯ Nas 4.1.2 svau. . . sed-us-as [22a] 6.4.131 vasoh . . pras ¯ . am sed-us . -ah . (other rules) is second derivation that requires the extended siddha-principle. is prevent i-insertion from taking effect while its environment still present as [25] sed-vas sed-ivas 7.2.35 ¯ ardhadh ¯ atukasyed . val ¯ adeh . sed-ivas- ¯ Nas 4.1.2 svau. . . sed-ius-as 6.4.131 vasoh . sampras ¯ aran . am *sed-yus . -ah . (other rules) Here the possible derivations (beginning with stage sed-vas): [26] sed-vas sed-ivas sed-vas-as sed-ivas-as sed-us-as sed-ius-as sed-yus-as stage sed-vas, there a between adding the case ending (sed-vas- ! sed-vas- ¯ Nas adding . ( ! sed-ivas- The output derived only afxation takes effect rst. at the stage sed-vas- ¯ Nas siddha-principle favors . pras ¯ aran . a (the nitya rule) over . , the output . ah . . the other hand, . is rst, derivation proceeds inexorably from sed-ivas wrong output *sed-yus . -ah . , as shown abov

9 e. question why , the stage sed-vas, cas
e. question why , the stage sed-vas, case afxation gets priority over . -augmentation. The -principle does not say anything about The extended siddha-principle does. the derivation because a siddha-relation between the sam . pras ¯ aran . a and . which Denition 2 present . takes effect before the case ending The relevant siddha-relations according two denitions and [18] are displayed [27] The of sam . pras ¯ aran . a (6.4.131) iT . (7.2.35) derivations [24] and [23]: by [17] by [18] siddha siddha siddha undened 8 the siddha-relation rule interactions, [18] characterizes the fact that derivation [24] optimal because maximizes interaction. not just avoid violations of the siddha-principle, but is a the siddha-principle does as much derivation as possible. The following example similar in involves relationship of sam . pras ¯ aran . a with augmentation rule. tense forms, verbs receive an augment, whose shape depends on whether root begins with a or with a vowel, Before a consonant, they get a short augment a- by 6.4.71 ¯ nla ¯ nlr . ¯ nks . v . ¯ . , and before a vowel, they get a long augment by 6.4.72 ¯ ad . aj ¯ ad ¯ ¯ . [28] a. 6.4.71 ¯ nla ¯ nlr . ¯ nks . v . ¯ . Short accented a before ending in) a ¯ n (aorist), la ¯ n (imperfect), . ¯ n tional) sufx. 6.4.72 ¯ ad . aj ¯ ad ¯ ¯ Long accented ¯ a (under the same conditions) before which begins with a vowel. Which of the augment depends not on the underlying shape of the root, but shape, as determined by morphological and phonological operations which sam . pras ¯ aran . a. [29] shows how this works derivation of aupyata sowed'. [29] vap-la ¯ N vap-ta (3.1.67 must here right output to derived) vap-yaK-ta 3.1.67 s ¯ arvadh ¯ atuke yak up-ya-ta 6.1.15 vacisvapiyaj ¯ adin ¯ am . kiti (6.4.72 but 6.4.71 wins the siddha-principle) ¯ a-up-ya-ta 6.4.72 ¯ ad . aj ¯ ad ¯ ¯ (now 6.4.72, not 6.4.71, aupyata (other rules) The augment must for the root vocalism changed by rule [30a], hence for the sufx yaK which that change added by [30b]. [30] a. 6.1.15 vacisvapiyaj ¯ adin ¯ am . kiti A semivowel by sam . pras ¯ aran . a roots vac, yaj,. . . before a sufx marked K. 3.1.67 s ¯ arvadh ¯ atuke yak yaK before a s ¯ arvadh ¯ atuka sufx which the Goal the Process. So, the stage

10 vap-ta why isn't the augment added immed
vap-ta why isn't the augment added immediately , which result ? [31] vap-la ¯ N vap-ta (suppose now 6.4.71 instead of 3.1.67 here) a-vap-ta 6.4.71 ¯ nla ¯ nlr . ¯ nks . v . ¯ . a-vap-yaK-ta 3.1.67 s ¯ arvadh ¯ atuke yak a-up-ya-ta 6.1.15 vacisvapiyaj ¯ adin ¯ am . kiti (other rules) The locally dened siddha-principle does not distinguish between this derivation and the one which before the augment. The extended, global siddha-principle does give the right result. the derivation over one because the former has two extra -relations, namely between the sam . pras ¯ aran . a rule 6.1.15 and the augmentation rules 6.4.71 (feeding, ¯ ade´ salaks . an . abh ¯ ava) and 6.4.72 (bleeding, utsargalaks . an . apratis . edha 9 vap-la ¯ N a-vap-la ¯ N vap-ta a-vap-ta vap-ya-ta a-vap-ya-ta up-ya-ta a-up-ya-ta ¯ a-up-ya-ta op-ya-ta a up-ya-ta 4 Are . pras ¯ aran . a The two examples, sedus . ah . and aupyata, have something common. the winning rule vo- a semivowel (sam . pras ¯ aran . a From cases, grammarians have drawn the generalization that sam . pras ¯ aran . a rules, such as [22a] and [30], have priority over rules is Pbh. 119 of N ¯ age´ sa'a paribh ¯ as . endu´ sekhara where is however rejected as superuous). This with gen- eral character of the grammar . The well-established unstated principles behind the grammar very Anything as parochial as a constraint on sam . pras ¯ aran . a rules would have been recorded as a rule gram- mar There a more objection, which simply sam . pras ¯ aran . a rules do not always have priority over rules. The derivation of ´ svayitv ¯ a `having swelled' a [33] ´ svi-Ktv ¯ a ´ svi-itv ¯ a 7.2.35 ¯ ardhadh ¯ atukasyed . val ¯ adeh . augment . ´ svi-itv ¯ a 1.2.18 ktv ¯ a . Ktv ¯ a marker K ´ sve-itv ¯ a 7.3.84 s ¯ arvadh ¯ atuk ¯ ardhadh ¯ ayukayoh . . a, triggered by sufxes not marked with K ¯ N ´ svayitv ¯ a (other rules) The relevant rules are given [34] a. 7.2.35 ¯ ardhadh ¯ atukasyed . val ¯ adeh . i before an ¯ ardhadh ¯ atuka sufx beginning with a other than y. 1.2.18 ktv ¯ a . Ktv ¯ a not have the marker K when the augment . . 7.3.84 s ¯ arvadh ¯ atuk ¯ ardhadh ¯ atukayoh . gun . ah . ) (6.4.1 a ¯ ngasya) Before s ¯ arvadh ¯ atuka and ¯ ardhadh ¯ atuka sufxes, (the last segm

11 ent of) by (gun . a In rule [34a] insert
ent of) by (gun . a In rule [34a] inserts the augment . a class of sufxes including the absolutive sufx -Ktv ¯ a. The augmented ¯ a the original triggering marker K, of rule [34b]. Therefore longer conditions sam . pras ¯ aran . a by [30], and triggers strong grade of the root by rule [34c]. derivation sam . pras ¯ aran . a replacement cannot be allowed to take place at the stage before the augment But contradicts the stipulation that sam . pras ¯ aran . a rules have priority over rules. That fact, predicts the wrong derivation 10 ´ svi-Ktv ¯ a ´ su-Ktv ¯ a 6.1.15 vacisvapiyaj ¯ adin ¯ am . kiti ´ su-itv ¯ a 7.2.35 ¯ ardhadh ¯ atukasyed . val ¯ adeh . ´ so-itv ¯ a 7.3.84 s ¯ arvadh ¯ atuk ¯ ardhadh ¯ atukayoh . ´ savitv ¯ a (other rules) the traditional nitya-principle, and the principle which they do not distinguish [35] from [33]. The extended siddha-principle, however, correctly selects the derivation [33] over The [33] instantiates two siddha-relations which not instantiated namely that [34a] 7.2.35 (via [34] 1.2.18) bleeds sam . pras ¯ aran . a by [30] 6.1.15, and that gun . a by [34c] 7.3.84. Thus, [33] derivation which rules interact maximally . The putative principle that sam . pras ¯ aran . a rules have priority over rules does not come even justice “lookahead” cases that the extended siddha-principle accounts for . Consider the derivation of forms like `they give'. The of the 3.Pl. ending depends on the form of the root according [36] a. 7.1.3 jho 'ntah . ant for jha a sufx. 7.1.4 ad abhyast ¯ at After a root, at (instead of ant) for jha a sufx. derivation shown 7.1.4 applies the the underlying the 3.Pl. ending by at because the root [37] dh ¯ a-jhi dh ¯ a- ´ SaP-jhi 3.1.68 kartari ´ dh ¯ a-( ´ Slu)-jhi 2.4.75 juhoty ¯ adibhyah . ´ sluh . da-dh ¯ a-jhi 6.1.10 ´ slau, etc. da-dh ¯ a-ati 7.1.4 ad abhyast ¯ at dadhati (other rules) After a simple have been replaced by ant instead by 7.1.3. The to prevent this from happening prematurely before reduplication actually takes effect: [38] dh ¯ a-jhi dh ¯ a- ´ SaP-jhi 3.1.68 kartari ´ dh ¯ a- ´ SaP-anti 7.1.3 jho 'ntah . dh ¯ a-( ´ Slu)-anti 2.4.75 juhoty ¯ adibhyah . ´ sluh . da-dh ¯ a-anti 6.1.10 ´ slau, etc. *dadhanti (other rules) Let' s at what wrong

12 here. stage dh ¯ a- ´ SaP-jhi two rule
here. stage dh ¯ a- ´ SaP-jhi two rules present themselves: [36a] 7.1.3 jho 'ntah . (jhi ! anti) and [39] 2.4.75 juhoty ¯ adibhyah . ´ sluh . ( ´ SaP ! ´ Slu 2.4.75 juhoty ¯ adibhyah . ´ sluh . After and the other roots of the ´ SaP by ´ Slu. 11 The -principle, and the traditional nitya-principle that do not decide between these two rules. These rules do not interact, so neither of them , or siddha by denition 1 (see [17]). And course there sam . pras ¯ aran . a rule involved. What ensure the correct derivation [37] extended siddha-principle. that [39] takes precedence because maximizes -relations derivation (under the denition of siddha The replacement of ´ SaP by ´ slu feeds 6.1.10 ´ slau, which bleeds 7.1.3 jho 'ntah . and feeds 7.1.4 ad abhyast ¯ at. jhi ! anti replacement applies rst, siddha-relations are lost. Thus, the correct derivation globally maximizes -relations, as the extended (lookahead) version of the siddha-principle requires. 5 Are morphologically conditioned The examples of the extended siddha-principle considered so far have something else common: the right derivation involves a “morphological” substitution before a “phonological” substitution. This gests yet another alternative extended siddha-principle. P ¯ an . ini could have made what modern terms amounts a between morphology and phonology all, distinguishes a class of rule which , which seems mean “phonological rule”). A derivation would proceed by lining up all morphemes, and then applying any substitution or augmentation rules that may applica- ble. This considered by the tradition under the rubric of the padasam . sk ¯ arapaks . a (and the v ¯ akyasam . sk ¯ arapaks . a, when same procedure level the whole sentence). Undoubtedly morphological operations do tend phonological operations Sanskrit, as they do languages. But neither be necessary nor sufcient impose a constraint effect. P ¯ an . inian derivations allow full range of interactions between morphological and phonological opera- tions. They be as the form built — the method called kramen . ¯ anv ¯ akhy ¯ anapaks . a. so is unlikely that P ¯ an . ini operated with a that gives morphological rules priority over logical rules. The tendency for morphology phonology derivations si

13 mply emerges from the extended siddha-pr
mply emerges from the extended siddha-principle, for afxation and other morphological operations usually determine the condi- tions under which processes apply , rather than vice versa. is an autonomous stipulation of the grammar, but a effect of the extended siddha-principle. Making P ¯ an . ini' s metatheory simpler and more this a attractive well historically plausible position. example the insufciency of stipulating the precedence of morphological operations over logical operations, derivation of asmai him', where extended siddha-principle the priority among morphological rules. The idam-e rules, 7.2.112 an ¯ apy akah . (id- ! ana [12] 7.2.102 tyad ¯ ad ¯ ¯ ah . (idam ! idaa, followed idaa ! ida ! a by other rules). two rules, is 7.2.102 which take effect: [40] idam-e (at this point, id- ! ana- by 7.2.112 an ¯ apy akah . must blocked) idaa-e [12] 7.2.102 tyad ¯ ad ¯ ¯ ah . ida-e 6.1.97 ato gun . e `short non word-nal a + a, o . a) by the . ' ida-smai 7.1.14 sarvan ¯ amnah . smai ` After a stem a, (Dat. Sg.) ¯ Ne by smai ' a-smai 7.2.113 hali lopah . ` ! ; a consonantal ending' A constraint which that morphological operations precede phonological operations does not decide between 7.2.102 and 7.2.112 at the rst of the derivation, since the losing rule 7.2.112 an ¯ apy akah . morphological. The -principle does not help either . But extended siddha- principle correctly chooses 7.2.102 over 7.2.112, because application of 7.2.102 creates an extra relation derivation, at the point where 7.1.14 sarvan ¯ amnah . smai feeds 7.2.113 hali lopah . . 12 Internal evidence for the extended siddha W e have seen that that the extended siddha-principle predicts the right rule interactions even where alternatives fail. W e now proceed a different kind of evidence which shows that just when extended siddha-principle predicts the wrong rule interaction, the grammar takes steps problem. The verb forms ¯ `he went' ¯ ayan `they went' offer a minimal which demonstrates that the siddha-principle its extended (lookahead) version plays a role design of the grammar, regardless of the morphological or phonological status of the rules. Both verb forms have the underlying root i, which by suppletive g ¯ a ¯ , and becomes y ¯ . Thus, the originally vocalic comes begin with a f

14 orms. Recall from [28] that the past ten
orms. Recall from [28] that the past tense augment ¯ before a vowel a- before a consonant. The -principle dictates that this distribution should be checked on the surface, predicting a short augment forms. This for ¯ `he went': [41] ¯ N i-ti 3.4.78 jhi. . . i-t 3.4.100 ´ s (up point, [28b] 6.4.72 ¯ ad . aj ¯ ad ¯ ¯ giving * ¯ ) g ¯ 2.4.45 . o g ¯ a lu ¯ ¯ a-t (now [28a] 6.4.71 ¯ nla ¯ nlr . ¯ nks . v . ¯ . inserts the augment a) Only extended siddha-principle works here. The augment a cannot be inserted until the sup- pletive root g ¯ a has replaced i, after which the root has a consonantal onset. This replacement by the inectional ending . stage , -principle (like tradi- tional nitya-principle) does not give priority replacement operation i ! g ¯ a because is nitya with augmentation. Nor, that matter, is it with lu ¯ N ! ti subsequent ! t. The extended siddha-principle works, however . the augment rule ahead” and apply a siddha relations are maximized over derivation. same token, however, the extended siddha-principle gives the wrong result for the other form, ¯ ayan `they went', where a by 6.4.72 ¯ ad . aj ¯ ad ¯ ¯ , as though the root begins with a vowel. begin with a vowel , but siddha-principle says that the relevant context surface y, which i before a vocalic by 6.4.81 . o . . With mind, P ¯ an . ini has put both relevant rules, 6.4.72 ¯ ad . aj ¯ ad ¯ ¯ and 6.4.81 . o . , into the special section headed by 6.4.22 asiddhavad ¯ a ¯ , which that all rules section are as with other . This sets aside the siddha-principle and that the root begins with a vowel, that the augment ¯ a as root vowel not been replaced by y. [42] ¯ N 3.4.78 jhi. . . , 7.1.3 jho 'ntah . point, the desired augment ¯ a could be derived by 6.4.72 ¯ ad . aj ¯ ad ¯ ¯ ) y-an 6.4.81 . o . ¯ ayan 6.4.72 ¯ ad . aj ¯ ad ¯ ¯ of 6.4.22 asiddhavad ¯ a ¯ ) This evidence that the construction of P ¯ an . ini' s grammar assumes the extended siddha-principle. Rules have been put under the scope of 6.4.22 only the siddha-principle. the extended siddha-principle provides sufcient reason for putting 6.4.81 into that section, that version that P ¯ an . ini must have worked with. 13 similar example 3.Pl. ¯ asan, where deletion of the root vowel as by 6.4.111 ´ snaso

15 r allopah . , so the long augment ¯ a-
r allopah . , so the long augment ¯ a- still Again, rules are put under the scope of 6.4.22 the siddha-principle. cakratus from kr -atus, the reduplication rule [22c] must syllabic form of the root, ignor- ing the replacement of r by r due glide formation rule [43]. [43] 6.1.77 iko yan . aci r . , l . (iK) ! y , v , r , l (yaN) . before a vowel () contact. The derivation is: [44] kr . . of kr . `protect') kr . 3.4.78 jhi. . . . -atus 3.4.82 parasmaipad ¯ an ¯ am. . . kr . -kr . -atus [22c] 6.1.8 . i ¯ anabhy ¯ asasya and other rules kr . -kr-atus [43] 6.1.77 iko yan . aci ca-kr-atus other rules antara ¯ nga-principle were operative inside words, predict this result (as K ¯ aty ¯ ayana points out). But right that the antara ¯ nga-principle does not apply inside words, nothing about the rule interaction example. contrary , the extended siddha-principle says that reduplication should “wait” for [43] the shape of the root. is signicant, therefore, that P ¯ an . ini has included a special rule for just these cases: [45] 1.1.59 dvirvacane The of a vowel, when by a following vowel, like original with The function of [45] really to things right where extended siddha-principle fails. It is instructive compare the derivation of dudy ¯ us . ati, where takes place before a conso- nant, the siddha-principle gives the correct result. [46] div-saN-tiP di ¯ u-sa-ti 6.4.19 chvoh . ´ s ¯ . anun ¯ ca dy ¯ u-sa-ti [43] 6.1.77 iko yan . aci dy ¯ u-dy ¯ u-sa-ti [22c] 6.1.8 . i ¯ anabhy ¯ asasya dudy ¯ us . ati other rules [45] makes vowel asiddha only before before a vowel , not before a consonant. This conrms that the author of the grammar well aware the extended siddha-principle and took care where Thus, our interpretation of the siddha-principle justies the wording of the grammar . Once proper domain of the antara ¯ -principle as giving priority operations over across word boundaries the issue whether siddha-principle than the antara ¯ -principle arises anew . There examples which that including the familiar gomatpriyah . `fond of cow-owners'. underlying ( (gomat+sU) + (priya+sU) )+sU, the rst member the compound can lose sufx -sU by one of two rules: 14 a. 6.1.68 hal ¯ ny ¯ abbhyo d ¯ rgh ¯ at sutisyapr . ktam . hal After a and after (

16 the feminine endings) long ¯ aP , ¯ n
the feminine endings) long ¯ aP , ¯ n ¯ , (the endings) -sU, , -siP deleted (replaced by the null element lopa) they consist of a single consonant. 2.4.71 supo dh ¯ atupr ¯ atipadikayoh . Case endings and stems are deleted (replaced by the null element luk The competition between lopa-deletion of the case sufx by rule 6.1.68, and -deletion by 2.4.71, has indirect repercussions. -deletion takes place, the stem will undergo the phonological changes triggered by the deleted ending (because of principle [48a] 1.1.62 pratyayalope pratyalalaks . an . am will develop into *gom ¯ an-. -deletion takes place, the stem will be affected by these changes (because of [48b] lumat ¯ a ¯ ngasya) and will emerge as gomat- [48] a. 1.1.62 pratyayalope pratyayalaks . an . am When a sufx the operations triggered by still . 1.1.63 lumat ¯ a
ngasya When a sufx by an element containing , operations on an a ¯ nga (stem) do not apply . priority between the two deletion rules were by the antara ¯ nga-principle, according which processes take precedence over processes, lopa deletion, which rst of the compound, should take precedence over deletion, which only whole compound. But would yield the wrong form *gom ¯ anpriyah . . 3 However, antara ¯ -principle extended siddha-principle, the correct result derived. extended siddha- principle gives priority deletion because (via 1.1.63) the phonological changes triggered by the deleted case ending, thereby establishing an extra relation. The principle the only trafc of the grammar works on a global, “lookahead” basis. The expressly says that the utsarga/apav ¯ ada (general/special rule) relation does so too. [49] Pbh. 64: upasam . janis . yam ¯ an . animitto 'py apav ¯ ada upasam . j ¯ animittam apy upasargam . b ¯ adhate “ An ¯ ada supersedes, even the causes of are to themselves, a general rule the causes of which already present. ” . Kielhorn) example derivation of ¯ ¯  with a (fem.), which must like [50] sva-ina kr¯ -Kta-sU svakr¯ 2.1.32 kartr . karan . e kr . t ¯ a bahulam svakr¯ ¯ . 4.1.50 ¯ ¯ karan . ¯ urv ¯ svakr¯ t¯  The feminine sufx - ¯ N ¯ . is ¯ in instrumental compound by the special rule 4.1.50. But the input stage, prior compounding, kr ¯ the ge

17 tting the general feminine sufx . ¯
tting the general feminine sufx . ¯ a. The sufxation rule must have the foresight until compounding establishes the proper environment for the feminine sufx. 4 3 T o get gomatpriyah . , Pata ˜ njali posits a exception the antara ¯ -principle: Pbh. 52 antara ¯ ¯ api vidh¯ bahira ¯ ngo b ¯ adhate bahira ¯ nga deletion luk even antara ¯ nga rules”. This quite obviously extremely unlikely to have a principle of . ini's grammar. The tradition (but not endorse) the idea that the “lookahead” effects a convention that operations when their trigger is eliminated (nimitt ¯ ¯ aye naimittikasyaiv ¯ apayah . This convention obviously too strong because forbids any type of utsargalaks . an . abh ¯ ava rule interaction. 4 See further the around Pbh. 75: gatik ¯ arakopapad ¯ an ¯ am . . . saha sam ¯ asavacanam . ¯ subutpatteh . . kr¯ -Kta kr¯ . ¯ a 4.1.4 aj ¯ ady atas . t . ¯ ap sva-ina kr¯ ¯ 4.1.2 svau. . . svakr¯ ¯ a 2.1.32 kartr . karan . e kr . t ¯ a bahulam 7 constraints The -principle, especially extended “lookahead” version, emerges rather naturally nonderiva- tional, constraint-based phonological theories. is of some interest whether P ¯ . ini' s gram- mar be reformulated terms without loss of generality . (This longer simply a matter interpreting P ¯ an . ini, of course, but using his work something about language.) The answer of the lookahead property , P ¯ an . ini' s system derivational. The context of rules can't necessarily be dened terms the output representation. Consequently , the rules must and the grammar be reformulated as an of constraint system without loss of generalization. The by the derivation of the perfect participle bibh ¯ -vas- (Nom. Sg. ¯ ¯ bhi- bh¯  bh¯ . 3.2.115 paroks . e . tense . is bh¯ -KvasU 3.2.107 kvasu ´ s . by KvasU bibh¯ -vas 6.1.8 . i ¯ anabhy ¯ asasya, etc. reduplication derivation, rule 7.2.67 vasv ¯ aj ¯ adghas ¯ am becomes applicable at the intermediate stage when sufx -KvasU has been added but has not taken effect yet: [53] 7.2.67 vasvek ¯ aj ¯ adghas ¯ am The augment . is ¯ ardhadh ¯ atuka sufx vasu after a monosyllabic stems, stems ending - ¯ and ghas This throws a potential spanner works, for interpolating an application

18 of [53] leads * ¯ an: [54] bh¯ 
of [53] leads * ¯ an: [54] bh¯  bh¯ . 3.2.115 paroks . e . bh¯ -KvasU 3.2.107 kvasu ´ s bh¯ -ivas 7.2.67 vasv ¯ aj ¯ adghas ¯ am bibh¯ -ivas 6.1.8 . i ¯ anabhy ¯ asasya, etc. bibhy-ivas 6.4.82 er anek ¯ aco `sam . ogap ¯ urvasya Each rule applies properly . The ¯ ac (monosyllable) condition on 7.2.67 satised both input and output. But fact 7.2.67 cannot apply . Our explains why . The derivation violates the extended siddha-principle at the stage bh ¯ -vas. The extended siddha-principle requires reduplication before . because the augment . is inserted before -vas because the reduplicated root and 7.2.67 vasv ¯ aj ¯ adghas ¯ am allows of . -vas only after a monosyllabic (ek ¯ ac) root. 16 The of this example lookahead allows “amnesties”: each rule applies only when conditions on are satised, and the derivation must obey the siddha-principle. That why P ¯ . ini' s phonological rules cannot be translated into constraints (even violable paper have shown three things: (1) that the two versions of the siddha-principle differ formally in how the siddha-relation dened, (2) that they differ substantively the version sub- sumes exactly traditional nitya-principle, whereas the extended version subsumes a generalized form of (3) that P ¯ an . ini' s grammar the extended siddha-principle. This nding further strengthens our claim that the siddha-principle fundamental principle governing the interaction of rules P ¯ . ini' s grammar . Bibliography RONKHORST OHANNES Asiddha . t . ¯ ¯ ay ¯ : a misunderstanding among the traditional commentators. UISKOOL OHANNES The T ¯ ad ¯ . Leiden: Brill. OSHI & AUL IPARSKY . Siddha and asiddha P ¯ . inian Phonology . Dan Dinnsen (ed.) Current Approaches Phonological Theory Bloomington, Indiana: Indiana University Press. OSHI & AUL IPARSKY . On antara ¯ nga-paribh ¯ . ¯ OSHI & J.A.F. OODBERGEN The of asiddhatva and sth ¯ anivadbh ¯ ava P ¯ . ini' s . t . ¯ ¯ ay ¯ . CASS , Number Publications of the Centre of Advanced Study Sanskrit. Class Number OSHI & J.A.F. OODBERGEN asiddha and sth ¯ anivat. Annals of the Bhandarkar Oriental Research Institute 68:541-549. IPARSKY , AUL Some theoretical problems P ¯ . ini's grammar . Poona: Bhandarkar Oriental Research Insti