You are on page 1of 15

On Local Ordering of Rules in Sanskrit Author(s): Elaine Phelps and Michael K.

Brame Reviewed work(s): Source: Linguistic Inquiry, Vol. 4, No. 3 (Summer, 1973), pp. 387-400 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/4177778 . Accessed: 09/04/2012 01:50
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

http://www.jstor.org

IV Number3 (Summer, LinguisticInquiryVolume 1973)

387-400.

Remarks and Replies

On Local Ordering of Rules in Sanskrit* Elaine Phelps and Michael K. Brame


i.

Introduction

An apparent ordering paradox in Sanskrit involving two rules, Grassmann's Law (GL) and Deaspiration (DaC), leads Anderson (I970) to the conclusion that linearly ordered rules cannot handle the facts of Sanskrit in a natural way. He proposes instead a theory of local ordering of rules that makes use of a modified version of Kiparsky's (I968) notion of marked and unmarked order of rules. We shall attempt to show that Anderson's proposal is defective in several ways and that the examples he cites from Sanskrit can be accounted for by linearly ordered rules.
2.

The Problem

Anderson posits three rules for Sanskrit which we reproduce below as (i), to be followed by examples drawn from Anderson's data. In the examples, we use a hyphen to separate elements of the stem, and a plus sign to separate the stem from the suffix. Aspiration is indicated by h.
(i)

a.

Law (GL) Grassmann's


[+ cons]
>

[-asp]

[+seg]0

+cons + asp

Root. L+ Law (BL) b. Bartholomae's


[-cont]
-

>

+ i [+asp_,

+ a [i + aspj

i - void

c.

Deaspiration (DaC)
[+ cons] > [-asp]/ {[+ obst]}

Copyright (?

973

by Elaine Phelps and Michael K. Brame

* We wish to thank Charles E. Cairns and Richard Demers for comments and criticism.

388

REMARKS AND REPLIES

The following examples illustrate GL, Rule (i a).


(2)

a. /pha - phal + a/ b. /bhudh + ati/ c. /bhi - bhr + tha/

[paphala] [budhati] [bibhrtha]

That GL deaspirates the first of two aspiratesin a stem is shown by (2a) and (2c) for reduplicated forms and by (2b) for diaspirate roots. That the second aspirate must be in the stem is shown by the fact that we get [bibhrtha] and not * [bibrtha] in (2C). The following examples illustrate BL, Rule (ib). (3) a. b. c. /rundh + tas/ /rundh + thas/ /labh + ta/ [runddhas] [runddhas] [labdha]

In Sanskrit, a preceding obstruent normally assimilates to a following obstruent in voicing. But, by BL, suffix-initial t and th (the only [-cont, -voice] segments that occur in this position) assimilate to a preceding voiced aspirate in both aspirationand voicing. The deaspiration of the stem-final aspirate results from DaC, Rule (ic). The following examples illustrate DaC, Rule (ic). (4) Root a. path b. -vrdh c. bhudh JNom. Sg. pat vrt bhut Acc.Sg. patham vrdham budham Instr.PI. padbhis vrdbhis bhudbhis

Deaspiration occurs before an obstruent (instr. pl.) and before word boundaries (nom. sg.), but not before sonorants (acc. sg.). We see the effect of GL in the acc. sg. (4c), and of the general voicing assimilation rule in the instr. pl. (4a), as well as devoicing before word boundary in the nom. sg. (4b) and (4c). We shall be concerned here only with the effects and interaction of the rules in (i). It can be shown that, given those rules and forms such as the ones listed in (5), no single linear ordering of the three rules in (i) will account for the phonetic variation in aspiration for all of these strings. (5) a. b. c. d. e. /bhudh + /bhudh + /bhudh/ /bhudh + /bhudh + taf am/ syati/ bhis/ [buddha] [budham] [bhut] [bhotsyati] [bhudbhis]

An ordering paradox seems to exist: the order must be DaC-GL for (sc), (5d), and (se) to prevent GL from applying (recall that GL deaspiratesthe first of two aspirates in a stem); but (5a) requires the reverseorder, GL-DaC, so that both those rules and BL will apply, resulting in the voicing and aspirating of the t of the suffix and the

REMARKS

AND

REPLIES

deaspirating of both aspirates of the stem. Either order of GL and DaC will give the correct result for (5b); DaC cannot affect this form, since it applies only to a sequence of two obstruents, or to an obstruent followed by word boundary. The problem of ordering these rules arises because of the approximately ten have three phonetic variants in terms diaspirate roots in Sanskrit, which, like bhudh, of aspiration: [bhud], [budh], and [bud]. Anderson rejects two solutions that preserve linear ordering, one proposed by Zwicky and one by Kiparsky, and concludes that there is no way to resolve the paradox in a natural manner if we maintain linear ordering. Although we do not agree with every aspect of Anderson's criticism of Zwicky and Kiparsky, several of the points he makes are sufficiently compelling to inspire a search for other alternatives. In the following section we offer two sets of linearly ordered rules, after which we discuss Anderson's proposal in detail.

3. Alternatives

Preserving

Linear Ordering

As we shall see in Section 4, there is no inherent relationship between GL and BL in Anderson's formulation of these two rules. Our rules, however, are formulated so that GL and BL interact, which we believe to be necessary in a correct account of the facts of Sanskrit. We shall postpone discussion of exceptions to our rules until after we have presented both alternatives. Our first set of linearly ordered rules establishes the relationship between GL and BL by means of a transformational rule which combines both processes. (6) a. GL and BL [+seg] [+seg]0
r

+ asp

+ Root

<<
I

+ voice

- cont
] c

voc

>

obst]

1
I

~~~~2
Kr

=~

asp ~~~~~~+
+ voice/

-asp]
2

b.

DaC [+ cons]
>

[-asp]/

{[+obst]}

Rule (6a) requires that v be an underlying [-obst], but there is good evidence that this is so. Whitney (I889, 3) lists v among the "semivowels" along with r, 1, andy, and there are sandhi rules that support this assignment. The expansion of (6a) consists of two disjunctively ordered rules:

390

REMARKS

AND

REPLIES

(7) [+ seg] [+ seg]0

+ asp

+Root
[+ voice]

oc

-cont]
o

[obst]

3
+ asp 1

[-asp]
I
2

+ voice] 3

(8)
2
2

+ asp 1
+ seg] [+ seg] 0 L+ RootJ [-obst]
I

[-asp] 4 The following derivations illustrate the application of the rules.


I

(g)
GL-BL DaC Other

a.
bhudh + ta budh + dha bud + dha e. bhudh + bhis

b.
bhudh + am budh + am

c.
bhudh# bhud# bhut#

d.
bhudh + syati bhud + syati bhot + syati

GL-BL
DaC bhud + bhis

Other
is an added restriction which prevents GL from applying to Term 4, [-obst], /bhudh#/, /bhudh + syati/, and /bhudh + bhis/, since in each of these forms the stem-final aspirate is not followed by t, th, or a [ - obst] segment. One of Anderson's arguments against Zwicky's statement of GL, and therefore presumably against (6a), is that it "seems strange" for a feature of the segment that follows the second aspirate to determine the aspiration of the first segment. We reject this criticism on several grounds. One could as well ask why a second aspirate in a root should affect the aspiration of the first aspirate, despite intervening segments, or why the second aspirate must be in the root. If the facts of the language require it, some rules may have to be more complex than we might wish. It is a fact of Sanskrit that GL, however formulated and however ordered with respect to the other rules, will never affect a string unless there is a sonorant ([- obst]) immediately after the second aspirate or unless the string also undergoes BL. It has been suggested that the rules in (6) make it appear to be merely a coincidence that GL fails to apply when the second aspirate is deaspirated by DaC. However, it is not always the case that GL fails to apply when the second aspirate is deaspirated, since we have forms like /bhudh + ta/ > [buddha],

REMARKS

AND

REPLIES

in which GL applies despite the fact that the second aspirate is deaspirated by DaC. This is precisely the source of the problem. GL deaspirates initial aspirates in two situations: (a) when the string meets the structural description of both BL and DaC, or (b) when the string meets the structural description of neither BL nor DaC. The only time that GL fails to apply to diaspirates is when the string meets the structural description of DaC but not that of BL. The relationship to be captured, therefore, is not only the one between GL and DaC, or the one between BL and DaC, but the interaction of all three processes with each other. The rules in (6) do this by having forms undergo GL and BL simultaneously (as seen in (7)), or by having GL apply in a context which excludes the possibility of having DaC apply (as seen in (8)), and finally by ordering DaC after the combined GL-BL rule. There is, however, another approach which attacks the problem via Bartholomae's Law. (io) a.

BL

- voice1
- cont
+-

+ voice]

+asp [-on ]+ Root

I I
/

+ voice]
asp

+ Root

b. DaC andGL

+ obst]
[+ cons] [-asp] / + asp

Rule (ioa) states that t and th assimilate to a preceding voiced aspirate of a root not only in voice and aspiration but in the feature [ + Root] as well. Equivalently, t and th are incorporated into the root. In the following derivations, the symbol "+" is used to indicate that every segment to its left is marked [ + Root].
(II)

a.
BL DaC GL Other bhudh + ta bhudhdh + a bhuddh + a buddh + a e. bhudh + bhis

b.
bhudh + am

c.
bhudh+# bhud +

d.
bhudh + syati bhud + syati bhot + syati

budh + am bhut +

BL
DaC bhud + bhis

GL Other
Every string that meets the structural description of BL also meets that of DaC: the input to BL and the output of BL contain a sequence [+cons] [+obst]. However, DaC can only deaspirate the first consonant in the cluster because, given the phono-

392

REMARKS

AND

REPLIES

logical structure of Sanskrit suffixes, the root-final aspirate that is created by BL is always followed by a sonorant ([- obst]), a context which does not meet the structural description of DaC. Thus, BL creates a root-final aspirate that cannot undergo DaC. This explains why it is that GL applies to strings that meet the structural description of DaC only if they also meet the structural description of BL, despite the fact that DaC precedes GL and removes input to GL. Again, it is both the form and the order of the rules that accounts for the seeming paradox. The rules of (i o) capture the interrelationship of the three processes by means that are quite different from those of (6). We prefer the rules of (io), but the standard theory of linear ordering does not at present provide an automatic choice between the two. It is most probable that the use of angled brackets must be restricted in some as yet undiscovered but principled fashion. Also unresolved is the question of what methods of dealing with exceptions are to be permitted within linear ordering. Anderson claims that his analysis under local ordering permits exceptions to be treated in the simplest way, since a form such as /dhadh + ta/, which is phonetically [dhatta] instead of the expected * [daddha], need only be marked [- BL]. By this criterion, the rules of (i o) are to be preferred to the rules of (6) since the exceptions can be marked [- BL] as in Anderson's analysis, while the rules of (6) require that exceptions be marked minus the first expansion of (6a); the question of whether an item can be marked as an exception to part of a rule has not yet been resolved within the standard theory of linear ordering. Other things being equal, the handling of exceptions can play a role in deciding among grammars. But things are not equaleach set of rules, as well as Anderson's proposal, makes a different claim about Sanskrit which can be evaluated only within the framework of a more extensive grammar of the language. For example, we have ignored, as has Anderson, the traditional distinction between roots and stems. According to Whitney, /dhadh/ comes from the root /dha:/ and is therefore not a root itself but a derived stem. A grammar of Sanskrit would have to account for this, and in so doing the status of exceptions such as /dhadh/ might be considerably altered. While nowhere in Whitney can one find any forms such as /dhadh/ or /bhudh/ actually cited with two aspirates, Anderson's interpretation of Whitney's views regarding the relationships between what we have been calling BL, GL, DaC, and diaspirate roots is different in almost every respect from the following interpretation (which is largely due to Robert Vroman), in support of which we cite a number of passages from Whitney (I889) in addition to those referred to by Anderson. Permitted Finals. 141. Of the non-nasal mutes, only the ... non-aspirate surd is allowed. W4Ia.In a few roots [the diaspirates], when their final (sonant aspirate) thus loses its aspiration, the original sonant aspiration of the initial reappears. Interpretation:When DaC deaspirates (and devoices) the root-final aspirate before word boundary, GL cannot apply and the initial aspirate remains.

REMARKS

AND

REPLIES

I53. An aspirate mute is changed to a non-aspirate before another non-nasal mute or before a sibilant. 155. In a few roots [the diaspirates], when a final aspirate ... thus loses its aspiration, the initial sonant consonant ... becomes aspirate. I55a. That is to say, the original initial aspirate of such roots is restored, when its presence does not interfere with the euphonic law . .. which (in Sanskrit as in Greek) forbids a root to both begin and end with an aspirate.

Interpretation:When DaC deaspirates the root-final aspirate before obstruents, GL cannot apply and the initial aspirate remains.
I59. The collision of surd and sonant sounds is avoided in combinations-and, regularly and usually, by assimilating the final to the following initial, or by regressive assimilation. I6o. If, however, a final sonant aspirate of a root is followed by t or th of an ending, the assimilation is in the other direction, or progressive: the combination is made sonant, and the aspiration of the final (lost according to I53), is transferred to the initial of the ending. In this combination, as the sonant aspiration is not lost but transferred, the G6ob. restoration of the initial aspiration (I55) does not take place.

Interpretation:Normally, a preceding obstruent assimilates in voicing to a following obstruent. But when the structural description of BL is met, t and th assimilate in both voicing and aspiration to the preceding voiced aspirate. GL deaspirates the initial aspirate of the root since the root-final aspirate is still present, although not in the original segment, occurring phonetically in the initial segment of the suffix as the result of assimilation. Taking all of Whitney's statements together we see that he does not regard aspiration as a separate segment: he does not permute root-final aspiration to rootinitial position ("the original sonant aspiration of the initial reappears" (i4ia), "the original initial aspirate . .. is restored" (I55a)); he does not permute root-final aspiration around the suffix-initial consonant but accounts for the change by assimilation (i6o, i6ob); and he does not voice t and th as the result of aspirate permutation but rather, again, as the result of progressive assimilation (i6o). Whitney does recognize Grassmann's Law ( I55a) and a class of diaspirate roots (I 4 Ia, I55, i 6ob), and does have general rules for DaC (I 4I, I53). Each of these points is at odds with Anderson's interpretation of Whitney, but each is consistent with Whitney's description translated into more modern form. Our one point of agreement with Anderson's interpretation is that Whitney was not concerned with describing these phenomena in terms of linearly ordered rules. Anderson himself rejects linearly ordered rules in favor of locally ordered rules.

4. A Critique of Local Ordering


Anderson's theory of local ordering incorporates an extended version of Kiparsky's definition of markedness of rule orders. Kiparsky considered the unmarked order of

394

REMARKS

AND

REPLIES

two rules Ri and R, to be the order that maximizes the applicability of Ri and R,. If one rule can increase the number of possible strings to which the other rule might otherwise apply, a potential feeding relationship exists between them. A feeding order is therefore an unmarked order; a counterfeeding order is marked. If one rule can decrease the number of possible strings to which the other rule might otherwise apply, a potential bleeding relationship exists between them. A bleeding order is therefore a marked order; a counterbleeding order is unmarked. Anderson adds to this concept of markedness the stipulation that the effect of a pair of rules on a given string determines the unmarked order of the rules for that string. The rules are in this sense locally ordered, since the unmarked order for one string may differ from the unmarked order for another string. Furthermore, since a given formative can occur in different strings, it is not necessarily the case that the same order of the rules will bear the same markedness value for every occurrence of that formative or that opposite orders of the rules will bear a different markedness value. Grammars which allow local ordering may violate the constraints of transitivity (if A precedes B and B precedes C, then A precedes C) and asymmetry (if A precedes B, B never precedes A) to which linearly ordered rules are subject. If a form is required to undergo two rules in the marked order, or if there is no marked or unmarked relationship between two rules, then the order must be explicitly stated in the grammar of the language for which it is required. On the other hand, for a given string, rules apply pairwise in their unmarked order, whatever that order is for the string, provided that the explicit ordering requirements are not violated. All unmarked applications of the rules not violating the language specific orders are "predicted" by the theory of local ordering, and thus are cost free, while all explicit ordering requirements are considered to be less natural, and therefore there is a cost assigned to them. Throughout his discussion of local ordering as it applies to Sanskrit, Anderson assumes that the rules in (i) are the correct rules. We have proposed alternatives which do not entail the apparent ordering paradox which results from that assumption. Each of our alternatives exhibits different markedness relationships among the rules; the form of the rules therefore plays an important role in determining such relationships. In the following discussion of local ordering in its application to Sanskrit, reference is to the rules as presented in (i). Since BL and GL have no markedness relationship with respect to each other, Anderson requires that an order be specified in the form of an explicit ordering statement, "GL precedes BL". For forms such as /bhudh + ta/ > [buddha], it is asserted that GL-BL, together with the unmarked (counterbleeding) order BL-DaC, enforces the order GL-BL-DaC, "the only order which is consistent" with them. But Anderson does not explain why this is the only possible order. In view of the fact that local ordering is not subject to the constraints of transitivity and asymmetry, and that it orders only two rules at a time (that is, it provides pairwise ordering), an

REMARKS

AND

REPLIES

explicit, automatic procedure must be provided by which two ordered pairs, such as GL-BL and BL-DaC, can induce an ordered triplet, GL-BL-DaC. In his thesis, to which he refers the reader, Anderson outlines a procedure that purports to order more than two rules.
(I2)

The following, then, is approximately the principle of rule application which is assumed here: given an underlying form in the language, which contains the rules R1, R2, . . . ,R, together with a collection of ordering relations of the form "R, must precede Rj" (or, equivalently, "Ri must not follow R1"), consider all possible sequences of the rules. Discard any such possible sequence which contains two rules whose order contradicts one of the restrictions. Of the remaining possible orderings, discard any which contains two rules in a marked order unless this order is explicitly provided for in one of the restrictions. The result of this procedure may not, in general, be a single ordering, but it is necessary to impose, as a general global condition of adequacy for grammars, the requirement that all such possible orders remaining after the eliminations described above yield the same result. (Anderson I969, I2)

"All possible sequences" must be understood to include sequences that may contain more than one occurrence of one or more rules. This follows from the fact that transitivity and asymmetry are not constraints on locally ordered rules (see, for In the present example, the discussion of Menomini in Anderson (I969, I o8- II2)). Now, in case this would include, for example, the sequence BL-DaC-GL-BL. what way could it be said of a sequence that it "contains two rules whose order contradicts one of the restrictions" or "contains two rules in a marked order"? Since all ordering relationships are pairwise, it can only be a pairwise order that could violate an ordering condition. However, in the absence of the constraints of transitivity and asymmetry, a sequence such as BL-DaC-GL-BL cannot be said to contradict the requirement that GL precede BL, because BL does not precede GL; rather, it precedes DaC and it follows GL. For the same reasons, the sequence does not contradict the unmarked order BL-DaC (nor, as we shall see, the "unmarked" order of DaC and GL). This sequence thus fulfills the two ordering conditions GL-BL and BL-DaC, but it incorrectly derives * [bhuddha] from /bhudh + ta/. Since there are at least two possible orders of the rules that are not eliminated by the instructions in GL-BL-DaC and BL-DaC-GL-BL, but that do not yield the same result, the (I 2), "global condition" of (I2) is not satisfied; and, presumably, a grammar that contains both is not an adequate grammar. On the other hand, if transitivity and asymmetry were constraints on local ordering, then we could say of this sequence that since BL precedes DaC and DaC precedes GL, then, transitively, BL precedes GL, which contradicts the requirement "GL precedes BL", and this sequence could be eliminated. But, we repeat, local ordering is not subject to these constraints, and we find no answer

396

REMARKS

AND

REPLIES

to our question of how GL-BL and BL-DaC result in the unique ordering GL-BLDaC. We call attention to the fact that, at this point, Anderson does not invoke a marked or unmarked order for DaC and GL to account for the order of the rules

for /bhudh + ta/. We shall see that this omission is not consistent with a subsequent argument, and that there are furtherdifficulties.For the moment, we simply emphasize the fact that Anderson attempts to induce the correct order of the rules for /bhudh +
ta/ by means of an explicit ordering statement for GL and BL, between which there is no markedness relationship. The fact is that either order of GL and BL would give the correct derivation for /bhudh + ta/.
(I3)

GL BL DaC

a. bhudh + ta budh + ta budh + dha bud + dha

BL GL DaC

b. bhudh + ta bhudh + dha budh + dha bud + dha

The most surprising part of Anderson's proposal for Sanskrit is his argument that DaC-GL is the unmarked order for these two rules. One would expect the reverse order, GL-DaC, to be unmarked since this is a counterbleeding order. But Anderson extends the notion of markedness of rule order to include the loss of a rule. He reasons as follows:
(I4)

Recall ... the original basis of the notion "unmarked order": this was "the order that maximizes the applicability of the rules". Though it appears that "GL-DaC" is such an order, consider the consequences of applying these rules in this way: the initial aspirate of every occurrence of a diaspirate root would be deaspirated; as a result, the class of diaspirate roots would cease to exist. That is, these roots would all be reanalyzed as ordinary roots with unaspirated initials and aspirated finals. Once this occurred, there would no longer be any reason for a rule like GL to be in the grammar, since its only remaining cases would be treated as a special fact about reduplication. Thus, while it appears that the order GL-DaC is one leading to wider applicability for GL, in fact such an order would lead to the complete elimination of GL from the grammar ... We are led to conclude, therefore, that the truly unmarked order of GL and DaC is DaC-GL ... In a form like /bhudh + ta/, however, the explicit ordering relationship "GL precedes BL" enforces another relative order on GL and DaC . . . It is only the fact that the normal case ... has the order which preserves the existence of the rule GL that permits other cases to exist with the other order of these two rules. (Anderson I970, 395)

Anderson does not show explicitly how the grammar mechanically determines the markedness values of pairs of rules as they relate to rule loss. It is clear from (I4),

REMARKS

AND

REPLIES

however, that the order GL-DaC is ruled out as a possible unmarked order on the basis of considering all possible derivations of diaspirates in which the rules apply in this order, for only by determining that the initial aspirate of all diaspirates is lost due to GL-DaC will the theory be able to predict that GL would be lost. The formal mechanism for deciding this is not mentioned in (I2), because the procedure outlined there provides for the determination of rule orders on the basis of individual forms. Thus, the argument for the unmarkedness of DaC-GL adds another dimension to the theory of local ordering. The unmarked order of two rules is no longer determined solely by the effect of the rules on a given form. In order to determine that a given order for two rules is an unmarked order for a particular form, it is now necessary to investigate the consequences that that order has for the loss or retention of those two rules. If we interpret (I4) correctly, this would involve knowing the order of application of the two rules for every form. Only after an order has been hypothesized for all forms can it be determined that the originally hypothesized unmarked order for a particular form will result in loss of a rule. If it does, then the hypothesized unmarked order is in fact a marked order. Given a single underlying form, presumably one of the many possible derivations for this form is chosen by the procedure in (i 2). If we take (I2) literally, then in a language that contains n rules, there are at least n! possible rulesequencesfor every form. Forjust ten rules, n! is I*2* 3 4 5 6 7*8 9 IO = 3,628,800 possible sequences, each of which must be evaluated for each form in the language! But given the further elaboration of local ordering that is provided for in (14), each of the 3,628,800 possible sequences must now be compared with all the 3,628,800 possible sequences for every other form in the language in order to determine that rule loss will not result. It is not clear what this process implies for the adult speaker if a form or rule is added to the language. Would a rule have to be incorporated into each of the sets of ordered rules by the original process? If not, how would it be dealt with? We see that local ordering involves much more than the simple statement "A and B apply in the unmarked order, whichever that is", modified by language specific ordering requirements. The "predictions" provided by cost free unmarked orders are surely made insignificant in the context of the incredibly prodigious ordering procedures required by the full theory of local ordering. The standard theory of linear ordering claims that a child learning a language must discover only one set of ordered rules, constrained by transitivity and asymmetry, that will hold for all forms in the language. Under that theory, a grammar is regarded as having predictive value in the sense that the rules and their order are established on the basis of a relatively small sample of the language to account not only for the already observed forms but for those as yet unobserved as well. If the preferable theory is the one that more narrowly restricts the number of grammars that must be considered as possible candidates, while at the same time providing grammars that have predictive value, the theory of local ordering fails on both counts when compared with the theory of linear ordering.

398

REMARKS

AND

REPLIES

Since Kiparsky's theory of marked/unmarked orders of rules is intended to provide insights into the direction of historical change, it is most appropriate for us to consider Anderson's innovation in that connection. Kiparsky suggests that when languages change they do so by reordering or simplifying rules so that the applicability of the rules increases. If Anderson's proposal is applied to historical change, the claim it makes is that loss of a rule is not one of the natural, expected kinds of changethat generalization through rule reordering is not a simplification of a grammar unless the rules are retained. What does it imply for rule loss that comes about not through reordering but through simplification of the structural description of a rule ? Is this also to be considered an unnatural kind of change ? If we accept Anderson's argument in (I 4), we would have to regard the loss of a rule as reduction of applicability to zero, and any order that prevented the loss of a rule would be unmarked. We would then have a hierarchy of markedness in language change, such that loss of a rule would be the most highly marked (least natural) kind of change even if it resulted from rule reordering that was in an unmarked direction. Since this claim has important consequences for the theory of historical change and the nature of language, we believe that a substantial amount of evidence must be adduced before we can regard it as more than conjecture. After establishing DaC-GL as an unmarked order, Anderson advances a final argument for GL-BL as the required language specific ordering statement.
(I

5)

Notice that while the ordering statement "GL precedes BL" seems arbitrary, it is in fact dictated by the facts of the language. That is, assume we chose instead "BL precedes GL". Now BL-DaC has been seen to be unmarked; hence both DaC and GL follow BL. The unmarked order of DaC and GL has just been seen to be DaC-GL; hence the only possible order of the three rules would be BL-DaC-GL. The result of applying the rules in this order, however, would be the incorrect *bhuddha. The facts, therefore, require that the order of BL and GL be GL-BL in order to enforce the correct order on GL and DaC in forms to which BL applies. (Anderson I970, 395)

In some unexplained way, the alternative explicit ordering statement, BL-GL, together with the unmarked orders BL-DaC and DaC-GL, is presumed to induce the unique order BL-DaC-GL. Since BL-DaC-GL must be interpreted as implying BL-GL (and how can this be true other than by transitivity?) if it is to satisfy the explicit ordering requirement, then GL-BL-DaC must be interpreted as implying GL-DaC, which contradicts the unmarked order DaC-GL. Recall that Anderson previously argued that GL-BL and BL-DaC induced the order GL-BL-DaC, but did not mention DaC-GL in that connection; apparently DaC-GL was not significant but BL-DaC was. Once again, he ignores the relationship between DaC-GL and GL-BL, while invoking the relationship between DaC-GL and BL-GL, in an

REMARKS

AND

REPLIES

attempt to prove that BL-GL cannot be the explicit ordering requirement. Note that in (I4) Anderson states that the explicit ordering requirement GL-BL enforces the order GL-DaC despite the fact that DaC-GL is the unmarked order. The instructions in (I 2) state that an acceptable sequence of rules cannot contain "two rules in a marked order unless this order is explicitly provided for in one of the restrictions". If GL-BL, which makes no mention of the ordering relation between GL and DaC, can nevertheless enforce the marked order GL-DaC, it must also be able to enforce the marked order DaC-BL. The sequence DaC-GL-BL is then as acceptable as GL-BL-DaC, since it satisfies GL-BL and DaC-GL, although it contradicts the unmarked order BL-DaC and incorrectly derives *[bhudta] (eventually *[bhutta]) from /bhudh + ta/. How does local ordering choose between this sequence and the one that contradicts DaC-GL, namely GL-BL-DaC? Whatever the explicit ordering requirement is for GL and BL, it must take precedence over the unmarked orders BL-DaC and DaC-GL, if local ordering is to be consistent. Consequently, if BL-GL is the ordering requirement, BL-DaC-GL is not the only possible order, for DaCBL-GL and BL-GL-DaC are also possible. The latter sequence satisfies BL-GL but, as is the case with GL-BL-DaC, it contradicts the unmarked order DaC-GL and correctly derives [buddha] from /bhudh + ta/, as we saw in (I3). On the other hand, DaC-BL-GL satisfies BL-GL but contradicts the unmarked order BL-DaC and gives the wrong derivation for /bhudh + ta/. Again, how does local ordering choose between these two sequences? In sum, we find that if BL-GL is the explicit ordering requirement, local ordering permits at least two sequences of rules which yield different results, and the same is true if GL-BL is the ordering requirement. Since BL-GL and GL-BL are both compatible with sequences of rules that do not yield the same results but which are not eliminated by the ordering procedures of (I 2), then, no matter which ordering requirement is chosen for BL and GL, there is no set of rules that satisfies the "global condition of adequacy for grammars [cited

in (I2)],

. . .

that all such possible orders remaining after the eliminations described

above yield the same result". As formulated by Anderson, BL and GL do not impinge upon each other at all, and it is therefore not surprising to discover that there is no order that can be imposed on these two rules that will account for the fact that GL and DaC seem to apply in the order GL-DaC if a form meets the structural description of BL, but in the reverse order otherwise, which was the original problem to be resolved by local ordering.

5. Conclusion We believe that we have shown that the inclusion of rule loss in the scale of rule order markedness has questionable implications for language acquisition and language change, and that it dramatically alters the theory of local ordering in such a way as to render the theory virtually unconstrained. We also believe that we have demon-

400

REMARKS

AND

REPLIES

strated that local ordering, contrary to the arguments advanced in its support, does not provide an automatic procedure for selection of a correct order of the rules for each form from among the many orderingsit makes available. In contrast to this, the more narrowly constrained and therefore preferable theory of linear ordering does succeed in accounting for the facts of Sanskrit under discussion, as shown by the linearly ordered rules we have proposed. One referee of our article has taken us to task for failing to criticize linear ordering for admitting more than one possible solution for the Sanskrit data. This criticism misses the point. Our article is directed against a much less constrained theory of phonology. It is well known that there are many problems with the present theory of linear ordering, but at present we have no better theory. The choice between the two alternativeswe have presented can only be decided by a detailed investigation of Sanskrit, and only with such detailed studies available can a reasonable evaluation criterion be proposed within linear ordering that will further constrain the theory. This, it seems to us, is the direction that current researchwithin generative phonology should be taking.
References VowelSystems and the Ordering Anderson, S. R. (I969) WestScandinavian of Phonological Rules, unpublished Doctoral dissertation, MIT, Cambridge, Mass. (reproduced by Indiana University Linguistics Club, Indiana University, Bloomington, Ind., 197I). Anderson, S. R. (1970) "On Grassmann's Law in Sanskrit," Linguistic Inquiry I, 387-396. Kiparsky, P. (I968) "Linguistic Universals and Linguistic Change," in E. Bach and R. Harms, eds., Universals in LinguisticTheory, Holt, Rinehart, and Winston, New York. Whitney, W. D. (I889) Sanskrit Grammar(republished by Harvard University Press, Cambridge, Mass. I960). Phelps Department of Linguistics University of Washington Seattle, Washington 98I95 Brame Department of Linguistics University of Washington Seattle, Washington 98I95

You might also like