GREGORY K.

IVERSON

ole of phonological rules is to express generazations about phonolog..< .

rhe :ructure. Rules are also ordinarily taken to represent specific derivational .
::uctions, or steps. along the path relating deep to s~rf~ce representa~ons. ,_ .
In this capacity, cert~ of the rules may affect the a~phcation of others, etther ~...
. creasing or decreasmg the number of forros to which the others may apply. :: .
1l1Under the assumptions of conventional generative phonology as laid out in . .
Chomsky and Halle (1968), the individual rules are all applied seq~entially, :":.
arrayed in an extrinsically ordered, or language-specifically determined lmear:
Iist. With respect t a rule A ordered before another rule B in such a list, A.will ~
bear one of three possible relations to B (terms adapted from Kiparsky 196$):
,
1 A Jeeds B just in case the applicatin of A increases the number o(.
forms to which B applies. . . .. .
2 A bleeds B just in case the application of A decreases the number pf;.
forms to which B applies. : ..!
3 A does not affect B just in case A neither feeds nor bleeds B.
. .
For example, contraction of the copula in English (e.g., Bill is here ~ Bill's
fzJ here) creates a syllabically stranded fricative, 1z/, which cliticizes onto the .
preceding noun; if this noun happens to end in a voiceless segment, then the .
lz/ further devoices to 1s/ because of the application of progressive devoicing
among obstruents clustered in. the syllable coda Uack is here ~ ]ack's [s] here).
~ntract!o~ thus feeds devoicing in derivations where both rules are. applicable
be use 1.t lS only through operation of the former that the 1z 1 of zs come~ to
syllabtcally affiliated with a voiceless segment in the preceding word,, t~e.,
:traction precedes devoicing. Applied in the reverse sequence, devoicing
ore contraction the ordering relation would be one in which contraction :~

:ntereeds devoidng (assuming that once a rule is sc~ed for application it

--~.

. ~- . ._'1

d~no~ be applied .again in the same derivatjon); in ~s .c-a_se, the _structural .

Ption of devotcing would not be met at the pomt 1t ts constdered for

f
L_
 . . b
appllcaton ecaus
e the vowe1 1n
the 1z 1 in this word wou
---
. is would not yet . have been contracted 1.e.
ld th n still be postvocahc, not part of an obstru
e
1
.
1

ent
cluster. ~ d. le interacti.on 15 actually a kind of nonaffecting relation
Counter1ee 1ng ru .
1 1 1.ther increases nor decreases the number f
since the counterfeed ng rut e fnde rule applies (contraction will take plao
f. orms t ~ w hch 1
the coun ere . .
ther devoicing precedes or follows 1t m the applicationa}
ce
urespective of whe . . led out from other nonaffecting rule interaction
sequence) Its status 1S Sing h . 1
b counterfeeding represents t e
. . potenha - albeit
typ es h owever, ecause . 1 f
. d f d nuc additive effect: while app Ication o contraction
unreal'1ze - or a yna ' hich

after d evmcrng wou Id not m crease the number of forms to
h. f d w devoicing
applies, the reverse applicational sequence does h~ve t IS <. ~e Ing) result. The
effect of d evmcrng on eontraction, by contrast' is neither additive
. . nor. subtractive
no matter what order the rules are applied in, hence .devoicr~g Slmply does not
affect contraction. Strictly speaking, then, the ordenng relation between any
pair of sequentially applied rules A and B is compound, such that A bears one
relation (possibly nonaffecting) to B, and B bea~s another to ~
The most common analysis of English inflec~ons ~hara~enzes the regular
noun plural affix also as 1z/, phonologically 1dentical. ~1th ~e con~acted
form of is. Progressive devoicing in coda clusters t~en ~Istingu1shes vo1celess
plurals such as in shacks [s] from voiced ones as m ptlls [z], but clusters of
sibilants, as arise in the derivation of plurals like raises [~z](< /rez + z/) and
races [~z ]( < 1res + z'/ ), are instead subject to a rule of epenthesis. Even when
the cluster which epenthesis interrupts is heterogeneous with respect to voicing,
as intermediate 1s + z/ is in the derivation of races, devoicing of the suffix
1z/ does not occur. The explanation from rule ordering for this nonapplication
of devoicing, despite the fact that its structural description is satisfied by
morpheme sequences like 1res + z 1, is that epenthesis is applied prior to
devoicing, and hence bleeds it. The number of forms to which devoicing applies
is accordingly decreased because of the applicational priority of epenthesis,
for were epenthesis not in the grammar to begin with, al1 things being otherwise
equal, devoicing would be able to apply even in sibilant clusters. Devoicing
could also apply in these environments, however, if the order of application
of the rules were reversed (to devoicing before epenthesis), or indeed if the
rules were applied simultaneously. Under either of these possibilities (/res+ z/
~ lfl~~s]), the o~de~g relation. would be one in which epenthesis counterbleeds
dev01~g, resu~tin? m the s.pecral kind of nonaffecting interaction in which a
rule fails to real1ze 1ts potential to reduce the number of forms to which another
rule applies. Irrespective of whether epenthesis bleeds or counterbleeds
devo~~g, .though, ther~ is no ~real or potentiai)effect on epenthesis, because
de~o1cmg 1s c~pable ne1ther of mcreasing nor decteasing the number of fonns
which are subJect to epenthesis. .
An~ther ~d of technically nonaffecting interaction between rules arises irt
cases In which the sets of inputs remain constant irrespective of applicational
sequence, but where ordering nonetheless makes a difference in the phonetic
 ' F t exampte, u u lt:a~ ~x1st ooth a rul
resaltS rdo and a rule to apocopate word-finealto stress the penultimate vowel
wo
of a neither wou
Id b e capable of feeding or blvowels
d. m . trisyllabic
. or longer
word~,ch they are applied is still crucial For a ee mg the other, but the order
whi represent ti
11 eh satisfies the structural req~irements of both of a on Ike /pine +tal'
whi 1 precedence of stress asstgnment ov these rules, the appli-
tiona er apocop Id
ca 1i d 1"\ the reverse . sequence, apocope befo e yte s [pint]; but
P e .u. th re stress ass th
ap d ce [pnet]. Stnce bo rules do apply in eith f Ignment, e rules
pro use that the application of one is facilitated (f e~)o t~e sequences, it is not
thfethc: other. Following I<iparsky (1971), the fo~ [ o,r tl]o~kethid~led) by th~t
o t th th . pme m s example 1s
'd to be transparen Wl respect to . e rules mvolved .t d . .
sal .. f h li . m 1 s envation, because
condttlons or ~ eu app cation are neither violated (th . no word-final .
the . . . ere 1s
'e vowel present to contravene fi . .apocope) nor concealed (the vowe1 whieh 1s .
11
stressed appe.ars d . super cra. hy m . penultimate position). The f orm . . [pmet
. , ], on
lt
y the other han , ts opaque wtt respect to stress assignment since that rule calls
e for stress to fall on the penult, not the ultima.
Feeding and bleeding (as well as nonaffecting) rule . interaction result in
r transpa~ency, as just .illu~trated, but counterfeeding and counterbleeding
1 interacton cause opactty m surface forms. Early work in rule applicational
relationships (Kiparsky 1968; Koutsoudas, Sanders, and Noll1974; Anderson
1974) suggested that the natural mode of rule interaction, presumably the
one easiest learned ~d to which phonological change would gravitate, was
that which resulted in maximal application of the rules, i.e., feeding and
counterbleeding interaction. Under the terms of local ordering (Anderson 1974),
infact, the same pair of rules might interact in one way in sorne derivations,
but in the reverse in others, particularly if the variation coincides with
differences attributable to presumed naturalness in the ordering relation. For
example, in its inflectional n1orphology, Modem lcelandic gives evidence of
feeding interaction between rules of syncope and u-umlaut,, the latter of which
modifies the vowel a to o before a u in the next syllable: katil + um (syncope)
~ katl + um (u-umlaut) ~ kiitlum "kettle" 4at. pl.; cf. katli ~at. sg., ~till. no~.
sg. But it appears the same rules interact in a counter~leed~g fashio?-, 1 ~, m
the reverse applicational sequence, when that is posstbl~ m the denvati~n~
~orphology: bagg + ul + i (u-umlaut) ~ biigg + ul + t (syncope) -7 bogglt
package" dat . s cf baggi "pack". 19
Other work .(K~~st~wicz and Kisseberth 1971; Kipars~ 1971: Ive;son ~;
Goldsmith 1991) leads to the conclusion that natural mteraction . ~vors nle
fee<t . cy with rules applymg o Y
. .mg and bleeding .relations of transparen ' . hich satisfy their
:mally
ofcttu:ai
(but. ~rsistently) .tO re~ove rep~~;e~:o:d':elf-bleeding mo~e
~escnptions. The predonunentl! se ee ie of applying to thetr
o application among iterative rules, ~hich ar: cae:rth 1977), wquld seem to
' : ' ou~ts ~Howard 1972; K~~t?WICZ and Ki~~the rinciples ~hich ?overn
thPl>ort this Vlew~ particularly tf It ts assumed th . hp . ~nteraction With one
.

e appl' . al . t play m t etr.l.U th . f

ano .tcation of individual rules are so a e led to the hypo esis 0
ther. In the 1970s, considerations such a~ thes
 OlL vregory r-... Jvc;,~v ----

. 1 lication1 i.e.1 to the idea that the ordering relati

universally detemm;:~ :~ications) is predictable from other aspects of ~n
between rules (an . 'pies with the universal force of the "Elsewh e
grammar and

1
(IG
from pnna d
k 1973) "Proper Inclusion Prece ence
, (S
anders 1974)
ere
Cond1tion" pars Y 1
d f th d ' or
"S
the urv1v . al C nstramt" (Anderson 1974). By the en o e ecadel however
o . chn f h '
attention had tumed more toward the mcreasmg ~ . ess ~ P .onologicaJ.
representa ti.on than to the specifics of rules and theu Interaction; . m . fact, the
entire class of stress rules had been supplanted by the configurations and
parameters of metrical theoryl and the ~se of autosegmental .and metrical
representation cast an entirely different hght onto the conv~ntional rules of
linear phonology (d. Goldsmith 1990 for summary elab~ration and chapters
5-8 of this volume ).
Today, sorne researchers assume that the ques~on of ~le order in phonology
has essentially been answered, and that rule. In~eractio~ ~e govemed by
general, though obviously still developing pnncrples (wtthin the context of
radical underspecification theory, d. especially Archangeli 1988, p. 185). Others
maintain that at least sorne language-specific rule ordering. stipulations may
be necessary. Bromberger and Halle (1989), in particular, consider that
phonology differs from the rest of grammar primarily in its requirement for
extrinsic ordering (Chomsky 1967; but d. Klausenburger 1990), a point which
they consider to be demonstrated by instances of language change or dialect
variation where differences in rule order alone might be at play. Certainly if
rule applicational interactions are specific to individual grammars, then it should .
be expected that rule order will vary from stage to stage of a language, or from
dialect to dialect, presumably in about the same magnitude as do other linguistic
constructs, including the lexicon and the phoneme inventory. Sueh variation
is in fact very difficult to establish, however; in the cases that have been
suggested, altemative explanations of equal or greater generality, with order
determined by principie, are easily available, and are often also necessary
(Iverson 1974; Koutsoudas, Sanders, and Noll 1974). The familiar case of
Canadian vowel rai~ing, for example, which Bromberger and Halle review in
this connection, has often been cited as justification for extrinsic rule ordering
but many alternatives have been offered, including oneby Kiparsky and Menrt
(1977) involving different phonemicization, and it is not clear to begin with,
following Chambers (1973), that exactly the same rules are.involved in both of
the dialects.
As first reported by Joos (1942), Canadian raising affects the diphthongs
1ay 1 and. 1aw (, centralizing them to [ay] and [~w] befare voiceless conso-
nants, as m .wnte [r~y~] versus ride [rayd]. Voicing (and tapping) of /ti ~so
takes place mtervocabcally, so that 1t/ merges with di in writer and rtder
~ut wh~er these w~~ds are homophonous .depends on the applicational
mteraction of the . v.o1ong and raising. rules. lf raising takes place before
(~ounterbl~ds) vo1~g, ~en, as characterizes one variety of Canadian Eng-
~sh,. the dtphthong m. umter [r~yd~r] will be centralized in comparison to th~t
m nder [rayd~]. But if voicing is applied before raising, and so bleeds it by
 --
elinnating the voicelessness which . .
ratsmg
Rule Ordering 613

be pronounced the s~me, i.e., both as ra r~quue~, then the two words will
dation of the other dtalect. As Kip k [ yd~ ], which represents the pronun-
. . ars y and Me (197
ever, this vanation can also be accounted for . nn 7, p. 48) observe, how-
/ ~y 1 as sep_arate phonemes, with in th Wtth the_assumption of 1ay 1 and
(which the first one has lost) 1ow . e second d1alect an additional rule
Under this interpretation, of cour~~ ~y 1 to [ay] befare voiced segments.
1
ation between the dialects is due e rules are not the same, and the vari-
themselves rather than to stipul ti a com~onplace difference in the rules
0

The model of lexical phonolo a on~ on theu mode of interaction.

how what appear to be the gy (Kiparsky 1985) offers another account of
different condttlons. 0n this view rules may
. . same . d"ff
. app1Y In d
1 erent or ers under
"deeper", less producti h of t~e lextcon a~d its relation to phonology,
and irregular inflectionv~~~rp o~ogcal ~perations involved in derivation
"'shallo , te Wtth certam phonological rules at Level 1,
while the
. wer, more general aspects of word formation
typtcal
of regular
inflection are govemed by another e1ass of rules at Level 2. Sorne phonologcal
rules apply throughout
. . the lextcon,
h owever, essenbally
whenever theu
s~ctural descrtptions are met. The lcelandic rules of syncope and u-urnlaut
dtscussed ab?ve would appear to be of this type, for they interact in the
counterbleedtng _sequence of u-umlaut before syncope in the derivational
morphology, which produces boggli from /bagg + ul + i/, but in an order
reverse of _this in the inflectional morphology, a feeding which produces kOtlum
from /katil + um/. Construed as cyclic, lexical rules available throughout the
derivation, syncope and u-umlaut will apply until their structural descriptions
are no longer satisfied, which yields the observed counterbleeding interaction
in sorne derivations, feeding in others. .
Even if the same phonemic inventory is assumed for the two Canadian
dialects under discussion, similarly, lexical phonology provides a natural basis
for distinguishing the "'bleeding" dialect (writer = [r~yd~r]) from the
"counterbleeding" one (writer = [rayd~j). As Chambers (1973) points out, there
are exceptions in the former to the raising rule (e.g., cyclops with [ay], not [~y]),
which would imply it is lexically restricted, whereas voicing has to be a
postlexical rule in order to be able to apply in multiword phrases .like fight it.
Since it is a general property of the theory that le~ical ~les precede postlex~cal
rules, the bleeding of voicing by raising is predtcted; m th~ counterbleed~ng
dialect, by contrast, which Chambe~s ma!t'tains no longe_r ~x1sts and for which
there is no evidence of exceptionahty wtth respect to r~tsmg, b~th rules were
presumably postlexical, free to apply, in counterbleedtng fa~hion, whenever
their structural descriptions were met.
The idea that at least sorne rules are persistent, applying_ t~oughoutfithe
tions are sabsfied, was rst
grammar whenever their structura1 d escnp . . h al f
d h f d new vitahty m t e -propos s o
suggested by Chafe (1968),_ an . as oun s on the basis of their form alone,
Myers (199l),who seeks to tdentify_s~ch ~~on. In other recent work, particu-
Wtthout reference to language-spectfic fun ti al ordering of rules confooods
larly Hvtnan ( .. 93 it is ar ed that conven on .. . . . .. -
 ~6~14~~G:r~eg~o~~~K~:Iv=e~~~o~n---------------------------------------
rather than facilitates the description of certain tonological phenomena i
that the rules involved cannot be ordered in any s:quence without signfic~~ "
':
Ioss of generalization. In the geometric representations of cu?'e~t phonological
description, (see chap; 7, this volume), feature.and node delinkin~ operations,
or neutralizations, are found always to feed mto feature spreadmg rules 0
assimilations, not to counterfeed them. In the th. e~ry of"h a~oruc phonology"
'r
(Goldsmith 1991), similarly, or in the connectiomst modeling of Wheeler and
Touretzky (1991), rule interactions are a1l determined by the nature of the
grammar itself, too: these approaches exploit the "natural" sequencing pro-
vided by, on the one hand, the precedence of rules relating morphological
representations to the word level, and the word level to phonetic representa-
tions (cf. also Anderson 1975), and, on the other hand, the predicted pre-
cedence of rules which determine syllabic and other prosodic structure over
those which impose traditionally segmenta! changes (cf. also Kenstowicz and
Kis~bertl_t 1971 ). Details vary from model to model still, but the machinery . -~
\~;.
',
/,
available m present phonological theory appears to provide sufficient descrip-
tive capacity without the stipulation of special constraints on rule ordering.

. ~ .

--~ .
'i
-~ .
' .
>

:; .

' .