Professional Documents
Culture Documents
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.
Linguistic Society of America is collaborating with JSTOR to digitize, preserve and extend access to Language.
http://www.jstor.org
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
TWO THEORIES OF NON-AUTOMATIC MORPHOPHONOLOGICAL
ALTERNATIONS: EVIDENCE FROM SPANISH
JAMESW. HARRIS
Massachusetts Institute of Technology
Evidence from vowel alternations in Spanish is presented which bears on two
theories of regular but non-automatic morphophonological alternations. On the
'disjunction' theory, alternating substrings of lexical entries are represented as
disjunctions of surface alternants. The appropriate alternants are selected by rules
triggered by the lexical disjunctions. On the 'rule-feature' theory, rule features are
attachedlexically to segmentsthat have surfacealternants.These featurestriggerrules
that apply only to forms bearing the correspondingrule feature. Though similar, the
two theories are empiricallydistinguishable.A particularanalysis of the Spanishdata
provided by the rule-featuretheory is descriptivelysuperior to one provided by the
rival theory. The inadequaciesof the disjunction analysis are direct consequences of
properties of that descriptive apparatus. The Spanish evidence thus supports the
rule-featuretheory.*
INTRODUCTION
1. What devices should linguistic theory make available for the description of
rule-governed but non-automatic morphophonological alternations? One mechan-
ism often employed in generative phonological practice is the 'diacritic feature/
minor rule' complex. In descriptions using this device, features without direct
phonetic interpretation are associated with segments that have surface alternants.
These features trigger rules that apply only to forms bearing the corresponding rule
feature (Chomsky & Halle 1968:172-6 et passim; Harris 1977c). We can call this
device the 'rule-feature' theory.1 However, Hooper 1976 has recently argued in
favor of another mechanism, which we might call the 'disjunction' theory. With this
mechanism, alternating substrings of lexical entries are represented as disjunctions
of surface alternants. The appropriate alternants are selected by rules triggered by
corresponding lexical disjunctions. As we shall see, the rule-feature and the dis-
junction theories, though similar, are empirically distinguishable.
Hooper claims that the disjunction theory is superior to the rule-feature theory
(1976:165-9).2 The substantive evidence for this claim comes from an analysis of
two sets of vocalic alternations in Spanish, 'diphthongization' and what I shall call
the 'high-mid alternation'. The purpose of the present paper is to argue that this
* A version of this paper was read at the Winter meeting of the Linguistic Society of
America, 1976, under the title 'How abstract is natural generative phonology?' I thank all
those who gave me helpful comments on the firstdraft. As usual, only the author is to be blamed
for deficiencies.Special thanks go to Joan Hooper for giving me access to the manuscriptfor
Chapter 8 of her book An introduction to natural generative phonology (1976) in advance of
publication, and to Rogelio Reyes for discussion of the Chicano data in ?4.5.
1'Rule features' are only one kind of 'diacritic'. Although the distinction is important, it
happens not to play a crucial role in the present study. Ringen & Iverson 1976 give interesting
relevantdiscussion.
2 Hereafter, all page numbers refer to Hooper 1976 unless otherwise indicated.
41
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
42 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
same empirical material, when examined more carefully than Hooper has done,
supports instead the rule-feature theory.
In subsequent sections, I first sketch briefly the basic data, and give an alter-
native analysis to Hooper's. Her analysis is summarized, and the two competing
analyses are compared.3 On the basis of general criteria of descriptive adequacy
which Hooper herself employs, I claim that my analysis is superior to hers.
Inadequacies in Hooper's account are directly traceable to properties of the
descriptive apparatus provided by the disjunction theory. The Spanish evidence
under consideration thus supports the rule-feature theory. I conclude with sug-
gestions as to the type of additional empirical evidence that might permit a choice
among theories of non-automatic morphophonological alternations.
This work is focused narrowly on issues raised in Chapter 8 of Hooper 1976. It
fits, however, into the larger context of evaluating the substantive achievements of
the style of investigation that has come to be known as 'natural generative phonol-
ogy', for which Hooper 1976 seems destined to become the standard reference
work.4 Insofar as the disjunction theory of alternation is taken as an integral part
of the theory of natural generative phonology, the latter is disconfirmed by its
choice of Hooper's account of the Spanish data under consideration, since this is an
incorrect analysis (if my arguments are valid).5 If, on the other hand, the disjunction
theory of alternation is disassociated from the general theory, then the general
theory evidently cannot choose between the two competing analyses at issue. In
this case, natural generative phonology is no more explanatory than other theories
that permit the same degree of latitude. The present study, then, if essentially
correct, indicates that Hooper's assessment of the empirical support for her
version of natural generative phonological theory is quite inflated, and that it
cannot be accepted uncritically.
3 In a sense, the two competing analyses do not differradically. It does not follow, of course,
that the differencesthat can be found are not worth scrutinizing. Progress at a given point in
the developmentof an empirical discipline often depends on clarificationof technical minutiae.
Morphonophologicaltheory seems to be at just such a point now.
4 See also Vennemann 1971, 1972a, 1972b, 1974; Hooper 1973, 1974, 1975; and other ref-
erences cited in Hooper 1976.
5 It should not be assumed that Hooper's version of natural generativephonological theory
representsa consensus which would be subscribedto in detail by everyone who would wish to
be called a 'natural generative phonologist'. By the same token, 'transformationalgenerative
phonology', which Hooper repeatedlyplaces in opposition to 'natural generative phonology',
certainly does not exist as a monolithic position either.
6
This topic is treated only very sketchily here. For details, see Harris 1977c, which includes
critical evaluation of several recent proposals other than Hooper's.
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 43
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
44 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
of the so-called first and second conjugations (5a-b), but also of the third conjuga-
tion (5c-d), with which the high-mid alternation is exclusively associated:8
(5) a. Ist conj. c[e]rrdmos c[ye]rran 'we they close'
b. 2nd conj. qu[e]remos qu[ye]ren 'we . they want'
c. 3rd conj. reqtu[e]rimos requ[ye]ren 'we they require'
d. 3rd conj. adqu[i]rimos adqu[ye]ren 'we . they acquire'
is confined to stems that are idiosyncratically
2.2. THE HIGH-MIDALTERNATION
(lexically) assigned to the third conjugation (see fn. 8, below).9 Some third-conjuga-
tion stems show no vocalic alternations at all, while others alternate. Excluding a
few defective and otherwise anomalous stems, there are five patterns of allomorphy.
These are illustrated as follows:
(6) Low V (no alternations):
[a]br- 'open', persu[a]d- 'persuade' etc.
(7) High-mid:
a. s[i]rv- - s[e]ru- 'serve' (ca. 30 verbs)
b. diphthongizing h[i]rv- h[e]rv- h[ye]rv- 'boil' (ca. 25 verbs)
(8) Not high-mid:
a. v[i]v- 'live', c[u]br- 'cover' (ca. 200 verbs)
b. diphthongizing inqu[i]r- inqu[ye]r-'inquire' (2 verbs)
There are no third-conjugation verb stems with non-alternating mid vowels;
every stem with a non-low vowel has a high vowel in some form.10We will soon see
that these gaps are systematic rather than accidental.
Let us examine first the 'pure' high-mid alternating group 7a, which contains a
little over half of all alternating stems. (Group 6 is shown only for the sake of
completeness.) Inspection of the roughly 60 inflected forms of each of the verbs in
this group shows that the distribution of high and mid stem vowels follows this
pattern:
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 45
1 (a)
(10)rv0 [+
[IVw]j [- high][HM ]Co hsyll] / [X rconjj[]
_ Yl]
The order and disjunctivity of the two cases of 10 are guaranteed by any and all
ordering principles of the 'elsewhere' and 'proper inclusion' types, so far as I know.
Recall that the alternation illustrated in 9 and formalized in 10 is found only in
third-conjugation verb stems-nowhere else in the language. This fact is reflected
in the outer portion of the environment of rule 10, which is thus a morphological
rather than a strictly phonological rule, in a fairly clear sense.
The feature [HM], mnemonic for 'high-mid', which appears in the inner en-
vironment of the first case of 10, represents the fact that membership in the class
of third-conjugation verbs that participate in the high-mid alternation is arbitrary
and unpredictable. This is illustrated here:
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
46 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
b. Unstressed:
[ - high] / Coi [ + high] / elsewhere
h[e]rvimos,m[o]rimos h[i]rvdmos,m[u]rdmos
h[e]rvird,m[o]rird h[i]rv[y]6, m[u]r[y]6
The generalization illustrated in 12b is obviously the same as that in 9-10. These
verbs undergo both rule 4 and rule 10; i.e. they are lexically specified both [D] and
[HM]. Rule 10 accommodates these additional data without alteration.
The remaining group of third-conjugationverb stems (8a) is more interesting. As
indicated, this group contains the vast bulk of all Spanish third-conjugation verbs,
almost four times as many as all the other groups combined. The stems of this
majority group have only non-alternating high vowels ([i] and [u] appear in
roughly equal numbers). Mirabile dictu, this most striking and pervasive fact about
the Spanish third conjugation is already predicted by rule 10: the second case of
this rule, the general or 'elsewhere' case, guarantees that all third-conjugation verb
stems with no exceptional lexical marks will have only high vowels throughout
their fifty to sixty inflected forms. Let us also observe at this point that rule 10 also
automatically accounts for some additional facts mentioned above, namely that
there are no non-alternating third-conjugation verb stems with mid vowels, and
that every third-conjugation stem with a non-low vowel has a high vowel in some
form. We see now in what sense these gaps are systematic rather than accidental:
precisely the missing forms are automatically excluded by rule 10.
2.3. SUMMARYAND FURTHERDETAILS.The data presented in ??2.1 and 2.2 are
exhaustively described by rule 4, 'diphthongization', and rule 10, 'high-mid'. The
latter is the more complex and interesting case. We began our investigation with the
class of verbs like s[i e]rv- 'serve', which are paradigm instances of the third-
conjugation high-mid alternation. We motivated rule 10 exclusively on the basis
of the vowel alternations in stems of this group. We then saw that rule 10 alone
constitutes the totality of descriptive machinery required to account for all the
systematic vocalic allomorphy peculiar to the third conjugation, including syste-
matic gaps.13
Some linguists would wish to add a certain type of footnote to this otherwise
apparently complete synchronic description. I am referring to 'lexical redundancy
statements' concerning the distribution in lexical items of the diacritics [D] and
[HM].
As indicated in ?2.1, [D] occurs commonly on the mid vowels /e o/, but rarely
on the high vowels /i u/. One might thus propose the following lexical redundancy
statement:
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 47
violate 13. (This is a familiar phenomenon: cf. Eng. svelte, sphere etc., whose
initial clusters violate the segment-sequence constraints of English.)
As indicated in ?2.2, [HM] occurs only on the (last) non-low vowel of a relatively
small number of third-conjugation stems. This suggests another lexical redundancy
statement:
? syll
(14) If [HM], then #X -low C,#
+ 3conj _
Finally, let us observe that [D] and [HM] are independent, as follows:
[D] [HM]
(15) a. 3rd-conjugation subgroup (7b) Yes Yes
b. all other diphthongizing forms (1) Yes No
c. 'pure' high-mid verbs (7a) No Yes
d. everything else No No
HOOPER'SANALYSIS
3.1. LEXICALREPRESENTATIONS. Hooper eschews rule-feature diacritics of the sort
employed in ?2 in favor of 'complex lexical entries' (157) in which alternating
segments are represented as a disjunction of surface alternants. All the relevant
cases are illustrated here:
(16) a. c[o]nt- c[we]nt- 'count', s[e]nt- s[ye]nt-'seat' etc.:
-
The appropriate
3.2. RULES AND REDUNDANCYSTATEMENTS. member of the
disjunctions contained in lexical entries of the sort illustrated in 16 is selected, in
Hooper's analysis, by this rule:'4
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
48 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
([-syll
{ ]0+ syll ]I )y
liback
[ 1
L+highJ[ hihJ
(19) +syll ]
r? CO]verb IS ALWAYS[+3rd conj.]
k-high]
?+syll]
L+highj
This expresses the fact that 'only third-conjugationverbs have the mid/high and the
mid/high/diphthong alternations'.
4.1. PREAMBLE. According to Hooper, both her analysis of the material at hand
and that of ?2 above 'are, in general, consistent with the constraints on N[atural]
G[enerative] P[honology]' (165).17Actually, it is not obvious that either analysis is
16 This is necessary to prevent 17 from applying generally; e.g., to block change of the e
of alfab[e]tico 'alphabetical', which conforms to the second case of 17, to i in alfab[e]to
'alphabet'.
16Hooper's version of 18, her 32, is more general, since she combines the content of 18 with
a similargeneralizationconcerningsecond-conjugationverbs. My alterationof her ruleis innocu-
ous, however, so far as the present discussion is concerned.
17More accurately, Hooper compares her analysis with that of Harris 1974b--rather than
with that of ?2, above. Although I consider the latter to contain certain technical improvements
on the earlier version, the differencesare irrelevantto the issues to be discussed below, so far
as I can see.
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 49
consistent with these constraints, not even Hooper's. For example, both rule 17 and
redundancy statement 18 violate the 'True Generalization Condition', which is one
of the cornerstones of NGP. This condition requires that 'ALLrules express trans-
parent surface generalizations, generalizations that are true for ALLsurface forms'
(13; emphasis supplied).
Both my analysis and Hooper's fail to account for some data. Rule 17 is not
correct with respect to surface forms such as the preterits d[i]jimos, d[i]jiste 'said'
and v[i]nimos, v[i]niste 'came'-though their stems must be subject to 17, as shown
by forms like d[e]cir, v[e]nir (infinitives) and d[i]c[y]endo, v[i]n[y]endo (gerunds).
Rule 18 is not true with respect to third-conjugation stems like ab[o]l-, conc[e]rn-,
and others (see fn. 10, above).
But the same sets of examples are, of course, exceptions to my rule 10.18The two
analyses are thus equivalent in this respect. I will not pursue this matter further,
because the object of the present study is to find evidence that distinguishes the
two analyses and their associated theories; and I assume, without furtherargument,
not that the exceptional forms just mentioned invalidate either analysis, but rather
that they (and many other cases that readily come to mind) suggest that the 'True
Generalization Condition' cannot be incorporated into any interesting morpho-
phonological theory.
Hooper states that both her analysis and mine 'account for the forms of standard
Spanish in practically equivalent ways' (166) and that 'both models account
adequately for the synchronic facts of standard Spanish, but only [the disjunction
theory] provides a natural account of the analogical changes that have occurred in
this system in non-standard dialects' (169). In ?4.5 I investigate the analogical
changes referred to, after examining the claim of synchronic adequacy in ??4.2-
4.4.
GAPS
4.2. ACCIDENTAL VS. INADMISSIBLE It is a fact that participa-
ALLOMORPHY.
tion in diphthongization and the high-mid alternation is unpredictable, a lexical
idiosyncrasy of individual stems. It is also a fact that, given a lexical representation
L which includes the information that L participates in one or both of the alter-
nations under discussion, the actual form of all possible alternants is predictable,
and NOTa lexical idiosyncrasy of L.
The first fact, that alternation is unpredictable, is recognized by both analyses.
This fact is reflected in my analysis by the appearance in lexical representations of
the rule-triggeringdiacritics [D] and [HM], and in Hooper's by the (equally rule-
triggering, diacritic-functioning) disjunctions of symbols in lexical entries. On this
count the two analyses are equivalent.
The second fact, that the shape of alternants is predictable, is reflected in my
analysis in that only the basic alternant (plus rule-triggering diacritics) is given
18 Hooper has pointed out to me (personal communication) that my redundancy
rule 13
also violates the'True Generalization Condition' because of the existence of the four cases
of diphthongizinghigh vowels mentioned just above 13. In fact, rule 19 also seems to violate
that condition. This is because the second-conjugationverb p[we&o u]d-'can, be able' has,
exceptionally, the three-way alternation allowed by 19 only in the third conjugation (e.g.
p[we]do, p[we]das; p[o]demos,p[o]dia; p[u]d[y]endo,p[u]d[y]eron.Thus, apparently, none of
Hooper's rules is consistent with her own theory.
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
50 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
lexically, while the other alternants are rule-generated. Thus the lexical entry
/m[e,D,HM]nt-/ encodes the information (a) that the basic phonological shape is
/ment-/ and (b) that this stem participates in two particular alternations. The
phonetic shapes m[ye]nt-, m[e]nt-, and m[i]nt- are generated in accordance with
rules 4 and 10. No other set of alternants could possibly be derived. In particular,
there is no way to derive a systematically impossible set of alternants such as *m-
[ye o u]nt-.
In Hooper's analysis, on the other hand, a lexical entry such as /m{ye,e,i}nt-/
encodes not only (a) that the stem alternates, but also (b) that the alternants are
precisely m[ye]nt-, m[e]nt-, and m[i]nt-. The latter information, however, is not
lexical information. That is, it is not information peculiar to this one stem, or to any
other individual stem. Thus, Hooper attributes unpredictability (correctly) to the
fact that a given stem alternates or not, but also (incorrectly) to the shape of each
alternant of every alternating stem in the lexicon.
This redundancy in Hooper's system has, as a consequence, a more serious
failing-namely, that the distinction between accidental lexical gaps (e.g. n[ye-
e i]nt-) and systematically inadmissible variation (e.g. *m[ye o u]nt-, *m[we
e i]nt- etc.) is not expressible in her system except by brute force, i.e. by additional
ad-hoc machinery. Nothing in her descriptive apparatus as formulated-rule 17 in
particular-relates to the distinction in question.19
The fact that my analysis automatically draws the correct line between accidental
and systematic lexical gaps-while Hooper's does not, and cannot without un-
motivated additional devices-is strong evidence of the descriptive superiority of
my analysis.
4.3. RULES VS. GENERALIZATIONS. Redundancy rule 19 expresses the fact,
Hooper says, that 'only third-conjugation verbs have the mid/high and the mid/
high/diphthong alternations' (161). The fact is essentially correct (but see fn. 18,
above); however, the generalization is spurious. As illustrated in 5, verbs of all
three conjugational classes (and non-verb forms as well) participate in the diph-
thongization alternation. There is nothing special in this respect about the third
conjugation. Indeed, it would be surprising if the third conjugation failed to
contain some diphthongizing stems. As illustrated in 15, diphthongization and the
high-mid alternation are independent.20The only generalization regarding diph-
thongization that might be appropriately expressed as a lexical redundancy state-
ment is that contained in 13, namely that only mid vowels diphthongize. (There
are no other lexical constraints on diphthongization.) This statement, however, is
found nowhere in Hooper's description; furthermore, it is apparently ruled out in
principle by her theory (cf. fn. 18, above).
Let us now collect the elements of Hooper's descriptive machinery that deal
only with the high-mid alternation, excluding anything that is concerned only with
19 This remains true, as far as I can see, when 17 is stated in maximally general form with
distinctive features (see fn. 14, above).
20
As we shall see in ?4.5, diphthongization and the high-mid alternation also show their
independenceof each other, in that the two phenomenacan change in differentways in different
dialects.
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 51
diphthongization. These are: rule 17 (second and third cases), rule 18, and rule 19
(second and third cases). For convenience, these are copied as follows:
high
-C. + l COlverbIS ALWAYS [+3rd conj.]
[?+highJj
We now compare this with the corresponding apparatus of my analysis in ?2,
namely the morphological rule 10 and the redundancy statement 14. For con-
venience, these are copied here:
I- ____I
[ [* _1_ yll ]
As far as I can tell, 20c and 21b may be taken as entirely equivalent, in their
respective contexts. Let us thus set them aside and concentrate on 20a-b and 21a.
Rules 20a-b are, jointly, the functional equivalent of 21a. There is, however, a
subtle but significant difference between 20a-b and 21a. Recall that the single most
characteristic feature of the Spanish third conjugation is that the overwhelming
bulk of its stems have non-alternating high vowels in verb forms. This is the
empirical content of 20b. Strikingly, Hooper's descriptive apparatus requires that
20b be stated separately from 20a. Thus, in Hooper's system, 20b, which reflectsthe
most pervasive characteristic of the third conjugation, could be true or false in-
dependently of 20a. That is, there is no inherent connection between 20a and 20b,
as they are formulated. In my analysis, on the other hand, the most characteristic
property of third-conjugation verb stems follows as an automatic consequence of
rule 21a, which is motivated by an independent set of facts (as observed in the earlier
discussion of rule 10).
It is important to see how the mechanisms characterizedin ? 1 as the 'disjunction'
theory force Hooper to formulate 20a-b as separate statements. Rule 20a is a
proper subpart of 17, which is in turn a joint statement of diphthongization and the
high-mid alternation. The disjunction theory requires that these two generaliza-
tions be conflated, despite the lack of any independent motivation, simply because
they happen to coincide in subclass 7b of third-conjugation verbs. As a result of this
conflation, 17 applies to forms both within and outside the third conjugation. Since
17 is not restricted to the third conjugation, it cannot be combined with 18 (=20b).
Thus the disjunction theory of non-automatic alternations is responsible for the
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
52 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 53
As in 22, non-verb forms of the class illustrated in 23 do not conform to the pattern
of alternation controlled in verb forms by rule 10, which would require, e.g.,
*s[i]rvomotdr,*conc[i]pc[y]6n,*conc[i]pto, if these forms were verbs. As was in fact
already illustrated in 9, both the stems in 23a, with basic mid vowels, and those in
23b, with basic high vowels, have the same distribution of high and mid vowels in
verb forms-namely, the high-mid alternation described by rule 10. The central
point is this: rule 10 is still seen to be (necessary and) sufficient to account for all
the systematic, predictable vocalic alternations in third-conjugation stems. This
rule was originally formulated to describe the alternation in high-mid alternating
stems without regard to the vocalism of related non-verb forms. The discovery
that the non-verb forms of some stems have mid vowels, while those of others have
high vowels, requires no additional machinery.
Let us turn to Hooper's analysis of this material. Her descriptive apparatus for
vowel alternations in verb stems has been exhaustively presented in 16-19, above.
The analysis embodying this apparatus 'attempts no account of stem vowels in
non-verb forms, nor were non-verb forms taken into consideration in the analysis.
The decision to exclude non-verb forms was made on the basis of the principles
established in Chapter 4 [of Hooper 1976]' (161). Hooper does, however, include a
discussion of non-verb forms, in order to make her analysis comparable in scope to
previous analyses such as Brame & Bordelois 1973 and Harris 1974b.
Hooper notes, correctly, that in many cases the apparatus presented in 16-19
above is sufficient to give an account of the systematic vowel alternations found in
sets of related verb and non-verb forms. These cases include some in which one of
the rules 17-19 functions as a 'via-rule'.21 She acknowledges that there are also
other cases for which her rules do not suffice (e.g. divert-/diversion,mor-/mortdl-a
good many more could be added), although 'the phonological relation is apparent
from the identity of the non-verb stem vowel with one of the verb stem vowels'
(164). We are told that 'these non-verbs will be related to verbs by some via-rules
but not by rule [17]' (164). These additional via-rules are never given, even inform-
ally, and nothing further is said about the matter.
It should be emphasized that Hooper does not deny that the relationships in
question exist, or that they are systematic. She simply does not deal with them.
Because of these omissions, her analysis cannot be compared with others with
respect to observational adequacy, or be considered 'practically equivalent' (166)
with respect to descriptive adequacy.
4.5. HISTORICAL DATA. In certain dialects, the high-mid alternation has been
leveled in favor of the high vowel.22 For example, the standard high-mid alternating
stem p[i\e]d- 'request' has become non-alternating p[i]d-, and the standard
diphthongizing high-mid stem m[ye e i]nt- 'lie' has become m[ye i]nt-. On my
analysis, formulated in terms of the rule-feature theory, loss of this alternation
implies loss of all and only the machinery associated with just this alternation,
21
For discussion of via-rules, see in particular Hooper 1973, 1974, 1976, and Vennemann
1972b.
22 Espinosa 1946, Reyes 1976. In this first pass through the data, I oversimplify in order to
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
54 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
name y ru
namely rules
es 14 and10a and and, obv ous y the tr
obviously, gger ng feature [HM]
triggering [HM]. Noth Nothing ng eelse
se
changes. part
changes In particular,
cu ar rule
ru e lOb
Obrema
remains ns in
n effect, guarantee
effect guaranteeing ng that the only
on y possible
poss b e
resultt of the loss
resu oss of the aalternation
ternat on iss prec
precisely
se y the survival
surv va of high
h gh vowels.
vowe s
In a numberof
number of d a ects-over app ng but not co-extens
dialects-overlapping co-extensive ve w
withth those in n wh
whichch the
gh-m d aalternation
high-mid
h ternat on iss lost-the
ost-the d phthong zat on aalternation
diphthongization ternat on has been partially
part a y
lost.23
ost 23 Here the loss oss iss sporad
sporadic, c affect
affectingng part ex ca items
cu ar lexical
particular tems at randomrandom. In
these cases
cases, the d phthong iss leveled
diphthong eve ed in n favor of the ssimple vowel. For examp
mp e vowe example, e
standardf[we o]rz- 'force'
force has becomef[o]rz- and qu[ye e]r- 'want' want has become
qu[e]r-.24
qu[e]r- 24W Withth respect to these cases cases, cclearly
ear y noth
nothing ng has happened to the d diph-
ph-
zat on process in
thongization
thong n genera
general. A nd v dua lexical
All that changes iss that individual tems lose
ex ca items ose
theirr except
the exceptionalona d diphthongization-marking
phthong zat on-mark ng property property, whether th thiss iss the d diacritic
acr t c
of the ru
rule-feature
e-featuretheory or the d sjunct on of the d
disjunction disjunction theory. Th
sjunct ontheory Thiss behav
behavior or
in n except
exceptionalona forms iss unremarkab
unremarkable.25 e 25
In a more interesting
nterest ng group of Ch Chicano
cano d a ects descr
dialects, describedbed in Reyes, the d
n Reyes diph-
ph-
ternat on has been lost
zat on aalternation
thongization
thong ost across the board w th n a pecu
within peculiarly
ar y
characterizable
character zab e set of formsforms. Here d phthongs have rep
diphthongs aced the correspond
replaced corresponding ng
mp e vowe
ssimple vowels, s inn unstressedas
unstressed as we well as stressed pos t on For examp
position. example, e the standard
verb stem c[weo]nt- 'count' count has become c[we]nt- c[we]nt-. Th Thiss part
particular eve ng has
cu ar leveling
occurred on onlyy in rst-conjugat on verbs (see fn
n ffirst-conjugation fn. 88, above) w with th the [we] [o]
aalternation
ternat on (and in n ffirst-conjugation
rst-conjugat onj[we u]g-). F
u]g-) rst-conjugat on verbs w
First-conjugation th the
with
[ye] [e] alternation
a ternat on are not affected,
affected and verbs of the other conjugations
conjugat ons with
w th
eeither
ther aalternation
ternat on are not affectedaffected. Th Thiss means that that, in n add t on to the genera
addition general
process of d phthong zat on (wh
diphthongization (whichch rema
remains ns una tered) the grammar of these
unaltered),
innovative
nnovat ve d a ects must conta
dialects contain n a spec
speciala statement referr referring ng to the absence of
rst-conjugat on verb stems w
ffirst-conjugation th the 'back'
with back branch of d diphthongization.26
phthong zat on 26
On Hooper
Hooper'ss account of the loss oss of the h ternat on the m
gh-m d aalternation,
high-mid mid-vowel
d-vowe
member of the d disjunction
sjunct on inn the lexical
ex ca representation
representat on of each relevant
re evant stem iss
lost.
ost For exampexample, e standard /p{e,i}d-/
p{e }d- becomes /pid-/,
p d- and standard /m{ye,e,i}nt-/
m{ye e }nt-
becomes /m{ye,i}nt-/
m{ye }nt- (168) (168). Concom
Concomitantly,
tant y the second (mid-vowel)
(m d-vowe ) case of e 19
rule
ru
iss lost.
ost RuRulee 17 rema
remains ns una tered except for the loss
unaltered oss ofof/ Coi,
Co since
s nce the second
d-vowe ) case iss st
(mid-vowel)
(m still needed for the genera
general process of d phthong zat on In sum
diphthongization. sum,
on Hooper
Hooper'ss ana analysis,
ys s the loss
oss of the high-mid
h gh-m d alternation
a ternat oncons
consistssts of three seemingly
seem ng y
unre ated changes: (a) loss
unrelated oss of m midd vowe
vowelss in n individual
nd v dua lexical ex ca d disjunctions,
sjunct ons (b) loss
oss
of the env ronment of the second case of ru
environment rulee 17
17, and (c) loss oss of the second case of
rulee 19
ru 19. Noth
Nothing ng in n th
thiss account pred
predictscts loss
oss of the m mid vowel in
d vowe n aall forms
forms, rather
than the h high
gh vowel
vowe in
n some and the mid
m d in
n others,
others or the high
h gh vowel
vowe in
n aall forms
forms.
(Contrast this
th s with
w th the prediction
pred ct on made by my analysis.)
ana ys s )
23
Boyd-Bowman
Boyd Bowman 1960 1960; La
Laferriere
err ere 1974 196 Jean Longm
1974:196; Longmire persona commun
re (personal communication
ca on
regard ngthe
regarding he Span sh oof Mer
Spanish Merida,
da Venezuela);
Venezue a Reyes 1976;
1976 Rosenblat
Rosenb a 1946.
1946
24
eas one case in
There iss aat least n wh ch sstandard
which andardddialects
a ec shave whilee some non
eve ed wh
have leveled, non-standard
s andard
dialects
d a ec s have re a ned the
retained he aalternation:
erna on sstandard
andard en r e g 'deliver'
entr[e]g- de ver versus non non-standard
s andard
entr[ye
en r ye - ee]g-.
g
25 The fact
ac that
ha the
he ssimple vowel ra
mp e vowe rather han the
her than he d ph hong surv
diphthong ves in
survives hese cases oof neu
n these neu-
tralization supportss the
ra za on suppor he cclaim ha the
a m that he former basicc and the
ormer iss bas he latter
a er der ved pace Hooper
derived,
(168-9).
168 9
26 Such a ormu a edin
a emen iss formulated
spec a sstatement'
'special n Harr 1974a. I wou
Harriss 1974a wouldd now aalter
er cer
certain
a ndedetails
a s
oof that
ha sstatement,
a emen bu but none oof them ec the
hem aaffect he presen
present line
ne oof argumen
argumentation.
a on
Th s con en down oaded om 150 214 146 47 on Thu 12 Nov 2015 16 42 33 UTC
A use sub ec o JSTOR Te ms and Cond ons
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 55
Th s con en down oaded om 150 214 146 47 on Thu 12 Nov 2015 16 42 33 UTC
A use sub ec o JSTOR Te ms and Cond ons
56 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
Th s con en down oaded om 150 214 146 47 on Thu 12 Nov 2015 16 42 33 UTC
A use sub ec o JSTOR Te ms and Cond ons
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 57
sider again only the subset of evidence that Hooper treats explicitly, ??4.2 and
4.3 show that my analysis yields desirable and correct predictions which Hooper's
is forced to treat as unrelated accidents. In sum, on the basis of the Spanish evidence
we have examined, the general scientific considerations enunciated at the beginning
of this summary dictate that the rule-feature theory of non-automatic morpho-
phonological alternations should be preferred over the disjunction theory.
UNRESOLVED
ISSUES
5.1. RULE INTERACTION.
It is obvious that there are representations that meet
the structural descriptions of both rules 4 and 10; e.g., Isg. pres. indic. /dorm+o/
of the stem /dorm-/ 'sleep', whose stem vowel is specified [D] and [HM]. The
phonetic output [dwermo] shows that rule 4 (diphthongization) must apply, but not
rule 10 (high-mid). (Application of rule 10 alone would give *[durmo], and of 4
and 10, in that order, *[dwirmo].) We must somehow guarantee that only rule 4
applies in such cases.
Note that we could mechanically block application of rule 10 to a representation
like [dwermo] by placing a condition on the rule to the effect that the input segment
must not be immediately preceded by a [-consonantal] segment. That is, the outer
environment of 10 could be revised as follows:
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
58 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
provides no principle that can impose the correct order over the members of these
sets. The cases of 17 are ordered as shown solely because they happen to work that
way. In other words, 17 restates, but does not explain, the observations made in the
first paragraph of the present section. Also, as has been pointed out before, there is
no independent motivation for conflating the separate generalizations concerning
diphthongization and the high-mid alternation into a single rule such as 17.
The two analyses under study, and their associated theories, are thus equivalent
insofar as neither provides an answer to the question raised in this section. Whatever
answer is ultimately chosen for one theory may well be compatible with the other.
5.2. NOTATION, 'ABSTRACTNESS', AND EMPIRICALQUESTIONS. Hooper gives this
description of her goals: 'The purpose of this study is to present and examine a
version of generative phonology that places the strongest possible constraints on
abstractness in phonological descriptions' (5). Lexical representations and rules
that contain diacritics like [D] and [HM] are 'abstract' in the sense that these
marks are unpronounceable, and do not occur in phonetic representations. On the
other hand, the 'disjunction' representationsillustrated in 16 and in rule 17 contain
only features (abbreviated as alphabetical symbols) that occur in phonetic repre-
sentations. It does not follow, however, that the latter type of representation is less
abstract, in any useful sense, than those containing rule features. We may observe
immediately that lexical representations like 16 are also abstract, in that the dis-
junctions themselves are unpronounceable and do not occur in phonetic repre-
sentations. Furthermore, 16 and 17 are abstract in another sense. Compare 26a-b
(which repeat 16d and 17) with 27a-b:
rye) Fwe)
(26) a. /m\ e tnt-/, /d o rm.l
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
NON-AUTOMATIC MORPHOPHONOLOGICAL ALTERNATIONS 59
I should clarify that Hooper does not claim in her book that representations like
26 are less abstract than those containing undisguised rule-feature diacritics. The
point of these observations-a serious one, I believe-is to suggest that even what
seem to be 'the strongest possible constraints on abstractness in phonological
descriptions' may turn out, on close examination, to have little or no empirical
content.
There remains a substantive empirical issue which Hooper does not address.
Disjunctive lexical representations like 28a, on the one hand, and lexical representa-
tions containing rule features like 28b, on the other hand, together with their
respectively associated rules, are superficially very similar:
We have seen above, however, that the two theories represented by 28a-b have
different empirical consequences. Still, we must ask whether there is an absolute
empirical refutation of either or both theories: Does some significant linguistic
generalization exist that is absolutely beyond the range of these theories, in that it
cannot be stated, even in an ad-hoc way? For example, suppose there is found in
some language a non-automatic alternation (i.e. one found in a lexically determined
subset of forms) in which consonants are deleted before consonants and vowels are
deleted before vowels. The rule-feature theory could express this generalization by
means of a diacritically-triggeredalpha-rule. The disjunction theory, on the other
hand, could not express the generalization. If such a case were found, then the
disjunction theory would be disconfirmed. Other types of hypothetical cases can
be imagined that would disconfirm the other theory, or both. I cannot at present
provide any such clear real cases, from Spanish or any other language. In principle,
however, this gap in our understanding can be closed by empirical investigation.33
For the time being, the evidence from Spanish studied above weighs heavily in favor
of the rule-feature theory over the disjunction theory of non-automatic morpho-
phonological alternations.34
33 If the analysis of Dell & Selkirk (MS)is tenable, then French presentsa case that absolutely
refutes the disjunction theory. This analysis involves a rule the correct statement of whose
environmentis
+L[ X +L[ Y ---- CO]+L Z]+L
Here the diacritic [+L] is a property of category nodes (alternatively, paired brackets), and
there are two nodes (pairs of brackets) with this property.
34 A serious issue that has not been touched on at all in this study is that of placing the
heaviest possible constraints on the rule-feature theory. For recent suggestions, see Harris
1974b and Ringen & Iverson 1976.
This content downloaded from 150.214.146.47 on Thu, 12 Nov 2015 16:42:33 UTC
All use subject to JSTOR Terms and Conditions
60 LANGUAGE, VOLUME 54, NUMBER 1 (1978)
REFERENCES
ARONOFFM
ARONOFF, M. 1976 orma on in
1976. Word formation n genera
generative ve grammar
grammar. Cambr Cambridge,dge MA MA: MIT
Press.
Press
BOYD-BOWMAN,
BOYD BOWMAN P. 1960
P 1960. E El hab
hablaa de Guana ua o Mex
Guanajuato. Mexico: Imprentaa Un
co Impren Universitaria.
vers ar a
BRAME,M
BRAME M. KK., and II. BORDELOIS
BORDELOIS. 11973.
973 Voca erna ons in
Vocalicc aalternations n Span sh LI 44.111-68.
Spanish. 111 68
CHOMSKY,
CHOMSKY N., and M
N HALLE.1968. The sound pa
M. HALLE1968 ern oof Eng
pattern sh New York
English. York: Harper
& RowRow.
DELL, F
DELL F. CC., and EE. OO. SELKIRK
SELKIRK.MS MS. On morpho og ca y governed vowe
morphologically vowel aalternations
erna ons in n
French. (Linguistic
French L ngu s c Inqu monographs, 33.) To appear
ry monographs
Inquiry appear, MIT Press Press.
DINNSEN,D
DINNSEN D. AA. 1974
1974. Cons ra n s on gglobal
Constraints oba ru es in
rules n phono ogy Lg
phonology. Lg. 5050.29-51.
29 51
ESPINOSA,A
ESPINOSA A. MM. 1946
1946. Es Estudios
ud os sobre eel espanespaniolo de Nuevo Me co II
Mejico, II: Mor
Morfologia.
o og a
B b o eca de d
(Biblioteca dialectologia
a ec o og a h spanoamer cana II
hispanoamericana, II.) Buenos A Aires:
res FacuFacultadad de
Filosofia
F oso a y Le ras de laa Un
Letras Universidad
vers dad de Buenos A Aires,
res Ins u e de F
Institute Filologia.
o og a
HARRIS,JJ. W
HARRIS W. 1974a
1974a. Morpho og za on oof phono
Morphologization og ca ru
phonological
phonologica rules: examplee from
es an examp rom Ch Chicano
cano
Spanish.
Span sh L ngu s c sstudies
Linguistic ud es in anguages ed
n Romance languages, ed. by R R. JJ. Campbe
Campbell, M M. G G.
dn &M
Goldin,
Go M. C C. Wang
Wang, 88-27.
27 Wash
Washington:
ng on George
Georgetown own Un vers y Press
University Press.
---. 1974b1974b. Lo mor o og co en una grama
morfologico gramatica ca genera va aalternancias
generativa: ernanc as voca cas en las
vocalicas as
formas
ormas verba
verbaleses de
del espan o E
espaniol. El espano
espanol y laa lingiiiustica
ng us cagenera
generativo-transformacional,
vo rans ormac ona
ed. by JJ. Gu
ed ar & JJ. Roy
Guitart Roy. Barce ona Ed
Barcelona: Edicions
c ons 6262-Peninsula.
Pen nsu a [In In press
press.]
--. 1977a. Aspec
1977a Aspectss oof Span
Spanishsh verb morpho
morphology.ogy S Studies
ud es in ngu s cs ed
n Romance linguistics, ed.
by M M. PP. Hag wara 44
Hagiwara, 44-60.
60 Row ey MA
Rowley, MA: Newbury House House.
1977b. Span
1977b Spanishsh vowe
vowel aalternations,
erna ons d diacritic ea ures and the
acr c features, he sstructure
ruc ure oof the he
lexicon.
ex con ProcProc. 77th h Annua
Annual Mee ng Nor
Meeting, North h Eas
Eastern
ern L Linguistic
ngu s c Soc e y 99
Society, 99-113.
113
Cambridge,
Cambr dge MAMA.
. 1977c
1977c. Remarks on d ph hong za on in
diphthongization n Span
Spanish.sh L ngua 41
Lingua 41.261-306.
261 306
HOOPER,J. B
HOOPERJ B. 1973 Aspectss oof na
1973. Aspec natural
ura genera
generative ve phono ogy UCLA d
phonology. dissertation.
sser a on
--. 1974. Ru
1974 Rulee morpho og za on in
morphologization n na
natural
ura genera
generative ve phono ogy Na
phonology. Natural
ura Phono
Phonologyogy
Parasession,
Parasess on 160160-70.
70 Ch cago CLS
Chicago: CLS.
. 1975
1975. The arch segmen in
archisegment n na
natural
ura genera
generative ve phono
phonology.
ogy Lg Lg. 51
51.536-60.
536 60
--. n roduc on too na
1976: An introduction
1976 natural
ura genera
generative ve phono ogy New York
phonology. York: Academ
Academicc
Press.
Press
HUDSON,G. 1975
HUDSONG 1975. Supp e on in
Suppletion n the
he represen a on oof aalternations.
representation erna ons UCLA d dissertation.
sser a on
LAFERRIERE,
LAFERRIERE M. 1974
M 1974. Some theoretical
heore ca properproperties es oof rurulee add on and ru
addition rulee loss.
oss
Papers fromrom the he 55thh mee ng Nor
meeting, North h Eas
Eastern
ern L Linguistic
ngu s c Soc e y 192
Society, 192-9.9 Cambr
Cambridge,dge
MA: Harvard Un
MA University.
vers y
LEBEN,W. R
LEBENW R., and 00. W ROBINSON.1977
W. ROBINSON 1977. 'Upside-down'
Ups de down phono phonology.
ogy Lg Lg. 5353.1-20.
1 20
REYES,R. 1976
REYESR 1976. S ud es in
Studies n Ch
Chicano
cano Span sh Harvard d
Spanish. dissertation.
sser a on
RINGEN,C., and G
RINGENC IVERSON.1976
G. IVERSON 1976. On cons ra n ng the
constraining he theory
heory oof excep ons Read aat
exceptions.
Winter
W n er mee
meetingng oof LSA
LSA, Ph Philadelphia.
ade ph a
ROSENBLAT,
ROSENBLAT A 1946. B
A. 1946 Biblioteca
b o eca de d dialectologia
a ec o og a h spanoamer cano II
hispanoamericano, II. Buenos A Aires:
res
Facultad
Facu ad de F oso a y Le
Filosofia ras de laa Un
Letras Universidad
vers dad de Buenos A res Ins
Aires, u o de
Instituto
Filologia.
F o og a
VENNEMANN,
VENNEMANN T. 1971
T 1971. Na Natural
ura genera
generativeve phono ogy Read aat W
phonology. Winter
n er mee
meeting ng oof LSA
LSA,
St. Lou
S Louis.
s
--. 1972a
1972a. Phono
Phonological
og ca un queness in
uniqueness n na
natural
ura genera
generative grammar. G
ve grammar ossa 66.105-16.
Glossa 105 16
. 1972b
1972b. Ru Rulee inversion.
nvers on L ngua 29
Lingua 29.209-42.
209 42
1974. Words and sy
--. 1974 ab es in
syllables n na
natural
ura genera
generative grammar. Na
ve grammar Natural
ura Phono
Phonology
ogy
Parasession,
Parasess on 346
346-74.
74 Ch cago CLS
Chicago: CLS.
Rece ved 5 Apr
[Received April 1977
1977.]
Th s con en down oaded om 150 214 146 47 on Thu 12 Nov 2015 16 42 33 UTC
A use sub ec o JSTOR Te ms and Cond ons