J. COMMUN. DISORD. 17(1984), 325-348
A NONLINEAR ANALYSIS OF PHONOLOGICAL DISABILITY ANDREW
SPENCER
Speech Therapy Department Central School of Speech and Drama and Engineering and Sciences Division Polytechnic of Central London London, England
In the spirit of Gandour (1981) a reanalysis is presented of the speech of a byr-old language delayed child, Rosey, whose phonologic disorder has been discussed by Grunwell (1982). A revealing account of the data can be given by adopting a nonlinear framework; specifically, recent amendments to Goldsmith’s (1976) theory of autosegmental phonology. This analysis borrows from autosegmental accounts of Arabic morphology, reduplication, and Kikuyu tone shift (all of which are briefly described). The processes postulated to explain the “deviant” pattern of data, hence, are all well motivated from the grammars of other languages. The analysis is contrasted with a “classical” (SPE) generative analysis and, thus, provides indirect evidence in favor of autosegmental theory over the classical theory. At the same time, the position of such phonologic analyses in psycholinguistic models of acquisition is briefly discussed.
INTRODUCTION Recently, Gandour (1981) has presented an analysis of data discussed by Lorentz (1976) from the point of view of Kahn’s syllabic theory of phonology (Kahn, 1976). Gandour argues that the child studied by Lorentz, Joe, displayed speech patterns that were abnormal by comparison with adult phonology but that nonetheless proved to be “normal” when assessed relative to general linguistic parameters. Gandour argues further that his analysis, which brings order to apparent chaos, provides support for the syllabic theories he discusses. He suggests that it is only when the appropriate descriptive system is employed that otherwise puzzling clinical data can be viewed in a proper perspective, with obvious consequences for remediation. Address correspondence to Andrew Spencer, Speech Therapy Department, Central School of Speech and Drama, Eton Avenue, London, NW3 3HY, England. This paper is a much expanded version of a talk given at the Third National Conference of Speech Therapy Lecturers, held in April 1983 in Oxford, England. 0 1984 by Elsevier Science Publishing Co., Inc. 52 Vanderbilt Ave., New York, NY 10017
325 002 l-9924/84/$03 .OO
A. SPENCER
326
In this paper I shall adopt a similar strategy, in a slightly different domain, and come to similar conclusions. I reanalyze data first presented by Grunwell and Pletts (1974) discussed in greater detail in a pedagogic context by Grunwell (1982). I shall show that when these data (from a child named Rosey) are examined from the point of view of nonlinear phonologic theory, in particular, autosegmental phonology (Goldsmith, 1976) they no longer appear idiosyncratic but, rather, turn out to be highly systematic simplifications of adult representations derived by dint of a small number of rules and principles some of which, according to current theory, are of very wide applicability. Rosey’s productions, far from being genuinely deviant, are just very immature and in fact pattern more or less as would be expected, given plausible psycholinguistic assumptions and motivated universal principles.
ROSEY’S DATA-A SOUND PATTERN OF ENGLISH ANALYSIS Grunwell (1982) presents Rosey’s data in a textbook of phonology for speech pathology students, and uses them to exemplify a “classical” analysis of children’s data within the framework of SPE (Chomsky and Halle, 1968). For the purposes of her pedagogic account Grunwell assumes that the child’s forms can be analyzed as derived from the adult surface forms by means of a set of “realization rules” (cf. Smith, 1973), mapping the adult forms onto the child’s output forms. In such a system, the adult surface forms fulfill the function of underlying representations (e.g., the phonologic representations given in the lexicon) in the derivation of the surface form. (It should be emphasized, however, that Grunwell adopts this stance for pedagogic purposes only; she is, in fact, a staunch critic of most varieties of posttaxonomic phonology, at least in the description of child language.) Table 1.
Data from Rosey
(1) monkey [‘mulri] (2) bottle [‘bDbD] (3) garage [‘dadi] (4) feathers [‘fEf&] (5) tclly [‘dek] (6) christmas [‘dIdI] (7) indian [‘AiLi] (8) scissors [W] (9) chimney [‘dlrli] (10) letter [‘lele] (11) pinny [‘bdi] (12) flowers [‘fafa] (13) budgie [‘buhi] (14) elephant [‘r(d) (15) li?rle [‘Ml]
(16) dressing [‘d&C] (17) fire [‘fafa] (18) pencil [‘brbc, ‘pcp~] (19) matches [‘maKi] (20) finger [‘fifi] (21) paper [‘bebc] (22) settee [‘d&G] (23) dolly [‘dDAi] (24) sugar [‘tutu, ‘dudu] (25) aeroplane [bc ‘be] (26) Cheryl [‘deli] (27) middle [‘mlml] (28) kitchen [‘dIdI, ‘tltl] (29) sleeping [Wi] (30) chocolate [‘dDdo]
From G~~nwell. 1982, p. 156. Reprinted with permission of Croom Helm.
PHONOLOGIC
327
DISABILITY
The main data set, consisting of 30 items elicited during assessment with the Edinburgh Articulation Test (Anthony et al., 1971), is given in Table 1. Grunwell and Pletts (1974) contains further examples of Rosey’s speech (it includes, for instance, monosyllabic items and gives much more phonetic detail). However, Table 1 represents the core component of Rosey’s phonology. As Grunwell points out, such a pattern of data is unusual for a child of this age, but similar kinds of phenomena have been reported for other children and, overall, the impression is one of extremely delayed rather than deviant speech. A cursory examination reveals widespread reduplication of syllables and a “favorite syllable” strategy for some second syllables, the vowel of which is [i] and the consonant of which is variously [XJ, [l], or lj]. I ignore this “allophonic” variation for expository purposes (in agreement with Grunwell) and notate the consonant as the palatal [Xl. The dialect Rosey was learning was that of the English town of Leicester (in the Midlands), which deviates in a number of ways from so-called RP. In most cases, this does not seem to affect interpretation of the data. I shall first present an analysis couched in the SPE framework (slightly modified). Unfortunately, there are a number of technical flaws in Grunwell’s treatment, from which I deviate somewhat. (I)
(2)
(3)
Cl
0/c-
[I kF3:ant] V
+ +
[+
anterior delayed release high back coronal
tense]/#(C)VC
+ L-l -
(4)
(5)
C
cv
(6)
0
(7)
(C(V))0
0 / # (C)/V
~
high back
328
A. SPENCER
These rules’ are extrinsically ordered and have the following effects: Rule 1 simplifies all consonant clusters by deleting all but the first member. Since there are no V V sequences presumably in underlying forms, this guarantees that all words have the shape (C)VCV , . . (C), strings of syllables, open except possibly for the last. The use of subscripting for the left hand side of this rule is nonstandard, but it could be interpreted as a right-to-left iterative rule (cf. Anderson, 1974; Kenstowicz and Kisseberth, 1977). Rule 2 neutralizes the distinction between velars and alveolars and, at the same time, ensures that all obstruents are stops (i.e., it neutralizes fricatives and affricates). (Notice that this rule makes use of the feature [ -+labial] which has been argued for by a number of researchers, including Gandour, 1981.) Rule 3 ensures that the distinction between tense /i/ and lax /I/ is neutralized in the second syllable, and rule 4 then deletes the consonant before /i/ (from either source). Rule 5 reduplicates the first syllable of a word beginning CVCV. Rule 6 inserts a palatal lateral [Q between two vowels (which, perforce will be derived through operation of rule 4) or before a word-initial vowel. Finally, rule 7 deletes all material following the second syllable. This is not intended to be a definitive analysis within an SPE-type framework, though all the rules except 1,2, and 7 represent rule types motivated in phonologies of the world’s grammars. The latter incur massive “aba controversial phenomenon in adult grammars, solute neutralization,” though unexceptional in generative accounts of child phonology, for obvious reasons. Particularly noticeable is the fact that the rule system conspires in a fairly transparent way to generate words with a simple caof this kind have been well known since nonical shape. “Conspiracies” the work of Kisseberth (1970), and constitute a serious conceptual problem for the classical SPE framework. For instance, very similar environmental specifications are given for several of the rules (especially 3, 4, 5, and 7), though there does not seem to be any obvious way of collapsing them so as to eliminate the redundancy. Moreover, rule 6, while reasonably simple in its formulation, involves an unpleasant disjunction, highly suggestive of a syllable-based process. A further criticism that might be leveled against a rule system such as l-7 is that it postulates a fairly complex set of rules whose purpose is to simplify the child’s output. Now, as the child’s speech improves and more closely approximates the adult norm, one would expect these rules to be discarded so that the child’s output eventually comes to resemble the input, the adult surface forms. This assumption, that learning a phonology involves loss of rules as competency develops, has been attacked (essentially) on conceptual grounds, e.g., by Grunwell(l981) and Donnegan ’ Voicing does not seem to be used contrastively her phonology both here and in the reanalysis.
by Rosey.
1 ignore
this dimension
of CI
PHONOLOGIC
DISABILITY
329
and Stampe (1979). While it seems to me that these attacks are largely misguided2 the fact remains that a theory that viewed development as an accretion of rules and representations, rather than a loss of them, might be considered intuitively more satisfying by some. I shall suggest a reanalysis of Rosey’s data that appeals to recent theories of autosegmental phonology. I shall make use of devices that have been proposed for the phonologies of languages such as Arabic, YidinY, and Kikuyu, in conjunction with minimal assumptions about the psycholinguistic mechanisms underlying the acquisition of phonology. I shall show not only that Rosey’s speech is not deviant, measured against the metric of universal grammar and markedness theory, but that several aspects of her phonology are precisely what would be expected, granted certain motivated assumptions. Thus, I shall arrive at conclusions very similar to those of Gandour, on the basis of a somewhat different data set, but with an analysis that is set within a closely related theoretical framework. NONLINEAR
PHONOLOGIC
THEORY
Nonconcatenative Morphology in Arabic It is sometimes naively assumed that the common processes of prefixation and suffixation, typical of well known Indo-European languages, are in some sense the “standard” ways of forming words. In point of fact, socalled nonconcatenative morphology (i.e., morphology that does not involve simply adjunction of affixes to a root or stem) very frequently is Z A lot of nonsense has been talked about the “counterintutiveness” of sets of realization rules that become simpler as the child gets older. Firstly, this ignores the fact that the child’s phonologic system, as a whole, is gaining in complexity, e.g., by virtue of reanalysis of underlying forms as adult underlying forms rather than adult surface forms. More importantly, those who question whether or not children can be thought of as applying realization rules to derive a simplified output are missing the point. The question to ask is not, “Why do children have phonologic rules?” but, “Why does anyone have phonologic rules?” In other words, “Why is there consistently a mismatch between underlying form and surface form, accounted for by rules?“. This boils down to the question, “Why is generative phonology a more explanatory theory than taxonomic phonemics?” Presumably, the reason for all the paraphernalia of the phonologic rule component of a grammar has to do with processing strategies, comprehension, ease and efficiency of articulation, and so forth, in the final analysis. These performance constraints surely loom even larger before the young learner whose cognitive faculties, in many respects, are less well developed than those of the adult. Given the fact that young children typically understand much of what is said to them, and given the assumption that this is something to do with their appreciation of the phonetic form of utterances they hear (to which only a radical behaviorist or a Gibsonian would object) it seems reasonable that the child will come to apply rules (whose format will be provided by universal grammar) that simplify even the simplified surface forms he hears from his fully competent interlocutors. Whether or not the resulting rule system is “complicated” in any revealing sense is not a question that can be left to raw intuition and, as far as I can see, our understanding of the nature of phonologic acquisition has not reached the point where the matter can be fruitfully discussed.
A. SPENCER
encountered among the world’s languages. There are some languages that make very little use of prefixes and suffixes, but show considerable reduplication and inlixation. These phenomena, until recently, have received scant attention from generative phonologists, in part because the SPE framework is very hard put to provide an illuminating account of them. The problems posed by nonconcatenative morphology, however, do intrude into Indo-European linguistics. Thus, it is common among the Germanic languages to find a series of verbs (“strong” verbs) that form past tense and past participle not by aflixation but by changing the stem vowel. In English, for instance, we observe cases such as sing/sang/sung. Arabic is a good example of a language whose morphology looks very different from that of Indo-European languages. A theory of Arabic morphology recently has been proposed by McCarthy (1979, 1981) based on Goldsmith’s (1976) autosegmental theory. (McCarthy’s work is also reviewed by van der Hulst and Smith, 1982, and by Halle and Vergnaud 1980). An essential notion in McCarthy’s theory is that of the prosodic template or CV skeleton, The basic idea is very simple: Words in Arabic tend to conform to fairly strict conditions on syllable structure. Thus, in verb paradigms only a certain number of the theoretically possible strings of consonant and vowel sequences (CV sequences) are actually found. Another recurrent feature of Semitic morphology is the triliteral root. Word roots in Arabic consist of three consonants. Vowels are slotted around these to form words in various inflectional and derivational guises. Thus, a word consists of (a) a CV skeleton, (b) a triliteral consonant root, (c) a vowel sequence, and (d) affixes (including infixes). Thus, from the root “k-t-b” (meaning “to do with writing”) we can form kataba-he wrote; kattaba-he caused to write; kuttaabun-Koran school; maktabun-office; makaatibu-offices (McCarthy, 1979). McCarthy shows that it is possible to provide a revealing account of Arabic morphology by assuming that the CV skeleton and all morphemes (affixes, root, and vowel sequence) exist on separate levels of representation. Hence, each word is given a multidimensional representation, as opposed to the two-dimensional representation of traditional theories in which a word consists of a linear string of segments or distinctive feature matrices. (It is for this reason that the theory is considered a nonlinear theory.) The CV level usually is called the segmental tier and the others, autosegmental tiers. The elements on these tiers reside in distinct planes of existence, and so customary relations of precedence (linear sequence) do not apply to them. A word in pronunciation, however, is linear in form so some mechanism must be available for achieving this linearization. This is done by associating the various autosegments (elements of autosegmental tiers) with the segmental tier (the CV skeleton). McCarthy’s rules of association
PHONOLOGIC
DISABILITY
331
are similar to those originally
proposed by Goldsmith, describing tone phenomena in African languages. In 8 are listed forms (“binyanim”) from the perfective active paradigm of the verb based on the root k-t-b, together with the characteristic CV skeleton of each of these eight binyanim (cf. McCarthy, 1979). (8) I. II. III. IV. VI. VII. VIII. x.
cvcvc cvccvc cvvcvc cvccvc CVCVVCVC ccvcvc ccvcvc ccvccvc
(The roman numerals
katab kattab kaatab ?aktab takaatab nkatab ktabab staktab
write cause to write correspond cause to write write to each other subscribe write; be registered write; make write
are from traditional
Arabic descriptive
practice.)
In addition to the consonants of the triliteral root k-t-b we see affixes (e.g.,? in IV, t in VI, n in VII, st in X), geminated consonants (e.g., tt in II), and recurrent consonants from the root (e.g., b in VIII). We begin
with the simpler cases, however, I and III. These contain no affixes and their underlying forms are given in 9-10. (9a) (9b) (9~) (lOa) (lob) (10~)
CVCVC ktb a CVVCVC ktb a
Association lines are drawn from the elements of the root to consonant slots in the CV skeleton one by one, left to right, and a similar (in this case trivial) procedure is followed with the vowel melody. Thus, we have
11 and 12:
katab
332
A. SPENCER
(12)
Ir.
a
a
A
CqV/J,$’
j
\
3
kaatab
1,
‘kid
ktb v
(The Greek letter p. represents
IJ-
morpheme.
Forms IV and VI contain prefixes. These are associated as separate morphemes as in 13 before the root consonants are associated. We then obtain derivations 14 and 15 (omitting the vowel melody for perspicuity): C VCCVC
(13a)
(14) c vc
cv
c
3
C
cvqFv,c 1, ‘\I
I
/
I
I
Fl vvcvc
3
c v c,,v
v ty v c
t
JL2
?aktab
kb
I P-1
PO
3
\I/
?
\Ii CL0 3
,’
‘\
I
cvccvc
kiti’
CL1 cvc
3
‘i v
‘7
(IS)
VCVVCv c
(13b)
I I’
‘\
i VIL0 CL? I cvc~vvfvF ‘\
3
takaatab
‘\
t I P-2
\ I I’ ktti
CL0
In example 8 III it will have been noticed that the vowel appeared as a geminate “aa” in the first syllable. Gemination or repetition of consonants also occurs. An interesting example is illustrated in derivation 16
PHONOLOGIC
(again,
omitting
333
DISABILITY
vowels): ccvcvc ’
l&l
ktb V
ktabab
CL0
What is unusual about this form, compared with those we have seen so far, is that the CV skeleton has more consonant slots than there are consonants in the root (there are no affixes to fill them either). A general, universal principle demands that the difficulty be resolved by a process of spreading. After each consonant slot is associated with a consonant from the root in a left to right fashion, the remaining slots are filled by spreading the nearest root consonant, in this case the final b of the root. Why can a different consonant from the root not spread to fill this unfilled slot? This, in fact, is prohibited unless other rules apply to alter the representation radically, for there is a very general principle of autosegmental phonology that states that association lines can never cross. Thus, a representation such as 17 would be completely illformed:
(“IY$y
*ktabat
To draw together this discussion I shall present the model of phonologic organization presupposed by McCarthy’s analysis of nonconcatenative morphology. Let us call the triliteral consonant root of Arabic words the root “melody.” The consonants comprising the root will then be “melody elements.” Likewise, the vowel sequence will be the vowel melody. Let us call the C and V slots making up the prosodic template (CV skeleton) the “melody-bearing elements”. Similarly, the “melody tier” and the “melody-bearing tier” will refer to the levels at which those representations exist. The association principles by which elements from the various tiers are associated ultimately, so as to form linearized strings of segments, are given in 18 (slightly adapted from McCarthy): (18) a. b. c.
Associate elements of the melody tier with elements of the melody-bearing tier one to one, left to right. After application of (a) associate a single melody element with all remaining unassociated melody-bearing elements. If all melody elements are associated and some melody-bearing elements remain unassociated, associate the latter with the nearest melody element on the left.
Finally, I state a modified version of the well-formedness condition (WFC) on autosegmental representations suggested by Goldsmith (1976) and revised by Halle and Vergnaud (1980).
A. SPENCER
334
(19) a. b.
Each melody-bearing element is associated with (at least) one melody element. Association lines never cross.
Principle 19a is subject to language-specific variation insofar as melodybearing elements, in some cases, can be associated with more than one melody element. This is stipulated by the rules of the language and is generally a marked option. Principle 19b is absolutely inviolable.
Reduplication An all-pervading feature of children’s phonology is the phenomenon of reduplication (Fee and Ingram, 1982). This is a process that is common in adult phonologies. McCarthy provides the outline for an analysis of this phenomenon but, for expository purposes, I shall describe a much more detailed analysis of reduplication presented by Marantz (1982). I shall restrict myself to one particular variety of reduplication, syllabic reduplication, found in the Australian language, YidinY. Marantz notes that this is the only language he has found, of the dozens he surveys, which shows this kind of reduplication and recent unpublished work by Juliette Levin (reported in a talk at the GLOW colloquium in York, England, March 1983) suggests that a more general theoretical analysis is possible. For descriptive and expository purposes, however, Marantz’s account suffices. I address the reasons why syllabic reduplication should be unexpectedly preponderant in child language later. In Yidiny, reduplication of the first two syllables of a word may occur, for example, in noun plural formation. Thus, we observe example 20 (syllabication is represented by a dot). (20) a.
b.
&.mu.rU gin.dal.ba
In agreement
‘house’ d_.mu.di.mu.rU ‘lizard’ gin.dal.gin.dal.ba with McCarthy,
ture is arranged hierarchically shown schematically in 21:
‘houses’ ‘lizards’
Marantz assumes that the skeleton strucinto syllables and ultimately morphemes.
(2 1) phoneme melody CV skeleton syllabic
C \jcvccii
skeleton
vu/v “\
morpheme
symbol
i/”
Y
Syllabic reduplication consists in copying the syllabic skeleton and adding it to the left of the stem. The syllabic skeleton then borrows both the
PHONOLOGIC
DISABILITY
335
CV skeleton and the phoneme melody from the stem, as shown in 22:
3
j
d.imud_imurU
gindalgindalba
li
F
By the principles of autosegmental phonology, unassociated elements such as the rU syllable of 22a or the ba syllable of 22b are deleted at the end of the derivation. The important point to bear in mind about this derivation is that it makes crucial appeal only to the level of syllable structure represented by the sigma level, apart from principles motivated elsewhere. This level df representation, itself, is necessary to describe
A. SPENCER
other syllabic processes (such as the syllabic phenomena mentioned in Gandour, 1981. See Selkirk, 1981, for further discussion of the necessity for prosodic levels of representation). Tonal Phenomena I now turn briefly to a somewhat different domain in which autosegmental phonology has been applied. Clements and Ford (1979) present a very interesting analysis of certain tonal phenomena in the Bantu language Kikuyu and its close relatives. In particular, they discuss the phenomenon of “downstepping,” and note that it can be explained if it is assumed that representations of the tone of Kikuyu words includes a so-called “floating tone,” which is not part of the pronunciation of the word as such but has tonologic effects on the following words. The tone systems of Bantu languages tend to be somewhat complex but, put in very simple terms, most of these languages have a tone system that places high (H) or low (L) tone on a vowel. This is generally interpreted phonetically as a contour tone. For example, a sequence HL on a single vowel will be a tone falling from high to low. In addition, different levels of tone are often found. Thus, a language may include extra high or extra low tones in its inventory. A recurrent phonetic phenomenon (which acquires phonologic status in some languages) is downstepping (or downshift). This is a process whereby a high tone is lowered slightly in certain phonetic/phonologic environments. Typically, a high tone that follows a low tone (or a falling tone) will be downstepped. In many cases, this is a straight forward case of phonetic assimilation rather similar to the assimilation to point of articulation found between /ml and a following labiodental fricative in English by which the nasal becomes labiodental [m]. In some cases, however, downstepping produces phonologic contrasts on the surface. In either case, a phonologic theory must be able to explain this pervasive phenomenon. Clements and Ford, along with other investigators, assume that the tone sequence in phonologic representations of a word or phrase in the Bantu languages exists on a separate plane from the nontonal segmental sequence. In other words, tone is autosegmental and can be operated upon by phonologic rules independently of the sequence of consonants and vowels with which it is associated. This phenomenon is apparent in English intonation, also. For example, it is well known that the English intonational system includes a fall-rise tone whose pitch peak is on the most accented syllable of the phrase, whose fall occurs on ihe next syllable, and whose final rise occurs on the final syllable. Schematically. this can be represented as a contour HLH, exemplified in 23:
PHONOLOGIC
(23)
337
DISABILITY
Tom may be able to do it 8 l
a.
e b.
.oro
l
Tom can 5
C. “;L
As 23c shows, the whole contour can fall on a single syllable. This indicates that the tone contour is independent of the segmental representation. In an autosegmental account we might suggest representations 24 for these cases (see Goldsmith, 1981, for further discussion of this line of reasoning). (24) a. b.
Tom may be able to do it H L H Tom can H LH
C.
EH (The case is complicated by accent phenomena and I shall not discuss it in further detail. The point of importance is that intonation in English is partially independent of segmental representation).
Returning to the Bantu data, Clements and Ford motivate an analysis of Kikuyu that posits a tone melody with each morpheme. The tone melody consists of sequences of high and low tones. These are associated with melody (tone)-bearing elements (which we can identify with vowels) in the segmental representation. Since the details of CV skeletal structure are not of importance for the argument, I shall follow Clements and Ford in omitting them. A tonologic rule that serves to differentiate Kikuyu from its immediate neighbors is given in 25: (25) Initial tone association rule: associate the first tone with the second tonebearing unit. By 25, no first syllable of a Kikuyu word will be given its own tone. However, all Kikuyu syllables bear some sort of tone so that the initial syllable must be assigned tone somehow. The principles of autosegmental phonology guarantee that the tone assigned is that of the second syllable. In other words, tone spreads to the left to associate with the unassociated tone-bearing element, in the manner shown in 26. This process is mandatory, indeed, for any other outcome would violate either the language-
338
A.SPENCER
particular rule of Kikuyu, tone or the WFC. (26) a.
mo e
r-E
ka nge
ri
c
rc
LH kv ngr
ri
c
LH b.
moe
; ,
c.
m"
way
’
I
mqe,,,,
LH ri
LH
(= ,
’
i 'g "t zI"_ LH
d.
which states that all syllables must receive a
k/
qw
LH
of releasing oneself quickly; - , low
tone; ‘,
high tone.)
In 26a we see the segmental representation of the word moeriklngerii and its independently represented tone contour, LHLH. By rule 2.5 (a rule peculiar to Kikuyu, recall) the first L of the contour is associate with the second tone-bearing element, the vowel e. By the principles of association, the rest of the HLH sequence is then mapped in a one by one fashion onto the next tone-bearing elements in the segmental sequence. In 26c the final H tone spreads to the unassociated tone-bearing elements to its right. In 26d the first syllable is associated with the nearest tone (L). Rule 26e shows the final representation. Now consider an example such as 27. (27) ikara!
charcoal
The raised exclamation point, !, means downstep. That is, it indicates that, if the first syllable of the word following ikara! bears a high tone. that tone will be downstepped, i.e., pronounced at a slightly lower pitch than usual. Judging from the surface tonetic form of the word ikara!, this is somewhat surprising, since downstepping is generally found after a low (or, of course, extra low) tone, yet, ikara! ends in a high tone. To explain this discrepancy, Clements and Ford suggest that the un
PHONOLOGIC
339
DISABILITY
derlying tonological representation of 27 is 28. The tone melody of this word is LHL, where -L signifies extra low tone. (28) i ka ra LHL -
The derivation of the word in its first stages is straight forward given the rules and principles so far described. The initial tone association rule 25 associates the first L with the second syllable, to give 29a. Then, the next tone-bearing element (the final syllable YU)is associated with the next tone, H, 29b. Next, the first syllable is associated with the first low tone by the process of spreading 29~: ka _,
(29) a. i *’
I’
i*
b. i> L c.
,,,,,,? H
!E
H
!c
i i+?/
L
What happens to the unassociated extra low tone (L> on the tone melody tier? A language-specific rule allows this tone to float, i.e., to remain unassociated. This is a marked option. In other cases (such as in other languages/dialects) one might expect the tone to be deleted by some (language-specific) rule or to be adjoined to the final syllable (along with H) to form a compound tone. Such options are unavailable in Kikuyu. The extra low tone, thus, is never pronounced as such, and only makes its presence felt by virtue of its effects on a following high tone. This effect is downstepping.
SUMMARY Much more could be said about these phenomena and these data. My purpose here, however, has not been detailed discussion of the merits of any particular analysis of the Arabic, YidinY, or Bantu data, but to give an illustration of the kind of reasoning that leads to the postulation of certain sorts of representation. I have omitted several important steps in the justification of these representations, particularly in the Arabic and Kikuyu cases. But the essential points to bear in mind are the notion of
340
A. SPENCER
separate, autosegmental levels of representation, particularly the CV skeleton and the morphemic and syllabic tiers; the WFC on autosegmental representations and the principles of association; and the notion of a floating element on an autosegmental tier.
A NONLINEAR ANALYSIS OF ROSEY’S DATA Let us now consider how the analytic techniques described can be brought to bear on the descriptive problem presented by Rosey’s data, provided in Table 1. To begin with, I shall make a number of psycholinguistic assumptions about the way Rosey has gone about the acquisition process. 1 assume, first, that Rosey’s phonologic system can be viewed as a mapping of adult surface forms onto Rosey’s own output forms. A set of phonologic rules then operate in conjunction with a set of WFCs, akin to morpheme structure conditions (MSCs), to generate Rosey’s final output. It seems to me that to postulate such an output lexicon is uncontroversial. It may appeal to representations different from those of the input lexicon (indeed, it must do so if, as is presumably the case, Rosey’s comprehension of adult language is significantly greater than her expressive ability). At the same time, the phonologic rules that ultimately account for the forms Rosey produces will include CV skeleta and other constraints on well-formed representations that, for example, can do the job of Smith’s (1973) MSCs without committing the analysis to the unreasonable view that Rosey is operating with “her own system.“‘3 This will allow us to circumvent the kinds of problems which, I believe, will inevitably arise if data such as Rosey’s are treated within the classical SPE model. The first part of the descriptive problem, then, is to provide Rosey with an output lexicon. It would appear (from data provided in Grunwell and Pletts, 1974, but not included in Grunwell, 1982) that Rosey can distinguish in her own speech between adult monosyllables and adult polysyllables, pronouncing the former as monosyllables and the latter as disyllables. Also, there may well be important effects due to internal syllable structure and stress, but the data are insufficient to support speculation and I shall ignore the problem, limiting myself to discussion of the data of Table 1.
3 Recall that Smith employed MSCs in an analysis of his son’s data “as a self-contained system” with a view to showing the inadequacy of such an approach. While his demonstration was valid, given assumptions then current about phonologic organization, presentday conceptions, especially McCarthy’s notion of the CV skeleton, allow some of the descriptive apparatus of Smith’s rejected analysis to be incorporated into a theory consonant with the philosophy of Smith’s preferred solution.
PHONOLOGIC
341
DISABILITY
The first rule of Rosey’s phonology as 30:
is segmental rule 2, repeated here
r + anterior1
I then assume that Rosey imposes the maximally simple CV skeleton (31) over the adult forms to derive her output lexical forms. (This assumption is by no means the only plausible or natural one compatible with the data, and a more sophisticated analysis might well link 31 with unmarked canonical syllable structure. Little hinges on the precise way these forms are.derived, in fact.) In this way, Rosey’s output lexicon will contain entries, such as 32 and 33, derived as in 34 and 35, respectively. The derivation has the effects of rule 1. (31)
c v
(32)
bottle:
bo
(33)
flower:
fa
(34)
bqt,l \ I \ c i,
3
f”” 6 1
*
hot! I \ c v
(35)
Suppose, “monkey” this fact is (analogous of Kikuyu (36)
f’aw I \ c v
now, that Rosey is cognizant of the fact that a word such as has a high front vowel in the second syllable and suppose that reflected in the word’s lexical entry by means of a floating/i/ to the floating extra low tone of Cements and Ford’s analysis tone shift). Thus, the lexical entry for monkey will be (36):
monkey:
mu + i
This effect can be achieved extremely easily by simply having a lexical redundancy rule to the effect that, if the second syllable of the target has a high front vowel, then a floating /i/ is added to the lexical entry. So far, we have ascribed to Rosey a lexicon populated by monosyllabic roots. However, we have seen that her productions of polysyllabic targets are all disyllabic. In fact, Rosey has lexical redundancy rule 37: (37)
All polysyllabic
words take the form CVCV.
342
A. SPENCER
Here “word” could as well be replaced by “morpheme” since, 1 take it, Rosey’s output system does not countenance word-formation rules and, hence, all words will be monomorphemic. Statement 37 will entail that all of Rosey’s productions of polysyllabic targets will be disyllabic, given a further condition that every vowel is the nucleus of a syllable. This, again, is the unmarked option as far as one can tell. There is now a difficulty. For we have yet to explain how it is that Rosey’s output takes the form not just of CVCV disyllables but of reduplicated disyllables. I claim that, unless other rules intervene, a “syllable harmony default rule,” 38, operates. This rule, I suggest, is common to many immature phonologies and represents, in some sense, an unmarked option. Indeed, it can be regarded as the simplest possible way of pronouncing a polysyllabic output. (38) Syllable
harmony
default
rule: syllable
2 is a copy of syllable
I.
Before we see how 38 operates note that on the universalist assumptions we have been adopting Rosey’s lexical entry will also contain information that each CV unit is a syllable (a unit of representation made available by universal grammar). Thus, we can refine 32 and 33 to 39 and 40. and 37 to 41: (39)
(40)
bottle:
flower:
b
I-I
I
I
f
a
I
I
To obtain the pronunciation of an actual word we associate the entry 39 and 40 with the CV skeleton 41 (as in 42a), then apply the syllable
PHONOLOGIC
DISABILITY
343
harmony default rule, 38, (as in 42b). Notice that this rule is stated at the level of syllables, which it reduplicates in a fashion similar to that suggested by Marantz for YidinY. (42)
a.
u
u
A 1I cvcv fa
vv u
CT
v CL
u
u
ilh fafa
IIII cvcv VV 0 u vF It is somewhat difficult to adequately represent the fact that the syllable harmony default rule operates at the level of syllables, while other rules operate at other levels. This is simply due to the inefficiency of two dimensional paper as a medium for representing multidimensional vectors. Formally, these processes can be described readily in terms of sets of representations with associated index, indicating the dimension or level at which the representation resides. Of much greater significance is the fact that, in a sense, operations on syllables are maximally simple/unmarked (at least from the developmental perspective), since syllables represent the smallest pronounceable prosodic structure (in the sense of Selkirk, 1980) and, thus, can be thought of as the simplest articulatory unit. In part, this may explain why syllable reduplication is so common in early acquisition (though it doesn’t explain why it should be so rare in adult phonologies). Left to its own devices, the syllable harmony default rule will apply to lexical items such as 36, mu + i, to produce “mumu.” This, however, is not what Rosey says. The reduplication rule, 38, must therefore be blocked for just these entries. Suppose that the floating /i/ is associated to an appropriate slot (i.e., a V slot in the skeleton) by 43, a rule specific
A. SPENCER
344
to Rosey. Rule 43 can be compared with the initial tone association rule of Kikuyu, which associates the first tone in the tone melody with the second tone-bearing element (rule 25). I suspect that in our case a universal principle can be invoked to the effect that, if a specific element of morphemic representation can be associated to a melody-bearing element in a tier, then it musr be so associated before any default rules, such as 38, apply. I shall not give detailed justification of this supposition here, however. If the suggestion is correct, no stipulation of the form (43) need be made in the description of Rosey’s phonology. In any case, the derivation of the entry for ‘monkey’ will proceed as in (44): (43)
Associate
i to V
Notice that this derivation cannot proceed further, say, by association of the m of the first syllable with the consonant slot of the second syllable, because this would violate the injunction against crossing association lines (rule 24b; cf. 22). Thus, we cannot have (45):
The derivation given in 44 leaves us with the form ‘mu% where C represents an untilled consonant slot. The syllable harmony default rule is unable to apply because its structural description is not met. All that remains is to account in a principled way for the appearance of a palatal consonant (variously [1], fi] or [Xl) in the C position. For. this purpose I assume a consonant default rule, 46: (46)
There
is a consonant
default
tier occupied
by the consonant
X, etc.
PHONOLOGIC
345
DISABILITY
Rule 46, like 38, is a default rule (i.e., a redundancy rule operating effectively as an output condition on well-formed surface representations). Again, it will apply (by universal principle or by stipulation) only when all other specific rules have applied (such as the floating /i/ association rule 43). It will apply after 38. The upshot of this discussion is that the form ‘m&i will be converted into [muAil by 46, in conjunction with association principles, as desired. Furthermore, we can also account for otherwise puzzling cases, such as the forms for ‘elephant’ and ‘indian’, [Xoii] and [Xtii], respectively. The derivation of [Adi] is given in 50 and 51. Rule 50 shows how the underlying representation is derived (i.e., the output lexicon entry), and 51 shows how the rules imputed to Rosey generate her output form without further stipulation. The derivation of the form for ‘indian’ is left to the reader as an exercise. (50)
(51)
c+i
3
E >>
I ‘\
i f \
circv
3
i
E
I I
3
cvcv
‘5
c
i
I CVCV
I
VIi
*
3
I(cAi
The ii-tier in 51 is on the other side of the CV skeleton from the morphemic entry “E + i” to show that it is on a separate tier. Phonologically, 79 0 “’ 3, and “X” are all on their own separate tiers. An accurate porl, 1, trayal of this would require three-dimensional paper. Left unexplained from Table 1 is the form [be’be] for “aeroplane.” Given the analysis so far, one might expect [&&] here. However, notice that, likely as not, the starting point for this lexical entry will be the variety “plane” and that this form can be expected to give exactly the result found (I am indebted to Martin Ball for pointing this out to me). Such a derivation also may explain the anomalous stress pattern on this production.
346
CONCLUDING
A.SPENCER
REMARKS
I have presented a phonologic analysis of a small corpus of data from a language delayed child. The data appear superficially puzzling and idiosyncratic (though, as Grunwell notes in her discussion of this case, 1982, all the processes exhibited by Rosey are found somewhere in the literature on normal language development). However, with the techniques of nonlinear phonologic theory and, in particular, recent applications of autosegmental phonology, we have been able to discern patterns that, plausibly, represent universal processes of development and bear an interesting correspondence to principles of markedness, often appealed to in the analysis of adult phonologies. Although these processes at first sight may appear exotic, they all correspond to processes that have been hypothesized for adult phonologies. The notion of CV skeleton is very fruitful, especially in solving descriptive problems related to morpheme structure redundancies often referred to as “conspiracies” (Kisseberth, 1970). Reduplication is widespread, both in child language and in adult phonologies. Although syllable reduplication is unusual in adult phonologies there are good grounds for believing that in child language it has an important organizing role, viz., that of allowing the child to articulate polysyllables at minimal computational cost. The existence of segments that reside on separate tiers is motivated in McCarthy’s analysis of Semitic and in many analyses of prosodic phenomena (e.g., nasal and glottal prosodies, harmony systems, and so on) and one interesting variety of such an autosegment, the floating tone, allows sense to be made of otherwise utterly inexplicable phenomena in Kikuyu, under Clements and Ford’s analysis. Even the default consonant, to which I appeal, is not especially exceptionable: A number of languages (German, Czech, and Arabic, for instance) have a rule adding a glottal stop to any word beginning with a vowel, in effect, a default glottal stop. In other words, Rosey’s speech is largely what one would expect from a phonologic system that made use of universal principles of phonologic organization, but that did not take much heed of the language-specific rules and representations of the target, English. At the same time, the analysis sketched here makes minimal though plausible assumptions about psycholinguistic mechanism. By linking the phonologic realization rules to forms in Rosey’s own internal lexicon, specifically an output lexicon, allowance is made for possible discrepancies between output lexical entries and input entries (that is, those used in comprehension). This may be seen as a significant departure from the model presented in Smith (1973), which is generally taken to be a paradigm for generative analyses of acquisitional data. However, very little of substance is changed. We must still assume some sort of mapping from the adult form to the child’s stored form and this mapping, as I have indicated,
PHONOLOGIC
347
DISABILITY
will be achieved by means of rules and representations made available to the child by universal grammar. Smith’s own analysis could easily be recast in the same terms without altering his conclusions (and, in particular, without committing him to the view that he, like I, strongly repudiates, that the child is operating with its “own system,” viewed as an autonomous system of contrasts independent of the adult’s). Indeed, following Macken’s (1980) critique of Smith’s original claims, Smith has conceded (1978) that perceptually-based alterations in the child’s lexicon must be responsible for some part of the patterning of data he discovered in his own informant’s speech. The view of the output lexicon outlined here can be regarded as simply an extension of such a perceptual filter (though one that possibly has a fairly rich internal structure, derived from universal grammar). I have suggested ways in which the extremely indirect link between competence theories (i.e., phonologic theory in this case) and performance theories (the theory of language acquisition in the present discussion) might be realized. Rosey is ascribed knowledge of universal grammar, but her ability to compute the sets of rules and representations required to generate English phonetic forms falls short of the norm. The result is a phonologic system that, while conforming to universal principles, comports rather poorly with the target system. The framework adopted here allows us a vision, to echo Gandour (1981), of the true underlying dimensions of Rosey’s phonologic disorder. The repercussions for diagnosis and treatment of phonologic disability are self-evident. I am grateful to several of the participants at the Third National Conference of Speech Therapy Lecturers for their helpful comments. I am also indebted to Pam G-unwell for providing me with a copy of G-unwell and Pletts, 1974, unpublished manuscript.
REFERENCES Anderson, S. R. (1974). The Organization of Phonology.
New York: Academic
Press. Anthony, A., Bogle, D., Ingram, T. T. S., McIsaac, M. W. (1971). Edinburgh Articulation Test. Edinburgh: Churchill Livingstone. Chomsky, N. and Halle, M. (1968). The Sound Pattern of English. New York: Harper and Row. Clements, G. N. and Ford, K. C. (1979). Kikuyu tone shift and its synchronic consequences. Linguistic Inquiry 10: 179-210.
Donegan, P. J. and Stampe, D. (1979). The study of natural phonology. nsen, D. H. (ed.), Current Approaches Indiana University Press.
to Phonological
Fee, J. and Ingram, D. (1982). Reduplication velopment.
J. Child Lang. 9:41-54.
In DinTheory. Bloomington:
as a strategy of phonological
de-
348
Gandour,
A. SPENCER
J. (1981). The nondeviant Dis. 14: 1l-29.
nature of deviant phonological
systems.
J.
Commun.
Goldsmith,
J. (1979). Autosegmental
Phonology.
New York: Garland
Goldsmith,
J. (1981). English as a tone language. nology in the 1980’s. Gent: Storia.
In: Goyvaerts,
Press.
D. (ed.), Pho-
Grunwell, P. (1981). The Nature of Phonological Disability in Children. London: Academic Press. Grunwell, P. (1982). Clinical Phonology. London: Croom Helm. Grunwell, P. and Pletts, M. M. K. (1974). Therapeutic a case study. (Unpublished manuscript.) Halle, M. and Vergnaud, 1:83-105.
guidelines from linguistics:
J-R. (1980). Three dimensional
phonology. J. Ling. Res.
Kahn, D. (1979). Syllable-Based York: Garland Press.
Generalizations
Kenstowicz, M. J. and Kisseberth, New York: Academic Press.
C. W. (1977). Topics in Phonological
Kisseberth,
C. W. (1971). On the functional
in English
Phonology.
New
Theory.
rules. Linguistic
unity of phonological
Znquiry 1:291-306.
Lorentz, J. P. (1976). An analysis of some deviant phonological rules of English. In: D. M. Morehead and A. E. Morehead (eds.), Normal and Dejicient Child Language. Baltimore: University Park Press. Macken, M. A. (1980). The child’s lexical representation: pickle” evidence. J. Linguistics 16:1-17.
The “puzzle-puddle-
McCarthy, J. (1979). Formal Problems in Semitic Phonology Bloomington Indiana University Linguistics Club.
and Morphology.
McCarthy,
morphology.
J. (1981). A prosodic theory of nonconcatenative
Lin-
guistic Inquiry 12:373-418.
Marantz,
A. (1982). Re reduplication.
Linguistic fnyuiry 13:435-482.
Selkirk,
E. 0. (1980). The role of prosodic Linguistic Inquiry 11:563-605.
categories
Smith, N. V. (1973). The Acquisition of Phonology. versity Press. Smith, N. V. (1978). Lexical representation (Revised forum lecture.)
in English word stress.
Cambridge:
and the acquisition
Cambridge
Uni-
of phonology.
van der Hulst, H. and Smith, N. (1982). An overview of autosegmental and metrical phonology. In: van der Hulst, H. and Smith, N. (eds.). The Structure of Phonological Representations, Part I. Dordrecht: Foris.