Phonological Features of Tone Author(S): William SY. Wang Source

Home , Tone letter, Yuen Ren Chao

Phonological Features of Tone Author(s): William S-Y. Wang Source: International Journal of American Linguistics, Vol. 33, No. 2 (Apr., 1967), pp. 93-105 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/1263953 Accessed: 31/05/2009 22:36

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=ucpress.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to International Journal of American Linguistics.

http://www.jstor.org International Journal of American Linguistics

VOLUMEXXXI TT April 1967 Number 2

PHONOLOGICAL FEATURES OF TONE1 WILLIAMS-Y. WANG

UNIVERSITY OF CALIFORNIA, BERKELEY

1. General discussion ing the literature on languages of this type, 2. Tone features and segmental features our attention is particularly drawn to three 3. Presentation of phonological features linguistic areas: (i) certain clusters of Ameri- 4. A proposed set of tone features can Indian languages, (ii) the vast majority 5. Redundancy conventions of African languages, and (iii) almost all of 6. Phonetic interpretation the languages of the Sino-Tibetan family 7. Marking conventions together with many neighboring languages 8. Tone circle in Min of Southeast Asian. Typically, tone systems of areas (i) and (ii) differ from those of area 1. 'Tone languages',2 in a broad sense, are (iii) in several ways. found in most parts of the world. In examin- One point of difference is in the use to which tones are put. In languages of area 1 Versions of this paper have been presented to tones are almost used lexi- the at and (iii), exclusively linguistic groups Berkeley, UCLA, with no correlation with the Honolulu. Work discussed in this paper is sup- cally, syntactic ported in part by the Office of Naval Research. I or morphological aspects of the language. am indebted to W. L. Ballard for his assistance in There are exceptions, of course, such as the collecting and systematizing the basic data on tone breathy fall-rise tone in Vietnamese which systems upon which the present discussion is is 'sometimes used to refer and for several discussions on anaphorically based, profitable back to some noun or nominal matters of interpretation. "key" expres- 2 The most comprehensive investigation of tone sion in what has gone before',3 or the modi- languages to date continues to be K. L. Pike's fied tones in several Chinese dialects which book of 1948, though new data have led to criti- serve a variety of connotative as well as cisms of some of Pike's assumptions; e.g. W. E. minor functions.4 in the and tonal syntactic Indeed, Welmers, Tonemics, morphotonemics, dialect there are two dozen or so morphemes, General Linguistics 4.1-9 (Spring, Peking 1949). A lucid discussion on the range of the term morphemes which change grammatical cate- 'tone language' by James D. McCawley in a paper gory according to tone. But these uses are entitled What is a tone language? was presented marginal when they are compared to the ex- to the Linguistic Society of America, August, 1964, tensive load that tones carry in the declen- in which he out that attempts at correctly pointed sional and of typology of this sort should be based primarily on conjugational morphology the abstract form of phonological rules which the many languages in America and Africa, as different tonal structures require. He argued con- exemplified in sections 3.11.2 and 3.11.3 of that with the so-called vincingly languages pitch Nida's Morphology,5 and documented abun- accent, e.g. Japanese, though frequently grouped with tone languages, are much more similar phono- logically to certain non-tone languages. Indeed, it 3 Eugenie J. A. Henderson, Tonal exponents of is an open question whether the distinction be- pronominal concord in Southern Vietnamese, tween these two types of accent can be given any Indian Linguistics 22.86-97 (1961). phonetic foundation. Although this distinction has 4 K. P. K. Whitaker, A Study of the modified played a prominent role at least since Karl Verner tones in Spoken Cantonese, Asia Major 5.9-36, used it as part of his famous historical thesis, our 184-207, (1956). understanding of the physical basis of this distinc- 6 Eugene A. Nida, Morphology, 2nd ed., Ann tion has not advanced much in the past century. Arbor (1949). 93 94 INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS VOL. XXXIII dantly in the literature which deals with area (iii), however, is by paradigmatic re- these languages.6 placement. Characteristically, tone x is re- The tone paradigms of languages of area placed by tone y when it is within some (iii) are typically more complex. If we count linguistic environment, and it is irrelevant each distinct pitch shape in citation mono- whether tone y is present elsewhere in the syllables as a tone, then paradigms of 6 or 7 sequence of tones. Frequently the phono- tones are quite common. According to a re- logical environment in which tone x occurs cent study, Cantonese may have as many as is also irrelevant for the sandhi. Some very 10 tones.7 On the other hand, although a few complex situations of paradigmatic sandhi American Indian languages also appear to are found in the Min and Wu dialects of have complex paradigms, languages of areas Chinese, an example of which is given in 8 (i) and (ii) in general do not have as many of this paper. distinct shapes. Most languages of these two areas have simply two or three noncontour 2. Recently I have examined a large num- tones; a relatively small number of these ber of tone languages, mostly of area (iii), languages have contour tones in addition.8 with a view towards constructing a set of Yet another point of difference can be phonological features of tone. These features seen in the way tone sandhi operates. The are proposed here as an addition to a general sequence of tones in many Bantu languages, theory of phonology. They are designed to for example, undergoes a form of sandhi that complement the two dozen or so features is essentially by syntagmatic displacement, which are currently used in the phonological i.e., a kind of tonal 'musical chairs' in which classification of segmental sounds. each syllable receives its tone from its Tone features, of course, are not com- (usually left) neighbor. The sandhi in the pletely independent of the segmental fea- so-called terrace-level tone languages of West tures. They have a particularly close rela- in that Africa is also syntagmatic the sense tion, synchronically and diachronically, with value of an unaccented is the pitch syllable features which are controlled primarily at predicted from the value of the usually pitch the larynx, e.g. voicing, aspiration, glottali- left neighbor.9 Tone sandhi in of languages zation, length, breathiness, etc. This relation 6 See, for example, Robert E. Longacre, Trique is easily understandable since the primary tone morphemics, A L 1.4.5-42 (April, 1959); A. Meeussen, Syntactic tones of nouns in Ganda: a preliminary synthesis, Linguistic Research in this tonal phenomenon within the broader frame- Belgium 77-86, Universa Wetteren, Belgium work of Niger-Congo languages, Lg 37.294-308, (1966). 1961. These African languages are terrace-level 7 Fd-BangZong, On the split of the Yin-ping languages of the descending variety in that the tone in Cantonese(in Chinese),Zhongguo Yuwen pitch of the voice characteristically progresses 132.376-89(1964). from high to low. Eunice Pike recently called 8 Some especially challengingcases are these my attention to a terrace-level language of the MexicanIndian languagesof the Oto-Manguean ascending variety, i.e., the Acatlan dialect of family:Robert E. Longacre,Five phonemicpitch Mixtec, where the voice pitch may be raised an levels in Trique,Acta Linguistica7.62-82 (1952); indefinite number of steps theoretically, by the FrankE. Robbins,Quiotepec Chinantec syllable repeated occurrence of the tone feature 'step-up'. patterning, IJAL 27.237-50(1961); William R. I am grateful to Miss Pike for showing me an Merrifield,Palantla Chinantec syllable types, unpublished paper on this subject, which she A L 5.5.1-16(1963). co-authored with Kent Wistram. The tone features 9 For a clearlypresented example of a terrace- 'step-up' and 'step-down' are not discussed in level tone languagein Ghana,see Paul Schachter, the present study since they call into play a type Phonetic similarity in tonemic analysis, Lg of phonological formalism that remains to be de- 37.231-8(1961). In the same issue of Language, veloped should they prove necessary at the sys- H. A. Gleason,Jr. gives additionaldiscussion of tematic phonemic level of representation. NO. 2 PHONOLOGICAL FEATURES OF TONE 95 determinant of tone is the rate of laryngeal are sometimes relevant for the initial conso- vibration. nant, sometimes for the nuclear vowel, and Indeed the development of some types of sometimes for the final consonant in various tonal distinctions probably can be explained phonological rules. If we were to add a in terms of oppositions which are originally column of tone features to a phonological segmental.10 As an example, although matrix of segmental features, then it be- Middle Chinese (ca. 600 A.D.) is tradition- comes arbitrary where precisely to insert ally regarded as having four distinct tones, this column. Furthermore, segmental fea- from physiological considerations we know tures are usually not relevant in the various that it must have had eight pitch shapes. types of tone sandhi; that is to say, the inter- Four lower pitch ones appeared only with action of tones in a sequence is independent syllables with voiced initials, the remaining of the nature of the segments which occur only with unvoiced initials. The phonetic dif- with the tones. Phonetically, of course, the ference between the four pairs of pitch domain of the tone is over the entire voiced shapes is intrinsic in that it can be predicted portion of the syllable. From these considera- from the mechanical properties of the speech tions, it is preferable to formalize the tone mechanism." Like the pitch difference be- features differently from the segmental fea- tween English /pin/ and /bin/ it is not tures and regard them as features of indi- phonemic even though it is perceptually vidual syllables. In Chinese, this is almost quite noticeable. When the voicing distinc- equivalent to marking these features on indi- tion was obliterated through historical vidual morphemes since almost all mor- change, as evidenced in most Chinese dia- phemes can be represented as single phono- lects today, the phonetic differences in the logical syllables. For languages with many pitch shapes in certain cases became dis- polysyllabic morphemes, it appears that the tinctive. syllable may need to be given independent It is more difficult to provide a phonetic theoretical status in order for it to bear the motivation for the extrinsic relationship that phonological features of tone. has been observed between tones and vowel height. This is reported to exist subpho- 3. The features of tone, much as other nemically in Peking Chinese, where certain phonological features, are selected for the mid vowels have a lower articulation with dual purpose of (i) describing the alterna- two of the four tones.'2 Much more striking, tions, both synchronic and diachronic, that however, are the morphophonemic alterna- are found in language, and (ii) providing the tions in Foochow Chinese, where whole sets abstract linguistic basis from which physical of the vowels are raised and diphthongs phonetic interpretations are made. As a monophthongized when certain tone sandhi consequence of the lack of a perfect match rules are applied.'3 between these two functions a universal set In languages like Chinese the tone features of phonological features usually contains a 10A broad-gauged study of this phenomenon is certain amount of internal redundancy in presented by A. G. Haudricourt in his Bipartition that some combinations of the feature speci- et tripartition dans les systemes de tons, Bulletin fications are not permitted in principle. State- la de Paris 56.163-80 de Societe de Linguistique ments regarding permitted feature combina- (1961). tions are conventions'. The 11A discussion of intrinsic pitch variation is 'redundancy given by Peter Ladefoged; see p. 42 of his A Pho- term 'convention' refers to language-uni- netic Study of West African Languages, Cam- versal statements while 'rule' refers to bridge University Press, 1964. language-specific statements. 12 Lawton Hartman, Lg 20. 28-42 (1944). 13 Ya-Xifi Lan, Foochow phonology (in Philosophy, National Taiwan University (Decem- Chinese), Journal of Literature, History, and ber 1953), 241-331. 96 INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS VOL. XXXIII

Of the permitted combinations of tone fea- factors'6 (emotive ones like speaker's mood tures, it appears from the languages ex- and voice volume; cognitive ones like in- amined that some combinations are favored tonation, contrastive accent; intrinsic ones over others, as is the case with segmental like tongue height and degree of supraglottal features. The present theory attempts to closure; etc.). What is said below about the incorporate this aspect of tone systems by phonetic value of the tone features must be 'marking conventions', using the notion of understood as descriptions of idealized pat- 'markedness' as it has been recently refined terns distilled out of speech events rather by Halle and Chomsky.14As will be seen in than, say, what can be measured directly 7, this notion allows us to state in a precise from narrow band spectrograms. Using the way the relative complexity of different 'tone letter' notation proposed by Y. R. types of feature combinations. Within such a Chao,17we present below an illustrative set framework the combinations which tend to of 13 tones as shown in Table I, together be favored are the ones which minimize the with a set of six features. Certain features total complexity of the paradigm. which are primarily associated with segmental oppositions, such as the laryngeal 4. The rapid advances in physical pho- ones mentioned above in 2, may also par- netics during the past two decades have ticipate in distinguishing tones in phono- made it increasingly clear that there is fre- logical representations. quently a considerable discrepancy between As can be seen from Table I, the -CONTOUR tones tones 1 are divided into a linguist's phonetic interpretation of a (i.e. through 5) five levels by the features HIGH, CENTRAL and speech wave and what the speech wave ac- MID. The +CONTOUR tones either in one tually contains.16 The parameter of voice go direction or in two directions; the former pitch varies according to a variety of other (i.e. tones 6 through 9) are either +RISING 14The notion of 'marked' and 'unmarked' has or +FALLING, while the latter (i.e. tones 9 been used by N. S. Troubetzkoy in connection through 13) are both +RISING and +FALL- with what he called 'privative oppositions', i.e. ING. The bidirectional tones are further oppositions which are characterized by the divided by the feature CONVEX. presence or absence of certain features such as tones in some have In his de Although languages voicing, nasality, rounding. Principes been transcribed with as as nine Phonologie, the notion was mentioned briefly but many pho- not developed. Recently Halle and Chomsky in- netic levels of pitch, I have not been able to tegrated this notion into their framework of find any language that has more than five generative phonology in an attempt to capture the distinctive pitch levels, which turns out to nature of see aysmmetric phonological features; be exactly the maximum number that Chao their Sound Pattern of English, now in press with in In Harper and Row. For reasons which have mostly allows for his insightful notation. fact, to do with the physical constraints on the speech the only clear cases I know of where there mechanism, the two specifications of a phonologi- are five distinctive pitch levels are some cal feature frequently are not symmetric with each Black Miao recorded Kun other. Furthermore, such asymmetry may be languages by context-sensitive, e.g. the dependence of voicing Chang and by F. K. Li. These are the only on consonantality, or it may be context-free, e.g. sounds are generally unglottalized. This observa- 16An attempt has been madeto categorizethe tion is equivalent to the one I am making regard- variouslinguistic functions of voice pitch in the ing certain combinations of feature specifications referencegiven in footnote11. A pioneeringpaper being favored over others. in this area is Y. R. Chao'sTone and intonation 15 For a discussion of a phonological entity in Chinese,Bulletin of the Institute of History that has been particularly elusive to phonetic and Philology (AcademiaSinica) 4.121-34(1933). investigation, see my Stress in English, Language 17Yuen-Ren Chao, A system of tone letters, Le Learning 12.69-77 (1962). Maitre Phon6tique 45.24-27 (1920). + +

NO. 2 PHONOLOGICAL FEATUR ES OF TONE 97 languages,18 in other words, which exploit distinguished from the level tones by some the feature MID of Table I. According to other feature.21 The recognition of contour Chang,"9 the Yong-C6ng language has a tones is crucial in the analysis of certain rising tone and two falling tones in addition types of tone systems if we are to capture to five level tones; the level tones may be all and only the consistent characteristics in exemplified by mo5 name, me4 eye, mjo3 ear, the phonological structure. If it turns out ko2 road and mjol fish. The Tai-Gong lan- that FALLING is the only relevant feature for guage20 has a falling tone and two rising a particular tone, then over-differentiation tones; its level tones may be exemplified by would only lead to chaos when we try to la5 short, la4 classifier, la3 cave, la2 to move mark what pitch level the tone falls from or away, and laWcandle. In these examples, what level it falls to. This would be like try- lower numbers indicate lower pitch levels. ing to mark, say, seven degrees of aspiration The above observation on the number of for English stop consonants. As Gleason re- TABLE I TONES AND THEIR FEATURES 1 2 3 4 5 6 7 8 9 10 11 12 13 IYY" 1-1r --_I _ n _ CONTOUR - - + + + + + + +

HIGHHIGH+ ' + -- +-1- - - 4-+ - - + - + - - 4-+ --

CENTRAL + + - -

MID - - - , - +- - -? --- -

RISING RISING - - - - + + + +

FALLING - - - + + + + + + L______

CONVEX - - - . . .- - . - - - I- - ++ L-_ - - ______levels is only valid, of course, if we recognize marked in connection with Doke's attempt in some cases contour tones in the paradigm. to describe Zulu tones with nine levels, "the Thus in Trique, only four phonemic levels phonemic contrasts in the pitch system in- (the tone features HIGH and CENTRAL) ap- volve other dimensions than mere level;" it pear to be necessary, since the fifth level comes as no surprise that the nine-level sys- goes only with a contour tone which can be tem of transcription proved 'utterly impos- sible to teach to Zulu students'.22 Doke's 18 Some Tai languages of Gtiizhou are also re- case of nine levels a rather ex- ported to have five level cf. of the pitch provides tones; Report treme of in Survey of Biyi Languages, Peking, 1959 (in example phonetics running wild Chinese). I have not studied whether any of these the absence of theoretic constraint. None- levels can be predicted by other features which are theless, I suspect that many tone descrip- independently distinctive, such as syllable length tions of the languages of areas (i) and (ii) or consonant voicing. can be re-examined with to see if new 9 Kun Chang, Some questions concerning profit Miao-Yao tones (in Chinese), Bulletin of the regularities can be discovered when contour Institute of History and Philology (Academia tones are added to the paradigm. Sinica) 16.93-110 (1947). 20 The data here were gathered by F. K. Li some 21 In Longacre's analysis a phonemic fifth level years ago and reported by Julia Kwan in a Uni- is posited but no contour tones are used; the versity of Washington M.A. thesis (1966), Phon- reference is given in fn. 8. ology of a Black Miao Dialect. 22 H. A. Gleason, Lg 32.570 (1956). 98 INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS VOL. XXXIII

Related to the difficulties due to overdif- 52/51/42 are obviously nondistinctive for ferentiation within single tone paradigms is any purpose, and are probably more spurious the situation we encounter when we compare than real. This overdifferentiation is due to tones of different paradigms or different a notational device which, in order to be uni- reports on the same tone paradigm. Here versally applicable, must provide for more again the crucial step is to separate pho- distinctions than any single paradigm can netic differences which are distinctive from have. (In fact, the tone letters allow for 5 those which reflect nondistinctive variations pitch levels at each of three positions, or a within or between speakers, or the habits of total of 125 tones.) the phoneticians who made the transcrip- Our line of reasoning is strengthened when tion. Consider, for example, the data on the we examine lines 9 and 10 above. This kind first eight sites reported in the dialect survey of discrepancy between different reports on of Jiangsu.23For each site the values for the the same site is not uncommon in the litera- four lexical tones, originally given in Chao's ture on Chinese dialects. Here we find the tone letters, are presented below in numeri- same type of variation that in all likelihood cal form: does not represent any consistent or systematic of Sizhou Whether the 1. 213 55 35 52 property speech. 2. 212 55 35 52 seventh tone in this dialect (which is a short 3. 313 55 35 51 tone) rises or not is a matter that needs more 4. 313 55 24 51 refined phonetic study. As for the difference 5. 313 55 24 52 between the two reports on the fifth tone, 6. 213 54 24 52 the of features 7. 213 55 24 51 system phonological proposed 8. 23A 55 24 52 here would predict it to be nondistinctive in principle: 331 is just a slowed down 31. Let us consider the above data in conjunc- Again, we would assume that lines 9 and 10 tion with the tones reported for Suizhou, site represent an identical tone paradigm in spite number 34 covered by the Jiangsu survey. of the phonetic differences reported for Line 9 gives the values of the Sfizhou tones almost each of the seven tones. according to the survey. Line 10 gives the 5. In principle, seven binary features can tone values as they are reported in another distinguish 128 objects. Since we are making recent source.24 use of only 13 tones in connection with these the is considerable. 9. 44 13 52 412 31 5 2 features, redundancy 10. 44 24 41 513 331 4 23 This redundancy can be formalized in conventions which predict certain combinations The lesson from the above data is quite of feature specifications to occur while pro- clear. It is evident from the values given in hibiting other combinations from occurring. lines 1 through 8 that these sites have the Of the total set of redundancy conventions same four tone paradigm, which can be dis- that can be stated in a general theory of pho- tinguished by the same two features, i.e. nology, only a subset is relevant for any par- FALLING and RISING. The only point that is ticular tone paradigm. This relevant set is somewhat uncertain is that the second tone determined by which features in the para- of site 6 is reported to fall slightly, whereas digm are considered to be lexically distinc- we are interpreting it here to be -FALLING. tive in the language. The variations 213/212/313, 35/24 and To illustrate this point, let us refer back to Table I. Let us makethe the unlikely assump- 23 Outline of the Dialects of Province far c Jiangsu tion that all the feature en- and ShAnghGai,ShAnghAi,Nanking (1960). In Chinese. specifications 24 Phonetic Dictionary of Chinese Dialects, closed by the dotted lines are distinctive, Peking (1962). In Chinese. hence yielding a paradigm of 13 tones. Given NO. 2 PHONOLOGICAL FEATURES OF TONE 99 these feature specifications, all of the re- FALLING CONTOUR feature of Table I can + maining specifications 7 + be predicted by the following redundancy conventions:

CONTOUR RISING 1. [--CONTOUR]-- - RISING -FALLING -CONVEX - 2. f[+HIGH]} - [-MID] -A+ \ I +A - CENTRAL] (1) (6) (8) (1) (6) (8) (6) 3. [+CENTRAL] -- [-CONT 'OUR] 4. Peking Chaozhou Canton > {[+RISING] + [+CONTN OUR] [+FALLING]j FIG. 1. Fragments of stone systems 5. [+CONTOUR] -*rF-CENT 'RAL1 L-MID J and RISING,even though no paradigm would 6.[ [-RISING] ever use all three of these. For the purpose -- [-CONVEX] of tone this \[-FALLING]/ describing alternations, particular redundancy is essential. The redundancy that is embodied in the To illustrate, let us take three tones which system of tone features is of three types. have roughly the values of tones 1, 6, and 8 First there is the type that follows directly in Table I. Tones of these values occur in from the phonetic meaning of the features; the Chinese dialects of Peking,25 Canton,26 this type of redundancy can be exemplified and Chaozhou27together with other tones. by convention 1 above. Secondly, redun- But the behavior of these tones in the three dancy arises when we make empirical claims dialects is quite different. In Peking tone 1 about what distinctions are actually used in alternates with 6; in Chaozhou 6 alternates lexical representations at the systematic with 8; and in Canton 8 alternates with 1. phonemic level. For instance, while we allow It can be seen that if the three tones are five pitch levels to be distinctive for -CON- coded as in Figure 1, making use of all three TOUR tones we are only allowing two pitch features, then each alternation can be de- levels to be distinctive for +CONTOURtones, scribed by changing the specification of one as can be seen from convention 5. Such re- feature. In particular, the type of alterna- strictions can always be removed when tion exemplified by the Chaozhsu tones pro- counter-evidence becomes available. vides the crucial evidence for the feature A third type of redundancy comes from CONTOUR. the fact that an identical set of tones may alternate differently in different phonologi- 6. Though subject to small variations from cal systems, a possibility that was high- language to language that have no cognitive lighted in the writings of Edward Sapir. import, the range of voice pitch remains re- This means that a phonological theory must markably uniform across languages. This is be able to provide alternative feature repre- true regardless of how many tones a lan- sentations for the same sounds as those guage has, or whether it has any at all. In sounds 'configurate' differently from lan- general I find it extremely difficult to dis- guage to language. Thus if [n] alternates with tinguish utterances in certain types of tone in one and with conso- liquids language stop 25 nants in then it should have dif- Yuen-Ren Chao, Mandarin Primer, Harvard another, Press 26. ferent in the two University (1948), underlying representations 26 See reference in Footnote 4. languages. These considerations lead us to 27 Y6ng-Ming LI, The Chsozh6u Dialect (in posit the three features CONTOUR, FALLING, Chinese), Shanghai (1959), 15. 100 INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS VOL. XXXIII languages (e.g. Mandarin) from those in a straightforward phonetic basis for inter- non-tone language (e.g. English) by just preting the features of tone. examining the pitch measurements of these A phonetic interpretation of the level fea- utterances. In languages with numerous tures is given schematically in Figure 2. The syllable-sized tones one would expect that pitch median is designated 0 and the range the utterances would have more complex is taken to be from -5 to +5. The effect of pitch contours, i.e. directional changes in the feature HIGH, viewed this way, is simply pitch occurring over shorter time spaces, but to provide a sign, '+' or '-'. The feature so far this problem has not been studied in a CENTRAL restricts the range to ?t3. In systematic way.27a actual speech, of course, the phonetic No matter how many tones a language boundaries are by no means as clearcut as has, the voice pitch traverses approximately suggested in this illustration. the same overall range. The difference re- There are advantages for considering the sides in how each pitch value is interpreted tone features to be binarily valued, as pro- vis-A-vis the particular tone paradigm. The posed here.29 The tone features can be re- greater the number of distinct tones in the garded as having theoretical status com- paradigm, the narrower the phonetic range parable to that of the segmental features of each tone would be. Furthermore, the (e.g. VOICE, NASAL, STRIDENT, etc.) and can phonetic differences among tones, in terms be manipulated by the same machinery of of pitch level, slope of contour, duration, rule application that has been developed on etc., are characteristically more pronounced the basis of segmental features, including the in deliberate speech and are reduced with use of the 2-valued variables. An example increased tempo.2 In these respects the pho- will be provided in Section 8 that makes netic variation of tone is essentially similar special use of phonological variables in solv- to that of vowel articulation, since both ing a problem of tone sandhi, though these voice pitch and vowels vary along physical variables were first conceived in connection dimensions which are continuous, as against with assimilation and dissimilation of seg- certain dimensions of consonant articula- mental features.30 tions which are discrete. An alternative would have been to propose It will be noted from Table I that a maxi- an n-ary system where a feature of LEVEL mum of five level tones are posited. As ob- would have five possible specifications, i.e. served earlier in this paper, systems which 1, 2, 3, 4, and 5. The specifications for this have five level tones are extremely rare, so feature would not be commensurate with the the feature MIDis rarely used. The feature specifications of the other tone features and HIGH specifies whether the pitch level of a the segmental features, many of which are tone is above or below some idealized pitch obviously binarily valued. The theoretical median. The feature CENTRAL specifies 29 Peter Ladefoged has expressed the view that whether the pitch level of a tone is close to the features should not be binarily-valued in a the median. These remarks, taken in con- paper entitled An attack on the number two, junction with the observations made in the presented to the Acoustical Society of America, in preceding paragraphs, should provide a June 1966; see UCLA Work Papers Phonetics No. 4 (July 1966), 7-9. I believe his arguments 27a In some languages with only noncontour against binary tone features are met in the present tones, especially those of the terrace-level type, paper. the pitch contours assume a remarkable step- 30 To my knowledge, variables were first intro- function appearance. duced by Morris Halle in his A descriptive 28For a more detailed discussion of these convention for treating assimilation and dissimila- matters, see K. L. Pike, Operational phonemics in tion, Quarterly Progress Report No. 66 of the relation to linguistic relativity, Journal of the M.I.T. Research Laboratory of Electronics (1962), Acoustical Society of America 24.618-24 (1952). 295-6. No. 2 PHONOLOGICAL FEATURES OF TONE 101

Number low Pitch Values high of Pitch levels -5 -4 -3 -2 -1 0 +1 +2 +3 +4 +5

2 levels -HIGH +HIGH

-HIGH +HIGH 3 levels -CENTRAL +CENTRAL -CENTRAL

-HIGH -HIGH +-HIGH +HIGH 4 levels -CENTRAL +CENTRAL +CENTRAL -CENTRAL

-HIGH -HIGH +HIGH +HIGH 5 levels -CENTRAL +CENTRAL +MID +CENTRAL -CENTRAL -MID -MID -MID -MID

FIG. 2. Illustration of the phonetic range of noncontour features. machinery of rule application cannot be high tone becomes the mid tone, the mid easily extended to this 5-valued feature. tone becomes the low tone, and the low tone We face a more serious difficulty when we becomes the falling tone in a given syntactic try to relate the specifications of LEVEL to environment. Let us consider this Gaoxiong their phonetic range in a fashion that is case together with a hypothetical 4-tone case illustrated in Figure II. With the features which has similar pitch lowering. illustrated there, the principle is self-evident Using the 5-valued feature to represent whereby the phonetic range of each addi- the three Gaoxiong tones as 5 LEVEL, 3 tional tone is marked off from the middle, LEVEL, and 1 LEVEL, the sandhi would have compressing the range at the two extremi- to be described by a sort of rule of phono- ties. To achieve the same desired effect, we logical arithmetic that subtracts 2 from each would be required to stipulate additionally of the first two tones. It is not clear what that if a language has two NONCONTOUR should be done with the 4-tone case since tones, then they are 1 LEVEL, and 5 LEVEL the tones are not 'equidistant' from each (or, perhaps, 2 LEVELand 4 LEVEL);if it has other. In both cases, we need some ad hoc three NONCONTOURtones, then they are 1 convention for what happens when the re- LEVEL, 3 LEVEL, and 5 LEVEL; and if it has mainder of the subtraction is less than one. four NONCONTOUR tones, then they are These difficulties do not present them- 1 LEVEL, 2 LEVEL, 4 LEVEL, and 5 LEVEL. selves within the binary framework proposed This difficulty of premature commitment to here. To describe a 3-tone paradigm that phonetic detail is not unlike the problems undergoes pitch lowering, the mid tone connected with overdifferentiation discussed should be specified as -HIGH and +CEN- in 4. TRAL. (If the paradigm undergoes pitch rais- The merit that has been argued for such a ing, the mid tone should be +HIGH and feature as LEVELhas to do with the descrip- +CENTRAL; cf. the discussion on alternative tion of a certain type of sandhi which system- phonological representations in Section 5 atically lowers or raises the pitch range of above.) The appropriate sandhi rule would each of the NONCONTOURtones. I cannot say be: at present how typical such sandhi is vis-h- [aHIGH] -- vis other observed types. We have such a aCENTRAL case, for instance, in Gaoxiong,31 where the spokenin the Kaohsiungarea, Journalof Taiwan 31Chao-Hui Tung, Phonology of Taiwanese as ProvincialNormal University 9.1-10 (June, 1964). 102 INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS VOL. XXXIII

For the 4-tone paradigm that undergoes are higher pitched than those derived from pitch lowering, the following rule must be forms with unvoiced initials." From the supplemented by a later rule that corrects physiological considerations alluded to in 2, the derivation of the lowest tone: we know that these dialects must have a historical of the aHIGH 1 - IGH 1 undergone flip-flop change sort exemplified in the above rule. Further- fLCENTRALJ L aCENTRALJ more, this change has left a synchronic The strongest evidence in support of the imprint in many Min dialects where words binary features comes from alternations that which had voiced initial consonants have may be called 'flip-flops.' These are cases sandhi forms that are lower pitched than the where, in certain linguistic environments, corresponding words which had unvoiced the high tones become low tones and the initial consonants. low tones become high tones. Such alterna- Such alternations pose a striking prob- tions have been reported for many Chinese lem for our understanding of phonological dialects, as well as for other languages. In change.34Assuming that there are no other some cases these alternations are synchronic; relevant factors, it is difficult to see how a in others, they are deducible only historically historical flip-flop such as one between high by comparing cognates. We have a syn- tone and low tone can be brought about chronic example in the dialect of Chaozhou,32 without the two tones merging with each where it is reported that before a high falling other at some stage of change, if one views tone, (i) high tone becomes low tone, (ii) this type of change as occurring in small. low tone becomes high tone, and (iii) mid cumulative phonetic increments. The pos- tone remains unchanged. These flip-flop al- sibility is always open, of course, for one of ternations which involve just the two ex- the tones to become something else during tremities of the pitch range cannot be easily an intermediate stage to avoid the collision stated with an n-ary framework. Using the course. However, the force of this explana- features of the present theory, however, they tion is diminished when we see that in many have a natural expression as follows: cases (i) the tones involved are phonetically short at both the initial and terminal stages [a HIGH] -) [-a HIGH]/ [-CENTRAL [HIGH +LFALLING of the change, making the distinctive use of CONTOUR features unlikely, and (ii) the pres- to such a situation as According Egerod,33 ent morphophonemic alternation is between that of Chaozhou is in the quite wide-spread +HIGH and -HIGH NONCONTOUR tones, southern Chinese dialects. He remarks that making the hypothesis of an intermediate in Middle Chinese tone 3 in the Min dialects, stage unlikely. One could interpret the and in Middle Chinese tone 4 in the Min and flip-flop as occurring in an all-or-none fashion Hakka dialects, "the words which have 34 A discussion of the theoretical developed from forms with voiced initials comprehensive issues involved in phonological change is given by 32 Reference given in Footnote 27. William Paul M. Postal in his Aspects of Phonological Merrifield recently called my attention to the ex- Theory, in press with Harper and Row. The prob- istence of a synchronic flip-flop in Palantla lem of flip-flop changes has been discussed in Chinantec. On p. 9 of his Korean Morphopho- detail by Robert P. Stockwell in a paper read to nemics, Baltimore, 1954, Samuel Martin describes the Linguistic Society of America called Realism a situation that seems to be a diachronic flip-flop, in historical English phonology (December 1964) which is not unlike what happened to the lexical and in some later unpublished papers. For an early accent in Japanese dialects, as he pointed out. Of statement to the effect that phonetic changes are related interest is the observation that some Brit- always by abrupt leaps rather than successive ish dialects of English reversed pitch contour. slides, see Alf Sommerfelt, Note Sur les change- 33See S0ren Egerod, The Lungtu Dialect, ments phon6tique; Bullentin de Socifte de Lin- Copenhagen (1956). The quote is from p. 272. guistique 24.138-41, 1923. NO. 2 PHONOLOGICAL FEATURES OF TONE 103

TABLE II RELATIVE COMPLEXITY OF TONES AS DEFINED BY MARKING CONVENTIONS 1 2 3 4 5 6 7 8 9 10 11 12 13 I q-HI 1 CONTOUR u u U 11 U m m m m m m m m HIGH U + - - + + - + + CENTRAL U u m Dn m u u u u u u u u MID Uu U u UI m u u u__ u__ uut u u u RISING U u U1 u + + + + + FALLING U U u 12 u + + + CONVEX U U Uu iU1 u U u uU uU uU u m m COMPLEXITY 1 1 2 2 3 4 4 4 4 4 4 5 5 over a gradually increasing sector of the TABLE III vocabulary and thereby avoid the concep- TONES OF AMOY HOKKIEN tual difficulties of the dilemma of inevitable Ia Ib II IIIa IIIb IVa IVb collision. Even in this view, however, inter- 7-1 'I - HIGH ++ + - _ 4- esting questions arise in connection with the FALLING __- + + - - effect this change has on intelligibility. RISING 7. Although the redundancy conventions LONG ++ + + + allow us to deduce which feature combinations are permitted by the theory, they do A preliminary attempt at formalizing not indicate which ones are favored. Lan- these preferences for combinations of feature guages with three tones would characteristi- specifications is made in Table II. It will be cally have tones 1, 2, and 5 of Table I, or seen that the 'u' specification (unmarked) perhaps 6, 8, and 5; we would hardly expect always happens to colrespond to the '-' them to have, say, tones 1, 3, and 6 or 1, 10, specification of Table I, and that the 'm' and 11. Languages with four tones may have specification (marked) always happens to tones 1, 2, 6, and 8 or 1, 2, 8, and 13; but correspond to the '+' specification. The hardly ever 1, 2, 3, and 4, or 6, 7, 8, and 9. 'u' specification, which is the favored The situation is not unlike that of segmental specification, does not add to the complexity features where we find that (a) certain of a sound system, whereas the specifications features are exploited more than others, i.e. '+', '-', and 'm' each add one unit to the HIGH more than CONTOUR which in turn is complexity. more exploited than CONVEX (cf. VOCALIC To convert the 'u' and 'm' to '+' and '-' more than STRIDENT, while STRIDENT is more all that is needed is the following general exploited than GLOTTALIZED); and (b) more marking convention which turns 'u' to '-' dimensions are utilized rather than more while its implied inverse convention turns distinctions within a single dimension. This 'm' to '+'. latter point explains why a four-tone para- [U TONE FEATURE] -> [-TONE FEATURE] has some CONTOUR even digm always tones, The assignment of '+' and '-' in Table do through many languages distinguish II is straightforward, since there is no four NONCONTOUR tones. CONVEX among empirical ground for favoring either +HIGH tones are only found in rather complex or -HIGH, or RISING or FALLING. The pref- paradigms containing at least five or six erence for --CONTOUR is justified by the tones. As mentioned above, only very rarely preponderance of languages with only is the feature MID to exploited distinguish -CONTOUR tones. Among the bidirectional five CONTOUR tones, though paradigms con- tones there seems to be a majority of taining more than five tones are not un- -CONVEX tones in the literature I surveyed. common. The bidirectional tones almost always be- 104 INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS VOL. XXXIII come unidirectional tones in sandhi; very of the values they have when pronounced rarely does any tone become a bidirectional as citation monosyllables. The numbers we tone in sandhi.35When a bidirectional tone assign to the tones correspond etymologi- like tone 10 changes, it may become either cally to the four tones of Middle Chinese, tone 6 or tone 8; i.e. tone 10 loses either its believed to have been spoken some 1,500 first half or its second half. In such cases years ago. The letters 'a' and 'b' indicate a one of the following two symmetric con- historical split that corresponds respectively ventions will apply. to unvoiced and voiced initial consonants. In most of the Min 1. [+RISING] -- [U RISING] /[+FALLING] dialects, of which Amoy Hokkien is one, Middle Chinese tone 2 did 2. [+FALLING] -- [U FALLING]/[+RISING] not participate in the split. The content of Table II and the associated In a large class of syntactic environ- marking conventions must be regarded as ments,37the tones undergo sandhi in a way highly tentative at this time. To provide an that is illustrated by Bodman's examples adequate empirical basis for such state- below. In the discussion here we will be ments, much more extensive results on concerned with the five long tones only.

(i) Ia -IIIb- sa: three but sa:kh6 three dollars (ii) Ib -* IIIb E-mfng Amoy but E-miugli-tlam Amoy Hotel (iii) IIIb -> IIIa go five but g6 kak fifty cents (iv) IIIa - II st four but si kak forty cents (v) II - Ia h6u good but h6u ciAq good eating diverse forms of tone alternations are re- A closer inspection of these alternations quired than are now available. Furthermore will show that there is some underlying there are great gaps in our present under- regularity that may be extracted. Indeed standing of the theoretical basis of marking Bodman presents these alternations in the conventions that need to be filled. However, form of a tone circle which can be linearized the exercise in this section does indicate that in the formula: 'n this area of phonological research tone la\ -* features can be treated in essentially the l) -- IIIb IIIa -- II -+ Ia same way as the segmental features. \lb The neutralization of tones la and lb in the 8. The use of the tone features presented sandhi position enables us to restate the here may be exemplified by a particularly above formula in terms of only two phono- interesting phenomenon of sandhi in Amoy logical features. It is understood, of course, Hokkien.36In Table III we present the seven that these features are in the environment lexical tones of this Chinese dialect in terms LONG, as discussed above.

+HIGH -HIGH -HIGH +HIGH +HIGH

-FALLING _ -FALLING _+FALLING_ _+FALLING _ -FALLING_

36 One case that has come to my attention "7A detailed study of the syntactic environ- recently is Zheng, Tone sandhi in the Wenzhou ments in which similar sandhi takes place in a dialect (in Chinese), Zhongguo Yuwen 129.106-52 Taiwanese variety of Min is available in a forth- (1964). coming paper by Robert L-W. Cheng. Roughly 36 Nicholas C. Bodman, Spoken Amoy Hokkien, speaking, the sandhi occurs on all syllables which Vol. 1, Kuala Lumpur (1955), especially 38-41. do not end major syntactic phrases. It seems to NO. 2 PHONOLOGICAL FEATURES OF TONE 105

The regularity in these alternations can have the same specification, then FALLING be clearly seen from the above formulation. will change its specification. Otherwise, HIGH In each alternation only one feature changes will change. These observations lead us to its specification. Furthermore we can pre- discover that the HIGHfeature of the derived dict which feature will change its specifica- tone takes on the same specification as the tion. If the two features of the basic tone FALLING feature of the basic tone, while the FALLING feature of the derived tone takes be general with Chinese dialects that sandhi is on the opposite specification of the HIGH conditioned by syntactic boundaries on the one feature of the basic tone. In all four hand and on the other. For sum, speech tempo examples alternations can be and discussion on this point see Cheng, Mandarin pair-wise captured by phonological structure, Journal of linguistics the single phonological rule: 2.135-58 (1966), especially pp. 150-1; and my Tone 3 in Pekinese, to appear in Journal of Speech and a HIGH HIGH Hearing Research (1967). L: FALLING -a FALLING