1

On the Nature of Changed Tones in Cantonese*

Te-hsin Liu, tehsinliu@ntu.edu.tw

National Taiwan University

Abstract

Yue dialects, i.e. Cantonese, Bobai and Xinyi, have a process whereby a rising tone

replaces the lexical tone of the head noun to derive diminutive forms, referred to as Pinjam

(changed tones) in the literature. Chao (1947) and Benedict (1942) noticed that, in Cantonese,

this derived rising tone has a slightly longer duration than the lexical rising tone. The same

phenomenon is observed in Bobai and Xinyi, where the derived rising tones are longer than

the lexical rising tone (Wang 1932, Ye & Tang 1982). Establishing a correspondence between

the Mandarin diminutive suffix [-ɻ] and the Cantonese high rising Pinjam, Chao (1959) used

mora to describe this additional length, suggesting that the Cantonese mora is a suffix taking

the form of a high tone rather than sound segments. This conjecture, capable of explaining

the additional length associated with the Pinjam, is contrary to current theories according to

which tones, being suprasegmental objects, have no temporal basis of their own. How to

solve this paradox?

Following O’Melia (1939) and Whitaker (1956) according to whom the additional

length is to compensate the elided diminutive suffix [¯in25] in Bobai, a more conservative

dialect compared to Cantonese, we claim that tones, rather than vowels, lengthen in order to

fill the vacuum left by the elision of the neighboring syllable. A conjecture based on the

segmental compensatory lengthening will encounter one problem: if the additional tonal

duration had to be explained by the compensatory lengthening of vowels, no change in length

would be expected to occur in closed syllables. Now, the additional length is observed in

both open and closed syllables in Pinjam. Consequently, it is tone that lengthens under

syllable elision, not vowel. In other words, the vowel lengthens under the pressure of the tone,

not the tone under the pressure of the vowel.

Subject keywords:

Diminutivization, changed tones, compensatory lengthening

* This paper is a chapter from my doctoral dissertation, and was presented at Berkeley Phonetics & Phonology

Forum at UC Berkeley in April 13, 2009. I would like to thank Joaquim Brandão de Carvalho, Chen Zhongmin, Samuel Cheung, Larry Hyman and Laurent Sagart for their comments and suggestions. Any opinions and shortcomings in the paper are those of the author.

2

1. Introduction

Yue dialects, i.e. Cantonese, Bobai and Xinyi, have a process whereby a rising tone

replaces the lexical tone of the head noun to derive diminutive forms, referred to as Pinjam

(changed tones) in the literature. Chao (1947) and Benedict (1942) noticed that, in Cantonese,

this derived rising tone has a slightly longer duration than the lexical rising tone. The same

phenomenon is observed in Bobai and Xinyi, where the derived rising tones are longer than

the lexical rising tones (Wang 1932, Ye & Tang 1982). Establishing a correspondence

between the Mandarin diminutive suffix [-ɻ] and the Cantonese high rising Pinjam, Chao

(1959) insightfully used mora to describe this additional length, suggesting that the

Cantonese mora is a suffix taking the form of a high tone rather than a sound segment. This

conjecture, capable of explaining the additional length associated with Pinjam, is contrary to

current theories according to which tones, being suprasegmental objects, have no temporal

basis of their own. How to solve this paradox?

Following O’Melia (1939) and Whitaker (1956) according to whom the additional

length is to compensate the elided diminutive suffix [¯in25] in Bobai, a more conservative

dialect compared to Cantonese, we claim that tones, rather than vowels, lengthen in order to

fill the vacuum left by the elision of the neighboring syllable. A conjecture based on the

segmental compensatory lengthening will encounter one problem: if the additional tonal

duration had to be explained by the compensatory lengthening of vowels, no change in length

would be expected to occur in closed syllables. Now, the additional length is observed in

both open and closed syllables in Pinjam. Consequently, it is tone that lengthens under

syllable elision, not vowel. In other words, the vowel lengthens under the pressure of the tone,

not the tone under the pressure of the vowel.

The goal of this paper is two-fold. Through two phonetic studies validating the

post-1940 data in §2.2, we shall account for the origin of this additional duration compared

with the lexical rising tone (Section 3). We claim that it is the tone that lengthens after the

elision of a diminutive morpheme. Taking the Cantonese Changed Tone as a jumping off

point, we shall show that compensatory lengthening is not specific to syllabic positions.

Tones can lengthen after the elision of a diminutive morpheme, demonstrating that they

cannot be treated as pure melodies in Chinese. Discussion and concluding remarks appear in

Section 4.

2. Prelimenaries

Yue dialects are spoken in the province of Guangdong, at the east of Guangxi, and in the

3

south-east of Asia. There are six lexical tones in Cantonese, i.e. 33, 22, 35, 23, 53 (55) , and

21. Three entering tones are observed in closed syllables, i.e. 5, 3, 2.

Apart from the rising tone 35, there exists also rising tones derived from morphological

alternations, referred to as Pinjam (Changed Tone) in the literature. Unlike tone sandhi in

most Chinese dialects, Pinjam is not caused by the contact with neighboring tones. In

Cantonese, the function of Pinjam is to derive, among others, nouns and dimunitives, the

derived elements bearing always a rising contour tone. In (1), nouns are derived from

semantically related verbs. The same thing goes for (2), where dimunitives are purely

indicated by a change in tone. Data are drawn from Yu (2007) et Chen (2000):

(1) Verb nominalization

Level tone Rising tone

sou33 “to sweep” sou35 “a broom”

pɔŋ22 “to weigh” pɔŋ35 “a scale”

wɑ22 “to listen” wɑ35 “an utterance”

tan22 “to pluck” tan35 “a missile”

(2) Derivation of diminutives

tʰoi21 “terrace” tʰoi35 “table”

kʷ ʰɐn21 “skirt” wɐi21 kʷ ʰɐn35 “apron”

kɛŋ33 “mirror” ŋan23 kɛŋ35 “eyeglasses”

tʰɔŋ21 “sugar” tʰɔŋ35 “candy”

nøy23 “women” nøy35 “girl”

ɒp3 “duck” ~ ɒp35 “little duck”

kɒt2 ʦɒt2 “cockroach” ~ kɒt2 ʦɒt 35 “little cockroach”

kɒp3 “pigeon” ~ kɒp35 “little pigeon”

Chen (2000) also notes that several aspectual markers in Cantonese which have either

an underlying high or rising tone often elide in colloquial speech, causing the tone of the

preceding syllable to change to rising, as in (3a) and (3b). Cantonese allows moreover the

ellipsis of certain morphemes in certain constructions. Thus iat 5 “one” is optionlly skipped

in examples such as in (3c). The elided entering tone is realized as 35:

(3) a. Omission of the perfective marker tsɔ35

pɔŋ22 tsɔ35 � pɔŋ35 “to weigh (PERFECTIVE)”

4

fan22 tsɔ35 � fan35 “to transgress (PERFECTIVE)”

ki:n33 tsɔ35 � ki:n35 “to meet (PERFECTIVE)”

b. Omission of the potential marker tɐk5

pɔŋ22 tɐk5 � pɔŋ35 “to weigh (POTENTIAL)”

fan22 tɐk5 � fan35 “to transgress (POTENTIAL)”

ki:n33 tɐk5 � ki:n35 “to meet (POTENTIAL)”

c. Omission of iat 5 “one”

iat tʰam i at t ʰam � iat tʰam t ʰam “puddle by puddle”

5 23 5 23 5 35 23

iat hɔŋ i at hɔŋ � iat hɔŋ hɔŋ “line by line”

5 21 5 21 5 35 21

iat tip iat tip � iat tip tip “plate by plate”

5 2 5 2 5 35 2

According to Bai (1989), two conditions are required so that Pinjam could occur in (3).

On the one hand, tones which give rise to the derived rising contour should be low tones such

as 21, 23, 33, 22, 3 and 2, high tones (53/55, 35, 5) being not influenced by the process. On

the other hand, the omitted syllable should carry 35 or 5.

2.1 Traditional representations of the rising tone derived from Pinjam

In order to represent the derived rising tones in (1) and (2), Yip (1980) and Chen (2000)

postulate a floating high tone associated to the end of the relevant syllable, creating thus a

new rising tone.

(4) sou “to sweep” sou “a broom”

g g2

M H M H

When the stem has a falling contour tone such as 21 or 53, a rule of tonal simplification

is added so that the floating high tone does not create complexe contours, forbidden in

Cantonese. However, when the stem has a rising contour tone such as 23, another rule is

necessary. Based on tonal geometry, Yip (1980) assumes that *25 is not allowed in the

surface form since “cross-register” contours are banned in Cantonese. Chen (2000) suggests

that Cantonese distinguishes only two rising contour tones, i.e. 23 and 35. As a result, *25 is

5

reduced to one of the two possible rising tones, and Cantonese chooses 35 as the final output.

Nevertheless, the reason why 35 is preferred over 23 remains unexplained.

To derive the rising contour in (3a), Yip (1989) proposes the reassociation of the rising

contour tone with the preceding host syllable. A set of tonal readjustment rules bans a

succession of two adjacent low level tones, causing the loss of the first low tone of the verb,

as is shown in (5). Concerning (3b), Yip (1980) and Chen (2000) hypothesize a floating high

tone attached to the previous syllable after the elision of the aspectual marker. A rule of

register readjustment gives rise to the appropriate MH cluster in the surface form, as in (6).

(5) a. pɔŋ < tsɔ > b. pɔŋ c. pɔŋ

g b g@ g

L L H L LH LH

(6) a. pɔŋ <tɐk> b. pɔŋ c. pɔŋ

g b g@ g2

L H L H M H

Albeit highly interesting, the above treatments are debatable for two reasons. A floating

high tone is postulated, in (4) and (6), in order to derive Pinjam. However, when the potential

marker bears already a rising contour tone, as in (5), this hypothetical floating high tone

disappears from the representation. If Pinjam always carries the same rising contour tone,

why to explain it by two distinct hypotheses?

Moreover, these analyses predict that the derived rising tone is homophonous to the

lexical rising tone. However, Chao (1959) notes that in Cantonese, when a word is

pronounced with Pinjam, its duration is slightly longer than that of a word having a lexical

rising tone. Chao uses mora to describe this relative long duration associated with Pinjam,

suggesting that mora is, in Cantonese, a high tone taking the form of a suffix, rather than

corresponding to a segmental melody. This conjecture, capable of explaining the additional

duration associated with Pinjam, is contrary to current theories according to which tones,

being suprasegmental objects, do not have temporal basis of their own. How to explain this

paradox?

2.2 A phonetic experiment on Cantonese Pinjam

In order to verify the duration difference between the lexical rising tone and the derived

rising tone, we conducted a phonetic experiment with a Cantonese native speaker of 77 years

6

old, born in the Province of Guangdong. Five words were read with five repetitions. The

duration of the tone was measured from the spectrogram window.

(7)

Duration of

lexical rising

tone

Duration of derived

rising tone

sister[ze] 0.29 girl [nøy] 0.51

bitter [fu] 0.21 candy [tHçN] 0.54

white [se] 0.25 apron [wɐi kʷ ʰɐn] 0.4

speak [koN] 0.29 call girl [mu nøy] 0.6

cause [saj] 0.24 table [tʰoi] 0.54

Average duration 0.256 Average duration 0.518

Standard deviation 0.03 Standard deviation 0.08

T-value : -7.211 P=0.00036

The average duration of the lexical rising tone is 0.256 seconds, whereas the average

duration of the derived rising tone is 0.518 seconds. A two-tailed t-test was conducted to see

if the mean duration of the two rising tones is statistically different from each other. As can

be seen from (7), the mean duration of the derived rising tone is significantly longer than that

of the lexical rising tone (T-value: -7.211; P=0.00036). One could argue that the subject

might be prompted to create contrasts, and that the difference in length may be due to

contrastive focus rather than the morphological differences. However, the contrastive focus

hypothesis cannot explain why the duration of the derived rising tone is systematically longer

than that of the lexical rising tone.

We are aware of the weak representativity of the data, obtained from only one speaker.

Consequently, we conducted another phonetic experiment with ten Cantonese native speakers

(4 males and 6 females). These speakers, aged from 20 to 25, are currently pursuing their

university studies at a Taiwanese university. More tokens were added in order to enlarge the

database, including words of similar syllable structures, such as [køy] ‘he/she’ vs. [nøy] ‘girl’

and [koN] ‘speak’ vs. [tHçN] ‘candy’. These participants were asked to repeat the data with

five repetitions. The recording result is shown below:

(8)

he/she [køy] 0.305 girl [nøy] 0.470

speak [koN] 0.380 candy [tHçN] 0.403

7

powder [fɐn] 0.330 apron [wɐi kʷ ʰɐn] 0.308

bitter [fu] 0.350 call girl [mu nøy] 0.388

cause [si] 0.313 table [tʰoi] 0.450

write [se] 0.340 glasses [Nan kEN] 0.325

egg [kåi tan] 0.325 nun [såU nøy] 0.420

Average 0.335 Average 0.395

Standard deviation 0.025 Standard deviation 0.060

Ratio

lexical : derived

1 :1.179

T-value : -2.434 P=0.0409

As can be observed from the above data, the duration difference between the lexical

rising tone and the derived rising tone is less important among young speakers of Cantonese.

Nevertheless, the mean duration of the derived rising tone is still significantly longer than

that of the lexical rising tone (T-value: -2.434; P=0.0409). The following figure illustrates the

difference between [koN] “speak” vs. [tHçN] “candy”. Both words have similar rime structure,

but the latter carries a derived rising tone. The duration difference conditioned by vowel

aperture or syllable structure can thus be excluded:

Figure 1 : [koN] “speak” (0.380 seconds) vs. [tHçN] “candy” (0.403 seconds)

The phonetic experiment confirms Chao’s observation that the derived rising tone has a

longer duration than the lexical rising tone.

2.3 Origin of Pinjam

From the phenomena observed in Yue dialects, a basic question follows: why does the

derived rising tone always have a superior duration to that of the lexical tone?

8

Chao (1948) and Wong (1982) suggest that Cantonese Pinjam originates from the elision

of certain morphemes which are no longer productive synchronically. Establishing a parallel

between the function of Pinjam in Cantonese and the retroflex suffix [-ɻ] in Mandarin

through a comparison of a hundred of lexical elements, Chao (1947, 1959) remarks that

around sixty nominal morphemes with the suffix [-ɻ] in Mandarin correspond to Pinjam

morphemes in Cantonese. Given this correspondance, Chao (1959) proposes to treat

Cantonese Pinjam as a suffix. Diminutive suffixation is a very frequent morphophonological

phenomenan in Chinese dialects. In Beijing Mandarin for exemple, the suffix takes a

retroflex feature which is realized in the rime (Chao 1968, Cheng 1973). In other dialects, as

a result of different historical developments, this suffix surfaces sometimes as an independent

syllable, sometimes as a nasal integrated to the coda of the lexical stem.

Whitaker (1956) establishes a more explicit relation between Cantonese Pinjam and the

suffix [-ɻ] 兒 in Mandarin, claiming that the latter is the ultimate source of Pinjam. She

postulates that, at an earlier historical stage, Cantonese might have used the suffix 兒,

pronounced as [-jı :]. At a later stage, the suffix disappeared, leaving a trace under the form of

Pinjam. Her theory is based on a comparison with Bobai, a Yue dialect spoken at the

southeast of Guangxi, where the process of suffixation might shed light on the origin of

Cantonese Pinjam. Citing the study of Wang Li (1931) entitled Une prononciation chinoise

de Po-pai, Whitaker remarks that at the time when Wang Li described the dialect, there were

two varieties. The more conservative variety used the dimunitive suffix [ɲin] carrying a

rising contour tone, as in (8a). When words ended in an occlusive coda, the final stop would

change to its homorganic nasal counterpart and the word would acquire a long rising tone

associated previously with [ɲin], as in (8b):

(8) a. kae32 ɲin25 “little chick”

iaŋ21 ɲin25 “little sheep”

ma21 ɲin25 “little horse”

b. œk54 “house” � oŋ25 “little house”

mat32 “thing” � man25 “little thing”

hɔp4 “box” � hɔm25 “little box”

pak1 “uncle” � a33-paŋ25 “my little uncle”

In addition to suffixation with [ɲin25] and the homorganic nasalization of final stops,

diminutivization in Bobai also involved the change of the original tone to long rising Pinjam

9

without suffixation, a pattern which closely resembles the Cantonese Pinjam. Whitaker sees

these three synchronic processes as three diachronic stages in the development of the long

rising tone: in the first stage, the suffix [ ɲin] with a rising tone is used; by the second stage,

the suffix has been lost, but left tonal (long rising contour) and segmental (nasalization of

final stops) traces; in the third stage, the nasalization has disappeared, leaving only the long

rising Pinjam2.

Tsao (2006) establishes the following diachronic cycle of the phonology of diminutive

suffix in Chinese dialects:

(9)

stage 1.

suffixation

2.

nasal coda

incorporation

3.

nasalization

4.

nasal coda

dropping

5.

de-nasalization

composition stem+

diminutive

suffix

words with

nasal endings

words with

nasalized

vowels and

nasal endings

words with

nasalized

vowels

words without

endings

Tsao (2006) notes that the diminutive tone emerges during the stage of coda loss. If we

compare this cycle with Pinjam in Yue dialects, it appears that Cantonese is less conservative

compared with Bobai which still retains the diminutive suffix.

Citing a study of Ye & Tang (1982), Bauer & Benedict (1997) remark that Xinyi,

another Yue diaelct, attests also similar Pinjam behaviors. There are two processes of nasal

suffixation to indicate diminutivization. In the first case, the nasal coda [ -n] is suffixed to the

lexical stem and the original tone of the stem becomes a rising contour, as in (10a). In the

second case, the same type of nasalization as in Bobai is found, with the final stops becoming

their homorganic nasal counterpart, as in (10b). Ye & Tang emphasize that the peak of this

derived rising contour is so high that the traditional scale of five degrees proposed by Chao

(1930) would not be capable of representing it. They use the symble � for notation. Words

2 Comparing several Wu and Yue dialects, Chen (1999, 2002) argues that the diminutive suffix in Yue dialects

might result from a glottal stop of the rime. At a later stage, this glottal stop gives rises to a rising tone

accompanied with a creaky voice. We share Chen’s argument that glottal stop and rising tone have a close

relationship, with the loss of the glottal stop giving rise to a rising tone during the processes of tonogenesis.

Nevertheless, more explanation will be needed in order to explain the relatively longer duration of Pinjam.

10

already ending in a nasal consonant, or in a diphtongue, do not take the nasal suffix, their

lexical tone simply transforming to a rising contour, as in (10c).

(10) a. tʃy 53 “pig” � tʃyn � “little pig”

lou 23 sy 35 “mouse” � lou 23 syn � “little mouse”

sam 53 fu 33 “clothes” � sam 53 fun � “clothes for kids”

hɔ 23 “river” � hɔn � “little river”

b. ʔap 22 “duck” � ʔam � “little duck”

fɐt 22 “Buddha” � fɐn � “little Buddha”

kɛk 33 “foot” � kɛŋ � “little foot”

c. kɐi 53 “chicken” � kɐi � “little chick”

kʰɐu 23 “ball” � kʰɐu � “little ball”

sɐu 35 kɐn 53 “towel” � sɐu 35 kɐn � “handkerchief”

kɔŋ 53 “jar” � kɔŋ � “little jar”

Unlike Cantonese Pinjam where 35(5) and 53(55) cannot be replaced, in Xinyi, the

derived rising contour can substitute any lexical tone, even 53 and 35. Moreover, according

to Ye & Tang, in Xinyi, the rising Pinjam can distinguish easily from the lexical contour tone

35 in perception. This phenomenon is different from Cantonese where speakers have

difficulties distinguishing them in perception (Yu 2007).

Given the data of Bobai and Xinyi, Bauer & Benedict (1997) consider that Cantonese

Pinjam might be derived from a monosyllable which had carried a rising contour and was

suffixed to words of different classes to represent diverse morphological functions. At an

intermediate stage, this suffix might have been reduced to a nasal coda and to a rising contour

tone which had replaced the lexical tone of the stem. The nasal coda is lost later on, and the

rising contour has become the only trace of the morphological process. We share Bauer &

Benedict’s hypothesis as it gives a whole picture of Pinjam in Yue dialects. The postulate of a

floating high tone, albeit highly interesting, needs several treatments depending on the lexical

tone of the stem, and fails to explain why the rising contour derived from Pinjam has a

superior duration to that of the lexical rising tone in Cantonese and Bobai. In Xinyi, the

derived rising tone rises much more higher than the lexical rising tone, indicating that it

might have a longer duration than the lexical rising tone as well3.

3 Articulatorily, a complicated tonal contour involving more pitch targets would involve more complicated

muscle state change, and thus prefer a longer duration to facilitate implementation. Sundberg (1973, 1979)

11

Before analyzing the long duration of Pinjam, it might be interesting to examine Daye, a

Mandarin dialect spoken in the province of Hubei. Just as in Cantonese, the diminutive

morpheme is purely tonal in Daye (Wang 1996). Wang (1996) calls it “emotive tone”, whose

function is similar to that of Pinjam in Cantonese. There are five lexical tones in this dialect,

i.e. 33, 31, 53, 35 and 13. The diminutive morpheme is realized on the stem by a high falling

contour tone that Wang transcribes as 553. Examples drawn from Wang (1996) are given in

(11a). When the head noun carries the lexical tone 31, the tonal morpheme 553 does not

attache to the stem, and it is necessary to add the morpheme [sai 35] “little” before the noun.

In this case, this morpheme also becomes 553, as is shown in (11b) :

(11) a. tɔ 33 “knife” � tɔ 553 “little knife”

tso 13 “table” � tso 553 “little table”

tɕʰi 35 tsʰe33 “car” � tɕʰi 35 tsʰe 553 “little car”

b. tʰɔŋ 31 “pond” � sai553 tʰɔŋ 31 “little pond”

sai 31 “fish” � sai553 sai 31 “little fish”

Wang remarks that the morpheme [sai 35] “little” is also used for nouns that carry tones

other than 31. In other words, there are two ways of expressing diminutivization in Daye : (i)

the tonal morpheme 553 replaces the tone of the head noun ; (ii) the morpheme [sai 35]

“little” is prefixed to the stem, and carries the tone 553. The first usage is more frequent

among speakers.

Why does the tonal morpheme always have a superior duration to that of the lexical tone

having a similar contour in Daye as well as in Yue dialects ?

In the following section, we shall propose an analysis of Pinjam in Yue dialects, and will

show that the phenomenon in Daye can be explained in the same fashion.

3. Compensatory lenghtening is not specific to syllabic segments

Following Bauer & Benedict (1997), we consider that Pinjam originates from a

dissapeared monosyllable bearing a rising contour. To explain the duration increase of

remarks that a contour with farther-apart pitch targets prefers a greater duration of its carrier. The greater the

pitch excursion of a tone, the longer duration it requires. 15 requires thus more time than 13. Ye & Tang does

not offer the spectrogram of the derived rising contour of Xinyi, and indicate only that the derived rising tone

rises much more higher than the lexical rising tone. We infer that the derived rising tone should also have a

longer duration than the lexical rising tone.

12

Pinjam, O’Melia (1939) et Whitaker (1956) had the intuition that it was to compensate the

loss of the elided syllable. A question arises as to the nature of this compensatory lenghtening:

how do we know if it is the vowel or the tone that lengthens? Two possibilities are at hand:

the tone lengthens due to vowel lengthening, and the vowel lengthens under the pressure of

tone. The first solution seems the only answer to the question: since Goldsmith (1976), only

syllabic constituents and/or segmental melodies are likely to lengthen through their

association to skeletal positions. Tones, being autosegments, cannot lengthen by themselves.

We shall begin by examining vowel compensatory lengthening in different languages in

order to see if this is the case in Pinjam.

3.1 The nature of CL

Compensatory lengthening (CL) designates a process whereby a segment lengthens to

compensate the loss of another segment. Based on the history of French, De Chene &

Anderson (1979) postulate that there was first a coalescence of two vowels with similar

timbres — such as in CŬCŬLLĂM > coole > coule “frock” or SĂTŬLLĂM > saoule, soûle

“drunk”. It was after this stage that the changes known under the name “compensatory

lengthening” — such as in MŬSCŬLŬM > mosle > moûle (loss of preconsonantal s) or

MŎDŬLŬM > molle [ mɔulə] (monophthongization) > moule [ mu:lə] — might contribute

to the formation of long vowels. They claim moreover that CL, as a process of structure

preservation, only occurs in a language where there is a pre-existing vowel length contrast.

However, it has been shown that a pre-existing vowel length contrast is not a necessary

condition for CL (Hock 1986, Hayes 1989, Morin 1992, Lin 1997). Hayes (1989) and Lin

(1997) particularly show that a distinction between heavy and light syllables is sufficient in

the moraic framework. In this sense, it is the mora and not the pre-existing long vowel which

is preserved.

Kavitskaya (2002) proposes another vision of CL, justifying that CL processes, i.e.

those through consonant loss and through vowel loss, are similar diachronically: both of them

result from the phonologization of phonetically long vowel duration, and none concerns the

transfer of phonetic duration or of phonological weight. Rather, intrinsic phonetic vowel

duration is reinterpreted as phonologically long because of a change in syllable structure.

Based on the listener-oriented model (Ohala 1981), she distinguishes two types of CL,

CVCV > CV:C through vowel loss, and CVC > CV: through consonant loss. We will be

primarily concerned with CVCV > CV:C here as the there are many words carrying the

derived rising tone in closed syllables in Yue dialects. It would be necessary to comprehend

13

the nature of CL in closd syllables in order to see if it is the case in Yue.

Regarding CVCV > CV:C, the change in syllable structure is responsible for the

reinterpretation of vowel length as phonemic given that vowels in open syllables are

phonetically longer than those in closed syllables. Consenquently, when the second syllable

is lost (failure to detect the final vowel), the duration of the vowel in the newly-closed

syllable becomes unexplainable, since it is longer than is expected in the closed syllable. The

listener therefore parses the longer duration as intended by the speaker, and reinterprets the

vowel in question as phonologically long. Here are some exemples drawn from Kavitskaya

(2002):

(12) a. Latin Friulian

lupum > *lupu > lo:f ‘wolf’

novum > *novu > nu:f ‘pure’

dekem > *detʃe > di:ʃ ‘ten’

nivem > *nive > ne:f ‘snow’

b. Old Church Slavic Serbo-Croatian

borʊ > bo:r ‘forest’

rogʊ > ro:g ‘horn’

medʊ > me:d ‘honey’

The above process is compensatory only in the large sense of the word, i.e. the

lengthening of a segment is correlated with the loss of a neighboring segment. However,

there is no compensation in the direct sense of the term, since there is neither transfer of

duration nor transfer of weight. In other words, the allophonic duration of the first vowel in

CVCV is phonologized when the second vowel is lost.

Citing Gess’ work on French, Morin (2007) provides a similar hypothesis: research on a

series of ancient texts convinced Gess (1988) that the coalescence of vowels in ancient

French happened only after the middle of the XIII century, longtime after the date normally

admitted for the deletion of preconsonantal s. The chronology proposed by De Chene &

Anderson is thus problematic. Morin remarks furthermore that Wallon de Liège, a French

dialect, conserves well the system of vowel length contrast of Western Romance, such as

NĀSŬM > [ne:] "nose", MŎSĂ > [mu:s] "Meuse" with a long vowel in open syllables in

Romance, and GRŎSSŬM / GRŎSSĂM > [grɔ] / [grɔs] with a short vowel in closed

syllables in Romance. In other words, the vowel lengthening in Wallon de Liège is

14

conditioned by the syllabic structure of Western romance: the vowel is already long in open

syllables, and short in closed syllables.

The hypotheses of Kavitskaya and Morin challenge the nature of CL: if lengthening

always precedes syncope, it is never “compensatory” in the sense that it does not occur “to

compensate” the loss of a segment. There is no transfer of weight, but a transfer of the

distinctive value of a disappeared element. We refer to these phenomena as a false

compensation, considering that finalism (a segment is lengthened in order to compensate the

loss of another segment) plays no role in CL. Just as in the case of palatalization: [tji] gives

rise to the phoneme /tj/ when /i/ is lost. The birth of a new phoneme /tj/ is not to compensate

the inevitable loss of /i/ given that the palatalization of /t/ has happened before vowel loss.

Returning to Yue dialects, what is the ultimate source of the additional duration

associated with Pinjam? If we explain this additional duration by vowel lengthening, we will

encounter the following problem: in the case of Frioulian [lupu] > [lo:f], the first vowel is

phonetically long at the beginning since it is in an open syllable. By contrast, in Yue, there is

no long vowel duration pre-existing to the loss of the diminutive suffix: in the sequence

[mat32 ɲin25] “little thing” in Bobai, the head noun is closed by a consonant; the suffix

begins with a consonant as well. The vowel of the head noun has no possibility to have a

long allophone. In other words, if the additional tonal duration had to be explained by vowel

CL, we would have difficulties in explaining why there is a change of duration in closed

syllables. Vowel lengthening in closed syllables, if any, is a rather marked process. Now, the

additional length is observed systematically in open and closed syllables. The possibility of

vowel CL is thus excluded. If we posit that the codas in Yue are moraic, we will encounter

another problem: it is difficult to understand why, in the case of three entering tones, tonal

duration is shortened by final stops in closed syllables in Cantonese, whereas the same final

stops are capable of bearing a long rising tone in Pinjam (cf. /ɒp35/ “little duck”, /kɒp35/

“little pigeon”)

Now, it remains the only possibility: there is tonal compensatory lengthening so that the

underlying structure could be preserved. The trigger of the additional duration associated

with Pinjam is tonal, vowel lengthening being only a correlate of tonal lengthening.

Unlike the two types of CL where the process is always accompanied by a change in

syllable structure, in Yue, an additional duration is systemetically found in Pinjam whereas

the syllable structure of the head noun remains unchanged: it is always in closed syllables.

Given the phonological aberration of vowel lengthening in closed syllables, it is the tone, the

only element active, which spreads to the following position released by the elision of the

15

diminutive suffix. We consider that Pinjam in Yue dialects constitutes a case of lengthening

which is truly compensatory: there is a compensation of the loss of the diminutive suffix by

an additional duration associated with Pinjam.

The compensatory nature of Pinjam being assured, the successive stages of its

development in Yue dialects can be illustrated as in (13):

(13) a. Suffixation

� ma21 ɲin25 “little horse” (Bobai)

b. Nasalization of the occlusive coda

� mat32 “chose” � man25 “little thing” (Bobai)

Vowel nasalization

� hɔ 23 “river” � hɔn � “little river” (Xinyi)

Emergence of a tonal suffixe with compensatory lengthening

� tʰÅŋ21 “sugar” � tʰÅŋ35 “candy” (Cantonese)

We can see that Cantonese is relatively innovative by comparison with other Yue

dialects, with the tonal suffix the only morphological marker nowadays.

Nasalization of occlusive codas by the diminutive suffixe is shown in (14)4:

(14) a. mat32 + ¯in25 � b. man25: “little thing”

H L ← L H L H

h0h h0h h0hh

x x + x x � x x + x x

0h0h 0h0h 0h0h

C V C C V C C V C

g g g!g g g g g g

m a t ¯ i n m a n

4 The interaction between tones and temporal positions is proposed as follows: the register is represented by the

spreading of a tonal segment to an adjacent position. Notice that the capital letter indicates the register.

a. falling contour tones b. level tones c. rising contour tones

H L H L H L H L L H L H

h1g g0g g1 0g g0gg g1g

x x x x x x x x x x x x

[Hl] (53) [hL] (31) [H] (44) [L] (22) [lH] (35) [Lh] (13)

16

In (14a), the nasal feature provided by the diminutive suffix [ɲin25] nasalizes the coda

of the preceding syllable. The rising tone of the diminutive suffix replaces that of the head

noun, but their temporal positions remain, allowing the second tonal segment of the rising

contour to spread to the following position released by tonal replacement. Hence there is the

additional duration of the rising contour derived from Pinjam.

In sum, it is the tone, the only element active even in closed syllables, which lengthens

after the elision of the dimunitive suffix, not the vowel. In other words, if there exists vowel

lengthening, it is the vowel which lengthens under the pressure of the tone, not the tone under

the pressure of the vowel.

3.2 Application to Daye

Recall that there are two ways of expressing diminutivization in Daye : (i) the tonal

morpheme 553 replaces the tone of the head noun ; (ii) the morpheme [sai 35] “little” is

prefixed to the stem, and carries the tone 553. Just as in Yue where the diminutive morpheme

is suffixed to the stem, we hypothesize that, in Daye, the prefixation of the morpheme [sai 35]

“little” might represent a more conservative usage of the dialect, and that a tone sandhi rule

specific to Daye transforms 35 to 53 in initial position5. At a later stage, this sandhi tone

gradually replaces the tone of the head noun.

The successive stages of diminutivization are given in (15), and the representation of

tonal CL is shown in (16) :

(15) tɔ 33 “knife” � tɔ 553 “little knife”

a. sai 35 + tɔ 33 � prefixation of the morpheme [sai 35] “little” to the head noun

b. sai 53 + tɔ 33 � 35 � 53/#__

c. Ø + tɔ 53 � substitution of the lexical tone of the stem by the prefix tone

d. tɔ 553 � spread of the first tonal segment to the preceding position

5 There are only three papers, all written in Chinese, on Daye. Two of them concern the emotive tone. The third

constitutes a comparative phonetic study of Mandarin and Daye. Our tone sandhi hypothesis conforms to the

characteristics of Mandarin dialects, to which Daye belongs: the syllable in final position retains its underlying

tone, and the preceding syllable is subject to tone sandhi rules.

17

(16) Ø + tɔ 53 � tɔ 553

H L

hh1h

x x + x x

qbq 0h0

C V C V

g g g g

s ai t ɔ

The sandhi tone replaces that of the head noun and spreads to the preceding position

released by the elided diminutive prefix. Consequently, the emotive tone is longer than the

lexical tone having a similar contour.

3.3 Implications of the present hypothesis

Tonal CL shows that tones cannot be treated as pure melodies in Chinese: they do have

direct acces to temporal positions just as syllabic elements onset and rime. Recent MRI work

conducted by Gandour (2000) and Gandour et al. (2005) compares the processing of various

aspects of prosody by Mandarin and English listeners. Their results show that, for Mandarin

listeners, but not for English listeners, the perception of consonants, rhymes and tones all

elicit predominantly the activation of left hemisphere, distributed over frontal, temporal and

parietal regions. They conclude that the processing of phonological contrasts, regardless of

whether they are segmental or suprasegmental, takes place in the same areas of the brain

among Chinese speakers. As a result, these results also echo our suggestion that tones have

the same status as segmental units in Chinese.

The duration increase of the contour tone derived from Pinjam reminds us of a study

conducted by Carter & Gerken (2004) on the omission of unstressed syllables by

English-speaking children. It is well admitted that pretonic syllables such as those in banana,

Cassandra are often omitted by children. Tranditional metric hypotheses assume that children

have similar lexical representations to those of adults, but production constraints prevent

them from prouncing their targets correctly, which impose a reduced prosodic structure

obeying to a trochaic feet (Gerken 1991, Demuth 1995). It follows from these hypotheses

that omitted unstressed syllables are absent from children’s output. Carter & Gerken consider

on the contrary that children simplify their production, deleting only the segmental material

of unstressed syllables. The underlying representation of unstressed syllables might be

18

present under the form of compensatory lengthening or other prosodic traces. They examined

the duration of reduced dissyllabic forms prounced by children, and compared them with

dissyllabic forms having a real trochaic structure s-w, i.e. Cassandra~Sandy, Lucinda ~ Cindy.

Their phonetic study shows that the average duration of verb-to-noun is longer for reduced

formes (“_cinda”, “_sandra) than for non reduced forms (“Cindy”, “Sandy”)6. The same

thing goes for Pinjam in Yue dialects, where the additional tonal duration preserves the

underlying structure of the diminutive suffix.

3.4 Neutralization in Modern Cantonese

Bauer & Benedict (1997) note that there is a neutralization of contrast between the

lexical rising tone and the rising tone derived from Pinjam in process in modern Cantonese:

“According to both Benedict (1942) and Chao (1947) who were writing their

descriptions of Cantonese tones in the 1940’s, the regular high rising tone had a mid to

high rising contour of 35, and the high rising Pinjam had a long, low to high rising

contour of 25:. Based on his experience of learning Cantonese in Guangzhou on

1949-1950, Egerod (p.c. with Bauer, 1994) directly observed this difference in the two

tones. However, at some time in the past 50 years the contours of the two tones have

become identical and indistinguishable for both Guangzhou and Hong Kong speakers.

Speakers now use the mid-low to high rising contour for both tones.”

This tendency is also confirmed by a study carried out recently by Yu (2007). He

remarks that speakers produce, with subtil phonetic differences, different rising contours

derived from Pinjam, but are unable to perceive the difference between the lexical rising tone

and the derived rising tone. Our two phonetic experiments also exhibit this neutralization in

process: the subject of the first experiment was 77 years old, and his pronunciation clearly

exhibits the duration difference between the lexical rising tone and the derived rising tone.

6 Here are some examples of stimuli used by Carter & Gerken (2004) :

Disyllabic target names Trisyllabic target names

Monosyllabic verbs He pushed Sandy. He pushed Cassandra.

He pushed Cindy. He pushed Lucinda.

Disyllabic verbs He pushes Sandy. He pushes Cassandra.

He pushes Cindy. He pushes Lucinda.

The average duration of verb-to-noun (He pushed Sandy vs He pushed Cassandra) was analyzed in order to see

if there was a trace left by an omitted syllable.

19

The subjects of the second experiment aged from 20 to 25 years old. It was observed that the

duration difference is less important among these young speakers of Cantonese, although the

mean duration of the derived rising tone is still significantly longer than that of the lexical

rising tone. In other words, this difference systemetically reported in the 1950s is

disappearing in modern Cantonese. Where does this neutralization come from? Moreover,

admitting that, in more conservative dialects, the duration increase results from an elided

diminutive suffix, another question arises: why is there no diminutive suffix in Cantonese?

Before tackling this problem, il would be necessary to discuss another problematic, which

appears unrelated to the neutralization in progress in Cantonese: verb nominalization through

tone change.

Mai (1995) observes that in Jiantsung, another Yue dialect, the word “bag” is realized

with the lexical tone 22 among elderly speakers, but among young speakers, “the bag”, be it

a big one or a small one, can be pronounced with a derived rising contour 35. This

phenomenon illustrates that the derived rising tone, as a nominal marker, comes from the

diminutive marker. Mai remarks, on the other hand, that the derived rising tone resulting

from the diminutive morpheme has gradually become a derivation marker in adjectival,

adverbial and verbal classes, where the function of Pinjam is to reduce the force transmitted

by verbs or to express an action of short duration. Here are some examples of Xinyi and

Cantonese:

(17) a. Xinyi, adjectival class (Ye & Tang 1982)

kam33 tɔ53 “so numerous” kam33 tɔ� “so few numerous”

kam33 tai53 “so tall” kam33 tai� “so little”

kam33 tʃʰɛŋ11 “so long” kam33 tʃʰɛŋ� “so short”

b. Xinyi, verbal class (Ye & Tang 1982)

mɔ53 “touch” mɔ53 mɔn� “touch gentlely”

tʃʰœ23 “sit down” tʃʰœ23 tʃʰœn� “sit down for a little while”

pʰak33 “hit” pʰak33 paŋ� “hit gentlely”

c. Xinyi, adverbial class (Mai 1995)

tai22 kʰɔi� “probably”

lœk22 lœk� “a little”

d. Cantonese, verb nominalization (Yu 2007)

sou33 “to sweep” sou35 “a broom”

pɔŋ22 “to weigh” pɔŋ35 “a balance”

20

In other words, in Yue, the derived rising tone was, at the beginning, a diminutive

marker. The diminutivization process has spread to other word classes later on. Given that it

is in the nominal class that diminutivization was developed the most completely, and that it is

in the nominal class that the rising contour is the most numerous, speakers have taken it as a

nominal marker.

Returning to Cantonese, we consider that, just as in Bobai, there was a diminutive suffix

with a rising contour to produce nominal diminutivization. At a later stage, there were two

ways of deriving diminutives: use of diminutive suffix with a rising contour, and tonal

replacement by a long rising tone. Diminutivization by tonal substitution has been spreading

to all classes of the language, and has become more and more productive and regular.

Consequently, it supplanted the diminutive suffix with a rising contour which was used only

in the nominal class. This substitution has provoked the loss of the suffix whose existence

became redondant. The suffix loss caused the dissapearance of all traces of affixal positions,

so CL cannot occur in modern Cantonese. As a consequence, the rising contour derived from

Pinjam confused with the lexical rising tone, and become the only marker of

diminutivization.

In sum, the derived rising contour, which was a diminutive marker in the nominal class,

has gradually become a nominal marker and a diminutive marker in other word classes. This

grammaticalization process creates new forms, introduces categories which had no linguistic

expressions, and transforms the whole system (Meillet 1912).

This grammaticalization reminds us of the adverbs formed with –ly in English, where

three forms were in competition, i.e. suffixation (–ly), zero formation (just, pretty, very), and

compounding (-wise, -ward, -style). The derivation by –ly has spread later on, and supplanted

other forms of adverbal suffixes, such as zero formation (Brinton & Traugott 2002).

4. Conclusion

CL which has been examined shows that tones are not pure melodies in Chinese: they

do have direct acces to temporal positions. A theory based on segmental CL will encounter

the following problem: if the additional tonal duration had to be explained by the CL of

vowels, no change in length would be expected to occur in closed syllables. Now, the

additional length is observed in both open and closed syllables in Pinjam. Consequently, it is

the tone that lengthens under syllable elision, not the vowel. In other words, the vowel

lengthens under the pressure of the tone, not the tone under the pressure of the vowel.

Moreover, we hope to have shown why there is neutralization between the lexical rising

21

tone and the derived rising tone in modern Cantonese: the diminutivization process by tonal

substitution has been spreading to all classes of the language, and has become more and more

productive and regular. Consequently, it supplanted the diminutive suffix with a rising

contour which was used only in the nominal class. This replacement has provoked the loss of

the suffix whose existence became redondant. The loss of the suffix caused the dissapearance

of all traces of affixal positions. Consequently, the process of CL cannot happen in modern

Cantonese.

5. References

Bai, Wanru.白宛如.1989.<廣州話中的省略性音變> 《方言》2:114-120.

Bao, Zhiming. 1990. On the Nature of tone. Ph.D. dissertation, MIT.

Bao, Zhiming. 1999. The Structure of Tone. Oxford, Oxford University Press.

Bauer, Robert S. & Paul K. Benedict. 1997. Modern Cantonese phonology. Berlin: Mouton

de Gruyter.

Benedict, Paul K. 1942. Cantonese Phonology, ms.

Brinton, Laurel J. & Traugott, Elizabeth Closs. 2002. Lexicalization and Language Change.

Cambridge: Cambridge University Press.

Carter, Allyson & Gerken, Louann. 2004. Do Children’s omissions leave traces? Journal of

Child Languages 31:561-586 Cambridge University Press.

Chao, Yuan Ren. 1930. A system of tone letters. Le maître phonétique. 45:24-27.

Chao, Yuan Ren. 1947. Cantonese Primer. Mass: Havard.

Chao, Yuan Ren. 1959. The morphemic status of certain Chinese tones. In Transactions of

International Conference of Orientalists in Japan IV:44-48.

Chen, Matthew. 2000. Tone Sandhi, Cambridge, Cambridge University Press.

Chen Zhongmin. 陳忠敏. 2002. <論廣州方言一種小稱的來源> 《東方語言與文化》

1:167-181，上海東方出版中心。

Chen Zhongmin. 1999. The Common Origin of Diminutives in Southern Chinese Dialects

and Southeast Asian languages” (漢語南方方言與東南亞語言小稱形式的共同來源),

Linguistics of the Tibeto-Burman Area: 22.2, 21-48.

Cheng, Teresa. 1973. The phonology of Taishan. Journal of Chinese Linguistics 1

(2):256-322.

Cheung, Hung-nin Samuel. 1997. Completing the Completive: (Re) Constructing Early

Cantonese Grammar. Journal of Chinese Linguistics, Monograph Series 10. Studies on

the History ofr Chinese Syntax, ed. by Chaofen Sun.

22

De Chene, E. Brent & Stephen R. Anderson. 1979. Compensatory lengthening. Language 55,

505–535.

Demuth, Katherine. 1995. Markedness and the development of prosodic structure. In. J.

Beckman (ed.), Proceedings of NELS 25, Amherst: GLSA, University of Massachusetts.

Gandour, J. 2000. Frontiers of brain mapping of speech prosody. Brain and Language 71:

75-77.

Gandour, J.,Tong, Y., Talavage, T., Wong, D., Dzemidzic, M., Xu, Y., Li, X. and Lowe, M.

2005. Neural circuitry underlying sentence-level prosody. NeuroImage 28: 417-428.

Gess, Randall. 1998. Compensatory lengthening and structure preservation revisited.

Phonology 15.353–366.

Gerken, LouAnn. 1991. The metrical basis for children’s subjectless sentences. Journal of

Memory and Language 30, 431-451.

Goldsmith, John. 1976. Autosegmental phonology. Ph.D. Dissertation, MIT. Published by

Garland Press, New York, 1979.

Hayes, Bruce. 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20:

253-306.

Hock, Hans Henrich. 1986. Compensatory lengthening: in defense of the concept `mora'.

Folia Linguistica 20: 431- 460.

Kao, Diane L. 1971. Structure of the Syllable in Cantonese. The Hague:Monton

Kavitskaya, Darya. 2002. Compensatory lengthening: phonetics, phonology, diachrony.

Routledge: New York,

Lam, kin-ping. 2002. A Study of Intonation in Hong Kong Cantonese. Dissertation doctorale,

Chinese University of Hong Kong.

Lin, Yen-Hwei. 1997. Syllabic and moraic structures in Piro. Phonology 14: 403 - 436.

Liu, Te-hsin (2008) Marque, registre et contour dans les systèmes tonals en

chinois (Markedness, Register and Contour in Chinese Tonal Systems), doctoral

dissertation, Université Paris 8.

Liu, Te-hsin (2009) On the Status of Tone in Compensatory Lengthening, talk given

at Berkeley Phonetics & Phonology Forum, UC Berkeley, April 13.

Luo, Jian-jun. 羅建軍 2006. 大冶陳貴方言口語中的合音詞，湖北師範學院學報 4:78-81.

Mai, Yun. 麥耘 1990. <廣州話的特殊 35調>《第二屆國際粵方言研討會論文集》, 廣州:

暨南大學出版社。

Mai, Yun. 麥耘 1995. <廣州話的語素變調及其來源與嬗變>，《音韻與方言研究》。

241-282， 廣州: 廣東人民出版社。

Meillet, Antoine. 1912. L’évolution des formes grammaticales. Scientia (Rivista di Scienza)

23

12, No. 26, 6. Linguistique historique et linguistique générale, 130-48. Paris: Librairie

Ancienne Honoré Champion, 130-148.

Morin, Yves-Charles. 1992. Phonological interpretations of historical lengthening. Paper

presented at the 7th International Phonology Meeting, Krems.

Morin, Yves-Charles. 2007. Sources et évolution des distinctions de durée vocalique en

gallo-romain. Paper presented at the conference GalRom07 - Diachronie du gallo-roman,

Nice, January 15-17.

Ohala, John. 1981. The listener as a source of sound change. In Masek, C.S., R.A. Hendrik,

M.F. Miller (eds.) CLS 17-2, 178-203. Chicago: CLS.

O’Melia, Thomas. 1939. First Year Cantonese. Hong Kong Maryknoll House, Hong Kong.

Sundberg, Johan. 1973. Data on maximun speed of pitch changes. Speech Transmission of

Technology. Stockholm, Sweden.

Sundberg, Johan. 1979. Maximum speed of pitch changes in singers and untrained subjects.

Journal of Phonetics 7: 71-79.

Tsao, Feng-Fu. 曹逢甫. 2006. <語法化輪迴的研究：以漢語鼻音尾/鼻化小稱詞為例> 《漢

語學報》2：2-15.

Wang, Guosheng. 汪國勝 1996. <湖北大冶話的情意變調>《中國語文》5:355-360。

Wang, Li. 1931. Une prononciation chinoise de Po-pei (Province de Kouang-si). Ph.D.

dissertation, Faculté des lettres de l’Université de Paris.

Whitaker, K. P. K. 1956. A study on the modified tones in spoken Cantonese. Asia Major

184-207.

Ye, Guocheng & Tang, Zhidong. 葉國泉、唐志東. 1982. <信宜方言的變音> 《方言》1:

47-51.

Yip, Moira. 1980. The Tonal Phonology of Chinese. Ph.D. dissertation, MIT.

Yip, Moira. 1989. Contour Tones, Phonology 6, 149-174

Yu, Alan. 2007. Understanding near mergers: the case of morphological tone in Cantonese.

Phonology 24 (1): 187-214

Yue-Hashimoto, Anne. 1972. Phonology of Cantonese. Cambridge: Cambridge University

Press.