a contour tone chain shift in jinhua wu sandhi tones

A Contour Tone Chain Shift in Jinhua Wu Sandhi Tones

A Contour Tone Chain Shiftin Jinhua Wu Sandhi Tones

Lucien Carroll ([email protected])

Workshop on East Asian LinguisticsUC Santa BarbaraFebruary 20, 2010


1 Jinhua sandhi in contextSociolinguistics of JinhuaIsolation tone systemTone sandhiConditions for change

2 A sociotonetic studyData collection and processingSpeaker similaritiesSandhi contours are shifting

3 Conclusions


Jinhua sandhi in context

Sociolinguistics of Jinhua

Inland southern Wu

●

●

Hangzhou

Nanjing

● Shanghai

●

●

●

●

●

●

●

●

●

●

●

●

●

Huzhou

Taizhou

Jiaxing

Lishui

Ningbo

Quzhou

Shaoxing

Suzhou

Wenzhou

Wuxi

Xuanzhou

Yongkang Tangxi

●

Jinhua

100 km

Figure: Wu is spoken in Shanghai, mostof Zhejiang, and neighboring corners ofAnhui and Jiangsu.

Inland Wu area charac-terized by:

High lexical andphonologicalvariation

Complicated tonalphonology

Language attritionin youngestgeneration



Isolation tone system

Tone categories and contours

Table: Tone categories of Jinhua (of generation born ca. 1930) comparedto Middle Chinese, according to Qian (1992)

MCÀ MCÁ MCÂ MCÃa MCÃb

MC [−vc] ons T 4351 T 544

3 T 455 T7

4P

MC [+vc] ons T 2132 T 24

6 T82P

Typical Wu tone inventory: similar categories, differentcontours (Notation: T contour

category )

MC voiced obstruents have devoiced in urban Jinhua, but aremurmured in some nearby areas; generation born 100 yearsago had a T4

T7 and T8 have very short contours, so will not be used inthis study → 5 isolation tones ∼ 6 historical categories



Tone sandhi

Beijing sandhi

Tone sandhi: phonological change to lexically specified tone

Example

Rule ExampleT3 → T214/ # foN55ùwej214 ‘feng-shui’→ T2

35/ T3 ùwej35kwo214 ‘fruit’→ T21/elsewhere ùwej21khu51 ‘reservoir’

The disyllabic sandhi of Beijing Mandarin is reducible, with fewexceptions, to this one three-part phonological rule.



Tone sandhi

Jinhua sandhi

The disyllabic sandhi of Jinhua is very complicated.A general summary:

5× 5 combinations of long tones produce about a dozendisyllabic sandhi contour categories

The sandhi patterns are not reducible to coarticulation nor toa small number of rules or constraints.

Jinhua exhibits ‘tone type recovery’: In some cases, historicaltone categories that have merged in isolation have differentsandhi patterns

Different grammatical structures have different sandhi, and inaddition, there is some unexplained lexical variation



Tone sandhi

Lexical variation in sandhi

For some combinations, as many as three different patterns areobserved, and the source of this lexical variation is largely unknownCao (2002, p. 111):

Example

T555 T2

313 →a. T33T14 pu.a tCju.dýjoN

cloth shoe bedbugb. T33T55 su.ju sje.mAw

veggie oil delicatec. T55T3 thja.m@N sje.dýju

steel door snowball



Conditions for change

Jinhua is ripe for change

Conditions predispose Jinhua phonology to change:

A generational shift from Wu dominance to Mandarindominance

Urbanization — contact with similar dialects

Likely recent loss of obstruent voicing contrast and T4

The sandhi system is difficult: highly complicated andidiosyncratic

Researchers as well as speakers mention that “young people speakdifferent”, but there has been no systematic study of variation inJinhua sandhi systems.


A sociotonetic study

Data collection and processing

Elicitation

Based on procedure of You and Yang (2001)

Stimuli are disyllabic words representing tone combinations

In carrier sentence thiN5 sa1 pje6 (‘Listen to three times’)

4 groups of speakers:

urban older group: 5 speakers born 1931-1952urban younger group: 5 speakers born 1984-1989rural older group: 2 speakers born 1946-1949rural younger group: 3 speakers born 1985-1986




Stimuli

Structure of compound or modifier+noun

Tone domain (domain of characteristic tone contour) assumedto be the final (yunmu) (Rose, 1998)

108 stimuli: 3 words for each combination of the 6 historicallong tones (MC À, Á, Â × [±vc])

Example

foN1 Cy3 ‘feng shui’




Data Processing

Tone domains annotated by hand

Pitch tracks extracted automatically in Praat

Each tone domain’s contour summarized as a quadratic curve,by polynomial regression with the intercept in the middle ofthe tone domain → each disyllabic contour is two parabolas

The collection of all speakers’ contours for a word is used asmathematical abstraction of that word

The collection of all words’ contours by a speaker is used asmathematical abstraction of that speaker




Principle Components Analysis (PCA)

The procedure is highly bottom-up, and a key tool here is PCA

1 Represent words or speakers as a vectors of contourparameters

2 Select dimensions that represent the most variance

3 Visualize similarities and differences in those dimensions

The resulting dimensions might not have obvious meaning, butsometimes they do.



Speaker similarities

PCA of speaker similarities

A B C D E F G H I J K L M N

Variance per dimension

Dimension Index

Pro

port

ion

of v

aria

nce

0.00

0.05

0.10

0.15

Figure: The 4dimensions with themost variancerepresent 47% of thebetween-speakervariance.




Dimensions B and D

Figure: Dimensions Band D show that thedata from speakers 18and 23 are atypical,so we set them aside.Borders: older (blue)and younger (red);Fill: urban Jinhua(blue) and Zhumavillage (yellow);Shapes: men (circles)and women (squares).




Dimensions A and C

Figure: Dimension Ais associated withgeneration, andDimension C isassociated withlocation. The circledgroups become thebasis of furtheranalysis.



Sandhi contours are shifting

PCA of word differences

1 2 3 4 5 6 7 8 9 11

Variance per dimension

Dimension Index

Pro

port

ion

of v

aria

nce

0.00

0.10

0.20

0.30

Figure: Dimensions 1and 2 togetherrepresent 61% of thevariance. We will alsolook briefly atDimensions 3 and 4.




Dimension meanings: archetype contours

0.0 0.2 0.4 0.6 0.8 1.0

−1.

0−

0.5

0.0

0.5

1.0

Older speakers

Normalized time

Arb

itrar

y F

0 un

its

1st domain 2nd domain

0.0 0.2 0.4 0.6 0.8 1.0

−1.

0−

0.5

0.0

0.5

1.0

Younger speakers

Normalized time

Arb

itrar

y F

0 un

its

1st domain 2nd domain

Dim 1Dim 2Dim 3Dim 4

Figure: Older speakers: D1 ∼ tone domains’ height difference; D2 ∼fall-rise vs rise-fall; D3 ∼ overall height; D4 ∼ contours rise/fall equally.Younger: D1 and D2 contours are shifted.




Contour changes

Figure: The change in D2captures the rising tonesgetting lower and the fallingtones getting higher. Thechange in D1 captures thelow turning towards fallingand the high turning towardsrising.




The shift is a generalization

−2 −1 0 1 2 3

−3

−2

−1

01

23

Counterclockwise words

Dimension 1

Dim

ensi

on 2

−2 −1 0 1 2 3

−3

−2

−1

01

23

Clockwise Words

Dimension 1

Dim

ensi

on 2

Figure: Each arrow indicates one word’s change from older speakers’contour to younger speakers’ contour. A large majority of words havemoved in a counter-clockwise direction.


Conclusions

Why are the contours changing this way?

Hypothesis A: The younger generation are reclassifying wordsaccording to Mandarin categories↪→ A few words are being reclassified, but most of thechanges are small, not categorical

Hypothesis B: The younger generation areperceiving/producing contours according to the Mandarincontour shapes↪→ This could cause falls to get higher (cf. Mandarin T 51

4 )and maybe lows to turn to falls (cf. Mandarin T 214

3 ∼ T 213 ).

But not rises to get lower (cf. Mandarin T 352 ), nor highs to

turn to rises (cf. Mandarin T 551 )


Conclusions

Why are the contours changing this way?

Hypothesis C: The younger generation misperceives ormisproduces the contour slightly later within the syllable.

1 The end of the contour is lost.2 The beginning is recreated from coarticulation.

Sanders (2008) provides a similar account for how Beijing T 352

became Taiwan Mandarin T 3242 .


Conclusions

Conclusions

The observed change is consistent with shifting of thecontours within the syllable

The change is not consistent with direct influence fromMandarin tone categories

The social conditions in Jinhua predispose it to change

We should expect to find similar tonal evolution in othercontour tone languages


Conclusions

Acknowledgements

Special thanks to:

The people of Jinhua, especially the native speakers recordedfor this study, and Emily Cen, Betty Hu, and Angie Kim

The linguists of UCSD, especially Sharon Rose (my advisor onthis project) and the other members of SaD PhIG


Conclusions

References

Cao, Z. (2002). Nanbu Wuyu yuyin yanjiu [Southern Wuphonology research]. Beijing: Shangwu Yin Shuguan.

Qian, N. (1992). Dangdai Wuyu yanjiu [Contemporary Wuresearch]. Shanghai: Shanghai Jiaoyu Chuban She.

Rose, P. (1998). The differential status of semivowels in theacoustic phonetic realisation of tone. In 5th ICSLP. (Paper298)

Sanders, R. (2008). Tonetic sound change in Taiwan Mandarin:The case of tone 2 and tone 3 citation contours. In 20thNACCL (pp. 87–107).

You, R., & Yang, J. (2001). Jinhua fangyan shengdiao shiyanyanjiu [Jinhua dialect tone experimental research]. In Wuyushengdiao de shiyan yanjiu [Wu tone experimental research].Shanghai: Fudan Daxue Chuban She.


Conclusions

PCA of groups’ mean contour representations

−0.2 −0.1 0.0 0.1 0.2

−0.

2−

0.1

0.0

0.1

0.2

PC1

PC

2

11_111_211_312_112_212_3

13_113_213_3

14_114_214_3

15_115_2

15_3

16_1

16_2

16_3

21_121_221_3

22_1

22_2

22_3

23_1

23_223_324_1

24_224_3

25_1

25_225_3

26_1

26_2

26_3

31_131_2

31_3

32_1

32_232_3

33_133_233_3

34_1

34_2

34_335_135_2

35_3

36_1

36_2

36_341_1

41_2

41_3

42_1

42_2

42_343_1

43_243_344_1

44_244_3

45_145_245_3

46_146_2

46_351_151_2

51_3

52_152_252_3

53_153_253_3

54_1

54_2

54_3

55_1

55_2

55_356_1

56_256_3 61_1

61_2

61_3

62_1

62_2

62_3

63_163_263_3

64_164_2

64_3

65_1

65_2

65_3

66_1

66_2

66_3

−5 0 5

−5

05

a1oa2o

b1o

b2o

c1o

c2oa1y

a2yb1y

b2y

c1y

c2y

Figure: Originalrepresentationsprojected onto firsttwo dimensions.Speaker groups: ‘o’,‘y’ and ‘z’ vectors;Heights: ‘a’ vectors;Slopes: ‘b’ vectors;Curvatures: ‘c’vectors


Conclusions

Word similarities and groups

A

AAA

AA

B

BBB

BB

AA

A

C

A

B

D D

D

E

E

E

A

BB B

DD

E

DD

E

E

E

F

F

F

G

GG

HHH

H

F

IJ

J

J

G

I

I

F

F

F

G

K

F

H

HH

B

B

H

JJJ

I

I

I

FA

F

I

I

I

BBB

C

C

I

F

A F

C

A

C J

J

J

C

I

K

H

C

H

BH

H

C

C

E

I

E

I

−4 −2 0 2 4

−2

02

4

Dim. 1

Dim

. 2

11_1

11_2

11_312_1

12_212_3

13_1

13_213_314_1

14_214_3

15_115_2

15_3

16_1

16_2

16_3

21_1 21_2

21_3

22_1

22_2

22_3

23_1

23_2

23_3 24_1

24_224_3

25_1

25_225_3

26_1

26_2

26_3

31_1

31_2

31_3

32_1

32_232_3

33_133_2

33_3

34_1

34_2

34_3

35_1

35_2

35_3

36_1

36_2

36_3

41_1

41_2

41_3

42_1

42_2

42_3

43_1

43_243_3

44_1

44_2

44_3

45_145_245_3

46_1

46_2

46_3

51_151_2

51_3

52_1

52_2

52_3

53_153_2

53_3

54_1

54_2

54_3

55_1

55_2 55_3

56_1

56_2

56_3 61_1

61_2

61_3

62_1

62_2

62_3

63_1

63_2

63_3

64_164_2

64_3

65_1

65_2

65_3

66_1

66_2

66_3

Figure: Plot of stimulicontour similarity, inthe first two principlecomponents. Wordclusters are indicatedby the colored letters.The first two digits ofthe word labels referto the tonecategories, numberedsequentially 1through 6.


Conclusions

Word familiarity vs contour change

●

●●

●

●

●

●

● ●●

●

●

●●

●●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●●

●

●●●

●

●

●

●●

●●

●

●

●

●●

●

●

●

●

●

●

●

●●

●●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●●

●

●

●

●

●

● ●

●●

●

●

−2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0

−3

−2

−1

01

2

Mean Familiarity sec. Young Speakers

Clo

ckw

ise

mot

ion

●

●

●

Figure: Familiarityof a word doesnot predict themagnitude ofcounter-clockwisemotion

a contour tone chain shift in jinhua wu sandhi tones

Documents