TAL 2012 ̶ Third International ISCA Archive Symposium on Tonal Aspects of http://www.isca-speech.org/archive Languages Nanjing, , May 26-29, 2012

Tone Sandhi in Jiaonan Dialect: an Optimality Theoretical Account

Zhao Cunhua 1, Zhai Honghua 2

1 College of Foreign Languages, University of Science and Technology, China 2 College of Foreign Languages, of Science and Technology, China [email protected], [email protected]

Abstract 2. Experimental Description Based on phonetic experiment, this paper aims at providing The raw phonetic data is gained through the acoustic records phonetic description on the tones and disyllabic tone sandhi in conducted in July 2009 and November 2010 respectively. Jiaonan dialect, and then illustrating the phonological Only one elder male’s and female’s sound were chosen from structures within the framework of Auto-segmental Phonology the six informants’ invited in 2009. The test words were from to conduct an OT analysis on the tone sandhi patterns of Fangyan Diaocha Zibiao and The Study of Shandong Dialect, Jiaonan dialect. including 160 monosyllabic words in citation forms and 130 Index Terms: Jiaonan dialect, phonetics, tone, tone sandhi, pairs in disyllabic sequences. Finally, 111 monosyllabic words OT in citation forms and 117 in disyllabic sequences are involved and taken for the next recording. The equipment used then 1. Introduction was SAMSUNG BR-1640 digital recorder. In 2011, 6 college Jiaonan, a county-level city of , locates in the students from Jiaonan City were invited for audio recording. southeast of . Jiaonan dialect (JND for Data recording was made by using a portable computer short hereafter) belongs to Jiao Liao Mandarin, a sub-dialect (Lenovo) and a microphone (Sennheiser PC 166) with the of the northern Mandarin family. According to Qian Zengyi correspondent sound board. The recording software was [1], JND is one representative of Dongwei Region, one of the CUHK-SIAT recording tool supplied by the Institute of four parts of Shandong dialect (namely Xiqi Region, Xilu Linguistics, Chinese Academy of Social Sciences (CASS). Region, Donglai Region and Dongwei Region). Up until now, The speech waveform was sampled at a rate of 16,000 Hz and Qian’s The Study of Shandong Dialect, an overall study on the resolution was 16 bit. Shandong dialect and its sub-dialects enjoys more authorities Annotation and pitch modification of the test syllables in some aspects of JND [2]. The collection of Wu Yonghuan’s were done with the help of software Praat 4.4.20 and Praat 6.0. study [3] on tones provides great reference for later studies. F0 values and durations were extracted for the first hand Focusing on tone sandhi types, many contributions which through Praat scripts provided by CASS. As pitch is based on are dedicated to the phonetic and phonological analyses in the fundamental frequency (F0) in vocal cord vibration, F0 Chinese (different dialects) have been made over the last values and durations were extracted for the first hand through thirty years. To our best knowledge, the previous studies have Praat script. Here, beginning and ending curves of an F0 explored Jiaonan tones on the basis of aural judgment and contour which are not easy to perceive by ear and do not there are all descriptive-oriented other than explanatory ones. influence the general tone, were neglected, leaving only the The main works include Wu Yonghuan [3], Jia Quanhong [4] tone-bearing sections taken into consideration. 10 points of F0 and Tangshun [5]. In their studies, the citation tones and tone values for each syllable were measured at equal intervals from changes are illustrated as in the following tables. the onset to the end of the whole F0 duration. Means of 10 points of pitch values for each tone type were obtained in Table 1. Previous descriptions on tone changes in disyllabic SPSS 13.0. sequences in JND Accordingly, the shape and mean pitch values of each tone type can be seen in Figure 1. Since the primary sources TSP Wu (2001) Tang (2007) Jia (2005) on the tones of JND adopted the five-scale system of Chao [6], I refer to the same method for convenience. T1+T1 213+213→ 21+213 214+214→ 24+214 213+213→ 42+213 T2+T4 42+21→ 55+21 53+31→ 55+31

T3+T1 55+213→ 42+213 55+214→ 53+214 55+213→ 42+213

T3+T3 55+55→ 42+55 55+55→ 53+55 55+55→ 42+55

T4+T4 21+21→ 213+21 31+31→ 214+31

T4+T1 211+213→ 42+213 (T1=yinping; T2=yangping; T3=shangsheng; T4=qusheng)

Figure 1: F0 Contours of the Four Citation Tones in JND (Horizontal axis: ten points of F0; Vertical axis: five scales)

1 1

Discrepancies between our result of T2 (51) and the yinping, while a slight falling tone in front of another previous results of Wu’s and Jia’s 42 and Tang’s 53 makes Shangsheng. T3 also undergoes changes in T3+T2 distinctions. So that of T4. But 41 may lead to totally different combinations, with a falling-rising shape. When two T4 are tonal representation from that of 21 or 31 in Wu[3] and Tang adjacent, the former shows a little tendency of falling, but it [5]. The differences between the two high falling tones in our should be treated as a high level tone according to Yip’s[7] results need further explanation. statement that level tones in Chinese dialect always have a Given the four-citation-tone system of JND, there are 16 lowering tendency in the end. combinational possibilities shown in four figures below. (for convenience of further research, the neutral-tone 5 disyllabic 3. Structural Analysis compounds were examined in the experiment, but neither the phonetic analysis nor the theoretical explanation has been T2 and T4 are both high falling tones. An assumption would done yet.) be probably efficient to explain the merging of T4 into T2, for which rich evidence is available from Wu’s [3] study on discussing the splitting and merging of the origins of tonal categories in system. Since tone sandhi plays a crucial role in the course of transition and the merging of some tone sandhi patterns will eventually result in the merging of the citation tones [8], it is necessary to reconsider about the tonal values of the second syllables in the disyllabic sequence, especially that of T2 and T4. Through observing the shape of the second syllables in disyllabic sequences, the following facts are available: Figure 2: T1+Tx (1) For T2 shapes, there are convex radians before the fourth point, at which begin sharp falling. 451 and 341 might be the only unlikeness. (2) T4 performs as 31 in the final position of the sequences, although the starting points are a little higher than 3 when T4 is after T2 or T3. From the aspect of co-articulation, we can say that tones of the second syllable keep constant in disyllable combinations in Jiaonan dialect. Citation tone and base tone are holding indispensable part of tone sandhi research. Citation tone and base tone are holding indispensable part of tone sandhi research [9]. It is true for the cases in which a Figure 3: T2+Tx same citation tone originates from different base tones in some Chinese dialects and then the underlying difference surfaces again in tonal sandhi form [10]. On this ground, it is reasonable to take the tones of second syllable as the base tones and underlying tones of Jiaonan dialect. Thus, six tone changes can also be well explained. In the underlying form, there are four tones, but in the surface form, T4 merges into T2.

Table 2. Tonal values of four base tones Figure 4: T3+Tx Tone Type T1 T2 T3 T4 Pitch 213 51 44 31

Table 3. Tone changes

S2 S1+S2 T1 T2 T3 T4

T1 42+T T+T T+T T+T Figure 5: T4+Tx T2 T+T T+T T+T 34+T S1 From the four figures above, it is proved that Jiaonan T3 52+T 213+T 42+T T+T dialect belongs to the right-syllable dominant type. The T4 T +T T +T T +T 44+T former T1 in a T1+T1 sequence changes from a falling-rising (“T” represents the unchanged ones.) shape into a high-falling one. In other forms, T1 remains relatively stable. T2 turns into a rising tone when it is followed by T4 while T2 in the other three types keeps its original forms. The only high level tone shows a quite active quality. T3 becomes into a relatively high-falling tone before

2 2

4. Tonal Representation (5c) *LEVEL: High/Lower level tones are not allowed. (5d) *X.Y ((X,Y∈{H.L},X≠Y): The final toneme and the To illustrate the tonal structure directly and clearly, Bao’s initial toneme of the adjacent syllables should be tonal representation model [11][12] is employed to serve as a identical. reasonable reference here, taking syllable as the TBU and (5e) *H.HH: The final toneme of the first syllable should not Yip’s [13] [±U] indicating the high and low registers. be high when followed by a high level tone. Six tone sandhi changes in Jiaonan dialect can be As the constraints have been assembled, the next step is to represented as follows: fix the ranking hierarchy to obtain the optimal output from the (3) [-U, LH][-U, LH] →[+U, HL][-U, LH] set of candidate outputs. One analytical strategy employed in [+U, HH][-U, LH]→[+U, HL][-U, LH] OT framework is to examine interactions of two constraints at [+U, HH][+U, HH]→[+U, HL][+U, HH] a time to get relative rankings, which can be couched in the [+U, HH][+U, HL]→[-U, LH][+U, HL] comparative tableau format introduced by Prince& Smolensky [+U, HL][-U, HL]→[+U, LH][-U, HL] [17]. It is gradient assessment that plays an important role to [-U, HL][-U, HL]→[+U, HH][-U, HL] distinguish the degree of constraint violation. The greater the mismatch, the higher degree the violation is.

Since Jiaonan dialect is right-prominent, it is the case that Obviously, the sandhi processes illustrated in (3) are in tonal changes only occur in the left position of disyllabic course of various kinds of tonal alternations. There is no sequences. RM-IDENT-IO-T will be never violated in register sandhi in Jiaonan dialect and tone sandhis occur at whatever context in Jiaonan dialect and it should rank at the both levels of register and contour. The tonal sandhi in T4+T4 highest position of the ranking hierarchy, whereas IDENT-IO- altering from a falling tone to a level tone is in line with one T(R) and IDENT-IO-T(C) the lowest position, for violations of the phonological processes, that is Leveling, summarized occur frequently to them while tonal alternations happen to by Hashimoto[14] and Bao[15], which is more specific in the initial syllables. The sandhi form of T3+T3 is in line with Dongwei Region of Shandong dialect. the phenomenon of TTSR(Third Tone Sandhi Rule)[18], with Here, several crucial points about disyllabic tone sandhi the preceding third tone changed to T2, rather than a possible features of Jiaonan dialect can be concluded: Firstly, the tone rising tone under the constraints of OCP-T and *LEVEL. This sandhi forms of disyllabic sequences show Jiaonan dialect constraint merely goes straight for the alternation of T3+T3 belongs to right prominent tone sandhi. Secondly, two and will never be violated in other conditions. Thus, it is identical registers are not allowed in adjacent position and the reasonable to rank *H.HH before OCP-T and *LEVEL. left one is bound to alter. However, the sandhi form of Here, one comparison between *H.HH and *[+U][+U] is adjacent high level tones is an exception. Thirdly, two made as an example to get a ranking hierarchy. identical tones are prohibited in a sequence and tone in the Putting *H.HH in a higher position than *[+U][+U] can initial position may undergo changes. This is essentially provide a good account for the maintenance of T2 (especially attributed to the motivation of OCP. Last but not least, the its register) in the first syllable of disyllabic sequences. final toneme of the first syllable and the initial toneme of the second syllable should be identical, unless they are both of Table 7．*H.HH>> *[+U][+U] upper register, with an exception that when preceding a high level tone, the high toneme would alter into its counterpart. In [+U, HL] [+U, HH] *H.HH *[+U][+U] addition, it is obvious to get the truth that if the register changes, the contour must change simultaneously, but not vice ) [+U, HL][+U, HH] * versa. [-U, LH] [+U, HH] * 5. OT Account of Tone Sandhi in JND If the order is put in another way, incorrect surface forms Constraints, major focus of research effort in OT, are ranked will be the optimal output. and violable. Generally, constraint conflict resulted from incompatible constraint requirements are solved by satisfying Table 8．*[+U][+U] >>*H.HH the requirement of a higher ranked constraint at the expense of violating the lower ones [16]. The constraint hierarchy [+U, HL] [+U, HH] *[+U][+U] *H.HH contains all universal constraints ranked in a language-specific way. [+U, HL][+U, HH] * Based on the five points resulted from the tonal #[-U, LH] [+U, HH] * representation, the faithfulness constraints and markedness constraints for the tone sandhi patterns in JND can be set as follows: Other constraints interactions can be obtained in the same (4) Faithfulness Constraints: way. It makes no strict difference between the expected (4a) RM-IDENT-IO-T: The feature of the tone in the right rankings of *H.HH and OCP-T, OCP-T and *[+U] [+U]. position keeps unchanged. According to the transitivity of ranking, the constraints are (4b) IDENT-IO-T(R): The register of the output tones are identical supposed to be put in the following ranking hierarchy. to that of input correspondents. (4c) IDENT-IO-T(C): The output contours are identical to that (6) RM-IDENT-IO-T >>*H.HH, OCP-T, *[+U] [+U] >> of input correspondents. *LEVEL >> *X.Y >> IDENT-IO-T(R), IDENT-IO-T(C) (5) Markedness Constraints: (5a) *[+U] [+U]: Two adjacent H/L-registered tones are No matter whether tone sandhi occurs or not, possible prohibited. disyllabic patterns should go through the same set of ranked (5b) OCP-T: Avoid adjacent identical tones at the tone level. constraints. Thus, it is clear to see if the optimal output corresponds to the attested data from the procedure of

3 3 evaluation. For the sake of efficiency and clarify, I would like following Tableaux) after evaluating all the possible disyllabic to choose some representative candidates (as shown in the sequences.

Table 9. T1+T1 Input: [-U, LH] [-U, LH] Output: [+U, HL] [-U, LH]

IDENT IDENT [-U, LH] [-U, LH] RM- IDENT *H.HH OCP-T *[+U][+U] *L *X.Y -IO- T -IO- T(R) -IO- T(C) a.[+U, LH] [-U, LH] *! * * b.[+U, LL] [-U, LH] * * *

)c.[+U, HL] [-U, LH] * **

d.[-U, LH] [-U, LH] *! * *

e.[-U, HL] [-U, LH] * **

f.[-U, LH] [+U, HL] ***! * **

(c) win over the other candidates listed in the above tableau, for it performs better in satisfying the constraint hierarchy, while the other competitors violate the higher-ranked constraints.

Table 10. T3+T3 Input: [+U, HH] [+U, HH] Output: [+U, HL] [+U, HH]

RM- IDENT IDENT IDENT [+U, HH] [+U, HH] *H.HH OCP-T *[+U][+U] *L *X.Y -IO- T -IO- T(R) -IO- T(C)

a.[+U, HH] [+U, HH] *! *! * *

b. [+U, LH] [+U, HH] *! * * *

)c.[+U, HL] [+U, HH] * * * *

d.[-U, LH] [+U, HH] *! * * *

e.[-U, HH] [+U, HH] *! *! * *

f.[-U, HL [+U, HL] *! *! * * **

Comparing with the optimal outputs listed in the tableau (9), (c) in Table 10 gets more violation marks, and the number equals to that of (b) and (d). But the later two both violate the even higher-ranked constraint *H.HH at a time. (c) should be the chosen as the optimal output. The unchanged patterns can also be attested with the same constraint ranking.

Table 11. T1+T2

Input: [-U, LH] [+U, HL] Output: [-U, LH] [+U, HL]

RM- IDENT IDENT IDENT [-U, LH] [+U, HL] *H.HH OCP-T *[+U][+U] *L *X.Y -IO- T -IO- T(R) -IO- T(C)

)a. [-U, LH] [+U, HL]

b.[-U, LL] [+U, HL] * * *

c.[-U,HH] [+U, HL] * *

d.[-U, HL] [+U, HL] *! * **

e.[+U, LH] [+U, HL] *

f. [+U, HL] [+U, LH] *! * * ****

Table 11 shows that those disyllabic sequences that do not Acoustic experiment is the basis of the study. The results undergo tone sandhi fit the constraints ranking hierarchy as suggest that both of yangping and qusheng are high falling well. Some violate no constraints, while some stay with the tones in citation forms and they should be of the same least violations. category, which thus proves the trend of transition from the four-tone system to the three-tone system in Jiaonan dialect. 6. Conclusion Although there are three surface citation tones in Jiaonan dialect, there are four base tones indicated by the disyllabic The paper presents a detailed phonetic description of tonal sequences, which are taken as the underlying tone forms. In information in Jiaonan dialect and a phonological explanation the process of analyzing the six kinds of disyllabic tone sandhi, of the experimental results within the framework of totally eight constraints are involved, including three Optimality Theory. faithfulness constraints and five markedness constraints. All

4 4 of them have been proved by other scholars, or can be found Rutgers University & University of Colorado, Boulder. in many other languages (for example: OCP, IDENT-IO) or in Revised version published in 2004 by Blackwell some northern Mandarin language (for example: *[±U][ ±U], publishers, Malden, MA, Oxford, UK; Carlton, Australia, *H.HH for Dialect), exhibiting the universality. 1993/2004. Through the interaction of the eight constraints, the tone [18] Chen, M. Y., Tone Sandhi: Patterns Across Chinese sandhi phenomena are well explained, showing the universal Dialects. Oxford University Press, Oxford, 2000. explanatory power of OT and may contribute to analysis of tone sandhi in Chinese topolects.

7. Acknowledgements The present analysis is financially sponsored by the Ministry of Education of the People’s Republic of China. Project name: Experimental Phonetic Study and Optimality Theoretical Analysis of Tonal Sandhi Patterns of Shandong Dialects (granted number: 11YJA740114).

8. References [1] Qian, Z. Y., et al., “Divisions of Shandong Dialect,” Fangyan, Vol. 4, 1985. [2] Qian, Z. Y., et al., The Study of Shandong Dialect. Qilu Press, , 2001. [3] Wu, Y. H., “The Study on the Tones of Shandong Dialect,” Ph. D. dissertation, Shandong University, Jinan, China, 2001. [4] Jia, Q. H., “A Study of the Jiaonan Dialect,” M.A. thesis, Suzhou University, Suzhou, China, 2005. [5] Tang, S., “A Study on the Phoneme of Jiaonan Dialect,” M.A. thesis, Shandong University, Jinan, China, 2007. [6] Chao, Yuen-Ren, “A system of tone letters,” La Maitre Phonetique 45: 24-27. Reprinted in Fangyan, Vol. 2, pp.81-82, 1980. [7] M. Yip, Tone. Cambridge University Press, Cambridge, 2002. [8] Zhao, R. X., and Gao, X.H., “A Review of “The Study of Shandong Dialect”,” Fangyan, Vol. 1, pp. 353-361, 2002. [9] Zhai, H. H., “Interaction of CR and CC in Chinese Tone Sandhi Patterning,” Ph. D. dissertation, Nankai University, Tianjin, 2009. [10] Ting, Pang-hsin, Diachronic Change in Chinese Tones. Proceedings of the Second International Sinology Conference, Taiwan: Academia Sinica, 1989. [11] Bao, Zhiming, “On the Nature of Tone,” Ph. D. dissertation, Massachusetts Institute of Technology, Massachusettes, United States, 1990. [12] Bao, Zhiming, The Structure of Tone. Oxford University Press, New York, 1999. [13] M. Yip, “Tone contours as melodic units: evidence from wuxi,” In Marjorie Chan and Thomas Ernst, (eds.), Proceedings of the Third Ohio State University Conference on Chinese Linguistics, 13-14 May, pp.37- 53, 1988, Reproduced by Indiana University Linguistics Club, 1989. [14] Yue-Hashimoto, and Ann Oikan, “Tone sandhi across Chinese dialects,” in Wang Li Memorial Volumes, Joint Publishing, pp. 445-474, 1987. [15] Bao, Zhiming. “Accentualism in Chinese,” Language and Linguistics, 2004. [16] J. McCarthy, Doing Optimality Theory. Oxford: Blackwell Publishing Ltd,. 2008. [17] A. Prince and P. Smolensky. Optimality Theory: Constraint Interaction in Generative Grammar. Ms.,

5 5