Tone sandhi in Changsha Xiang: A possible tone-stress interaction?

Miao Zhang (University of Zurich)
Shuxiao Gong (Beijing Language and Culture University)
Chengyu Guo (Beijing Normal University (Zhuhai))

Goal

To investigate the phonetic properties of tone sandhi in Changsha Xiang, a Sinitic language with both left- and right-dominant tone sandhi.
To explore the possibility that the tone sandhi in Changsha Xiang reflects metrical prominence (stress) at the word level.

Background

Changsha Xiang

A Sinitic language of the Xiang group.
Spoken in Changsha, the capital city of Hunan Province, China.
Has a 6-tone system.

The location

Changsha Xiang tones

Tone	Tone category	Example	Meaning
T13	阳平 Yangping	头 /təʊ˩˧/	‘head’
T21	阳去 Yangqu	豆 /təʊ˨˩/	‘bean’
T24	入声 Rusheng	毒 /təʊ˨˦/	‘poison’
T34	阴平 Qusheng	都 /təʊ˧˦/	‘capital’
T42	上声 Shangsheng	陡 /təʊ˦˨/	‘steep’
T45	阴去 Yingqu	斗 /təʊ˦˥/	‘fight’

The tones in isolation

Sociolinguistic notes on the tones

T13 (Yangping) and T24 (Rusheng) are merging in young speakers raised in the city, either due to similar tonal contours or due to the influence of Mandarin.
Some T21 syllalbes are merging with T45 due to the influence of Standard Mandarin.
The falling of T42 (T42 -> T44) is disappearing in young speakers.

Tone sandhi in Changsha Xiang

Tone sandhi: a phonological process where the tone of a syllable changes based on the context in which the tones occur.
Changsha Xiang has a position-sensitive tone sandhi system, exhibiting both left- and right-dominant patterns.

Tone sandhi in Changsha Xiang

Bao et al. (1998), Zhong (2003), Zhong (2007):
- Tones are partially neutralized and leveled in the non-dominant syllables in disyllables.
  - T45 -> T5
  - {T42, T24} -> T4
  - {T13, T34} -> T3
  - T21 -> T2
- Paradigmatic neutralization (or “default insertion”, see J. Zhang (2007)), a type of tone sandhi usually found in right-dominant tone sandhis across Sinitic languages.

Structure-dependent tone sandhi

Structure	Word	Meaning	Tone sandhi
Subj-pred	牛贵 [lʲəu¹³⁻³.kʷeɪ⁴⁵]	‘cows are expensive’	Right
Verb-result	磨破 [mo¹³⁻³.pʰo⁴⁵]	‘to wear out’	Right
Verb-obj	停电 [tin¹³⁻³.tʲẽ⁴⁵]	‘to cut off electricity’	Right

(Lin, 2011)

Structure-dependent tone sandhi

Structure	Word	Meaning	Tone sandhi
Coordinative	呈现 [t͡sən¹³.ɕẽ⁴⁵⁻⁵]	‘to demonstrate’	Left
Modifier-head	咸菜 [xã¹³.t͡sʰaɪ⁴⁵⁻⁵]	‘salted vegetables’	Left

(Lin, 2011)

Structure-dependent tone sandhi is also found in other Sinitic languages.
- E.g., Shanghai Wu (J. Zhang & Meng, 2016).

Structure-dependent tone sandhi

A famous minimal pair:

炒饭 /t͡sʰaʊ^42-4.fã²¹/ ‘to stir-fry rice’ (V-Obj)
炒饭 /t͡sʰaʊ⁴².fã^21-1/ ‘fried rice’ (Noun)

However,

Things are more complicated beyond disyllables.

干燥 /kã³⁴.tsʰaʊ^45-4/ ‘dry’ (Coordinative)
烘干 /xən^34-3.kã³⁴/ ‘to dry’ (V-Result)
风干 /xən³⁴.kã^34-3/ ‘to air dry’ (Manner-V)
烘干机 /xən³⁴.kã^34-3.t͡ɕi_z^34-3/ ‘drying machine’ (Noun)
风干鸡 /xən³⁴.kã^34-3.t͡ɕi_z^34-3/ ‘dry-cured chicken’ (Noun)

Tone sandhi at what prosodic level?

Previous documentation may have conflated word- and phrase-level tonology in Changsha Xiang.
There are words in Changsha Xiang in which the final syllable is always dominant, regardless of the syllable count.
E.g., proper nouns.
- 文明 /wən¹³.min^13-3/ ‘civilization/civilized’.
- 文明 /wən^13-3.min¹³/ ‘Civilization (A video game series)’.

Common noun vs. person names

This applies to person names as well.

红旗 /xɤn¹³.t͡ɕi_z^13-3/ ‘red flag’ (Common noun)
洪齐 /xɤn^13-3.t͡ɕi_z¹³/ ‘Hong Qi’ (Person name)
洪明齐 /xɤn^13-3.min^13-3.t͡ɕi_z¹³/ ‘Hong Mingqi’ (Person name)
乌鲁木齐 /uᵝ^34-3.ləʊ^42-4.mo^24-4.t͡ɕi_z¹³/ ‘Urumqi’ (Place name)

I’ll focus on the tone sandhi patterns in disyllabic common nouns and person names to avoid the confound of phrase-level prominence.

Languages with both tone and stress

Stress and tone can co-occur in the same language (Hyman, 2006).
Changsha Xiang tone sandhi resembles tone-stress interactions in languages such as Swedish (Riad, 2012), Goizueta Basque (Hualde et al., 2008), Lamba (Bickmore, 1995), Curaçao Papiamentu (Remijsen & Heuven, 2005), Itsunyoso Triqui (DiCanio, 2008), Quiaviní Zapotec (Peón, 2010) and Serbian (Zec & Zsiga, 2010).

Languages with both tone and stress

In these languages, full tonal contrast is only found in the stressed syllables.
The unstressed syllables are either toneless (Swedish, Serbian, Lamba, and Goizueta Basque) or have a reduced tonal inventory (Itsunyoso Triqui and Curaçao Papiamentu).

Languages with both tone and stress

E.g., Itsunyoso Triqui (DiCanio, 2008)

9 lexical tones.
Most words are disyllabic, and final syllables are obligatorily stressed.
6 tones (both level and contour tones) can occur in the final stressed syllable,
But only 4 level tones can occur in the initial unstressed syllables.

Itsunoyo Triqui tones (DiCanio, 2008)

Phonetics of tone-stress interaction

Duration was found to be the most reliable acoustic correlate of stress, probably because F0 has been utilized in the language’s tonal phonology.
- Curaçao Papiamentu (Remijsen & Heuven, 2005; Rivera-Castillo & Pickering, 2004), Pirahã (Everett, 1998), Thai (Potisuk et al., 1996), Itunyoso Trique (DiCanio, 2010), Ixcatec (DiCanio, submitted), Kurtöp (Hyslop, 2021), Central Pame (Torres & Herce, 2025), Tetsǫ́t’iné (Jaker & Howson, 2022), Chácobo (Tallman & Elías-Ulloa, 2020), Ei (Wang & Lin, 2016) and Fuzhou (Chan, 1985).

Phonetics of tone-stress interaction

In contrast, in non-tonal languages, F0 has been reported to be the most reliable acoustic correlate of stress (Gordon & Roettger, 2017).
Changsha Xiang has already shown tonal alternation, but does it also affect duration?
Very few studies investigated tone-stress interaction in Sinitic languages (Duanmu, 1995; Duanmu, 1999; Wright, 1983; Zhong, 2007).
Even fewer studies have explored its phonetic properties (Guo & Chen, 2022; Hu & Feng, 2024; Qing, 2014; Sui, 2016; Wang & Lin, 2016; Wen et al., 2023).

Phonetics of tone-stress interaction in Sinitic languages

Positive evidence:

Standard (Sui, 2016) and Chengdu Mandarin (Liu et al., 2022): Initial dominant syllables are longer.
Guo & Chen (2022): Dominant syllalbes are longer in Changsha Xiang (initial), Xiamen min (final), and Fuzhou min (final) but not in Chengdu Mandarin (initial).
Hu & Feng (2024): Dominant syllables are longer in Chongqing Mandarin (initial), Kunming Mandarin (final), and Xiamen Min (final).

Phonetics of tone-stress interaction in Sinitic languages

Negative evidence:

Wen et al. (2023): Dominant syllables are not longer in Shaoxing Wu.
Hsieh (2021): Metrical prominence is not perceptually salient in Standard Mandarin.

Research questions

Stressed syllables should show:
- longer duration,
- more distinct tonal contrast,
- less tonal coarticulation.

Research questions

Specifically, we ask the following questions:

How does tone sandhi affect duration?
How does tone sandhi affect tonal production and coarticulation in terms of F0 contour?
Can we observe tonal simplification (neutralization and leveling) in non-dominant syllables in phonetics?

Methods

Speech materials

For common noun trials, 4 sets of disyllabic words were created as the target words, with each set containing 36 words for each of the tonal combinations (144 in total).
For person-name trials, 2 distinct sets of disyllabic person names were used (72 in total).
Exp 1 tested only the person names, while Exp 2 tested only the common nouns.
Each trial is repeated twice in Exp 1, while four times in Exp 2.

Participants

16 native speakers of Changsha Xiang.
Participants have to have grown up and currently live in Changsha, and have to be able to distinguish T13 and T24 in order to be included in the study.
7 of the 16 participants completed both experiments, with 12 participants in each Exp.

Recording procedure

Recording took place in participants’ homes using a Zoom H4n recorder and a Shure SM35 headset condenser microphone.
They are instructed to read the trial sentences shown on a computer screen at a normal pace, and to repeat the sentences if they make any mistakes.

Data extraction

Syllable and rhyme boundaries were manually annotated in Praat.
Syllable duration was measured from the annotated boundaries.
F0 was extracted from 10 equidistant points in the rhyme of each syllable, and normalized into z-scores for each speaker.
- F0 values that are more than 2sd away from the speaker’s mean were excluded as outliers.

Data analysis

Linear mixed-effects models (LMM) were used to analyze the effects of tone sandhi on syllable duration.
Generalized additive mixed-effects models (GAMM) were used to analyze the effects of tone sandhi on 1) tonal production and 2) coarticulation.
GAMMs were fit separately for each tone in both analyses.

See details of the models in my paper: M. Zhang et al. (2026).

Data analysis

To investigate tonal coarticulation, I encoded T45 and T21 as H and L tones in the initial and final syllables.

Data analysis

Linear Discriminant Analysis (LDA) was used to analyze the tonal merging and leveling patterns.
To perform LDA, each f0 contour is fit with a cubic polynomial regression, and the coefficients of the fitted polynomial are used as features for LDA.
The mean intensity of each syllable and the syllable duration were also included.

See details of the models in my paper: M. Zhang et al. (2026).

Results

Syllable duration

Findings on syllable duration

There is always a difference between dominant syllables and non-dominant syllables across all tones.
The only insignificant difference is dominant or non-dominant syllable-internal comparisons.

GAMM results on tonal production

Findings on tonal production I

In dominant syllables,
- T21 and T42 exhibit a slightly smaller falling in s1.
- The rising tones (T13, T24, T34, and T45) show a rising contour only in s1.
In non-dominant syllables,
- T21 and T42 show a slightly smaller falling in s1.
- The rising tones (T13, T24, T34, and T45) are more or less leveled in s2.

GAMM results on tonal production

Findings on tonal production II

In s1,
- In dominant syllables, T21 and T13 are entirely lower.
- The other rising tones (T24, T34, and T45) show a steeper rising contour in dominant syllables.
- T42 is unaffected.
In s2,
- T21 reaches a lower F0 minumum in dominant syllables.
- T13, T24, T34 and T45 show rising contour only in dominant syllables.

Overall tone contours

GAMM results on anticipatory coarticulation

GAMM results on carryover coarticulation

Findings on coarticulation

Anticipatory coarticulation only occurs for T24 and T34. It is more prominent in dominant syllables.
Carryover coarticulation observed for T21, T24, T42, and T45 tones, but is more extensive in non-dominant syllables.

LDA results

Findings in LDA analysis

Prediction accuracy is higher in dominant syllables.
In final non-dominant syllables, the prediction accuracy of T13 and T24 becomes very low (chance level is 16.7%), suggesting tonal merging.
Initial non-dominant syllables behave similarly to dominant syllables, although the prediction accuracy is lower, probably due to less distinct F0 contours as a consequence of short duration.

Discussion

How does tone sandhi affect duration?

Dominant syllables are longer than non-dominant syllables (~ 100ms), consistent with the metrical prominence of the dominant syllables.

How does tone sandhi affect tonal production?

Contour tones retain their contour not only in dominant syllables, but also in initial non-dominant syllables, while the slope of the contour differs.
Final non-dominant syllables show significantly less F0 excursion, indicating tonal leveling.
LDA analysis suggests T13 and T24 do lose their tonal identity, yet the picture is more complex than categorical merging.

How does tone sandhi affect tonal coarticulation?

Only the carryover effect is extensively influenced by tone sandhi.
Non-dominant syllables show more carryover coarticulation than dominant syllables.

Asymetric tone sandhi

Left- and right-dominant tone sandhi in Changsha Xiang is asymmetric.
Paradigmatic neutralization (J. Zhang, 2007) is only found in the final non-dominant syllables, while those in initial non-dominant syllables only undergo tonal undershoot, with less F0 excursions than in dominant syllalbes due to short duration.

Asymetric tone sandhi

This is Similar to the Shanghai Wu tone sandhi, which is also asymetric:

Shanghai Wu left-dominant tone sandhi: the tone spreads from the initial syllable to the final syllable.
Shanghai Wu right-dominant tone sandhi: a pattern called paradigmatic neutralization (J. Zhang, 2007).
- Paradigmatic neutralization: some default surface tones are inserted.
- They are the four short-level tones (T2-5) in Changsha Xiang.

Asymetric tone sandhi

A similar pattern was found in Chongqing Mandarin (Qing, 2014) as well.

Only non-dominant syllablers demonstrate tonal leveling.
- E.g., T35 -> T33 when in non-dominant syllables.
Tones in dominant syllables can largely retain their underlying contours.

The possible cause of the asymmetry

Scenario 1:

The right-dominant tone sandhi is reusing a sandhi process for phrasal tonologly.
There might have been a word boundary between the two syllables.
Tones adjacent to a stronger boundary (word or phrase) are less likely to lose their tonal identity.

The possible cause of the asymmetry

Scenario 2:

The non-dominant syllable in right-dominant disyllables are word-initial.
Word-initial position has in general been considered a prosodically strong position such that it blocks tonal neutralization.
More data on tones in the word-initial syllables in Changsha Xiang multisyllabic words are needed to test this hypothesis.

Typology of tone sandhi in Sinitic languages

Paradigmatic neutralization is usually found in right-dominant tone sandhi (J. Zhang, 2007; J. Zhang, 2014).
In Changsha Xiang, it is the left-dominant tone sandhi that shows paradigmatic neutralization.
A typological hypothesis (in the context of Sinitic languages): In terms of the degree of grammaticalization: tonal spreading > paradigmatic neutralization > tonal undershoot.

Possible tone-stress interaction?

Multiple evidence suggests that the tone sandhi in Changsha Xiang may reflect metrical prominence at the word level:

Dominant syllables are much longer than non-dominant syllables, and have more distinct tonal production than non-dominant syllables.
Non-dominant syllables show more tonal coarticulation than dominant syllables.

More research, both phonetic and phonological, is needed to establish stress in Changsha Xiang.

Conclusion

Tone sandhi in Changsha Xiang is asymmetric, involving paradigmatic neutralization only in the final non-dominant syllables. This process may reflect metrical prominence at the word level, with dominant syllables being longer and having more distinct tonal production than non-dominant syllables. Further studies that investigate how other acoustic measures are influenced by tone sandhi are needed.

References

Bao, H., Cui, Z., Shen, R., & Wu, Y. (1998). The Dictionary of Changsha Dialect of Chinese. Jiangsu Education Press.

Bickmore, L. S. (1995). Tone and stress in Lamba. Phonology, 12(3), 307–341. https://doi.org/10.1017/S0952675700002542

Chan, M. K. M. (1985). Fuzhou Phonology: A Nonlinear Analysis of Tone and Stress (chinese, Dialect, Foochow) [PhD thesis, University of Washington]. https://www.proquest.com/docview/303427121/abstract/5A02E93F720A4748PQ/1

DiCanio, C. T. (2008). The Phonetics and Phonology of San Martín Itunyoso Trique [University of California, Berkeley]. https://arxiv.org/abs/1011.1669v3

DiCanio, C. T. (2010). Itunyoso Trique. Journal of the International Phonetic Association, 40(2), 227–238. https://doi.org/10.1017/S0025100310000034

DiCanio, C. T. (submitted). The phonetics of word-prosody structure in Ixcatec.

Duanmu, S. (1995). Metrical and Tonal Phonology of Compounds in Two Chinese Dialects. Language, 71(2), 225–259. https://doi.org/10.2307/416163

Duanmu, S. (1999). Metrical Structure and Tone: Evidence from Mandarin and Shanghai. Journal of East Asian Linguistics, 8(1), 1–38. https://doi.org/10.1023/A:1008353028173

Everett, K. M. (1998). The acoustic correlates of stress in Pirahã. The Journal of Amazonian Languages, 30.

Gordon, M., & Roettger, T. (2017). Acoustic correlates of word stress: A cross-linguistic survey. Linguistics Vanguard, 3(1), 1–11. https://doi.org/10.1515/lingvan-2017-0007

Guo, C., & Chen, F. (2022). Phonetic Realizations of Metrical Structure in Tone Languages: Evidence From Chinese Dialects. Frontiers in Psychology, 13, 945973. https://doi.org/10.3389/fpsyg.2022.945973

Hsieh, F. (2021). On the Existence of Word Stress in Mandarin Chinese. Studies in Prosodic Grammar, 7(1), 27–58.

Hu, Y., & Feng, H. (2024). Key Acoustic Cues for the Realization of Metrical Prominence in Tone Languages: A Cross-Dialect Study. Interspeech 2024, 3130–3134. https://doi.org/10.21437/Interspeech.2024-666

Hualde, J. I., Lujanbio, O., & Torreira, F. (2008). Lexical tone and stress in Goizueta Basque. Journal of the International Phonetic Association, 38(1), 1–24. https://doi.org/10.1017/S0025100308003241

Hyman, L. M. (2006). Word-prosodic typology. Phonology, 23(02), 225–257. https://doi.org/10.1017/S0952675706000893

Hyslop, G. (2021). Between Stress and Tone: Acoustic Evidence of Word Prominence in Kurtöp. Language Documentation, 15, 25.

Jaker, A., & Howson, P. J. (2022). An acoustic study of Tetsǫ́t’ıné stress: Iambic stress in a quantity-sensitive tone language. Phonology, 39(1), 1–39. https://doi.org/10.1017/S0952675722000069

Lin, H.-S. (2011). Changsha Tone Sandhi. Journal of Chinese Language Teaching, 8(2).

Liu, 春卉. C. 刘, Li, 可纯. K. 李, & Nolan, F. (2022). 论声调语言的节奏与重音模式 On the Speech Rhythm and Stress Pattern of Tone Languages. Journal of Sichuan University (Philosophy and Social Science Edition), 3, 151–163. https://wkxb.scu.edu.cn/zh/article/58937311/

Peón, M. E. C. (2010). The interaction of metrical structure, tone, and phonation types in Quiaviní Zapotec [PhD thesis, University of British Columbia]. https://doi.org/10.14288/1.0071253

Potisuk, S., Gandour, J., & Harper, M. P. (1996). Acoustic Correlates of Stress in Thai. Phonetica, 53(4), 200–220. https://doi.org/10.1159/000262201

Qing, W. (2014). An Analysis on the Disyllabic Tone Sandhi of Chongqing Dialect from the Perspectives of Word Stress and Tonal Coarticulation. National University of Singapore.

Remijsen, B., & Heuven, V. J. van. (2005). Stress, tone and discourse prominence in the Curaçao dialect of Papiamentu. Phonology, 22(2), 205–235. https://doi.org/10.1017/S0952675705000540

Riad, T. (2012). Culminativity, stress and tone accent in Central Swedish. Lingua, SI: Varieties of Pitch Accent Systems, 122(13), 1352–1379. https://doi.org/10.1016/j.lingua.2012.07.001

Rivera-Castillo, Y., & Pickering, L. (2004). Phonetic correlates of stress and tone in a mixed system. Journal of Pidgin and Creole Languages, 19(2), 261–284. https://doi.org/10.1075/jpcl.19.2.02riv

Sui, Y. (2016). The interaction of metrical structure and tone in standard Chinese. Dimensions of Phonological Stress, 101–122.

Tallman, A. J. R., & Elías-Ulloa, J. (2020). The acoustic correlates of stress and tone in Chácobo (Pano): A production study. The Journal of the Acoustical Society of America, 147(4), 3028–3042. https://doi.org/10.1121/10.0001014

Torres, C., & Herce, B. (2025). Word-prosodic typology of Central Pame: Acoustic evidence for stress and tone. The Journal of the Acoustical Society of America, 158(2), 1478–1490. https://doi.org/10.1121/10.0038802

Wang, X., & Lin, Y.-H. (2016). Metrical Structure and Tone Sandhi: Evidence from Ei Tonal Reduction. 5th International Symposium on Tonal Aspects of Languages (TAL 2016), 70–74. https://doi.org/10.21437/TAL.2016-15

Wen, X., Chen, Y., & Cheng, L. L.-S. (2023). Disyllabic tone sandhi patterns in Shaoxing Wu Chinese. ICPhS 023.

Wright, M. S. (1983). A metrical approach to tone sandhi in Chinese dialects [PhD thesis]. University of Massachusetts Amherst.

Zec, D., & Zsiga, E. (2010). Interaction of tone and stress in Standard Serbian. Formal Approaches to Slavic Linguistics (FASL), 18, 536–555.

Zhang, J. (2007). A directional asymmetry in Chinese tone sandhi systems. Journal of East Asian Linguistics, 16.

Zhang, J. (2014). Tones, Tonal Phonology, and Tone Sandhi. In The Handbook of Chinese Linguistics (pp. 443–464). John Wiley & Sons, Ltd. https://doi.org/10.1002/9781118584552.ch17

Zhang, J., & Meng, Y. (2016). Structure-dependent tone sandhi in real and nonce disyllables in Shanghai Wu. Journal of Phonetics, 54, 169–201. https://doi.org/10.1016/j.wocn.2015.10.004

Zhang, M., Gong, S., & Guo, C. (2026). Tone sandhi and tonal coarticulation in disyllabic sequences in Changsha Xiang. Journal of Phonetics, 114, 101472. https://doi.org/10.1016/j.wocn.2025.101472

Zhong, Q. (2003). Light Tones in Changsha Dialect. Fangyan, 3, 255–264.

Zhong, Q. (2007). A study of stress patterns in some major Chinese dialects. National University of Singapore.