Tone Adaptation in Nuosu Yi Loanwords: �comparison between corpus and experiments

Yao Zhang

Cornell University, Department of Linguistics

PLC 50

02/28/2026

Relevant terms

Adaptation: the systematic modification of a borrowed form

English /baɪk/ → Japanese [baiku]

​

Faithfulness to source vs. native language markedness

​

2

Loanword adaptation…

• …is the systematic modification of borrowed forms;
• …loanwords sometimes conform to the phonological grammar of the borrowing language;
• …showing how speakers’ native productive knowledge interacts with external input.

​

This project focuses on tone adaptation.

3

Motivation: why Tone?

• Suprasegmental loans usually prefer native unmarked features (Kang & Katsuda 2024), contrasting with segmental adaptation, which often tolerates marked structures (e.g. Kenstowicz & Suchato 2006; Zuraw 2010).
• Tone-to-tone adaptation remains understudied.
• Tone-to-tone adaptation between similar tonal systems is even less studied.

​

4

Motivation: why Corpus + experiment?

Corpus:

• established loanwords
• conventionalized adaptation, community grammar

​

Experiment:

• real-time adaptation of new words, even nonce words
• individual variation

​

5

Take-aways

• Corpus study shows amplification of high-tone markedness in Nuosu Yi loanwords.
• Experimental adaptation partly reproduces this pattern, but with stronger source-faithfulness effects than the corpus.
• The balance between markedness and faithfulness varies across speakers.

6

Introduction:

Nuosu Yi & Mandarin

​

sustaining contact; similar tone systems

7

Geographic relation

8

• Recipient: Nuosu Yi

Spoken mainly in Liangshan, Sichuan

• Donor1: Chengdu Mandarin

Spoken in Chengdu and throughout the province

Used as a second language by non-Chinese speakers.

• Donor2: Standard Mandarin

National prestige language

​

​

Phonological comparison

9

​

​

High=5

Low=1

Five scale tone letters

Tone adaptation patterns in corpus

​

High-tone avoidance

​

10

Corpus data

462 disyllabic loanwords:

• extracted from Azimo (2019)
• double checked with the original source Han-Yi Dictionary (1979/1989)

​

Donor language: CdM (Pan 1990; Zhang 2025)

​

Recipient language: Standard Nuosu Yi

11

Example

12

​

​

Corpus results: tone adaptation heatmap

X-axis:

borrowing language

13

Y-axis:

borrowed language

High

High

Non-high

Non-high

Corpus results: non-high favored, high disfavored

CdM non-high → Yi 21

​

14

CdM high → Yi 33 or 21

• 42 → 33/21
• 45 → 33

​

Yi high tones rarely output in loanwords.

Statistical learning of native tone frequencies?

Corpus vs. Native

corpus:

21> 33 > 55 > 44

=

Markedness: 44>55>33>21

15

native: 33 > 21 > 55 > 44

=

Markedness: 44>55>21>33

Native high-tone markedness is amplified in loanwords!

Tone adaptation patterns in experiments

​

Weaker high-tone avoidance; individual variation

​

16

Method

• Participants: 20 native speakers of Yi
• undergraduate students from Southwest Minzu University, China
• m=9, f=11
• 4 dialects (6 Yinuo, 4 Sundi, 6 Adu, 4 Shengzha=Standard Nuosu Yi)

​

• Process: speakers listen to audio stimuli and produce the loanword forms
• Part 1: 48 CdM real words
• Part 2: 120 StM nonce words

17

Stimuli example: balanced for tone, coda, and frequency

18

Stimuli example: balanced for tone, coda, and frequency

19

Statistical modeling

20

Model: LoanTone ~ SourceTone + SyllablePosition + Frequency + Coda + Dialect +(1 | Stimuli) + (1 | Subject)

​

Result:

• Global predictor: SourceTone
• Secondary predictor: Dialect (Suodi & Yinuo differ from Shengzha & Adu significantly on 33)
• Weak tendency: Frequency
• Random effect: Subject

​

​

​

​

CdM 48RW results: CdM non-high → Yi non-high

Dominant patterns:

• CdM 213 → Yi 21
• CdM 21 → Yi 21

21

Subdominant patterns:

• CdM 213 → Yi 33
• CdM 21 → Yi 33

CdM 48RW results: CdM high → Yi non-high

Dominant patterns:

• CdM 42 → Yi 21/33
• CdM 45 → Yi 33/21

22

Subdominant patterns:

• CdM 42 → Yi 55
• CdM 45 → Yi 55

CdM 48RW results: more 55

Yi high tone 55 outputs more in the CdM real-word experiment than in the corpus.

23

StM 120NW results: StM non-high → Yi non-high

Dominant patterns:

• StM 214 → Yi 21

24

Subdominant patterns:

• StM 214 → Yi 33

StM 120NW results: StM high → Yi non-high

Dominant patterns:

• StM 51 → Yi 21
• StM 55 → Yi 33

25

Subdominant patterns:

• StM 51 → Yi 55
• StM 55 → Yi 55/44

StM 120NW results: more 55 & 44

Yi high tone 55 &44 output more in the StM nonce-word experiment than in the corpus.

26

Why more high tone(s) in the experiments?

Individual variation: H-avoidance & H-preference

27

2 strategies in CdM 48RW

H-avoidance

28

H-preference

​

Proportion of 45 → 55/44 adaptations by subject

48RW Inter-speaker variance: H-avoidance

• Native pattern

​

• Corpus pattern

29

48RW Inter-speaker variance: H-preference

• H-dominance

​

• H-preference

30

2 strategies in StM 120NW

H-avoidance

31

H-preference

​

Proportion of 55 → 55/44 adaptations by subject

120NW Inter-speaker variance: H-avoidance

• Native pattern

​

• Corpus pattern

32

120NW Inter-speaker variance: H-preference

• H-dominance

​

• H-preference

33

Individual-level variation contributes to the community-level difference

34

​

Conclusion

​

35

Conclusions on Nuosu Yi loanwords

Corpus Results:

• Established loanwords systematically prefer non-high tones and avoid high tones.
• Reflects amplification of high-tone markedness from native phonology.

Experimental Results:

• Real-time adaptations show substantial inter-speaker variation.
• Most speakers replicate the established patterns to different degrees.
• A minority favor high-tone outputs, showing stronger faithfulness to the source tone.

​

​

​

36

Contributions to loanword phonology

• Loanwords are not necessarily more marked than native lexicon: tone adaptation can amplify native markedness.

​

• Corpus vs. online adaptation reveal different grammatical states: corpus patterns reflect a stabilized community grammar; experimental data reveal active competition between constraints.

​

• Loanword grammars are not categorical: adaptation reflects gradient constraint competition; speakers differ in how they weigh markedness vs. faithfulness.

​

37

Acknowledgement

I am deeply grateful to:

• the Nuosu Yi participants, consultants (Qiu, Fuyuan; Liu, Xiaomei), and community members;
• my advisor Jennifer Kuo and committee members Draga Zec, Abby Cohn;
• Cornell Phonetics Lab

The project is supported by Cornell East Asia Program Travel Grant, East Asia Program International Research Travel Grant, and Timothy Murray Graduate Travel Grant

38

Thank you!

39

Selected references

• Bradley, D. (1990). The Status of the 44 Tone in Nosu. La Trobe. Journal contribution.
• Kang, Y., & Katsuda, H. (2024). Loanword phonology. In K. Nasukawa, B. Samuels, G. Schwartz, & M. Törkenczy (Eds.), Companion to Phonology (pp. 1-28). Wiley Blackwell.
• Katsuda, H. (2025). A probabilistic model of loanword accentuation in Japanese. Phonology, 42, e9, 1-26.
• Kenstowicz, M., & Suchato, A. (2006). Issues in loanword adaptation: A case study from Thai. Lingua, 116(7), 921-949.
• Zuraw, K. (2010). A model of lexical variation and the grammar with application to Tagalog nasal substitution. Natural Language & Linguistic Theory, 28(2), 417-472.
• 阿子莫小英. (2019). 凉山彝语中的汉语借词研究. 西南民族大学硕士论文.
• 潘正云. (1990). 现代凉山彝语音译词规范初探. 西南民族大学学报: 人文社会科学版, (4), 33-39.

​

40

Q&A

41

• “Sichuanese/Sichuan Madarin” is often synonymous with the Chengdu-Chongqing (Cheng-Yu) dialect.

​

• The map shows dialectal distribution among Chinese speakers.

​

• Cheng-Yu (sub)dialect is also used by many minority groups as a second language, including Nuosu Yi people.

Cheng-Yu

Min-Chi

Jiang-Gong

Ya-Gan

Corpus source tone distribution: �roughly balanced for tone

42

Pnon-high: 0.52

Phigh: 0.48

Corpus

43

​

​

​

Chengdu Mandarin 48 RW by subject

44

Standard Mandarin 120 NW by subject

45

2 strategies in CdM 48RW

H-avoidance

46

H-preference

​

48RW Inter-speaker variance: H-avoidance

• Native pattern

​

• Corpus pattern

47

48RW Inter-speaker variance: H-preference

• H-dominance

​

• H-preference

48

2 strategies in StM 120NW

H-avoidance

49

H-preference

​

120NW Inter-speaker variance: H-avoidance

• Native pattern

​

• Corpus pattern

50

120NW Inter-speaker variance: H-preference

• H-dominance

​

• H-preference

51