Content uploaded by Yike Yang
Author content
All content in this area was uploaded by Yike Yang on Oct 28, 2020
Content may be subject to copyright.
F0 patterns in Mandarin statements of Mandarin and Cantonese speakers
Yike Yang1, Si Chen1,2, Xi Chen1
1Department of Chinese and Bilingual Studies, The Hong Kong Polytechnic University,
Hong Kong S.A.R., China
2The Hong Kong Polytechnic University-Peking University Research Centre on Chinese
Linguistics, Hong Kong S.A.R., China
yi-ke.yang@connect.polyu.hk, sarah.chen@polyu.edu.hk, skye.chen@polyu.edu.hk
Abstract
Cross-linguistic differences of F0 patterns have been found
from both monolingual and bilingual speakers. However,
previous studies either worked on intonation languages or
compared an intonation language with a tone language. It still
remains unknown whether there are F0 differences in bilingual
speakers of tone languages. This study compared second
language (L2) Mandarin with Cantonese and first language
(L1) Mandarin, to test whether the L2 speakers of Mandarin
have acquired the F0 patterns of Mandarin and whether there
are influences from their L1 Cantonese. Different F0
measurements (including maximum F0, minimum F0, mean
F0 and F0 range) were examined with linear mixed-effects
models. Cantonese and Mandarin showed different F0
patterns, the source of which still requires further
investigation. The L2 Mandarin data resembled the F0 patterns
of Cantonese and were different from L1 Mandarin, for which
we provided different explanations: assimilation of L1
Cantonese and L2 Mandarin, the negative transfer from native
Cantonese, and similarities in the nature of tone languages.
Suggestions for testing these assumptions are proposed.
Lastly, our data provided conflicting results concerning the
role of gender in F0 pattern realisation.
Index Terms: speech production, second language, F0
pattern, Mandarin, Cantonese
1. Introduction
Languages differ in their F0 profiles, and studies have been
conducted to test how speakers of various languages make use
of F0-related cues, especially mean F0 and F0 range [1]–[6].
For example, Mandarin speakers were found to have higher
mean F0 and more F0 variation than American English
speakers in read speech, and the divergence was attributed to
the different intonation patterns of a tone language and a stress
language [1]. Even within the system of intonation languages,
English female speakers showed a higher F0 level and a wider
F0 span than German female speakers [4]. Several sources of
cross-linguistic F0 differences have been proposed, such as
differences in the intonation of languages, divergence in
cultural and social norms and differences in physiology of
speakers [2], but no consensus has been reached on this issue.
Also, such divergence in F0 profiles is observed for the two
languages of simultaneous bilinguals. In Japanese-English
bilinguals’ read speech, Japanese had higher sentence-initial
F0, higher first and second peaks as well as steeper declination
lines than English [3]. Moreover, there might be a gender
effect that interacts with the F0 profiles of languages, but the
mechanism behind remains unclear [2], [5]. For instance,
consistent cross-linguistic differences in F0 range were
observed in Welsh-English female bilinguals but not in male
bilinguals [2].
However, there is little information about F0 patterns of
the two languages in late second language (L2) learners, or
sequential bilinguals, who usually start to learn their L2 after
complete acquisition of their first language (L1) and are
associated with accented speech in their L2 [7]. Work into this
direction can deepen our understanding on issues of L2 speech
and foreign accent, which will shed light on the postulations
from models such as Speech Learning Model (SLM) [8], [9]
and Perceptual Assimilation Model (PAM) [10]. Although
such models were originally proposed to account for the
learning of L2 segments, it is possible to borrow the ideas into
the investigations on L2 prosody. Nguyễn [11] examined the
production of English by English and Vietnamese speakers.
The results showed that advanced Vietnamese speakers and
native English speakers shared similar F0 patterns, which had
greater F0 variation than beginners. The beginners’ English,
on the other hand, resembled the F0 patterns of Vietnamese,
and this was interpreted as being transferred from Vietnamese.
To the best of our knowledge, no thorough investigation
has been conducted on tone language pairs, so this study
attempts to fill this gap with data from speakers of two closely
related tone languages, namely, Mandarin and Cantonese.
Mandarin and Cantonese are tone languages from the
Sino-Tibetan language family, but they have different
phonological systems and are not mutually intelligible [12].
There are four lexical tones in Mandarin, among which Tone 1
(represented as 55) is the high-level tone and the remaining
ones are contour tones [13]. In Cantonese, there are six lexical
tones and three of them are level tones: high-level Tone 1
(55), mid-level Tone 3 (33) and low-level Tone 6 (22) [14]. To
accommodate more tones, Cantonese is shown to have a wider
F0 space than Mandarin (e.g. F0 range of male speakers for
the six Cantonese tones: 80-170 Hz [15]; F0 range of male
speakers for the four Mandarin tones: 90-140 Hz [16]).
Despite the phonological differences, Chinese characters are
used in the orthography of both languages.
The current study first compared the F0 patterns of
Mandarin and Cantonese, and then compared L2 Mandarin
with Cantonese and L1 Mandarin, to test whether the L2
speakers of Mandarin have acquired the F0 patterns of
Mandarin and whether there are influences from their L1
Cantonese. Also, the effect of gender was taken into account
in this study. The F0 profiles tested in this study included
maximum F0, minimum F0, mean F0 and F0 range.
Copyright © 2020 ISCA
INTERSPEECH 2020
October 25–29, 2020, Shanghai, China
http://dx.doi.org/10.21437/Interspeech.2020-25494163
2. Methods
2.1. Participants
Eleven native speakers of Mandarin (six females, five males;
aged: 24.72 f 4.39) and 12 native speakers of Hong Kong
Cantonese (six females, six males; aged: 20.41 f 2.97)
participated in a production experiment at the Speech and
Language Sciences Laboratory of the Hong Kong Polytechnic
University. The Mandarin participants started to speak
Mandarin from birth and had spent most of their lives in
Mandarin-speaking regions. The Cantonese participants speak
Cantonese as the native and dominant language and also
learned to speak Mandarin as a second language at an average
age of three years and nine months old. To assess their
language profile of Cantonese and Mandarin, the Cantonese
speakers completed a language background questionnaire,
which was an adapted version of Bilingual Language Profile
[17]. Information concerning language history, language use,
language proficiency and language attitudes was collected and
converted to both module and global scores, the results of
which suggested that the Cantonese speakers are Cantonese-
dominant but are also fluent in Mandarin. No participants
received formal musical training, and none reported any
history of speaking, hearing or language difficulty.
2.2. Materials
The experiments were designed for a larger project on
Mandarin and Cantonese prosody and part of the Mandarin
data has been reported in [18]. Short subject-verb-object
(SVO) statements were used as the test sentences. To make the
data of Cantonese and Mandarin comparable, exactly the same
syntactic structure was adopted for Cantonese and Mandarin
stimuli, with seven syllables in each sentence. A determiner
and a classifier (DET and CL; both monosyllabic) were added
to the beginning of each sentence so that the stimuli make
more sense. We were more interested in the global F0 patterns
than the tonal effect, so for the remaining five syllables, only
Tone 1 in both languages was used in the sentences (it is very
difficult, if not impossible, to control the tones of the
determiners and classifiers in both languages). Sentences (1a)
and (1b) are examples of the Mandarin and Cantonese target
sentences, respectively.
(1) a. na4 wei4 yi1sheng1 he1 ka1fei1
DET CL doctor drink coffee
‘The doctor drinks coffee.’
b. go2 di1 can1cik1 zong1 faa1dang1
DET CL relative install lantern
‘The relatives install lanterns.’
The Mandarin speakers recorded six target sentences in
Mandarin, and the data were then labelled as ‘Man_L1’ (L1
Mandarin). The Cantonese speakers recorded four target
sentences in Mandarin and another four in Cantonese, which
were labelled as ‘Man_L2’ (L2 Mandarin) and ‘Can’ (L1
Cantonese), respectively. Three repetitions were collected for
each target sentence. In total, there were 198 tokens (6
sentences * 11 speakers * 3 repetitions) in the ‘Man_L1’ data
group, 144 tokens (4 sentences * 12 speakers * 3 repetitions)
in the ‘Man_L2’ data group, and 144 tokens (4 sentences * 12
speakers * 3 repetitions) in the ‘Can’ data group.
2.3. Data collection
The sentences, written in Chinese characters, were divided
into different blocks and randomly presented on a computer
screen in E-Prime 2.0 [19]. The sentences were elicited as
answers to questions the experimenter asked. This procedure
has the following advantages: 1) this semi-simultaneous
elicitation made the data collection more natural than read
speech; and 2) with such method, the experiment also
managed to control the recording materials and made them
comparable across speakers and languages. The dialogues
were recorded at a sampling rate of 44,100 Hz in Audacity
[20] on another computer. Only the answers were further
segmented and analysed.
This project has been approved by the Human Subjects
Ethics Sub-committee (HSESC) of the Hong Kong
Polytechnic University (Reference #: HSEARS20190102001).
All participants gave their written informed consent prior to
the recording sessions.
2.4. Data analysis
Each syllable of the target sentences was manually segmented
in Praat [21], and the maximum F0, minimum F0, mean F0
and F0 range were extracted using the ProsodyPro Praat script
[22]. The F0 values, originally measured in Hz, were
converted to semitones (st) individually, with mean F0 of each
speaker as reference [23].
As mentioned in Section 2.2, each sentence starts with a
determiner and a classifier, which are not in Tone 1. In our
analysis, we first created two subsets of data, one with the
complete sentences and another with the remaining five Tone
1 syllables only, and analysed each subset separately with
linear mixed-effects models using the ‘lme4’ package [24] in
R [25], [26]. According to the models, there was no significant
difference between the two subsets of data, so in the results,
we presented the sentences with five Tone 1 syllables only.
Because we wanted to examine the F0 patterns at both the
syllable and utterance levels, we further divided the subsets
into two groups. The F0 values were our dependent variable.
Group, Gender and their interaction were the fixed effects.
Sentence, Repetition, Speaker and Syllable (for the syllable
level only) were included as the random effects. P-values were
obtained by likelihood ratio tests of the full model with the
effect in question against the model without the effect in
question. If the effect of Gender was significant, we would
separate the data into two subsets of female and male
speakers. Also, to further test the difference among the groups,
pairwise comparisons of different language groups were
conducted. The figures were plotted with the ‘ggplot2’
package [27].
3. Results
3.1. F0 patterns at the utterance level
We first fit models for the F0 patterns at the utterance level
and plotted the results in Figure 1. There were main effects of
Group for minimum F0, F0 range and mean F0 (ps < .001)
and main effects of Gender for the maximum F0 and
minimum F0 (ps < .002). There was no interaction of Group
and Gender across all the variables.
For female speakers, main effects of Group were found for
minimum F0, F0 range and mean F0 (ps < .005). Pairwise
comparisons suggested that the L1 Mandarin data had lower
4164
maximum F0 than the Cantonese data (p = .049) and higher
minimum F0 and mean F0 than the Cantonese and L2
Mandarin data (ps < .006). The Mandarin female speakers also
had a smaller F0 range than the other groups (ps < .003). No
differences were found between the Cantonese and L2
Mandarin of the female speakers. For male speakers, main
effects of Group were found for minimum F0, F0 range and
mean F0 (ps < .019). The Mandarin male speakers showed
higher minimum F0 and mean F0 (ps < .001) as well as a
smaller F0 range (ps < .02) than the other two groups. There
was no difference between the Cantonese and L2 Mandarin
data of the male speakers.
Figure 1: F0 values of the utterances.
At the utterance level, the variables Group and Gender
affected the F0 values. A closer examination suggested that
the L1 Mandarin data differed from the Cantonese and L2
Mandarin data in having higher minimum F0 and mean F0 and
a smaller F0 range, and there was no difference in the F0
profiles of the Cantonese and L2 Mandarin data. Similar
patterns were found in female and male speakers.
3.2. F0 patterns at the syllable level
We then employed the linear mixed-effects models to analyse
the F0 patterns at the syllable level, the results of which are
presented in Figure 2. There were main effects of Group for
the maximum F0, minimum F0, F0 range and mean F0 (ps <
.001) and main effects of Gender for the maximum F0,
minimum F0 and mean F0 (ps < .012). Interactions of Group
and Gender were also found for F0 range and mean F0 (ps <
.043).
For female speakers, there were effects of Group for the
minimum F0, F0 range and mean F0 (ps < .001). The L1
Mandarin data exhibited higher maximum F0 than the L2
Mandarin data (p = .036) and higher minimum F0 and mean
F0 than the Cantonese and L2 Mandarin data (ps < .001).
Also, there was a smaller F0 range for the L1 Mandarin
speakers (ps < .001). Again, no difference was found in the
Cantonese and L2 Mandarin of the female speakers. For male
speakers, significant effects of Group were also found for the
minimum F0, F0 range and mean F0 (ps < .001). The L1
Mandarin data were highest in minimum F0 and mean F0 (ps
< .001), and the L2 Mandarin data were lower than the
Cantonese data in minimum F0 (p = .049). In terms of F0
range, L1 Mandarin was lower than L2 Mandarin (p = .001)
and Cantonese (p = .057), and L2 Mandarin was higher than
Cantonese (p = .009).
Figure 2: F0 values of the syllables.
At the syllable level, the variables Group and Gender also
affected the F0 values. The female data at the syllable level
resembled those at the utterance level, where the L2 Mandarin
and Cantonese data showed very similar patterns and the L1
Mandarin data revealed systematic differences with the other
two groups. For male speakers, the three groups of data
showed noticeable differences, although the L1 Mandarin data
had the highest mean F0 and the smallest F0 range, consistent
with data from female speakers.
4. Discussion
The first aim of the study was to examine the F0 profiles of
Mandarin and Cantonese. Across different levels (syllable and
utterance) and genders, the general patterns identified were: 1)
Mandarin and Cantonese speakers had comparable maximum
F0 in their native languages, suggesting that they share similar
upper limit in their use of F0 (at least for the normalised data
according to our analysis); 2) Mandarin speakers showed
higher minimum F0 and consequently a smaller F0 range than
Cantonese speakers, which falls well in line with previous
findings that Cantonese speakers have a wider F0 span than
Mandarin speakers [15], [16]; and 3) Mandarin speakers had
higher mean F0 than Cantonese speakers.
There are six lexical tones in Cantonese, and three of them
are level tones. Among the four lexical tones in Mandarin,
there is only one level tone. It is thus reasonable for Cantonese
to have a larger F0 space than Mandarin to maintain the tonal
contrasts. Yet it is unclear why Cantonese has lower mean F0
than Mandarin. Further investigations are required as to
whether the difference is due to physiological constraints of
speakers, the relatively small sample size of this study, or
language-specific features of Mandarin and Cantonese.
Our second aim was to investigate the L2 Mandarin data
produced by Cantonese speakers and compare them with
native Mandarin and Cantonese. The data revealed that the L2
Mandarin resembled Cantonese in our measurements and were
strikingly different from L1 Mandarin. Specifically, compared
with the L2 Mandarin, the L1 Mandarin had a smaller F0
range but higher mean F0. Given the fact that all Cantonese
learners speak fluent Mandarin, it is surprising to find such
huge divergence from the L1 and L2 Mandarin data. In a
4165
recent study on Mandarin tone sandhi production [28],
Cantonese speakers also had lower F0 values than Mandarin
speakers under some tonal combinations. This consistent
pattern might be a result of the Cantonese learners’ perceptual
mapping of Cantonese and Mandarin tones [29], because there
are differences in the tonal systems of these two languages.
The category assimilation hypothesis (CAH) from SLM
may provide another possible explanation. According to SLM,
there is a common phonological space for storing L1 and L2
sounds [8]. The CAH suggests that an L2 sound perceived
similar to an L1 sound does not form a new category, and is
understood as a variant of the L1 sound at an allophonic level.
Such mapping eventually gives rise to a new merged category
in mental representation of a bilingual speaker. In our test, the
Cantonese speakers may have stored the prosodic features of
F0 in L1 Cantonese and L2 Mandarin in the common
phonological space and do not specifically differentiate them.
As a result, the features of F0 in L1 and L2 have undergone
assimilation processes and become more similar to each other.
To test whether this is the case, two directions can be
considered. The participants from the current study are fluent
L2 speakers of Mandarin, so subsequent studies may invite
learners with much less exposure to Mandarin and at lower
proficiency levels (beginner and intermediate levels). Also,
Mandarin sentences varying in F0 values can be used as the
stimuli for perceptual experiments to test Cantonese speakers’
sensitivity to Mandarin F0 levels.
However, one might argue that the similarities in
Cantonese speakers’ Mandarin and Cantonese actually reflect
their non-nativeness in Mandarin, although they speak fluent
Mandarin. This claim seems convincing as L2 speech
learning, especially L2 prosody, is very challenging, and the
attainment of native pronunciation is unlikely for late L2
learners [30]. It may be the case that our participants have not
fully acquired the features of Mandarin and still have trouble
with the production of Mandarin prosody. As a result, their L2
Mandarin production reveals a negative transfer from their
native Cantonese. However, only when Cantonese learners of
Mandarin at different proficiency levels are tested (as
suggested in the previous paragraph), can we verify whether
this claim holds.
Also, the similar F0 profiles in Cantonese speakers’
Cantonese and Mandarin observed in our study lend some
support to the physiology-based claim for cross-linguistic F0
differences. However, special caution must be taken before we
can reach a conclusion because previous studies showed
differences in bilinguals’ language pairs (e.g. English and
Korean [5], Welsh and English [2]). Unlike previous studies,
our data were collected from speakers of two tone languages,
and we thus provided new data regarding cross-linguistic
differences of F0 patterns. For tone languages, each syllable
determines its local F0 contour, which interacts with
intonation in the realisation of the global F0 contour in an
utterance. There are remaining issues to be examined. First, as
stated before, it is yet unknown whether of Mandarin and
Cantonese share similar intonation patterns. Second, because
individual differences are common in speech production [31],
it is necessary to test the F0 patterns of the two tone languages
within the same speaker and across different speakers. Third,
more speakers with diverse language backgrounds (especially
Cantonese-Mandarin bilinguals that are more Mandarin-
dominant) are needed to address this issue.
The third aim of the study was to examine the role of
gender in cross-linguistic F0 formation. The L1 Mandarin was
distinct from the L2 Mandarin and Cantonese regardless of
gender, but the effect of gender was found in Cantonese
speakers’ Cantonese and L2 Mandarin. For female speakers,
there was no difference between the Cantonese and L2
Mandarin data at both the utterance and syllable levels. For
male speakers, although the patterns of Cantonese and L2
Mandarin were similar at the utterance level, the two sets of
data diverged from each other at the syllable level. Previous
investigations on this issue provided conflicting results. While
some studies suggested that female bilinguals tended to switch
the F0 patterns consistently between the two languages [2],
others showed no effect of gender on the F0 patterns [3].
Contradicting previous findings, our results complicated the
issue of gender effect, which remains to be explored.
To conclude, Cantonese and Mandarin showed different
F0 patterns, the source of which still requires further
investigation. The L2 Mandarin data resembled the F0 patterns
of Cantonese and were different from L1 Mandarin, for which
we provided different explanations: assimilation of L1
Cantonese and L2 Mandarin, negative transfer from native
Cantonese, and similarities in the nature of tone languages.
Suggestions for testing these assumptions are proposed.
Lastly, our data provided conflicting results concerning the
role of gender in F0 pattern realisation.
5. Acknowledgements
The authors thank all the informants for their participation in
this project and the anonymous reviewers for their comments
on this paper.
6. References
[1] S. J. Eady, “Differences in the F0 Patterns of Speech: Tone
Language Versus Stress Language,” Lang. Speech, vol. 25, no.
1, pp. 29–42, 1982.
[2] M. Ordin and I. Mennen, “Cross-Linguistic Differences in
Bilinguals’ Fundamental Frequency Ranges,” J. Speech Lang.
Hear. Res., vol. 60, no. 6, p. 1493, 2017.
[3] C. Graham, “Fundamental Frequency Range in Japanese and
English: The Case of Simultaneous Bilinguals,” Phonetica, vol.
71, no. 4, pp. 271–295, 2014.
[4] I. Mennen, F. Schaeffler, and G. Docherty, “Cross-language
differences in fundamental frequency range: A comparison of
English and German,” J. Acoust. Soc. Am., vol. 131, no. 3, pp.
2249–2260, 2012.
[5] A. Cheng, “Cross-linguistic f0 differences in bilingual speakers
of English and Korean,” J. Acoust. Soc. Am., vol. 147, pp. EL67–
EL73, 2020.
[6] P. Keating and G. Kuo, “Comparison of speaking fundamental
frequency in English and Mandarin,” J. Acoust. Soc. Am., vol.
132, no. 2, pp. 1050–1060, 2012.
[7] M. S. Schmid and H. Hopp, “Comparing foreign accent in L1
attrition and L2 acquisition: Range and rater effects,” Lang. Test.,
vol. 31, no. 3, pp. 367–388, 2014.
[8] J. E. Flege, “Second Language Speech Learning: Theory,
Findings, and Problems,” in Speech Perception and Linguistic
Experience: Issues in Cross-language research, W. Strange, Ed.
Timonium, MD: York Press, 1995, pp. 233–277.
[9] J. E. Flege, “Interactions between the native and second-
Ianguage phonetic systems,” in An integrated view of language
development: Papers in honor of Henning Wode, T. Piske, A.
Rohde, and P. Burmeister, Eds. Trier: Wissenschaftlicher Verlag,
2002, pp. 217–244.
[10] C. T. Best and M. D. Tyler, “Nonnative and second-language
speech perception: Commonalities and complementarities,” in
4166
Second language speech learning: The role of language
experience in speech perception and production, M. J. Munro
and O.-S. Bohn, Eds. Amsterdam: John Benjamins, 2007, pp. 13–
34.
[11] A. T. T. Nguyễn, “F0 patterns of tone versus non-tone languages:
The case of Vietnamese speakers of English,” Second Lang. Res.,
vol. 36, no. 1, pp. 97–121, 2020.
[12] X. Zhang, “Dialect MT: A case study between Cantonese and
Mandarin,” in Proc. COLING 1998, 1998, vol. 2, pp. 1460–1464.
[13] Y. R. Chao, Mandarin Primer. Cambridge: Harvard University
Press, 1948.
[14] R. S. Bauer and P. K. Benedict, Modern Cantonese Phonology.
Berlin: Walter de Gruyter, 1997.
[15] A. L. Francis, V. Ciocca, L. Ma, and K. Fenn, “Perceptual
learning of Cantonese lexical tones by tone and non-tone
language speakers,” J. Phon., vol. 36, no. 2, pp. 268–294, 2008.
[16] Y. Xu, “Contextual tonal variations in Mandarin,” J. Phon., vol.
25, no. 1, pp. 61–83, 1997.
[17] D. Birdsong, L. M. Gertken, and M. Amengual, “Bilingual
Language Profile: An Easy-to-Use Instrument to Assess
Bilingualism,” COERLL, University of Texas at Austin, 2012.
https://sites.la.utexas.edu/bilingual/.
[18] Y. Yang and S. Chen, “Revisiting focus production in Mandarin
Chinese: Some preliminary findings,” in Proc. Speech Prosody
2020, 2020, pp. 260–264.
[19] W. Schneider, A. Eschman, and A. Zuccolotto, E-Prime User’s
Guide. Pittsburgh: Psychological Software Tools Inc, 2012.
[20] Audacity Team, “Audacity(R): Free Audio Editor and Recorder.”
2019.
[21] P. Boersma and D. Weenink, “Praat: doing phonetics by
computer.” 2015.
[22] Y. Xu, “ProsodyPro - A tool for large-scale systematic prosody
analysis,” in Proc. TRASP’2013, 2013, pp. 7–10.
[23] F. Nolan, “Intonational equivalence: an experimental evaluation
of pitch scales,” in Proc. ICPhS 2003, 2003, pp. 771–774.
[24] D. Bates, M. Mächler, B. Bolker, and S. Walker, “Fitting linear
mixed-effects models using lme4,” J. Stat. Softw., vol. 67, no. 1,
pp. 1–48, 2015.
[25] R Core Team, “R: A Language and Environment for Statistical
Computing.” R Foundation for Statistical Computing, Vienna,
Austria, 2018.
[26] RStudio Team, “RStudio: Integrated Development for R.”
RStudio, Inc., Boston, MA, 2016.
[27] H. Wickham, ggplot2: Elegant Graphics for Data Analysis.
Cham: Springer, 2016.
[28] S. Chen, Y. He, R. Wayland, Y. Yang, B. Li, and C. W. Yuen,
“Mechanisms of tone sandhi rule application by tonal and non-
tonal non-native speakers,” Speech Commun., vol. 115, pp. 67–
77, 2019.
[29] Y. C. Hao, “Second language acquisition of Mandarin Chinese
tones by tonal and non-tonal language speakers,” J. Phon., vol.
40, no. 2, pp. 269–279, 2012.
[30] D. Singleton, “The Critical Period Hypothesis: A coat of many
colours,” IRAL - Int. Rev. Appl. Linguist. Lang. Teach., vol. 43,
no. 4, pp. 269–285, 2005.
[31] Y. Yang and S. Chen, “Individual differences in Mandarin focus
production,” in Proc. ExLing 2020, 2020.
4167
