Production of phonetic and phonological contrast by heritage speakers of Mandarin

Abstract

This study tested the hypothesis that heritage speakers of a minority language, due to their childhood experience with two languages, would outperform late learners in producing contrast: language-internal phonological contrast, as well as cross-linguistic phonetic contrast between similar, yet acoustically distinct, categories of different languages. To this end, production of Mandarin and English by heritage speakers of Mandarin was compared to that of native Mandarin speakers and native American English-speaking late learners of Mandarin in three experiments. In experiment 1, back vowels in Mandarin and English were produced distinctly by all groups, but the greatest separation between similar vowels was achieved by heritage speakers. In experiment 2, Mandarin aspirated and English voiceless plosives were produced distinctly by native Mandarin speakers and heritage speakers, who both put more distance between them than late learners. In experiment 3, the Mandarin retroflex and English palato-alveolar fricatives were distinguished by more heritage speakers and late learners than native Mandarin speakers. Thus, overall the hypothesis was supported: across experiments, heritage speakers were found to be the most successful at simultaneously maintaining language-internal and cross-linguistic contrasts, a result that may stem from a close approximation of phonetic norms that occurs during early exposure to both languages.

Full text

Production of phonetic and phonological contrast by heritage
speakers of Mandarin
Charles B. Chang
a),b)
University of Maryland, College Park, Center for Advanced Study of Language,
7005 52nd Avenue, College Park, Maryland 20742
Ya o Ya o
b)
Hong Kong Polytechnic University, Department of Chinese and Bilingual Studies, GH626,
Hong Hum, Kowloon, Hong Kong
Erin F. Haynes and Russell Rhodes
University of California, Berkeley, Department of Linguistics, 1203 Dwinelle Hall #2650, Berkeley,
California 94720
(Received 3 December 2009; revised 28 February 2011; accepted 1 March 2011)
This study tested the hypothesis that heritage speakers of a minority language, due to their child-
hood experience with two languages, would outperform late learners in producing contrast: lan-
guage-internal phonological contrast, as well as cross-linguistic phonetic contrast between similar,
yet acoustically distinct, categories of different languages. To this end, production of Mandarin and
English by heritage speakers of Mandarin was compared to that of native Mandarin speakers and
native American English-speaking late learners of Mandarin in three experiments. In experiment 1,
back vowels in Mandarin and English were produced distinctly by all groups, but the greatest sepa-
ration between similar vowels was achieved by heritage speakers. In experiment 2, Mandarin aspi-
rated and English voiceless plosives were produced distinctly by native Mandarin speakers and
heritage speakers, who both put more distance between them than late learners. In experiment 3,
the Mandarin retroflex and English palato-alveolar fricatives were distinguished by more heritage
speakers and late learners than native Mandarin speakers. Thus, overall the hypothesis was sup-
ported: across experiments, heritage speakers were found to be the most successful at simultane-
ously maintaining language-internal and cross-linguistic contrasts, a result that may stem from a
close approximation of phonetic norms that occurs during early exposure to both languages.
V
C2011 Acoustical Society of America. [DOI: 10.1121/1.3569736]
PACS number(s): 43.70.Kv, 43.70.Fq, 43.70.Bk, 43.70.Ep [AL] Pages: 3964–3980
I. INTRODUCTION
Although there exists a wide range of scholarship on the
linguistic competence of child first-language (L1) and adult
second-language (L2) acquirers, researchers have only begun
to examine the linguistic knowledge of heritage-language
speakers—that is, individuals whose current primary lan-
guage differs from the language they spoke or only heard as
a child (i.e., the heritage language or HL). HL speakers are a
group of interest because they often have a rich knowledge
of their HL, even when they do not actively speak the lan-
guage. Typical HL re-learners are predicted to have acquired
“nearly 90% of the phonological system” and “80% to 90%
of the grammatical rules” of the HL—a significantly more
extensive command of the language than second-year col-
lege L2 learners (Campbell and Rosenthal, 2000, p. 167).
Indeed, studies that have examined the phonological compe-
tence of HL speakers have found that childhood experience
with a minority language, even if merely overhearing, can
provide a significant boost to a speaker’s production and per-
ception of that language later in life in comparison to L2
learners with no prior experience (Tees and Werker, 1984;
Knightly et al., 2003;Oh et al., 2003). Similarly, studies that
have examined the grammatical competence of HL speakers
have found that they tend to be more native-like than L2
learners in their morphosyntax as well, although they none-
theless pattern differently from native speakers (Montrul,
2008;Au et al., 2008;Polinsky, 2008). There seems to be
something special about early linguistic experience acquired
in childhood, and this point has been made especially clear
in studies of HL phonology.
A. Heritage-language phonology
Studies of HL phonology have been conducted on a num-
ber of languages, including Armenian (Godson, 2003, 2004),
Korean (Au and Romo, 1997;Oh et al., 2002, 2003;Au
and Oh, 2009), Russian (Andrews, 1999), and Spanish (Au and
Romo, 1997;Au et al., 2002;Knightly et al., 2003;Oh and
Au, 2005;Au et al., 2008), the majority of this research coming
out of joint work by Au, Jun, Knightly, Oh, and Romo on HL
speakers of Korean and Spanish. In their series of studies,
which include acoustic measures such as voice onset time
(VOT) and degree of lenition, holistic measures such as overall
a)
Author to whom correspondence should be addressed. Electronic mail:
cbchang@umd.edu
b)
Also at: University of California, Berkeley, Department of Linguistics,
1203 Dwinelle Hall #2650, Berkeley, CA 94720.
3964 J. Acoust. Soc. Am. 129 (6), June 2011 0001-4966/2011/129(6)/3964/17/$30.00 V
C2011 Acoustical Society of America
Author's complimentary copy

accent ratings, and perceptual measures such as phoneme iden-
tification accuracy, the recurring theme is that HL speakers
tend to have a phonological advantage over L2 learners. How-
ever, whether HL speakers show an advantage over L2 learners
just in perception or in both perception and production of the
HL seems to be related to the nature of their HL experience. In
this regard, Au and colleagues have distinguished between
“childhood hearers” and “childhood speakers.”
Knightly et al. (2003), for example, focused on childhood
overhearers of Spanish—Spanish speakers who had regular
childhood experience with overhearing Spanish, but not with
speaking or being spoken to—and found that these childhood
overhearers were measurably better than L2 learners at pro-
ducing individual Spanish phonemes as well as whole Spanish
narratives. Similarly, Oh et al.(2003)found that individuals
with HL experience in Korean had a phonological advantage
over L2 learners of Korean; however, they examined not only
childhood hearers, but also childhood speakers who spoke
Korean regularly during childhood. Comparing these two HL
groups, they found that while childhood speakers were meas-
urably more native-like than L2 learners in both perception
and production of Korean, childhood hearers were more
native-like than L2 learners only in perception. This discrep-
ancy with the results of Knightly et al. (2003) was attributed
to two possible factors: the difference in average duration of
HL re-learning (longer in the case of the HL Spanish speak-
ers) and the difference in complexity between the two con-
trasts examined (a two-way laryngeal contrast in Spanish
between voiced and voiceless stops vs a three-way laryngeal
contrast in Korean among lenis, fortis, and aspirated
stops=affricates). In short, the findings of Au and colleagues
have suggested that previous HL speaking experience confers
an advantage in both production and perception of the HL,
and that previous HL listening experience confers an advant-
age in perception of the HL, even when this experience is lim-
ited to just the first year of life (Oh et al., 2010).
1
The benefit
conferred by HL listening experience in production of the
HL, however, appears to be mediated by additional factors.
Although studies on HL phonology have investigated
the authenticity of HL speakers’ production, few have explic-
itly examined the question of categorical merger—that is,
whether HL speakers merge different sound categories rather
than producing them distinctly. This question merits investi-
gation even if only HL categories are considered, as phono-
logical merger is commonly attested in cases of L1 attrition
(Andersen, 1982;Campbell and Muntzel, 1989;Goodfellow,
2005), which bears a number of similarities to L2 and HL ac-
quisition (see, e.g., Montrul, 2008). Moreover, HL speakers’
production of categories of the dominant language relative to
those of the HL has yet to be fully addressed: although HL
speakers may make all the phonological contrasts in each of
their languages, do they also make phonetic contrasts across
their two languages between similar, yet acoustically distinct
phones? Suggestive results were obtained by Godson (2003,
2004), who found that HL speakers of Western Armenian
showed some influence of English vowels in their pronuncia-
tion of the Armenian back vowels closest to English vowels,
but this influence did not necessarily result in the merger of
similar Armenian and English vowels.
B. Second-language phonology
While few HL studies have investigated the extent to
which HL speakers produce cross-linguistic contrast
between similar categories in their two languages, this ques-
tion has long been a subject of inquiry in research on L2
speech and bilingual phonology (see, e.g., Flege, 1995;
Laeufer, 1996), a field which has been informed by two in-
fluential models: the Perceptual Assimilation Model (PAM;
Best, 1994) and the Speech Learning Model (SLM; Flege,
1995). The PAM is applicable to the process of L2 phono-
logical acquisition at its very beginning stages. Principally a
model of non-native speech perception by naive listeners
(i.e., those who have no knowledge of the non-native lan-
guage), the PAM sets forth a typology of ways in which non-
native speech contrasts may be interpreted by naive listeners
relative to L1 phonological categories (so-called perceptual
assimilations). The type of perceptual assimilation that
occurs with members of a non-native contrast predicts the
degree of difficulty that learners will have with perceiving
that contrast. If the members of the contrast are assimilated
to different L1 categories, the contrast will be perceived
accurately; if not, the contrast will be perceived less accu-
rately, to a degree depending upon how equally well the
members of the contrast are assimilated to the same L1 cate-
gory. In a more recent version of this model, the PAM-L2
(Best and Tyler, 2007), the connection between non-native
speech perception and L2 speech perception is made
explicit. The PAM-L2 expands upon the PAM by incorporat-
ing the influence of an L2 learner’s developing phonetic and
phonological knowledge of L2, thus allowing for perceptual
assimilation at the gestural, phonetic, and phonological lev-
els. The novel possibility of assimilation at the phonological
level in particular is one of the features of this model that
most distinguishes it from the SLM.
The SLM, in contrast to the PAM(-L2), is mainly a model
of later stages of L2 speech acquisition, focusing on proficient
bilinguals rather than novice learners. The model posits that
phonetic categories are continually modified in response to
sounds in another language that are identified with these cate-
gories. Furthermore, categories of L1 and L2 are said to exist
in a common phonological space for bilinguals, who tend to
keep them distinct under a general pressure to maintain con-
trast between different sounds. Central to the SLM is its
account of inaccurate production of an L2 sound in terms of
the recruitment of a similar L1 category. While a “new” L2
sound—one that has no clear parallel in L1—will motivate
the formation of a new phonetic category, a “similar” L2
sound tends to undergo “equivalence classification” with a
close L1 counterpart, a phenomenon that becomes increas-
ingly likely as age of L2 learning increases. In this way, an L1
sound and an L2 sound may become linked to each other per-
ceptually. A major way in which the SLM differs from the
PAM—and the principal reason the SLM is more relevant to
the present study—is that the SLM overtly addresses the con-
nection to L2 production. Perceptually linked L1 and L2
sounds are predicted to eventually approximate each other in
production. At the same time, however, following from the
notion of L1 and L2 sounds existing in the same phonological
J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin 3965
Author's complimentary copy

space, the model allows for the possibility of an L2 category
dissimilating from an L1 category for the sake of maintaining
contrast between them. In other words, the existence of L1
and L2 sounds in a shared space may lead to either conver-
gence or divergence between the sounds.
That similar L1 and L2 sounds undergo equivalence clas-
sification and influence each other in production has been
demonstrated in a number of studies (e.g., Flege and Hillen-
brand, 1984;Flege, 1987;Major, 1992;Sancier and Fowler,
1997). In one such investigation focusing on L2 speakers of
English and French, Flege (1987) found that native English
speakers who had learned French and native French speakers
who had learned English produced French /u/ differently from
monolingual native French speakers. Both groups produced
French =u=with significantly higher second-formant values in
approximation to the high second-formant norms of English
=u=. Moreover, with regard to the realization of French =t=
(unaspirated) vs English =t=(aspirated), speakers did not typi-
cally reach the L2 phonetic norm for voicing lag and the L2
phonetic norm had an effect on their L1 =t=,suchthatboth
groups ended up over-aspirating French =t=and under-aspirat-
ing English =t=. On the other hand, native English speakers’
production of French =y=(a “new” sound with no counterpart
in English) was comparable to native French =y=.
C. The present study
It remains to be seen whether this sort of subphonemic,
bidirectional cross-linguistic influence or even, as alluded to
above, categorical merger is found in HL speakers (individu-
als who, unlike typical adult L2 learners, received some
degree of early exposure to both of their languages). Thus,
the present study reexamined phonological production by HL
speakers—in this case, HL speakers of Mandarin Chinese—
in order to address three main questions. First, do HL speak-
ers in fact reliably produce the phonological contrasts in both
the HL and the dominant language? Second, do HL speakers
produce phonetic contrasts between similar, yet acoustically
distinct, categories in their two languages? Third, how do HL
speakers compare to native speakers and late learners in their
production of phonetic and phonological contrast? Like previ-
ous studies on HL phonology, one objective of the current
study was to see how closely HL Mandarin speakers would
pattern with native speakers vs late learners with respect to
production of language-internal phonemic contrasts. How-
ever, unlike previous HL studies, another objective was to see
how HL Mandarin speakers would compare to these other
groups in production of cross-linguistic contrast between sim-
ilar categories in their two languages (Mandarin and English).
Another way in which the current study differed from
previous HL studies was in the treatment of variability
among HL speakers. Although HL speakers have been noted
to outpace novice L2 learners of a language in a number of
ways, the population of language users referred to as HL
speakers has also been noted to be an extremely heterogene-
ous group. Li and Duff (2008, p. 17), for instance, note of
Chinese HL speakers that “even within a proficiency-defined
‘HL’ group, learners generally have a very uneven grasp of
the HL, falling along a continuum of having very little HL
knowledge to being highly proficient.” In this study, the het-
erogeneity of the HL group—rather than being artificially
reduced via the detailed sort of screening of participants
used in previous studies—was instead accepted as represen-
tative of the larger population under study. The only require-
ment for HL speakers to be included in the current HL
speaker sample (see Sec. II A) was that their primary HL ex-
perience was with Mandarin, as opposed to another variety
of Chinese. Although inclusion of a wider spectrum of HL
experience than examined in previous studies increased the
probability of inter-speaker variation within the HL group
(and, thus, the probability of obtaining null results), it also
served to maximize the generalizability of the results, which
emerged in spite of the variability purposefully left within
this HL speaker sample. In other words, the main findings
are expected to be robust and reproducible with a different
pseudorandom sampling of HL Mandarin speakers.
The research questions in this study were addressed via
an acoustic investigation of American HL speakers’ produc-
tion of Mandarin Chinese and American English. The produc-
tion of both Mandarin and English phonological categories by
HL speakers of Mandarin (whose dominant language was
English) was compared to that of native L1 speakers of Man-
darin (who were late L2 learners of English) and late L2 learn-
ers of Mandarin (who were native L1 speakers of English) in
a series of experiments designed to investigate the realization
of three different types of phonemic categories: vowel quality
categories, plosive voicing (i.e., laryngeal) categories, and fri-
cative place categories. These categories in particular were
examined for two reasons. On the one hand, focusing on
vowel quality and laryngeal categories facilitated comparison
with previous studies on HL and L2 phonology, as both of
these category types have figured prominently in earlier work
(e.g., Flege and Hillenbrand, 1984;Flege, 1987;Godson,
2003, 2004;Knightly et al., 2003;Oh et al., 2003); on the
other hand, extending the domain of inquiry to consonantal
place of articulation categories allowed for an examination of
whether previous findings would generalize to new dimen-
sions of phonological contrast that have not yet been exam-
ined in this regard. As the study was concerned with the
production of similar categories, the specific categories chosen
for investigation mostly comprised pairs of similar Mandarin
and English categories that stood to be identified with each
other: rounded vowels (Mandarin and English /o
u
,u=,Man-
darin =y=), short-lag stops (Mandarin unaspirated, English
voiced), long-lag stops (Mandarin aspirated, English voice-
less), and post-alveolar fricatives (Mandarin retroflex /§/and
alveolo-palatal //, English palato-alveolar =$=). The acoustic
data comprised measurements of formant resonances (Sec. III
A), VOT (Sec. III B), and spectral features such as center of
gravity, or centroid (Sec. III C). These measurements, as well
as the phonetic norms against which they were compared, are
described in more detail in Secs. II D and III A–III C.
There is little literature on L1 Mandarin speakers’ pro-
duction of L2 English segmentals as opposed to prosody
(e.g., Zhang et al., 2008) and even less literature on L1 Eng-
lish speakers’ production of L2 Mandarin segmentals that
offers predictions regarding the sort of patterns one might
expect to find in the current study of cross-language
3966 J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin
Author's complimentary copy

production in Mandarin and English. The few studies that
have examined Mandarin speakers’ production of English
vowels (Jia et al., 2006;Jiang, 2008, 2010) have generally
examined accuracy via listener ratings rather than acoustic
analysis, although some acoustic data from Chen (2006) sug-
gest that Mandarin speakers produce English =u=with sec-
ond-formant values lower than those of native English
speakers, in keeping with the respective phonetic norms of
Mandarin and English back rounded vowels (Sec. III A). As
for Mandarin speakers’ production of English laryngeal con-
trast, Zhang and Yin (2009, p. 144) noted in a review article
that due to the different features at work in the two languages
(voicing in English, aspiration in Mandarin), “Chinese learn-
ers of English often neglect the differences between voiced
and voiceless sounds in English”; however, this statement
was not about plosives specifically, and no data were pre-
sented in support of this claim. No known studies exist on
Mandarin speakers’ production of English fricatives or on
English speakers’ production of the corresponding Mandarin
segmentals. In summary, previous studies offer little in the
way of specific predictions regarding cross-language produc-
tion in this study, although the data that do exist suggest that
in L2 production there is some influence of the phonetic
norms for similar L1 categories, as has been found in many
studies of L2 speech including that of Flege (1987).
Specific hypotheses regarding the research questions
follow from the principles of the SLM. Since equivalence
classification and concomitant linking of similar L1 and L2
categories is thought to occur more often with increasing age
of L2 learning, it was hypothesized that heritage speakers of
a minority language, due to their early childhood experience
with two languages (the dominant language and the HL),
would outperform late L2 learners in producing contrast
between distinct sounds: language-internal phonological
contrast between phonemic categories, as well as cross-lin-
guistic phonetic contrast between similar, yet acoustically
distinct categories of different languages. On the other hand,
HL speakers and L2 learners were both predicted to do well
with producing HL=L2 categories that are substantially dif-
ferent from those of the dominant language. Thus, the results
were expected to show HL speakers producing contrast
between relatively similar categories (e.g., Mandarin /§/ and
//; Mandarin =o
u
=and English =o
u
=) more often and more
effectively than L2 learners, but not necessarily producing
significantly more accurately than L2 learners those catego-
ries that would be, for the L2 learner, new categories vis-a-
vis L1 (e.g., Mandarin =y=).
As for the manner in which L2 sounds are deemed
“similar” to L1 sounds, it is argued that the PAM-L2 offers
the most comprehensive view of how these equivalences
may arise. As discussed above, this model (unlike the SLM)
is explicit about allowing category equivalence to be based
on gestural, phonetic, and phonological considerations,
rather than strictly phonetic proximity. Given the abundant
evidence that has been found for the role played by phono-
logical information in determining cross-language category
equivalence in loanword adaptation (see, e.g., LaCharite´ and
Paradis, 2005;Kang, 2008;Chang, 2009), it follows that
phonological information is indeed likely to play an impor-
tant role in determining category equivalence in L2 acquisi-
tion. However, there may be cases where the phonological
level conflicts with the phonetic level and=or gestural level
with respect to cross-language proximity between categories,
and the PAM-L2 does not indicate which level prevails in
these cases. In fact, the current study concerned one such
case, where phonological considerations are at odds with
phonetic ones. Experiment 1 examined two Mandarin high
rounded vowels, back =u=and front =y=, which are each
similar to English =u=, but in different ways; consequently,
it is unclear which of these vowels should be considered
“similar” and liable to be linked to English =u=in percep-
tion=production. When only acoustic measures of vowel
quality are considered, English =u=is more similar to Man-
darin =y=(which is on the order of 3 Bark away from Eng-
lish =u=in second-formant frequency; see Table I) than to
Mandarin =u=(which is twice as far away). When the phono-
logical statuses of these vowels are considered, however,
Mandarin =u=emerges as the clear counterpart of English
=u=, since they both function in their respective vowel in-
ventory as high back rounded vowels. Due to this ambiguity
in cross-language proximity, both possible vowel equivalen-
ces (English =u=-Mandarin =u=, English =u=-Mandarin =y=)
were considered in this study. However, it was predicted
that, as with French =y=in Flege (1987), Mandarin =y=
would constitute a “new” vowel for L2 learners—on the ba-
sis of its phonological, rather than phonetic, deviance from
English =u=—and, thus, that it would be produced relatively
accurately. This point is further discussed in Sec. IV.
The paper is organized as follows. Section II provides
an overview of the characteristics of the speakers who par-
ticipated in the study, the procedure and stimuli used in the
experiments, and the acoustic analyses conducted on partici-
pants’ productions. Section III presents the results of each of
the three experiments (experiment 1: vowel categories,
experiment 2: laryngeal categories, experiment 3: fricative
categories), and, finally, Sec. IV discusses the findings in
light of the hypotheses discussed above.
II. METHODS
A. Participants
A total of 28 Mandarin speakers and learners partici-
pated, with two excluded from the final analysis due to lan-
guage backgrounds inconsistent with the focus of the study.
All were recruited at the University of California, Berkeley,
and paid for a single session that encompassed all three
experiments. Participants who were included in the analysis
comprised 15 females and 11 males ranging in age from 18
to 40 years, none of whom reported any history of speech or
hearing impairments. They were each presented with the
same set of stimuli, described in Sec. II C.
Demographic information about all participants is pre-
sented in Appendix A, listing each participant’s identifier
(PID), gender, age at the time of the study (in years), place of
birth or residential history (including ages), and where appli-
cable: age of arrival to the U.S. (in years), other languages
spoken or exposed to at home, frequency of current Mandarin
use, and general experience with Mandarin (including the
J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin 3967
Author's complimentary copy

ages at which the experience occurred). For the purposes of
analysis, participants were divided into groups according to
their responses on a detailed questionnaire about their life his-
tory and family background, language background, current
language use, formal language education, and Mandarin pro-
ficiency.
2
If participants had not received exposure to Man-
darin until the age of 18 years, they were classified as late L2
learners. If, on the other hand, they were born and schooled
in a Mandarin-speaking region, reported their current Man-
darin proficiency level to be native-like, and judged Mandarin
to be their best language, they were classified as native Man-
darin (NM) speakers. Anyone with prior Mandarin experi-
ence in the home who did not fulfill all of the criteria for the
native speaker group was classified as an HL speaker.
HL speakers were divided into high-exposure (HE) and
low-exposure (LE) groups using self-reported frequency of
current Mandarin use as the primary consideration and the
number of years lived in a Mandarin-speaking region as a
secondary consideration. For HL speakers, there was a gen-
eral trend for people who had lived longer in Mandarin-
speaking regions to also use Mandarin more often with
their family. Three exceptional cases were participants H9,
H13, and H20. Participant H9, who had visited Mandarin-
speaking regions many times but was born and educated
entirely in the U.S., reported extensive use of Mandarin in
her family, both with parents and siblings and with other
relatives. Participant H13, on the other hand, did not come
to the U.S. until she was 10 years old, yet reported using
Mandarin only half of the time with parents and not with
anyone else currently. Finally, participant H20 was born
and spent the first two years of life in China, but Mandarin
was spoken to her at home only by her nanny; although her
father was also a Mandarin speaker, both parents spoke to
her in English, and she only started to hear and speak Man-
darin again when taking a Chinese language class during
the semester of recording. These three participants were di-
vided into the HE and LE groups by simultaneously consid-
ering both their current use of the language and the amount
of time they had spent in Mandarin-speaking areas. In gen-
eral, however, HL speakers were put into the HE group if
they reported using Mandarin at home more than half of the
time and into the LE group if they reported using Mandarin
at home half of the time or less.
The participants in each resulting group possessed sev-
eral shared background characteristics. The six participants
(four females, two males; mean age 29.8 yr) in the NM
group were all NM speakers who were born and educated
(up to at least seventh grade) in mainland China or Taiwan.
The 15 HL speaker participants reported speaking English
most of the time overall, but they were all born to Mandarin-
speaking parents. Generally speaking, the nine participants
(four females, five males; mean age 21.0 yr) in the HE group
were heritage speakers who had extensive exposure to Man-
darin as children and reported using Mandarin to communi-
cate with both parents most or all of the time. Most of the
HE participants were either born in a Mandarin-speaking
region (H8, H10, H11, H12, H13), with a mean age of arrival
to the U.S. of 6.9 yr, or had otherwise lived for a number of
years in a Mandarin-speaking region. In contrast, the six par-
ticipants (four females, two males; mean age 20.0 yr) in the
LE group were heritage speakers who had limited exposure
to the language and reported using Mandarin with their
parents half of the time or less. With the exception of H20,
all of the LE participants were born in the U.S. and had
never lived in a Mandarin-speaking region. The five partici-
pants (three females, two males; mean age 21.6 yr) in the
second-language (L2) learner group were native English
speakers who were born and educated in the U.S., grew up
in English-speaking families, and started to learn Mandarin
after the age of 18. Three (L22, L23, L24) grew up in a
monolingual home environment, while the other two (L25,
L26) had some degree of exposure to other languages as
well. With the exception of L26, all had received formal
Mandarin language instruction, ranging in duration from
three months in an immersion environment to two years in
an American university setting. Nearly all reported their cur-
rent Mandarin proficiency to be quite poor and estimated
that they understood 10%–25% of normal conversational
Mandarin; the exception is L25, who had received the most
formal instruction and reported understanding 30%–50% of
conversational Mandarin.
B. Procedure
Study participants were recorded reading aloud 59
Mandarin items and 32 English items presented via indi-
vidual index cards in random order by language. Each
language block was repeated four times in a single ses-
sion for a total of 364 tokens in all. Participants com-
pleted all blocks in one language before moving on to the
second language, with the order of the languages (Man-
darin-English or English-Mandarin) balanced across par-
ticipants. English words were written in English
orthography, and Mandarin words in Mandarin orthogra-
phy (traditional or simplified characters) and phonetic
spelling (pinyin, the spelling system used in mainland
China, and=or zhuyin=Bopomofo, the spelling system
used in Taiwan). The recordings were made in a sound-
attenuated booth with 48-kHz sampling and 16-bit resolu-
tion using either a Marantz PMD660 solid-state recorder
and an AKG C420 head-mounted condenser microphone,
or a Dell desktop computer connected to an M-AUDIO
Mobile-Pre USB preamp audio interface and an AKG
C520 head-mounted condenser microphone.
TABLE I. Native F
1
and F
2
norms (in Bark) for rounded vowels in Man-
darin and English. The vowels compared are Mandarin and English =o
u
=,
Mandarin and English =u=, and Mandarin =y=.
F
1
F
2
Mandarin English Mandarin English
=o
u
=Male 5.38 4.36 6.61 9.59
Female 6.72 5.06 8.02 10.60
=u=Male 3.54 3.26 4.51 10.72
Female 4.12 3.97 6.06 11.92
=y=Male 2.93 — 13.53 —
Female 3.23 — 14.71 —
3968 J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin
Author's complimentary copy

C. Stimuli
In choosing Mandarin and English stimuli, segmental
context was matched across language as much as possible,
and Mandarin items with falling tones were selected (when
such words existed) so as to make the pitch contour of the
Mandarin items maximally similar to the falling pitch con-
tour of English words spoken in isolation (e.g., English boot
vs Mandarin “not”; English tote [t
h
o
u
t] vs Man-
darin “transparent”; English shot [$At] vs Mandarin
“suddenly” and “below”). In addition, the
most common character corresponding to the phonological
shape of each Mandarin item was selected, minimizing the
possibility of participants being unfamiliar with any of the
items. In the end, the stimulus items chosen were all com-
mon for native speakers of that language, although L2 learn-
ers—particularly the L2 Mandarin learners, who had
relatively little experience with Mandarin—were not neces-
sarily familiar with all of the items. Nonetheless, because
multiple sources of information were provided about each
item, participants were able to complete the task described
in Sec. IIBwith little trouble.
The speech stimuli used in experiments 1–3 are pre-
sented in Appendix B. Critical stimuli (i.e., the non-filler
items subjected to acoustic analysis; see Sec. II D) were gen-
erally of the form consonant-vowel (CV) in the case of Man-
darin and of the form consonant-vowel-consonant (CVC) in
the case of English. In experiment 1, critical stimuli contained
one of five rounded vowel categories: Mandarin =u=appeared
in ten items, Mandarin =o
u
=in seven, Mandarin =y=in three,
English =u=in eleven, and English =o
u
=in ten. With the
exception of Mandarin =y=, which is phonotactically re-
stricted to coronal contexts, all of these vowels occurred fol-
lowing the onsets of several different places of articulation
and laryngeal types. In experiment 2, critical stimuli contained
a word-initial plosive of one of four laryngeal categories and
three places of articulation: Mandarin unaspirated =p,t,k=,
Mandarin aspirated =p
h
,t
h
,k
h
=, English voiced =b,d,g=,and
English voiceless =p,t,k=. With one exception (due to the ab-
sence of =po
u
=in Mandarin), all of these plosives preceded
back rounded vowels. There were two items per combination
of laryngeal category and place of articulation, for a total of
12 Mandarin items and 12 English items. In experiment 3,
critical stimuli contained one of three post-alveolar sibilant
fricatives: Mandarin retroflex /§/, Mandarin alveolo-palatal //,
and English palato-alveolar =$=. These fricatives appeared
prevocalically in seven Mandarin items and two English
items. All occurred in a low vowel context.
D. Acoustic analysis
All acoustic measurements were taken manually in
PRAAT (Boersma and Weenink, 2008) using a 5-ms analysis
window and 50-dB dynamic range. In experiment 1, vowel
quality was analyzed by measuring average values of the
first (F
1
) and second (F
2
) formants (Ladefoged, 2005, pp.
40–43) over the whole duration of the vowel, from the be-
ginning of the first glottal pulse to the end of the last visible
glottal pulse (Mandarin tokens) or the beginning of the final
consonant constriction (English tokens). In experiment 2,
voicing lag in word-initial plosives was analyzed by meas-
uring VOT as time at the onset of periodicity minus time at
plosive release (Lisker and Abramson, 1964;Ladefoged,
2003, pp. 96–101). In experiment 3, peak amplitude fre-
quency (PAF) and centroid frequency (Ladefoged, 2003, pp.
156–158) were measured over an average spectrum of the
middle 100 ms of the fricative. A low-frequency stop-band
filter was applied to this spectrum to remove frequencies
from 0 up to the F
2
region (so as to get a better measure of
specifically front cavity resonances varying with place of
articulation). The location of the F
2
region (the endpoint of
the band filter) was estimated for each subject as three-fifth
of the speaker’s average third formant in the vowel =a=(Li
et al., 2007). Frequency measurements were later converted
to Bark for an acoustic perceptual view of participants’
vowel and fricative productions using the following formula
from Traunmu¨ller (1990):z¼[26.81=(1 þ1960=f)] 0.53.
To ensure that the measurements were reliable, 25% of
the measurements from each experiment were double-
checked by a second researcher in a pseudorandom fashion.
Any discrepancy between the two researchers’ measure-
ments in excess of 100 Hz (for formants, PAFs, and cent-
roids) or 5 ms (for VOT) was checked again by a third
researcher. In experiment 1, 8% of formant measurements
were triple-checked in this fashion. Final calculations of the
differences between researchers’ measurements here
revealed an average difference of 13 Hz in F
1
measurements
(81% were less than 25 Hz apart) and 24 Hz in F
2
measure-
ments (63% were less than 25 Hz apart). In experiment 2,
9% of VOT measurements were triple-checked, with an av-
erage difference of 1.4 ms between different researchers’
measurements. In experiment 3, 3% of the measurements
were triple-checked. There was an average difference of 12
Hz in PAF measurements (72% were less than 25 Hz apart)
and 33 Hz in centroid measurements (41% were less than 25
Hz apart). If after a third measurement there still remained a
discrepancy between different researchers’ measurements of
greater than 100 Hz=5 ms, all of these measurements were
discarded; however, this resulted in the discarding of less
than 1% of the total number of measurements.
III. RESULTS
A. Experiment 1: Vowels
On the basis of relative acoustic phonetic similarity as
well as place in the relevant vowel inventory, the “similar”
vowels compared to each other were Mandarin =o
u
=-English
=o
u
=and Mandarin =u=-English =u=, while Mandarin =y=
was predicted to constitute a “new” vowel vis-a-vis English.
Differences between formant norms for the Mandarin and
English vowels under study are summarized in Table I, con-
verted to Bark according to the formula given in Sec. II D
(Mandarin figures from Wu and Lin, 1989 and Lin and
Wang, 1992; English figures from Hagiwara, 1997). On av-
erage, Mandarin =u=and English =u=are quite similar in F
1
,
but differ substantially in F
2
. The average F
2
for English =u=
is approximately 6 Bark higher than that of Mandarin =u=
for both male and female speakers. On the other hand, Man-
darin =o
u
=and English =o
u
=differ in both F
1
and F
2
, English
J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin 3969
Author's complimentary copy

=o
u
=being 1–1.5 Bark lower in F
1
and approximately 2.5–3
Bark higher in F
2
. Mandarin =y=is similar to English =u=in
F
1
, but approximately 3 Bark higher in F
2
. Thus, if speakers
with experience in both languages closely approximate these
phonetic norms, they are expected to produce a slight differ-
ence in F
1
and a substantial difference in F
2
between the two
mid vowels, as well as a large difference in F
2
between the
two high vowels. Furthermore, they are expected to produce
the front vowel =y=with the highest F
2
of all.
Mean F
1
and F
2
in participants’ productions of the mid
rounded vowels are plotted in Fig. 1. For each group the
Mandarin and English vowels occupied distinct phonetic
spaces, English =o
u
=being produced with higher F
2
values
than Mandarin =o
u
=. The NM and L2 groups each produced
the =o
u
=of their non-native language with F
2
values approx-
imating the =o
u
=of their native language, while HL speakers
patterned somewhat in between these two groups. For exam-
ple, in the case of Mandarin =o
u
=, most NM speakers had
lower F
2
values of approximately 8.0–8.5 Bark, whereas
most L2 learners had higher F
2
values of approximately 8.6–
9.0 Bark. The majority of HE speakers were located in the
same region as NM speakers, while the majority of LE
speakers were located in the same region as L2 speakers;
both these groups, however, spanned a wide phonetic space
that extended across the regions occupied by NM and L2
speakers. Figure 1also shows some differentiation of the
two vowels in terms of F
1
. In accordance with the small dif-
ference in native F
1
norms seen in Table I, for all speaker
groups the space for Mandarin =o
u
=extended into a higher
F
1
region than the space for English =o
u
=.
Mean F
1
and F
2
in participants’ productions of the high
rounded vowels are plotted in Fig. 2. There are several pat-
terns of note here. First, all groups distinguished Mandarin
=u=and English =u=, producing the latter with substantially
higher F
2
values. However, the groups differed in terms of
their location in F
1
-F
2
space. NM speakers produced both
vowels with the lowest F
2
values, while L2 learners (native
English speakers) produced both vowels with the highest F
2
values, with HL speakers located somewhat in between these
two groups for both vowels. Thus, similar to the case of
=o
u
=, both NM speakers and L2 learners appeared to be
influenced in their pronunciation of the =u=of their second
language by the phonetic characteristics of the =u=of their
first language: NM speakers produced English =u=with a
relatively low F
2
approximating the low F
2
of Mandarin
=u=, whereas L2 learners produced Mandarin =u=with a rel-
atively high F
2
approximating the high F
2
of English =u=.
On the other hand, HL speakers generally produced Man-
darin =u=and English =u=with F
2
values that were rela-
tively close to native values. To put it another way, for most
HL speakers F
2
for Mandarin =u=was not as high as it was
for L2 learners, nor was F
2
for English =u=as low as it was
for NM speakers.
With regard to the Mandarin high front rounded vowel
=y=, all groups produced this vowel in a distinct phonetic space
with much higher F
2
values than the Mandarin and English
back vowels, and the groups did not differ from each other
appreciably with respect to their location in F
1
-F
2
space. The
results of a two-way analysis of variance (ANOVA) with fac-
tors Group and Gender
3
were consistent with this impression.
There was a main effect of Gender on F
1
[F(l,18) ¼16.27,
p<0.001] as well as F
2
[F(l,18) ¼14.79, p<0.01], but no
main effect of Group on either F
1
or F
2
and no interaction with
Gender. In other words, although men and women (unsurpris-
ingly) had different formants for =y=,L2learnersandHL
speakers did not differ statistically from NM speakers in their
production of Mandarin =y=as it was measured here.
Formant data for the mid and back rounded vowels were
subjected to mixed-model ANOVAs, with Group and Gender
as between-subjects factors and Language, Vowel (=o
u
=or
=u=), and Place
4
(of articulation of the onset consonant) as
within-subjects factors. With respect to F
1
,therewasnomain
effect of Language, but there were highly significant main
effects of Vowel [F(l,5) ¼563.58, p<0.001], Place
FIG. 1. Bark plot of the first two formants in mean productions of Mandarin
=o
u
=(gray symbols) and English =o
u
=(white symbols). NM speakers are
plotted in circles, HE speakers in triangles, LE speakers in upside-down tri-
angles, and L2 learners in squares.
FIG. 2. Bark plot of the first two formants in mean productions of Mandarin
=u=(light gray symbols), English =u=(white symbols), and Mandarin =y=
(dark gray symbols). NM speakers are plotted in circles, HE speakers in tri-
angles, LE speakers in upside-down triangles, and L2 learners in squares.
3970 J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin
Author's complimentary copy

[F(4,51) ¼115.17, p<0.001], and Gender [F(l,5) ¼20.28,
p<0.01], as expected: =o
u
=>=u=; velar >alveolar >
bilabial >glottal=post-alveolar; and female >male. As one
would predict from the formant norms cited in Table I,there
was also a two-way interaction between Language and Vowel
[F(l,5) ¼73.69, p<0.001], attributable to only Mandarin =u=
and English =u=not being produced with distinct F
1
values.
Males were more successful than females at producing an F
1
difference between the high back vowels, resulting in a three-
way interaction of Gender, Language, and Vowel
[F(l,5) ¼10.99, p<0.05]. However, Group did not have a
main effect on F
1
, nor did it interact significantly with any
other factors. In short, although the various vowels were over-
all produced differently in terms of F
1
, which was moreover
affected by the consonantal context in which the vowels
occurred, the participant groups did not differ from each other
statistically with respect to production of F
1
.
While there was no main effect of Group on F
1
,there
was a main effect of Group on F
2
[F(3,4) ¼11.24, p<0.05],
although no main effect of Gender. There were also highly
significant main effects of Language [F(l,4) ¼704.62,
p<0.001] and Place [F(4,48) ¼315.82, p<0.001]:
English >Mandarin, and post-alveolar >alveolar >velar
>glottal=bilabial. Although the effect of Vowel was only
marginally significant, a two-way interaction between Lan-
guage and Vowel [F(1,4) ¼316.25, p<0.001] arose from the
greater effect of Language on F
2
in the case of =u=than in the
case of =o
u
=. Group not only had a main effect on F
2
, it also
interacted with Language [F(3,4) ¼11.20, p<0.05] and with
Language and Vowel [F(3,4) ¼15.93, p<0.05]. The
Group Language interaction was attributable to the pattern
seen in Figs. 1and 2: English back vowels were produced
with greater F
2
values than Mandarin back vowels in all
groups, but this language effect differed across the groups,
which produced disparate F
2
values and unequal distances
between languages. The Group Language Vowel interac-
tion arose from the fact that the Group Language interaction
was more pronounced for =u=than for =o
u
=.
5
In summary,
the F
2
results contrasted with the F
1
results in two main ways:
the English vowels were found overall to be produced with
higher F
2
values than the Mandarin vowels, and the partici-
pant groups (but not the genders) differed from each other sig-
nificantly in F
2
production.
To examine between-group differences in the realiza-
tion of cross-linguistic contrasts between similar vowel cat-
egories, the mean differences in F
1
and F
2
produced
between corresponding back vowels (Mandarin and English
=o
u
=, Mandarin and English =u=) were calculated for each
participant. One-way ANOVAs showed a highly significant
main effect of Group on F
2
distances between Mandarin
and English =u=[F(3,22) ¼7.85, p<0.001], but not on F
2
distances between Mandarin and English =o
u
=.Thesemean
F
2
distances are presented in Fig. 3, where it can be seen
that both the HE group and the LE group put more acoustic
distance between Mandarin and English =u=than did the
L2 group (HE vs L2: Mann–Whitney U¼38, n
1
¼9,
n
2
¼5, p<0.05 two-tailed; LE vs L2: Mann–Whitney
U¼30, n
1
¼6,n
2
¼5, p<0.01 two-tailed). The LE group
also surpassed the NM group (Mann–Whitney U¼34, n
1
¼n
2
¼6, p<0.01 two-tailed) and the HE group (Mann–
Whitney U¼45, n
1
¼6, n
2
¼9, p<0.05 two-tailed) in this
regard. In short, HL speakers separated their two high back
rounded vowels in F
2
to a greater degree than L2 learners
did, and LE speakers in particular also produced greater F
2
separation than NM speakers.
B. Experiment 2: Plosives
Differences between VOT norms for Mandarin and Eng-
lish plosives are summarized in Table II (Mandarin figures
from Wu and Lin, 1989; English figures from Lisker and
Abramson, 1964). On the basis of their acoustic phonetic
similarity, the categories compared to each other were the
two short-lag VOT categories (Mandarin unaspirated and
English voiced) and the two long-lag VOT categories (Man-
darin aspirated and English voiceless), which in initial posi-
tion are all typically realized without vocal fold vibration
during closure. Of the two short-lag categories, Mandarin
unaspirated plosives are on average characterized by the
longer VOT, with the VOT of English voiced plosives being
similar, but 2–9 ms shorter or longer at the same place of
articulation. With respect to the long-lag categories, Man-
darin aspirated plosives are significantly more aspirated than
English voiceless plosives, by as much as 48 ms at the same
FIG. 3. Mean differences in F
2
produced between Mandarin and English
back rounded vowels, by participant group (from left to right: NM, HE, LE,
L2). Differences between Mandarin and English =o
u
=are in dark gray bars,
differences between Mandarin and English =u=in light gray bars. Error bars
indicate 61 standard error about the mean.
TABLE II. Native VOT norms (in milliseconds) for plosives in Mandarin
and English. The laryngeal categories compared are Mandarin unaspirated,
Mandarin aspirated, English voiced, and English voiceless.
Short-lag Long-lag
Mandarin English Mandarin English
Unaspirated Voiced Aspirated Voiceless
Labial 10 1 106 58
Coronal 7 5 113 70
Dorsal 15 21 116 80
Average 11 9 112 69
J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin 3971
Author's complimentary copy

place of articulation. In short, both pairs of similar laryngeal
categories differ in VOT, although the difference between
the long-lag categories is much greater than that between the
short-lag categories. If speakers with some degree of experi-
ence in both languages closely approximate these phonetic
norms, then, it is expected that they will produce a subtle dif-
ference between the short-lag categories and a pronounced
difference between the long-lag categories.
As a first step toward testing this prediction, the VOT
data collected in experiment 2 were subjected to a mixed-
model ANOVA, with Group and Gender as between-subjects
factors and Language, Voicing Type (short-lag vs long-lag),
Place (of articulation), and Vowel
6
(environment) as within-
subjects factors. As expected, the ANOVA results showed
highly significant main effects of every within-subjects fac-
tor: Language [F(1,6) ¼46.49, p<0.001], Voicing Type
[F(1,6) ¼613.05, p<0.001], Place [F(2,18) ¼93.52,
p<0.001], and Vowel [F(1,6) ¼19.49, p<0.01]. These
main effects were all in the expected direction:
Mandarin >English; long-lag >short-lag; velar >alveolar >
bilabial; and =u=>=o
u
=. There was also a two-way interac-
tion between Language and Voicing Type [F(1,6) ¼18.64,
p<0.01], an effect mostly attributable to there being no sig-
nificant difference in VOT between Mandarin unaspirated
and English voiced stop productions. While there were no
main effects of Group or Gender on VOT, there was a signif-
icant six-way interaction between these factors and the
four within-subjects factors: Group Gender Language
Voicing Type Place Vowel [F(3,17) ¼3.41, p<0.05].
This interaction occurred due to between-group differences
only for comparisons of a few combinations of the within-
subjects factors. For instance, in comparing the HE and L2
groups, Tukey’s HSD (Honestly Significant Difference) test
showed a reliable difference only between HE and L2
females and only with respect to Mandarin long-lag velar
stops preceding =u=[p<0.05]. In summary, Mandarin VOT
was produced as longer overall than English VOT (due to
the long-lag VOT of Mandarin aspirated stop productions
being longer than the long-lag VOT of English voiceless
stop productions), and there were no significant differences
among groups with respect to overall VOT levels.
When the VOT data were examined by participant, it was
apparent that there was a strong tendency for HL speakers to
make a VOT distinction between cross-linguistically similar
laryngeal categories. The short-lag categories were produced
with reliably distinct VOTs by only six participants. Of these
six, half came from the HE or LE groups; the other three were
divided between the NM and L2 groups. The long-lag catego-
ries, on the other hand, were distinguished by 18 participants
(Fig. 4). These participants were concentrated in the NM, HE,
and LE groups, such that all NM speakers, all but one HE
speaker, and half of LE speakers produced a reliable differ-
ence in VOT between the long-lag categories. In contrast, all
but one L2 learner produced no reliable difference in VOT
between these two categories. Note that this pattern still held
after adjusting for multiple comparisons. When the Bonferroni
correction was applied, five NM speakers and eight HL speak-
ers, but only one L2 learner, were found to produce a signifi-
cant difference in VOT.
The results of experiment 2 thus indicated that while all
participants reliably produced the language-internal contrasts
between Mandarin unaspirated and aspirated plosives and
between English voiced and voiceless plosives, the same
could not be said of their realization of cross-linguistic con-
trasts. Few made the cross-linguistic contrast between the
short-lag categories of Mandarin and English. On the other
hand, many participants produced a contrast between the
long-lag categories. However, these were nearly all partici-
pants with the greatest Mandarin experience—namely, NM
and HL speakers. Most L2 learners failed to distinguish the
long-lag categories.
Between-group differences in the realization of cross-
linguistic contrasts were further examined by calculating for
each participant the mean difference in VOT produced
between similar laryngeal categories. The mean VOT distan-
ces produced by all groups are presented in Fig. 5. A one-
way ANOVA showed no main effect of Group on the VOT
distances established between the short-lag categories, but a
marginally significant main effect of Group on the VOT dis-
tances established between the long-lag categories
[F(3,22) ¼2.27, p¼0.1]. Here the HE group produced reli-
ably greater distance between the two categories than the L2
group (Mann–Whitney U¼26, n
1
¼9, n
2
¼6, p¼0.05 two-
tailed), as did the NM group (Mann–Whitney U¼38,
n
1
¼n
2
¼6, p<0.05 two-tailed). These results were consist-
ent with the findings of the participant analysis described
above: NM speakers and HL speakers—HE speakers, in par-
ticular—established a greater acoustic distance between the
long-lag VOT categories of Mandarin and English than did
L2 learners of Mandarin.
C. Experiment 3: Fricatives
Before the results of Experiment 3 are presented, the pho-
netics of the three post-alveolar fricatives under investigation
are first reviewed. These fricatives have been described in
detail by Ladefoged and Maddieson (1996, pp. 148–154).
FIG. 4. VOT in Mandarin aspirated plosives (triangles) and English voice-
less plosives (circles), by participant. Error bars indicate 95% confidence
intervals. Participants who produced reliably different means are marked
with stars: * (p<0.05),
**
(p<0.01),
***
(p<0.001).
3972 J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin
Author's complimentary copy

While the English palato-alveolar =$=is described as
“domed” (i.e., with the front of the tongue raised) and
rounded, the Mandarin retroflex /§/ is described as “flat” (i.e.,
without the front of the tongue raised), laminal, and not truly
retroflexed, having a location and width of constriction that
are “very comparable with those for English $.” The Mandarin
alveolo-palatal //, on the other hand, is described as signifi-
cantly “palatalized,” with a long, flat constriction formed by a
greater degree of raising of the blade and front of the tongue.
These descriptions suggest that, compared to //, /§/ is closer
phonetically to =$=and, consequently, that /§/ismorelikely
to be merged with =$=in production. Note, however, that
merger of //and=$=has been found before (Young, 2007).
Thus, it is possible that both Mandarin fricatives might be
merged with the English fricative, although the influence of
phonological knowledge in perceptual assimilation (i.e., that
the Mandarin sounds serve to distinguish words; see the
PAM-L2) is likely to prevent such dual merger from
happening.
In fact, differences between the centroids of Mandarin
and English post-alveolar fricatives suggest that, at least
with respect to centroid frequency, both Mandarin fricatives
differ significantly from the English palato-alveolar. The
centroid norms are summarized in Table III, converted to
Bark according to the formula given in Sec. II D (Mandarin
figures averaged from Svantesson, 1986; English figures
from Jongman et al., 2000). Mandarin // is characterized by
the highest centroid frequency, followed by English =$=and
then Mandarin /§/. Taking into account that the average cent-
roid for =$=is likely to be slightly higher than the figure
given in Table III (an average that includes the correspond-
ing voiced fricative =Z=, whose centroid will be drawn down
by the lower frequencies of voicing), one can see that the
centroid of /§/ is slightly closer to that of =$=than is the cent-
roid of //, but each Mandarin centroid lies on the order of 1
Bark away from the English centroid. Thus, if speakers with
some degree of experience in both languages closely approx-
imate these phonetic norms, it is expected that they will pro-
duce a three-way contrast in centroid among these fricatives.
Conversely, if there is merger of any two of these categories,
it is predicted that the fricatives merged will be the more
similar /§/ and =$=.
The centroid and PAF data
7
collected in experiment 3
were subjected to mixed-model ANOVAs, with Group
and Gender as between-subjects factors and Fricative as
a within-subjects factor. As expected, the ANOVA
results showed a highly significant main effect of Fricative
on both centroid [F(2,36) ¼52.33, p<0.001] and PAF
[F(2,36) ¼87.47, p<0.001]: in both cases, //>/§/>=$=(in
contrast to the predictions of Table III). There was also a
main effect of Gender on PAF [F(l,14) ¼13.99, p<0.01]:
female >male. There was no main effect of Group on cent-
roid or PAF, although there was a significant three-way
interaction between Group, Gender, and Fricative with
respect to PAF [F(6,36) ¼3.07, p<0.05], an effect due
mainly to between-group differences only for comparisons
of particular combinations of Gender and Fricative. For
example, in comparing the HE and L2 groups, Tukey’s HSD
test showed no reliable difference between HE and L2 males
or between HE and L2 females with respect to //or=$=, but
did show a reliable difference between HE and L2 males
with respect to /§/[p<0.001]. In summary, although the fri-
catives were produced as spectrally distinct in general, the
groups did not differ from each other significantly in terms
of overall centroid or PAF.
Between-group differences in the realization of cross-lin-
guistic contrasts between similar fricative categories (espe-
cially Mandarin /§/ and English =$=) were examined by
calculating for each participant the mean distances in centroid
and PAF established between each pair of fricatives. In con-
trast to the results of Young (2007), which suggested that HL
speakers might tend to merge Mandarin //withEnglish=$=,
the HL speakers in this study did not differ from other groups
with respect to producing acoustic distance between //and
=$=: all produced a robust contrast between these two frica-
tives. With respect to /§/and=$=, on the other hand, the HL
and L2 groups appeared to separate these categories to a
greater degree than the NM group, particularly with respect to
centroid (Fig. 6). Nevertheless, one-way ANOVAs showed no
main effect of Group on centroid distances or PAF distances.
However, when the centroid data for /§/ and =$=were
examined by participant, it was apparent that HL speakers
and L2 learners more often made a distinction between the
two fricatives than NM speakers. These fricatives were dis-
tinguished in centroid by a total of 14 participants, who were
FIG. 5. Mean differences in VOT produced between Mandarin and English
plosives, by participant group (from left to right: NM, HE, LE, L2). Differ-
ences between Mandarin unaspirated and English voiced plosives are in
dark gray bars, differences between Mandarin aspirated and English voice-
less plosives in light gray bars. Error bars indicate 61 standard error about
the mean.
TABLE III. Native centroid norms (in Bark) for post-alveolar fricatives in
Mandarin and English. The places of articulation compared are Mandarin
retroflex, Mandarin alveolo-palatal, and English palato-alveolar.
Mandarin English
Retroflex
/§/
Alveolo-palatal
//
Palato-alveolar
=$,Z=
16.80 19.12 17.79
J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin 3973
Author's complimentary copy

unevenly distributed across groups (Fig. 7). While the major-
ity of HL speakers (five of nine HE speakers and half of LE
speakers) and the majority of L2 speakers (four of five) pro-
duced a reliable difference in centroid, the majority of NM
speakers (four of six) did not. Again, the pattern held after
Bonferroni correction, in which case six HL speakers and
four L2 learners, but no NM speakers, were found to produce
a reliable difference in centroid between /§/ and =$=.
In short, the results of experiment 3 showed no overall
differences between groups in the realization of contrast
between Mandarin and English post-alveolar fricatives (as it
was measured here). However, on an individual level HL
speakers and L2 learners were found more often to achieve a
reliable distinction between Mandarin /§/ and English =$=
than NM speakers.
IV. DISCUSSION AND CONCLUSIONS
To summarize, in experiments 1–3 evidence was found
that HL speakers were more successful than NM speakers
and L2 learners at producing cross-language contrasts
simultaneously with language-internal contrasts. In experi-
ment 1, participants in all groups were found to make an
F
2
distinction between Mandarin and English back vowels,
with NM speakers’ back vowels having lower F
2
values in
both languages than those of HL speakers and L2 learners.
However, HL speakers—in particular, LE speakers—
clearly outperformed both NM speakers and L2 learners in
achieving acoustic separation between similar vowel cate-
gories. In experiment 2, few participants distinguished
Mandarin unaspirated and English voiced plosives, but HL
and NM speakers distinguished Mandarin aspirated and
English voiceless plosives; furthermore, they put more
acoustic distance between these categories than L2 learn-
ers, who mostly failed to distinguish them. In experiment
3, HL speakers produced a contrast between the two Man-
darin post-alveolar fricatives and were also more likely to
produce a contrast between Mandarin /§/ and English =$=
than NM speakers.
Thus, it was found that HL speakers maintained not
only language-internal “functional” contrast (that is, con-
trast that functions to distinguish words, e.g., English =u=
vs =o
u
=), but also cross-linguistic “non-functional” contrast
(that is, contrast that has no function in distinguishing
words by virtue of the members of the contrast belonging
to different languages, e.g., English =u=vs Mandarin =u=).
On the first point, HL speakers did not differ significantly
from other groups, as almost no speaker in any group
failed to distinguish the phonemic categories of their L1
and L2. HL speakers did not all realize categories in the
same way as more Ll-dominant native speakers (e.g., F
2
values for Mandarin =u=were slightly higher for several
HL speakers than those of NM speakers), but on average
they came very close—much closer than L2 learners—and
this close approximation of phonetic norms seems to lie at
the heart of why HL speakers were more successful than
L2 learners at maintaining contrasts between similar L1
and L2 categories, which for the most part they would
never need to distinguish for the purposes of being
understood.
A. Approximation of phonetic norms
In the present study, it is somewhat difficult to tell how
closely speakers approached the phonetic norms of Mandarin
and English, given the amount of inter-speaker variation and
the limited nature of the acoustic norms available in the liter-
ature (e.g., the Mandarin figures provided by Wu and Lin,
1989 are based on only a few speakers). However, if the
numbers cited in Tables I–III are indeed representative of
the relevant speech communities, then it seems that at least
some of the current data show the same sort of bidirectional
cross-linguistic influence found in Flege (1987). For exam-
ple, the phonetic norm for F
2
in Mandarin =u=is cited as
approximately 450–650 Hz (equivalent to 4.5–6.2 Bark), but
speakers in this study produced this vowel with F
2
values of
FIG. 6. Mean differences in centroid and PAF produced between Mandarin
/§/ and English =$=, by participant group (from left to right: NM, HE, LE,
L2). Differences in centroid are in gray bars, differences in PAF in white
bars. Error bars indicate 61 standard error about the mean.
FIG. 7. Centroids in Mandarin retroflex /§/ (squares) and English palato-al-
veolar =$=(triangles), by participant. Error bars indicate 95% confidence
intervals. Participants who produced reliably different means are marked
with stars: * (p<0.05),
**
(p<0.01),
***
(p<0.001).
3974 J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin
Author's complimentary copy

approximately 6.9–9.7 Bark. Similarly, the phonetic norm
for VOT in Mandarin unaspirated plosives is estimated at 7–
15 ms, but speakers in this study produced these with VOTs
of approximately 15–25 ms. What is most significant about
the findings of this study, however, is that when taken to-
gether, the results of experiments 1–3 showed HL speakers
to have been the most successful at approximating the pho-
netic norms of both of their languages.
As for why HL speakers would tend to be more suc-
cessful than late learners at maintaining contrasts between
similar categories in their two languages, there are two
possible explanations. First, early exposure to both lan-
guages might simply make HL speakers better able to hit
close targets accurately, due to the existence of more fine-
grained, less language-specific perceptual capabilities
early in life (see, e.g., Werker and Tees, 1984;Kuhl et al.,
1992). Alternatively, similar categories that are acquired
early may interact with each other in a shared phonological
system and dissimilate. The results of experiments 2 and 3
are more consistent with the former hypothesis, as similar
laryngeal and place categories in these experiments were
not produced by the HL groups as “too native” with respect
to the productions of the NM and L2 groups (e.g., Man-
darin unaspirated stops were not produced with VOTs that
were even shorter than NM VOTs). On the other hand, the
results of experiment 1 showed signs that some HL speak-
ers had dissimilated similar vowel categories, resulting in a
“polarized” phonetic space that went past native targets
(Laeufer, 1996). In both Figs. 1and 2, it can be seen that
there were HL speakers who went lower in F
2
for their
Mandarin vowels than the NM group, as well as HL speak-
ers who went higher in F
2
for their English vowels than the
L2 group.
Thus, there are two ways to arrive at the patterns
observed among HL speakers in this study, but it should be
noted that these accounts of how HL speakers come to pro-
duce cross-linguistic phonetic contrasts are not mutually
exclusive. Perhaps, as the data suggest, close approximation
of native phonetic norms occurs generally during HL speak-
ers’ relatively early exposure to both languages, but the pres-
sure to keep categories distinct within a speaker’s
phonological system (regardless of which language they
come from) is what serves to keep similar L1 and L2 catego-
ries apart—close to the native phonetic norms—and prevent
them from merging on a “compromise” value. Apparently
this pressure may even push the categories further apart than
they need to be, although the present results suggest that this
is very much the minority case.
The ways in which the linguistic input received by NM
speakers in mainland China and Taiwan differs from the lin-
guistic input received by the other two groups in the U.S.
must also be considered. In particular, NM speakers’ initial
English input is likely to have been accented, making it pos-
sible that the amount of non-approximation to English pho-
netic norms seen for a given NM speaker, rather than being
attributable to that one speaker, had actually accumulated
over a chain of L2 acquirers. For that matter, one wonders
whether the early Mandarin input received by HL speakers
born in the U.S. (e.g., the Mandarin spoken by their parents,
who had for the most part been living in the U.S. for a con-
siderable period of time prior to their birth) would have dif-
fered significantly from the Mandarin input they would have
received in a country where English is not so widely spoken.
These are questions that will require more detailed study of
the relevant acquisition situations to be able to answer, but
there is reason to believe that if there were such an effect of
inaccurate input here, it would stand to be the strongest in
the NM speakers, who might have been exposed to heavily
accented L2 English, whereas HL speakers were probably
exposed to no worse than native Mandarin that had “drifted”
(Sancier and Fowler, 1997;Chang, 2010) in an English-
speaking environment.
Finally, it should be observed that although in experi-
ment 1, HL speakers seemed to outperform both the NM
group and the L2 group in producing cross-linguistic con-
trast, in experiments 2 and 3 they appeared to pattern to-
gether with one other group in outperforming the third
group. In experiment 2, both the HL group and the NM
group surpassed the L2 group in producing acoustic distance
(in terms of VOT) between the similar Mandarin aspirated
and English voiceless plosives; likewise, in experiment 3,
both the HL group and the L2 group surpassed the NM group
in distinguishing the similar fricatives /§/ and =$=. Why did
this occur?
As for why NM speakers, themselves L2 learners of
English, outperformed L2 learners of Mandarin in distin-
guishing the two long-lag VOT categories in experiment 2,
one possibility is that NM speakers, being accustomed to
very long VOTs for their native long-lag VOT category, are
attuned to picking out VOTs that are too short to qualify as
Mandarin aspirated stops, thus leading them to perceive Eng-
lish voiceless stops as significantly less aspirated than Man-
darin aspirated stops. On the other hand, L2 learners of
Mandarin might simply be focused on whether a VOT is
long enough to be an exemplar of English voiceless as
opposed to English voiced, in which case they may be rela-
tively insensitive to the difference between Mandarin aspi-
rated and English voiceless, since in initial position both are
aspirated enough to pass the VOT boundary that is salient
for them.
The explanation for L2 learners of Mandarin outper-
forming NM speakers in distinguishing Mandarin /§/ and
English =$=is likely quite different. Here it should be noted
that these two types of L2 learners face very different tasks:
Mandarin-speaking L2 learners of English, who already
have two L1 post-alveolar fricatives, need to learn just one
L2 post-alveolar fricative category, while English-speaking
L2 learners of Mandarin, who have only one L1 post-alveo-
lar fricative, need to learn to distinguish two L2 post-alveo-
lar fricative categories. This one-to-many vs many-to-one
contrast does not in itself account for why the two groups
seem to have reached disparate learning outcomes, but it
does suggest a possible difference in learning strategies and
perhaps instructional input as well. While NM speakers can
afford to produce English =$=relatively inaccurately (e.g., as
[§]) with no serious consequences for intelligibility, L2 learn-
ers of Mandarin cannot similarly afford to produce the Man-
darin fricatives inaccurately because there is a real chance
J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin 3975
Author's complimentary copy

they will be misunderstood, due to the relative crowdedness
of the Mandarin fricative inventory. As a result of this pres-
sure, they are probably highly conscious of their pronuncia-
tion of these segments, and formal instruction might serve to
amplify their efforts to differentiate these fricatives from
each other and from their L1 inventory by, for example, exo-
ticizing the pronunciation of /§/ as “retroflex” (which, as
mentioned in Sec. III C, is actually a misnomer). Thus, the
role played by explicit knowledge and instruction in L2 pro-
duction may be largely responsible for this latter result.
B. “New” vs “similar” categories
Just as the cross-linguistic influence seen in experiments
1 and 2 was consistent with the cross-linguistic influence
documented in Flege (1987), the results obtained in experi-
ment 1 for the production of Mandarin =y=were also con-
sistent with Flege’s (1987) results for French =y=:L2
learners did not differ appreciably from NM speakers in their
phonetic space for =y=, suggesting that it was indeed per-
ceived as a “new” sound. However, an alternative explana-
tion for L2 learners’ relatively accurate production of =y=
exists—namely, that they were not producing anything new
at all. In other words, they may have simply been drawing
upon vowel tokens that were already in their repertoire: par-
ticularly fronted versions of English =u=. It has been pro-
posed that due to the fact that American English =u=, which
is already relatively front on average, is further fronted in
the context of alveolars, the front vowel =y=may not
actually constitute a “new” vowel for English-speaking L2
learners, but instead a “similar” vowel, at least in the context
of alveolars (Strange et al., 2007;Levy, 2009;Levy and
Law, 2010). Documented perceptual assimilation patterns in
which English speakers identify German and French =y=as
closest to English =u=(e.g., Polka and Bohn, 1996;Strange
et al., 2004) are consistent with this idea. Nevertheless, this
proposal does not provide a convincing account of the cur-
rent findings for three reasons.
First, it is unclear whether the requisite pattern of per-
ceptual assimilation occurs for =y=to become categorically
linked to English =u=. Although German =y=in the context
of alveolars, for example, is assimilated to English =u=in
cross-language labeling tasks, German =u=in the same con-
text is also assimilated to English =u=and is, moreover, rated
as a better exemplar of English =u=than =y=is (Polka and
Bohn, 1996). These data suggest that, if L2 German learners
are to maintain contrast between German =u=and German
=y=, only one of these vowels will be perceptually linked to
English =u=, and that vowel will be German =u=, not Ger-
man =y=(which is actually the acoustically closer vowel).
Such a linkage, in which the L2 category linked to the L1
category is not the phonetically closest L2 category, is plau-
sible in light of findings showing that perceptual assimilation
of vowels does not necessarily follow from strict phonetic
proximity (e.g., French =y=is acoustically closer to English
=i=than to English =u=, yet is consistently assimilated to
=u=; see Strange et al., 2004).
The account of =y=as a “similar” vowel is also not sup-
ported by the distributional characteristics of L2 learners’
vowel productions in this study. For L2 learners’ accurate
production of =y=to be an artifact of the existence of pho-
netically similar, fronted realizations of English =u=, there
should be significant overlap in the F
2
distributions of these
two categories, yet most L2 learner participants showed little
to no overlap between these two distributions. For example,
participant L22 had an F
2
range of 8.63–12.33 Bark for Eng-
lish =u=vs 12.30–13.39 Bark for Mandarin =y=; similarly,
the relevant F
2
ranges for participants L24 and L25 were,
respectively, 9.02–13.17 vs 13.47–14.59 and 7.74–12.78 vs
12.66–14.01. Thus, it is not the case that L2 learners pro-
duced =y=by simply recruiting their most fronted exemplars
of English =u=; in fact, they produced =y=with significantly
higher F
2
values than those of their most fronted =u=
productions.
Finally, given the phonetic norms of Mandarin =y=and
English =u=, the hypothetical classification of =y=as a vowel
“similar” to English =u=is inconsistent with how =y=was
produced by the L2 learners in this study. If Mandarin =y=
were treated as a similar vowel, according to the SLM it
would be perceptually linked to English =u=and, therefore,
produced inaccurately (with too-low F
2
values) under influ-
ence from English =u=, since English =u=is on average char-
acterized by an F
2
that is lower than that of Mandarin =y=
(Table I), even if its most fronted realizations might show F
2
values that overlap with those of =y=. What was found
instead, however, was that =y=was produced by L2 learners
relatively accurately—right in the same region with the =y=
productions of NM speakers, rather than retracted in the F
2
dimension.
Thus, the findings of this study support the initial pre-
diction that Mandarin =y=would constitute a “new” vowel
for English speakers, confirming that—consistent with the
PAM-L2—category equivalence in L2 speech acquisition
can be based on phonological proximity. Furthermore, the
findings suggest that when phonetic and phonological con-
siderations conflict in these situations, the higher-level
phonological considerations override the lower-level pho-
netic ones. However, it may not always be the case that
higher-level proximity preempts lower-level proximity.
The exact nature of their interaction with respect to deter-
mining category equivalence remains a question for future
research.
ACKNOWLEDGMENTS
This work was supported in part by a National Science
Foundation Graduate Research Fellowship to the first author
and a grant from the Abigail Reynolds Hodgen Publication
Fund. The authors are grateful to Susanne Gahl, Sharon Ink-
elas, Keith Johnson, participants in a fall 2007 UC Berkeley
seminar on phonological learning, several anonymous
reviewers, and audiences at the University of California,
Berkeley, the University of Pennsylvania, and the University
of Chicago for helpful comments and discussion. Portions of
these data are also discussed in the University of Pennsylva-
nia Working Papers in Linguistics (Chang et al., 2009) and
the Proceedings from the Annual Meeting of the Chicago
Linguistic Society (Chang et al., 2010).
3976 J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin
Author's complimentary copy

APPENDIX A: PARTICIPANT INFORMATION
See Tables IV,V, and VI.
TABLE IV. Background information, NM speaker group. The NM group comprised NM speakers who were born and educated up to at least seventh grade in
mainland China (MC) or Taiwan (TW).
PID Gender Age Place of birth Age of arrival Other languages spoken
N1 F 33 MC 30 English
N2 M 32 TW 26 English
N3 M 40 MC 26 Shanghainese, English
N4 F 35 MC 34 English
N5 F 20 MC 16 Cantonese, English
N6 F 19 TW 13 English
TABLE V. Background information, HL speaker group. The HL group comprised Chinese Americans who were born to Mandarin-speaking parents and had
prior experience with Mandarin in the home. Participants in the HE subgroup had extensive exposure to Mandarin, while participants in the LE subgroup had
more limited exposure to Mandarin.
Current use of Mandarin
PID Gender Age Places lived
Language exposure
at home To grandparents To parents To siblings
HE H7 M 24 USA (0–6); Mandarin All the time Mostly Half the time
TW (6–18);
USA (18–24)
H8 M 21 TW (0–9); Mandarin, Japanese, Taiwanese All the time Mostly Half the time
Singapore (9–13);
USA (13–21)
H9 F 19 USA (0–19) Mandarin — Mostly Half the time
H10 F 20 TW (0–5); Mandarin, Taiwanese All the time Mostly Seldom
USA (5–20)
H11 M 20 TW (0–3); Mandarin, English All the time All the time (mixed with English) Seldom
USA (3–20)
H12 M 20 MC (0–3.5); Mandarin All the time Mostly —
USA (3.5–20)
H13 F 23 MC (0–10); Mandarin, occasional — Half the time —
USA (10–23) Fuzhounese
H14 F 20 USA (0–20) Mandarin, occasional All the time Mostly Seldom
Taiwanese
H15 M 22 USA (0–5); Mandarin Mostly Mostly Seldom
Singapore (5–11.5);
Qatar (11.5–14);
TW (14–15);
USA (15–22)
LE H16 F 18 USA (0–18) English, Mandarin — Half the time Seldom
H17 F 20 USA (0–20) Mandarin, occasional All the time Seldom Seldom
Taiwanese
H18 M 21 USA (0–21) Mandarin All the time Sometimes Never
H19 M 20 USA (0–20) Mandarin, English All the time Rarely Never
H20 F 21 MC (0–2); Mandarin, English Never Never Never
USA (2–21)
H21 F 20 USA (0–20) English, occasional Mandarin Sometimes Never Never
TABLE VI. Background information, late L2 learner group. The L2 group comprised native English speakers who were born and educated in the U.S. and
started to learn Mandarin after the age of 18.
PID Gender Age Other languages spoken in family Mandarin experience
L22 M 19 — 2 semesters of college-level Mandarin (18–19)
L23 F 19 — 1 intensive summer session (¼2 regular semesters) of college-level Mandarin (18)
L24 F 27 — 1 year living in Beijing (26–27), including a 3-month conversation course
L25 F 24 Cebuano 2 years of college-level Mandarin (20–23)
L26 M 19 Korean 2.5-week trip to Taiwan
J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin 3977
Author's complimentary copy

APPENDIX B: LIST OF STIMULI
See Tables VII,VIII, and IX.
1
Note that contradictory null results have emerged from the work of Pallier
and colleagues (Pallier et al., 2003), who, examining subjects from a dif-
ferent HL situation (Korean adoptees in France), have failed to find a per-
ceptual or even low-level neural advantage for the HL in individuals with
early HL exposure, suggesting that some re-learning of the HL, or at least
intermittent exposure to it, may be necessary for distant HL experience to
become readily accessible.
2
An anonymous reviewer questioned the analysis of participants in terms of
groups, as well as the basis for dividing participants into the groups
described in this section. It should be noted that there are a number of ways
the data in this study could have been analyzed; however, a group approach
was the most appropriate because the focus was on differences between HL
speakers (i.e., individuals with early exposure to two languages) and indi-
viduals without early exposure to two languages. The advantages of a group
approach over, e.g., a correlational approach were twofold. First, a group
approach obviated the need to arbitrarily define one background characteris-
tic as the independent variable or to compute a holistic index of Mandarin
experience based on several background characteristics as the independent
variable. This was beneficial because HL speakers often have complex lan-
guage backgrounds and residential histories (as seen in Table V)thatmake
it unclear how two relatively non-proficient HL speakers should stack up
relative to one another. Second, a group approach made it possible to
include variability in the analysis of HL speakers, in keeping with the char-
acteristics of this population. In the interest of yielding more generalizable
results, the decision was made not to arbitrarily examine one specific level
of HL experience. Instead, HL speakers with a range of experience were
included to more accurately represent the population of HL Mandarin
speakers (as described in Sec. IC), and they were divided into subgroups in
a manner that was straightforward and highly replicable (as described in this
section).
3
Gender was entered as a factor in all of the statistical analyses in experi-
ments 1–3 because it has been shown to have an effect on all of the acous-
tic dimensions investigated (an effect that is, moreover, usually
perceptible to listeners): vowel formants (Whiteside, 1998b,c), VOT
(Swartz 1992;Whiteside and Irving 1998; though see Morris et al., 2008),
and spectral properties of frication (Whiteside, 1998a). For a recent review
of gender differences in speech, see Simpson (2009). Thus, even though
gender was not the focus of any analysis, its potential effect on the data
was accounted for explicitly, rather than ignored, so that any possible
interactions with other factors of interest could be accounted for.
4
The segmental context in which a vowel occurs—in particular, the place
of articulation of flanking consonants—has been shown to have a signifi-
cant, predictable effect on the quality of the vowel. American English =u=,
for example, has been shown to be significantly fronted in the context of
alveolars (Hillenbrand et al., 2001;Strange et al., 2007). Place of articula-
tion has also been shown to have a predictable effect on VOT (e.g., Lisker
and Abramson, 1967;Nearey and Rochet, 1994;Liu et al., 2007), which
was measured in experiment 2. Thus, although place was not the focus of
any analysis, given its probable effect on the data it was entered into the
analyses so that any possible interactions with other factors could be
accounted for.
5
For the dependent variables of F
1
and F
2
in experiment 1 as well as the de-
pendent variable of VOT in experiment 2 (Sec. III B), there were signifi-
cant two- and three-way interactions involving Place: Language Place,
Place Vowel, and Language Place Vowel. However, these interac-
tions are not of concern here, so they will not be discussed further.
6
Vowel environment, like place of articulation, was entered into the analy-
ses because it has been shown to have a significant effect on VOT (Klatt,
1975;Nearey and Rochet, 1994).
TABLE VII. Critical stimuli in experiment 1 (vowel quality). The vowels
of interest were Mandarin and English =o
u
=, Mandarin and English =u=, and
Mandarin =y=.
Mid back
rounded Mandarin =o
u
=English =o
u
=
“bean” dote =do
u
t=
“enough” goat =go
u
t=
“to dissect” pope =po
u
p=
“through” tote =to
u
t=
“button” coat =ko
u
t=
“Europe” oat =o
u
t=
“meat” wrote
boat =bo
u
t=
cope =ko
u
p=
host =ho
u
st=
High back
rounded Mandarin =u=English =u=
“no; not” boot =but=
“belly” dupe =dup=
“to take care of” goose =gus=
“waterfall” poop =pup=
“rabbit” toot =tut=
“garage” coup =ku=
“household” hoot =hut=
“to enter” root
“not have” choose =t$uz=
“summer” shoe =$u=
shoot =$ut=
High front
rounded Mandarin =y=
“green”
“female”
“encounter”
TABLE VIII. Critical stimuli in experiment 2 (laryngeal contrast). The la-
ryngeal categories of interest were Mandarin unaspirated, Mandarin aspi-
rated, English voiced, and English voiceless.
Short-lag stops Mandarin unaspirated English voiced
“father” boat =bo
u
t=
“no; not” boot =but=
“bean” dote =do
u
t=
“belly” dupe =dup=
“enough” goat =go
u
t=
“to take care of” goose =gus=
Long-lag stops Mandarin aspirated English voiceless
“to dissect” pope =po
u
p=
“waterfall” poop =pup=
“through” tote =to
u
t=
“rabbit” toot =tut=
“button” coat =ko
u
t=
“garage” coup =ku=
TABLE IX. Critical stimuli in experiment 3 (place of articulation). The pla-
ces of articulation of interest were the post-alveolar places: Mandarin retro-
flex, Mandarin alveolo-palatal, and English palato-alveolar.
Mandarin English
Retroflex Alveolo-palatal Palato-alveolar
“suddenly” “below” shot =$At=
“sand” “shrimp” shop =$Ap=
“what” “govern”
“stupid”
3978 J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin
Author's complimentary copy

7
Formant transition data showed that alveolo-palatal //, having by far the
lowest F
1
onset (approximately 1 Bark lower than that of /§/ and =$=in all
groups) and highest F
2
onset (approximately 2 Bark higher than that of /§/
and =$=in all groups), was clearly the most “palatalized” of the three post-
alveolar fricatives. However, the formant data did not clearly differentiate
/§/ and =$=. Thus, the focus in this study was on centroid frequency data,
which were supplemented with PAF data as described in Sec. II D.
Andersen, R. W. (1982). “Determining the linguistic attributes of language
attrition,” in The Loss of Language Skills, edited by R. D. Lambert and B.
F. Freed (Newbury House, Rowley, MA), pp. 83–118.
Andrews, D. R. (1999). Sociocultural Perspectives on Language Change in
Diaspora: Soviet Immigrants in the United States (John Benjamins Pub-
lishing, Amsterdam), pp. 1–200.
Au, T. K., Knightly, L. M., Jun, S.-A., and Oh, J. S. (2002). “Overhearing a
language during childhood,” Psychol. Sci. 13, 238–243.
Au, T. K., and Oh, J. S. (2009). “Korean as a heritage language,” in Hand-
book of East Asian Psycholinguistics, Vol. III: Korean, edited by C. Lee,
G. B. Simpson, and Y. Kim (Cambridge University Press, Cambridge),
pp. 268–275.
Au, T. K., Oh, J. S., Knightly, L. M., Jun, S.-A., and Romo, L. F. (2008).
“Salvaging a childhood language,” J. Mem. Lang. 58, 998–1011.
Au, T. K., and Romo, L. F. (1997). “Does childhood language experience
help adult learners?,” in The Cognitive Processing of Chinese and Related
Asian Languages, edited by H.-C. Chen (Chinese University Press, Hong
Kong), pp. 417–443.
Best, C. T. (1994). “The emergence of native-language phonological influ-
ences in infants: A perceptual assimilation model,” in The Development of
Speech Perception: The Transition from Speech Sounds to Spoken Words,
edited by J. C. Goodman and H. C. Nusbaum (MIT Press, Cambridge,
MA), pp. 167–224.
Best, C. T., and Tyler, M. D. (2007). “Nonnative and second-language
speech perception: Commonalities and complementarities,” in Language
Experience in Second Language Speech Learning: In Honor of James
Emil Flege, edited by O.-S. Bohn and M. J. Munro (John Benjamins Pub-
lishing, Amsterdam, The Netherlands), pp. 13–34.
Boersma, P., and Weenink, D. (2008). “PRAAT: Doing phonetics by computer
(version 5.0.26) [computer program],” http://www.praat.org (Last viewed
June 15, 2008).
Campbell, L., and Muntzel, M. C. (1989). “The structural consequences of
language death,” in Investigating Obsolescence: Studies in Language Con-
traction and Death, edited by N. C. Dorian (Cambridge University Press,
Cambridge, UK), pp. 181–196.
Campbell, R. N., and Rosenthal, J. W. (2000). “Heritage languages,” in
Handbook of Undergraduate Second Language Education, edited by J. W.
Rosenthal (Erlbaum, Mahwah, NJ), pp. 165–184.
Chang, C. B. (2009). “English loanword adaptation in Burmese,” J. South-
east Asian Linguist. Soc. 1, 77–94.
Chang, C. B. (2010). “First language phonetic drift during second language
acquisition,” Ph.D. dissertation, University of California, Berkeley, CA.
Chang, C. B., Haynes, E. F., Yao, Y., and Rhodes, R. (2009). “A tale
of five fricatives: Consonantal contrast in heritage speakers of Man-
darin,” University of Pennsylvania Working Papers in Linguistics 15,
37–43.
Chang, C. B., Haynes, E. F., Yao, Y., and Rhodes, R. (2010). “The phonetic
space of phonological categories in heritage speakers of Mandarin,” in
Proceedings from the 44th Annual Meeting of the Chicago Linguistic Soci-
ety: The Main Session, edited by M. Bane, J. Bueno, T. Grano, A. Grot-
berg, and Y. McNabb (Chicago Linguistic Society, Chicago, IL), pp. 31–
45.
Chen, Y. (2006). “Production of tense-lax contrast by Mandarin speakers of
English,” Folia Phoniatr. Logop. 58, 240–249.
Flege, J. E. (1987). “The production of ‘new’ and ‘similar’ phones in a for-
eign language: Evidence for the effect of equivalence classification,”
J. Phonetics 15, 47–65.
Flege, J. E. (1995). “Second language speech learning: Theory, findings,
and problems,” in Speech Perception and Linguistic Experience: Issues in
Cross-Language Research, edited by W. Strange (York Press, Baltimore,
MD), pp. 233–272.
Flege, J. E., and Hillenbrand, J. (1984). “Limits on phonetic accuracy in for-
eign language speech production,” J. Acoust. Soc. Am. 76, 708–721.
Godson, L. (2003). “Phonetics of language attrition: Vowel production and
articulatory setting in the speech of Western Armenian heritage speakers,”
Ph.D. dissertation, University of California, San Diego, CA.
Godson, L. (2004). “Vowel production in the speech of Western Armenian
heritage speakers,” Heritage Lang. J. 2, 44–69.
Goodfellow, A. M. (2005). Talking in Context: Language and Identity in
Kwakwaka’wakw Society (McGill-Queen’s University Press, Montreal,
Canada), pp. 1–219.
Hagiwara, R. (1997). “Dialect variation and formant frequency: The Ameri-
can English vowels revisited,” J. Acoust. Soc. Am. 102, 655–658.
Hillenbrand, J. M., Clark, M. J., and Nearey, T. M. (2001). “Effects of con-
sonant environment on vowel formant patterns,” J. Acoust. Soc. Am. 109,
748–763.
Jia, G., Strange, W., Wu, Y., Collado, J., and Guan, Q. (2006). “Perception
and production of English vowels by Mandarin speakers: Age-related dif-
ferences vary with amount of L2 exposure,” J. Acoust. Soc. Am. 119,
1118–1130.
Jiang, H. (2008). “Effect of L2 phonetic learning on L1 vowels,” Ph.D. dis-
sertation, Simon Fraser University, Vancouver, Canada.
Jiang, H. (2010). “Effect of L2 phonetic learning on the production of L1
vowels: A study of Mandarin-English bilinguals in Canada,” in New
Sounds 2010: Proceedings of the 6th International Symposium on the Ac-
quisition of Second Language Speech, edited by K. Dziubalska-Kolaczyk,
M. Wrembel, and M. Kul (Adam Mickiewicz University, Poznan´, Poland),
pp. 227–232.
Jongman, A., Wayland, R., and Wong, S. (2000). “Acoustic characteristics
of English fricatives,” J. Acoust. Soc. Am. 108, 1252–1263.
Kang, Y. (2008). “Interlanguage segmental mapping as evidence for the na-
ture of lexical representation,” Language and Linguistics Compass 2, 103–
118.
Klatt, D. H. (1975). “Voice onset time, frication, and aspiration in word-ini-
tial consonant clusters,” J. Speech Hear. Res. 18, 686–706.
Knightly, L. M., Jun, S.-A., Oh, J. S., and Au, T. K. (2003). “Production
benefits of childhood overhearing,” J. Acoust. Soc. Am. 114, 465–474.
Kuhl, P. K., Williams, K. A., Lacerda, F., Stevens, K. N., and Lindblom, B.
(1992). “Linguistic experience alters phonetic perception in infants by 6
months of age,” Science 255, 606–608.
LaCharite´, D., and Paradis, C. (2005). “Category preservation and proximity
versus phonetic approximation in loanword adaptation,” Linguistic Inquiry
36, 223–258.
Ladefoged, P. (2003). Phonetic Data Analysis (Blackwell Publishing, Mal-
den, MA), pp. 96–101 and 156–158.
Ladefoged, P. (2005). Vowels and Consonants, 2nd ed. (Blackwell Publish-
ing, Malden, MA), pp. 40–43.
Ladefoged, P., and Maddieson, I. (1996). The Sounds of the World’s Lan-
guages (Blackwell Publishers, Oxford, UK), pp. 148–154.
Laeufer, C. (1996). “Towards a typology of bilingual phonological sys-
tems,” in Second-Language Speech: Structure and Process, edited by A.
James and J. Leather (Mouton de Gruyter, Berlin), pp. 325–342.
Levy, E. S. (2009). “Language experience and consonantal context effects
on perceptual assimilation of French vowels by American-English learners
of French,” J. Acoust. Soc. Am. 125, 1138–1152.
Levy, E. S., and Law, F. F. (2010). “Production of French vowels by Ameri-
can-English learners of French: Language experience, consonantal con-
text, and the perception-production relationship,” J. Acoust. Soc. Am. 128,
1290–1305.
Li, D., and Duff, P. A. (2008). “Issues in Chinese heritage language educa-
tion and research at the postsecondary level,” in Chinese as a Heritage
Language: Fostering Rooted World Citizenry, edited by A. W. He and Y.
Xiao (National Foreign Language Resource Center, University of Hawaii,
Honolulu, HI), pp. 13–32.
Li, F., Edwards, J., and Beckman, M. (2007). “Spectral measures for
sibilant fricatives of English, Japanese, and Mandarin Chinese,” in
Proceedings of the 16th International Congress of Phonetic Sciences,
edited by J. Trouvain and W. J. Barry (Pirrot, Dudweiler, Germany),
pp. 917–920.
Lin, T., and Wang, L. (1992). Yu Yin Xue Jiao Cheng (in Chinese) (Beijing
University Press, Beijing, China), pp. 1–208.
Lisker, L., and Abramson, A. S. (1964). “A cross-language study of voicing
in initial stops: Acoustical measurements,” Word 20, 384–422.
Lisker, L., and Abramson, A. S. (1967). “Some effects of context on voice
onset time in English stops,” Lang. Speech 10, 1–28.
Liu, H., Ng, M. L., Wan, M., Wang, S., and Zhang, Y. (2007). “Effects of
place of articulation and aspiration on voice onset time in Mandarin esoph-
ageal speech,” Folia Phoniatr. Logop. 59, 147–154.
Major, R. C. (1992). “Losing English as a first language,” The Modern Lan-
guage Journal 76, 190–208.
J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin 3979
Author's complimentary copy

Montrul, S. A. (2008). Incomplete Acquisition in Bilingualism: Re-examin-
ing the Age Factor (John Benjamins Publishing, Amsterdam), pp. 1–312.
Morris, R. J., McCrea, C. R., and Herring, K. D. (2008). “Voice onset time
differences between adult males and females: Isolated syllables,” J. Pho-
netics 36, 308–317.
Nearey, T. M., and Rochet, B. L. (1994). “Effects of place of articulation
and vowel context on VOT production and perception for French and Eng-
lish stops,” J. Int. Phonetic Assoc. 24, 1–18.
Oh, J., Jun, S.-A., Knightly, L., and Au, T. (2003). “Holding on to childhood
language memory,” Cognition 86, B53–B64.
Oh, J. S., and Au, T. K. (2005). “Learning Spanish as a heritage language:
The role of sociocultural background variables,” Language, Culture and
Curriculum 18, 229–241.
Oh, J. S., Au, T. K., and Jun, S.-A. (2002). “Benefits of childhood language
experience for adult L2 learners’ phonology,” in Proceedings of the 26th
Annual Boston University Conference on Language Development, Vol. 2,
edited by B. Skarabela, S. Fish, and A. H.-J. Do (Cascadilla Press, Somer-
ville, MA), pp. 464–472.
Oh, J. S., Au, T. K., and Jun, S.-A. (2010). “Early childhood language mem-
ory in the speech perception of international adoptees,” J. Child Lang. 37,
1123–1132.
Pallier, C., Dehaene, S., Poline, J.-B., LeBihan, D., Argenti, A.-M., Dupoux,
E., and Mehler, J. (2003). “Brain imaging of language plasticity in adopted
adults: Can a second language replace the first?,” Cereb. Cortex 13, 155–
161.
Polinsky, M. (2008). “Gender under incomplete acquisition: Heritage speak-
ers’ knowledge of noun categorization,” Heritage Lang. J. 6, 40–71.
Polka, L., and Bohn, O.-S. (1996). “A cross-language comparison of vowel
perception in English-learning and German-learning infants,” J. Acoust.
Soc. Am. 100, 577–592.
Sancier, M. L., and Fowler, C. A. (1997). “Gestural drift in a bilingual
speaker of Brazilian Portuguese and English,” J. Phonetics 27, 421–436.
Simpson, A. P. (2009). “Phonetic differences between male and female
speech,” Language and Linguistics Compass 3, 621–640.
Strange, W., Levy, E., and Lehnholf, R., Jr. (2004). “Perceptual assimilation
of French and German vowels by American English monolinguals: Acous-
tic similarity does not predict perceptual similarity,” J. Acoust. Soc. Am.
115, 2606.
Strange, W., Weber, A., Levy, E. S., Shafiro, V., Hisagi, M., and Nishi, K. (2007).
“Acoustic variability within and across German, French and American English
vowels: Phonetic context effects,” J. Acoust. Soc. Am. 122, 1111–1129.
Svantesson, J.-O. (1986). “Acoustic analysis of Chinese fricatives and
affricates,” J. Chin. Linguist. 14, 53–70.
Swartz, B. L. (1992). “Gender difference in voice onset time,” Percept. Mot.
Skills 75, 983–992.
Tees, R. C., and Werker, J. F. (1984). “Perceptual flexibility: Maintenance
or recovery of the ability to discriminate non-native speech sounds,” Can.
J. Psychol. 38, 579–590.
Traunmu¨ller, H. (1990). “Analytical expressions for the tonotopic sensory
scale,” J. Acoust. Soc. Am. 88, 97–100.
Werker, J. F., and Tees, R. C. (1984). “Cross-language speech perception:
Evidence for perceptual reorganization during the first year of life,” Infant
Behav. Dev. 7, 49–63.
Whiteside, S. P. (1998a). “Identification of a speaker’s sex: A fricative
study,” Percept. Mot. Skills 86, 587–591.
Whiteside, S. P. (1998b). “Identification of a speaker’s sex: A study of vow-
els,” Percept. Mot. Skills 86, 579–584.
Whiteside, S. P. (1998c). “The identification of a speaker’s sex from synthe-
sized vowels,” Percept. Mot. Skills 87, 595–600.
Whiteside, S. P., and Irving, C. J. (1998). “Speakers’ sex differences in
voice onset time: A study of isolated word production,” Percept. Mot.
Skills 86, 651–654.
Wu, Z., and Lin, M. (1989). Shi Yan Yu Yin Xue Gai Yao (in Chinese)
(Higher Education Press, Beijing, China), pp. 1–347.
Young, N. (2007). “A case study of phonological attrition of Taiwanese
Mandarin in California,” UC Berkeley Phonology Lab Annual Report
2007, pp. 71–126.
Zhang, F., and Yin, P. (2009). “A study of pronunciation problems of Eng-
lish learners in China,” Asian Social Science 5, 141–146.
Zhang, Y., Nissen, S. L., and Francis, A. L. (2008). “Acoustic characteristics
of English lexical stress produced by native Mandarin speakers,” J.
Acoust. Soc. Am. 123, 4498–4513.
3980 J. Acoust. Soc. Am., Vol. 129, No. 6, June 2011 Chang et al.: Contrast in heritage speakers of Mandarin
Author's complimentary copy