Content uploaded by Amalia Arvaniti
Author content
All content in this area was uploaded by Amalia Arvaniti on Apr 19, 2016
Content may be subject to copyright.
The form and use of uptalk in Southern Californian English
Amanda Ritchart1, Amalia Arvaniti2
1 University of California, San Diego
2 University of Kent
aritchart@ucsd.edu, a.arvaniti@kent.ac.uk
Abstract
This study examines the phonetics, phonology and pragmatic
function of uptalk, utterance-final rising pitch movements, as
used in Southern Californian English. Twelve female and
eleven male speakers were recorded in a variety of tasks.
Instances of uptalk were coded for discourse function
(statement, question, confirmation request, floor holding)
based on context. The excursion of the pitch rise and the
distance of the rise start from the onset of the utterance’s last
stressed vowel were also measured. Confirmation requests and
floor holding showed variable realization. Questions, on the
other hand, showed a rise that typically started within the
stressed vowel and had a large pitch excursion, while uptalk
“proper”, i.e. uptalk used with statements, exhibited both a
smaller pitch excursion and a later rise that often started after
vowel offset. This pattern suggests that statements have a L*
L-H% melody while questions have L* H-H%. Gender
differences were also found: female speakers used uptalk more
often than males, and showed greater pitch excursion and later
alignment, all else being equal. Other social parameters,
however, such as social class and linguistic background, did
not affect the use of uptalk.
Index Terms: intonation, HRT, uptalk, English,
sociolinguistics, gender
1. Introduction
Rising melodies used with statements, commonly referred to
as uptalk or high rise terminals are common in many varieties
of English. Here we use the term uptalk which better reflects
the Southern Californian patterns that are the focus of our
investigation. Research on uptalk in some varieties is quite
extensive, but has often been impressionistic [1]. The varieties
that have been most investigated include those spoken in
Australia and New Zealand as well as UK varieties from
Glasgow and Belfast [1]–[5] (and references therein).
These studies document that different tunes are used for
uptalk across varieties. Thus, [1] report that Australian uptalk
is realized as either L* H-H% or H* H-H%. For Glasgow,
L*H H-L% is proposed for the “rise-plateau-slump” type of
uptalk, with suspension of the rule that in other English
varieties upsteps a L% after a H- phrase accent [2]. In [1],
New Zealand uptalk is analyzed as reflecting two main
patterns, LH* H-H% and L* H-H% (based on [3]), but a
newer study suggests that New Zealand English may exhibit
change in progress with respect to uptalk [4].
In addition to differences in form, uptalk across varieties
of English is used for different purposes. Thus, [1] report that
in Australian English upstep is used both with questions and
declarative statements; upstepped statements are particularly
frequent when the speaker wishes to hold the floor. This leads
[1] to suggest that the intonational difference between
statements and questions and that between statements and
continuation is neutralized in Australian English. New Zealand
English also uses uptalk for both statements and questions but
the tunes used for each function are becoming increasingly
distinct [4]. Research on Glasgow and Belfast English, e.g.,
[2] and [5], focused on form rather than function, but recent
research suggests that uptalk, in Belfast at least, may have its
origins in list intonation [6].
One of the varieties that is stereotypically known as
exhibiting use of uptalk is Californian English, particularly the
varieties spoken in the south (henceforth SoCal). The use of
uptalk in SoCal is often referred to as “valley girl speak” and
is often assumed to be a feature of younger females only,
though no studies exist, to our knowledge, confirming or
refuting this general lay perception.
Here we present data from SoCal English which show
that the use of uptalk is widespread in this variety and exhibits
gender-related variation. We further show that SoCal uptalk
tunes are different from those reported for other varieties of
English, and that speakers retain systematic differences
between uptalk used in statements and other types of uptalk,
such as pitch rises used with questions. Differences apply both
to the tunes employed and to the scaling of the rise.
2. Methods
2.1. Speakers
Twenty-three speakers were recorded for the study, eleven
male and twelve female. They were all native speakers of
SoCal English, from San Diego (N = 7), Orange (N = 6), Los
Angeles (N = 8), and Riverside (N = 2) counties. Fifteen were
monolingual, while the other eight reported being bilingual in
English and one of the following languages: Vietnamese (N =
3), Japanese (N = 1), Armenian (N = 1), Assyrian (N = 1),
Spanish (N = 1), and Cantonese (N = 1). The speakers’ ethnic
backgrounds varied: twelve self-identified as Asian, six as
Hispanic and five as White.
The MacArthur Scale of Subjective Social Status (cf.
[7], [8]) was used to determine the speakers’ socioeconomic
status or SEC, a rather fluid concept in California. Participants
found the use of the scale easy and intuitive. They were
classed into three groups based on their responses: lower
(rungs 1-4, N = 4), middle (rungs 5-7, N = 13) and upper
(rungs 7-10, N = 6).
2.2. Materials, Tasks and Procedures
Recordings took place in the recording studio of the UCSD
Phonetics Lab, using an AD converter at 48 KHz and 16-bit
quantization. Four types of data were collected from each
speaker: (a) map task; (b1) reading of the transcript of a
popular sitcom scene; (b2) retelling of the sitcom scene; (c)
controlled materials consisting of isolated questions and
statements. For the first 17 participants, tasks were presented
in the following order: (c), (b), (a). (For task (b), the retelling
of the clip always followed the reading of the transcript.) To
control for possible order effects, the order of tasks was
counterbalanced for the other recordings and each participant
was randomly assigned to one of three possible orders (Latin
square design): abc, bca, or cab. However, given that the data
consist largely of spontaneous speech it is unlikely that order
could have severely biased speaker productions with respect to
uptalk.
For the map task, maps with local (or local sounding)
landmarks were designed as illustrated in Figure 1. In the map
task, the participants acted as leaders with the follower being
either the first author or an undergraduate research assistant
(both females and native SoCal speakers). For task (b1), a
scene from either Scrubs or How I met your mother was used;
the show chosen was the one the participant was less familiar
with. Lack of familiarity was sought so that speakers would
not imitate the actors’ accents. The scene was muted and
participants were given a transcript of the dialogue while they
watched the clip. When they were ready, they chose which
character they were most comfortable reading from, and
participated in reading aloud the transcript in a dialogue with
the experimenter. In task (b2), participants had to retell the
same scene in their own words. For task (c), the participants
read aloud a list of 49 sentences. These were statements and
questions constructed for the study. In these sentences, the
number of syllables and position of stress was controlled in
order to examine the realization of specific tonal events; see
(1) for an illustration. Here we report on the results from tasks
(a) and (b2).
(1) a. Did Anne and Mel eat the lime?
b. Did Annabelle and Melinda eat the lime?
Figure 1: The Instruction Giver’s map used in the map task.
2.3. Analysis and Measurements
The analysis involved both a categorization of each instance of
uptalk in terms of its discourse function and acoustic
measurements with respect to the alignment and scaling of the
pitch rise associated with each uptalk token.
Specifically, instances of uptalk were classed in one of
four discourse functions: question, statement, holding the
floor, and confirmation request. A token was considered to be
a question when it was syntactically marked as such, for
example by showing inversion. Confirmation requests were
indirect questions: they were not syntactically questions, but
the context and interlocutor response indicated that the speaker
was indirectly asking if their interlocutor was paying attention,
agreed or understood. Holding the floor was defined as an
utterance indicating that the speaker did not intend to cede the
floor, in that s/he continued talking with either a minimal or no
pause and was not interrupted by their interlocutor. All other
instances of uptalk were identified as statements. These were
regular declaratives for which no other discourse function was
apparent from context; e.g., such utterances did not elicit
information from the interlocutor. For cases in which the
discourse context was ambiguous, a forced choice was made
by the first author who is a native speaker of the dialect.
In addition, the scaling and alignment of the rise was
annotated using the facilities of Praat [9]. The beginning of the
rise was manually located as the point at which an upward
trend was apparent in which successive F0 values differed by
more than 5 Hz (this was done to exclude microprosodic
variation); see Figure 2 for an illustration. Scaling was
measured in Hz and defined as the difference between the F0
at the beginning of the rise and the highest F0 point at its end.
After the F0 information was extracted, values in Hz were
converted to ERB in order to better compare male and female
voices. The alignment of the F0 rise was defined as the
distance of the point annotated as the start of the rise from the
onset of the last stressed vowel in the utterance. This
measurement was based on the assumption (supported by the
data) that the last content word is typically the one carrying the
nuclear pitch accent.
3. Results
Results presented here are related to the function, scaling and
alignment of rises and to differences in gender. We note that
ethnicity, SEC status and bilingualism did not affect the use of
uptalk; thus they will not be discussed further. All significance
testing was determined using linear mixed-effects models with
Speaker as a random intercept. P-values are given with respect
to model comparisons and are reported with the χ2 statistic,
which compares the (reduced) model without the fixed effect
and the (full) model with the fixed effect.
3.1. Discourse Functions and Distribution of Uptalk
The best-fit model for comparing uptalk against other
utterances in the corpus included discourse function, task type,
gender and an interaction between gender and discourse
function as fixed effects. Uptalk was more frequent in the map
task than in clip retell: 34% of the utterances in the map task
ended in uptalk as opposed to only 20% of utterances in clip
retell [χ2(1) = 37.4, p < 0.001]. Uptalk was also used more
frequently, approximately twice as often, by female than male
speakers: uptalk comprised 42% of the female speakers’
utterances vs. 20% of the male speakers’ utterances [χ2(1) =
14.1, p < 0.001].
Gender also interacted with discourse function [χ2(3) =
16.9, p < 0.001]. First, no gender or discourse differences were
found for uptalk in questions and confirmation requests in the
corpus: uptalk was used for both types of utterances in 100%
of the tokens independently of speaker gender. Statements and
floor holding, on the other hand, showed different frequencies
for uptalk, with floor holding being signaled by uptalk
significantly more frequently than statements: 45% of floor
holding ended in uptalk vs. 16% of statements [χ2(3) = 244.7,
p < 0.001]. However, while females and males used uptalk
with statements equally frequently, females used uptalk to
hold the floor significantly more frequently than males; indeed
females used uptalk more than twice as much as males for
floor holding. This is illustrated in Figure 3.
Figure 2: Example of data annotation from the map task. LSV = last stressed vowel; Us = start of uptalk rise; Ue = end of uptalk
rise; Q = question; S = statement; FH = floor holding. The follower’s response (“yes, I do”) which followed the question in this
example has been removed for clarity.
Figure 3: Proportion of uptalk used by discourse function and
gender.
3.2. Alignment of Uptalk Rise
The best-fit model for the alignment of the uptalk rise included
discourse function and gender as fixed effects. In this model,
only two levels of discourse function were included, statement
and question. Floor holding and confirmation requests were
omitted from the model as their alignment was too variable.
The results from statements and questions showed a
consistent difference between the onset of the rise in
statements vs. questions, with the former having significantly
later alignment than the latter [χ2(1) = 19.3, p < 0.001].
Specifically, the rise in the questions included the last stressed
vowel (which is presumed to carry the nuclear pitch accent)
while in statements the rise started after this vowel. The
difference in the alignment of the rise in statements and
questions is illustrated in Figure 4, which also shows the effect
of gender on alignment. Specifically, uptalk produced by
female speakers showed later alignment than uptalk produced
by male speakers both for statements and questions [χ2(1) =
5.6, p = 0.02]. The differences were quite substantial,
particularly for the questions: male speakers started the rise
just before the last stressed vowel on average, while the rise
for the female speakers started within this vowel.
3.3. Scaling of the Uptalk Rise
The best-fit model for the scaling of the rise included
discourse function [χ2(3) = 19.4, p < 0.001], gender [χ2(1) =
27.01, p < 0.001] and task type [χ2(1) = 20.03, p < 0.001] as
fixed effects. The major difference in pitch excursion with
respect to discourse function was between statements and the
other functions, with statements showing approximately half
the pitch rise than questions, confirmation requests and floor
holding (see Figure 5). Differences between these last three
discourse functions were also statistically significant but
minimal in actual terms [questions, confirmation requests >
floor holding].
Figure 4: Mean rise alignment (with standard error bars) per
type of discourse function and gender. Negative values
represent a rise beginning before the onset of the last stressed
vowel (LSV).
Figure 5: Mean scaling of rises (with standard error bars) per
discourse function.
In addition, the data showed that female speakers had
generally greater pitch excursions than males (see Figure 6a),
presumably a reflection of gender differences in the use of the
LSVUs Ue LSVUs Ue LSV Us Ue LSV Us Ue
an’ then do you see Valley Mall ok so go past Valley Mall go in that direction
Q S FH S
Time (s)
0 4.663
59
28
17 16
0
20
40
60
80
female male
% of utterances in task
floor holding statement
185 126
74
-38
-200
-100
0
100
200
300
female male
ms from LSV onset
statement question
1.37 1.39 1.32 0.75
0
0.4
0.8
1.2
1.6
2
Question Confirmation
Request
Floor Holding Statement
ERB
frequency and effort codes [10]. Further, pitch excursions
associated with uptalk were significantly larger in the map task
than in clip retell (see Figure 6b). Neither result interacted
with discourse function, however, suggesting these are
independent effects and not the result of, e.g., female talkers
asking more questions, or speakers in general making more
confirmation requests in the map task than in clip retell.
Figure 6: On the left, mean scaling of uptalk (with standard
error bars) by gender; on the right, mean scaling of uptalk
(with standard error bars) by type of task.
4. Discussion and Conclusions
Given the above results, we propose that the melody typically
used with questions in SoCal English is L* H-H% and that
used with statements is L* L-H%. The difference in
phonological composition accounts both for the difference in
alignment reported above but also for the difference in the
scaling of the pitch rise: L-H% results in a lower rise than H-
H%. Questions, as noted, can show a rise on the stressed
syllable, a contour that could be interpreted as the reflex of a
bitonal LH accent. However, the auditory impression is that of
a low pitch accent, while the use of either L*H or LH* in
questions is pragmatically doubtful (cf. [11] on the pragmatics
of L*H when followed by a rise). Independently of the
representation adopted for the question tune, the fact remains
that questions and statements are not relying on the same
melody as is often assumed; to put it differently, SoCal
statements with uptalk do not sound like questions.
Our results further show that SoCal English makes a
distinction between uptalked statements and questions even
when the same melody is used (as happens occasionally). In
particular, although the distinction is typically realized as a
choice of tune, as noted above, it can also be signaled by just
differences in the pitch scaling of the final rise (cf. the
question and statements in Figure 2). The difference in pitch
rise scaling is particularly evident when the tune used is H* H-
H%, a variant that was attested but was not as frequent in our
data as the L* accent variants. If such differences in the
scaling of the rise turn out to be used by listeners to interpret
the pragmatic intent of an utterance, this would suggest the
need to incorporate scaling contrasts beyond H vs. L in
phonological representations of intonation.
The two main melodies L* H-H% and L* L-H% are also
used for floor holding and confirmation requests except that
these two functions do not have as consistent a connection
with a specific melody. In the case of confirmation requests
this could be due to their dual role as questions and statements:
speakers are making a statement but simultaneously requesting
that their interlocutor confirm that what is said is understood
or accepted. Thus, speakers use L* H-H%, L* L-H% or H* H-
H% in these instances. Regarding floor holding, one of the
most noticeable features was the use of high plateaux, rather
than rises per se. Plateaux are particularly prevalent when
speakers are listing items or instructions in the map task (cf.
[12] on the intonation of lists). Plateaux are possible
realizations of high tones [13] and thus they can perhaps create
the impression of a rise; however, in our data they were clearly
different from uptalk “proper” both acoustically and
impressionistically and thus best represented phonologically as
L* H-L% where the L% is upstepped.
The patterns described above document the use of tunes
that are different from those described for other varieties of
English that use uptalk. In particular, the prevalence of L* is
not reported for other varieties of English (but see [4] on New
Zealand English). As noted, for example, Australian English
uses mostly a H* accent and it is precisely this use that has
given rise to the term High Rising Terminal. Thus, the present
study underlies the importance of including dialectal variation
in the investigation of intonation and gives support to the
claim that such variation exists even within dialectal areas
often described as uniform, like the USA West [14].
Regarding the demographic factors in our study, we note
that there are consistent differences between genders, with
females using uptalk twice as often as males. This difference is
presumably what has given rise to the stereotype that uptalk is
used by females only; among women uptalk is sufficiently
frequent to be identified as a distinctive characteristic of their
way of speaking. Contrary to the popular stereotype, no gender
differences in the use of uptalk were observed for statements:
approximately 16% of statements ended in uptalk in the
speech of both men and women in our sample. However,
differences are evident in the use of uptalk for floor holding: in
this use, uptalk is twice as frequent in the data from female
speakers (a result that in itself suggests that the similar
frequency of uptalk with statements cannot be attributed to the
fact that our male speakers interacted with female researchers).
At present we do not have a good explanation for this but offer
some suggestions. One possibility is that women wish to hold
the floor longer and use uptalk as a device to indicate this
intent. This explanation however does not quite tally with
existing research suggesting that women do not take longer
turns than men ([15] and references therein). Another
possibility is that women wish to indicate their intent to hold
the floor because they are generally interrupted more often
than men [15]. Again, this is not entirely satisfactory as our
data were based on monologues (clip retell) and a cooperative
task in which the interlocutor was always female. Thus, this
aspect of the data clearly requires further investigation. At the
same time, we do find that gendered use of uptalk did not
interact with task. From this we can infer that the gender effect
is not due, e.g., to women asking more questions, but rather to
their general preference for certain uses of uptalk.
Unlike gender, which was a clear determiner of the
frequency, function and form of uptalk, we did not find
differences relating to ethnicity, SEC status or the language
background of our speakers. Although it is possible that such
differences could emerge with a larger sample, the ubiquitous
use of uptalk in our corpus rather suggests that uptalk is
sufficiently widespread in SoCal to transcend social barriers.
In turn this tallies with the speakers’ attitude to uptalk: for
SoCal speakers it is not a feature that attracts attention.
5. Acknowledgements
We thank our participants, our research assistants Annabelle
Cadang and Andy Hsiu for help with constructing and labeling
the corpus, and the members of the UCSD Phonetics Lab for
valuable feedback on this project.
1.57 0.7
0
0.5
1
1.5
2
Female Male
ERB
1.26 0.89
0
0.5
1
1.5
Map Task Clip Retell
ERB
(a)
(b)
6. References
[1] J. Fletcher, E. Grabe, and P. Warren. “Intonational variation in
four dialects of English: the high rising tune,” in Prosodic
Typology: The Phonology of Intonation and Phrasing, S-A. Jun,
Ed. Oxford: Oxford University Press, 2005, pp. 390-409.
[2] C. Mayo, M. Aylett, and D. R. Ladd. “GlaToBI prosodic
transcription of Glasgow English: An evaluation study of
GlaToBI,” in Proceedings of the ESCA Workshop on Intonation:
Theory, Models and Applications, 1997, pp. 231-234.
[3] N. Daly and P. Warren, “Pitching it differently in New Zealand
English: Speaker sex and intonation patterns,” Journal of
Sociolinguistics, vol. 5, no. 1, pp. 85-96, 2001.
[4] P. Warren, “Patterns of late rising in New Zealand: Intonational
variation or intonational change?” Language Variation and
Change, vol. 17, no. 2, pp. 209-230, 2005.
[5] E. Jarman and A. Cruttenden, “Belfast intonation and the myth
of the fall,” Journal of the International Phonetic Association,
vol. 6, no. 1, pp. 4-12, 1976.
[6] J. Sullivan. “The why of Belfast rises,” in New Perspectives on
Irish English, B. Migge and M. Ní Chiosáin, Eds. John
Benjamins Publishing Company, 2012, pp. 67–84.
[7] N. E. Adler, E. S. Epel, G. Castellazzo, and J. R. Ickovics,
“Relationship of subjective and objective social status with
psychological and physiological functioning: preliminary data in
healthy white women,” Health psychology: official journal of the
Division of Health Psychology, American Psychological
Association, vol. 19, no. 6, pp. 586-592, 2000.
[8] A. Singh-Manoux, M. G. Marmot, and N. E. Adler, “Does
subjective social status predict health and change in health status
better than objective status?” Psychosomatic Medicine, vol. 67,
no. 6, pp. 855-861, 2001.
[9] P. Boersma and D. Weenik. (2013). Praat: doing phonetics by
computer [Computer program], Version 5.3.59. Available:
http://www.praat.org
[10] C. Gussenhoven, The Phonology of Tone and Intonation.
Cambridge University Press, 2004.
[11] J. Hirschberg and G. Ward, “The influence of pitch range,
duration, amplitude and spectral features on the interpretation of
the rise-fall-rise intonation contour in English,” Journal of
Phonetics, vol. 20, pp. 241-251, 1992.
[12] M. Liberman and J. Pierrehumbert. “Intonational invariance
under changes in pitch range and length,” in Language Sound
Structure, M. Aronoff and R. Oehrle, Eds. Cambridge, MA:
MIT Press, 1984, pp. 157-233.
[13] R-A. Knight and F. Nolan, “The effect of pitch span on
intonational plateaux,” Journal of the International Phonetic
Association, vol. 36, no. 1, pp. 21-38, 2006.
[14] W. Labov. “The three dialects of English,” in Handbook of
Dialects and Language Variation, M. D. Linn, Ed. San Diego:
Academic Press, 1998, pp. 39-81.
[15] J. Coates, Men, women and language. Pearson Education, 2004.
