linguistics.byu.edu · Web viewGlottal stops in English appear in a variety of contexts. They...
Transcript of linguistics.byu.edu · Web viewGlottal stops in English appear in a variety of contexts. They...
A production and perception study of /t/ glottalization and oral releases following glottals in the US
David Eddington, Brigham Young University, [email protected] (corresponding author)
Earl K. Brown, Brigham Young University, [email protected]
4041 JFSB
Provo, Utah 84602
(801) 422-7452
Short title (<50 characters):
/t/ glottalization and oral releases
Page 1 of 38
A production and perception study of /t/ glottalization and oral releases following glottals in the US
Abstract:
The articulation of /t/ in American English varies according to linguistic and extralinguistic factors. Concerning social factors, word-final /t/ glottalization is seen more among speakers of African American English (Farrington 2018), younger speakers (Partin-Hernandez 2005, Roberts 2006), and women (Byrd 1994, Eddington and Channer 2010).
This paper examines the production and perception of /t/ in five US states: Indiana, Mississippi, New Mexico, Utah, Vermont. For the production study, participants read a letter containing 24 prenasal word-medial /t/s (e.g., kitten) and 28 prevocalic word-final /t/s (e.g., not ever). For the perception study, 22 speakers recorded a unique sentence, each of which was manipulated acoustically in order to yield both oral and nasal releases of prenasal word-medial /t/ (e.g. button [bʌʔən] vs [bʌʔn̩]), as well as tap and glottal stop pronunciations of prevocalic word-final /t/ (e.g. not ever [nɑɾɛvɚ] vs. [nɑʔɛvɚ]). Next, these recordings were presented to participants who rated the speakers in terms of their perceived age, friendliness, pleasantness, rurality, education level, and whether they were from the same state as the participants.
The production results for prenasal word-medial /t/ (e.g. button) indicate that younger speakers produced oral releases more often than their older counterparts. Age also was related to the realization of prevocalic word-final /t/ as a glottal stop (e.g., not ever), such that younger speakers and women produced glottal stops more often than older speakers. In the perception study, speakers who used glottal stops were viewed as less educated and less friendly. Speakers who used oral releases were perceived as more rustic and less educated.
This paper contributes to the literature documenting the production and perception of /t/ in American English and to the literature that demonstrates the usefulness of using both production and perceptual data to study language variation (e.g., Brown 2015).
Keywords:
glottal stop, tap, oral release, English /t/, American English, experiment, matched guise
Acknowledgements:
ACKNOWLEDGEMENTS GO HERE, IF/ONCE ACCEPTED FOR PUBLICATION
1. Introduction
Glottal stops in English appear in a variety of contexts. They often precede vowel-initial words
such as ‘into’ [ʔɪnthuʊ] and serve to break up hiatus between vowels: ‘spa appointment’
[spaʔəpɔɪntmɨnt]. They also appear as an allophone of /t/, occurring most often in the syllable coda as
Page 2 of 38
in ‘catfish’ [khæʔfɪʃ] (Seyfarth and Garallek 2015). The particular realization of /t/ in word-internal
prevocalic position is a salient characteristic that marks speakers of particular varieties. For example,
North American and some Irish varieties favor taps in this position (e.g., ‘better’ [bɛɾɝ]), while many
Scottish and English varieties prefer glottal stops (e.g., [bɛʔɜ]).
The first documented evidence of glottal pronunciations are found in Britain in the late 19th
century (Andrésen 1968, Collins and Mees 1996), and in the 20th century they have been observed in
many different varieties of English: Received Pronunciation (RP) (Fabricius 2000; Wells 1997),
London (Wells 1982), Scotland (Marshall 2003; Reid 1978), Estuary English (Coggle 1993),
Newcastle (Docherty and Foulkes 1999), Ipswich (Straw and Patrick 2007), Cardiff (Mees 1987), U.S.
(Byrd 1994), New York City (Wells 1982; Levon 2006), Vermont (Roberts 2006), Appalachia (Wells
1982), and California (Partin-Hernandez 2005).
Word-final /t/ is of particular interest because it may be followed by either a consonant, where a
glottal realization is more common, or by a vowel, in which case the tap [ɾ] often occurs. Word-
finally, /t/ occurs in an alternating context where a great deal of variation is expected, and where the
variation in pronunciation is often associated with social factors (Straw and Patrick 2007). For
example, word-final t-glottalization is a characteristic of African American English (Farrington 2019).
In the U.K., younger speakers appear to glottalize more (Klánová 2016, Macaulay 1977, Marshall
2003, Tollfree 1999), a trend also observed in New Zealand (Holmes 1995), and the U.S. (Eddington
and Taylor 2009, Eddington and Channer 2010, Partin-Hernandez 2005, Roberts 2006). Gender is also
a factor, and many studies cite women as more prone to glottalization in the U.S. (Byrd 1994,
Eddington and Taylor, 2009, Eddington and Channer 2010), in New Zealand (Holmes 1995), and in the
U.K. (Klánová 2016, Milroy, Milroy, Hartley, and Walshaw 1994).
Nevertheless, there are some exceptions to these trends. In RP, men are more likely to glottalize
(Barrera 2015). In one study of the Reform American Jewish community, male speakers were also
Page 3 of 38
observed to glottalize more than women (Levon 2006). In like manner, men in Vermont used more
glottal stops than women (Roberts 2006). However, in another study (Bellavance 2017) consisting of
Vermont children at the fourth grade through high school levels, age and gender were not significant
predictors of glottalization. In Ohio, Kazmierski (2018) observed an interaction between age and
gender to the effect that younger women used more glottal stops than both older women, and men in
both age groups.
A number of linguistic variables have been considered in addition to the social variables. In
Barrera’s study of RP (2015), lexical frequency, number of syllables, and preceding vowel were
significant factors. The least clear factor is the role of the vowel following /t/. Ostalski (2013) found
that back vowels favor glottal stops, while Eddington and Taylor (2009) reported that front vowels
favored glottal realizations of /t/. Kazmierski (2018), in contrast, observed no significant influence of
vowel backness. More agreement exists in the literature on the role of the stress on the syllable of the
following vowel, where a following stressed vowel appears to favor glottalization (Eddington and
Channer 2010, Kazmierski 2018).
In American English, glottal stops tend to occur before consonants, while taps generally
precede vowels. If this is the case, why would word-final /t/ be realized as a glottal stop before a word-
initial vowel (e.g. abou[ʔ] any)? Eddington and Channer (2010) searched the spoken section of the
Corpus of Contemporary American English (Davies 2008) for all bigrams in which the first word
ended in /t/ and observed that such words are more than twice as likely to be followed by a consonant
than a vowel. According to exemplar theory, this means that there would be more glottalization, hence
more instances stored in memory with glottal stops than with other pronunciations. This would result in
a frequency effect (Bybee 2017) that gives some advantage to the glottal realization even in cases when
it appears before a vowel. Kazmierski (2018) tested this hypothesis more rigorously using the Buckeye
Corpus (Pitt, Johnson, Hume, Kiesling, and Raymond 2005) by calculating the ratio of consonant-
Page 4 of 38
initial to vowel-initial words following each word ending in /t/. He found this ratio to be a significant
predictor of t-glottalization in the corpus.
While the realization of /t/ as a glottal stop has received attention in the literature, less has been
said about how glottal stops are released before nasals in words such as mitten and beaten. The most
general pronunciation releases the glottal nasally ([mɪʔn̩]), while oral releases (e.g. [mɪʔən]) have been
documented in Utah (Eddington and Savage 2012), occurring mainly among younger females. Oral
releases have also been observed anecdotally in Indiana1 and New Mexico2.
Clearly the geographic, social, and linguistic factors that govern the glottalization of word-
final /t/ in prevocalic position in American English, as well as those that influence how glottal stops are
released before nasals, merit further investigation. However, extant studies of t-glottalization have
focused principally on variables related to pronunciation. More specifically, little is known about the
geographical distribution of these pronunciations in the US. Anecdotal evidence suggests that there is
some social stigma attached to them, but this has not been explored in the extant literature on the
subject. Therefore, we attempt to investigate how glottal stops and their releases are perceived in terms
of positive and negative social traits. The present study is therefore designed to examine both the
production and social perception sides of the topic by examining the speech and perceptions of
speakers from five U.S. states: Indiana, Mississippi, New Mexico, Utah, and Vermont.
1.1. Research questions
The previous studies on this topic motivate the following research questions:
RQ1: Which linguistic and social variables are correlated with the production of an oral release after
glottalized /t/ in prenasal word-medial position (e.g., kitten)?
1 https://www.waywordradio.org/american-glottalization/2 https://www.youtube.com/watch?v=N5Yy0iWVC00
Page 5 of 38
RQ2: Which linguistic and social variables are related to the production of a glottal stop for /t/ in
prevocalic word-final position (e.g., not ever)?
RQ3: What social traits are perceived to be associated with oral releases of prenasal word-medial /t/
(e.g., button)?
RQ4: What social traits are perceived to be associated with glottal realizations of prevocalic word-
final /t/ (e.g., fat or)?
2. Production Study
2.1. Test items
Participants recorded themselves reading the letter in Appendix A twice. The letter includes 24
words containing a /t/ followed by nasal (PreN, e.g. kitten, written), and 28 words ending in /t/ which
are followed by a vowel-initial word (PreV, e.g. about a, fat or).
2.2. Method
The majority of participants were recruited on Amazon Mechanical Turk and were paid USD $2
for providing a recording. Separate announcements were posted to solicit participants from Indiana,
Mississippi, New Mexico, Utah, and Vermont. A smaller number were recruited using social media and
through third person contacts.3 The participants were directed to a survey placed on the Qualtrics
survey platform in which their demographic information was gathered, and they were given
instructions on how to make an audio recording. Separate recording instructions were given for iOS
(iPhones), and Android OS (Android phones), and well as how to install and make a recording on a
computer using the sound editing software Audacity (Audacity Team 2018). The instructions directed
them to read the letter that appeared in the survey twice in a natural voice, in a quiet location. Once
3 Z-scores were used to avoid a warning message about the model failing to converge.
Page 6 of 38
they completed the recording, they uploaded it on wetransfer.com. This step was included so that they
could send the recording without revealing their email address and thus maintain their anonymity.4
2.3. Participants
Recordings from 94 participants were used. Others were eliminated for a number of reasons
including providing a poor quality recording, not following instructions correctly, or not having lived at
least 60%5 of their life in either Indiana, Mississippi, New Mexico, or Utah. Although efforts were
made to obtain recordings of speakers from Vermont, only five people from that state responded and
provided a recording—this was too few to warrant their inclusion in the production study. The final
result was 22 recordings from Indiana, 27 from Mississippi, 23 from New Mexico, and 22 from Utah.
There were 35 male participants and 59 females, and the mean age of the respondents was 32.2 years
(SD = 10.7).
As far as education is concerned, 11 participants had high school diplomas, 40 had attended
college without graduating, 29 had undergraduate degrees, and 14 had graduate degrees. The majority
of the participants (63) self-identified their race as White, while the remainder consisted of 12 African
Americans, 9 Hispanics, 8 people of multiethnic origin, and 2 belonging to other ethnicities.
2.4. Results of the tests for prevocalic word-final /t/
The authors each examined approximately half of the tokens spectrographically and auditorily.
Cases that were difficult to categorize were examined by both authors. Glottal stops were identified by
irregularly spaced striations on the spectrograms (see Figure 1).6 Taps, in contrast, appeared as brief
voiced occlusions (see Figure 2). Alveolar stops were those that had voiceless occlusions with a
following release burst and aspiration (see Figure 3). Instances which were not clear enough to be
4 We thank Joey Stanley for this idea.5 We did not attempt to distinguish glottal reinforcement of the stop from pure glottal stops.6 This 60% cut-off is an arbitrary threshold that we adopted in order to identify people who had lived at least more than half of their life in their state.
Page 7 of 38
unambiguously categorized were not included in the analysis (N=478). The remaining 4,858 responses
yielded 59.1% taps, 39% glottal stops, and 1.9% alveolar stops.
Figure 1: Prevocalic word-final glottal stop (speaker A2Q3, 12s).
Figure 2: Prevocalic word-final tap (speaker A3N7, 52s).
Figure 3: Prevocalic word-final stop (speaker AW3Z, 253s).
Page 8 of 38
The data were analyzed using the R programming language (R Core Team 2018). In it, a mixed
effects logistic regression was carried out in which participants and test words were included as random
intercepts and the BOBYQA optimizer7 (Powell 2009) was used (Table 1). The dependent variable was
whether the pronunciation was glottal versus other pronunciations combined (i.e. taps and alveolar
stops). The independent variables for the analysis were both social and linguistic. The social variables
included participant gender, age, and state of origin, as well as whether the participant was a college
graduate or not. Race consisted of four categories: Black, Hispanic, White, Mixed/Other. Linguistic
variables included the phones preceding and following the word-final /t/, as well as the stress on the
vowel following /t/.
As far as the statistical analyses are concerned, the idea that the alpha level of 0.05 is special or
ideal has been criticized (Amrhein, Greenland and McShane 2019; Amrhein, Trafimow, and Greenland
2019, Cohen 1994). Kline (2004) actually suggests that an alpha level of 0.1 would be more
appropriate in the behavioral sciences. In either case, this suggests that an alpha level should not be
used to blindly dichotomize factors into those that do and those that do not affect the results. Instead, it
should be taken as a point at which the data are most compatible, with less compatible, but still
influential, data falling farther away (i.e., higher alpha levels). Therefore, we choose to discuss results
that approach, but do not reach an alpha of 0.05 as important trends.
7 Bound Optimization BY Quadratic Approximation
Page 9 of 38
Table 1. Analysis of deviance of the fixed effects of the glottal stop pronunciation of word-final
prevocalic /t/.
χ2 df p
(Intercept) 5.78 1 0.016
Following vowel 117.00 6 0.000
Age 7.12 1 0.008
Gender 7.03 1 0.008
College Graduate 1.85 1 0.174
State 5.54 3 0.136
Preceding phone 7.57 9 0.578
Stress on following vowel 0.99 1 0.319
In order to determine how following vowels influenced the use of glottal stops, a Bonferroni
post hoc analysis was carried out that compared how each vowel differed from the average of all
vowels in terms of their favoring a glottal pronunciation. It revealed that the vowel [o] favored glottal
stops to a significantly greater extent, while the vowel [ə] disfavored them. However, given the limited
variety of words in the present study, these results may be due to the particular words used in the study
rather than to specific vowels. For instance, all cases of [o] occur in the word or, found in only two
bigrams: lit or and fat or. The vowel [ə] is the only vowel that appears in a variety of words, which
may suggest that schwa itself, rather that the particular word it appears in, statistically disfavors a
glottal realization of a preceding /t/. Perhaps prosodic features also play a part in this alternation, and
future studies might investigate this possibility.
Page 10 of 38
Several social variables proved significant as well. Women, for example, used glottal stops in
42.6% of the cases and men in only 33%. As far as age is concerned, the general trend is that younger
people used more glottal stops (see Figure 4: the solid line is a regression line, with 95% confidence
intervals around it, and the dotted curved line is a loess line). Conspicuously absent is an influence of
the state the participant calls home. The fact that there are no differences in glottalization rates across
these four states gives credence to the idea that the use of the glottal stop may be widely dispersed
across the U.S., rather than more common in western states, as Eddington and Channer (2010) suggest.
However, the present study confirms previous observations that glottalization in the U.S. is more
prevalent among younger speakers and females, with the exception of male speakers from Vermont (cf.
Roberts 2006).
Figure 4. Percent glottal stop of prevocalic word-final /t/ by age.
Page 11 of 38
2.5. Results of the tests for prenasal word-medial /t/
Prenasal word-medial /t/ was also analyzed with a mixed effects logistic regression as in the
previous analysis in which participants and test words were included as random intercepts (Table 2).
After removing responses which were not clear enough to be categorized (N=240), there were 4,272
responses of which 82% were glottal stops followed by nasal releases (e.g. button [bʌʔn̩], see Figure 5),
11.3% were glottal stops followed by oral releases (e.g. [bʌʔən], see Figure 6), 3.2% were nasally
released [d] (e.g. [bʌdn̩]), 2.9% were [t], and the remaining 0.5% were taps (e.g. [bʌɾən]). The
regression analysis contrasted the use of glottal stops followed by oral releases from all other
pronunciations. The exact quality of the vowel produced between the glottal stops and nasals varied
Page 12 of 38
between speakers and was not deemed relevant for the purposes of the present study. However,
impressionistically it appeared to be a lax vowel that varied between [ɪ], [ə], and [ɨ].
Figure 5: Prenasal word-medial glottal stop followed by nasal release (speaker A3N7, 4s).
Figure 6: Prenasal word-medial glottal stop followed by oral release (speaker QUEZ, 205s).
Table 2. Analysis of deviance of the fixed effects of the oral release pronunciation of word-medial /t/ followed by nasal.
χ2 df p
(Intercept) 2.70 1 0.101
Age8 5.99 1 0.014
Page 13 of 38
Preceding vowel 38.53 9 0.000
Race 4.84 3 0.184
Gender 2.04 1 0.153
State 2.79 3 0.426
College Graduate 0.40 1 0.526
As is apparent in Figure 7, oral releases, such as in written [ɹɪʔən], are more common among
younger speakers. The preceding vowel is also significant, with /ɪ/ and /ə/ favoring an oral release
and /æ/ and /o/ disfavoring it. As far as the influence of state is concerned, we specifically chose Utah
as a test state because oral releases have been reported there (Eddington and Savage 2012). They have
also been observed anecdotally in New Mexico and Indiana (see footnotes 1 and 2 above). Mississippi,
on the other hand was included as a kind of control state, as we had no reason to believe that oral
releases are commonly produced there. For this reason, it is an interesting finding that there are no
significant differences between the four states that are compared. However, there is a definite trend in
that direction in terms of the percentage of oral releases (NM 24%, UT 12%, IN 9%, MS 2%).
Figure 7. Percent of glottal stop followed by an oral release of prenasal word-medial /t/ according to
age.
8 We thank Felice Coles, Dawn Nordquist, Mark Waltermire, Karol Ibarra Zetter, and Debbie Wager for their help with recruiting participants.
Page 14 of 38
3. Perception Study
The matched-guise method (cf. Lambert, Hodgson, Gardner & Fillenbaum 1960) in
experimental sociolinguistic research involves the same person reading or speaking in two different
parts of the experiment in such a way that the listener does not realize that the same person is speaking.
The speaker modifies his voice or language, or regional accent, to achieve this deceptive effect. The
purpose of the matched-guise experimental method is to elicit attitudinal data from listeners, with the
intent to attach specific language attitudes to specific pronunciations. A variant of the matched-guise
method is to digitally manipulate the pronunciation of a speaker in a recording with acoustic software
in order to control the pronunciation with precision. For example, Campbell-Kibler (2010) manipulated
word-final -ing (e.g., going) in English, while Podesva, Reynolds, Callier, and Baptiste (2015) changed
the release of /t/ in the speech of six prominent U.S. politicians, and Chappell (2016) altered the
voicing of intervocalic /s/ in Costa Rican Spanish. The authors then gathered attitudinal data from
listeners and correlated social characteristics with specific sounds.
Page 15 of 38
3.1. Test items
The present study consisted of six test sentences with words containing a final /t/ followed by a
vowel (PreV, e.g. met on, wait until, get another) and six sentences with words containing a /t/
followed by a nasal (PreN, e.g. curtain, tighten). A complete list of test sentences appears in Appendix
B. Another six sentences that did not include the variables in question were included as distractors,
along with four lie detector sentences, which are discussed more in the next section. Each sentence was
recorded by a different speaker, all of whom were white females in their 20s from the western U.S.,
who were university students.
The speakers rendered the /t/ in the relevant words in the PreV sentences with a tap
pronunciation (e.g. me[ɾ] on, wai[ɾ] until), while the /t/ plus nasal sequences in the words in the PreN
sentences were pronounced as glottal stops followed by a schwa (e.g. tighten [taɪʔən]). For most of the
volunteer speakers, achieving an orally released glottal stop word-medially required some training and
practice, as this pronunciation was not natural for them. In contrast, articulating a tap for prevocalic
word-final /t/ was more natural for them.
Using Praat we then created another twelve test sentences by manipulating these original
sentences. Specifically, taps in the PreV sentences (e.g., not ever) were converted into synthesized
glottal stops, primarily by lowering the fundamental frequency of the tap segment of the word
(Pierrehumbert and Frisch 1997). However, in order to achieve natural sounding glottal stops, we also
lengthened the tap to varying degrees and added silence between glottal pulses to render their spacing
more irregular. This was done on a case by case basis until a natural sounding glottal stop was
achieved. Additionally, the PreN words were converted from oral releases (e.g. tighten [taɪʔən]) to
nasal releases (e.g. [taɪʔn̩]) by simply removing the vowel following the glottal stop.
Page 16 of 38
3.2. Participants
A total of 229 people participated in the perception study, 99 males and 130 females. Of these,
70 were from Indiana, 46 from Mississippi, 26 from New Mexico, 66 from Utah, and 21 from
Vermont. Only two respondents had less than a high school degree, while 31 had completed high
school, 83 had attended some college, 98 were college graduates, and the remaining 15 had graduate
degrees. The great majority of the participants (185) were White, while 13 were Black, 4 were Asian, 6
were Hispanic, and 1 Native American. The remaining 20 were of mixed ethnicity.
3.3. Method
Participants from the five states of interest were recruited on Mechanical Turk and paid USD $2
for completing the survey. After answering a few sociodemographic questions, the participants were
given these instructions:
You will hear a sentence then be asked to judge how the person sounds. Please listen to the
entire audio BEFORE answering any of the questions. If you don't follow the instructions the
survey will automatically stop. That means that you won't be able to try it a second time, and
you won't be paid for completing it.
In surveys of this nature it is possible for the participants to answer without actually listening to the
recording. In order to weed out inattentive participants four lie detector recordings were included. In
those recordings a test sentence was followed by explicit instructions to leave all of the answers related
to that recording blank. In order to emphasize the importance of following the instructions, the first test
item was a lie detector. When a participant answered any questions about this first sentence they were
given this message:
Page 17 of 38
You didn't listen to the whole clip and follow the instructions!! If you don't follow the
instructions the survey will automatically stop. That means that you can't try it a second time or
be paid for completing it. Please continue to the next audio clip.
This served as a warning. The remaining three lie detector items were randomly interspersed among the
other test items and distractors. When a participant answered any question about the other three lie
detector sentences the survey would cease and the responses of these participants were eliminated from
the study. Each recording was preceded with these instructions:
Listen to this woman's voice then answer the following 6 questions about how she sounds.
Please listen to the entire recording before answering any of the questions.
The participants were then able to start the recording and answer a series of six questions about the
speaker in the recording (cf. Appendix C). The first question asked the listeners to guess the speaker’s
age on a scale between 12 to 40 years old. Subsequent questions were answered on a Likert-type scale
and solicited the degree to which the speaker in the recording sounded like she was from a town or city,
how pleasant, friendly, and educated she sounded, as well as how much she sounded like she was from
the state the participant was from. For example, for pleasantness the scale was: Very pleasant,
Somewhat pleasant, Neither pleasant nor unpleasant, Somewhat unpleasant, Very unpleasant.
Participants were randomly assigned to one of two groups. The members of each group heard
the same four lie detector recordings, and the same six distractor recordings, however, the groups
differed in terms of what pronunciations they heard in the test sentences (Table 3). This division meant
that no listener heard the same sentence or speaker twice, and allowed a comparison of the resulting
evaluations in terms of the phones that varied. Aside from the first item played to the listeners, which
was the same for all listeners, as mentioned above, the items were played in a random order.
Table 3. Assignment of test recordings to the participants.
Page 18 of 38
Test Group 1 Test Group 2
Lie detector sentences 1-4 Lie detector sentences 1-4
Distractor sentences 1-6 Distractor sentences 1-6
PreV sentences 1-3 with taps PreV sentences 1-3 with glottal stops
PreV sentences 4-6 with glottal stops PreV sentences 4-6 with taps
PreN sentences 1-3 with nasal releases PreN sentences 1-3 with oral releases
PreN sentences 4-6 with oral releases PreN sentences 4-6 with nasal releases
3.4. Results
Since the purpose of the present study was to determine how pronunciation differences affected
the participants’ perceptions of the speakers, the only independent variable that was relevant was the
particular pronunciation and the interaction of pronunciation with the other biographical variables, state
and gender. However, in order to save space, we do not discuss the findings that are not related to
pronunciation. For instance, we observed that female participants rated the speakers as more friendly
than did the male participants, but that result is not germane to the research questions in the present
study. In any event, our analyses investigated how the above mentioned variables are related to
participants’ perceptions of the speaker’s age, friendliness, pleasantness, education level, how much
they sounded urban vs. rural (rurality), and how much they sounded like they were from the state the
participant was from. In the analyses, random intercepts were included for participant and for the
speaker who made the recording. These analyses were carried out using the Jamovi statistical package
(Jamovi Project 2018) which is a GUI for R. It simplifies running the analyses since it does not require
line commands, and has a number of dependencies, including the Gamlj R package, the lmerTest R
package (Kuznetsova, Brockhoff and Christensen 2017) and the lme4 R package (Bates, Mächler,
Bolker, Walker 2015).
Page 19 of 38
Perhaps the most unexpected result is that there were no significant interactions between
pronunciation and state. In fact, no significant interactions between any other variable and
pronunciation occurred at or below the p = 0.05 alpha level. However, we examine a number of
predictors that were not significant at an alpha level of 0.05. In such cases it is common to state that
there was a trend toward significance. The first trend (p = 0.058) is that when rating the speaker’s
friendliness with prevocalic word-final /t/ (e.g., not ever), men perceived speakers who used glottal
stops as less friendly than those who used taps, judging by the means of their responses (glottal stop =
0.57, tap = 0.73), while women’s ratings differed little (glottal stop = 0.79, tap = 0.82). Another
noteworthy trend (p = .083) was seen in perceptions of whether the speakers who were heard in the
recording were from the same state as the participants when the speakers were heard using either oral
or nasal releases with prenasal word-medial /t/ (e.g., kitten). Keep in mind that higher numbers in Table
4 indicate greater perceptions that the speaker is from the same state as the participant. The means in
Table 4 suggests that Mississippians perceive oral releases as more ‘foreign’ sounding, while in
Vermont, Utah, and New Mexico oral releases were viewed as more likely spoken by natives of those
three states. In spite of the anecdotal reports of oral releases in Indiana, no trend was apparent in the
data from that state.
Table 4. Means for ratings of ‘in-stateness’ according to pronunciation of prenasal word-medial /t/.
Pronunciation State Means
Nasal Release IN .21
Oral Release IN .19
Nasal Release MS -.10
Oral Release MS -.31
Nasal Release NM -.04
Page 20 of 38
Oral Release NM .04
Nasal Release UT .66
Oral Release UT .74
Nasal Release VT -.18
Oral Release VT .19
Judgments of the speaker’s rurality were also influenced by pronunciation (p = .073) to the
effect that glottal stops (-.16) were associated with town dwellers more than taps and other
pronunciations (-.04). Friendliness was likewise affected to a small degree by the pronunciation the
participants heard (p = .091); nasal releases (.98) sounded friendlier than oral releases (.90).
The statistically significant findings relating to pronunciation are summarized in Table 5. The
full statistical results appear in Appendix D. This table contains the results of both the prevocalic (e.g.,
not ever) and the prenasal (e.g., kitten) context in which /t/ occurs. It shows that recordings of speakers
containing oral releases with prenasal, word-medial /t/ (e.g., kitten [kɪʔVn]) were perceived to be less
educated and more rural than when a nasal release [kɪʔn̩]) was heard. The use of glottal stops with
prevocalic, word-final /t/ (e.g., not ever [nɑʔɛvɚ]) was associated with a lower degree of friendliness
and education. Glottal stops were also significantly associated with a slight decrease in the speaker’s
perceived age.
Page 21 of 38
Table 5. Significant variables that involve pronunciation according to the influencing independent
variables and means for each pronunciation.
Independent Variables Mean Phonological context Dependent Variable Values
Rurality -0.27 PreN Oral release
-0.04 PreN Nasal release
Education 0.61 PreN Oral release
0.78 PreN Nasal release
Age 22.3 PreV Glottal Stop
22.6 PreV Non-glottal
Friendliness 0.70 PreV Glottal Stop
0.78 PreV Non-glottal
Education 0.49 PreV Glottal Stop
0.58 PreV Non-glottal
4. Discussion and Conclusions
With respect to RQ1, which asked where prenasal word-medial /t/ is realized with an oral
release (e.g., button [bʌʔən]), we can conclude that it occurs in all four states that we chose to include
in the study. As already mentioned earlier, few participants from Vermont provided a recording for the
production study which is why Vermont was not included. Despite the oral release being present in
Indiana, Mississippi, New Mexico, and Utah, the form used the most was a nasal release (e.g., [bʌʔn̩]).
With respect to which state oral releases were used the most in, New Mexico had the highest rate
(24%) followed by Utah (UT 12%), Indiana (IN 9%), and Mississippi (2%). Despite these differences
in rates of usage of oral releases, state as an independent variable was not found to be a significant
Page 22 of 38
predictor of oral releases when the response variable was specified as the binary distinction between an
oral release and all other articulations (e.g., a nasal release [bʌʔn̩], the alveolar stop [t] [bʌtn̩], a tap
[bʌɾVn̩], and a nasally released [d] [bʌdn̩]).
One variable that shows a significant effect in the regression analysis was the preceding vowel,
and another was age, with younger speakers producing more oral releases than older speakers, despite
the fact that even for younger speakers, nasal releases were the most common pronunciation. This
finding concurs with previous studies in the U.S. (Eddington and Taylor 2009, Eddington and Channer
2010). Whether this trend represents a change in progress in apparent time (cf. Bailey, Wikle, Tillery
and Sand 1991; Cukor-Avila and Bailey 2013) or stable variation is an interesting question, but falls
outside the scope of the current paper and warrants further investigation.
Concerning RQ2, which asked where prevocalic, word-final /t/ is realized as a glottal stop, the
results of the production study show that it is produced in the four states examined, with rates of glottal
stop realization ranging from 35.5% in Mississippi to 45.9% in New Mexico, with a mean of 39.1%,
and a standard deviation of 4.8. As with prenasal word-medial /t/, state did not make a significant
contribution to the prediction of prevocalic word-final /t/, despite differences in rates of usage of glottal
stops versus other realizations of /t/. Although differences of production rates between states were
minimal, gender and age proved important; women and younger speakers produced more glottal stops.
Additionally, the following vowels were found to significantly condition glottal stops: /o/ favored a
preceding glottal stop, and /ə/ disfavored a preceding glottal stop. As pointed out above, with only two
unique two-word strings with a following /o/ (i.e., lit or, fat or) this effect from the following vowel
may be due to the specific word rather than being due to the vowel itself. We trust that future studies
will include a wide variety of words so that the effect of following vowel may be more precisely
determined.
With regard to RQ3, which asked which social qualities are perceived to be associated with an
oral release of prenasal word-medial /t/, the results of the perception study revealed several that were
Page 23 of 38
connected with an oral release. The participants perceived speakers who used an oral release in this
context to be less educated and more likely to be from a small town rather than from a city.
Furthermore, they perceived the speakers who used an oral release to be less friendly and slightly
younger. The latter finding concurs with anecdotal observations that younger speakers in Indiana
produce oral releases more than older speakers (see footnote 1).
Finally, with respect to RQ4, which asked about social qualities that are perceived to be
associated with a glottal realization of prevocalic word-final /t/, the question can be answered in the
following manner: the presence of a glottal stop instead of a tap or other realizations influenced the
listener’s perceptions about the speakers. When speakers produced a prevocalic word-final /t/ as a
glottal stop (e.g., not ever [nɑʔɛvɚ]) rather than a tap (e.g., [nɑɾɛvɚ]) or other realization, listeners
perceived them to be less educated, less friendly, and younger.
A connection between production and perception is apparent with prevocalic word-final /t/ and
age of speaker. As reported above, younger speakers produce more glottal stops than older speakers,
and listeners associate a glottal realization of /t/ in this context with slightly younger speakers. This
may indicate that listeners subconsciously associate this pronunciation with younger speakers. In
contrast, there does not seem to be a connection in these data between the production and perception of
prenasal word-medial /t/ and age of speaker. While the younger speakers produced more oral releases
in this context than older speakers, the listeners did not associate a difference in the perceived age of
speakers who produced oral releases. It may be that in the future an oral release of prenasal word-
medial /t/ will eventually become aware of the use of this articulation and associate it with particular
social characteristics.
Concerning the connection between the perceived level of education of speakers and the
pronunciation of /t/ in the two phonological environments analyzed in this study, listeners perceived
speakers to be less educated both when they articulated an oral release of prenasal word-medial /t/ and
a glottal stop for prevocalic word-final /t/. This perception may be due to the effect of the association of
Page 24 of 38
age with these phenomena, in that younger speakers in general have received less education than older
speakers.
The present paper contributes to the field by documenting certain pronunciations of /t/ in
American English as well as contributing to the literature documenting the connection between sub-
phonemic sounds and social evaluations, or sociophonetic variation (cf. Thomas 2011). The
participants in this study, who came from five U.S. states, perceived different social traits about
speakers based on phonetic differences in /t/ as it appeared in two phonological contexts. This finding
is in line with other studies of sub-phonemic variability and social evaluation cited above (e.g.,
Campbell-Kibler 2010; Podesva, et al. 2015, Chappell 2016).
In summary, this paper shows the utility of employing both production and perceptual data
collected online to study regional linguistic variation (cf. Brown 2015) as well as the social
implications and perceptions of that variability. Further, this paper contributes to the field detailing the
articulation of /t/ in American English.
Page 25 of 38
References
Amrhein, Valentin, Sander Greenland, and Black McShane. 2019. Retire statistical significance.
Nature 567. 305-307.
Amrhein, Valentin, David Trafimow, and Sander Greenland. 2019. Inferential statistics as descriptive
statistics: There is no replication crisis if we don’t expect replication. The American Statistician
73. 262-270.
Andrésen, Bjørn Stålhane. 1968. Pre-Glottalisation in English Standard Pronunciation. Olso:
Norwegian Universities Press.
Audacity Team. 2018. Audacity: Free Audio Editor and Recorder [Computer application]. Version
2.3.0. Retrieved from https://audacityteam.org.
Bailey, Guy, Tom Wikle, Jan Tillery and Lori Sand. 1991. The apparent time construct. Language
Variation and Change 3(3). 241–264.
Barrera, Berta Badia. 2015. “A Sociolinguistic Study of T-Glottalling in Young RP: Accent, Class and Education.” University of Essex.
Bates, Douglas, Martin Mächler, Ben Bolker and Steve Walker. 2015. Fitting Linear Mixed-Effects
Models Using lme4. Journal of Statistical Software 67(1). 1–48.
Bellavance, Sarah. 2017. “Co-Occurrence of /t/ Variants in Young Vermont Speakers.” Burlington,
VT: University of Vermont.
Boersma, Paul and David Weenink. 2019. Praat: doing phonetics by computer [Computer program].
www.praat.org
Brown, Earl K. 2015. On the utility of combining production data and perceptual data to investigate
regional linguistic variation: The case of Spanish experiential gustar “to like, to please” on
Twitter and in an online survey. Journal of Linguistic Geography 3(2). 47–59.
Byrd, Dani. 1993. “54,000 American Stops.” UCLA Working Papers in Phonetics 83: 97–116.
Page 26 of 38
Campbell-Kibler, Kathryn. 2010. The sociolinguistic variant as a carrier of social meaning. Language
Variation and Change 22(3). 423–441. doi:10.1017/S0954394510000177.
Chappell, Whitney. 2016. On the social perception of intervocalic /s/ voicing in Costa Rican Spanish.
Language Variation and Change 28(3). 357–378. doi:10.1017/S0954394516000107.
Coggle, Paul. 1993. Do You Speak Estuary? London: Bloomsbury.
Cohen, Jacob. 1994. The earth is round (p < .05). American Psychologist 49, 997-1003.
Collins, Beverley, and Inger M. Mees. 1996. “Spreading Everywhere?: How Recent a Phenomenon Is
Glottalisation in Received Pronunciation?” English World-Wide 17: 175–87.
Cukor-Avila, Patricia and Guy Bailey. 2013. Real Time and Apparent Time. In J.K. Chambers and
Natalie Schilling (eds.), The Handbook of Language Variation and Change, 239–262. 2nd edn.
Malden, MA: Wiley-Blackwell.
Docherty, Gerry, and Paul Foulkes. 1999. “Sociophonetic Variation in ‘Glottals’ in Newscastle
English.” In Proceedings of the 14th International Congress of Phonetic Sciences, 1037-1040.
Berkeley: University of California.
Eddington, David, and Caitlin Channer. 2010. “American English Has Goʔ a Loʔ of Glottal Stops:
Social Diffusion and Linguistic Motivation.” American Speech 85 (3): 338–51.
Eddington, David, and Matthew Savage. 2012. “Where Are the Moun[ɁƏ]Ns in Utah?” American
Speech 87 (3): 336–49.
Eddington, David, and Michael Taylor. 2009. “T-Glottalization in American English.” American
Speech 84 (3): 298–314.
Fabricius, Anne H. 2000. “T-Glottaling Between Stigma and Prestige: A Sociolinguistic Study of
Modern RP.” Ph.D Dissertation, Copenhagen Business School.
Farrington, Charlie. 2019. “Incomplete Neutralization in African American English: The Case of Final
Consonant Voicing.” University of Oregon.
Page 27 of 38
Holmes, Janet. 1995. “Two for /t/: Flapping and Glottal Stops in New Zealand English.” Te Reo 38:
53–72.
Jamovi project. 2018. Jamovi (Version 0.9) [Computer Software]. www.jamovi.org
Kazmierski, Kamil. 2018. “Word Boundary Intervocalic T-Glottaling in Midland American English.”
Adam Mickiewicz University.
Kim, Chaeyoon, Sravana Reddy, James Stanford, Ezra Wyschogrod, and Jack Grieve. 2018. “Bring on
the Crowd! Using Online Audio Crowdsourcing for Large-Scale New England Dialectology
and Acoustic Sociophonetics.” American Speech, 1–38.
Klánová, Aneta. 2016. “Sociophonetic Study of Substitutional Glottalization in Native English
Speakers.” Univerzita Karlova v Praze.
Kline, R. B. 2004. Beyond significance testing: Reforming data analysis methods in behavioral research. Washington D.C.: American Psychological Association.
Kuznetsova, Alexandra, Per B. Brockhoff and Rune H. B. Christensen. 2017. lmerTest Package: Tests in Linear Mixed Effects Models. Journal of Statistical Software 82(13). 1–26.
Lambert, Wallace E., Richard. C. Hodgson, Richard C. Gardner and Samuel Fillenbaum. 1960. Evaluational reactions to spoken languages. The Journal of Abnormal and Social Psychology 60(1). 44–51.
Levon, Erez. 2006. “Mosaic Identity and Style: Phonological Variation among Reform American
Jews.” Journal of Sociolinguistics 10: 181–204.
Macaulay, Ronald K.S. 1977. Language, Social Class, and Education: A Glasgow Study. Edinburgh:
The University Press.
Marshall, Jonathan. 2003. “The Changing Sociolinguistic Status of the Glottal Stop in Northeast
Scottish English.” English World-Wide 24: 89–108.
Mees, Inger. 1987. “Glottal Stop as a Prestigious Feature in Cardiff English.” English World-Wide 8:
25–39.
Milroy, James, and Lesley Milroy. 1978. “Belfast: Change and Variation in an Urban Vernacular.” In
Sociolinguistic Patterns in British, edited by Peter Trudgill, 19–36. Baltimore: University Park
Press.
Page 28 of 38
Milroy, James, Lesley Milroy, Sue Hartley, and David Walshaw. 1994. “Glottal Stops and Tynside
Glottalization: Competing Patterns of Variation and Change in British English.” Language
Variation and Change 6: 327–357.
Ostalski, Przemyslaw. 2013. “Glottal Stops in General American (Intervocalic Environments).” In
Teaching and Researching English Accents in Native and Non-Native Speakers, edited by E.
Waniek-Klimczak and L.R. Shockey, 241–51. Second Language Learning and Teaching.
Berlin: Springer-Verlag.
Partin-Hernandez, Allyn. 2002. “You So Don’t Talk Like Me: An Exploration of Southern California
Sound Changes.” presented at the American Dialect Society, San Francisco, CA.
Pierrehumbert, Janet B., and Stefan Frisch. 1997. “Synthesizing Allophonic Glottalization.” In
Progress in Speech Synthesis, edited by Jan P. H. van Santen, Richard W. Sproat, Joseph P.
Olive, and Julia Hirshberg, 9–26. New York: Springer.
Pitt, Mark A., Keith Johnson, Elizabeth Hume, Scott Kiesling, and William Raymond. 2005. “The
Buckeye Corpus of Conversational Speech: Labeling Conventions and a Test of Transcriber
Reliability.” Speech Communication 45 (1): 89–95.
Podesva, Robert J, Jeremy Reynolds, Patrick Callier and Jessica Baptiste. 2015. Constraints on the
social meaning of released /t/: A production and perception study of U.S. politicians. Language
Variation and Change 27(1). 59–87.
Powell, M. J. D. 2009. The BOBYQA algorithm for bound constrained optimization without
derivatives. Report, Department of Applied Mathematics and Theoretical Physics, Cambridge
University.
Preston, Dennis R. 1996. Where the worst English is spoken. In E. Schneider (ed.), Focus on the USA
(Varieties of English around the World), 297–360. Amsterdam: John Benjamins.
Preston, Dennis R. (ed.). 1999. A Handbook of Perceptual Dialectology. Vol. 1. Amsterdam: John
Benjamins.
Page 29 of 38
R Core Team. 2013. R: A language and environment for statistical computing. R Foundation for
Statistical Computing, Vienna, Austria
Reid, Euan. 1978. “Social and Stylistic Variation in the Speech of Children: Some Evidence from
Edinburgh.” In Sociolinguistic Patterns in British, edited by Peter Trudgill, 158–171.
Baltimore: University Park Press.
Roberts, Julie. 2006. “As Old Becomes New: Glottalization in Vermont.” American Speech 81: 227–
49.
Seyfarth, Scott, and Marc Garellek. 2015. “Coda Glottalization in American English.” In Proceedings
of the 18th International Congress of Phonetic Sciences, 5. Glasgow: The Scottish Consortium
for ICPhS 2015.
Straw, Michelle, and Peter L. Patrick. 2007. “Dialect acquisition of glottal variation in /t/: Barbadan in
Ipswich.” Language Sciences 29: 385–407.
Thomas, Erik. 2011. Sociophonetics: An Introduction. New York: Palgrave.
Tollfree, Laura. 1999. “South-East London English: Discrete versus Continual Modelling of Consonant
Reduction.” In Urban Voices: Accent Studies in the British Isles, edited by P.Foulkes and P.
Docherty, 163–184. London: Arnold.
Wells, J.C. 1982. Accents of English. Vol. 3. Cambridge: Cambridge University Press.
Page 30 of 38
Appendix A
The version of the letter that the speakers had access to did not have boldface or underlined keywords.
Underlined words are those containing a /t/ followed by nasal sequence (24), while boldface words are
those test items ending in /t/ that are followed by a following vowel-initial word. (28)
Dear Mom,
I know it's been a while since I've written. In the last letter I told you about a kitten I found that
looked like it hadn't eaten for a month. Well, it's living in my apartment now. It was in pretty bad
shape, definitely not fat or clean. It actually looks like it'd been beaten but it wasn’t frightened of
people at all. I've named it Keaton and my goal is to fatten it up a bit and show it people aren't so bad.
There are good things and bad things about having a cat in my apartment. The bad thing is that
it's started to climb up the drapes which made the curtain rods wobbly so I had to tighten them. He has
a certain fascination with buttons, and has bit a few off my sweaters. Now, I make sure to shut the
closet anytime I’m home, and not ever use the expensive satin sheets. He shreds the paper for the toilet
almost every day. I went into the bathroom one day and it looked like there were cotton balls by the
toilet and on the floor. Yesterday when I came home from work he'd done a number on the carpet in
the bedroom. I about flattened him but then realized that I'd forgotten to let him out in the morning, so
it really wasn't his fault.
In spite of the problems it's fun to have him. He seems to know when it's lunch time because
when I go toward the kitchen at about eleven, his face brightens up, he straightens his tail and starts to
flick it around, then he tries to beat me to the kitchen. I usually take him outside and let him sniff
around while I sit on the edge of the fountain and study for my Lit or Latin class. A guy from my
building saw me sitting there with Keaton. He sat on the bench and started playing with him. His name
is Lance Patton. He's an English major and is Colton Smith's roommate. They met on a study abroad
Page 31 of 38
trip to Great Britain. We got into a conversation and I think he may be interested in me. He actually
put a note with my number in his cell phone. What a way to sweeten my day!
I can't wait any longer to be home.
Love your daughter,
Candace
Page 32 of 38
Appendix B
Underlined words are those containing a /t/ followed by nasal sequence, while boldface words are those
test items ending in /t/ that are followed by a following vowel-initial word.
Test items containing /t/ followed by nasal (PreN)
1. I know it's been a while since I've written. In the last letter I told you about the kitten I found
that looked like it hadn't eaten for a month. Well, it’s living at my house now.
2. The cat started to climb up the drapes which made the curtain rods wobbly so I had to tighten
them. I hope my landlord doesn’t notice.
3. So, I learned that my new cat has a certain fascination with buttons, and likes to bite them off
my sweaters. I’m hoping he’ll stop when he’s older.
4. Yesterday when I came home from work he'd done a number on the carpet. I about flattened
him but then realized I'd forgotten to leave the door to the basement open, so it really wasn't his
fault.
5. He seems to know when it's close to lunch time because when I go toward the kitchen, his face
brightens up, he straightens his tail and then he tries to beat me there.
6. I always thought that cotton was a Latin word, but it’s actually an Arabic word that the Romans
borrowed.
Test items containing word-final /t/ followed by a vowel (PreV)
1. They make a mess in the house so now I make sure to shut the closet anytime I’m home, and
not ever let the cat around the laundry either. That’s taken care of the problem.
2. I totally forgot about all the new carpet on the floor, and left my muddy boots on when I came
in. Luckily, they were not even going to mention it.
Page 33 of 38
3. So, we met on Saturday, and I can’t wait until we get another chance to see each other again.
Can you believe we both like hip hop?
4. He’s not fat or anything, maybe a bit overweight if you consider his height. But I just want
people to stop making rude comments about him.
5. They usually just sit around playing video games, but the other day they started to fight over
some stupid thing and Jason put a major hole in the closet door with his fist.
6. The worst thing was that the toilet overflowed the night Irene came over so we spent the whole
time cleaning up the mess.
Appendix C
Test questions
(The numbers in the responses are the values used in the statistical analyses. They were not visible to
the participants.)
1. How old does she sound? Move the slider to the age.
(The slider ranged from 12 to 40.)
2. How much does she sound like she's from a small town or from a city?
-2 Definitely from a small town
-1 Probably from a small town
0 Can’t tell
1 Probably from a city
2 Definitely from a city
3. How pleasant does she sound?
2 Very pleasant
1 Somewhat pleasant
0 Neither pleasant nor unpleasant
Page 34 of 38
-1 Somewhat unpleasant
-2 Very unpleasant
4. How educated does she sound?
2 Very educated
1 Somewhat educated
0 Neither educated nor uneducated
-1 Somewhat uneducated
-2 Very uneducated
5. How much does she sound like she is from [state participant is from]?
2 Definitely from [state participant is from]
1 Probably from [state participant is from]
0 Can’t tell
-1 Probably not from [state participant is from]
-2 Definitely not from [state participant is from]
6. How friendly does she sound?
2 Very friendly
1 Somewhat friendly
0 Neither friendly nor unfriendly
-1 Somewhat unfriendly
-2 Very unfriendly
Page 35 of 38
Appendix D
Statistical results of the matched guise experiment.
PreN dependent variable: Education F Num df Den df pGENDER 2.734 1 219 0.100PRONUNCIATION 13.795 1 1125 < .001STATE 2.530 4 219 0.041GENDER ✻ PRONUNCIATION 0.375 1 1125 0.541GENDER ✻ STATE 0.113 4 219 0.978PRONUNCIATION ✻ STATE 0.923 4 1125 0.450
GENDER ✻ PRONUNCIATION ✻ STATE 0.566 4 1125 0.687
PreN dependent variable: Age (of speaker)STATE 0.4679 4 218 0.759GENDER 0.3035 1 218 0.582AGE (of listener) 1.3567 1 219 0.245PRONUNCIATION 0.5739 1 1126 0.449STATE ✻ GENDER 1.1932 4 218 0.315STATE ✻ PRONUNCIATION 0.7856 4 1125 0.535GENDER ✻ PRONUNCIATION 0.0635 1 1125 0.801STATE ✻ GENDER ✻ PRONUNCIATION 0.3976 4 1126 0.810GENDER ✻ STATE ✻ PRONUNCIATION 1.2764 4 1126 0.277
PreN dependent variable: FriendlinessGENDER 3.6513 1 219 0.057STATE 1.2798 4 219 0.279PRONUNCIATION 2.8534 1 1127 0.091GENDER ✻ STATE 0.9293 4 219 0.448GENDER ✻ PRONUNCIATION 0.0410 1 1126 0.840STATE ✻ PRONUNCIATION 0.5902 4 1126 0.670GENDER ✻ STATE ✻ PRONUNCIATION 1.2764 4 1126 0.277GENDER ✻ PRONUNCIATION 0.4701 1 1126 0.493STATE ✻ GENDER ✻ PRONUNCIATION 0.7745 4 1126 0.542
PreN dependent variable: PleasantnessSTATE 1.8597 4 219 0.119GENDER 3.9449 1 219 0.048PRONUNCIATION 0.9013 1 1126 0.343STATE ✻ GENDER 0.0619 4 219 0.993STATE ✻ PRONUNCIATION 0.4805 4 1126 0.750GENDER ✻ PRONUNCIATION 0.4701 1 1126 0.493STATE ✻ GENDER ✻ PRONUNCIATION 0.7745 4 1126 0.542
PreN dependent variable: StateGENDER 4.144 1 219 0.043
Page 36 of 38
STATE 14.614 4 219 < .001PRONUNCIATION 1.255 1 1126 0.263GENDER ✻ STATE 0.320 4 219 0.864GENDER ✻ PRONUNCIATION 1.414 1 1126 0.235STATE ✻ PRONUNCIATION 2.293 4 1126 0.058GENDER ✻ STATE ✻ PRONUNCIATION 0.389 4 1126 0.817
PreN dependent variable: RuralitySTATE 2.999 4 219 0.019GENDER 0.825 1 219 0.365PRONUNCIATION 17.199 1 1126 < .001STATE ✻ GENDER 0.445 4 219 0.776STATE ✻ PRONUNCIATION 1.534 4 1126 0.190GENDER ✻ PRONUNCIATION 1.841 1 1125 0.175STATE ✻ GENDER ✻ PRONUNCIATION 0.233 4 1126 0.920
PreV dependent variable: EducationSTATE 3.377 4 219 0.010GENDER 9.226 1 219 0.003PRONUNCIATION 3.907 1 1127 0.048STATE ✻ GENDER 0.596 4 219 0.666STATE ✻ PRONUNCIATION 0.566 4 1127 0.687GENDER ✻ PRONUNCIATION 1.708 1 1127 0.191STATE ✻ GENDER ✻ PRONUNCIATION 0.542 4 1127 0.705
PreV dependent variable: Age (of speaker)STATE 0.240 4 218 0.916GENDER 1.495 1 218 0.223AGE (of listener) 6.736 1 219 0.010PRONUNCIATION 1.651 1 1127 0.199STATE ✻ GENDER 1.010 4 218 0.403STATE ✻ PRONUNCIATION 1.188 4 1127 0.314GENDER ✻ PRONUNCIATION 0.625 1 1126 0.429STATE ✻ GENDER ✻ PRONUNCIATION 0.737 4 1127 0.567
PreV dependent variable: FriendlinessSTATE 0.670 4 219 0.614GENDER 4.349 1 219 0.038PRONUNCIATION 7.167 1 1127 0.008STATE ✻ GENDER 0.351 4 219 0.843STATE ✻ PRONUNCIATION 1.350 4 1127 0.249GENDER ✻ PRONUNCIATION 3.014 1 1127 0.083STATE ✻ GENDER ✻ PRONUNCIATION 0.415 4 1127 0.798
Page 37 of 38
PreV dependent variable: StateSTATE 4.7322 4 219 0.001GENDER 0.9830 1 219 0.323PRONUNCIATION 0.0238 1 1128 0.878STATE ✻ GENDER 0.2746 4 219 0.894STATE ✻ PRONUNCIATION 1.3917 4 1128 0.235GENDER ✻ PRONUNCIATION 2.2890 1 1127 0.131STATE ✻ GENDER ✻ PRONUNCIATION 1.2613 4 1128 0.283
PreV dependent variable: PleasantnessSTATE 1.141 4 219 0.338GENDER 8.267 1 219 0.004PRONUNCIATION 1.076 1 1127 0.300STATE ✻ GENDER 0.519 4 219 0.722STATE ✻ PRONUNCIATION 0.219 4 1127 0.928GENDER ✻ PRONUNCIATION 1.244 1 1127 0.265STATE ✻ GENDER ✻ PRONUNCIATION 1.002 4 1127 0.405
PreV dependent variable: RuralitySTATE 3.4671 4 218 0.009GENDER 2.4652 1 218 0.118AGE 0.0352 1 221 0.851PRONUNCIATION 3.2141 1 1128 0.073STATE ✻ GENDER 0.3559 4 218 0.840STATE ✻ PRONUNCIATION 1.3898 4 1128 0.235GENDER ✻ PRONUNCIATION 1.3530 1 1127 0.245STATE ✻ GENDER ✻ PRONUNCIATION 1.4871 4 1128 0.204
Page 38 of 38