The fluency principle: Why foreign accent strength negatively biases language attitudes

Abstract

Two experiments tested the prediction that heavy foreign-accented speakers are evaluated more negatively than mild foreign-accented speakers because the former are perceived as more prototypical (i.e., representative) of their respective group and their speech disrupts listeners’ processing fluency (i.e., is more difficult to process). Participants listened to a mild or heavy Punjabi- (Study 1) or Mandarin-accented (Study 2) speaker. Compared to the mild-accented speaker, the heavy-accented speaker in both studies was attributed less status (but not solidarity), was perceived as more prototypical of their respective group, disrupted listeners’ processing fluency, and elicited a more negative affective reaction. The negative effects of accent strength on status were mediated by processing fluency and sequentially by processing fluency and affect, but not by prototypicality. Theoretical, methodological, and practical implications are discussed.

Full text

Human Communication Research ISSN 0360-3989
ORIGINAL ARTICLE
I Don’t Like You Because You’re Hard to
Understand: The Role of Processing Fluency
in the Language Attitudes Process
Marko Dragojevic1& Howard Giles2
1 Department of Communication, University of Kentucky, Lexington, KY 40506-0042, USA
2 Department of Communication, University of California, Santa Barbara, CA 93106-4020, USA; School of
Psychology, The University of Queensland, Australia
Two experiments examined the eects of processing uency—that is, the ease with which
speech is processed—on language attitudes toward native- and foreign-accented speech.
Participants listened to an audio recording of a story read in either a Standard Ameri-
can English (SAE) or Punjabi English (PE) accent. ey heard the recording either free of
noise or mixed with background white noise of various intensity levels. Listeners attributed
more solidarity (but equal status) to the SAE than the PE accent. Compared to quieter lis-
tening conditions, noisier conditions reduced processing uency, elicited a more negative
aective reaction, and resulted in more negative language attitudes. Processing uency and
aect mediated the eects of noise on language attitudes. eoretical, methodological, and
practical implications are discussed.
Keywords: Language Attitudes, Processing Fluency, Accent, White Noise, Aect.
doi:10.1111/hcre.12079
Language is a powerful social force that conveys more than just referential informa-
tion. Research on the social evaluation of speech styles, or language attitudes,has
shown that people routinely make various judgments about others based simply on
how they speak (Garrett, 2010; Giles & Watson, 2013). For example, in the United
States, foreign-accented speakers tend to be judged less favorably on a host of traits,
including intelligence and friendliness, compared to their native-speaking counter-
parts. Because of this, these speakers oen face profound challenges and barriers to
access and opportunities in a wide range of social and professional settings (see Drago-
jevic, Giles, & Watson, 2013).
Such ndings have typically been explained with reference to categorization and
stereotyping (Lambert, 1967; Ryan, 1983). People use language cues (e.g., accent) to
make an inference about speakers’ social group membership(s) and, in turn, attribute
stereotypical traits associated with those inferred group memberships to them. In
Corresponding author: Marko Dragojevic; e-mail: marko.dragojevic@uky.edu
Human Communication Research (2016) © 2016 International Communication Association 1

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
other words, language attitudes reect people’s stereotypes about dierent linguistic
groups. Although this explanation has garnered considerable empirical support over
the years (see Giles & Watson, 2013), other mechanisms may also contribute to
the frequently observed downgrading of foreign- and other nonstandard-accented
speakers worldwide.
One such mechanism involves the ease or diculty with which listeners can
process a person’s speech. A growing body of literature has shown that human
judgment reects not only the content of our thoughts (e.g., stereotypes) but also
the metacognitive experience of processing those thoughts (see Alter & Oppen-
heimer, 2009; Pearson & Dovidio, 2014). Processing uency,denedastheeaseor
diculty with which information is processed, is one such metacognitive experi-
ence that has been shown to be a potent cue to judgment, independent of thought
content.
Any cognitive task— including speech processing— can be dened along a
continuum ranging from highly eortless to highly eortful,whichproducesacor-
responding metacognitive experience ranging from highly uent to highly disuent
(Alter & Oppenheimer, 2009). Researchers have manipulated processing uency
using a wide range of experimental methods, including varying visual and audio
clarity, frequency and duration of exposure, and vocabulary complexity, all produc-
ing remarkably similar eects on judgment across a wide array of domains (for a
review, see Alter & Oppenheimer, 2009). In general, high uency has promoted more
favorable judgments, including higher ratings of truth (Reber & Schwarz, 1999),
intelligence (Oppenheimer, 2006), and liking (Reber, Winkielman, & Schwarz, 1998),
among others.
Despite growing evidence of its ubiquitous inuence, the impact of processing u-
ency on language attitudes has remained largely unexplored (but see, e.g., Dovidio &
Gluszek, 2012; Lev-Ari & Keysar, 2010; Sebastian, Ryan, Keogh, & Schmidt, 1980).
is is surprising given that the ease with which a given person’s speech is processed
varies considerably from one situation to the next and is, at least in part, a function of
that person’s linguistic style (see Cristia et al., 2012). Indeed, a considerable amount of
research has shown that speech produced in accents dierent from one’s own, partic-
ularly foreign accents, is more dicult to process than speech produced in one’s own
accent (e.g., Floccia, Butler, Goslin, & Ellis, 2009; Gass & Varonis, 1984; Munro &
Derwing, 1995). Moreover, listeners have oen rationalized and justied their nega-
tive evaluations of foreign- and other nonstandard-accented speakers precisely based
on comprehensibility, or uency, concerns (Shuck, 2004, 2006).
Accordingly, one plausible hypothesis is that negative attitudes toward particular
accents are triggered, in part, by the diculty associated with processing speech
produced in those accents. We conducted two experiments to test this processing
uency hypothesis.Inthesectionsthatfollow,werstprovideabriefoverview
of the language attitudes literature. Next, we discuss the ways in which process-
ing uency can inuence language attitudes. We then describe two experiments
designed to test the role of processing uency in the language attitudes process
2Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
and discuss the theoretical, methodological, and practical implications of our
ndings.
Language attitudes
Past research on language attitudes has focused primarily on documenting the stereo-
types surrounding so-called “standard” and “nonstandard” language varieties (Gar-
rett, 2010; Giles & Watson, 2013). Standard varieties are those that adhere to codied
norms dening “correct” spoken and written usage in terms of grammar, vocabulary,
and pronunciation, whereas nonstandard varieties are those that depart from such
codied norms in some way (e.g., pronunciation; St. Clair, 1982; Milroy & Milroy,
1999). Examples of standard varieties include Standard American English (SAE) in
the United States, whereas examples of nonstandard varieties include most regional
(e.g., American Southern English) and ethnic (e.g., African American Vernacular
English) dialects as well as foreign accents (e.g., Spanish accent in United States).
Not unlike other social stereotypes (see Fiske, Cuddy, & Glick, 2007), language-
based stereotypes are organized along two primary evaluative dimensions: status
(e.g., intelligent, competent) and solidarity (e.g., friendly, sociable). Operating within
these two dimensions, past research has found that speakers of standard and non-
standard varieties elicit dierent evaluative reactions (Garrett, 2010; Giles & Watson,
2013). Status attributions were based primarily on perceptions of socioeconomic
status (Fiske et al., 2007; Woolard, 1985). Consequently, because standard varieties
tend to be associated with dominant socioeconomic groups within a given society,
standard speakers have typically been attributed more status than nonstandard
speakers (Fuertes, Gottdiener, Martin, Gilbert, & Giles, 2012).
Solidarity attributions, on the other hand, have tended to reect in-group loyalty
(for a discussion, see Dragojevic et al., 2013). Language is an important symbol of
socialidentity,anduseofthein-groupstylecanenhancefeelingsofsolidaritywithin
one’s own linguistic community (Giles, Bourhis, & Taylor, 1977). Accordingly, peo-
ple tend to attribute more solidarity to members of their own linguistic community,
particularly when that community is characterized by a high or increasing number of
speakers and enjoys institutional support (see ethnolinguistic vitality; Giles & John-
son, 1987). In this way, despite being downgraded on the status dimension, nonstan-
dard varieties can possess covert prestige, with users of those forms sometimes being
upgraded on the solidarity dimension by others who speak that variety (Luhman,
1990; Powesland & Giles, 1975).
Language-based stereotypes are socialized early in life (Day, 1982) and may be
transmitted in various ways, including through language criticism (Marlow & Giles,
2010), biased media portrayals (Dragojevic, Mastro, Giles, & Sink, in press), and sub-
tle linguistic biases (for a discussion, see Beukeboom, 2014), among others. Although
this stereotype-based account of the language attitudes process has garnered con-
siderable empirical support (Giles & Watson, 2013), other factors, such as listeners’
processing uency, may also play a role.
Human Communication Research (2016) © 2016 International Communication Association 3

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
Figure 1 eoretical model of the eects of processing uency on language attitudes via naïve
theories (path c)andaect(pathab).
Processing fluency and language attitudes
Processing uency can inuence judgment in two basic ways (for a schematic
diagram of these processes, see Figure 1). First, processing uency can inuence
judgment through the application of naïve theories, which provide relevant inference
rules (path c, Figure 1; Alter & Oppenheimer, 2009; Schwarz, 2004). People’s naïve
theoriesaboutprocessinguencyreecttheirassumptionsaboutwhatmakesiteasy
or dicult to process information, thus linking their metacognitive experiences to
the external world, the situation, and/or their state of knowledge.
For example, people believe that familiar information is more easily processed
than unfamiliar information (Schwarz, 2004). Consequently, they tend to interpret
high uency as an indicator of familiarity (e.g., Whittlesea, Jacoby & Girard, 1990).
Similarly, people believe that the more they know about something, the easier it is to
come up with examples. Accordingly, accessibility experiences can positively aect
expertise judgments (see Schwarz, 2004). People have a wide range of domain-specic
naïve theories about processing uency (Alter & Oppenheimer, 2009). e naïve
theory they employ to make sense of their metacognitive experiences— and thus the
inferences they make about those experiences—depends on the judgment task itself.
e application of one theory typically renders other theories inapplicable (Schwarz,
2004).
People may have several naïve theories that link their metacognitive experiences
of uency to their language attitudes. For example, people may believe that the ease
or diculty they experience processing a communicative message depends on the
sender’s competence and/or goodwill. People commonly place the burden of com-
munication disproportionately or even entirely on the sender (Lippi-Green, 2012).
at is, they believe that it is the sender’s responsibility to communicate his or her
message in a way they can easily understand.
Accordingly, they may interpret any diculty they experience processing a mes-
sage as indicative of the sender’s inability or unwillingness to communicate more
eectively and thus rate the sender lower on status and solidarity traits, respectively.
4Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
Consistent with this claim, Oppenheimer (2006) found that subjects in disuent con-
ditions (e.g., high lexical complexity) rated a target person as less intelligent than
subjects in uent conditions and that this eect was mediated by perceptions of com-
prehensibility (e.g., uency). By the naïve theory account, then, processing uency is
itself a cue to judgment and can exert a direct eect on language attitudes.
Second, processing uency can also inuence judgment indirectly via aect
(path ab, Figure 1; Dovidio & Gluszek, 2012; Reber, Schwarz, & Winkielman, 2004).
Processing uency is hedonically marked because it says something about a positive
or negative state of aairs, either in the world or within one’s cognitive system
(Winkielman, Schwarz, Fazendeiro, & Reber, 2003). Fluency provides feedback
about ongoing cognitive operations. High uency is indicative of facilitated cognitive
operations, whereas high disuency is indicative of hindered cognitive operations.
Accordingly, because people tend to experience cognitive progress— that is, the
successful decoding of information— as rewarding, high uency tends to be marked
with high positive aect, whereas high disuency tends to be marked with high
negative aect (Schwarz & Clore, 2007). Fluency-based aective reactions have been
demonstrated through both self-report and physiological measures (for an overview,
see Winkielman et al., 2003).
isuency-basedaectivereactioncan,inturn,biasjudgment.Indeed,acon-
siderable body of literature has shown that people’s aective reactions can inuence
a wide range of judgments, including liking, intelligence, and honesty, among others
(Forgas, 1990, 1995; Schwarz, 2012; for a recent review see, Greifeneder, Bless, &
Pham, 2011). In general, high positive aect has promoted favorable evaluations,
whereas high negative aect has promoted unfavorable evaluations (Byrne, 1971;
Byrne & Clore, 1970). For example, Forgas (1990) found that subjects rated a target
person lower on both solidarity- and status-related traits when the person was asso-
ciated with negative aect rather than positive or neutral aect. By this account then,
processing uency can have an indirect eect on language attitudes via aect, wherein
easier (more dicult) processing elicits a more positive (negative) aective reaction
that, in turn, promotes more (less) favorable status and solidarity attributions.
etworoutesdescribedabovearenotmutuallyexclusive.Processinguencymay
simultaneously exert both a direct eect on language attitudes, through the applica-
tion of naïve theories, as well as an indirect eect, via aect. Regardless of the route,
both accounts predict the same outcome: decreased uency should result in less favor-
able language attitudes. Two experiments were conducted to test this processing u-
ency hypothesis.
Overview of studies
To test the processing uency hypothesis, showing that the diculty of understanding
accentedspeechhasauniqueeectonlanguageattitudesthatcannotbeattributedto
stereotypes is important. A fairly direct test of this can be made by experimentally
varying the diculty with which listeners process the speech of the same speaker,
Human Communication Research (2016) © 2016 International Communication Association 5

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
speaking with the same accent. Previous research has shown that the presence of back-
ground white noise can make speech more dicult to process (e.g., Adank, Evans,
Stuart-Smith, & Scott, 2009; Munro, 1998; Van Engen & Bradlow, 2007) as well as
eliciting a negative aective reaction (e.g., Sebastian et al., 1980). Accordingly, if the
processing uency hypothesis is correct, then the same speaker should be evaluated
less favorably when his or her speech is heard in the presence of background white
noiseratherthanfreeofnoise,andthiseectshouldbemediatedbyprocessingu-
ency and sequentially by processing uency and aect.
In partial support of this hypothesis, Sebastian et al. (1980) had participants lis-
tentoanSAE-accentedspeakeroveraquiettape,freeofnoise,oranoisytapemixed
with background white noise. Consistent with the processing uency hypothesis, the
speaker was perceived as more dicult to understand and rated more negatively on
status and solidarity traits in the noisy condition than the quiet condition. Although
these results provide preliminary evidence that factors which decrease processing
uency can also negatively inuence language attitudes, they are limited to a single
accent and do not provide direct empirical evidence of the mechanisms underlying
this eect. Two experiments were designed to provide a more stringent test of the
processing uency hypothesis.
Experiment 1
On the basis of the above reasoning, participants in the rst experiment listened to
an audio recording of a speaker reading a short story in either an SAE or a Punjabi
English (PE) accent. e recording was played to participants either free of noise or
mixed with background white noise of one of four signal-to-noise (SNR) ratios: +30,
+20, +10, and 0 decibels (dB). SNRs represent the intensity of a signal (e.g., speech)
relative to the intensity of background noise. Lower SNRs represent more noisy lis-
tening conditions.
e SAE and PE accent were selected for several reasons. First, past research has
suggested that naïve listeners can reliably identify both varieties from dierences in
pronunciation alone (Clopper & Pisoni, 2004; Dragojevic & Giles, 2014). Second, PE is
a variety that many Americans frequently encounter in their daily lives due to increas-
ing globalization of the world’s economies and the outsourcing of many American
companies’ call centers to India (Chand, 2009). ird, the two varieties are associ-
ated with distinct stereotypes. Previous research has specically shown that American
listeners tend to attribute less status and solidarity to PE than SAE speakers (e.g.,
Dragojevic & Giles, 2014). Expecting to replicate these latter ndings pertaining to
accent and stereotypes, we predicted that:
H1a-b: e SAE speaker will be rated more favorably on (a) status and (b) solidarity traits
than the PE speaker.
We also expected noise to inuence listeners’ evaluations of the two speakers. As
noted above, the presence of background white noise can make speech more di-
cult to process by degrading the quality of the speech signal (e.g., Clopper & Bradlow,
6Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
2008). In turn, the increased diculty associated with processing speech in noisy lis-
tening conditions is likely to elicit a more negative aective response (e.g., frustration)
in those conditions compared to quieter listening conditions (Sebastian et al., 1980).
Accordingly, we predicted that:
H2: Noisier listening conditions will result in lower processing uency than quieter
listening conditions.
H3: Noisier listening conditions will elicit a more negative aective response than
quieter listening conditions.
In turn, the increased diculty and more negative aect associated with process-
ing speech in noisy listening conditions is likely to promote more negative speaker
evaluations in those conditions compared to quieter conditions. As explained ear-
lier, processing uency can have a direct eect on language attitudes, through the
application of naïve theories, as well as an indirect eect, via uency-based aective
reactions (Alter & Oppenheimer, 2009; Schwarz & Clore, 2007). Regardless of route,
both accounts predict that factors which make speech more dicult to process should
also lead to more negative speaker evaluations. We predicted that:
H4a-b: Noisier listening conditions will result in lower (a) status and (b) solidarity
ratings than quieter listening conditions.
Implicit in the above rationale is the notion that processing uency and aect
mediatetheeectsofnoiseonlanguageattitudes;thatis,noisedisruptslisteners’pro-
cessing uency, which elicits a negative aective reaction. Processing uency and the
associated uency-based aective reaction, in turn, both exert an eect on language
attitudes. us, we also predicted that:
H5a-b: e eects of noise on language attitudes will be mediated by (a) processing
uency and (b) sequentially by processing uency and aect.
Method
Design
e study was a 2 (accent: SAE, PE) ×5(noise:quiet,+30 dB, +20 dB, +10 dB, 0 dB)
between-subjects factorial design.
Participants
Participants were 351 undergraduate students from a large university on the West
Coast of the United States. ese participants (83.8% women) ranged in age from 18
to 26 years old (M=19.54) and reported their ethnicity as follows: White (46.7%),
Hispanic (22.2%), Asian (21.9%), African American (4.8%), and other (4.4%).
Voice stimuli
e present study employed the matched-guise technique (MGT; e.g., Lambert,
1967), which involves bidialectical or bilingual speakers producing the same pas-
sage of text in dierent linguistic styles, or guises. is procedure ensures that
dierences across guises reect only the features of the linguistic style itself rather
Human Communication Research (2016) © 2016 International Communication Association 7

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
than extraneous variables (e.g., pitch). e voice stimuli for the present study were
produced by a female Californian native of Indian descent in her mid-20s. She was
recorded reading a short, ctional story in an SAE accent and a PE accent. e
speaker was instructed to adopt a moderate pace throughout and to keep all aspects
of her speech other than accent constant across the guises. Both recordings were 60
seconds long.
e two recordings were normalized at 70 dB sound pressure level (SPL). ese
recordings were then mixed with white noise of varying intensity levels to produce
four dierent noisy conditions, characterized by dierent SNRs: +30, +20, +10,
and 0 dB. is yielded a total of 10 recordings— that is, SAE or PE recording free of
noiseormixedwithwhitenoiseatoneoffourSNRs.Weemphasizethattherangeof
SNRsselectedforthisstudy,includingthelowest(i.e.,noisiest),aretypical of most
real-world situations (see Pearsons, Bennett, & Fidell, 1977).
Procedure
e experiment was introduced to participants as being concerned with how peo-
ple rate others’ personalities based on limited information, such as their voice. ey
were told that they would listen to a speaker reading a short, ctional story and then
answer several questions about the speaker and themselves. Participants were ran-
domlyassignedtolistentooneofthe10recordingsdescribedabove.
Having listened to the recording, participants completed a questionnaire contain-
ing the dependent measures. First, participants indicated the extent to which they
felteachofthreenegative(i.e.,annoyed,irritated,frustrated)andthreepositiveemo-
tions (i.e., interested, happy, enthusiastic) on a 5-point scale (1 =very slightly or not
at all,5=extremely). ese items were adapted from Watson, Clark, and Tellegen’s
(1988) positive and negative aect schedule (PANAS). e three positive and three
negative emotion items were averaged to form the positive aect (α=.74) and nega-
tive aect scale (α=.84),respectively.Becausetheprimarygoalofthepresentstudy
wastoassesstheoverall valance of listeners’ aective responses, an aect (or hedonic)
balance score was calculated by subtracting mean negative aect from mean positive
aect and adding a constant of 4 to avoid negative values. is procedure has been
used successfully in past research (e.g., Diener & Seligman, 2002). e aect balance
score had a theoretical range from 0 (high negative aect)to8(high positive aect)
(M=4.49; SD =1.27).
Second, participants rated the extent to which each of 10 personality traits—ve
relating to status (i.e., intelligent, educated, smart, competent, successful) and ve
relating to solidarity (i.e., friendly, nice, pleasant, honest, sociable)— characterized
the speaker on a 7-point scale (1 =not at all,7=very). ese items were adapted from
past research on language attitudes (Heaton & Nygaard, 2011; Zahn & Hopper, 1985).
e ve status and ve solidarity items were averaged to form the status (α=.93)
and solidarity scale (α=.91), respectively. Next, participants indicated the extent to
which the speaker was comprehensible, easy to understand, clear to understand, and
eortful to understand (reverse coded) on a 7-point scale (1=notatall,7=very).
8Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
ese items were averaged to form the processing uency scale (α=.92). Participants
then indicated where they thought the speaker was from in an open-ended question
(i.e., Where is the speaker you heard from?). Finally, participants provided standard
demographic information.
Results
Manipulation check
Processing uency was manipulated by varying the listening η2
pconditions under
which participants heard the recording. us, it was necessary to conrm that the
dierent noise conditions did, in fact, produce dierences in processing uency as
intended. ere was a main eect of noise on processing uency, F(4, 341) =27.48,
p<.001, η2
p=.24. Pair-wise comparisons showed that uency was lower in the 0 dB
SNR condition (M=2.78) than the +10 dB SNR condition (M=3.86) which, in turn,
was lower than uency in the +20 dB SNR (M=4.43), +30 dB SNR (M=4.36), and
quiet conditions (M=4.56), ps<.05; uency in these latter three conditions, how-
ever, was not reliably dierent (ps>.31). ere was also a main eect for accent, F(1,
341) =286.27, p<.001, η2
p=.46. e PE guise was perceived as less uent (M=2.93)
than the SAE guise (M=5.04). e interaction was not signicant (p=.14).
Given that the manipulation failed to yield dierences in uency between the
quiet, +30 dB, and +20 dB noise conditions, data across these three conditions was
collapsed. Accordingly, a new, 3-level noise factor was constructed: low noise (quiet,
+30 dB, +20 dB conditions collapsed), moderate noise (+10 dB), and high noise
(0 dB). Fluency was higher in the low-noise (M=4.45) than the moderate-noise con-
dition (M=3.86) which, in turn, was higher than uency in the high-noise condition
(M=2.78), ps<.001. is 3-level noise factor was used for all subsequent analyses.
Focal analyses
Aect,status,andsolidarityweresubmittedtoa2(accent:PE,SAE)×3(noise:low,
moderate, high) analysis of variance (ANOVA). Hypotheses pertaining to the eects
of noise on the dependent measures were tested using repeated (or sequential) con-
trasts (see Hayes & Preacher, 2014; Wendorf, 2004). ese contrasts compare the mean
at each level of a factor to the mean one step sequentially lower in the ordered system.
Such contrasts are especially useful when a multicategorical predictor is an ordered
variable,aswasthecaseinthepresentstudy.Torepresentthe3-levelnoisefactor,two
contrastswerecreated.erstcontrast(c1=011)comparedthemeanofthemoder-
ate noise condition to the mean of the low noise condition. e second contrast (c2=0
0 1) compared the mean of the high noise condition to the mean of the moderate noise
condition. Cell means and standard deviations are displayed in Table 1.
Aect
erewasamaineectfornoise,F(2, 345) =10.73, p<.001, η2
p=.06, and accent,
F(1, 345) =9.70, p<.01, η2
p=.03. ese eects were subsumed by an interaction, F(2,
345) =3.22, p<.05, η2
p=.02.Simplemaineectanalysesshowedthatnoisehadan
Human Communication Research (2016) © 2016 International Communication Association 9

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
Table 1 Eects of Accent and Noise on Dependent Measures in Experiment 1
Noise
Measure Accent Low Moderate High Signicant Eects
Processing uency SAE 5.54 5.03 3.60 N, A
(1.10) (1.53) (1.31)
PE 3.35 2.66 1.94
(1.17) (1.14) (0.81)
Aect SAE 4.73 4.66 4.35 N, A, NA
(0.94) (1.31) (1.57)
PE 4.70 4.10 3.54
(1.10) (1.42) (1.59)
Status SAE 4.39 4.13 4.21 N, A, NA
(0.97) (1.16) (0.96)
PE 4.32 3.83 3.22
(0.99) (0.95) (1.26)
Solidarity SAE 5.07 4.97 4.94 N, A, NA
(1.05) (1.20) (1.17)
PE 4.57 4.10 3.67
(1.09) (1.02) (1.38)
Notes: Standard deviations appear in parentheses below their respective means. In the “Accent”
column, SAE =Standard American English accent, PE =Punjabi English accent. In the “Sig-
nicant Eects” column, N =main eect for noise, A =main eect for accent, NA =two-way
interaction. All reported eects are signicant, p<.05.
eect on aect for the PE guise [F(2, 345) =12.57, p<.001] but not for the SAE guise
[F(2, 345) =1.26, p=.29]. Accordingly, planned contrasts were tested only for the
Punjabi guise. ese revealed that the high-noise condition elicited a more negative
aective response (M=3.54) than the moderate-noise condition (M=4.10), p<.05,
which, in turn, elicited a more negative aective response than the low-noise con-
dition (M=4.70), p<.01. In other words, as noise increased, participants’ aective
response became more negative, but only for the PE guise.
Status
ere was a main eect for noise, F(2, 345) =11.64, p<.001, η2
p=.06, and accent,
F(1, 345) =13.43, p<.001, η2
p=.04. ese eects were subsumed by an interaction,
F(2, 345) =5.23, p=.006, η2
p=.03. Simple main eects analyses showed that noise
hadaneectonstatusattributionsforthePEguise[F(2, 345) =15.64, p<.001] but
not for the SAE guise [F(2, 345) =1.07, p=.34]. Accordingly, planned contrasts were
onlycarriedoutforthePEguise.eserevealedthatstatusratingswerelowerinthe
high-noise condition (M=3.22) than in the moderate-noise condition (M=3.83),
p<.01, which, in turn, were lower than status ratings in the low-noise condition
(M=4.32), p<.01. In other words, as noise increased, status attributions decreased,
but only for the PE guise. Pair-wise comparisons also showed that the PE and
10 Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
SAE guise were attributed equal status in the low- (MPE =4.32; MSAE =4.39) and
moderate-noise (MPE =3.83; MSAE =4.13) conditions (ps>.21). However, in the
high-noise condition, the PE guise was attributed less status (M=3.22) than the SAE
guise (M=4.21), p<.001.
Solidarity
ere was a main eect for noise, F(2, 345) =6.14, p<.01, η2
p=.03, and for accent,
F(1, 345) =42.01, p<.001, η2
p=.11. ese eects were subsumed by an interaction,
F(2, 345) =3.26, p<.05, η2
p=.02. Simple main eects analyses showed that noise
inuenced solidarity ratings of the PE guise [F(2, 345) =9.06, p<.001] but not the
SAE guise [F(2, 345) =0.24, p=.79]. Accordingly, planned contrasts were only car-
ried out for the PE guise. ese revealed that the moderate-noise condition elicited
lower solidarity attributions (M=4.10) than the low-noise condition (M=4.57),
p<.05. Although the high-noise condition also elicited lower solidarity attributions
(M=3.67) than the moderate-noise condition, this dierence was only marginally
signicant, p=.053. In sum, as noise increased, solidarity attributions decreased, but
only for the PE guise. Pair-wise comparisons also showed that the Punjabi guise was
attributed less solidarity than the SAE guise in the low- (MPE =4.57, MSAE =5.07),
moderate- (MPE =4.10, MSAE =4.97), and high-noise (MPE =3.67, MSAE =4.94)
conditions (ps<.01).
Mediation analyses
Hayes’s (2013) PROCESS macro (Model 6) was used to test whether the signicant
eects of noise on status and solidarity for the PE guise were mediated by processing
uency and sequentially by processing uency and aect. Repeated contrast codes
were created for the noise factor (as described above) and entered as predictors. With
repeated codes, the relative indirect eects can be interpreted as the eects of mem-
bershipinonegroup(e.g.,moderatenoise)relativetothegrouponestepsequentially
lower in the ordered system (e.g., low noise; Hayes & Preacher, 2014; Wendorf, 2004).
Processing uency was entered as the rst mediator in the sequential chain, followed
by aect. e analysis used 10,000 bootstrap resamples and a bias corrected and accel-
erated 95% condence interval (CI) as recommended (e.g., Hayes & Preacher, 2014).
A given indirect eect was considered signicant if its respective CI did not contain 0
(Hayes, 2013). All coecients reported below are unstandardized. Standardized coef-
cients are displayed in the path diagrams in Figures 2 and 3.
When processing uency, aect, and the contrast-coded noise factor were
included in the model simultaneously; processing uency (B=.35, p<.001) and
aect (B=.23, p<.001) were both signicant predictors of status attributions,
whereas all direct eects of noise were rendered nonsignicant (ps>.26). e
indirect eects of the moderate-noise condition relative to the low-noise condition
via processing uency (B=−.24, CI =−.49, −.08) and sequentially via processing
uency and aect (B=−.06, CI =−.17, −.02) were both signicant. Similarly, the
indirect eects of the high-noise condition relative to the moderate-noise condition
Human Communication Research (2016) © 2016 International Communication Association 11

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
Figure 2 Indirect eects of noise on status attributions for the Punjabi English (PE) guise
via processing uency and aect (Experiment 1). Unstandardized path coecients are listed
rst followed by standardized path coecients in parentheses. Signicant paths (p<.05) are
denoted by solid lines and bolded coecients. Nonsignicant paths are denoted by dashed
lines.
via processing uency (B=−.25, CI =−.48, −.10) and sequentially via processing
uency and aect (B=−.07, CI =−.17, −.02) were also both signicant. Stated
dierently, processing uency mediated the eects of noise on status attributions and
aect mediated the eects of processing uency on status attributions (see Figure 2).
Similarly, when processing uency, aect, and the contrast-coded noise factor
were included in the model simultaneously, processing uency (B=.20, p=.01) and
Figure 3 Indirect eects of noise on solidarity attributions for the Punjabi English (PE) guise
via processing uency and aect (Experiment 1). Unstandardized path coecients are listed
rst followed by standardized path coecients in parentheses. Signicant paths (p<.05) are
denoted by solid lines and bolded coecients. Nonsignicant paths are denoted by dashed
lines.
12 Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
aect (B=.23, p<.001) were both signicant predictors of solidarity, whereas all
direct eects of noise were rendered nonsignicant (ps>.38).eindirecteects
of the moderate-noise condition relative to the low-noise condition via processing
uency (B=−.14, CI =−.36, −.03) and sequentially via processing uency and aect
(B=−.07, CI =−.17, −.02) were both signicant. Similarly, the indirect eects of
the high-noise condition relative to the moderate-noise condition via processing
uency (B=−.14, CI =−.34, −.03) and sequentially via processing uency and aect
(B=−.07, CI =−.17, −.02) were also both signicant. Stated dierently, processing
uency mediated the eects of noise on solidarity attributions and aect mediated
theeectsofprocessinguencyonsolidarityattributions(seeFigure3).
Discussion
Experiment 1 examined the eects of noise and speakers’ accent on listeners’ language
attitudes. Consistent with H1a, the SAE guise was rated higher on solidarity than the
PE guise. However, contrary to H1b, the two guises were rated similarly on the sta-
tus dimension. In addition to accent, noise also had an eect on listeners’ language
attitudes, but only for the PE guise. Noisier listening conditions specically reduced
processing uency (H2), elicited a more negative aective reaction (H3), and resulted
in lower status and solidarity attributions (H4). Also consistent with predictions (H5),
theeectsofnoiseonlanguageattitudesweremediatedbyprocessinguencyand
sequentially by processing uency and aect; that is, compared to quieter listening
conditions, noisier listening conditions made speech more dicult to process, which,
in turn, elicited a more negative aective reaction. is increased diculty and more
negative aect reaction both had a negative eect on listeners’ language attitudes (see
Figures 2 and 3).
Although results of Experiment 1 provide preliminary evidence of systematic
eects of processing uency on language attitudes, there are several issues that
need to be resolved before rmer conclusions can be drawn. First, the study did
not explicitly control for categorization and stereotyping. Although listeners heard
the same speaker in the dierent noise conditions, they may have categorized the
speaker dierently across the conditions. Past research has found that variables which
make speech more dicult to process, such as noise, can also make categorization
more dicult (see Clopper & Bradlow, 2008). Inspection of listeners’ responses to
the open-ended question of where the speaker was from allowed us to examine the
extent to which the two guises were categorized correctly across the dierent noise
conditions— that is, as being from California/West Coast or India — based on accent
condition. Categorization accuracy for the SAE guise was equal across the three noise
conditions (low =78.1%; moderate =69.4%; high =80.6%), χ2(2) =1.48, p=.48.
In contrast, categorization accuracy for the PE guise varied across the three noise
conditions, χ2(2) =37.95, p<.001. Categorization accuracy was specically equally
high in the low- (83.7%) and moderate-noise conditions (71.4%) but signicantly
lower in the high-noise condition (28.6%). us, less favorable evaluations of the
PE guise in the high-noise condition relative to the other two conditions cannot be
Human Communication Research (2016) © 2016 International Communication Association 13

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
unequivocally attributed to processing uency; rather, they may simply reect less
favorable stereotypes stemming from a dierent categorization (e.g., foreigner vs.
Indian) (cf. Lindemann, 2003).
Second, the reason that noise had no eect on evaluations of the SAE guise remains
uncleardespitethefactthatitdisrupteduencyforthatguiseaswell.Processingdi-
culties may be more likely to prompt source derogation when listeners are at least min-
imally motivated to understand one another and when they perceive those diculties
as likely to impair their ability to carry on a conversation or execute other interaction
demands successfully (cf. Sebastian et al., 1980). Accordingly, one possible explana-
tion for why noise had no eect on evaluations of the SAE guise is that the processing
diculties listeners experienced when listening to SAE speech, even in the noisiest
condition, were insuciently large to be perceived as communicatively consequential.
Foreign-accented speech is itself more dicult to process than native-accented speech
(e.g., Floccia et al., 2009; Munro & Derwing, 1995; for a discussion, see Dovidio &
Gluszek, 2012). us, even in the presence of considerable background noise, SAE
speech is likely to remain signicantly more uent than PE speech.
Our data support this conclusion. Inspection of the cell means (see Table 1)
shows that listeners perceived SAE speech in the noisiest condition as easier to
process (M=3.60) than PE speech in all the noise conditions, including the quietest
(M=3.35). In other words, even in the noisiest listening condition, the SAE guise
remained relatively uent. is point is further reinforced by the fact that noise
had no eect on aect for the SAE guise, suggesting that any diculties listeners
experienced processing SAE speech were insucient to elicit a negative aective
reaction (e.g., frustration, annoyance). us, had the processing diculties been
greater or had participants been cognizant of the communicative consequences of
those diculties—as they would be in most real-world situations — then perhaps
noise would have inuenced evaluations of the SAE guise as well. To address the two
issues above, a second experiment was conducted.
Experiment 2
e second experiment was a replication of Experiment 1, with two notable excep-
tions. First, to provide a more stringent test of the processing uency hypothesis, we
controlled for categorization (and thus stereotyping); namely, prior to listening to the
recording, participants were explicitly told where the speaker they would hear was
from—that is, from California or India, depending on accent condition. is ensured
that each guise was categorized correctly and in the same manner across the dierent
noise conditions.
Second, the communicative consequences of listeners’ processing diculties were
made salient to them by having them complete a memory task prior to rating the
speaker. If, as suggested above, noise had no eect on evaluations of the SAE guise in
the rst study because listeners perceived any processing diculties they experienced
as communicatively inconsequential, then making participants explicitly aware of the
14 Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
communicative consequences of those diculties should produce signicant eects
of noise on language attitudes even toward the SAE guise. In line with the processing
uencyhypothesis,thesamepredictionsweremadeasintherststudy(seeabove).
Method
Design
e experiment was a 2 (accent: SAE, PE) ×2 (noise: quiet, noisy) between-subjects
factorial design.
Participants
Participants were a dierent group of 173 undergraduates from the same West Coast
U.S. university. ese participants (73.4% women) ranged in age from 18 to 24 years
old (M=19.63) and reported their ethnicity as follows: White (34.7%), Asian (32.4%),
Hispanic (22.5%), African American (6.4%), and other (4.0%).
Voice stimuli
e same SAE and PE voice stimuli were used as in the rst experiment. ese record-
ings were mixed with background white noise at a SNR of +2dB. is yielded four
recordings: SAE or PE guise, in quiet or in noise.
Procedure
Participants were randomly assigned to listen to one of the four recordings described
above. Experiment 2 followed the same general procedure as the rst experiment, with
two exceptions. First, prior to hearing the recording, the researcher explicitly told the
participants where each speaker was from— that is, “e speaker you will hear is a
woman from California/India,” with location varied according to accent condition.
Second, having heard the recording and prior to completing the dependent measures,
participants completed a ll-in-the-blank memory task. Participants were specically
presented with a transcript of the story they had heard with 12 words omitted. eir
task was to write in the missing words. All omitted words were of low predictability
from the context of the passage to minimize guessing.
Raw scores on the memory task were converted to percent correct scores and then
to rationalized arcsine units (RAUs; Studebaker, 1985). is transformation is custom-
ary for speech recognition performance scores expressed in proportional units (e.g.,
Adank et al., 2009) and ensures that the mean and variance of the data are uncorre-
latedandthatthedataareonalinearandadditivescale(Sherbecoe&Studebaker,
2004). Scores on this scale range from −23RAU(correspondingto0%correct)to
+123 RAU (corresponding to 100% correct).
Immediately aer the memory task, participants completed the same dependent
measures used in the rst experiment, indicated where they thought the speaker
was from, and provided standard demographic information. All scales were reliable:
negative aect (α=.89), positive aect (α=.70), status (α=.94), solidarity (α=.91),
and processing uency (α=.88).Onceagain,anaectbalancescorewascreatedby
Human Communication Research (2016) © 2016 International Communication Association 15

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
subtracting mean negative aect from mean positive aect for each participant and
adding a constant of 4 to avoid negative values (M=4.28, SD =1.42).
Results
Manipulation check
Inspection of listeners’ responses to the open-ended question of where the speaker
was from showed that all participants categorized the guises as intended—that is, the
SAE guise as being from California and the PE guise as being from India. us, the
categorization manipulation was successful. Also as expected, processing uency was
lower in the noisy (M=2.37) than the quiet condition (M=4.42), F(1, 169) =188.05,
p<.001, ηp2=.53. Processing uency was also lower for the PE guise (M=2.79) than
the SAE guise (M=3.98), F(1, 169) =63.01, p<.001, ηp2=.27. e interaction was
nonsignicant (p=.17). e negative eects of noise on processing uency were fur-
ther reinforced by participants’ actual performance on the memory task. As expected,
performance in the noisy condition (MSAE =21.52; MPE =−8.51) was lower than in
the quiet condition (MSAE =35.70; MPE =33.44), and this was more pronounced for
the PE guise as evidenced by a signicant interaction, F(1, 169) =17.44, p<.001,
ηp2=.09. us, the noise manipulation was also successful.
Focal analyses
Status, solidarity, and aect were each submitted to a separate ANOVA. Cell means
and standard deviations are displayed in Table 2.
Aect
Amaineectfornoise[F(1, 169) =12.26, p=.001, ηp2=.07] revealed that the noisy
condition elicited a more negative aective response (M=3.93) than the quiet condi-
tion (M=4.66). All other eects were nonsignicant (ps>.13).
Status
A main eect for noise [F(1, 169) =8.77, p<.01, ηp2=.05] showed that status attri-
butions were lower in the noisy (M=4.01) than the quiet condition (M=4.52). All
other eects were nonsignicant (Fs<1).
Solidarity
A main eect for accent [F(1, 169) =4.22, p<.05, ηp2=.02] showed that the SAE
guise was attributed more solidarity (M=5.10) than the PE guise (M=4.73).
Although the main eect for noise was not signicant [F(1, 169) =1.31, p=.25], the
pattern of means was in the predicted direction (Mquiet =5.01; Mnoisy =4.81). e
interaction was nonsignicant (F<1).
Mediation analyses
Hayes’s (2013) PROCESS macro (Model 6) was used to test whether the signicant
eect of noise on status attributions was mediated by processing uency and sequen-
tially by processing uency and aect using the same procedure employed in the rst
16 Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
Table 2 Eects of Accent and Noise on Dependent Measures in Experiment 2
Noise
Measure Accent Quiet Noisy Signicant Eects
Processing uency SAE 5.13 2.88 N, A
(1.23) (0.84)
PE 3.74 1.90
(1.05) (0.76)
Aect SAE 4.82 4.08 N
(1.04) (1.58)
PE 4.50 3.78
(1.22) (1.55)
Status SAE 4.48 4.03 N
(0.80) (1.08)
PE 4.56 4.00
(1.21) (1.29)
Solidarity SAE 5.22 4.98 A
(1.19) (1.27)
PE 4.81 4.65
(1.02) (1.22)
Notes: Standard deviations appear in parentheses below their respective means. In the “Accent”
column, SAE =Standard American English accent, PE =Punjabi English accent. In the “Sig-
nicant Eects” column, N =main eect for noise, A =main eect for accent, NA =two-way
interaction. All reported eects are signicant, p<.05. e 173 participants were randomly
distributed among the four between-subjects experimental conditions as follows: quiet/SAE
(n=41); noisy/SAE (n=43); quiet/PE (n=43); noisy/PE (n=46).
experiment. Noise was dummy coded (0 =quiet, 1 =noisy) and entered as the predic-
tor. Processing uency was entered as the rst mediator in the sequential chain fol-
lowed by aect. Given that the eects of noise on all the dependent variables were con-
sistent across the two guises, guise was dummy coded (0 =SAE, 1 =PE) and treated
as a covariate. All coecients reported below are unstandardized. Standardized coef-
cients are displayed in the path diagram in Figure 4.
Whenprocessinguency,aect,andthedummy-codednoisefactorwereincluded
in the model simultaneously, processing uency (B=.27, p<.01) and aect (B=.18,
p<.01) were both signicant predictors of status attributions, whereas the direct eect
of noise was rendered nonsignicant (B=.18, p=.45).eindirecteectofnoiseon
status via processing uency was signicant (B=−.55, CI =−.93, −.21) as was the
indirect eect of noise on status via processing uency and aect (B=−.09, CI =−.24,
−.01). Stated dierently, processing uency mediated the eect of noise on status attri-
butionsandaectmediatedtheeectofprocessinguencyonstatusattributions(see
Figure 4).
Human Communication Research (2016) © 2016 International Communication Association 17

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
Figure 4 Indirect eects of noise on status attributions via processing uency and aect, con-
trolling for guise (Experiment 2). Unstandardized path coecients are listed rst followed by
standardized path coecients in parentheses. Signicant paths (p<.05) are denoted by solid
lines and bolded coecients. Nonsignicant paths are denoted by dashed lines.
Discussion
Experiment 2 provided a more stringent test of the processing uency hypothesis
by controlling for categorization and, thus, stereotyping. In addition, the potential
communicative consequences of listeners’ processing diculties were made salient to
them by having them complete a memory task prior to rating the speaker. Consistent
with H1a, the SAE guise was attributed more solidarity than the PE guise. However,
contrary to H1b, the two guises were rated similarly on the status dimension. Noise
also had an eect. Compared to the quiet condition, the noisy listening condition
specically reduced processing uency (H2), elicited a more negative aective reac-
tion (H3), and resulted in lower status (but not solidarity) attributions (H4) for both
the SAE and PE guise. Consistent with the processing uency hypothesis, the signif-
icant eects of noise on status attributions were mediated by processing uency and
sequentially by processing uency and aect (H5) (see Figure 4). Given that listeners
heard the same speaker and categorized her in the same manner across the two noise
conditions, these ndings cannot be attributed to stereotyping.
General discussion
Past research on language attitudes has shown that, worldwide and cross-culturally,
foreign- and other nonstandard-accented speakers tend to be rated less favorably on
status (e.g., intelligent) and solidarity (e.g., friendly) traits compared to native- and
standard-accented speakers, respectively (Giles & Watson, 2013). ese ndings have
typically been explained with reference to categorization and stereotyping (Ryan,
1983); namely, listeners use language cues (e.g., accent) to make an inference about
speakers’ social group membership(s) and, in turn, attribute to them stereotypical
traits associated with those inferred group memberships.
Findings of the two present experiments provide compelling evidence for anaddi-
tional explanatory mechanism of the language attitudes process by showing that more
negative attitudes toward a particular accent can be triggered simply by the diculty
associated with processing speech in that accent (i.e., listeners’ processing uency). In
18 Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
both experiments, participants listened to an audio recording of a short story read in
either an SAE or a PE accent. ey heard the recording either free of noise or mixed
with background white noise of various intensity levels.
Listeners attributed more solidarity (but not status) to the SAE than PE accent.
Noise also had an eect; compared to quieter listening conditions, noisier condi-
tions reduced listeners’ processing uency, elicited a more negative aective reaction,
and resulted in more negative language attitudes. Moreover, the signicant eects of
noise on language attitudes were mediated by processing uency and sequentially by
processing uency and aect; that is, noisier listening conditions made speech more
dicult to process, which, in turn, elicited a more negative aective reaction. In turn,
this increased diculty and more negative aect both exerted a negative aect on lis-
teners’ language attitudes. Given that listeners heard the same speaker in the dierent
noise conditions (both experiments) and categorized the speaker in the same manner
across the dierent noise conditions (Experiment 2), these results cannot be attributed
to stereotyping.
Together, these ndings provide compelling evidence of systematic eects of pro-
cessing uency on language attitudes, independent of stereotyping,by demonstrating
that factors which reduce listeners’ processing uency (e.g., noise) can also have a
negative eect on listeners’ language attitudes. Given that foreign-accented speech
tends to be more dicult to process than native-accented speech (e.g., Munro & Der-
wing, 1995), results of the present research suggest that one reason foreign-accented
speakers tend to be evaluated more negatively than native-accented speakers is simply
because they are harder to understand (see also, Dovidio & Gluszek, 2012). Our nd-
ings suggest that this evaluative downgrading is likely to occur even in the absence of
negative stereotypes.
e uency- and stereotype-based accounts of the language attitudes process are
neither mutually exclusive nor contradictory. Rather, they complement one another
and represent two parallel mechanisms of the same process. Processing uency, just
like inferred group membership, is simply one of many cues to judgment. Sometimes
these cues may reinforce one another—for example,, an accent is dicult to process
andassociatedwithnegativestereotypes.Othertimes,thesecuesmaycontradictone
another—for example, an accent is dicult to process but associated with positive
stereotypes.
People’s reliance on uency as a cue to language attitudes— either in place of or in
addition to inferred group membership — is likely to vary from one situation to the
next, depending on several factors. First, processing diculties may be more likely to
have an eect on language attitudes when people perceive them as communicatively
signicant— that is, likely to impair their ability to carry out a conversation or execute
other interaction demands successfully. While noise had no eect on evaluations
of the SAE accent in the rst study, it did in the second study presumably because
listeners were explicitly made aware of the communicative consequences of their
processing diculties.
Human Communication Research (2016) © 2016 International Communication Association 19

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
Second, the nding that noise had an eect on both status and solidarity attribu-
tions in the rst study but only on status attributions in the second study suggests
that processing diculties may exert a stronger eect on status ratings. One possible
explanation for this is that people’s naïve theories about processing uency may be
more strongly linked to status than solidarity inferences; that is, when listeners expe-
rience diculty processing a person’s speech, they may be more likely to interpret
those diculties as indicative of the speaker’s lack of competence (i.e., status) than
lack of goodwill (i.e., solidarity).
ird, processing diculties may have a stronger eect on attitudes toward unfa-
miliar than familiar accents. When a speaker’s accent is unfamiliar, listeners cannot
make an accurate inference about the speaker’s social group membership and may
rely on overly inclusive categories (e.g., foreigner) (Lindemann, 2003; Ryan, 1983). In
such situations, inferred group membership becomes a relatively less informative cue
to evaluations, which may lead people to rely more on experiential information, such
as their uency experience. Future research should examine how (and whether) these
and other factors moderate the eects of processing uency on language attitudes.
Finally, past research has shown that uency has an eect on judgment only if it
is perceived to bear on the target of judgment (see Alter & Oppenheimer, 2009). If
uency is attributed to an incidental source irrelevant to the target of judgment, its
eects tend to be reduced or altogether eliminated (e.g., Oppenheimer, 2006). e
attributions people make about their uency-based experiences may, in part, be a
function of their expectations. People have stereotypes about how comprehensible
(i.e., uent) the speech of dierent linguistic groups is. For instance, people tend to
dichotomize native- and foreign-accented speakers as comprehensible and incompre-
hensible, respectively (Lindemann, 2005; Shuck, 2004, 2006). In other words, they
expect the speech of native-accented speakers to be easy to process and the speech
of foreign-accented speakers to be dicult to process.
Accordingly, when people experience diculty processing the speech of
foreign-accented speakers, they may be inclined to automatically attribute the
diculty internally to the speaker him or herself because it conrms their expecta-
tions. Such an internal attribution reinforces the informational value of the uency
cue (Alter & Oppenheimer, 2009). Given that people prefer to attribute an event to one
rather than multiple causes (Kelley, 1973), listeners are likely to (mis)attribute their
processing diculties to foreign-accented speakers even if theactualsourceofthe
diculty is due to an obvious incidental source unrelated to the target of judgment.
In contrast, when people experience diculty processing the speech of
native-accented speakers, they may be more inclined to attribute the diculty
externally to an incidental source unrelated to the speaker because it violates their
expectations.Suchanexternalattributionunderminestheinformationalvalueofthe
uency cue. Based on this, then, listeners’ processing uency may have a stronger
eect on their evaluations of foreign- than native-accented speakers in some sit-
uations. Future research should examine this possibility as well as more explicitly
explore the attributions listeners make about their uency experiences.
20 Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
In addition to the theoretical implications described above, ndings of the present
research also have important methodological implications; namely, researchers work-
ing in the language attitudes domain should becognizantofhowlisteners’processing
dynamics —due to the quality of listening conditions, questionnaire wording, or some
other source—can inuence their language attitudes. ey should also recognize the
possibilitythatmorenegativeevaluationsmaysometimesnot necessarily reect more
negative stereotypes but rather be the result of more dicult processing (cf. Dovidio
& Gluszek, 2012).
Similarly, a lack of evaluative dierences between two varieties should not imme-
diately be taken as evidence that the two varieties are associated with the same stereo-
types. It may be that one variety is associated with more negative stereotypes than the
other,yetisalsoeasiertoprocess—thatis,theuencyandinferredgroupmembership
cues contradict one another.
Moreover, assuming that the majority of past research has examined language atti-
tudesinwhatcanbecharacterizedas“good”(i.e.,highlyuent)listeningconditions
and that such listening conditions are atypicalofmostreal-worldsituations(cf.Pear-
sons et al., 1977), ndings of the present research suggest that past studies may have
actually underestimated theevaluativedowngradingthatspeakersofsomevarieties
face in most real-world situations.
Our ndings also have important practical implications. Most notably, results
of the present study suggest that, in addition to improving listeners’ stereotypes,
the evaluative downgrading that foreign-accented speakers oen face vis-à-vis
native-accented speakers could be minimized by increasing listeners’ processing u-
ency. is can be achieved in a number of ways. Most obviously, listening conditions
characterized by higher SNRs—either due to lower background noise or louder
speech— should result in easier processing and, thus, more favorable evaluations.
Additionally, past research has suggested that mere exposure to a given foreign accent
can facilitate later processing of that accent (e.g., Gass & Varonis, 1984).
isstudyhasseverallimitations.First,itexaminedtheeectsofprocessing
uency on only two language varieties. Future research should extend this nding
to other foreign accents as well as other nonstandard native varieties. Second, the
present study used background white noise to manipulate listeners’ processing
uency. Although this method has been used widely in past research (e.g., Clopper &
Bradlow, 2008; Sebastian et al., 1980), future research should examine whether other
uency manipulations (e.g., multitalker babble) yield similar eects.
Future research should also examine whether the eects of uency on lan-
guage attitudes dier based on context. Given that the present study was framed
as an impression formation task, participants had little motivation to attend to the
actual content of the speaker’s message. Future studies should examine whether the
eects of uency on speaker evaluations are more pronounced in contexts where
understanding the speaker is more consequential (e.g., lecture).
In sum, ndings of the present research provide compelling evidence of system-
atic eects of processing uency on listeners’ language attitudes by demonstrating that
Human Communication Research (2016) © 2016 International Communication Association 21

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
morenegativeattitudestowardagivenaccentcanbetriggeredsimplybythedi-
culty associated with processing speech in that accent. As such, they shed light on an
additional explanatory mechanism of the language attitudes process.
References
Adank, P., Evans, B. G., Stuart-Smith, J., & Scott, S. K. (2009). Comprehension of familiar and
unfamiliar native accents under adverse listening conditions. Journal of Experimental
Psychology: Human Perception and Performance,35, 520–529.
Alter, A. L., & Oppenheimer, D. M. (2009). Uniting the tribes of uency to form a
metacognitive nation. Personality and Social Psychology Review,13, 219–235.
Beukeboom, C. J. (2014). Mechanisms of linguistic bias: How words reect and maintain
stereotypic expectancies. In J. Laszlo, J. Forgas & O. Vincze (Eds.), Social cognition and
communication (pp. 313–330). New York, NY: Psychology Press.
Byrne, D. (1971). e attraction paradigm. New York, NY: Academic Press.
Byrne, D., & Clore, G. L. (1970). A reinforcement model of evaluation responses. Personality:
An International Journal,1, 103– 128.
Chand, V. (2009). [v]at is going on? Local and global ideologies about Indian English.
Language in Society,38, 393– 419.
Clopper, C. G., & Bradlow, A. R. (2008). Perception of dialect variation in noise: Intelligibility
and classication. Language and Speech,51, 175 – 198.
Clopper, C. G., & Pisoni, D. B. (2004). Some acoustic cues for the perceptual categorization of
American English regional dialects. Journal of Phonetics,32, 111– 140.
Cristia, A., Seidl, A., Vaughn, C., Schmale, R., Bradlow, A., & Floccia, C. (2012). Linguistic
processing of accented speech across the lifespan. Frontiers in Psychology,3, 1 –15.
Day, R. R. (1982). Children’s attitudes toward language. In E. B. Ryan & H. Giles (Eds.),
Attitudes toward language variation: Social and applied contexts (pp. 116– 131). London,
England: Edward Arnold.
Diener, E., & Seligman, M. E. P. (2002). Very happy people. Psychological Science,13, 81– 84.
Dovidio, J. F., & Gluszek, A. (2012). Accent, nonverbal behavior, and intergroup bias. In H.
Giles (Ed.), e handbook of intergroup communication (pp. 87– 99). New York, NY:
Routledge.
Dragojevic, M., & Giles, H. (2014). e reference frame eect: An intergroup perspective on
language attitudes. Human Communication Research,40, 91–111.
Dragojevic, M., Giles, H., & Watson, B. (2013). Language ideologies and language attitudes: A
foundational framework. In H. Giles & B. Watson (Eds.), e social meanings of language,
dialect, and accent: International perspectives on speech styles (pp.1–25).NewYork,NY:
Peter Lang.
Dragojevic, M., Mastro, D., Giles, H., & Sink, A. (in press). Silencing nonstandard speakers: A
content analysis of accent portrayals on American primetime television. Language in
Society.
Fiske, S. T., Cuddy, A. J. C., & Glick, P. (2007). Universal dimensions of social cognition:
Warmth an d c o m p e t en c e . Trends in Cognitive Sciences,11, 77– 83.
Floccia, C., Butler, J., Goslin, J., & Ellis, L. (2009). Regional and foreign accent processing in
English: Can listeners adapt? Journal of Psycholinguistic Research,38, 379 – 412.
Forgas, J. P. (1990). Aective inuences on individual and group judgments. European
JournalofSocialPsychology,20, 441– 453.
22 Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
Forgas, J. P. (1995). Mood and judgment: e aect infusion model. Psychological Bulletin,
117, 39–66.
Fuertes, J. N., Gottdiener, W., Martin, H., Gilbert, T. C., & Giles, H. (2012). A meta-analysis
oftheeectsofspeakers’accentsoninterpersonalevaluations.European Journal of Social
Psychology,42, 120 –133.
Garrett, P. (2010). Attitudes to language. Cambridge, England: University Press.
Gass, S., & Varonis, E. M. (1984). e eect of familiarity on the comprehensibility of
nonnative speech. Language Learning,34, 65–89.
Giles, H., Bourhis, R., & Taylor, D. (1977). Towards a theory of language in ethnic group
relations. In H. Giles (Ed.), Language, ethnicity and intergroup relations (pp. 307 –348).
London, England: Academic Press.
Giles, H., & Johnson, P. (1987). Ethnolinguistic identity theory: A social psychological
approach to language maintenance. International Journal of the Sociology of Language,68,
69–99.
Giles, H., & Watson, B. (Eds.) (2013). e social meanings of language, dialect, and accent:
International perspectives on speech styles.NewYork,NY:PeterLang.
Greifeneder, R., Bless, H., & Pham, M. T. (2011). When do people rely on aective and
cognitive feelings in judgment? A review. Personality and Social Psychology Review,15,
107–141.
Hayes, A. F. (2013). Introduction to mediation, moderation, and conditional process analysis: A
regression-based approach.NewYork,NY:GuilfordPress.
Hayes, A. F., & Preacher, K. J. (2014). Statistical mediation analysis with multicategorical
independent variables. British Journal of Mathematical and Statistical Psychology,67,
451–470.
Heaton, H., & Nygaard, L. C. (2011). Charm or harm: Eect of passage content on listener
attitudes toward American English accents. Journal of Language and Social Psychology,
30, 202–211.
Kelley, H. H. (1973). e process of causal attribution. American Psychologist,28, 107–128.
Lambert, W. E. (1967). A social psychology of bilingualism. JournalofSocialIssues,23,
91–109.
Lev-Ari, S., & Keysar, B. (2010). Why don’t we believe non-native speakers? e inuence of
accent on credibility. Journal of Experimental Social Psychology,46, 1093–1096.
Lindemann, S. (2003). Koreans, Chinese or Indians? Attitudes and ideologies about
non-native English speakers in the United States. Journal of Sociolinguistics,7, 348–364.
Lindemann, S. (2005). Who speaks “broken English”? US undergraduates’ perceptions of
non-native English. International Journal of Applied Linguistics,15, 187– 212.
Lippi-Green, R. (2012). English with an accent (Vol. ,2nd ed.). London, England: Routledge.
Luhman, R. (1990). Appalachian English stereotypes: Language attitudes in Kentucky.
Language in Society,19, 331– 348.
Marlow, M. L., & Giles, H. (2010). ‘We won’t get ahead speaking like that!’ Expressing and
managing language criticism in Hawai’i. Journal of Multilingual and Multicultural
Development,31, 237–251.
Milroy, J., & Milroy, L. (1999). Authority in language: Investigating Standard English (Vol. ,3rd
ed.).NewYork,NY:Routledge.
Munro, M. J. (1998). e eects of noise on the intelligibility of foreign-accented speech.
Studies in Second Language Acquisition,20, 139–154.
Human Communication Research (2016) © 2016 International Communication Association 23

Processing Fluency and Language Attitudes M. Dragojevic & H. Giles
Munro, M. J., & Derwing, T. M. (1995). Processing time, accent, and comprehensibility in the
perception of native and foreign-accented speech. Language and Speech,38, 289– 306.
Oppenheimer, D. M. (2006). Consequences of erudite vernacular utilized irrespective of
necessity: Problems with using long words needlessly. Applied Cognitive Psychology,20,
139–156.
Pearson, A. R., & Dovidio, J. F. (2014). Intergroup uency: How processing experiences shape
intergroup cognition and communication. In J. P. Forgas, O. Vincze & J. Laszlo (Eds.),
Social cognition and communication (pp. 101– 120). New York, NY: Psychology Press.
Pearsons, K. S., Bennett, R. L., & Fidell, S. (1977). Speech levels in various noise environments
(Report No. EPA-600/1-77-025). Washington, DC: U.S. Environmental Protection
Agency.
Powesland, P., & Giles, H. (1975). Persuasiveness and accent-message incompatibility.
Human Relations,28, 85–93.
Reber, R., & Schwarz, N. (1999). Eects of perceptual uency on judgments of truth.
Consciousness and Cognition,8, 338– 342.
Reber, R., Schwarz, N., & Winkielman, P. (2004). Processing uency and aesthetic pleasure: Is
beauty in the perceiver’s processing experience? Personality and Social Psychology Review,
8, 364–382.
Reber, R., Winkielman, P., & Schwarz, N. (1998). Eects of perceptual uency on aective
judgments. Psychological Science,9, 45–48.
Ryan, E. B. (1983). Social psychological mechanisms underlying native speaker evaluations of
non-native speech. Studies in Second Language Acquisition,5, 148–159.
Schwarz, N. (2004). Metacognitive experiences in consumer judgment and decision making.
Journal of Consumers Psychology,14, 332 – 348.
Schwarz, N. (2012). Feelings-as-information theory. In P. Van Lange, A. Kruglanski & E. T.
Higgins (Eds.), Handbook of theories of social psychology (pp. 289–308). ousand Oaks,
CA: Sage.
Schwarz, N., & Clore, G. L. (2007). Feelings and phenomenal experiences. In A. Kruglanski &
E. T. Higgins (Eds.), Social psychology: Handbook of basic principles (Vol. ,2nd ed., pp.
385–407). New York, NY: Guilford.
Sebastian, R. J., Ryan, E. B., Keogh, T., & Schmidt, A. (1980). e eects of negative aect
arousalonreactionstospeakers.InH.Giles,W.P.Robinson&P.M.Smith(Eds.),
Language: Social psychological perspectives (pp. 203–208). Oxford, England: Pergamon.
Sherbecoe, R. L., & Studebaker, G. A. (2004). Supplementary formulas and tables for
calculating and interconverting speech recognition scores in transformed arcsine units.
International Journal of Audiology,43, 442–448.
Shuck, G. (2004). Conversational performance and the poetic construction of ideology.
Language in Society,33, 195– 222.
Shuck, G. (2006). Racializing the nonnative English speaker. Journal of Language, Identity,
and Education,5, 259 –276.
St. Clair, R. N. (1982). From social history to language attitudes. In E. B. Ryan & H. Giles
(Eds.), Attitudes towards language variation: Social and applied contexts (pp. 164–174).
London, England: Edward Arnold.
Studebaker, G. A. (1985). A ‘rationalized’ arcsine transform. Journal of Speech and Hearing
Research,28, 455– 462.
24 Human Communication Research (2016) © 2016 International Communication Association

M. Dragojevic & H. Giles Processing Fluency and Language Attitudes
Van Engen, K. J., & Bradlow, A. R. (2007). Sentence recognition in native- and
foreign-language multi-talker background noise. Journal of the Acoustical Society of
America,121, 519–526.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures
of positive and negative aect: e PANAS scales. Journal and Personality and Social
Psychology,54, 1063 –1070.
Wendorf, C. A. (2004). Primer on multiple regression coding: Common forms and the
additional case of repeated contrasts. Understanding Statistics,3, 47 –57.
Whittlesea, B. W. A., Jacoby, L. L., & Girard, K. (1990). Illusions of immediate memory:
Evidence of an attributional basis for feelings of familiarity and perceptual quality.
Journal of Memory and Language,29, 716–732.
Winkielman, P., Schwarz, N., Fazendeiro, T. A., & Reber, R. (2003). e hedonic marking of
processinguency:Implicationsforevaluativejudgment.InJ.Musch&K.C.Klauer
(Eds.), e psychology of evaluation: Aective processes in cognition and emotion (pp.
195–225). Mahwah, NJ: Lawrence Erlbaum.
Woolard, K. A. (1985). Language variation and cultural hegemony: Toward an integration of
sociolinguistic and social theory. American Ethnologist,12, 738 –748.
Zahn, C. J., & Hopper, R. (1985). Measuring language attitudes: e speech evaluation
instrument. Journal of Language and Social Psychology,4, 113–123.
Human Communication Research (2016) © 2016 International Communication Association 25