Effects of Task Performance Expectancy Violations on Processing Fluency and Speaker Evaluations

Processing fluency refers to the subjective ease or difficulty people experience processing information, such as a person's speech (Schwarz, 2010). According to the fluency principle of language attitudes (FPLA; Dragojevic, 2020), the more easily listeners process a person's speech, the more favorably they evaluate that person, all else being equal. Americans typically experience more difficulty processing foreign-accented (FA) than native-accented (NA) speech, and this reduced fluency is one reason why they tend to evaluate FA speakers less favorably than NA speakers (Dragojevic & Goatley-Soan, 2022). In addition to speakers’ accents, other factors can also affect listeners’ fluency—and thereby their evaluations of speakers— such as the presence of background noise (Dragojevic & Giles, 2016) and subtitles (Dragojevic et al., 2017).

We propose that another factor which may influence listeners’ fluency and evaluations of speakers is listeners’ performance on tasks which require speech processing. Americans often expect communicative problems in interactions with FA speakers, and expect to perform relatively poorly on tasks which require them to process FA speech (Gluszek & Dovidio, 2010; Lindemann, 2002, 2005; Rubin & Smith, 1990; for a discussion, Lippi-Green, 2012), such as learning from an FA instructor, or following directions given by an FA speaker. Drawing on expectancy violations theory (EVT; Burgoon, 1993) and the FPLA (Dragojevic, 2020), we argue (and show) that violations of American listeners’ performance expectancies on such tasks can influence their evaluations of FA speakers, and that this effect is mediated by listeners’ retrospective fluency and affective reaction.

Task Performance Expectancies and Processing Fluency

According to EVT (Burgoon, 1993), people enter social interactions with various expectancies, or cognitions about what they anticipate will occur during the interaction. When interactions involve the completion of a specific task, people develop performance expectancies, or cognitions about their anticipated performance on the task. Such expectancies can be based on various factors, including one's fluency: The more fluency people experience performing a task, the better they expect to perform on that task (Rawson & Dunlosky, 2002; Reber et al., 2006). In line with this, Americans often expect to perform relatively poorly on tasks which require them to process FA speech, in part because they experience considerable difficulty processing it (Lippi-Green, 2012).

Upon completing a task, listeners’ performance expectancies may be confirmed (i.e., performance consistent with expectancies) or violated (i.e., performance inconsistent with expectancies). According to EVT, expectancy violations are arousing and prompt a cognitive appraisal of the violation as positive or negative, depending on the desirability of the outcome relative to the expectancy (Burgoon, 1993). Task success is typically a desirable outcome; consequently, better-than-expected performance constitutes a positive violation, whereas worse-than-expected task performance constitutes a negative violation.

We contend that performance expectancy violations can influence speaker evaluations through two key mechanisms (see Figure 1). First, expectancy violations are hedonically marked: Compared to expectancy confirmations, positive or negative violations tend to elicit more positive or negative affective reactions, respectively (Figure 1, path a; Bartholow et al., 2001; Hansen et al., 2017). These affective reactions can, in turn, bias listeners’ evaluations of speakers: More positive affective reactions promote more favorable speaker evaluations (Figure 1, path b; Forgas & Bower, 1987; Schwarz, 2010). Consistent with this claim, several studies have shown that affect mediates the effect of expectancy violations on speaker evaluations (Biernat et al., 1999; Dragojevic et al., 2019).

OPEN IN VIEWER

Figure 1.

Theoretical model depicting the effects of performance expectancy violations on speaker evaluations.

Second, expectancy violations can prompt people to engage in retrospective sensemaking in an attempt to explain the violation (Kramer, 2017). This process is likely adaptive, as it can inform future expectancies and potentially increase their accuracy (Gasiorek & Aune, 2021). Processing fluency is a subjective experience that is constructed online during task performance and can be revised following task completion. One reason Americans typically expect to perform relatively poorly on tasks involving FA speech is that they experience considerable disfluency processing it (Lippi-Green, 2012). We contend that violations of these performance expectancies could prompt listeners to revise their recollections of fluency in an attempt to explain the violation (cf. Gasiorek & Dragojevic, 2018). Namely, if listeners perform better or worse than they expected, they may conclude that the speaker was easier or harder to understand than they initially thought, which should elevate or depress (respectively) their retrospective experience of fluency during the task. Stated differently, compared to performance expectancy confirmations, positive or negative violations are likely to elicit higher or lower retrospective judgments of fluency, respectively (Figure 1, path c).

This variation in fluency can be consequential. According to the FPLA (Dragojevic, 2020), increases in listeners’ fluency can promote more favorable speaker evaluations via two routes. First, fluency can exert a direct effect on speaker evaluations through the application of naïve theories, or commonsense explanations for why a person's speech is easy or difficult to process (Figure 1, path d; Schwarz, 2010). Listeners disproportionately place the communicative burden on the speaker (Lippi-Green, 2012). Consequently, they may interpret the ease with which they process a speaker's message as indicative of the speaker's ability and/or willingness to communicate effectively. Accordingly, increased fluency can prompt listeners to rate the speaker as more competent and/or warm. People's naïve theories may also link their fluency experience to other outcomes. For instance, people may interpret fluency as indicative of psychological closeness (Alter & Oppenheimer, 2008). Consequently, increased fluency may prompt a greater sense of connection with the speaker (see also communication accommodation theory: Giles, 2016). Second, fluency can exert an indirect effect on speaker evaluations via affect (Roessel et al., 2019). Like expectancy violations, processing fluency is hedonically marked; increases in fluency elicit more positive affective reactions (Figure 1, path e; Schwarz, 2010). As noted earlier, these affective reactions can, in turn, bias listeners’ evaluations of speakers (Figure 1, path b; Forgas & Bower, 1987; Schwarz, 2010).

In sum, performance expectancy violations can influence speaker evaluations through two mechanisms. First, violations can elicit an affective reaction (Figure 1, path a), which can bias speaker evaluations (Figure 1, path b). Second, violations can prompt people to revise their recollections of fluency (Figure 1, path c), which can bias speaker evaluations directly through the application of naïve theories (Figure 1, path d), and indirectly through affect (Figure 1, path eb). Both mechanisms predict the same outcome: Compared to confirmations, positive or negative expectancy violations should elicit a more positive or negative affective reaction, higher or lower retrospective fluency, and more or less favorable speaker evaluations, respectively.

The Present Study

In this study, participants listened to a foreign-accented speaker reading a story, completed a 10-item memory test on the story's content, and indicated their expected performance on the test. They then received either positive, negative, or no performance feedback.

We anticipated that listeners would expect to perform relatively poorly (∼50% accuracy) on the test. Based on this, we constructed a positive feedback condition in which participants were told they correctly answered 8/10 questions; this was intended to positively violate listeners’ performance expectancies. Similarly, we constructed a negative feedback condition in which participants were told they correctly answered 2/10 questions; this was intended to negatively violate listeners’ performance expectancies. Other participants received no feedback. We reasoned that if participants were not provided with information about their performance on the test, they would, by default, assume that their expectancies were accurate (i.e., confirmed). Participants were randomly assigned to one of these three feedback conditions (i.e., positive, negative, or no feedback).

To the extent that positive performance feedback positively violates listeners’ expectancies, we expected the positive feedback condition to elicit (H_1a) a more positive affective reaction; (H_1b) higher fluency; (H_1c) a higher sense of connection with the speaker; and (H_1d) higher ratings of speaker competence and (H_1e) warmth, compared to the no feedback condition. Similarly, to the extent that negative performance feedback negatively violates listeners’ expectancies, we expected the negative feedback condition to elicit (H_2a) a more negative affective reaction; (H_2b) lower fluency; (H_2c) a lower sense of connection with the speaker; and (H_2d) lower ratings of speaker competence and (H_2e) warmth, compared to the no feedback condition. Consistent with the theoretical model depicted in Figure 1, we expected the effects of feedback on speaker evaluations to be mediated by (H_3a) affect; (H_3b) fluency; and (H_3c) sequentially by fluency and affect.

Method

Procedure and Measures

The study was conducted online. After providing informed consent, participants first completed a soundcheck item to ensure their computer audio was working properly. They then listened to the audio recording described earlier. The audio recording began playing automatically and participants could not advance to the next page until it finished nor replay it, ensuring that all participants listened to the recording once in its entirety. Participants then completed a memory test on the story's content, consisting of 10 multiple-choice questions, each with four response options. They then indicated how many questions (0–10) they thought they answered correctly, which served as our measure of expected performance. Participants were then randomly assigned to one of the three conditions described earlier (i.e., negative, positive, or no feedback) and received feedback on their performance accordingly. Participants were not told their actual performance on the test.

Participants then completed dependent measures (adapted from Dragojevic, 2020). They reported their affective reaction toward the speaker using a 100-point (1 = very negative; 100 = very positive) feeling thermometer; scores were rescaled by dividing by 10. Participants reported their processing fluency by indicating how easy to understand, clear, and comprehensible the speaker was (1 = not at all; 7 = very). They indicated their sense of connection with the speaker by selecting one of seven pairs of circles with varying degrees of overlap (ranging from no to nearly complete overlap; adapted from Aron et al., 1992). Participants then rated the speaker on five competence­- (i.e., competent, intelligent, educated, smart, successful) and five warmth-related traits (i.e., warm, friendly, nice, sociable, pleasant; 1 = not at all; 7 = very; see also Fiske et al., 2002). Items comprising each multi-item scale were averaged to form a composite score. Means, standard deviations, reliabilities, and zero-order correlations between all dependent variables appear in Table 1.

OPEN IN VIEWER

Table 1.

Means, Standard Deviations, Reliabilities, and Zero-Order Correlations Between Dependent Variables.

Variable	Items	α	M	SD	1	2	3	4
1. Affect	1	–	5.55	2.26
2. Fluency	3	.88	2.23	1.14	.59***
3. Connection	1	–	1.92	1.18	.43***	.58***
4. Competence	5	.95	4.16	1.31	.61***	.40***	.27***
5. Warmth	5	.92	4.55	1.28	.58***	.38***	.27***	.56***

* p < .05, ** p < .01, ***, p < .001.

As a check on our accent manipulation, participants indicated how strong, familiar, and similar to their own accent the speaker's accent sounded (1 = not at all; 7 = very), as well as where they thought the speaker was from via an open-ended question (i.e., “Where do you think the speaker is from?”). Finally, participants provided demographic information and were debriefed.

Results

Focal Analyses

Dependent measures were submitted to a series of one-way ANOVAs. Significant omnibus tests were followed by one-tailed planned contrasts. Omnibus test statistics, cell means, and standard deviations appear in Table 2.

OPEN IN VIEWER

Table 2.

Effects of Feedback on Dependent Variables.

	Negative Feedback	No Feedback	Positive Feedback	F(2,289)	η2
Affect	5.12a (2.17)	5.40a (2.20)	6.13b (2.31)	5.33**	.04
Fluency	2.09a (1.11)	1.93a (0.91)	2.66b (1.24)	11.88***	.08
Connection	1.79a (1.08)	1.66a (0.81)	2.33b (1.46)	9.38***	.06
Competence	4.03a (1.39)	4.09a (1.26)	4.36a (1.28)	1.81	.01
Warmth	4.32a (1.33)	4.50a (1.22)	4.84b (1.25)	4.29*	.03

Note. Means appear first, followed by standard deviation in parentheses. Within each row, means with different superscripts are significant, p < .05 (one-tailed).

* p < .05, ** p < .01, ***, p < .001.

Feedback influenced affect, fluency, sense of connection, and warmth ratings. Compared to no feedback, positive feedback elicited a more positive affective reaction, t(289) = 2.29, p = .01, d = .27 (H_1a); higher fluency, t(289) = 4.62, p < .001, d = .54 (H_1b); a stronger sense of connection with the speaker, t(289) = 4.07, p < .001, d = .48 (H_1c); and higher ratings of speaker warmth, t(289) = 1.91, p = .03, d = .22 (H_1d). Contrary to predictions, no significant differences emerged between the negative and no feedback conditions on affect (H_2a), fluency (H_2b), sense of connection (H_2c), or warmth (H_2d), ps > .31. Feedback did not influence competence ratings (H_1e, H_2e).

Mediation Analyses

To test for mediation, the model depicted in Figure 1 was specified in Mplus 8.5 (Muthén & Muthén, 1998–2017). Given that feedback influenced sense of connection and warmth, but not competence, perceptions, only the former two outcomes were included in the model. The model with corresponding path coefficients and fit statistics appears in Figure 2. All indirect effects appear in Table 3.

OPEN IN VIEWER

Figure 2.

Obtained path model depicting the effects of feedback on speaker evaluations.

OPEN IN VIEWER

Table 3.

Specific Indirect Effects of Feedback on Speaker Evaluations.

	Sense of Connection	Warmth
Positive Feedback
→ fluency →	.38 [.18, .65]	.05 [–.03, .15]
→ affect →	−.01 [−.06, .03]	–.04 [–.19, .14]
→ fluency → affect →	.06 [.01, .13]	.26 [.15, .40]
Negative Feedback
→ fluency →	.08 [–.06, .26]	.01 [–.01, .07]
→ affect →	–.03 [–.11, .01]	–.14 [–.32, .01]
→ fluency → affect →	.01 [–.01, .06]	.06 [–.03, .17]

Note. Unstandardized effect estimates appear first, followed by 95% confidence intervals (CIs) in brackets. CIs are based on 10,000 bootstrap resamples. Bolded effects are significant.

Positive feedback exerted an indirect effect on sense of connection via fluency (H3b) and sequentially via fluency and affect (H_3c), but not via affect alone (H_3a). All indirect effect of negative feedback on sense of connection were nonsignificant. Positive feedback exerted an indirect effect on warmth sequentially via fluency and affect (H_3c), but not via fluency (H_3b) or affect alone (H_3a). All indirect effects of negative feedback on warmth were nonsignificant.

Discussion

This study examined how task performance feedback influenced Americans’ attitudes toward FA speakers. As anticipated, positive feedback positively violated listeners’ task performance expectancies and, compared to the no feedback (i.e., expectancy confirmation) condition, elicited higher retrospective reports of fluency, a more positive affective reaction, and more favorable speaker evaluations. The effect of positive feedback on speaker evaluations was mediated by fluency and sequentially by fluency and affect. Also as anticipated, negative feedback negatively violated listeners’ expectancies, at least objectively. Contrary to predictions, however, no differences emerged on any dependent variables between the negative and no feedback conditions. One possibility is that, because of its relatively small magnitude, the negative violation was not significant or meaningful enough perceptually to produce detectable differences in the outcomes.

These findings have several theoretical implications. First, they lend insight into the mechanisms underlying the effects of expectancy violations on speaker evaluations (cf. Burgoon, 1993). Namely, our findings suggest that when listeners base their task performance expectancies on their fluency, violations of those expectancies can initiate a sensemaking process and prompt listeners to retrospectively revise their fluency judgments to bring them in line with their actual performance, which, in turn, can bias their evaluations of speakers. Second, our findings lend further support to the FPLA (Dragojevic, 2020) by showing that factors which influence listeners’ fluency—in this case, task performance expectancy violations—can also have corresponding effects on listeners’ evaluations of speakers. Third, and related, our findings provide further insight into why fluency influences speaker evaluations. Consistent with the FPLA (Dragojevic, 2020), our results indicate that fluency can exert an indirect effect on speaker evaluations through affect: Increases in listeners’ fluency elicit a more positive affective reaction, which, in turn, positively biases their evaluations of speakers (Roessel et al. 2019). Our findings also show that fluency can itself be a cue to speaker evaluations through the application of naïve theories. In this study, fluency exerted a direct positive effect on listeners’ sense of connection with the speaker, arguably because listeners interpreted their increased fluency as indicative of increased psychological closeness (Alter & Oppenheimer, 2008).

Our findings also have practical implications. In the US, FA speakers tend to be evaluatively downgraded relative to NA speakers and often face prejudice and discrimination (Roessel et al., 2020). Factors which increase listeners’ fluency have the potential to attenuate some of this evaluative downgrading (Dragojevic, 2020); our findings suggest that task performance may be one such factor. Participants in this study significantly underestimated their actual task performance, and we suspect that people often underestimate their performance on tasks which require them to process FA speech. If people are made aware of their actual performance (e.g., students receive a grade on an exam testing their comprehension of content delivered by a foreign-accented instructor) and it is better than their expected performance—as we suspect it often is for tasks involving FA speech processing—this positive expectancy violation could increase listeners’ retrospective fluency and prompt more favorable speaker evaluations.

This study has several limitations. First, our findings are based on a single experiment, using a single speaker and foreign accent. Theoretically these effects should generalize to other speakers and other foreign (and native) accents; however, this remains an empirical question. Second, our study was limited to a single, arguably low-stakes context, where listeners’ actual performance on the task was inconsequential. Had listeners’ actual task performance been more consequential (i.e., associated with valued symbolic or material outcomes), the observed effects might have been more pronounced. Third, we did not assess listeners’ perceptions of expectancy violations, but rather assumed, based on objective criteria, that participants interpreted the feedback manipulations as intended; future studies should assess listeners’ perceptions of expectancy violations more directly.

In sum, our results demonstrate that violations of listeners’ task performance expectancies can influence how they evaluate speakers and that this effect is mediated by listeners’ retrospective judgments of their fluency during the task, and the affective reaction those judgments engender.