The medium helps the message: Early sensitivity to auditory fluency in children’s endorsement of statements

Bernard, Stéphane; Proust, Joëlle; Clément, Fabrice

doi:10.3389/fpsyg.2014.01412

ORIGINAL RESEARCH article

Front. Psychol., 04 December 2014

Sec. Developmental Psychology

Volume 5 - 2014 | https://doi.org/10.3389/fpsyg.2014.01412

The medium helps the message: Early sensitivity to auditory fluency in children’s endorsement of statements

$\r\nStphane Bernard*$ Stéphane Bernard^1*

Joëlle Proust²

Fabrice Clément¹

¹Cognitive Science Centre, Neuchâtel, Switzerland
²Institut Jean Nicod, École Normale Supérieure, Paris, France

Recently, a growing number of studies have investigated the cues used by children to selectively accept testimony. In parallel, several studies with adults have shown that the fluency with which information is provided influences message evaluation: adults evaluate fluent information as more credible than dysfluent information. It is therefore plausible that the fluency of a message could also influence children’s endorsement of statements. Three experiments were designed to test this hypothesis with 3- to 5-year-olds where the auditory fluency of a message was manipulated by adding different levels of noise to recorded statements. The results show that 4 and 5-year-old children, but not 3-year-olds, are more likely to endorse a fluent statement than a dysfluent one. The present study constitutes a first attempt to show that fluency, i.e., ease of processing, is recruited as a cue to guide epistemic decision in children. An interpretation of the age difference based on the way cues are processed by younger children is suggested.

Introduction

Recently a growing number of studies in developmental psychology have underlined the importance of testimony in knowledge acquisition (e.g., Clément, 2010; Harris, 2012). Given the risk of being misguided by manipulative or misinformed sources (Sperber et al., 2010), one of the objectives of this line of research is to determine whether children are able to assess the comparative credibility of information provided by a speaker. A typical paradigm for testing endorsement of statements or of object labels involves contrasting two sources stating contradictory propositions and asking children to choose between the two statements (the conflicting sources task). So far, the main focus has been on properties of the source of the testimony, such as reliability (e.g., Koenig and Harris, 2005; Pasquini et al., 2007; Birch et al., 2008; Scofield and Behrend, 2008; Corriveau and Harris, 2009a), expertise (Sobel and Corriveau, 2010; Koenig and Jaswal, 2011) or the emotions displayed by informants (Clément et al., 2013). Little is known, however, about the influence of the message itself on testimony selection. Bernard et al. (2012) asked whether children would more readily follow an informant using the connective because to link an argument in a statement than an informant giving very similar information without the connective. They showed that while 4- and 5-year-olds used the connective because to guide their selective trust, 3-year-olds demonstrated indiscriminate selective trust. The same authors have recently shown that 3-year-olds could take into account certain informational properties of a message in selecting testimonies (Mercier et al., 2014). In fact, 3- to 5-year-olds preferred to endorse statements including a non-circular argument over a circular one (see also Corriveau and Kurkul, 2014, and for older children: Baum et al., 2008). The three experiments reported in this paper explore a cue embedded in the message that has not been yet investigated: the fluency of the communicated information.

Processing fluency can be defined as “the ease or difficulty with which new, external information can be processed” (Schwarz, 2004, p. 338). Fluency, or ease of processing, is what a person may experience when attempting to perceptually discriminate objects with a given property (perceptual fluency), retrieve items from episodic or semantic memory (retrieval fluency), or understand the communicated content (conceptual fluency; see Alter and Oppenheimer, 2009). Dysfluent information is information that is more difficult to process, requiring, for instance, a longer latency before any decision can be reached by adults (see Koch and Forgas, 2012). Several studies on adults support the prediction that fluent statements are more readily endorsed than dysfluent ones: fluent stimuli subjectively appear to be more credible, and are more frequently judged as probably true on a Likert scale than dysfluent ones (for a meta-analysis of this effect, see Dechêne et al., 2010). Regarding this effect, fluency has been, for instance, manipulated through repeated exposure to statements, which enhances the ease of processing (Johnston et al., 1985). A number of studies have shown that repeated statements were judged as more credible in comparison to statements that had not been presented before (e.g., Hasher et al., 1977; Begg et al., 1992; Brown and Nix, 1996). The effect of repeated exposure on credibility judgment could be explained as follows: processing a repeated statement is experienced as unexpectedly fluent. This unexpected fluency is experienced as discrepant from a comparison standard. Research has shown that an experienced discrepancy between expected and actual fluency of a stimulus generates feelings of familiarity (e.g., Whittlesea et al., 1990; Whittlesea and Williams, 1998, 2000; Whittlesea, 2004). These feelings involve the sense that something has been encountered before. They are generated by the higher fluency which previous exposure confers to perceptual and memorial processing of a given item. It has been hypothesized that these feelings of familiarity in turn affect credibility judgments because experiencing a stimulus as familiar could lead subjects to feel that they have seen or heard this stimulus before, suggesting that it is probably more credible (e.g., Reber and Schwarz, 1999; Weaver et al., 2007; Schwarz, 2010).

Arguably, any other variable that increases processing fluency might have the same epistemic effect as prior exposure, and in fact people make similar credibility judgments about statements that are fluently processed for other reasons than repetitive presentation. Studies have demonstrated that other instantiations of fluency, such as rhyme in aphorisms (McGlone and Tofighbakhsh, 2000) or high contrast in the visual presentation of statements (Reber and Schwarz, 1999; Koch and Forgas, 2012), also influence credibility judgments. For example, Reber and Schwarz (1999) manipulated the visual fluency of statements: various color fonts were used on a computer screen to make the statements more or less difficult to read against a white background. After each presentation, adult participants were asked if the statement presented (e.g., “Osorno is in Chile”) was true or false. The statements were more likely to be judged as true when presented in a highly visible color, i.e., in a fluent presentation.

The three experiments reported in this paper manipulated the auditory fluency of the message by adding different levels of noise to statements presented to children. Even though the effect of fluency on credibility with this type of auditory manipulation has not been previously tested in adults, research on memory has shown that this kind of manipulation generates different levels of fluency. For instance, in a study by Goldinger et al. (1999), adults were briefly presented with lists of words and the auditory fluency was manipulated by presenting the words with soft or loud background noise. Adults were then presented with a recognition test (recognition judgments). Results indicated that adults were more likely to consider a word as “old” (sense of familiarity) when it was presented with the soft background noise (fluent presentation) than with the loud background noise (dysfluent presentation; for other operationalizations of auditory fluency see Rhodes and Castel, 2009; Besken and Mulligan, 2013). From a more general perspective, it has been shown that fluency influences memory assessment. In particular, it has been established that a wide range of perceptual fluency affects judgments of learning (i.e., people’s estimations of how well they have learned something) and recognition judgments (e.g., Kelley and Rhodes, 2002; Kornell et al., 2011; McDonough and Gallo, 2012; Yue et al., 2013).

Given that adults judge a fluent statement to be more credible than a dysfluent one, one can expect that, in the absence of other cues, children should be more willing to endorse a statement that is easier to process. There are no studies directly addressing this question in the case of preschoolers, and, more generally, nothing is known about the possible influence of fluency on child development. If two conflicting statements, one fluent and the other dysfluent, are presented by two different informants (as in the classical paradigm used in research on testimony with children), children should selectively endorse the more fluent statement. This selectivity would be based only on the fact that higher fluency makes a statement more credible, as in adult studies. In the three experiments described below, the possible role of the auditory fluency of a message on endorsement of statements was investigated. Participants were as young as 3-years-old, an age at which it has been demonstrated that children are able to take part in the conflicting sources task (Koenig et al., 2004), and as old as 5, to see if there were any developmental differences. These possible developmental differences are investigated given that previous studies have reported developmental changes in testimony selection between 3- and 5-year-olds when the message itself was manipulated and not the properties of the informants (Bernard et al., 2012; Corriveau and Kurkul, 2014; Mercier et al., 2014).

Experiment 1

Method

Participants

This experiment involved 81 children: 27 3-year-olds (11 girls, Mage = 42.66 months, SD = 3.80, range 36–47 months), 26 4-year-olds (11 girls, Mage = 52.88 months, SD = 3.69, range 48–59 months), and 28 5-year-olds (15 girls, Mage = 67.03 months, SD = 3.52, range 61–71 months) from a school in Lyon (France). All the participants were French, and all the experiments were conducted in French. Most children were from middle- and upper-middle-class families. Each child was tested individually in a quiet room by a single experimenter for about 10 min. Informed parental consent was obtained for each child. All participants were treated according to the Declaration of Helsinki.

Materials and procedure

To test whether children are more likely to endorse a fluent than a dysfluent statement, a conflicting sources task was used. Three stories were presented to children on a computer screen. All stories were constructed on the same model: in the first vignette, a young Playmobil girl arrived at a place where two closed colored boxes are sitting on a table (places counterbalanced). This girl was carrying an object, and the experimenter explained that the girl was going to put this object in one of the two boxes. A second vignette showed the same girl without the object. The experimenter explained that she was leaving after having put the object in one of the boxes; he specified that he did not know which of the two boxes contained the object. Then, the experimenter showed a third vignette where two new female Playmobil characters were depicted (Figure 1).

FIGURE 1

FIGURE 1. Vignettes for the ball story.

The experimenter gave the names of the two characters and said that each one of them would help him find the object. Children wore headphones in order to listen to what the characters said.

For the first character, an animation bubble popped up and a voice recording was activated; the child heard, for instance, “The ball is in the orange box.” The same procedure was repeated with the second character, who stated, for instance, “The ball is in the pink box” (the characters’ locations, the order of bubble activation, the fluency of each sentence and the voice attributed to a character were all counterbalanced).

Finally, the experimenter asked the children, “According to you, where is the ball?” If the children did not respond, the experimenter asked: “Is the ball here (pointing at one of the boxes) or is the ball here (pointing at the other)?”

The two other stories were built on the same model. In each of them, the young girl, the object carried, the colors of the boxes and their places (right/left), the two female informants, their names and the voices associated with them were varied. The child could obtain a maximum score of 3 points (1 point for each story when the box linked to the more fluent statement was chosen).

After the three stories, children listened to each statement again to evaluate their comprehension—this was done by asking them to repeat each statement. This step was used to ensure that children’s choices were not linked to any difficulty of comprehension. All children succeeded at this comprehension test.

All sentences were recorded and modified using Audacity software. The sentences were digitized (32-bit resolution) at a 44.1-kHz sampling rate. For each story (N = 3), two different volunteers were asked to say the two sentences corresponding to the two locations in a neutral tone.

The sentences were normalized to produce an output level of 80 dB-A (A-weighted decibels: audible decibels measured with an artificial ear) and standardized to 2.1 s in duration. Two separate copies were created for each sentence and modified by adding Brownian noise to obtain two levels of auditory fluency: a fluent statement (amplitude of Brownian noise = 50 dB-A) and a dysfluent statement (amplitude of Brownian noise = 65 dB-A). Brownian noise (or red noise) is the kind of signal noise produced by Brownian motion (random process). The sound is a low roar resembling a waterfall and has a “soft” quality compared to white and pink noise.

To test whether these two levels of fluency were perceptually discriminated, 23 adults were recruited (13 women, Mage = 23.88 years, SD = 5.5 years, age range 19–44 years). The adults listened to the stimuli and rated the recording quality for each sentence on a scale from 1 (very good quality) to 7 (very bad quality). The results showed that the fluent statements (M = 2.39, SD = 0.89) were evaluated as having better auditory quality than the dysfluent statements [M = 5.26, SD = 1.17, t(44) = 9.32, p < 0.001].

Results

The percentage of choices elicited by fluent sentences was 44.4% for the 3-year-olds, 58.9% for the 4-year-olds, and 65.5% for the 5-year-olds (Figure 2). A 3 (age group: 3-year-olds, 4-year-olds, 5-year-olds) × 2 (gender: girl, boy) analysis of variance (ANOVA) with the score of fluent choices as a dependent variable yielded a significant main effect only for the age group factor, F(2,75) = 4.75, p < 0.05. The 5-year-olds chose the fluent statements significantly more¹ (M = 1.96, SD = 0.84) than the 3-year-olds (M = 1.33, SD = 0.78). 4-year-olds’ performance (M = 1.77, SD = 0.65) was not significantly different from that of the 3-year-olds or from that of the 5-year-olds.

FIGURE 2

FIGURE 2. Percentage of fluent choices in each Experiment and in each age group (the chance line is at 0.50). *p < 0.05, **p < 0.01, one-sample t-test.

However, children’s choices elicited by fluent sentences were significantly above chance for the 4-year-olds, t(25) = 2.11, p = 0.04, and 5-year-olds, t(27) = 2.93, p = 0.01, but not for the 3-year-olds, t(26) = -1.10, p = 0.28. These findings could be explained by the fact that the ANOVA did not take the chance level into account or test hierarchical relations between means (in this case, the ANOVA tested the null hypothesis M1 = M2 = M3). Following the difference between the groups in terms of chance and the percentages of choices elicited by fluent sentences, another hypothesis can be tested: the 3-year-olds choose the fluent statements less often than the 4-year-olds; and the performance of the 4-year-olds is similar to that of the 5-year-olds. To test this hypothesis more specifically, a contrast analysis was used (see e.g., Wendorf, 2004). This kind of analysis is more powerful for testing specific hierarchical hypotheses than classical analyses, and can be used even if the classical test is not significant (Brauer and McClelland, 2005). Two contrasts were tested in a regression analysis: a contrast of interest², corresponding to the previous hypothesis regarding the age difference, and an orthogonal contrast, which tested the residual variance. The hypothesis can be accepted if the contrast of interest predicts the choices of the fluent statements and if the orthogonal contrast does not. The contrast analysis confirmed this hypothesis: F(1,78) = 2.96, p = 0.004, for the contrast of interest, and F(1,78) = 0.94, p = 0.35, for the orthogonal contrast.

Discussion

This first experiment showed that 4- and 5-year-olds use auditory fluency to select which testimony to endorse. These results are the first to indicate that fluency could play a role in testimony selection by children and, from a wider perspective, could play an epistemic role in child development.

The fact that sensitivity to fluency is only elicited in 4- and 5-year-olds demands an explanation. First, it could be hypothesized that 3-year-olds do not discriminate between fluent and dysfluent sentences. In fact, anatomical data has shown that children do not reach the adult level of myelination for the auditory system until around age 4–5 (Kinney et al., 1988; Moore and Linthicum, 2007). Thus it is likely that 3-year-olds cannot use auditory fluency because their perceptual system does not provide sufficiently contrasted information about the quality of the signals. Given that Experiment 1 tested only how discrepancy in fluency was perceived by adults, we decided to include in a second experiment a measure to evaluate if the two levels of fluency manipulated were perceptually discriminated by 3-, 4-, and 5-year-old children.

The difference between the 3-year-olds’ results and those of older children could also be explained in another way: one could consider that children needed to reason about the knowledge of others in order to succeed in the conflicting sources task of Experiment 1. If this is the case, one should expect that this task required social cognitive abilities such as a theory of mind. Indeed, given that the ability to pass false belief tasks develops between 3 and 4 years of age (Wellman et al., 2001), it is plausible that the difference between 3-year-olds and older children was related to a difference in theory of mind development, even if the relation between theory of mind and testimony selection is a matter of debate (e.g., Proust, 2012; Lucas et al., 2013). In order to test this hypothesis, children were therefore presented with three false belief tasks in Experiment 2.

Finally, regarding the results of the older children, one could argue that the discrepancy in fluency may lead children to attribute characteristics such as familiarity or in-group membership to the informant who used the fluent statement. In fact, recent studies have shown that familiarity and group membership influence testimony selection in young children (e.g., Corriveau and Harris, 2009b; MacDonald et al., 2013). In order to ensure that children’s choices were not linked to a preference for the informant whose message was more fluent, a preference question was also introduced in Experiment 2.