RESULTS AND DISCUSSION

For each child, we summed the number of training and practice blocks as an index of initial difficulty with the task. As can be seen in Figure 3, younger children required more blocks than did older children, indicating greater initial difficulty in differentiating questions from statements. Data from the test phase were analyzed by converting responses to d´ (d-prime) scores using proportions of hits and false alarms. Whereas d´ scores provide an index of listeners’ sensitivity to question intonation (i.e. the issue of interest in the present investigation), percent correct scores reflect bias as well as sensitivity. For the present purposes, question responses to question stimuli constituted hits, and question responses to statement stimuli constituted false alarms. A score of d´ = 1 corresponds to 69% correct.

Fig. 3.

Number of training and practice blocks in Experiment 1 (natural utterances) and Experiment 2 (low-pass filtered utterances) as a function of age group. Error bars are standard errors.

The stimulus set consisted of forty questions and forty statements, which meant that the maximum number of hits or false alarms was forty. Because d´ scores cannot be computed when the proportion of hits is 1·0 or the proportion of false alarms is 0 (i.e. statistically infinite scores), proportions of hits and false alarms were calculated by adding 0·5 to the number of hits and also to the number of false alarms and dividing those numbers by 41 (total possible hits or false alarms + 1), as in previous developmental studies (e.g. Thorpe, Trehub, Morrongiello & Bull, Reference Thorpe, Trehub, Morrongiello and Bull1988). These proportions were converted to z-scores and then to d´ scores (d´ = z[hit rate] – z[false-alarm rate]). The maximum d´ score was 4·5. Raw data (percent correct) and standard errors are illustrated in Figure 4 separately for each age group.

Fig. 4.

Performance in the test sessions of Experiment 1 (natural utterances) and Experiment 2 (low-pass filtered utterances) as a function of age group. Error bars are standard errors.

One-sample t-tests conducted separately for each age group revealed that performance was significantly better than chance (i.e. chance or d´ = 0 results from an equal number of hits and false alarms) for each age group (ps < .001). A one-way analysis of variance (ANOVA), with age (5–6, 7–8, 9–10, and adults) as the between-subjects variable and d´ as the dependent variable, revealed a significant effect of age (F(3,118) = 35·78, p < .001, η ² = ·48). Follow-up pairwise comparisons (Tukey HSD) revealed that the performance of five- to six-year-olds was significantly poorer than all other age groups (ps < .001), and that seven- to eight-year-olds performed more poorly than adults (p = ·001). The performance of nine- to ten-year-olds did not differ significantly from adults or from seven- to eight-year-olds (ps > .1).

We also examined the percentage of individuals in each group whose correct responses (i.e. correct identification of questions and statements) exceeded chance levels. According to the normal approximation to the binomial test (one-tailed, correcting for continuity), 48 or more correct responses out of 80 is significantly better than chance. Only 63% (19 of 30) of five- and six-year-olds obtained a score of 48 or more, in contrast to 100% in each of the three older groups. Fisher's exact tests confirmed that the proportion of individuals performing at chance levels was significantly higher in the youngest group of children compared to any other group (ps < .001).

We then asked whether participants had a response set, specifically a bias to respond ‘telling’ (or ‘statement’) more or less often than ‘asking’ (or ‘question’). For each participant, we calculated the total number of ‘telling’ responses. Comparisons with 50% (40 of 80) revealed that the youngest children had no response bias (p > .5), but each of the three older groups tended to respond ‘telling’ more than half of the time (ps < .03), presumably because of the syntax of the stimulus sentences. Nevertheless, the mean number of ‘telling’ responses was under 42 in each instance.

The variance in the number of ‘telling’ responses was considerably higher for the youngest children (SD = 15·79) compared to the other three groups (all SDs < 4·10), which motivated us to examine the number of participants who exhibited a bias to respond either ‘telling’ or ‘asking.’ Using the criterion described above (i.e. 48 or more ‘telling’ responses, or 48 or more ‘asking’ responses), we identified 11 participants who exhibited such a bias: 10 five- to six-year-olds and 1 seven- to eight-year-old. Fisher's exact tests confirmed that this bias was greater for the youngest group of children than for any of the other three groups (ps < .003).

Finally, we examined whether children who required more training and practice blocks performed more poorly in the test phase than those who required fewer blocks. Positive skewness of the training/practice variable prompted the use of a Spearman's correlation. Test scores were negatively correlated with number of training and practice blocks (r _s(n = 92) = –0·48, p < .001).

In sum, the results indicated that children and adults identified questions based on prosodic information alone. Nevertheless, five- to six-year-old children performed more poorly than older children and adults. In fact, 37% of children in the youngest age group performed at chance levels, but no participant in any other group did so, and one-third of the youngest children tended to respond consistently with ‘telling’ or ‘asking’ regardless of the stimuli. The performance of seven- to eight-year-old children was also less accurate than that of adults. Finally, children who required more training and practice trials had poorer outcomes on test trials.

In light of young children's propensity to focus on irrelevant verbal content or situational context when judging a speaker's feelings (Aguert et al., Reference Aguert, Laval, Bigot and Bernicot2010; Friend, Reference Friend2000; Morton & Trehub, Reference Morton and Trehub2001), the irrelevant verbal content of declarative questions may have distracted them from the critical prosodic cues. Just as young children achieve greater success in identifying a speaker's feelings when the verbal content is obscured (Morton & Trehub, Reference Morton and Trehub2001), they may achieve greater success in identifying declarative questions when the verbal content is unintelligible. This possibility was examined in Experiment 2.

RESULTS AND DISCUSSION

As shown in Figure 3, children required more training and practice blocks in the present experiment (M = 3·40, SD = 1·49) than in Experiment 1 (M = 2·91, SD = 1·33) (t(187·55) = 2·37, p = ·019) (unequal variances test). The data from the test phase were converted to d´ scores, as in Experiment 1. Raw data (percent correct) are illustrated in Figure 4. One-sample t-tests for each age group revealed that performance was significantly better than chance in all cases (ps < .001). An ANOVA with age (5–6, 7–8, 9–10, and adults) as the between-subjects variable and d´ as the dependent variable revealed a significant effect of age (F(3,122) = 33·13, p < .001, η ² = ·45). The performance of five- to six-year-olds was significantly worse than that of all other age groups (ps < .001), and the performance of seven- to eight-year-olds was worse than that of nine- to ten-year-olds (p = ·038) and adults (p < .001). The performance of nine- to ten-year-olds and adults did not differ (p > .2).

We then examined the percentage of individuals in each group whose correct responses exceeded chance levels (48 or more correct responses). Because performance was not as consistently good as it was in Experiment 1, we used a 4 (age [5–6, 7–8, 9–10, adults]) by 2 (above chance, chance) chi-square test of independence to examine age-related differences in the percentage of individuals who successfully distinguished questions from statements (all cells had expected frequencies > 5). Successful performance varied across age groups (χ ²(3, N = 126) = 54·70, p < .001, Φ = ·66). All adults performed above chance levels, as did all but one of the nine- to ten-year-olds, and all but three of the seven- to eight-year-olds. For the five- to six-year-olds, far fewer children – just slightly more than a third (36%, 12 of 33) – exceeded chance levels of performance.

Performance on the natural utterances from Experiment 1 and the low-pass filtered utterances from the present experiment was compared by means of a two-way ANOVA with stimulus type (natural, filtered) and age group (5–6, 7–8, 9–10, adults) as between-subjects variables and d´ as the dependent variable. There were main effects of age (F(3,240) = 67·92, p < .001, η ² = ·43) and stimulus type (F(1,240) = 26·86, p < .001, η ² = ·06). Overall, performance improved with age, and participants could more easily identify natural rather than low-pass filtered questions and statements. The interaction between age and stimulus type was not significant (F < 1). In other words, the decrement in performance for filtered compared to natural stimuli was similar across age groups.

If the unusual acoustic quality resulting from low-pass filtering contributed to the unexpected reduction in performance, then performance may have improved from the first to the second half of the test session, after listeners adapted to the spectral degradation. Because individual participants had different numbers of questions and statements in the first and second half due to randomization of order, we analyzed the number of correct responses rather than d´ scores. A mixed-design ANOVA, with age group (5–6, 7–8, 9–10, adults) as the between-subjects variable and test phase (first or second half) as the within-subjects variable revealed a main effect of age (F(3,122) = 37·14, p < .001, η _p ² = ·48), reflecting age-related improvement, and a small but significant effect of test phase (F(1,122) = 8·40, p = ·004, η _p ² = ·06). Performance was better during the second half of the test session (M = 33·63, SD = 7·49) than in the first half (M = 32·62, SD = 7·73), but the improvement represented only one additional correct answer on forty trials. There was no interaction between age group and test phase (F < 1), which indicated that all age groups had comparable adaptation to the low-pass filtered stimuli. In the second half of trials, overall performance with filtered stimuli (84% correct) remained substantially below overall performance with natural stimuli in Experiment 1 (91%) (t(237·04) = 3·19. p = ·002) (unequal variances test).

Examination of a possible response set (i.e. responding ‘telling’ more than 50% of the time) revealed no such bias among the three groups of children (ps > .1), but a small bias for adults (p = ·011) (M = 40·93). Categorization of participants into those with or without a systematic bias to respond either ‘telling’ or ‘asking’ revealed 21 participants with such a bias: 14 five- to six-year-olds, 6 seven- to eight-year-olds, and 1 nine- to ten-year-old. The proportion of participants with this bias varied reliably across age groups (χ ²(3, N = 126) = 25·42, p < .001, Φ = ·45).

As in Experiment 1, we examined the relation between number of training and practice blocks and subsequent test scores for the child participants. A Spearman's correlational analysis revealed a significant negative correlation between test scores and number of training and practice blocks (r _s(n = 98) = –0·50 p < .001).

In sum, children differentiated questions from statements in low-pass filtered utterances at better than chance levels, with the two oldest age groups (9–10, adults) performing significantly better than the two youngest (5–6, 7–8). Contrary to expectations, all age groups performed worse on the filtered utterances than on the original versions, which indicates that young children's difficulty in differentiating question from statement intonation cannot be attributed to interference from the verbal content. As in Experiment 1, a substantial portion of children in the youngest group tended to respond ‘telling’ or ‘asking’ in general, and children who required more training and practice trials tended to perform poorly in the actual test session.

GENERAL DISCUSSION

The present study examined the identification of declarative questions and statements by English-speaking children (five to ten years of age) and adults. Age-related differences in performance were similar for natural utterances (Experiment 1) and low-pass filtered utterances (Experiment 2), but performance in general was poorer for the filtered utterances. Although five- and six-year-olds performed above chance levels, they performed more poorly than all other age groups despite having numerous training and practice trials and continuous feedback about response accuracy throughout the test session. The seven- and eight-year-olds performed no differently than nine- and ten-year-olds on natural utterances but they performed more poorly on filtered utterances. Finally, the nine- and ten-year-olds performed no differently than adults.

The findings from the five- and six-year-olds are consistent with limited knowledge of the relevant intonation categories and, consequently, poor mapping between intonational contours and communicative functions. These results are in line with preschool children's challenges in linking cartoon characters with rising or falling melodic sequences (Creel, Reference Creel2014) and six-year-olds’ difficulty with conventional verbal labels for rising and falling pitch (Costa-Giomi & Descombe, Reference Costa-Giomi and Descombes1996).

It is likely that young children's difficulties were exacerbated by limitations in attention allocation and working memory (Cowan, Morey, AuBuchon, Zwilling & Gilchrist, Reference Cowan, Morey, AuBuchon, Zwilling and Gilchrist2010). To succeed on the present task, children had to focus on the terminal pitch contour, determine if it was rising, and designate it as a question if it was or as a statement otherwise. Young children's habitual focus on lexical cues at the expense of prosodic cues (Morton & Trehub, Reference Morton and Trehub2001; Snedeker & Trueswell, Reference Snedeker and Trueswell2004) and limited cognitive flexibility (Munakata, Snyder & Chatham, Reference Munakata, Snyder and Chatham2012), even in the face of continuous feedback, may have interfered with consistent allocation of attention to the relevant cues in Experiment 1, which featured natural utterances.

Although all participants were required to meet the same training criterion before proceeding to the test phase, they did not achieve comparable understanding, as reflected in persistent performance differences in the test phase. In fact, the number of trials required to meet the training criterion – one index of initial comprehension – was predictive of subsequent comprehension, as indexed by performance in the test phase, which also featured feedback on every trial. Fully 37% of the five- to six-year-olds failed to identify utterance type, as reflected in their chance-level performance.

Low-pass filtering that made the verbal content unintelligible was expected to facilitate young children's attention to the relevant acoustic cues, in line with previous studies of emotional prosody (Morton & Trehub, Reference Morton and Trehub2001). Instead, such filtering had the opposite effect, reducing performance comparably for all age groups. Low-pass filtering decreases the salience of the component pitches that are relevant for differentiating statements from questions (Cruttenden, Reference Cruttenden1981; Eady & Cooper, Reference Eady and Cooper1986; Gårding & Abramson, Reference Gårding and Abramson1965; Peng et al., Reference Peng, Lu and Chatterjee2009; Studdert-Kennedy & Hadding, Reference Studdert-Kennedy and Hadding1973). Young children's difficulty with the filtered utterances confirmed that their problems with the unfiltered utterances in Experiment 1 did not stem from lexical biases. Although 63% of the five- and six-year-old children performed above chance levels on the natural utterances, only 36% performed above chance on the filtered utterances. Presumably, their problems with categorizing pitch contours were exacerbated by decreased salience of the relevant cues.

Low-pass filtering preserves the pitch directional differences of the original utterances, but the resulting speech is perceived as being reduced in pitch range (i.e. compressed pitch contours) and pitch variability relative to unfiltered speech (Scherer, Koivumaki & Rosenthal, Reference Scherer, Koivumaki and Rosenthal1972; van Bezooijen & Boves, Reference van Bezooijen and Boves1986). It is likely, then, that reduced pitch salience in the filtered utterances contributed to the uniform reduction in performance across age.

The unusual sound quality of the filtered speech could have impaired performance even though listeners adapt to distorted speech after limited exposure (Hervais-Adelman, Davis, Johnsrude & Carlyon, Reference Hervais-Adelman, Davis, Johnsrude and Carlyon2008; Hervais-Adelman, Davis, Johnsrude, Taylor & Carlyon, Reference Hervais-Adelman, Davis, Johnsrude, Taylor and Carlyon2011). Indeed, performance on the filtered utterances improved modestly (i.e. one additional item correct) during the second half of the test session although it remained below the levels achieved with unfiltered stimuli. Presumably, adaptation to the filtered speech began during the training phase and continued during the test phase. Further exposure, either during the training phase or in a subsequent session, could result in performance levels approaching those attained with unfiltered utterances.

If young children have difficulty discerning the communicative intent of declarative questions, one would expect caregivers to avoid such questions. There is evidence, however, that parents make regular use of declarative questions in their conversations with very young children (Estigarribia, Reference Estigarribia2010). One might further expect the acoustic cues to the question/statement distinction to be exaggerated in child-directed speech, as are other cues to the speaker's intentions (Foulkes, Docherty & Watt, Reference Foulkes, Docherty and Watt2005; Jacobson, Boersma, Fields & Olson, Reference Jacobson, Boersma, Fields and Olson1983). Moreover, declarative questions in parent–child discourse are likely to be restricted to face-to-face contexts that provide supplementary visual cues such as raised eyebrows and head movements (Ekman, Reference Ekman1976, Reference Ekman, von Cranach, Foppa, Lepenies and Ploog1979; Srinivasan & Massaro, Reference Srinivasan and Massaro2003). Most importantly, such questions would not occur in isolation, as in the present study. Instead, parents would adhere to the contextual constraints on declarative questions (Gunlogson, Reference Gunlogson2003), which would highlight their communicative intentions. For example, a child's request for a cookie just before dinner might receive an incredulous reply such as “You want a cookie?” Despite the limitations of five- and six-year-olds, as observed in the present study, it is likely that they would comprehend the gist of the message.

In conclusion, children between five and six years of age have considerably greater difficulty than older children and adults in differentiating isolated declarative questions from statements on the basis of intonation alone. We contend that this difficulty stems primarily from absent or unstable intonational categories and secondarily from immature working memory (Cowan et al., Reference Cowan, Morey, AuBuchon, Zwilling and Gilchrist2010), which precluded successful mapping of terminal pitch contours onto communicative functions. An important challenge for future research is to ascertain the factors that support children's transition from continuous representations of pitch contours to categorical representations of rising and falling contour.