A profile of prosodic speech differences in individuals with autism spectrum disorder and first-degree relatives

Abstract

Background:

Impairments in prosody (e.g., intonation, stress) are among the most notable communication characteristics of individuals with autism spectrum disorder (ASD) and can significantly impact communicative interactions. Evidence suggests that differences in prosody may be evident among first-degree relatives of autistic individuals, indicating that genetic liability to ASD is expressed through prosodic variation, along with subclinical traits referred to as the broad autism phenotype (BAP). This study aimed to further characterize prosodic profiles associated with ASD and the BAP to better understand the clinical and etiologic significance of prosodic differences.

Method:

Autistic individuals, their parents, and respective control groups completed the Profiling Elements of Prosody in Speech-Communication (PEPS-C), an assessment of receptive and expressive prosody. Responses to expressive subtests were further examined using acoustic analyses. Relationships between PEPS-C performance, acoustic measurements, and pragmatic language ability in conversation were assessed to understand how differences in prosody might contribute to broader ASD-related pragmatic profiles.

Results:

In ASD, receptive prosody deficits were observed in contrastive stress. With regard to expressive prosody, both the ASD and ASD Parent groups exhibited reduced accuracy in imitation, lexical stress, and contrastive stress expression compared to respective control groups, though no acoustic differences were noted. In ASD and Control groups, lower accuracy across several PEPS-C subtests and acoustic measurements related to increased pragmatic language violations. In parents, acoustic measurements were tied to broader pragmatic language and personality traits of the BAP.

Conclusion:

Overlapping areas of expressive prosody differences were identified in ASD and parents, providing evidence that prosody is an important language-related ability that may be impacted by genetic risk of ASD.

1. Introduction

Impaired use of prosody (e.g., intonation and volume modulation, stress, rhythm) is a hallmark feature of the speech of individuals with ASD¹, dating back to Kanner’s original description of ASD (Kanner, 1943; Mesibov, 1992; Van Bourgondien & Woods, 1992; Wells & Peppé, 2003). Prosody plays a significant role in social communication by conveying information related to affective state, pragmatic intent (e.g., sarcasm, questioning), and word boundaries (e.g., chunking) (Wells & Peppé, 2003). Differences in prosody are often one of the first noticeable traits when interacting with an autistic person, which can impact communicative success with peers (Bent et al., 2016; Shriberg et al., 2001). Specific areas of difficulty include stress, speech rate, affect expression, and intonation modulation (Bone et al., 2012; Macdonald et al., 1989; McCaleb & Prizant, 1985; McCann & Peppé, 2003; Szatmari et al., 1989). Difficulties in these key aspects of prosody may impact an individual’s ability to clarify a message, convey a preference or choice, align affective tone to facial expressions, and appropriately vary vocal register for different social functions (Diehl & Paul, 2012; Filipe et al., 2014; Peppé et al., 2007). For instance, if one says “Mom went to the store” with a rising intonation pattern, rather than the expected falling intonation pattern, the statement may be perceived as a question. Patterns of subtle prosodic differences have also been noted among parents of individuals on the autism spectrum (Landa et al., 1992; Losh et al., 2008; Losh et al., 2012; Patel et al., 2020; Piven et al., 1997), suggesting that prosodic abilities may be impacted by genetic liability to ASD, even among clinically unaffected relatives.

Though prosody is a well-documented area of impairment in ASD, findings suggest substantial heterogeneity in the prosodic differences observed across the spectrum. Additionally, as noted by others (McCann & Peppé, 2003), many early studies of prosody in autistic individuals included relatively small sample sizes (e.g., typically fewer than 30 participants in each clinical group) and wide age ranges across studies, and sometimes an absence of control groups. Changes in ASD diagnostic criteria over time and lack of standardized assessments used to characterize prosody have also complicated efforts to identify specific, core prosodic impairments in ASD. Many studies of prosody in ASD have relied exclusively on trained listeners’ ratings of autistic people’s speech, with fewer studies examining acoustic measurements of prosody or receptive prosodic skills. Although listeners’ ratings can provide valuable information about general aspects of prosody that may contribute to a sense of atypicality, objective acoustic measures are needed to reliably and objectively characterize prosodic differences, and also to understand what specific features of speech (e.g., fundamental frequency, intensity, jitter, shimmer) are impacted in ASD. To this end, several studies have utilized a standardized prosodic assessment, the Profiling Elements of Prosody in Speech-Communication (PEPS-C; McCann and Peppé 2003), which yields measurement of receptive prosodic abilities, listener ratings (from trained raters) of expressive prosody, and also provides data that can be subjected to objective measurement of acoustic properties of prosody.

Studies using the PEPS-C’s listener rating scoring have demonstrated increased difficulty with prosodic discrimination (i.e., determining if auditory stimuli are produced with the same or different prosody), as well as difficulty using prosodic cues to identify phrasal boundaries, affect, turn-end type (i.e., understanding intonation of a question versus statement), and focus (i.e., contrastive stress) among individuals with ASD (Diehl & Paul, 2012, 2013; Järvinen-Pasley et al., 2008; McCann et al., 2007; Peppé et al., 2007). Poorer expression of phrasal boundaries, affect, contrastive stress, lexical stress, turn-end intonation, and prosodic imitation were also reported from studies using the PEPS-C (Diehl & Paul, 2012; Hesling et al., 2010; McCann et al., 2007; Peppé et al., 2007).

With regard to acoustic prosodic differences, autistic individuals were reported to display a wider fundamental frequency (F0) range during narrative production (Diehl et al., 2009; Nadig & Shaw, 2012), which was replicated in a separate sample on the contrastive stress expression subtest of the PEPS-C (Diehl & Paul, 2012, 2013). Additionally, on the PEPS-C, individuals with ASD produced utterances with overall longer duration compared to controls when expressing negative affect, turn-end, and imitation of prosody of single words (Diehl & Paul, 2012, 2013). One study investigated the extent to which specific acoustic prosodic differences in speech of autistic individuals during a narrative task mapped to raters’ perceived differences in conveyed meaning, and found that increased F0 range and decreased speech rate were correlated with greater differences in intonation, as well as pragmatic difficulties with presupposition during conversation (e.g., failure to provide appropriate background information or signal humor) (Patel et al., 2020). These important findings begin to identify the acoustic features that may contribute to perceived prosodic and more overarching pragmatic language differences in ASD. This study attempted to build on these findings through a comprehensive analysis of prosodic ability, using the PEPS-C (including application of acoustic analyses to speech samples) and directly examining relationships with ASD symptomatology and pragmatic language ability. The study additionally considered such questions in parents of autistic individuals to understand how potentially subtle prosodic profiles might include heritable traits that mark genetic risk of ASD.

1.1. Prosodic differences in first-degree relatives

Studies of first-degree relatives have proven to be a powerful approach in informing the underlying biology impacting subclinical expression of heritable features in various disorders/conditions (Almasy et al., 1999; Gottesman & Shields, 1972; Gur et al., 2007; Mitchell et al., 1999; Williams et al., 2010), including ASD (Bolton et al., 1994; Frazier et al., 2015; Piven et al., 1997). A growing body of research has identified differences among first-degree relatives of individuals with ASD across domains, such as social cognition (Adolphs et al., 2008; Baron-Cohen & Hammer, 1997; Billeci et al., 2016; Losh et al., 2009; Losh & Piven, 2007; Palmen et al., 2005; Sasson et al., 2013; Yucel et al., 2015), visual attention patterns during language production tasks (Hogan-Brown et al., 2014; Nayar et al., 2018), and broader differences in pragmatic skills (Landa et al., 1992; Losh et al., 2008; Losh et al., 2012). The collection of subclinical traits noted among clinically unaffected relatives of individuals with ASD are referred to as the broad autism phenotype (BAP) and may also include prosody (Bailey et al., 1995; Bolton et al., 1994; Landa et al., 1992; Losh et al., 2008; Piven et al., 1997). Ratings from trained listeners have identified increased or decreased intonation variability (e.g., sing-songy or monotone, respectively), and differences in rhythm and volume modulation among parents of autistic individuals. A recent study examining acoustic properties of speech during narratives in parents of individuals with ASD found that a subset of parents demonstrated an overall slower speech rate with increasing utterance length, as well as a reduced mean and increased range of F0 (Patel et al., 2020). As in ASD, these acoustic prosodic features (specifically, decreased F0 range and increased variability between syllable duration (a measure of speech rhythm)) were related to increased pragmatic language violations, suggesting that prosodic differences may play an important role in the broader pragmatic language profiles associated with ASD and the BAP. Although some global differences in prosodic expression have been noted in parents of individuals with ASD, prior work has not yet discerned how specific functions of prosody (e.g., contrastive stress, boundary expression) may differ in parents.

Differences in audio-vocal integration may be one source of prosodic differences in autistic individuals and first-degree relatives. For instance, auditory information in loud settings is typically accommodated through a compensatory tendency to increase one’s volume to be better heard by others, and to better hear oneself (Lombard, 1911). Studies investigating how individuals with ASD and their parents integrate auditory information to modify vocal output (i.e., audio-vocal integration, specifically with regard to pitch) have demonstrated an exaggerated pattern of compensatory vocal responses to pitch-perturbed auditory feedback in both ASD and ASD parent groups (Patel et al., 2019). In this task, participants completed a pitch-perturbed auditory feedback task in which they heard their voice in real-time as they vocalized. Computer-generated perturbations were presented during vocalization to elicit a compensatory vocal response. While all participants generated a compensatory vocal response (e.g., lowering pitch if they heard a computer-generated increase in pitch which may have sounded like one’s voice cracked), autistic individuals and their parents overcorrected for the perturbation by producing an even lower pitch than expected, indicating less efficient audio-vocal integration in both of these groups. Further, across individuals with ASD and their parents, increased vocal response magnitudes were associated with greater prosodic differences across subtests of the PEPS-C (Patel et al., 2019), highlighting audio-vocal integration as a potential heritable mechanism contributing to the prosodic differences in autistic individuals and the BAP.

1.2. Predictions

This study aimed to build on prior work to investigate the clinical and etiologic significance of prosodic impairments in ASD and subclinical differences evident among some first-degree relatives (Landa et al., 1992; Losh et al., 2012; Patel et al., 2019, 2020), by using a standardized test of receptive and expressive prosodic skills, and complementary acoustic analyses, to answer the following research questions:

Q1: What functions of prosody differ among individuals with ASD and their clinically unaffected relatives, possibly reflecting genetic liability to ASD?

Prediction: As discussed above, previous literature in ASD has identified differences across discrimination, imitation, phrase stress understanding and expression, affect understanding and expression, turn-end type understanding and expression, contrastive stress understanding and expression, and lexical stress expression (Diehl & Paul, 2012, 2013; Hesling et al., 2010; Järvinen-Pasley et al., 2008; McCann et al., 2007; Peppé et al., 2007). We aimed to reproduce these results in a new sample of autistic adolescents and examine parallel aspects of prosody in parents. Specifically, we predicted that parents of autistic individuals would show subtle prosodic differences in a subset of the PEPS-C domains impacted in ASD, and these overlapping domains would be candidates for indexing genetic risk to ASD. We predicted that prosodic differences would be apparent based on listener ratings and acoustic measurements.

Q2: How do acoustic measurements relate to patterns of differences identified by listener ratings across ASD, Control, ASD Parent, and Parent Control groups?

Prediction: In narrative and conversational settings, prior research has identified a complex relationship between acoustic measurements and listener ratings of prosody (Nadig & Shaw, 2012; Patel et al., 2020), such that there is not consistency in direct mapping between these two measurements. Acoustic measurements in ASD have been found to correlate with broader social communication differences characteristic of ASD (Diehl et al., 2009). Capitalizing on the structured nature of the PEPS-C, we predicted that acoustic markers of prosodic differences (i.e., mean, standard deviation, and range of F0) would emerge in subtests in which listeners identified prosodic differences across all groups.

Q3: How are prosodic differences captured by listener ratings and acoustic measurements related to broader clinical-behavioral measures across groups?

Prediction: Differences in prosody will cluster with the broader language and personality traits associated with ASD and the BAP, potentially reflecting an ASD-specific communication profile that may also be subtly expressed in relatives who are genetic carriers.

2. Methods

2.1. Participants

Participants were recruited through existing studies, the Northwestern University Communication Research Registry (P30DC012035), the Northwestern Child Studies Group, and by study advertisement. Thirty-one autistic individuals (ASD group), 24 typically developing controls (Control group), 50 parents of individuals with autism (ASD Parent group), and 39 parents of typically developing controls (Parent Control group) participated in this study. Efforts were made to recruit intact parent-child dyads. However, in some cases, participants were included in the study even if their parent or child were not available to participate. All participants were native English speakers with no history of hearing loss, brain injury, or the presence of a known genetic condition (e.g., Down syndrome) or syndromic forms of ASD (e.g., fragile X syndrome). Inclusion in both control groups required no personal history of language related impairments or family history of first- or second-degree relatives with ASD. Autistic individuals had an existing diagnosis of autism or autism spectrum disorder, which was confirmed for the study using the Autism Diagnostic Observation Schedule-2nd Edition (ADOS-2) (Lord et al., 2012) by research reliable study personnel. The Autism Diagnostic Interview-Revised (ADI-R) (Lord et al., 1994) was also used to confirm diagnoses when time permitted. The Wechsler Abbreviated Scale of Intelligence (WASI) (Wechsler, 1999) was used to assess intellectual functioning in individuals 16 years of age or older, while the Wechsler Intelligence Scale for Children-Fourth Edition (WISC-IV) (Wechsler, 2003) was used for individuals younger than 16 years of age. See Tables 1 and 2 for full sample characteristics.

Table 1.

Group characteristics.

	ASD (n = 31)	Control (n = 24)	Group Comparison (ASD vs. Control)	ASD Parent (n = 50)	Parent Control (n = 39)	Group Comparison (ASD Parent vs. Parent Control)
Sex	23:8	15:9		13:37	12:27
Males:Females
Chronological Age M (SD)	16.79 (5.38)	15.21 (6.99)	t(53)= − 0.95, p = .72, d = 0.26	48.90 (9.01)	48.27 (9.67)	t(87) = − 0.31, p = .49, d = 0.07
Full scale IQ M (SD)	100.13 (17.45)*	116.96 (12.52)	t(53) = 3.39, p = .05, d = 1.08	112.50 (12.07)*	119.51 (9.22)	t(87) = 3.01, p = .03, d = 0.64
Verbal IQ M (SD)	100.17 (19.64)	116.17 (13.79)	t(53) = 3.33, p = .11, d = 0.92	110.38 (12.88)*	115.59 (9.39)	t(87) = 2.10, p =0.03, d = 0.45
Performance IQ M (SD)	100.58 (18.48)	117.07 (15.73)	t(53) = 3.43, p = .30, d = 0.95	110.91 (11.94)	117.24 (9.98)	t(87) = 2.64, p = 0.39, d = 0.57
ADOS Severity Score M (SD)	8.37 (1.40)***	1.22 (0.52)	t(53) = 23.23, p < .001, d = 6.44	–	–	–

Note. M = mean; SD = standard deviation.

^*p < .05.

^**p < .01.

^***p < .001.

Table 2.

Ethnicity and education.

	ASD (n = 31)	Control (n = 24)	Group Comparison (ASD vs. Control)	ASD Parent (n = 50)	Parent Control (n = 39)	Group Comparison (ASD Parent vs. Parent Control)
Ethnicity N (%)			X2(1, 55) = 1.05, p = .31, w = 0.14			X2(1, 89) = 1.15, p = .28, w = 0.12
Hispanic	4 (12.90%)	1 (4.17%)		4 (8.00%)	1 (2.56%)
Non-Hispanic	27 (87.10%)	21 (87.50%)		45 (90.00%)	36 (92.31%)
Unknown	0 (0.00%)	2 (8.33%)		1 (2.00%)	2 (5.13%)
Education N (%)						X2(5, 89) = 10.12, p = .07, w= 0.34
Less than high school	–	–		1 (2.00%)	0 (0%)
High school or GED	–	–		6 (12.00%)	0 (0%)
Associate’s or Technical Degree	–	–		6 (12.00%)	1 (2.56%)
Bachelor’s Degree	–	–		8 (16.00%)	11 (28.21%)
Master’s Degree	–	–		17 (34.00%)	15 (38.46%)
Doctorate	–	–		4 (8.00%)	5 (12.82%)
Unknown	–	–		8 (16.00%)	7 (17.95%)

2.2. Profiling elements of Prosody in Speech-Communication (PEPS-C; McCann and Peppé 2003)

The PEPS-C includes seven receptive prosodic subtests (discrimination, turn-end type, affect, lexical stress, phrase stress, boundary, contrastive stress) and seven expressive subtests (imitation, turn-end type, affect, lexical stress, phrase stress, boundary, contrastive stress). Each domain further assesses two target responses (see Table 3 for summary of target response types for each subtest). Participants completed 16 items (8 items per each response type) for each subtest. During the receptive subtests, the participant identified their response by pointing or verbally stating their response, and then responses were automatically scored as incorrect or correct by the PEPS-C computer program. During the expressive subtests, the participant vocalized a response, which was recorded via a USB microphone (Shure MV51 or Blue Snowball) or head-mounted microphone (Audio-Technica ATW-1101/H92-TH). The audio data underwent extensive qualitative review for sound quality. All recorded responses were rated as correct, incorrect, or ambiguous by a team of three coders (authors SPP, EL, CR). Ambiguous responses were re-coded as incorrect for analysis. Fifteen percent of all files split across groups were coded by two raters for reliability. Inter-rater agreement for scoring of the expressive subtests was calculated using Fleiss’ kappa, and showed that there was substantial (Landis & Koch, 1977) agreement between raters, κ=0.771 (p < .001) for the ASD and Control groups and almost perfect agreement, κ=0.851 (p < 0.001) for the parent groups.

Table 3.

PEPS-C expressive subtests and target response types.

	Response 1	Response 2
Turn-end type	Statement	Question
Affect	Like	Dislike
Lexical Stress	Stress on first syllable (e.g., DIScount)	Stress on second syllable (e.g., disCOUNT)
Phrase Stress	Producing two separate words (e.g., green house)	Producing a compound word (e.g., greenhouse)
Boundary	Boundary after first word (e.g., chicken, fingers, and fruit)	Boundary after second word (e.g, chicken fingers and fruit)
Contrastive Stress	Stress on first item (e.g., BLUE cow)	Stress on second item (e.g., blue COW)

2.3. Acoustic analyses

The onset and offset of responses to the expressive subtests of the PEPS-C were manually marked using custom software and subsequently force aligned using the Montreal Forced Aligner (MFA) (McAuliffe et al., 2017). A Praat (Boersma, 2001) script with pitch tracking ranges dependent on speaker age and sex (see Table 4) was implemented in order to minimize pitch tracking errors in Praat. The script extracted F0 (in Hz) within each force-aligned utterance with a timestep of 0.01 s. In order to approximate the scale on which pitch is perceived, logarithmic transformation was applied to the F0 values. Subsequently, the mean, SD, and range of F0 for each response was calculated. Range of F0 was calculated by subtracting the minimum F0 from the maximum F0 for each utterance. Mean F0 is the physical correlate of pitch, thereby providing information regarding how “high” or “low” an individual’s voice is. SD and range of F0 measure variation in an individual’s F0. Acoustic analyses were conducted separately by task and target response.

Table 4.

Fundamental frequency detection ranges.

	Minimum (Hz)	Maximum (Hz)
Males 11 years and younger	130	400
Males 12–18 years of age	70	400
Males 19+ years	70	250
Females (all ages)	130	400

2.4. Clinical-behavioral correlates

2.4.1. ASD symptom severity

ASD Symptom Severity was assessed using calibrated overall severity comparison scores for both Modules 3 and 4 of the ADOS-2 (Gotham et al., 2009; Lord et al., 2012). Higher scores indicated greater symptom severity.

2.4.2. BAP personality traits

The Modified Personality Assessment Schedule-Revised (MPAS-R) (Piven et al., 1994; Tyrer & Alexander, 1988) was used to assess social (aloof or untactful) and rigid personality traits of the BAP in parents. Two trained coders independently rated personality traits on a scale of 0 to 2 from video. A score of 0 indicated the trait was absent, a score 1 indicated the trait was mildly or questionably present, and a score of 2 indicated the trait was definitely present. Discrepancies between coders were discussed until a consensus was reached for each participant. The rated personality features are thought to mirror in quality the social and repetitive behavior domains of impairment in ASD, have been shown to reliably distinguish clinically unaffected relatives of autistic individuals from controls (Losh et al., 2008; Piven et al., 1997), and relate to a variety of language skills in parents (Losh et al., 2010; Nayar et al., 2018; Patel et al., 2019; Patel et al., 2020).

2.4.3. Pragmatic language skills

Pragmatic language skills in the ASD and Control groups were assessed using the Pragmatic Rating Scale-School Age (PRS-SA) (Landa, 2011). The PRS-SA was rated from video based on semi-structured play and conversation from the ADOS-2 (Lord et al., 2012). In the ASD Parent and Parent Control groups, pragmatic language skills were assessed using the Pragmatic Rating Scale (PRS) (Landa et al., 1992). The PRS was coded from video based on a semi-structured conversational interview in which an examiner asked the parent a series of questions about their childhood, schooling, social relationships, and occupation. Fleiss’ kappa was used to assess interrater agreement, and showed that there was moderate agreement (Landis & Koch, 1977) between raters, κ=0.568 (p < 0.001) for the PRS-SA and κ=0.536 (p < 0.001) for the PRS.

For both the PRS-SA and the PRS, two coders blind to group independently rated the videotaped conversational interactions for pragmatic language features (e.g., overly detailed, response elaboration) on a three-point scale, with 0 indicating absent, 1 indicating mild, and 2 indicating present. Any coding discrepancies were resolved through discussion. Importantly, both the PRS and PRS-SA include items related to suprasegmental features (e.g., intonation variability, rate, volume, rhythm) examined in this study.

2.5. Statistical analysis plan

A series of one-way analyses of covariance (ANCOVAs) examined group differences (ASD vs. Controls, ASD Parent vs. Parent Controls) in prosodic skills across the receptive and expressive subtests of the PEPS-C including age as a covariate. Of note, not all participants completed each subtest of the PEPS-C (see Table 5 for sample size by subtest). This variation was largely due to time constraints during testing rather than varying ability levels of participants. Acoustic measures were examined in a series of mixed effects linear regression models investigating mean F0, SD of F0, and range of F0. Analyses were conducted in the ASD and parent groups separately for each subtest (e.g., turn-end expression, affect expression) and target response type (e.g., question vs. statement, like vs. dislike). All models included fixed and random effects of group and chronological age. Though the groups differed in IQ, the inclusion of full-scale IQ did not strengthen the model, so it was excluded. The models were fit using the lme4 package (Bates et al., 2014) for R statistical software. To explore interrelationships between each receptive and expressive subtest of the PEPS-C, as well as associations between expressive prosody scales on the PEPS-C and acoustic measurements, exploratory Pearson correlations were conducted within each group. Correlations with subtests of the PEPS-C in which group differences were identified are presented below and the remaining correlations are included as Supplementary Material. As reported below, these correlations did not withstand Bonferroni correction (p < .0009), and so should be interpreted with caution. They are reported to guide further investigation in this understudied area of research, particularly with regard to prosodic speech in first-degree relatives of individuals with ASD. In order to assess broader relationships between primary measures, ASD symptoms (including subclinical features of the BAP in parents), and pragmatic language skills, Pearson correlations with the ASD and Control groups combined were conducted. Combining groups allowed for assessment of relationships across the full range of abilities (i.e., typical to atypical in the child groups). This approach was similarly applied for correlations in the ASD Parent and Parent Control groups to examine the full range of subclinical features (Wainer et al., 2011).

Table 5.

Sample size for PEPS-C subtests.

	ASD	Control	ASD Parent	Parent Control
Discrimination	17	17	35	15
Imitation	30	24	50	39
Turn-end Understanding	31	24	49	38
Turn-end Expression	31	23	50	39
Affect Understanding	31	23	50	38
Affect Expression	31	23	50	38
Lexical Stress Understanding	31	23	50	38
Lexical Stress Expression	29	23	50	39
Phrase Stress Understanding	17	15	35	25
Phrase Stress Expression	17	15	35	25
Boundary Understanding	17	15	35	20
Boundary Expression	17	14	34	20
Contrastive Stress Understanding	31	21	50	38
Contrastive Stress Expression	31	22	50	39

3. Results

3.1. Group differences in PEPS-C listener ratings and acoustic measures

3.1.1. Receptive prosody skills

The ASD group exhibited significantly poorer ability understanding contrastive stress (F(1,49) = 10.81, p = 0.002), compared to the Control group. The ASD and Control groups did not differ significantly on any other receptive prosody subtests (ps ≥ 0.05; see Table 6 for results of ANCOVAs for each subtest). The ASD Parent and Parent Control groups did not differ significantly along any of the PEPS-C receptive prosody subtests (ps ≥ 0.12; see Table 7 for statistical results of ANCOVAs for the parent groups).

Table 6.

PEPS-C group comparisons in ASD and control groups.

	ANCOVA F	ANCOVA p-value	ANCOVA Effect Size ηp2	ASD M (SD)	Control M (SD)
Receptive Subtests
Discrimination	F(1,31) = 0.97	p = .33	.03	14.00 (2.03)	14.41 (1.50)
Turn-end Type	F(1, 52) = 0.97	p = .33	.02	14.87 (2.62)	15.38 (1.35)
Affect	F(1,51) = 0.24	p = .63	.01	14.06 (2.08)	14.13 (1.91)
Lexical Stress	F(1,51) = 4.23	p = .05	.08	10.19 (2.76)	11.43 (2.94)
Phrase Stress	F(1,29) = 0.24	p = .63	.01	11.71 (3.02)	12.07 (2.55)
Boundary	F(1,29) = 1.41	p = .25	.05	14.00 (3.24)	14.80 (2.11)
Contrastive Stress	F(1,49) = 10.81	p = .002 **	.18	12.90 (2.82)	14.90 (1.61)
Expressive Subtests
Imitation	F(1,51) = 30.08	p < .001 ***	.37	9.33 (4.56)	14.79 (1.32)
Turn-end Type	F(1,51) = 1.99	p = .17	.04	13.39 (2.79)	14.26 (2.67)
Affect	F(1,51) = 0.17	p = .68	.00	12.06 (4.01)	11.78 (3.99)
Lexical Stress	F(1,49) = 5.39	p = .02 *	.10	10.90 (4.06)	12.96 (2.79)
Phrase Stress	F(1,29) = 4.55	p = .04 *	.14	12.00 (3.04)	13.80 (1.97)
Boundary	F(1,28) = 0.36	p = .56	.01	12.88 (2.91)	13.36 (2.41)
Contrastive Stress	F(1,50) = 8.96	p = .004 **	.15	11.77 (3.29)	14.23 (2.51)

Note. M = mean; SD = standard deviation.

Note. Due to time constraints, some participants completed a reduced battery.As a result, sample size varied across subtests.

^*p < .05.

^**p < .01.

^***p < .001.

Table 7.

PEPS-C group comparisons in ASD parent and parent control groups.

	ANCOVA F	ANCOVA p-value	ANCOVA Effect Size ηp2	ASD Parent M (SD)	Parent Control M (SD)
Receptive Subtests
Discrimination	F(1,57) = 0.22	p = .64	.00	14.14 (1.48)	14.36 (1.55)
Turn-end Type	F(1, 84) = 0.51	p = .48	.01	15.94 (0.24)	15.89 (0.39)
Affect	F(1,85) = 1.46	p = .23	.02	14.94 (1.67)	15.32 (1.09)
Lexical Stress	F(1,85) = 0.48	p = .49	.01	11.70 (2.47)	12.03 (2.56)
Phrase Stress	F(1,57) = 0.73	p = .40	.01	12.77 (2.13)	13.48 (2.47)
Boundary	F(1,52) = 2.54	p = .12	.05	14.94 (1.53)	15.65 (0.81)
Contrastive Stress	F(1,85) = 1.11	p = .30	.01	14.30 (2.77)	14.82 (1.71)
Expressive Subtests
Imitation	F(1,86) = 5.28	p = .02 *	.06	13.61 (2.25)	14.59 (1.55)
Turn-end Type	F(1,86) = 2.03	p = .16	.02	15.14 (1.07)	15.44 (0.79)
Affect	F(1,85) = 0.45	p = .51	.01	13.96 (2.41)	14.32 (3.11)
Lexical Stress	F(1,86) = 4.40	p =.04 *	.05	13.92 (2.25)	14.74 (1.43)
Phrase Stress	F(1,57) = 2.89	p = .09	.05	14.23 (2.17)	15.12 (1.48)
Boundary	F(1,51) = 1.07	p = .31	.02	14.59 (1.84)	15.20 (0.83)
Contrastive Stress	F(1,86) = 7.67	p = .007 **	.08	14.02 (2.23)	15.13 (1.22)

Note. M = mean; SD = standard deviation.

Note. Due to time constraints, some participants completed a reduced battery. As a result, sample size varied across subtests.

^*p < .05.

^**p < .01.

^***p < .001.

3.1.2. Expressive prosody skills

The ASD group exhibited reduced accuracy imitating prosody (F(1,51) = 30.08, p < 0.001), as well as significantly poorer ability expressing lexical stress (F(1,49) = 5.39, p = 0.02), phrase stress (F(1,29) = 4.55, p = 0.04), and contrastive stress (F(1,50) = 8.96, p = 0.004) than Controls. No significant differences were detected on the remaining subtests (ps ≥ 0.17; see Table 6 for results of ANCOVAs for each subtest).

Similar to the ASD group, the ASD Parent group exhibited reduced accuracy imitating prosody (F(1,86) = 5.28, p = 0.03), expressing lexical stress (F(1,86) = 4.40, p = 0.04), and contrastive stress (F(1,86) = 7.67, p = 0.007) compared to Parent Controls. The parent groups did not differ on other subtests (ps ≥ 0.09; see Table 7 for results of ANCOVAs for each subtest).

3.1.3. Acoustic measurements of prosody

Acoustic differences between the ASD and Control groups were detected in SD and range of F0. The ASD group exhibited increased SD and range of F0 on lexical stress (first syllable) and boundary expression subtests, but reduced SD and range of F0 on the phrase stress (two words) subtest (ps ≤ 0.05). See Table 8 for results of linear mixed-effects models for each response type across subtests.

Table 8.

Acoustic group comparisons across child and parent groups.

		ASD vs. Controls			ASD Parent vs. Parent Controls
		Mean F0	SD F0	Range F0	Mean F0	SD F0	Range F0
Imitation		ß = 0.012	ß = − 0.061	ß = − 0.097	ß = 0.066	ß = 0.092	ß = 0.011
Turn End	Statement	ß = 0.050	ß = 0.321	ß = 0.371	ß = 0.008	ß = 0.092	ß = 0.117
	Question	ß = 0.029	ß = 0.078	ß = 0.134	ß = 0.005	ß = 0.029	ß = 0.058
Affect	Positive	ß = 0.010	ß = 0.018	ß = .− 0.007	ß = 0.010	ß = 0.175 *	ß = 0.207 *
	Negative	ß = 0.012	ß = − 0.076	ß = − 0.094	ß =−0.069 *	ß = − 0.118	ß = − 0.101
Lexical Stress	Stress on First Syllable	ß = 0.053	ß = 0.526 **	ß = 0.574 **	ß = − 0.012	ß = 0.030	ß = 0.037
	Stress on Second Syllable	ß = 0.027	ß = 0.334	ß = 0.343	ß = 0.002	ß = 0.169 *	ß = 0.193 *
Phrase Stress	Compound Word	ß = 0.017	ß = 0.071	ß = 0.045	ß = 0.001	ß = − 0.022	ß = 0.001
	Two Words	ß = 0.051	ß = 0.334 *	ß = 0.415 *	ß = − 0.000	ß = 0.026	ß = 0.035
Boundary	Boundary after First Word	ß = − 0.001	ß = 0.356 **	ß = 0.361 **	ß = − 0.019	ß = − 0.004	ß = 0.024
	Boundary after Second Word	ß = − 0.010	ß = 0.270 *	ß = 0.341 *	ß = − 0.020	ß = 0.028	ß = 0.049
Contrastive Stress	Stress on First Word	ß = − 0.024	ß = 0.230	ß = 0.208	ß = − 0.037	ß = 0.128	ß = 0.135
	Stress on Second Word	ß = − 0.011	ß = 0.209	ß = 0.234	ß = − 0.037	ß = 0.086	ß = 0.066

Note. Due to time constraints, some participants completed a reduced battery. As a result, sample size varied across subtests.

^*p < .05.

^**p < .01.

^***p < .001.

In the parent groups, reduced mean F0 was evident on the affect (dislike) expression subtest (p = 0.03). The ASD Parent group exhibited increased SD and range of F0 across the affect (positive) and lexical stress (second syllable) subtests (ps ≤ 0.05). See Table 8 for results of linear mixed-effects models for each response type across subtests.

3.2. Associations between expressive PEPS-C ratings and acoustic measurements

Associations between subtests in which groups differed on expressive PEPS-C subtests are presented here. Despite moderate to large correlation coefficients, associations did not withstand Bonferroni correction (p < 0.0009) and should be interpreted accordingly. Poorer contrastive stress expression (second word) was associated with increased mean F0 (r = − 0.49, p = 0.03) in the ASD group. In the Control group, reduced scores on phrase stress (two words) expression were associated with greater mean F0 (r = − 0.58, p = 0.04).

In the ASD Parent group, poorer scores on contrastive stress (second word) expression (r = − 0.49, p = 0.03) were associated with increased mean of F0. There were no significant associations in the Parent Control group within subtests that groups differed on for expressive ratings.

3.3. Associations with ASD symptoms, BAP personality traits, and pragmatic language skills

3.3.1. ASD and control groups

In the ASD and Control groups combined, greater ASD symptom severity was associated with poorer prosodic imitation, lexical stress expression, phrase stress expression, and contrastive stress expression (rs ≤ − 0.32, ps ≤ 0.05). Greater ASD symptom severity was associated with poorer contrastive stress understanding (r = − 0.42, p = 0.002). Greater overall pragmatic language impairment was associated with poorer performance on the imitation, phrase stress, and contrastive stress expression subtests (rs ≤ − 0.43, ps ≤ 0.02), as well as understanding contrastive stress (r = − 0.39, p = 0.004). See Fig. 1 for correlations between pragmatic language domains, ASD symptom severity, and PEPS-C subtests. With regard to acoustic measures, greater ADOS symptom severity was associated with increased SD and range of F0 on the contrastive stress (first and second words) and lexical stress (first syllable) subtests (rs ≥ 0.33, ps ≤ 0.05). Examining associations between pragmatic skills and acoustic measures of prosody revealed that increased rates of pragmatic language violations in a conversational context were associated with increased SD and range of F0 on the lexical stress (first syllable) subtest (rs ≥ 0.41, ps ≤ 0.01) and increased SD of F0 on the contrastive stress (first word) subtest (r = 0.47, p = 0.04).

Fig. 1.

Correlations between PEPS-C performance, ASD symptom severity, and pragmatic language ability in the ASD and control groups.

*p < .05.

**p < .01.

***p < .001.

3.3.2. ASD parent and parent control groups

Social personality traits of the BAP were associated with better imitation skills (r = 0.36, p = 0.006) (see Fig. 2). Rigid traits of the BAP were associated with poorer contrastive stress understanding (r = −0.29, p = 0.03). There were no statistically significant associations between pragmatic language and performance on the PEPS-C in the parent groups (rs ≤ 0.21 or ≥ −0.15, ps ≥ 0.11). Acoustic measurements were related to BAP personality traits and pragmatic language. The presence of a socially reticent personality style was associated with increased SD and range of F0 on lexical stress expression (first syllable) (rs ≥ 0.37, ps ≤ 0.05). Higher pragmatic language violations were associated with increased range of F0 on the imitation subtest (r = 0.28, p = 0.03).

Fig. 2.

Correlations between PEPS-C performance, BAP personality traits, and pragmatic language ability in the parent groups.

*p < .05.

**p < .01.

***p < .001.

4. Discussion

Impairments in the perception and production of prosody have been consistently reported in ASD, and more subtle prosodic differences have also been described as a component of the language-related features of the broad autism phenotype (BAP) in clinically unaffected relatives, suggesting that such differences are not only clinically, but also etiologically meaningful. This study (1) investigated receptive and expressive prosodic abilities using a standardized assessment and applied objective acoustic measurements to the speech samples obtained during standardized testing in autistic individuals and their parents, (2) examined the relationship between listener ratings of expressive prosody and acoustic measurements, and (3) examined relationships between prosody and broader clinical-behavioral features. Consistent with prior literature, the ASD group displayed few difficulties with receptive prosody, but showed more pronounced differences in expressive prosody that in some cases overlapped with patterns of differences observed in parents— namely, imitation, lexical stress, and contrastive stress. Interestingly, across these domains acoustic differences were only noted in lexical stress, and overall, measures of acoustic properties of speech samples from the PEPS-C revealed relatively minimal group differences. Given varying sample sizes across subtests of the PEPS-C, it is possible that differences may reflect variation in power to detect statistically significant effects. Therefore, the presented profile of differences should be interpreted with caution. Ratings from the PEPS-C and acoustic measurements of prosody were associated with broader pragmatic language profiles in the ASD and Control groups, whereas only acoustic measurements of prosody that related to pragmatic abilities among parent groups. Together, findings confirm and extend prior work demonstrating several key prosodic skills impacted in ASD, which also appear to be influenced by broader genetic liability to ASD as evidenced by differences in biological parents of autistic individuals.

4.1. Profiles of receptive and expressive prosody in ASD and parents

Contrary to prior research suggesting that adolescents with ASD tend to exhibit parallel trajectories of receptive and expressive language development beyond their early childhood years (Kwok et al., 2015), the ASD prosodic profile that emerged from analyses of receptive and expressive subtests on the PEPS-C is one of discordance, where receptive prosodic skills generally outstripped performance on expressive subtests. One notable exception concerned contrastive stress understanding, an area in which impaired comprehension in ASD has been reported in prior work (Diehl & Paul, 2013; Peppé et al ., 2007). Contrastive stress is critical for imparting communicative intent, and misunderstanding of contrastive stress may derail a social interaction. For example, if a communication partner states, “I WALKED to your house,” the stress on walked may be correcting an implication that the individual drove to the house. In this situation, understanding contrastive stress is imperative to interpreting the communication partner’s intent (e.g., perhaps subtly requesting a ride home because it is raining outside) and may impact the listener’s response. Interestingly, autistic individuals, as well as their parents, exhibited reduced accuracy expressing contrastive stress, suggesting that this domain may be central to the prosodic profile of ASD. Together with parallel patterns of reduced lexical stress expression in the ASD and ASD Parent groups, as well as reduced accuracy producing phrase stress in the ASD group, findings suggest that stress more broadly may be a key prosodic feature impacted in ASD and mark genetic liability to ASD.

4.2. Acoustic features associated with receptive and expressive prosody

In contrast to the similarities observed among PEPS-C subtests in the ASD and ASD parent groups, where discordance across receptive and expressive prosodic ability, and imitation, lexical stress, and contrastive stress differences were observed in both the ASD and ASD parent groups, acoustic analyses revealed less consistent patterns. In the autistic group, acoustic differences aligned with ratings of poorer lexical stress and phrase stress expression, with findings indicating differential patterns of intonation variability (SD and range of F0). Interestingly, differences in ratings on the contrastive stress expression subtest were not apparent in acoustic analyses, conflicting with a prior report of increased SD and range of F0 as contributors to contrastive stress differences in ASD (Diehl & Paul, 2013). These differing results may be due to analysis of all responses in the present study compared to analysis of only responses judged as accurate in the Diehl and Paul (2013) study.

While some acoustic differences emerged in the ASD Parent group, these differences did not correspond to subtests in which parents received lower ratings on the PEPS-C. Comparisons did not withstand Bonferroni corrections, and so may reflect some spurious findings or possibly some significant findings that were not strong enough to withstand the correction. As such, further research with a larger sample size and fewer variables will be necessary to derive meaning from correlational findings. In general, lack of consistency between listener ratings on the PEPS-C and acoustic differences suggests that the acoustic differences detected in parents reflect subtle prosodic variation among clinically unaffected relatives that are not perceptually salient to listeners compared to the more pronounced and heterogenous acoustic variation present in autistic individuals. It is possible that a confluence of acoustic measurements rather than individual measures assessed in the present study impacted listeners’ ratings of poorer prosodic ability. Consistent with some previous literature demonstrating that acoustic differences do not regularly map to listener-rated differences in ASD (Patel et al., 2020), the lack of consistent associations between ratings on the PEPS-C and variation in acoustic measurements underscores the complexity of acoustic properties that likely contribute to perceptions of differences in such skills as contrastive stress marking. The acoustic measurements deployed in this study were limited to isolated features (e.g., measures of average pitch [F0]) that may not capture the more complex integration of subtle variations across multiple acoustic components that contribute to a listener’s perception of prosodic differences.

4.3. Relationships between prosody and broader clinical-behavioral traits

In the ASD and Control groups combined, reduced accuracy across several receptive and expressive subtests (imitation, lexical stress, phrase stress, contrastive stress) and differences in acoustic patterns of speech (generally increased SD and range of F0) were related to more severe pragmatic language impairment and elevated ASD symptom severity. Associations suggest that impairments in these areas of prosody may be more closely tied to the core symptoms of ASD. In parents, lower expressive prosodic skills (imitation, affect) and differences in acoustic patterns of speech (generally increased SD and range of F0) were more common in parents who exhibited social personality (i.e., aloof) and pragmatic language traits of the BAP, suggesting that prosodic abilities co-segregate with BAP features to represent a potentially biologically significant communication profile in parents. These findings in parents highlight expressive prosodic abilities as integral components of the complex social communication profile associated with ASD in clinically unaffected relatives. Additionally, findings support accumulating evidence implicating prosody as an important phenotype indexing genetic risk of ASD among first-degree relatives (Patel et al., 2019, 2020). Of note, agreement between raters for pragmatic language was in the moderate range for this study, and while that is generally an acceptable range and, particularly so, considering the difficulty inherent in coding pragmatic language (Adams, 2002; Brinton & Fujiki, 1989; Roberts et al., 2007), understanding of relationships between prosody and broader pragmatic language skills may become clearer in with greater consistency between raters.

5. Implications

This study identified a discordant pattern of prosodic skills in autistic individuals and their parents, with receptive prosodic skills exceeding expressive prosodic skills. Individuals with ASD showed some evidence of receptive prosody deficits compared with controls, but more pronounced expressive deficits. Notably, expressive prosodic differences overlapping with those observed in ASD, but more subtly expressed (and not warranting clinical intervention), were also evident in clinically unaffected parents of individuals with ASD, suggesting that key elements of prosodic variability reflect genetic liability to ASD. Across the ASD and ASD parent groups, difficulties in prosodic expression were only inconsistently supported by acoustic differences. Together, findings suggest that key aspects of prosody, and particularly expressive prosody, may be important targets for social-communication interventions in ASD, and that interventions may not need to focus on specific acoustic measurements, but rather may benefit from focusing on listener ratings of key domains (e.g., contrastive stress) of expressive prosodic deficits. Of course, intervention studies are necessary to evaluate the effectiveness of these recommendations in a clinical context. In conjunction with these findings, relationships between prosody and broader clinical-behavioral features of ASD and the BAP (e.g., reduced accuracy on the PEPS-C associated with increased ASD symptomatology and social personality in the BAP) highlight prosody as an important marker of genetically meaningful communication characteristics in ASD.

6. Future directions

The present findings support several possible lines of future work. First, data-driven approaches incorporating machine learning or computational prosodic modeling may help to further depict the dynamic and often subtle acoustic variations in speech contributing to audibly detectable differences in ASD and parents. The inclusion of additional acoustic features, such as intensity, speech rate, and rhythm may also be important to include in future work, as these areas have been shown to be impacted in ASD in prior literature (Bone et al., 2012; Hubbard et al., 2017). Similarly, it is important to note that the structured nature of the PEPS-C may limit generalization of present findings to other contexts in which prosodic differences have commonly been reported. As such, further investigation of prosody across contexts (e.g., narrative, conversation) is important to understand how the specific deficits in ASD and more subtle differences in parents noted here manifest in more naturalistic interactions, as well as how other contexts might reveal differences not tapped by highly structured tasks such as the PEPS-C. For example, the PEPS-C provides a limited sampling of prosodic targets that do not fully account for the full range of prosodic patterns (e.g., prosodic patterns characteristic of sarcasm) that may be impacted in ASD and in parents who show more subtle differences. Studying more complex and naturalistic contexts may be particularly important for future studies of parents, considering that the parent groups achieved near-ceiling results on the PEPS-C. Relatedly, acoustic analyses in the present study focused primarily on foundational acoustic measurements that could be easily extractible in a clinical setting, which, although valuable, should be expanded on to include more varied and sensitive acoustic measures (e.g., speech rate, speech rhythm measured using a variable such as nPVI, which assesses durational variability across syllables within an utterance) that might help to more precisely define the acoustic properties of perceived differences comprising the prosodic profile of ASD and the BAP.

An additional important focus for future research will be to examine whether the present findings extend to autistic individuals with reduced cognitive and language abilities. Studies have shown links between prosodic abilities and certain cognitive abilities, such as divided attention, working memory, and inhibition, suggesting that the present findings may not generalize to the full range of the autism spectrum of ability (Filipe et al., 2018). Future research assessing prosodic skills in autistic individuals with more varied cognitive and language abilities should include instructional modifications to the PEPS-C to match the needs of the participants. Finally, it will be important for future work to expand on these findings in samples enriched for females, to further explore possible sex differences that could inform specific treatment protocols for autistic individuals, and further illuminate the expression of genetic liability to ASD in first degree relatives.