Correspondence between consonant inventories obtained using modified Naturalistic Listening in Real Time (NLRT) and parent report

Introduction

Studies of early speech sound development in English speaking children have demonstrated that infants’ early vocalizations progressively increase in both volume and complexity from 7 – 18 months, and while a great deal of individual variability is present there are distinct patterns that emerge (Morgan and Wren, 2018). Stops, nasals, and glides are evident in the majority of utterances produced during this period (Gildersleeve et al., 2000; Robb and Bleile, 1994), with alveolar stops occurring most frequently between 7 to 12 months (MacNeilage et al., 1997; Robb and Bleile, 1994). Fricatives and affricates are not typically evident in the syllable-initial position during this time but may be produced in the syllable-final position (Gildersleeve et al., 2000; Robb and Bleile, 1994; Vihman et al., 1986). Although variability is evident in consonant inventories across young children, size of consonant inventory should also progressively expand as the child enters the linguistic period of development (Robb and Bleile, 1994; Stoel-Gammon, 1985).

The study of early vocal development has historically involved phonetic transcription. Some authors have argued that phonetic transcription of early prelinguistic vocalizations is problematic because infant productions are often immature and lack the precision needed to use the International Phonetic Alphabet (Oller and Ramsdell, 2006). In their series of investigations, Ramsdell and colleagues (2012) compared syllable inventories obtained from infants using three different methods: parent report, phonetic transcription, and naturalistic listening (i.e., listening to recorded samples of vocalizations in real time, then listing syllable types from memory). Although similar inventories were obtained between parent report and naturalistic listening, larger inventories were obtained using phonetic transcription. The authors hypothesized that parents probably attend more to global sound patterns than to phonetic detail and proposed that “the functional repertoire of infant syllables is best seen as that repertoire recognized by caregivers” (p. 1627). They argued that phonetic transcription can overestimate a child’s consonant inventory and compromise efforts to predict future performance or compare differences between groups.

Willadsen and her colleagues (2020) were interested in determining the reliability of naturalistic listening in real time (NLRT) compared to phonetic transcription while also assessing the validity of NLRT for use in a large-scale study of early infant vocal development of young children with cleft palate. Similar to Ramsdell and colleagues, they also reported smaller syllable inventory sizes when using NLRT compared to phonetic transcription and noted that the latter method was also more time consuming. Large variation in intercoder reliability was evident, however, for specific syllable types. Since coders listened to a 22-minute recording before writing down syllable types, the authors speculated that reliability might be enhanced by having coders listen to shorter samples.

Willadsen and her colleagues (2020) acknowledged that while phonetic transcription may overestimate a child’s functional phonetic inventory, underestimating a child’s inventory may be equally problematic depending on the purpose of an investigation. They pointed out that phonetic transcription might be more appropriate in investigations that are attempting to examine the emergence of consonants or to document the frequency of different syllable types. Hardin-Jones and her colleagues (2022) concurred with the authors and stated that the emergence of oral stops in babbling and early words of toddlers with repaired cleft palate is of particular interest to SLPs because those productions provide early evidence of attempted velopharyngeal closure. Since NLRT will likely miss emerging consonants that are produced infrequently and since phonetic transcription is a time- consuming option, they recommended the use of a modified NLRT procedure. This latter method involved coders listening to recorded samples of vocalizations in real time, identifying syllables using TimeStamper software (Willadsen et al., 2018), and documenting each consonant they heard on a coding form. The authors reported excellent intercoder reliability for size of consonant inventory and noted that this method is more easily used than traditional phonetic transcription in both research and in clinical settings because it is less labor intensive. Further, while Willadsen et al. reported that twice as many consonants were identified during phonetic transcription compared to the NLRT condition, Hardin-Jones et al. found that the average number of different sounds identified as in-inventory consonants in the two different contexts for their 20 participants was only one. They attributed the closer match to the coder’s ability to write down consonants as they were heard in real time, which minimized the influence of memory.

Overestimation of a child’s phonetic inventory using phonetic transcription is likely related, at least in part, to how an in-inventory consonant is defined. Historically, the majority of investigations examining vocal development in young children have arbitrarily defined an in-inventory consonant as one that occurs at least two or three times in a sample (e.g., Chapman et al., 2001; Scherer et al 2008; Stoel-Gammon, 1985; Robb and Bleile, 1994). While this approach may lack the ecological validity of naturalistic listening, it has provided a standard approach to characterizing early sound inventories for both research and clinical purposes. The correspondence between parent reported phonetic inventory and that obtained through phonetic transcription might be closer if a more stringent criteria were used to define an in-inventory consonant. The purpose of the present investigation was to extend the Hardin-Jones et al. (2022) study and examine the correspondence between consonant inventories obtained using the modified NLRT approach and parent report. We were also interested in determining whether increasing the number of consonant tokens required for inclusion in the consonant inventory from two to ten would result in a closer match between parent-reported inventories and those obtained through phonetic transcription.

Methods

This study was reviewed and approved using a Single Institutional Review Board model with the University of Utah as the Institutional Review Board of record (IRB#00105205). Informed consent was obtained at each of the sites participating in the Cleft Outcomes Research Network (CORNET) for each child participant following federal law, regulation, and policy, including the policies set forth by the National Institutes of Health.

Participants

Participants included 70 toddlers (44 males and 26 females) with repaired cleft palate ranging in age from 14–18 months (mean age=16 months) who were participating in the multisite CORNET study. These children were selected from a larger cohort because parent reported consonant inventories were available for them. They included seven children with bilateral cleft lip and palate, 23 with unilateral cleft lip and palate, 31 with clefts of the hard and soft palate, and 9 with clefts of the soft palate only. Fifty-one children had isolated clefts, 15 had Pierre Robin Sequence, while the remaining four had syndromes. All participants were from English-speaking homes.

All participants had undergone cleft palate repair. Age at time of palatal surgery ranged from 9–16 months (mean= 11.83 months). Fifty-three of the 70 children (76%) had received myringotomy and ventilation tubes by the time of their 16-month evaluation.

Sixty-seven (96%) of the participant’s parents completed the Ages and Stages Questionnaire (ASQ-3; Brookes Publishing, 2009). The Problem-Solving domain score (which estimates whether a baby’s development is on schedule, borderline, or in need of additional assessment) was used as a proxy for cognition. Forty-eight of the 70 participants (69%) demonstrated problem solving skills that were considered to be on schedule. Scores were in the borderline range for 11 (16%) children while the remaining 8 (11%) children had scores suggesting the need for additional assessment.

Procedures

Each participant’s parent was provided with either (1) a written checklist of 15 consonants (h, w, j, m, n, b, p, d, t, g, k, f, s, ʃ, l) and asked to identify the consonants they have heard their child produce in words or babbling, or (2) this list was reviewed verbally with them by the study coordinator (if the visit was conducted remotely, over the phone), at the time of the 16-month study visit. To further ensure that parents understood the information being requested, examples of “sounds” were provided (e.g., Does your child produce the “b” sound as in “bye-bye”, the “k” sound as in “kitty”, etc.). In addition, they were asked to record their child in the home for approximately two-to-four hours using a Language ENvironmental Analysis (LENA) recorder (a digital language processor, DLP), within 2 weeks of the study visit/receipt of the DLP. The DLP is a small device with a microphone that can be placed in a vest that the child wears throughout the recording period.

The audio recording was processed using LENA software and hourly Child Vocalization Counts (CVC) were obtained using automated processing algorithms. Eight consecutive five-minute audio-recorded segments were then extracted from the child’s most vocal hour of the recording. The audio segments were imported in pairs and aligned using Audacity audio editor software to create four ten-minute audio-recorded samples of vocalizations for each participant. At least 100 syllables were required across the four ten-minute segments to ensure that an adequate sample size was available for analysis. Children whose samples did not meet this syllable criterion were excluded from the analysis. Sennheiser (Model eH 1430) headphones were used by the coders to listen to the audio recordings.

Modified NLRT Analysis

A modified Naturalistic Listening in Real Time (NLRT) analysis was used in this study and has been previously described (Hardin-Jones et al., 2023). Each ten-minute recording was loaded into the TimeStamper software program (Willadsen et al., 2018) and assessed in real time without pause. The software identifies annotations of a keystroke for canonical syllables and another keystroke for noncanonical syllables. For the purposes of this study, only one keystroke was used to maintain a cumulative count of syllables. Each coder was provided with a list of consonants and asked to place a vertical line next to each consonant they heard the child produce. Only consonants that were produced a minimum of two times were considered to be in the child’s inventory in the analysis performed for the larger multisite cohort.

In addition to the 15 consonants provided on the parent form, coders were provided with the opportunity to identify compensatory articulations (e.g., glottal stops, pharyngeal fricatives) as well as consonants more commonly identified in older children (e.g., ʧ, ʤ, r). It should be noted, however, that while coders were provided with the opportunity to identify more consonants (n=29) than were included on the list provided to parents, only those 15 consonants provided on the parent form were ever identified by coders as “in inventory” for the participants in this study.

Two coders listened to the audio recordings, one certified SLP and one graduate student in speech-language pathology. The certified SLP participated in an NLRT training session using TimeStamper software conducted by Dr. Elisabeth Willadsen, and then replicated the training protocol with the graduate student several months later. Major types of vocalizations including canonical syllables, marginal syllables, fully resonant nuclei, and quasi-resonant nuclei were reviewed. Reflexive vocalizations (e.g., cry, laughter, laughed syllables without a consonant, hiccups, burps, sneezes, sounds of effort/grunts) and babbling sequences that did not contain a vowel were excluded from analysis as were vocalizations produced when the child had something in his/her mouth (ex. pacifier; food). Sustained consonants that conveyed a specific message (e.g., “shhh” to indicate “be quiet”) were included.

Data Analysis

Size of consonant inventory, including those reported by parents and those identified by the coders, were entered into IBM SPSS Statistical Program (version 26). Descriptive analyses were performed to determine the total number of syllables and consonants produced by the participants per caregiver and coder report. Size of consonant inventory identified by the coders was analyzed in two ways: 1) counting the number of consonants produced by each participant a minimum of two times (per consonant) and 2) counting the number of consonants produced a minimum of 10 times.

Reliability

Both coders independently identified the number of consonants from 18 of the 70 recordings (26%) using the modified NLRT. Interjudge reliability for total number of consonants in inventory calculated using Pearson’s product moment correlational analysis, was 0.90.

Results

Consonant inventory comparisons: parent report versus coder count for 2+ occurrences

Size of phonetic inventory identified in the modified TimeStamper analysis for the 70 participants ranged from 4–13 consonants (mean=7.90), while size of phonetic inventory reported by the participant’s parents ranged from 1 to 15 (mean=6.06) (Table 1). Coders identified more consonants than parents reported for 49 of the 70 (70%) children. In contrast, only 12 parents (17%) identified more consonants in their child’s inventories than the coders. Coders and parents agreed on size of consonant inventory for the remaining nine children (13%). Interestingly, when coders and parents did not agree on size of consonant inventory, disagreements were within 2 consonants for 27 of the 61 (44%). A positive correlation of .34 (p<.01) was obtained between size of consonant inventories reported by parents and identified by coders.

Table 1.

Comparison of parent and coder consonant inventory counts.

Participant	Diagnosis	Cleft Type	Parent # Consonants	Coder # Consonants 2+ occurrences	Coder # Consonants 10+ occurrences
1	CLP/CP	ULP	2	5	3
2	CLP/CP	ULP	2	7	1
3	CLP/CP	HSP	3	11	8
4	CLP/CP	HSP	1	9	8
5	CLP/CP	ULP	2	10	4
6	CLP/CP	ULP	3	6	3
7	CLP/CP	ULP	2	7	1
8	CLP/CP	HSP	9	10	6
9	CLP/CP	HSP	5	7	0
10	CLP/CP	HSP	2	6	3
11	CLP/CP	BLP	9	9	5
12	CLP/CP	HSP	4	4	2
13	CLP/CP	ULP	3	6	3
14	CLP/CP	BLP	3	9	7
15	CLP/CP	BLP	6	7	4
16	CLP/CP	ULP	5	6	5
17	CLP/CP	ULP	15	8	7
18	CLP/CP	ULP	9	11	6
19	CLP/CP	HSP	10	10	7
20	CLP/CP	ULP	8	10	7
21	CLP/CP	HSP	7	7	5
22	CLP/CP	ULP	15	5	4
23	CLP/CP	HSP	10	9	7
24	CLP/CP	ULP	4	8	5
25	CLP/CP	ULP	10	9	3
26	CLP/CP	ULP	6	8	3
27	CLP/CP	BLP	7	7	4
28	CLP/CP	HSP	9	6	4
29	CLP/CP	SPO	5	11	7
30	CLP/CP	BLP	7	13	11
31	CLP/CP	ULP	6	8	4
32	CLP/CP	BLP	8	9	6
33	CLP/CP	HSP	7	8	5
34	CLP/CP	ULP	4	10	6
35	CLP/CP	ULP	8	8	5
36	CLP/CP	SPO	11	11	5
37	CLP/CP	ULP	4	6	5
38	CLP/CP	ULP	2	6	4
39	CLP/CP	SPO	8	6	3
40	CLP/CP	HSP	8	12	8
41	CLP/CP	BLP	9	11	4
42	CLP/CP	HSP	6	11	10
43	CLP/CP	ULP	4	6	5
44	CLP/CP	SPO	11	9	3
45	CLP/CP	BLP	4	8	4
46	CLP/CP	ULP	2	5	1
47	CLP/CP	HSP	13	9	4
48	CLP/CP	SPO	5	9	7
49	CLP/CP	ULP	5	7	4
50	CLP/CP	BLP	5	5	4
51	CLP/CP	HSP	4	7	4
52	PR	HSP	4	8	3
53	PR	HSP	5	7	4
54	PR	ULP	2	9	3
55	PR	HSP	3	7	5
56	PR	HSP	3	8	3
57	PR	HSP	11	12	10
58	PR	HSP	6	6	2
59	PR	HSP	5	8	5
60	PR	ULP	8	6	3
61	PR	HSP	6	7	2
62	PR	BLP	3	6	2
63	PR	HSP	10	8	4
64	PR	HSP	2	5	2
65	PR	HSP	4	6	4
66	PR	HSP	3	7	2
67	syndrome	BLP	10	9	5
68	syndrome	BLP	6	8	3
69	syndrome	HSP	11	7	5
70	syndrome	HSP	5	7	2

Note: Diagnosis: CLP/CP= Cleft lip and palate/Cleft palate only, PR= Pierre Robin Sequence, Syndrome= craniofacial syndrome

Cleft Type: BLP= Bilateral cleft lip and palate, ULP= Unilateral cleft lip and palate, HSP= Cleft of the hard and soft palate, SPO= Cleft of the soft palate only

Parents and coders agreed on total manner categories present in inventory for only 10 children. Table 2 shows the agreement between parents and coders for individual manner of production categories. Agreement between the groups for presence or absence of nasals (94%) and stops (87%) was high while agreement for glottal (60%) and glides (56%) (both categories that are typically present in this age group) was much lower.

Table 2.

Agreement between parents and coders for manner of production categories in inventory

	Coder 2+ occurrences # (%) agreement	Coder 10+ occurrences # (%) agreement
Glottals	42 (60)	45 (64)
Glides	39 (56)	40 (57)
Nasals	66 (94)	52 (74)
Stops	61 (87)	44 (63)
Fricatives	56 (80)	55 (79)
Liquids	37 (53)	45 (64)

As evident in Table 3, agreement was high for presence of [m] in the children’s phonetic inventories with coders identifying it in 66 children and parents reporting it in 69. In contrast, disagreements between the coder and parents frequently occurred for [h], [w], and [j]. The [h] sound was identified as an in-inventory consonant for all 70 children by the coders, but only 42 (60%) parents reported that their child produced it. The [w] and [j] were identified in the coder’s transcripts for 67 children but were reported by parents for only 19 and 35 children, respectively.

Table 3.

Number of children with specific consonants in inventory: parent report versus coder count for 2+ and 10+ occurrences.

	Parent	Coder 2+ consonant occurrence	Coder 10+ consonant occurrence
h	42	70	53
w	19	67	48
j	35	67	40
m	69	66	42
n	44	64	41
p	27	11	2
b	38	45	17
t	10	9	1
d	45	47	26
k	17	24	8
g	26	48	29
f	5	0	0
s	12	3	1
ʃ	13	6	2
l	21	23	5

Consonant inventory comparisons: parent report versus coder count for 10+ occurrences

When only those consonants that were identified 10 or more times within the forty-minute recording period were considered as “in-inventory” consonants, differences in size of phonetic inventory between the two groups were reduced (see Table 1). Size of phonetic inventory identified in the modified TimeStamper analysis ranged from 1–11 consonants (mean=4.46) while that reported by the participant’s parents ranged from 1 to 15 consonants (mean=6.06). Interestingly, parents reported more consonants than coders for 43 of the 70 (61%) children while coders identified more consonants than parents for 17 (24%) children. Coders and parents agreed on size of consonant inventory for the remaining ten children (14%). When coders and parents did not agree on size of consonant inventory, disagreements were within 2 consonants for 29 of the 60 (48%) children. Although the mean number of consonants in-inventory between parents and coders was closer when a minimum of 10 productions was required, the correlation between inventories was similar to that obtained in the previous comparison (r=.31; p<.01).

Agreement between parents and coders on manner of production categories was lower when a minimum of 10 tokens of consonants was used to define an in-inventory consonant. Both groups agreed on total manner categories present in inventory for only 5 children. As evident in Table 2, agreement between the groups for presence or absence of nasals (74%) and stops (63%) was substantially lower in this comparison than when only a minimum of two tokens was required for inclusion of a consonant in inventory.

Individual Consonant Production

Table 3 shows the number of children who produced or were reported to produce each consonant. The top five most frequently reported consonants by parents included [m], [d], [n], [h], and [b]. In contrast, the top five consonants most frequently transcribed by the coders included [h], [w], [j], [m], and [n] when 2 or more occurrences were required for inclusion and [h], [w], [m], [j], [n] when 10 or more occurrences were required for inclusion. Although parents reported the stop consonants [b] and [d] in the inventories of more than 50% of the children, stops were present in more than 50% of the coded samples only when the two or more occurrence criterion was used. Further, even though size of consonant inventory was more similar between parent report and coded samples that included ten or more occurrences of consonants, composition of the parent inventories was more similar to the transcriptions that required only two or more occurrences. While fricatives were rarely identified by the coders for children in this study, parents reported [s] and [ʃ] in the inventories of 12 and 13 children, respectively.

Discussion

The current investigation was conducted to compare size of consonant inventory reported by parents with inventories coded using two different consonant inventory inclusion criteria. As expected, larger consonant inventories were typically coded using the modified NLRT procedure when fewer tokens of a consonant (i.e., a minimum of two) were required for inclusion in a child’s inventory. Coders identified more consonants than parents reported for 70% of the participants when a minimum of two consonant tokens were used for inclusion in the children’s consonant inventories, while parents reported more consonants for 61% of the participants when the inclusion criteria required 10 or more consonant tokens.

Previous research has demonstrated that while similar consonant inventory sizes are obtained between parent report and those identified through naturalistic listening, larger inventories are obtained when phonetic transcription is used (Ramsdell et al., 2012). The expectation, then, might be that parents will recall consonants they have heard their child say repeatedly or clearly, and ignore consonants that are produced infrequently and/or without precision. Although this seems intuitive, our findings suggest that this may not always be the case. While the average number of consonants identified by coders more closely mirrored that reported by parents when a minimum of 10 productions were required for inclusion in the child’s consonant inventory, differences were evident in the actual consonants reported for many children. Parents and coders agreed on inventory size for nine children but agreed upon exact consonants for only one. In addition, while more than 50% of parents reported the presence of [b] and [d] in their child’s inventories, stops were coded for more than 50% of the children only when the criterion of 2+ occurrences was used. Interestingly, a number of children in this study produced more than 40 tokens of one or more consonants within a 40-minute recording session that were not included in the parent-reported inventory.

There are a number of reasons why consonants identified by coders may differ from those reported by parents. Sampling is always a concern when transcribing a young child’s vocal behavior (Ramsdell et al., 2012). Does the child’s spontaneous vocalizations represent the full repertoire of sounds they produce on a routine basis or only a subset? Because most vocalization samples reflect a small window of time in a child’s day, one might expect parents to report more consonants in their child’s inventory than would be reflected in a restricted sample. Why, then, would phonetic transcription of a young child reveal more consonants than a parent would report? Our findings revealed that parents were more likely to report nasals and stops in their child’s inventories than the glottal [h] or glides. In the absence of an intended word that begins with [h], it is easy to see why parents might ignore the production of that consonant in prelinguistic utterances since it is not a highly salient consonant. Further, it seems possible that [w] and [j] may be ignored by parents and perhaps overestimated in phonetic transcriptions of early vocalizations since both are produced not only as consonants but as transitions between vowels. So, a parent might include them as a consonant when they have heard a recognizable word that begins with [w] or [j] but exclude them in prelinguistic productions. Stops and nasals, on the other hand, are more salient productions- largely because they are the earliest consonants typically heard in babbling and are linked to early words that parents might be more responsive to. These findings are consistent with Ramsdell and colleagues (2012) who also reported that parents identified more productions containing obtruents (i.e., stops) and nasals compared to glides.

We performed a post-hoc analysis to determine whether a closer relationship was evident between coded and parent-reported consonants when only true consonants (defined as any consonant other than a glide (e.g., [j] & [w]) or glottal (e.g., [h]) were included in inventory. As expected, the mean number of true consonants reported by parents (4.69) more closely corresponded to the mean number identified by the coders at least twice (4.99) for the children. Despite the similarity in mean number of consonants, however, identical inventory sizes were evident for only 19 of the 70 (27%) children. When differences occurred, inventories were within two consonants for 28 of the remaining 51 children (55%). These differences could be related to sampling error (as discussed above) since the coders listened to a much smaller vocalization sample than parents are exposed to over time. Although parents were instructed to interact with their children while obtaining the vocalization sample, not all parents did so and so it seems likely that some consonants were not adequately stimulated through parent-child interaction and/or toys/items in the child’s environment.

Additional posthoc analyses were performed to examine differences between coded and parent report consonant inventories across the three ASQ problem solving category subgroups (on schedule, borderline, needs additional assessment) to examine the impact (if any) of developmental status. Mean size of phonetic inventory identified in the modified TimeStamper analysis were similar across all three ASQ subgroups (onschedule mean=7.94; borderline mean=7.73; additional assessment=7.50) as were those report by parents (onschedule mean=6.04; borderline=5.36;additional assessment=6.25). Group differences were also small when only true consonants were examined. Mean size of consonant inventory coded using the TimeStamper analysis was 5.00, 4.91, and 4.50 for the on schedule, borderline, and additional assessment groups, respectively. Mean size of consonant inventory reported by parents was 4.65, 4.00, and 4.88 for the on schedule, borderline, and additional assessment groups, respectively. So, developmental status did not appear to have an appreciable impact on either the coded or reported size of consonant inventory.

The participants in this study ranged in age from 14 to 18 months and thus differed in the size of their expressive vocabulary (range= 0–147; mean= 17). According to information provided by parents on the MacArthur Communicative Development Inventory (CDI: Toddler; Fenson et al., 2007) (available for 68 of the 70 children), the majority of children were producing one or more words (90%) and 29 (43%) had 10 or more words. When examining consonant inventory in this study, we coded consonants heard in both prelinguistic utterances and recognizable words. Although parents were instructed to identify consonants they had heard their child produce in babble or in words, it seems possible that some parents reported only those produced in words since they would be particularly salient, and because all examples on the parent checklist were only provided for words (not babble). Further, since only words containing each consonant in the word-initial position were provided as examples to parents, it is also possible that this may have led them to underreport sounds that occurred in the final (or medial) position. In any study relying on parent report of a child’s consonant inventory, it is important to recognize that recall of consonants in a child’s inventory may be particularly problematic for parents of children with large expressive vocabularies since recall can be impacted by memory constraints. Further, parent recall can at times be influenced by the spelling of a word. Although parents in this study were provided with explicit instruction to focus on the sound their child produces, it is always possible that some may have reported the letter that corresponds to the sound they have heard their child say (e.g., /k/ vs. “c” for “car” or /s/ vs “c” for “ice”).

Finally, it is important to acknowledge that young children with repaired cleft palate may demonstrate vocal characteristics that can impede a listener’s identification of specific consonants. Eshghi and her colleagues (2017) have reported that not all children with repaired CP achieve consistent velopharyngeal closure for stop consonants and vowels immediately following palatal surgery. Problems with velopharyngeal closure that occur consistently or inconsistently can lead to hypernasality as well as either audible nasal emission or weak pressure during production of oral stops and each of these attributes could contribute to differences among listeners in identification of the consonant produced.

Clinical Implications

Documenting the emergence of consonants in a young child’s inventory is important when determining the normal course of vocal development. As a child moves from babbling to production of early words, the expectation is that size of consonant inventory will continue to grow to permit production of new words. Although consonant inventory size is a useful measure of vocal development in young children, it is important to recognize that it is a gross measure that is influenced by many factors for the very young child- regardless of the method used to obtain that information. Because naturalistic listening identifies consonants that are produced frequently by a young child, it is useful in identifying consonants that are available for early word learning. It may be less useful in documenting the emergence of new consonants in children acquiring their single word expressive vocabulary. Documenting consonant inventory size using phonetic transcription also has limitations. Not only is the task time consuming, but the limited window of time spent transcribing a child’s productions may result in sampling error. So, both methods of estimating consonant inventory may underestimate the full inventory of consonants that a child produces. While clinicians may attempt to augment their clinical findings by asking parents to provide information about the consonants they have heard their child say, our findings suggest that some parents may ignore glottals (h) and glides in prelinguistic vocalizations. Since some parents may pay closer attention to consonants they have heard their child produce in recognizable words, especially at the beginning of words, reliability of parent recall may be enhanced by requesting consonant inventory information separately for babbling and word production and providing examples of the sound in different word positions. Confidence may be further increased by asking parents to provide an example of a word that they have heard their child say containing each consonant and then verbally confirming their perception of how the sounds in the word are pronounced.

Conclusions

Consonant inventories obtained using the modified NLRT approach and parent report were only moderately correlated in this study. Although the mean number of consonants in inventory between coders and parents was slightly closer using the 10+ versus 2+ criterion for consonant inclusion, the difference was not significant enough to recommend one protocol over the other. The use of parent report for purposes of screening and/or assessment of speech and language performance of young children is widely used for both clinical and research purposes. As with direct assessment of a young child’s speech and language skills, it has limitations that must be acknowledged. In-depth studies such as these are important as we were able to identify exactly where the agreements and disagreements occurred. For children with cleft palate, the emergence of stops is an important milestone post-surgery as the lack of stop development suggests that early intervention is needed and may also be an indicator of questionable velopharyngeal function (Hardin-Jones, et al., 2023). Fortunately, parents and coders agreed 87% of the time on the presence/absence of this sound category. As part of our ongoing longitudinal research, we will continue to follow these children until 3 years of age when we will examine their sound inventories and also examine how these early productions “predict” later speech and language performance.

Highlights.

1. Parent reported consonant inventories only moderately correlated with coders

2. High agreement between parents and coders on presence of stops and nasals

3. Poor agreement between parents and coders on presence of [h] and glides