On the role of syllable structure in atypical phonological development: Evidence from the acquisition of /l/ by Portuguese children

ABSTRACT

The alveolar lateral is phonetically and phonologically complex. Previous studies have shown that /l/ is one of the last segments to be acquired by typically developing Portuguese children. However, little is known about how Portuguese children with atypical development acquire /l/. In this paper, we investigate the acquisition of /l/ by Portuguese children with protracted phonological development (DLD; SSD). We explore the effect of syllable structure and segmental properties in the acquisition of /l/ and describe mismatches used for target /l/, thus contributing empirical evidence to the ongoing discussion on differential diagnoses for children with primary phonological disorders. Our results show that the lateral is more problematic in SSD than in DLD, with the manner of articulation being more problematic than its place. A syllable-segment interface effect was attested. Mismatches showed a preference for [w, ɾ, ø]. The results are discussed considering their implications for clinical practice and the role of target phonetic and phonological properties in the /l/ acquisition path.

KEYWORDS:

• Developmental language disorder
• speech sound disorders
• alveolar lateral
• mismatches
• European Portuguese

Introduction

The alveolar lateral is one of the four European Portuguese (EP) liquids (/l, ʎ, ɾ, ʀ/). As observed for other languages (Bernhardt & Stemberger, 1998; Fikkert, 1994; Kim et al., 2016; Smit et al., 1990), liquids tend to be the last natural class to become stable in EP child phonology (Amorim, 2021; Costa, 2010; Freitas, 1997; Mendes et al., 2013; Ramalho, 2017). In the case of /l/, although it is present in 80% of the world’s languages (Maddieson, 2013), its categorisation is challenging for children. This may be due to the co-occurrence of different phonetic and phonological properties. Phonetically, a plosive-like constriction at the place of articulation occurs. However, the oral resonance, visible in formant structures, is consistent with vocalic segments (Ladefoged & Maddieson, 1996); vowel-like productions for /l/ have been attested in several languages, including [w] in Brazilian Portuguese (Bisol, 1996). These vowel-like productions are used by some scholars as an argument favouring the proposal that laterals have the highest degree of sonority among consonants and are adjacent to vowels in phonetically based sonority scales, like the one in Ladefoged (1982). This proximity in sonority may contribute to /l/ complexity. Maximal sonority contrasts within the syllable are preferred by children (Goad & Rose, 2004; Jakobson, 1968). Unlike plosive+vowel sequences, the first universal syllable template for children’s early words, tautosyllabic lateral + vowel or vowel + lateral sequences do not assure a maximal sonority contrast. Moreover, difficulties with the acquisition of /l/ in EP may be related to syllable distribution: The fact that it may occur as singleton, in onset cluster, or in coda may enhance its complexity. The tendency for the scalar acquisition of /l/ in several languages (mostly, singleton >> coda >> onset cluster) matches the syllable – segment interface effect, largely demonstrated in typical child phonology (Bernhardt & Stemberger, 1998; Fikkert, 1994; Freitas, 1997), and also reported for atypical development (Almeida et al., 2019; Gallon et al., 2007; Kavitskaya et al., 2011; Martins et al., 2021; Tamburelli & Jones, 2013).

Little is known of the acquisition of /l/ by Portuguese children with atypical development. Reis (2021) interviewed 282 Portuguese Speech and Language Therapists (SLTs), who unanimously identified /l/ as the most resistant segment to intervention. We, therefore, examine the acquisition of /l/ by Portuguese children with Protracted Phonological Development (PPD).¹ In this paper, we provide empirical data on success rates and mismatches for target /l/ in different syllable positions, and compare results with the ones available in the literature for typically developing (TD) Portuguese children. Two types of diagnoses are considered: Speech Sound Disorders (SSD) and Developmental Language Disorder (DLD). Based on the inherent complexity of EP /l/, we discuss the implications of our results for clinical practice, contributing empirical data to the debate on the nature of SSD and DLD. Before delving into our study, we present relevant previous research on the acquisition of /l/ by Portuguese children with typical development and review the debate on DLD and SSD.

The alveolar lateral in EP

The alveolar lateral is one of the 19 possible phonological singletons in EP and the 9th most frequent consonant in the system (5%) (the most frequent is /t/ (13%); the least frequent is /ʎ/ (1%) (Costa, 2010). In onset, /l/ occurs in combinations of plosive + /l/ and fricative + /l/ clusters. In plosive + /l/ onsets, all plosives are possible as C1 except /d/; for fricative + /l/ sequences, only /f/ is possible as C1. The coda inventory in EP includes three consonants (/l/, /ɾ/, /s/). The lateral is the least frequent (27% of all codas; Vigário et al., 2010). Syllables with /l/ occur in stressed and unstressed positions, although unstressed ones are more frequent (64%; Frota et al., 2010). Singleton /l/ and onset clusters with /l/ are possible word-initially and word-medially; the coda lateral is possible word-medially and word-finally. EP /l/ displays two allophonic variants ([l, ɫ]), depending on syllabic distribution: [l] in onsets and [ɫ] in codas. Table 1 illustrates the syllabic distribution of /l/ in EP:

Table 1. Distribution of /l/ in EP according to the syllable structure.

[l] singleton	lupa ‘magnifier’ [ˈlupɐ]	bola ‘ball’ [ˈbɔlɐ]
[ɫ] coda	calças ‘trousers’ [ˈkaɫsɐʃ]	papel ‘paper’ [pɐˈpɛɫ]
[l] onset cluster	planta ‘plant’ [ˈplɐ̃tɐ]	biblioteca ‘library’ [bibljuˈtɛkɐ]

This categorical view of /l/ allophones has been questioned by phonetic studies. From an articulatory perspective, [l] shows an occlusion on the mid-sagittal line of the vocal tract, with the apex of the tongue touching the alveolar region; the airflow passes along one or both lateral areas of the tongue. The velarised variant [ɫ] preserves all properties of [l], but a retraction of the tongue’s posterior area is added. The constriction is weaker in coda, attesting to the possible production of a labial glide (Oliveira et al., 2011). Acoustic studies have shown that the lateral exhibits velarisation in all syllable positions, although the degree of velarisation is stronger in coda than in onset (Oliveira et al., 2011; Rodrigues et al., 2019). Rodrigues et al. (2019) showed that, despite the close values for F2 in both allophones (about 1000 Hz), F3 (another acoustic correlate of velarisation) is significantly higher in coda than in onset, which favours the phonological analysis that assumes different allophones for onset ([l]) and coda ([ɫ]).

The acquisition of the alveolar lateral by Portuguese children

As mentioned above, liquids are reported as problematic for children worldwide. Results from different languages, including EP, show the impact of the syllable – segment interface in the acquisition of liquids, both in typical (Bernhardt & Stemberger, 1998; Fikkert, 1994; Freitas, 1997) and in atypical development (Almeida et al., 2019; Gallon et al., 2007; Kavitskaya et al., 2011; Martins et al., 2021; Tamburelli & Jones, 2013). In EP, /ɾ/ and /l/ are the only segments licenced in three syllable positions, i.e., singleton, coda and onset cluster. Recent studies have shown that /ɾ/ reaches high success rates before /l/ in TD Portuguese children. Specifically, singleton /ɾ/ is acquired at 4;1–5;0; coda /ɾ/ is acquired but not stabilised at 6;06; and in the onset cluster, although the success rates are always higher than those observed for /l/, /ɾ/ is not acquired by the oldest group of children observed (Ramalho, 2017). Regarding the alveolar lateral, studies on EP have shown that it is acquired after 5;0 (Amorim, 2021; Mendes et al., 2013; Ramalho, 2017). Table 2 provides information on accuracy rates for /l/, by syllable position and age group, in Ramalho (2017) and Freitas et al. (2022); the sample contains 87 TD children organised by age group (G1: 3;0–4;0; G2: 4;1–5;0; G1: 5;1–6;06). These two studies used the methodological procedures adopted in the current study, as developed in our Method section.

Table 2. Accuracy rates for /l/ in Ramalho (2017) and Freitas et al. (2022).

	G1	G2	G3
[l] singleton	63.5%	62.4%	66.1%
[ɫ] coda	33.5%	46.1%	53.7%
[l] onset cluster	8.3%	24.7%	35.9%

Most studies on child phonology focus on match rates. Data on mismatches used for target /l/ are scarce. Bernhardt and Stemberger (1998) point out that the feature [lateral] is universally considered to be late acquired, with [d, n, j, w] as the most frequent mismatches. Few studies on L1 acquisition in EP describe mismatches. Amorim and Veloso (2021) report the following mismatches for singleton /l/: [w] in 56% and [g] in 32% of all mismatches; these are more frequent word-medially. Gliding was also attested by Guerreiro (2007), mostly for word-medial codas (48%), as well as word-medial coda deletion and word-final vowel insertion; for onset clusters, she reports cluster reduction (with /l/ deletion), particularly in fricative+lateral sequences, and vowel insertion in C /l/ clusters, already attested in Freitas (1997) for EP. Mendes et al. (2013) propose a typology of phonological processes affecting liquids that includes gliding, final consonant omission and consonant cluster reduction. Lousada (2012) set the offset of these processes at 6;0–6;11, which expresses the late acquisition of liquids in EP. Freitas and colleagues (2022) listed the following mismatches for target /l/ in TD Portuguese children: [w, Ø, ʋ] for singletons and codas, and vowel epenthesis and [Ø, w] for C2 in onset clusters.

On the nature of DLD and SSD

Children may show exclusively phonological difficulties or difficulties in language domains other than phonology. Sometimes, these difficulties are associated with a biomedical condition, such as intellectual development disorders or hearing loss. Nevertheless, in most cases, language problems occur without a known cause, indicating a primary disorder (Shriberg et al., 2010; Waring & Knight, 2014).

Phonological alterations are common to several disorders and may occur in association with SSD, DLD or a specific learning disorder (Ramalho et al., 2022). Disorders related to language development have been described in the literature since the mid-19th century. The term Language Development Disorder was used in Rapin and Allen’s (1983) landmark publication. The authors described a role of ‘syndromes’ whose common denominator was linguistic alteration: i) auditory verbal agnosia; ii) verbal dyspraxia; iii) phonological programming deficit syndrome; iv) lexical-syntactic deficit syndrome; v) phonological-syntactic programming deficit syndrome; and vi) semantic-pragmatic deficit syndrome. The phonological domain is involved in perceptual aspects or in the way this knowledge impacts the children’s productions (e.g., subtypes ii), iii) and iv)). Currently, children with the ii) and iii) profiles are labelled as having SSD.

Since Rapin and Allen (1987), various terminological proposals have been discussed. The most recent terminology assumes the possibility of a phonological language development disorder (Bishop, 2017). Although several authors have described diagnostic criteria, there is no consensus on the criteria and their application, making it difficult to separate DLD from SSD. The existence of linguistic alterations outside the phonological domain and the persistence of phonological difficulties after 5;0 have been pointed out as crucial factors for differential diagnosis (Bishop, 2017; Hayiou‐Thomas et al., 2017); these factors are associated with a high severity degree and, therefore, with DLD. However, some authors have suggested that phonological difficulties may persist beyond the typical period of phonetic-phonological development, linking them to persistent SSD (e.g., Shriberg et al., 2010).

Although SSD constitutes a heterogeneous group, children with SSD tend to show a delayed initial acquisition of speech sounds compared to children with TD (Shriberg et al., 1997). SSD may present different levels of severity, underlying causes and/or types of mismatches, with or without impairment of other linguistic domains and with different types of response to treatment (Bowen, 2014; Dodd, 2011). To focus the discussion on the impairment of phonological development, the term PPD was proposed (e.g., Mason et al., 2015), with the aim of emphasising the analysis of phonological changes, challenging the traditional taxonomic classifications (DLD versus SSD). The possibility of the existence of a continuum of phonological difficulties has also been suggested; therefore, it can be assumed that there is a greater severity of extra-phonological linguistic difficulties associated with DLD and a lower severity associated with SSD (IALP, 2021; Ramalho et al., 2022; Stringer et al., 2023).

Various proposals for classifying SSD were presented, with Dodd et al.’s (2005) being one of the most used. The author focuses on the typology of mismatches, which may reflect difficulties at different underlying levels of processing. A model with five sub-types of SSD is proposed for differential diagnosis (see Dodd et al., 2005, for further details). Additionally, there is Shriberg’s classification, which is primarily a medical aetiology-based concept (Shriberg et al., 1997, 2010). This classification incorporates both medical-diagnostic and descriptive-linguistic data, based on clinical data, and considers speech and hearing mechanisms, cognitive-linguistic performance, and psychosocial characteristics. This classification includes three major categories: speech delay (phonological), speech disorder (mostly articulatory phonetic difficulties), and motor speech disorder (including apraxia and dysarthria).

The debate on the place of SSD and phonological DLD has been reignited since Bishop (2017; IALP, 2021; Stringer et al., 2023). Stringer et al. (2023) discuss how the terminology associated with SSD has evolved over time, becoming increasingly specific as theoretical and clinical knowledge in the field has increased. The authors argue that Bishop's (2017) proposal has limited clinical applicability in a clinical context and overlaps with existing terminology. Furthermore, they suggest that children with DLD with phonological difficulties should be diagnosed with a co-occurring diagnosis of SSD when they also have observable changes in production. Therefore, the discussion on the nature of children’s phonological representations and their impact on the description of linguistic profiles and the use of diagnostic labels continues to be relevant.

Purpose of the current study

In this paper, we focus on how monolingual Portuguese children with atypical development acquire the alveolar lateral. Underlying the focus on /l/ is the fact that it is currently considered by Portuguese SLTs as the most resistant segment to clinical intervention. Our main goals are as follows: (i) to explore the effect of the syllable – segment interface in the acquisition of /l/ by Portuguese children with atypical development; (ii) to use mismatches as windows to /l/ phonological processing in child phonology; (iii) to contribute with empirical evidence to the ongoing debate on differential diagnoses in children with primary phonological disorders. To achieve our goals, we describe the acquisition of /l/ according to syllable constituency (singleton, coda, and onset cluster) and the manner and place of articulation (respectively, MoA and PoA). We look at types of mismatches to build hypotheses on the representation of /l/ in child phonology. Finally, differences and/or similarities between children with DLD and SSD are explored, aiming to contribute to the ongoing debate on differential language diagnoses and the implications for clinical intervention. Our results also shed some light on the role of /l/ as a potential clinical marker in EP.

Method

Participants

The data presented in this study are stored in the Phonodis corpus (Freitas et al., 2019), which is available at PhonBank (MacWhinney & Rose, 2014). The data in the corpus was collected using the Crosslinguistic Child Phonology Project – European Portuguese assessment tool (Ramalho et al., 2014; available at http://www.clul.ul.pt).² Currently, PhonoDis stores production data from 26 Portuguese children diagnosed with different types of speech and language disorders, all EP monolinguals. All children were followed up in speech and language therapy and were diagnosed using a language assessment test available for EP and standardised for their chronological age (i.e., Teste de Avaliação da Linguagem na Criança (Sua-Kay & Tavares, 2007)). No evidence of other sensory, emotional, neurological, or cognitive disorders was found, and there were no significant abnormalities in oral-motor structure as evaluated by SLTs. In the current study, only data from children with DLD and children with SSD were analysed. We examined productions for /l/ by eight Portuguese children with DLD, aged 4;00 to 8;03, and seven with SSD, aged 3;02 to 6;00. To provide a complete characterisation of each child’s clinical profile, the mean correct consonants (PCC) and the mean CV match per child are presented below. See Tables 3 and 4 for more details:

Table 3. Characteristics of the participating children with DLD.

Code (gender: M/F)	Age	Time since SLT intervention	PCC	CV Match
D (M)	4;11	February 2017	58.7%	18.5%
F (M)	8;03	Started intervention in March 2015 (first intervention in speech therapy)	65.5%	40.1%
G (M)	7;09	October 2018 (had already started speech and language therapy with another therapist in Dec. 2017)	69.9%	33.1%
I (F)	6;10	Started SLT at 4;06	58.5%	28.7%
L (F)	7;06.10	Started SLT at 3;03	64.7%	36.9%
N (M)	4;00.24	Started SLT at 3;06	82.7%	70.7%
O (M)	4;09	Started SLT at 3;00	55.5%	24.2%
R (M)	5;11.27	Started SLT at 4;00	70%	24.2%

Table 4. Characteristics of the participating children with SSD.

Code (gender: M/F)	Age	Time since SLT intervention	PCC	CV Match
K (F)	5;09	Started SLT at 3;03	49.5%	20.4%
M (M)	4;00.09	Started SLT at 3;00	40.9%	16.1%
Q (M)	3;02	Started SLT at 3;00	28%	10.8%
S (M)	05;02.29	Started SLT at 3;06	76.8%	58.6%
T (M)	5;03	Started SLT at 4;00	62.2%	30.6%
U (M)	05;05.07	Started SLT at 3;06	61.7%	23.7%
V (M)	6;00	Started SLT at 3;06	53.3%	26.9%

In the DLD group, the range of the PCC rate is 55.5% to 82.7%. CV match also shows a wide range of percentages (18.5% to 70.7%). Similar rate amplitudes are observed for SSD (PCC: 28% to 76.8%; CV match: 10.8% to 58.6%).

Procedures

The assessment tool used for data collection was the Crosslinguistic Child Phonology Project – European Portuguese (CLCP-EP; Ramalho et al., 2014). The CLCP-EP is available via open access and was validated for EP by Ramalho (2017). The tool contains drawings that elicit the story of Cenourinha, a bunny who lives with a family of humans. The tool is composed of 150 target lexical stimuli, grouped into 42 thematic scenarios. The lexical targets are controlled for segmental and prosodic properties (segmental inventory, syllable structure, word stress, position within the word, word length). Our data collection procedures included i) informed consent according to the CLCP project guidelines; ii) 42 digital images presented to each child individually, on a tablet or computer screen, in a picture naming format; and iii) recordings made in the therapy session by the child’s SLT. The process is as follows: The child is led to produce the target lexical items by finishing the sentences initiated by the SLT, whereby the aim is to obtain spontaneous naming. However, if the child does not know the word, first semantic and then phonological cues are provided; if both strategies fail, the child is asked to repeat the word after the SLT. The task takes about 10 to 15 minutes.

All productions of lexical targets with /l/ were extracted from PhonoDis. The first author, a linguist and native speaker of EP, transcribed all the audio files in the corpus for children in our sample, with the assistance of the Phon program (Rose et al., 2006). For phonetic transcription that did not match the ones initially available in the corpus, a review was performed by the second and third authors, an STL and a linguist, both native speakers of EP with extensive experience in phonetic transcription. In cases of disagreement, the tokens (a total of 6) were excluded from the analysis. Tokens with a noisy background or tokens in which the child’s speech overlapped with the adult’s speech were not considered for data analysis. Moreover, only spontaneous productions were analysed. A total of 597 tokens were considered (see Table 5).

Table 5. Number of target tokens by syllable constituent and by clinical group.

	[l] singleton	[ɫ] coda	[l] onset cluster
DLD	177	90	58
SSD	146	77	49

Statistical analysis was conducted in R (version 4.3.1, R Core Team 2023). To minimise error variance, we applied the arcsine transformation to global accuracy proportions before running statistical analysis (Ambridge & Rowland, 2013).

For age thresholds, we adopted the criteria used in Ramalho (2017), our reference study for /l/ acquisition in EP: (i) success rates from 0% to 50% – segment not acquired; (ii) success rates from 51% to 75% – segment under acquisition; (iii) success rates from 76% to 85% – segment acquired but not stabilised; (iv) success rates from 86% to 100% – segment stabilised.

To test the impact of MoA and PoA in the acquisition of /l/, we compared its success rates with the ones for /t/ and /ʎ/: (i) the contrast /l/–/t/ (similar PoA; different MoA) provided data on the impact of MoA; (ii) the contrast /l/–/ʎ/ (similar MoA; different PoA) provided data on the impact of PoA. This contrastive analysis was performed only for singleton /l/, due to the low success rates for /l/ in coda and in onset cluster. The total number of tokens used for /t/ was: DLD = 383 and SSD = 335; the total of tokens for /ʎ/ was: DLD = 52 and SSD = 44.

Given (i) the paucity of data on the phonological profile of Portuguese children with primary phonological disorders and (ii) the non-homogeneous samples, we describe the results per child with PPD and compare them with those of age-matched TD children, collected with the same test and data collection criteria (cf. Ramalho, 2017). For instance, child Q (3;02 years old), with SSD, is compared with the children of G1 ([3;00–4;00]) from Ramalho (2017). Children D, N, and O, with DLD, and child M, with SSD, are between 4;00 and 4;09 years old; their results are compared with those of G2 ([4;00–5;00]) from Ramalho (2017). As children F, G, I, L, R, with DLD, and K, S, T, U and V, with SSD, are older than 5;00 years, their results are compared with those of G3 ([5;00–6;06]) with TD.

Results

The results are reported first in terms of overall accuracy for each of the two clinical groups and then in terms of individual patterns. For the individual analysis, the success rates are compared with the children’s chronological pairs in Ramalho (2017; for further details see the Method section).

Global accuracy

Success rate for /l/

Figure 1 illustrates the accuracy rates across syllable positions for the two clinical groups. The mismatches employed for /l/ by each group are presented in Table 6.

Figure 1. Match rates for /l/ across syllable positions by clinical group.

Table 6. Mismatches for /l/ across syllable positions by clinical group (%).

	[l] Singleton		[ɫ] Coda		[l] Onset cluster
DLD	[w]	33.73	[w]	41.94	C2 deletion	47.37
	∅	19.28	∅	35.48	[w]	23.68
	[ɾ]	15.66	[ɾ]	12.9	Others	28.95
	Others	31.33	Others	9.68
SSD	[w]	34.95	∅	37.68	C2 deletion	45.45
	∅	19.42	[w]	33.33	[ɾ]	18.18
	[ɾ]	18.45	[ɾ]	11.6	[w]	16.37
	Others	27.18	Others	17.39	Others	20

The results demonstrate different match rates according to syllable constituency and a tendency for higher accuracy in singleton in the two clinical groups (DLD: 44.63%; SSD: 39.73%); however, neither group reached 50% accuracy in this syllable position. No significant effects were found for singleton /l/ between children with DLD and those with SSD (Welch’s t-test, t(10.34) = 0.52, p = 0.611, 95%CI[−0.34, 0.54]).

In coda position, a major difficulty in the production of /l/ was attested, especially in the clinical group with SSD, with children showing poorer performance compared to the singleton position (paired t-test, t(5) = 5.67, p = 0.002, 95%CI[0.31, 0.82]). A significant difference in the performance of the two clinical groups was attested in coda position (Welch’s t-test, t(12.37) = 2.79, p = 0.015, 95%CI[0.1, 0.65]).

With respect to /l/ in onset cluster, both clinical groups showed the lowest accuracy (for DLD = 18.97%, for SSD = 6.12%). The two groups did not differ significantly (Welch’s t-test: t(9.99) = 1.38, p = 0.197, 95%CI[−0.14, 0.59]).

Mismatches

When failing to produce singleton /l/, the children with SSD and DLD used similar strategies, with the same pattern of occurrence: [w], deletion (∅), and [ɾ]. The preferred mismatches for coda /l/ were also similar ([w], [∅], and [ɾ]) in both clinical groups, and occurred in the same proportion.

Regarding the onset cluster, C2 deletion was the most frequent mismatch. Additionally, [w] was used in both groups, although more frequently for DLD. The use of [ɾ] was attested only in the SSD group.

Manner and place of articulation: Success rates for singleton /l/, /t/, and /ʎ /

Considering the low success rates for /l/ in coda and onset cluster in both clinical groups, the contrastive analysis to test the impact of MoA and PoA was only performed for singletons (see the Method section). From now on, we compare the singleton /l/ success rates with the ones for /t/ (for MoA effects) and /ʎ/ (for PoA effects). Figure 2 illustrates the accuracy rates for the three segments in singleton.

Figure 2. Match rates for singleton /l, t, ʎ/ by clinical group.

Regarding the effect of MoA on singleton position, Figure 2 shows that both clinical groups had accuracy rates for /t/ above 80% (DLD = 89%; SSD = 81.5%), showing that the segment is acquired. No significant effects between the two groups were attested (Welch’s t-test: t(9.05) = 0.96, p = 0.363, 95%CI[−0.1, 0.3]). For both groups, a major difficulty was observed for the palatal /ʎ/. Significant group differences were observed (Welch’s t-test: t(12.75) = 2.51, p = 0.026, 95%CI[0.1, 0.9]), showing lower performance in the group of children with SSD (20.5%) when compared to the group of children with DLD (44.63%). We will come back to the contrasts for MoA and PoA in the Discussion section.

Individual accuracy

The individual analysis is performed only for singleton /l/, due to the low match rates for /l/ in coda and in onset cluster. Again, to test the possible effects of MoA and PoA in the acquisition of/l/, the success rates for /t/ and /ʎ/ are presented. The results are compared with data available in the literature and collected from TD chronological pairs using the same test (cf. Ramalho, 2017; see Method).

Figure 3 shows the results for child Q’s accuracy rates and for the TD children matched in age (G1: [3;00–4;00]).

Figure 3. Match rates for singleton /l, t, ʎ/ by group (DLD and TD age-matched (G1: [3;00–4;00]).

Overall, child Q showed a lower performance in all three segments when compared to the age-matched TD children: only target-like productions for /t/ were attested, with a success rate below 70%.

Figure 4 (DLD) and Figure 5 (SSD) display match rates for singleton /l, t, ʎ/ by the children with PPD and their age-matched TD pairs (G2: [4;00–5;00]).

Figure 4. Match rates for singleton /l, t, ʎ/ by group (DLD and TD age-matched (G2: [4;00–5;00]).

Figure 5. Match rates for singleton /l, t, ʎ/ by group (SSD and TD age-matched (G2: [4;00–5;00]).

For all children, the results revealed a major difficulty in producing /l/ and /ʎ/. However, child M, with SSD, achieved a match rate for /l/ (50%) close to that of the age-matched TD pairs (62.4%). Moreover, child N, with DLD, also achieved a match rate for /ʎ/ (57.1%) similar to that achieved by age-matched TD children (62.9%). In both clinical groups, all the children achieved higher match rates for /t/, with target-like productions of over 70%.

Figures 6 and 7 show the results for the children with DLD and SSD and their age-matched TD children (G3: [5:00–6:00]).

Figure 6. Match rates for singleton /l, t, ʎ/ by group (DLD and TD age-matched (G3: [5;00–6;06]).

Figure 7. Match rates for singleton /l, t, ʎ/ by group (SSD and TD age-matched (G3: [5;00–6;06]).

The oldest participants with DLD (subjects G, I and L³) and SSD (K and V) achieved match rates of over 60% for /l/. As a result, these children obtained similar or even higher success rates than the G3 age group [5;00–6;06], who registered a mean rate of 66.1% for singleton /l/ (see Figures 6 and 7). For both clinical groups, a tendency towards a higher accuracy for /t/ was attested (above 70%). Concerning /ʎ/, in general, children with DLD achieved higher success rates when compared to children with SSD: children F, G and L (with DLD) stand out, with success rates higher than 65%.

Discussion

In this paper, we focus on how monolingual Portuguese children with atypical development acquire the alveolar lateral. Underlying this research question is the fact that /l/ is currently referred to by Portuguese SLTs (Reis, 2021) as the most resistant segment to clinical intervention. Considering the effect of the syllable – segment interface in phonological development (e.g., Almeida et al., 2019; Bernhardt & Stemberger, 1998; Fikkert, 1994; Gallon et al., 2007; Kavitskaya et al., 2011; Martins et al., 2021; Tamburelli & Jones, 2013), we tested the role of syllable structure in our atypical samples. Singleton and coda /l/ as well as C + /l/ clusters produced by monolingual Portuguese children in two clinical groups (DLD and SSD) were described and compared with those of their TD pairs, whose data was collected with the same assessment tool and similar methodological procedures (Ramalho, 2017).

Global accuracy rates showed more success for singleton /l/ than for coda or onset cluster (see Figure 1). However, the mean rates were always below 50%, demonstrating that /l/ is a complex segment in all syllable positions. Note that low success rates were also attested among TD Portuguese children for /l/ in all syllable positions (Freitas et al., 2022; Ramalho, 2017; see Table 2): mean rates ranged from 62.4% to 66.1% for singleton /l/, with no development attested across the three age groups (total time interval: 3;00–6;06); as for coda /l/, and only in the last age group (G3: [5;0–6;06]) did children show a mean rate above 50%; in onset cluster, /l/ was not acquired for all age groups (G1, G2, G3). Therefore, the data from Portuguese children with typical and atypical development corroborate the difficulties expressed by the Portuguese SLTs interviewed in Reis (2021). The syllable – segment interface was attested in our SSD and DLD groups, matching the behaviour of TD Portuguese children (Amorim, 2021; Freitas, 1997; Ramalho, 2017). Furthermore, as observed across languages, our data provide additional evidence to support the difficulty of mastering liquid sounds by children with both TD and PPD (Almeida et al., 2019; Fikkert, 1994; Gallon et al., 2007; Kavitskaya et al., 2011; Kim et al., 2016; Martins et al., 2021; Smit et al., 1990; Tamburelli & Jones, 2013).

The only statistically significant difference concerning the data in Figure 1 was attested for DLD and SSD coda /l/ rates. The DLD group performed better (31.11%) than the SSD group (6.67%), suggesting that the coda lateral may be a good candidate for testing as a clinical marker in EP, i.e., the use of /l/ in coda may allow Portuguese SLTs to distinguish children with DLD from children with SSD. However, it should be noted that (i) our samples are small, and (ii) the two groups of children were of different ages and had different intervention times. Therefore, more data are needed to test this hypothesis (larger and controlled samples of Portuguese children with atypical development), since age (Montgomery, 2005) and clinical intervention (Broomfield & Dodd, 2011) impact language development in children with phonological deficits, which may have constrained the research results.

Due to the general difficulties with /l/ across syllable constituents, we decided to inspect the impact of internal segmental properties in its acquisition. We selected singleton /l/, the target gathering higher success rates in our data, and compared it with two other targets, singletons /t/ and /ʎ/. The pair /l/–/t/ shares the dento-alveolar PoA and was used to test an MoA effect (lateral versus plosive). The pair /l/–/ʎ/ shares the lateral MoA and was used to test a PoA effect (dento-alveolar versus palatal). Global accuracy (Figure 2) showed that MoA was harder to process than PoA (see the higher rates for /t/, above 80%, and the low rates for /ʎ/, below 50%, in both clinical groups). Despite the individual differences inherent to the subjects in our non-homogeneous samples (for instance, age and time since the first SLT intervention), children in both clinical groups revealed similar trends. The same PoA and MoA effects were observed for their TD pairs (G1 in Figure 3; G2 in Figures 4 and 5; G3 in Figures 6 and 7). Higher success rates for /t/ than for /l/ showed that the dento-alveolar PoA does not explain the difficulties attested. This result matches the unmarked nature of the coronal [+anterior] PoA mentioned in phonological models (Paradis & Prunet, 1991), and assumed in EP phonology (Mateus & Andrade, 2000). Low rates for /ʎ/ in most children with SSD and DLD, as well as for laterals, in general, show that MoA, and not PoA, plays a negative role in the acquisition of EP /l/.

The individual analysis and the results for both laterals (see Figures 3–7) challenge the generalisation based on global accuracy rates: Most of the children with DLD showed a better performance with laterals than the children with SSD. This emphasises the importance of providing global and individual rates when studying clinical samples. Again, the DLD group shows better performance when compared to the SSD group. However, this asymmetry needs to be tested with larger samples.

The mismatches used by the children in our study (Table 6) showed similar tendencies across clinical groups. The segmental preference for [w] across syllable constituents argues for the processing of /l/ as a vowel-like segment by the children in our samples, with a high degree of sonority, bringing it close to laterals in a phonetically based sonority scale (Ladefoged, 1982). This processing may be triggered by the oral resonance in the acoustic input (Bisol, 1996; Ladefoged & Maddieson, 1996). The preference for [w, ɾ] contrasts with the residual or absent use of plosives in our samples. Note that the preference for [g, w] is described by Amorim and Veloso (2021) for EP; Bernhardt and Stemberger (1998) report the use of [d, n, j, w] for target /l/ in different languages. The preference for [w, ɾ] shows that the children in our samples are processing /l/ as a sonorant, not a plosive (which could be triggered by the occlusion in the dento-alveolar articulation). The target minimal sonority contrast is being processed by our children in tautosyllabic lateral+vowel sequences. Since maximal sonority contrasts are universally preferred from the early stages of production (Goad & Rose, 2004; Jakobson, 1968), the presence of a minimal sonority contrast in these target sequences may delay the acquisition of /l/.

Similar mismatch tendencies ([w, ɾ, Ø]) were attested for singleton and coda /l/, although deletion was higher in coda than in singleton /l/. The highest rates of deletion were attested in onset cluster, where [w] was also possible. Also, the higher deletion rates in coda and in onset cluster may be due to syllable development and the gradual emergence of the syllable constituents in child grammar: The acquisition of one specific segment with different syllable roles is rarely categorical and tends to follow the order singleton >> coda >> onset cluster, illustrating the effect of the syllable – segment interface in phonological development (e.g., Almeida et al., 2019; Bernhardt & Stemberger, 1998; Fikkert, 1994; Freitas, 1997; Ramalho, 2017). The mismatch [ɾ] was used in our samples more frequently for singleton than for coda /l/, and it was residual or absent in onset cluster. This scalar usage of [ɾ], a good candidate to replace /l/ due to its similar presence in all three syllable positions in EP phonology, is also consistent with the acquisition order above ([ɾ] for singleton /l/> [ɾ] for coda /l/ > [ɾ] for /l/ in onset cluster).

A constellation of phonetic and phonological properties associated with EP /l/ may underline the delay in its acquisition. Despite global tendencies for success rates and mismatches, we were able to identify two measurements showing better performance in the DLD sample when compared to the SSD sample: (i) a significant statistical difference for coda match rates and (ii) a general tendency towards better performance for singleton laterals.

The present study has relevant implications for clinical practice in terms of not only clinical diagnosis but also target selection for intervention. Our research confirms that phonologically complex structures are valid clues for identifying children with primary phonological disorders. Our findings suggest that the lateral coda is a possible reliable clinical marker of a phonological deficit in EP, to be further tested in larger samples of Portuguese children with atypical development. Consequently, coda /l/ may be used by clinicians as a differentiator between children with phonological deficit and children with TD.

Furthermore, our study suggests that, especially for children with SSD, the MoA (lateral) is more difficult to process than the PoA (dento-alveolar) (see scores for /l, t, ʎ/ in Figures 2–7), and that this may extend to other segments, although further research is needed to substantiate this hypothesis. Therefore, our results suggest that SLTs may use the [lateral] feature to distinguish children with DLD from children with SSD, pointing to a fruitful area for future research. Thus, SLTs may have to spend more time and effort helping children with SSD to process the [lateral] feature.

Conclusion

As suggested by Portuguese SLTs, our findings show that EP /l/ is a problematic segment for Portuguese children with atypical development. As observed across languages, our data highlight the importance of using the phonological hierarchy in assessment and intervention: (i) an effect of syllable structure was attested, confirming the importance of considering the syllable – segment interface in clinical assessment and intervention; predominant mismatch patterns reflected negative structural constraints (deletion of C2 or deletion of coda); (ii) an effect of internal segmental properties was attested, showing that MoA is more difficult to process than PoA.

The match rates suggested that the coda /l/ is more difficult for children with SSD than for children with DLD. Thereby, it appeared to be the best candidate for a clinical marker to differentiate children with DLD from children with SSD in EP. The results also revealed that MoA plays a more significant role in the performance of children with SSD than in children with DLD, indicating a potential clinical difference between the two samples. Our results suggest that clinical intervention needs to consider both segmental (e.g., feature cooccurrence) and prosodic (e.g., syllable structure) properties, since they contribute to discriminate phonological profiles (typical versus atypical; DLD versus SSD). Nevertheless, larger samples of Portuguese children with atypical development will be needed to test the adequacy of our generalisations.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the FCT – Fundação para a Ciência e a Tecnologia, I.P., under project UIDB/00214/2020.

Notes

¹ PPD emphasises the perspective that phonological development can extend over a long period of time, with the possibility of improvements, regardless of the clinical labels SSD or DLD (Mason et al., 2015).

² The assessment tool used follows the criteria adopted by the Crosslinguistic Child Phonology Project, coordinated by Bernhardt and Stemberger.

³ Note that subjects G (7;09), I (6;10), L (7;06) are older than children from G3 (Ramalho, 2017).