Individual profiles in protracted phonological development across languages: introduction to the special issue

ABSTRACT

Although group studies provide necessary information about the range and frequency of phenomena in phonological development, individual profiles (case studies) can be used to describe entire phonological systems in detail. Profiles from different languages can highlight similarities and differences across languages that may be less obvious in group studies. The current issue presents profiles of children with protracted phonological development (PPD: speech sound disorders) from 16 languages (Akan, Kuwaiti Arabic, Bulgarian, Canadian English, Farsi, Canadian French, German, Greek, Icelandic, Japanese, Mandarin, Polish, European Portuguese, Slovenian, Granada Spanish, Swedish). Utilising a constraints-based nonlinear phonological framework, each profile describes a child’s strengths and needs in word structure, segments, features and their interactions and suggests an intervention plan. Where available, follow-up data from after clinical intervention are included. This introductory paper provides the theoretical background for the papers and reflects on the findings, drawing out particular themes and implications for phonological and developmental theories and clinical intervention.

Video Abstract

Read the transcript

Watch the video on Vimeo

©2022 Taylor & Francis Group, LLC

KEYWORDS:

• Phonological development
• phonological acquisition
• phonological complexity
• crosslinguistic
• speech sound disorders
• case studies

Introduction

This special issue presents individual profiles in protracted phonological development (PPD)¹ across 16 languages (listed in the following paragraph) using a constraints-based nonlinear phonological framework. This introductory paper is a companion to each of the articles, providing background on the theories and methods underlying the papers, and drawing out themes and implications for theory and clinical practice from the articles.

In phonological development research, group studies with children are crucially needed to determine general developmental patterns and timelines. However, such studies generally focus on specific aspects of phonological development, possibly missing key interactions between various components. Individual profiles in contrast provide opportunities to delve deeply into one phonological system in its entirety. Although individual profiles go at least as far back as the diary study of Louis XIII’s speech development (Héroard, 1601), few profiles in PPD are available for languages other than English, and those that exist differ in scope, structure, methodology, and theoretical approach. By adopting a uniform theoretical framework and basic methodology, the individual profiles presented in this special issue provide a solid basis for identifying similarities and differences across children speaking a range of diverse languages. The language families and specific languages represented in the issue are listed below:

Table

(1) Indo-European:
	(a) Germanic:	North (Icelandic, Swedish); West (Canadian English, German)
	(b) Hellenic:	Greek
	(c) Indo-Aryan:	Farsi
	(d) Romance:	Canadian French, European Portuguese, Granada Spanish
	(e) Slavic:	East (Polish), South (Bulgarian, Slovenian)
(2) Isolate:	Japanese
(3) Niger-Congo:	Akan
(4) Semitic:	Kuwaiti Arabic
(5) Sino-Tibetan:	Mandarin

Table 1 presents key characteristics of the children with PPD by language and authors. Most children were monolingual (if a child was exposed to a second language, this is acknowledged in the article; e.g. the Canadian French-speaking four-year-old was starting to learn English). Although most children had no other major developmental challenges, one child with Down Syndrome was included in the set (a Kuwaiti Arabic speaker), in order to underline the perspective that PPD is not restricted to people with otherwise typical development. In terms of degree of PPD (‘severity’), the mean Whole Word Match² for children in the papers was 14% (SD 11.2%: range 3.7%-38.7%), i.e. overall in the more severe range for PPD across children by age (according to Bernhardt et al., 2020).

Table 1. Salient characteristics of the individual profiles across languages.

Domain(s) of interest (of several)	Foci of interest (of several)	Language	Author(s)	Follow-up data?
Word shape	WI C deletion WM [j] default	Swedish	Lundeborg Hammarström and Stemberger (2022)
Segmental	Stops < other consonants	Arabic (Kuwait)	Ayyad (2022)
	Alveolar obstruents	Japanese	Ueda and Bernhardt (2022)
	[θ] default	German	B. M. Bernhardt and Ullrich (2022)
	[t] default	French (Canada)	Bérubé and Spoor (2022)	Intervention, nonlinear basis
Structure-segment interactions	Onset [h/ʔ] default	English (Canada)	Chung et al. (2022)	Nonlinear intervention study
	Onset [ʔ] default	Mandarin (Shanghai)	Bernhardt et al. (2022)
	Unstressed initial syllables	Bulgarian	Ignatova et al. (2022)
	Complex Cs Syllabic Cs Sequences	Akan	Amoako and Stemberger (2022)
	V: for non- initial CC WI CC	Icelandic	Másdóttir and Bernhardt (2022)	Intervention, nonlinear basis.
	[-vd] > [+] in WM codas Default stops CC reduction	Farsi (Iran)	Shooshtaryzadeh and Stemberger (2022)
	Sequences: VV, CC C_C	Spanish (Granada, Spain)	Pérez et al. (2022)
	Sequences Coronal combinations Multisyllabic words	Portuguese (Portugal)	Ramalho et al. (2022)
	Fricative coda constraint Sequences	Slovenian	Ozbič and Bernhardt (2022)	Intervention, nonlinear basis.
	Alveolars Clusters Imprecision	Polish	Zydorowicz (2022)
	Rhotic in CC Obstruent-Sonorant CC	Greek	Babatsouli and Geronikou (2022)	Five-month follow-up

WI = word-initial; WM = word-medial; WF = word-final; C = consonant; CC = consonant cluster; V = vowel; [vd] = [voiced].

Each article has the same structure. A short introduction highlights noteworthy characteristics of the case profile and briefly describes the phonology of the language and what is known about phonological development in that language. A constraints-based nonlinear phonological analysis of the child’s data follows, culminating in an intervention plan and outcomes data where available. On-line supplemental files complement each issue, including full phonetic transcriptions, which readers may explore to investigate their own hypotheses.

This overview discusses first the theoretical foundations and basic methodology for the profiles, and then themes arising in the articles, including theoretical and clinical implications.

Theoretical foundations

Constraints-based nonlinear phonology

The papers in this issue are framed in a consistent fashion within a single approach (constraints-based nonlinear phonology); strengths of this approach are highlighted in the analysis of each child’s data. We do not mean to question the usefulness of other theories, but simply note that it is difficult to compare across studies that use different theoretical approaches with different foci and different types of explanation. We note also that previous empirical findings and theoretical ideas have led to many similarities across current theories, and thus, conclusions drawn from these case studies in the approach taken here may correspond to similar interpretations in other theoretical approaches.

Phonology and PPD

Following Bernhardt et al. (2010), the terms phonology and phonological are broadly defined, to encompass cognitive, linguistic, motor-planning, articulatory-implementation and perceptual factors, as well as their interactions. Phonological development depends not just on exposure to the ambient language(s), but maturation of a child’s anatomy and physiology related to speech perception and production (including of the brain) and development of cognitive processes necessary for phonological development, e.g. memory (storage and retrieval), attention, lexical access, and phonological and phonetic planning. If key structures or cognitive processes develop more slowly than is typical, a child may show limitations in output (speech production). The authors in this volume assume that all of the above factors are relevant to phonological development, assessment and intervention, with relative importance/impact of the various factors differing across time for a given child and across children.

Nonlinear phonology

Nonlinear phonology posits independent tiers (levels) for all aspects of the phonological system, from the highest level of the phrase to the lowest level of the phonological feature (Figures 1 and 2). A segmental representation, e.g. the camp as [ðəˈkʰæ̃mp] implies a single line of elements (segments). However, nonlinear phonology posits multiple independent lines of elements that occur in parallel across time, with some elements lasting for longer periods of time than others (e.g. Goldsmith, 1990, based on his observations of tonal features in African languages). In [ðəˈkʰæ̃mp], for example, the second syllable lasts a long time, while the feature complex [Dorsal,+back,+high] is limited to the first consonant of that syllable, the feature complex [Dorsal,-back,+low] is limited to the vowel, the feature [+nasal] lasts for the vowel and the immediately following consonant, and the feature [Labial] lasts for the final two consonants; the voicing tier shows more activity, starting with [+voiced] in [ð] and then [ə], changing to [-voiced] in [kʰ], back to [+voiced] in [æ̃] and then [m], and back to [-voiced] in [p]. Each feature has a separate line, but the feature lines are grouped together into sub-segmental complexes (Figure 1) and segments (Root Nodes). Segments are then grouped together into larger and larger groupings (Figure 2), from parts of the syllable up to phonological words (which are syntactic words plus unstressed clitics that are dependent on them, so that the camp is one phonological word). This basic framework is shared by all the papers in this special issue. (See, B. M. H. Bernhardt & Stemberger, 1998, for a detailed description of the framework.).

Figure 1. Structure within the segment.

Figure 2. Structure above the segment.

There is a fundamental distinction between the content (features) and the structure of a word (subgroupings of content elements with temporal structure). Content comprises features such as [Labial] and [+continuant]. Features that occur at roughly the same time are combined into segment-sized units, although features can vary in duration (e.g. the [-continuant] feature of [pʰ] ends long before [+spread-glottis] for the aspiration does, and the changing values of [continuant] ([-] to [+]) each last only briefly in affricates such as [ʧ]). Furthermore, some features may belong to more than one segment (e.g. the place feature [Labial] in the consonant sequence /mp/). Note that these conceptualisations (duration of features, their ability to extend beyond one segment) are among those that distinguish nonlinear phonology from ‘more linear, sequential’ accounts of features and segments. ‘Linear’ phonology assumes indivisibility of segments (with features being inherent properties of individual segments, the primes of the system); in nonlinear phonology, features are the primes and relationships between segments can be accounted for through interactions of features, which can extend beyond a single segment (Goldsmith, 1990) and even ‘hop’ over segments in a linear sequence (‘distant’ consonant assimilation across vowels, as in [ɡaɡ] for /daɡ/, where the [Dorsal] feature of /ɡ/ extends to the onset position of the syllable/word).

Features (and their combinations, or segments) are part of larger structures that group them in specific ways; for example, in the word abrupt /əˈbɹʌpt/, the /b/ is not grouped with the preceding vowel /ə/ but with the following stressed vowel /ʌ/, the /b/ and the /ɹ/ are grouped into a complex onset and the /pt/ appears in a complex coda in the same syllable. Segments in the nucleus (the vowel or syllabic consonant) and the coda are viewed as having ‘weight’, which is represented with a mora (μ); a short vowel has one mora, a long vowel or diphthong has two moras (and so last longer), and codas add another mora. Note that word forms such as /iːpa/ and /ipːa/ are identical segmentally (with three Root Nodes each), and differ in that the /i/ in /iːpa/ has two moras while the /p/ has no mora, while the /i/ in /ipːa/ has only one mora and the /p/ also has a mora (e.g. Hayes, 1989). Syllables are combined with other syllables into feet (encoding prominence differences such as stress), and feet are combined into phonological words; for example, hippopotamus /ˌhɪpəˈpɑɾəməs/ is a phonological word with the two feet /ˌhɪpə/ and /ˈpɑɾəməs/, each of which begins with a stressed syllable, but with primary stress on the second foot. Phonological words are combined into phrases. (See, B. M. H. Bernhardt & Stemberger, 1998).

The nonlinear perspective assumes that each element is independent. However, everything must be coordinated, which brings a measure of dependence at each point in time and across nearby points in time.

Interactivity versus modularity

The phonological system and the motor system can also be viewed as independent but interacting systems. B. M. H. Bernhardt and Stemberger (1998) noted that language has sometimes been conceptualised as a set of separate modules (e.g. semantics, syntax, lexicon, phonology, phonetics), which can interact with each other only in specific and limited ways. Fodor and Pylyshyn (1988) maintained that information feeds forward from an early module to a later module, but that earlier modules work completely independently of later modules. For example, syntax and the lexicon each influence each other, and both syntax and the lexicon influence phonology, but the phonology cannot feed back and influence the lexicon (choice of words to produce) or the syntax (word order or case). However, connectionist-interactionist theories (e.g. Dell, 1985) posit a bi-directional relationship: processing in later modules can impact on processing in earlier modules and alter the output in systematic ways (even though the primary flow of information is from an early module to a later module). B. M. H. Bernhardt and Stemberger (1998) argue that motor limitations lead to feedback that alters phonological processing (competition-based access of target phonological elements), leading to phonological accommodation for motor limitations: the phonetic and phonological systems can never have completely separate, independent impacts on speech output, just as the tiers within the phonological system may be autonomous but yet interact with other tiers. We elaborate input and output factors in the next sections.

Output constraints

No output is free: everything requires time and resources (for memory and processing), and these requirements constrain possible outputs. A speaker must learn to do everything: to produce a given feature at all, or in a particular location, with another feature at the same time, at several points in time, or in competition with other features at nearby points in time. Constraints impact on the production of particular content or structures; positive constraints (faithfulness = survival in the output) promote the successful production of target features or structures, and negative constraints (markedness) can prevent output. There must be a positive constraint (promoting the target) and a negative constraint (reflecting the level of resources needed for the output) for every element and every combination of elements, and a learner must learn the appropriate relationship (ranking) of those constraints for that language. Depending on whether the positive or negative constraint is dominant, at the feature level, a feature such as [Labial], or a feature combination such as [Labial,+continuant], may be produced successfully, replaced, moved or deleted. Constraints on word structure tiers can impact on complexity, facilitating a target onset cluster such as [pl] or reducing it to a single consonant such as [p] or [f]; facilitating or preventing unstressed syllables, everywhere or in particular locations in the word; allowing or preventing sequences of unstressed syllables or sequences of vowels; etc. Constraints can also facilitate or prevent the coordination of features across sequences of contiguous or separated segments; e.g. whether coronal [d] can occur before labial [u] or a labial [p] later in the word.

As noted, tiers can operate independently. Thus, as we will see in the papers in this issue, some children may have more difficulties with content than structure (or vice versa), or with structure at one tier more than another (e.g. with complex syllable onsets more than with foot structure of multisyllabic words, or vice versa). However, tiers also can interact, with developmental mismatches (‘errors’) reflecting interactions between content and structure, e.g.: (1) a particular feature combination (e.g. [Labial,+continuant] for /w/, /f/ and /v/) may be restricted to one word position (e.g. word-initial, WI); or (2) mismatches may vary by context, e.g. /s/ may be deleted word finally but appear as [t] word medially.

Competition between outputs is important. An input element such as [Dorsal] in /k/ must out-compete alternatives to be in the output, alternatives such as [Labial] ([p]) and [Coronal] ([t]) – or even no place features at all ([ʔ]). Many constraints interact, each favouring or disfavouring particular outputs, and a particular output, whether accurate or inaccurate, emerges from this mass activity. Two important points are that: (1) constraints and constraint rankings have a number of different sources; and (2) ‘inaccuracies’ in output (mismatch between the ‘adult/ambient’ target and child production) are not intentional (unless for humourous purposes) or ‘created’ by rules, but emerge from the joint action of many independent constraints. We elaborate these points below.

To the first point, Prince and Smolensky (2004) assumed that constraints and a learner’s initial constraint rankings are specific to phonology, innate and universal, while others presuppose that constraints emerge as a phonologization of phonetics (e.g. Hayes, 1999). Articulatory Phonology (e.g. Browman & Goldstein, 1989) presupposes that the motor system and the structure of the vocal tract automatically entail that certain outputs are more difficult than others (implying that humans without physical or neural anomalies are effectively identical for the purposes of phonetics). B. M. H. Bernhardt and Stemberger (1998) argue that constraints are all cognitive in nature (how to access and coordinate elements), but constraint rankings reflect cognitive and phonetic factors, with a random component. Humans without physical or neural anomalies show individual differences from the earliest point in development, including in brain organisation (which can pre-adapt the brain to find some elements difficult and others easy), hearing and perceptual mechanisms and the vocal tract. These individual differences lead to different initial relative rankings of constraints across TD children, to different solutions to the same challenges, and to different developmental paths. Learning to do the same outputs requires the systems of different people to converge over time, but they are unlikely to become identical.

Regarding the second point, the approach taken here assumes that a child’s inaccurate pronunciations arise in an emergent fashion from competition between elements, and do not explicitly exist anywhere in the system. This contrasts with Smith (1973), who posited explicit rules to replace a given input with a given output, with no guidance as to how or why that particular rule came into existence. It also contrasts with Vihman and Croft (2007) and Sosa and Bybee (2008), who suggest that the child’s mismatching output patterns are explicitly represented and stored as elements in the system, and are selected by the child as an overt change of output goals in place of the adult target. Articulatory Phonology (e.g. Namasivayam et al., 2020) assumes that a child’s mismatches may arise from small differences in amplitude or timing (not explicitly represented), but is vague about where this happens (e.g. early in planning versus during articulation in the vocal tract) and rarely addresses mismatches with non-target elements (such as [+constricted glottis] [ʔ] as a replacement of /kʰ/). Constraints-based non-linear phonology remains the most explicit, elaborated and flexible approach available for exploration of phonological development.

Input: context, perception, attention, and memory

Cross-linguistic differences arise because humans learn in ways that are sensitive to frequency of occurrence in the input. Insofar as different languages have different statistical properties, frequency will skew different languages in different directions. Both the order of mastery and the types of mismatches made by children learning different languages will reflect these differences in input.

Research has shown that by 11 months of age perception is finely tuned to the target language (e.g. Werker & Curtin, 2005), so that, given normal hearing, a child’s perception of adult forms in their own language(s) might be largely accurate (e.g. Swingley, 2009). However, ‘finely tuned’ does not imply perfect performance in all circumstances, and some researchers presuppose that inaccurate perception may underlie some individual variation or mismatch (error) patterns (e.g. Rvachew & Brosseau-Lapré, 2018; Sosa & Bybee, 2008). In the atypical developmental context of phonological intervention, intensive focused stimulation and speech perception training have been observed to effect positive change in perception and production (Rvachew & Brosseau-Lapré, 2018). Questions remain, however, as to what the observed changes reflect: changes in perception or phonological working memory, or focusing of attention. Furthermore, while perceptual training may show developmental effects in intervention with children with PPD, it is not clear whether and/or to what extent TD children might show an acceleration of phonological development given an increase in focused input. We return to this topic in the section on phonological intervention below, following an overview of the case profiles.

Basic methodology for the profiles³

Data collection, analysis and reporting

The data are from single-word elicitation probes developed for each language following a set of uniform principles (Bérubé et al., 2015). Most lists contain about 100 different words, eliciting all segments (consonants, vowels) at least twice across word positions in a variety of word lengths and shapes. Copies of most of these probes (tests) are available for free download at https://phonodevelopment.sites.olt.ubc.ca/ (Bernhardt & Stemberger, 2015, updated 2021). Information was also collected about the children’s family history, hearing and oral mechanism status, and general language and cognitive development, using standard tests and/or protocols for the relevant dialect/language.

Speech samples were audio-recorded (with or without video) by native speakers, usually in an uncompressed format with high quality digital equipment in quiet environments. Some samples were collected in less than optimal conditions in terms of ambient noise or recording equipment. Recordings were transcribed phonetically by at least one trained native speaker according to narrowness conventions developed for each language and often supplemented with acoustic analysis (Bernhardt & Stemberger, 2012). Transcription reliability was checked with a second (and sometimes third) transcriber, with final transcriptions arrived at through discussion and consensus, and further support from acoustic analysis.

The papers include a variety of common measures, presented in the same general order and format (including similar tables and supplemental files). For word structure, match proportions (accuracy) and mismatch (error) analyses are provided as relevant to the language for word length, stress, tone, timing units (presence of segments, i.e. consonant or vowel units) and word shapes (e.g. CVC, CCVC, VCCCVC, etc.). At the segmental level, matches and mismatches are described for specific consonants and vowels as singletons and in sequences by word position. Sequence analyses may concern contiguous segments (clusters, diphthongs) and/or more distant sequences (e.g. consonants separated by vowels), and generally address the features of the sequence, e.g. comparing the place feature sequence [Coronal]-[Labial] in the words tap, dip versus the [Labial]-[Coronal] sequences in pat, boat. In-depth feature analyses are provided where they illustrate particular patterns. A Whole Word Match measure evaluates whole word accuracy.

Most tables are straightforward, but we explain a more complex feature mismatch table (presented in some papers) for readers less familiar with nonlinear theories (Table 2, derived from table 4 in the German paper).

Table 2. Feature mismatches summed across word positions: Extract from a German example (B. M. Bernhardt & Ullrich, 2022).

	Manner						Laryngeal			Labial			Coronal					Dorsal
	+lat	+nas	+son	-cont	+cont	-/+cont	+vd	-vd	+sg	Lab	-ld	+ld	Cor	+ant	-ant	+gd	-gd
p^(h)					θ – 1								θ – 1, t^h-1
s						ʦ – 1											θ −1 $\mathrm{n}$tθ – 1
	Plus data from 14 additional phonemes
All	4	2	1 (6)	5	15 (9)	4	3	4	4	0	2	0	50	15	0	0	15	1
	Total Manner: 31(25)						Total laryngeal: 11			Total major place mismatches: 52 (51 produced as coronal); Total minor place mismatches: 32 (30 coronals)

Substitutions are entered in the cells for every mismatched feature for that substitution. Grey cells indicate target features for the consonant on that row. Underlining indicates possible assimilation. lat = lateral, nas = nasal, son = sonorant, cont = continuant, vd = voiced, sg = spread glottis Lab = Labial, ld = labiodental, Cor = Coronal, ant = anterior, gd =grooved. Columns labelled Lab, Cor and Dorsal designate major place changes, e.g. Coronal to Labial, without reference to more specific features within that type.

A feature mismatch analysis examines substitutions to determine (a) which features are affected, (both individually and in combination with other features), and (b) types and proportions of feature changes. In Table 2, the greyed out boxes show the expected features for consonants on each row. Mismatches (deletions/substitutions/additions) are entered into each relevant column where there are feature differences for the consonant on that row. For the German child, [θ] ([+continuant], [-voiced, +spread glottis], [Coronal, -grooved]) replaced both /p^h/ and /s/, but the mismatch entries differ for the two targets. For /p^h/, [θ] is a mismatch only for target features [-continuant] and [Labial], and so the substitution is entered into those two columns. For/s/, [θ] is amismatch only for [-grooved], and so the substitution is entered only into that column. Summing the total number of substitutions in each feature column allows acomparison across features and feature types (manner, place, laryngeal). Scanning within each row reveals the extent to which substitutions affect single versus multiple features for given targets.

Intervention planning

Each profile includes an intervention plan (actually implemented or just proposed) derived from the constraints-based nonlinear analysis. Treatment goals and strategies take all aspects of the phonological hierarchy into account: (a) individual word structures (word lengths, stress patterns, word shapes); (b) individual segments and features; and (c) interactions (word position or sequence constraints on segments/features; feature combinations: Bernhardt & Stemberger, 2000). The various phonological elements are designated as strengths (higher match levels: positive constraints) or needs (low accuracy or absence: negative constraints). Before selecting specific treatment goals and strategies, other factors about the client are reviewed: their skills, interests and context as per the International Classification of Functioning, Disability and Health (World Health Organization [WHO], 2021).

In order to promote system-wide change, the plan generally includes at least two treatment targets from different levels of the phonological hierarchy in the same time period, with the assumption that targets will be recycled as needed in subsequent time periods. The plan generally includes at least one more marked (more complex) target, but the phonological system or other factors about the client may also show a need for less marked, developmentally earlier targets. If a very strong negative constraint particularly impacts output overall, that constraint is generally addressed early in the intervention programme, whether or not the constraint affects developmentally early targets (Akan, English, Icelandic, Slovenian).

In developing therapy strategies, strengths (established forms) are often exploited when addressing needs (strong word structures for new segments/features, or vice versa). If match levels are low, developing or emergent/marginal, structures or segments may be designated as (relative) strengths because they are more advanced than absent elements. Even though they are accessible outputs, pervasive substitutions are not generally considered strengths, but may be useful comparative elements in contrast-based intervention.

The intervention framework is relatively neutral with respect to therapy techniques, but assumes that: (1) perception and awareness activities may enhance attention and self-monitoring skills, even if a child appears not to have misperceptions; (2) visual, auditory and tactile cueing may enhance acquisition of new elements (Bernhardt & Stemberger, 2015, updated 2021); and (3) most importantly, intervention (direct or indirect) needs to be relevant and motivating to the client. Phonological theories suggest some treatment strategies, for example: (a) for word structure development, drawing attention to timing units through rhythm-based activities, and/or manipulating onsets and rimes in alternation patterns (VC > CV; CV … CV > CVCV, etc.); and (b) for segments, building new segments through focus on component features already present in established segments.

Data from the Profiles

The data in the case profiles allow us to address several broad questions. Here we summarise data relating to word structure and structure-segment interactions as a basis for the subsequent discussions on theoretical and clinical implications. On-line Supplemental File 1 also provides overall global measures and a summary of segmental development organised by manner, laryngeal and place features.

Word length, stress and suprasegmentals

For the languages in this issue, target words ranged in length from one to four or more syllables. Word length (in syllables) presented challenges for some children (Arabic, Bulgarian, English, Portuguese, Spanish), often related to constraints on WI unstressed syllables and sequences of unstressed syllables. WI unstressed syllables presented challenges for half the children, the syllables either deleting (Arabic, Bulgarian, English, Spanish), showing more substitutions (French, German), or both (Portuguese). One child (English) deleted many syllables, even WI stressed syllables, so that monosyllabic outputs predominated. In certain cases, word length sometimes increased due to epenthesis to break up CCs or to support a coda consonant (Akan, Portuguese).

Suprasegmentals (stress, tone, pitch accent) were otherwise relatively well-preserved. Pitch accent in Japanese and tone in Mandarin were relatively accurate (with some differences for Mandarin complex tones), but much lower in Akan, where tone is mastered relatively late by TD children (Amoako et al., 2020). Furthermore, WI syllables in tone and pitch accent languages (Akan, Japanese, Mandarin) were not prone to deletion as seen above.

Word position

Simple WI onsets were generally present. However, the Arabic, English, Farsi, Mandarin, Polish, and Swedish children showed both full deletion and glottal stops for some WI consonants. The Spanish and Portuguese children deleted the WI onset only in unstressed syllables. Simple word-medial (WM) onsets rarely deleted (except for the English child). Onsets in WM stressed syllables sometimes behaved more like WI onsets than like WM onsets in unstressed syllables (e.g. Swedish). Word-final (WF) singleton codas (in languages that have them) were never entirely absent for any child, though many showed some deletion. One child (English) in contrast had a greater diversity of consonants word finally than elsewhere.

Syllable complexity

Complex onsets were often simplified, with loss of one consonant or both. Many children followed the ‘standard’ pattern: deletion of /s/ from s-stop clusters such as /sp/, deletion of the sonorant from C-sonorant clusters such as /pr/ (Bulgarian, Greek, Icelandic, Polish, Swedish, Spanish). A few languages with a richer set of WI consonant sequences (Arabic, Greek, Polish) did not follow sonority in quite the expected way, either because of different treatment of clusters with equivalent sonority (Greek, with accuracy in Nasal-Nasal clusters but deletion in Stop-Stop clusters) or because of similar reduction in clusters with diverse sonority relations (Arabic, Polish).

WF clusters are common only in seven of the adult languages, were fairly accurate in only two languages (German, Slovenian), and were extensively simplified (to one consonant or deleted entirely) in five languages (Arabic, English, Farsi, Icelandic, Swedish). For English, simplification was driven mostly by segmental issues involving low-frequency consonants. For the remaining children, the deleted consonant was usually a sonorant (nasal or liquid) or a continuant (fricative), showing a bias towards retention of oral stops. Farsi, French and Icelandic have final obstruent-sonorant clusters (e.g. /Vpr/, /Vpn̥/); for Farsi, such clusters were simplified in the same way as sonorant-obstruent clusters (loss of sonorant, which for French is a possible output even for adults), while for Icelandic these clusters were the only type of WF cluster usually produced accurately.

Compensatory lengthening was associated with the simplification of certain consonant sequences in two languages, most clearly for Icelandic, with a previously unreported developmental pattern of [VːCV] for target sequences with a short vowel followed by a pre-aspirated stop or a cluster beginning with voiceless sonorant or fricative. Akan showed a number of instances where syllabic /r̩/ was realised as doubling of a following vowel and/or non-adjacent nasal (e.g. /ń.sɹ́.ʊ̃́.má/ [ʰḿ.ʃʊ̃̀ʊ̃̀.ḿ.mʏ̃̀.áʰ]). That timing could be transferred to another segment suggests that timing is independent of segmental content (e.g. that there are timing units separate from other features, in accordance with assumptions about the nonlinear phonological hierarchy). Timing units in most languages were maintained only via substitution (e.g. /pr/ as [pl]).

Complexity constraints sometimes affected vowels. As is common (e.g. B. M. H. Bernhardt & Stemberger, 1998), monophthongs were relatively advanced, with only three children at 60% match or below (Akan, Arabic, English). Diphthongs showed reduction in Mandarin and Spanish (and to some extent in Farsi and Portuguese, where they are transcribed as coda consonants, e.g. /ej/).

Non-contextual substitution versus contextual influences

RAM patterns (reduplication, assimilation, migration – plus coalescence of two segments) arise due to interference between co-activated segments. RAM patterns sometimes made up more than half of a child’s substitutions (Bulgarian, Slovenian, Spanish), suggesting that segmental issues reflected interference from other segments (sequence constraints) more than constraints on features/segments per se. Some children’s substitutions were 15%-49% RAM (Akan, Japanese, Polish), suggesting basic difficulty with features plus some difficulty with interference.

Theoretical implications

While an exhaustive theoretical discussion is beyond the scope of this introductory paper, we highlight key implications below regarding areas of strength and difficulty, independence versus interdependence of phonological tiers, types of mismatches, variability, and motor-based challenges. Much of the data is compatible with many theoretical approaches, and we indicate potential points of similarity and difference.

Areas of strength and difficulty

As generally predicted by markedness, simple word structures and segments generally had higher match levels than complex ones (Jakobson, 1941/1968). Monosyllables and stress-initial disyllables with basic CV structure had greater accuracy overall than words with WI unstressed syllables, sequences of unstressed syllables, consonant clusters or diphthongs. Complex consonants with secondary articulations were also challenging. Children varied, however, in terms of whether any given point of complexity was problematic. Thus, complexity (markedness) appears to be relative, reflecting each individual’s own particular constraint rankings.

Segments also generally followed markedness predictions (Jakobson, 1941/1968). Many children had higher match with WI non-continuants (stops and nasals) than WI continuant sounds (fricatives and liquids, but also approximants). This has been characterised as making sense phonetically, given that stops and nasals can be produced accurately even if they are made with a forceful constriction, whereas fricatives require a constriction that is sufficient to cause turbulence in the airstream but weak enough not to halt airflow entirely (e.g. Browman & Goldstein, 1989). The Arabic child, however, showed the opposite effect (continuants better than stops), but this also makes sense if the child (who had Down Syndrome) has a physical system that tends to limit the forcefulness of gestures (hypotonicity), making it more difficult to stop airflow entirely, and pre-adapting the child to continuant sounds (but see below). Again, individuals can have their own complexity (markedness) constraints.

Independence and interdependence of levels

Across children, the data showed that segmental and structural information is independent to some degree. Some children showed high levels of accuracy for most consonants in ‘simple’ environments (singletons, in environments without problematic sequences of segments) but low accuracy for complex word structures (e.g. complex onsets, WI unstressed syllables): Bulgarian, Greek, Icelandic, Polish, Slovenian. Some children showed high accuracy for complex word structures, but low accuracy for consonants in simple environments: Farsi, French, German, Japanese, Mandarin, Akan. Some children showed low accuracy with both: Arabic, English, Portuguese, Spanish, Swedish. (No child showed high accuracy with both segmental and structural information; this is found only with TD children at these older ages.) That there can be a (statistical) double dissociation between segmental and structural information, plus interactions between them, demonstrates their independence, and the importance of both in development and in the identification of PPD. Of course, difficulty with complex structures has consequences for segmental accuracy: when a consonant is deleted from a cluster (the dominant mismatch for all children), the deleted consonant is inaccurate, but for structural rather than segmental reasons.

Mismatches

Mismatches in each child’s system bore characteristics of the adult language. Segmental substitutions usually fell within the phonetic inventory of the language, suggesting overgeneralisation of elements learned as targets. However, sometimes non-target substitutions were used extensively: e.g. [θ] in German and Slovenian, alveolopalatals in many languages, and affricates in Portuguese. Phonotactically, mismatches included non-adult clusters such as [tl] (Icelandic), [tj] (Akan), and [dt] (Farsi). In Akan, syllabic nasals sometimes occurred with onsets, which is not permitted in adult Akan. Constraints-based phonology allows the child’s outputs to fall outside the adult system; error-driven learning leads the system into accidental temporary states on the way to mastering the adult system.

Sometimes the child matched one aspect of the adult target but with a mismatch that violated the constraints of the languages or the majority dialectal variant. The Icelandic child followed the adult pattern of complementarity of length between stressed vowels and a following consonant or cluster by lengthening vowels when WM or WF consonant sequences were not possible, something never before reported in Icelandic phonological development (Másdóttir, 2008). For the Akan child the feature [+round] moved from consonants and (deleted) vowels onto other vowels, creating front rounded vowels; this pattern was not observed for any other language, and may have reflected an overgeneralisation of variation in adult Akan, where the labial component of labiopalatals can shift onto a following vowel (e.g. [dᶣi]~[dy]).

Variability

There was a large amount of variability across the children in this issue, with some children showing primarily segmental/featural difficulties, and others showing primarily problems with combining segments (and their features) into larger structures. Children differed on whether the difficulties involving larger structures had a structural basis (e.g. constraints on complex CC onsets) versus interference between different features in segments (RAM patterns). This is exactly the sort of diversity that a constraints-based nonlinear approach predicts. It is unclear whether alternative approaches such as Articulatory Phonology or Usage-Based Grammar (e.g. Browman & Goldstein, 1989; Bybee, 2006; Namasivayam et al., 2020; Vihman & Croft, 2007) predict such diversity – mostly because such approaches are under-developed about how inter-individual variation arises.

Phonological ‘versus’ motor-based difficulties

Distinctions are often made concerning whether a child has phonological versus motor-based difficulties in speech-language pathology (e.g. Rvachew & Brosseau-Lapré, 2018). The Arabic child with Down Syndrome had some motor-based weakness, possibly explaining why this child had more difficulties with stops and nasals than with fricatives, liquids, and approximants. However, that does not exclude a phonological component to her atypical system. The child produced some stops as fricatives (e.g. /d/ as [ð] or [j]), which is what would result from incomplete gestures, but also had a high rate of glottal substitution ([ʔ]), involving a complete loss of the oral gesture with addition of glottal closure (not resulting from weak articulation of gestures, but rather from a change in target at the level of phonological planning). The phonological system accommodated to the motor-based restrictions, providing work-around solutions to certain possibly motor-based constraints.

Similarly, to account for the fronting of velars to coronal articulations, some have assumed that this is not a phonological pattern (see, e.g. McAllister Byun, 2012). Limitations in the control of the tongue result in an undifferentiated tongue shape that is more similar to [t] than to [k]: distinct from both, but perceived and transcribed as [t] for both. The Farsi child, however, showed velar fronting only in WM and WF positions. [ʔ] appeared in WI position in place of dorsal target consonants, showing a phonological resolution to the difficulty of WI target /k, ɡ/. (Since the child did not tolerate the undifferentiated tongue shape for WI target /k, g/, it is not obvious why it was tolerated for WI target /t, d/). While there may possibly be motor implementation difficulties, these lead to changes in the phonological system, which is evidence against modularity, and against the easy dichotomous compartmentalisation of difficulties into phonological versus phonetic/motor.

Other theories

We have noted points where Articulatory Phonology and Usage-Based Grammar do not necessarily predict the findings of the articles in this special issue, but it should be noted that all extant theories have more similarities than differences, drawing on research on phonetics and phonology from the past 50 years, as well as language processing and general cognition. B. M. H. Bernhardt and Stemberger (1998) observe that a nonlinear constraints-based phonology approach provides an elaborated set of tools for analysing child data, while other approaches tend to be more focused and to date have provided no or minimal guidance on many topics, especially about how particular mismatch patterns arise. We have noted that constraints-based systems are about competition and not just about whether a child can articulate elements. The Swedish child produced WM [j] for all voiced consonants (including /m/), losing target place and manner features that were actually accurate for voiceless consonants; [-voiced] prohibited a default [j] substitution, thus allowing output of e.g. fricatives and labials, a wonderful example of the effects of competition. It is simply unclear how Articulatory Phonology accounts for patterns of this sort. While usage-based approaches can technically describe such patterns, it is unclear how and why they arise. Nonlinear constraints-based phonology remains the most elaborated and flexible approach available, but different theories sometimes suggest different analyses and alternate ways of thinking about certain aspects of the data.

Clinical implications

Finally, what do the data imply about PPD and clinical practice? Speech development can proceed in many ways, because of the characteristics of the target linguistic system, an individual child’s aptitudes for speech development and basic happenstance. In the words of Robert Frost in the poem ‘The Road Not Taken’:

”Two roads diverged in a wood, and I–

I took the one less travelled by,

And that has made all the difference.”

When a child is identified as having PPD, the interventionist needs to determine a particular child’s needs and strengths (not just in phonology), and capitalise on those strengths to stimulate accelerated development. Doing comprehensive analyses takes time, but time spent in analysis can reduce time spent in treatment. Constraints-based nonlinear analyses investigated entire phonological systems, culminating in identification of needs across the phonological hierarchy, but also strengths that could be exploited in designing treatment strategies; where there were anomalous and sometimes puzzling patterns, detailed analyses (e.g. of sequence constraints) provided clues for intervention strategies that could potentially lead a child to progress along a more typical developmental path. After a relatively short period of intervention addressing anomalous patterns, the English, Icelandic and Slovenian children showed more typical phonological patterns, and afterwards progressed as other children do (though taking time to fully catch up to age-level). We cannot know whether the plans for the children who did not receive therapy as part of this project would have had similar positive outcomes, but the analyses led to a plan with potential to do that. For most of these cases, this was the first full published case analysis with intervention plan for that language.

Conclusion

These articles take a broad view of phonology, where all contributing factors to speech production are implicated (Bernhardt et al., 2010). They illustrate the degree of variability between different children, and each child’s output reflects the characteristics of their own language, but with commonalities across languages. The data are rich and available for further analysis (see the Supplemental files online). We encourage readers to ‘try this at home’, integrating what they already know into this holistic view of phonology.

Acknowledgments

The authors thank the various contributors for their commitment to crosslinguistic study in phonological development and particularly the children and their families for their participation.

Disclosure statement

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

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Notes

¹ The terms Speech Sound Disorders, Speech Sound Difficulties and Developmental Phonological Disorders are often used in the field of speech-language pathology but we utilise the term ‘protracted phonological development’ with the perspective that phonological development can extend over a long period of time but that hope remains for positive change.

² Whole Word Match: the proportion of words in the sample that match the adult targets exactly, ignoring slight deviations in tongue placement (e.g. [s] versus [s]) or full voicing (e.g. [b] versus [b]).

³ All data were collected according to ethical procedures in each country with the informed written consent of parents or guardians, including permission to disseminate anonymised data.