Markedness and implicational relationships in phonological development: A cross-linguistic investigation

Abstract

Purpose:

The complexity approach to speech disorders, based on the theoretical notion of phonological markedness, has been gaining interest over the last decade. In a nutshell, this approach suggests that the acquisition of phonologically marked units (e.g. complex onsets) implies the acquisition of less marked ones (e.g. singleton onsets). However, because the notion of markedness is, itself, subject to controversies, we need to constrain what types of implications can be generalized among language learners, within and across languages.

Method:

We report on longitudinal data from one phonologically-disordered and five typically-developing children documented across four different languages (English, French, German, Portuguese), using data from the PhonBank database (https://phonbank.talkbank.org). Using the Phon software program (https://www.phon.ca), we systematically analysed each longitudinal study for consonants in singleton onsets and codas as well as in onset clusters.

Result:

The implicational relationships supported by our study involve units of similar types (e.g. relations between different segmental categories), while relationships that involve different types of units or processes cannot be generalized across learners.

Conclusion:

A better understanding of implicational relationships makes the complexity approach more predictive of developmental patterns of phonology and related phonological disorders.

Introduction

The complexity approach to speech disorders, based on the theoretical notion of phonological markedness, has been gaining interest over the last decade (Baker & Williams, 2010; Peña-Brooks & Hedge, 2015). From the perspective of phonological development, this approach holds that the acquisition of phonologically complex, or marked, sounds or structures (e.g. complex onsets) implies the acquisition of less marked sounds or structures (e.g. singleton onsets). Likewise, in the area of clinical phonology, this approach suggests that a focus on more complex linguistic stimuli helps promote generalization to untreated but related targets which are considered less complex, or “unmarked.” For example, on the “Speech Sound Disorders” web page of the American Speech-Language-Hearing Association (ASHA) website, approaches based on these types of implications are listed under both Target Selection and Treatment Options:¹

Complexity—focuses on more complex, linguistically marked phonological elements not in the child’s phonological system to encourage cascading, generalized learning of sounds

(Gierut, 2007; Storkel, 2018).

Central to the complexity approach is the notion of markedness which, as in the quote above, tends to be used interchangeably with different notions of phonological complexity. While markedness is central to numerous studies of phonology based on typological universals, in the tradition of Jakobson (1941) and Chomsky & Halle (1968, chap. 9), this notion also suffers from degrees of vagueness, especially concerning how it should be generalized across languages. As Culicover (2013) remarks, “we do not yet have an independent measure of complexity [and thus] have to make inferences about what is more or less complex based on our understanding of language processing, language acquisition, language change, and variation” (p. 11). While we maintain that markedness is both a valid notion for theoretical inquiry and a useful metric to predict developmental or clinical patterns of speech production, we emphasize in this paper that markedness and the implicational relationships it entails for phonological development can only be established between units which can be directly compared in terms of structural complexity (e.g. onset clusters vs. singletons) or, for individual phones, phonological features, or the acoustic or articulatory dimensions of speech which are involved in their production.

The discussion below revolves around the theoretical notion of markedness and, in particular, markedness-based implicational relationships. We begin by situating these notions within the current literature.

Markedness

Within phonological theory, the notion of markedness is generally invoked to capture the relative complexity of different phonological units or processes. Markedness can, in this respect, refer to widely different things, including frequency of occurrence, from either typological or language-specific perspectives (infrequent units are more marked than frequent ones), structural complexity within phonological representations (branching structures are more marked than non-branching ones) or, at the segmental level, in terms of the number of phonological features or particular combinations of phonological features involved in the representation of a phone (Fikkert & Levelt, 2008; Levelt, 1994; Rice & Avery, 1995). Within more phonetically grounded approaches to phonology, markedness can refer to varying degrees of auditory perceptibility (e.g. Steriade, 2001), articulatory complexity (e.g. Gayraud et al., 2018; Menn, 1983), or both (e.g. McAllister Byun et al., 2016).

In the universalist tradition of phonological inquiry set by Jakobson (1941), markedness refers to principles or laws that govern the functioning of all languages; typological evidence is thus taken as the primary source to establish cross-linguistic markedness relations, often referred to as linguistic universals. More recently, however, scholars have advocated for more language-specific approaches to markedness, in particular by including phonetic aspects of speech perception and production, the parameters of which can only be described on a language-specific basis. For example, while English and French are phonologically similar in that they both display a phonological rhotic in their respective inventory of consonants, the rhotic of most dialects of English is a lingual approximant, either retroflexed or bunched, also with some degree of labialization (e.g. Keyser & Stevens, 2006), while the most common rhotic of French is a uvular fricative or trill (Fougeron & Smith, 1993). Phonetically, and skipping over several other details, for example about allophonic distributions, the rhotics of these languages thus differ tremendously from one another.

Likewise, because additional factors beyond unit types, such as usage frequency, can also serve as metrics for markedness, the ultimate computation of markedness relations is often possible only on language-specific and, at times, speaker-specific grounds (e.g. Edwards & Beckman, 2008; Edwards et al., 2015; Ingram, 1999). Given this, claims about universal markedness can only be interpreted in terms of general trends or expectations, not in absolute terms. Hume (2011) summarizes this as the markedness-through-mechanism approach, which “attributes markedness patterns to a confluence of factors that interact with grammatical systems, and relate to physical, cognitive, and social mechanisms shared by all humans” (Hume, 2011, p. 81). We list, in Table I, some of the most general descriptors of markedness under this approach.

Table I:

Markedness descriptors (adapted from Hume 2011: 79–81)

Unmarked	Marked
Simple	Complex
More frequent	Less frequent
Articulatorily simple	Articulatorily difficult
Perceptually salient	Perceptually opaque
Acquired earlier	Acquired later

In the argument below, we fully embrace the tenets of markedness-through-mechanism, but discuss ways in which we can, or should, impose principled limits on its application. In a nutshell, we propose that formal markedness-based implicational relationships should not combine different types of descriptors such as those in Table I. For example, issues in syllable structure complexity should not be implicationally related to issues in segmental development. That is not to say that segmental and syllable structure development do not interact; they often do, for example in the context of positionally-conditioned phonological substitutions (e.g. Inkelas & Rose, 2007). Our argument is that these types of interactions emerge from a confluence of otherwise independent factors and, as such, should not be construed in terms of formal implications.

Implicational relationships in phonological development and speech disorders

Taking a universalist stance on implicational relationships, Gierut (2007) proposed a series of 22 implicational relationships, which she claims can be generalized to all language learners on universal markedness grounds:

“Drawing from the literature, phonological complexity is defined from epistemic, ontological, and functional perspectives, with specific emphasis on the application of language universals in the selection of target sounds for treatment.”

(Gierut, 2007, p.6)

The general logic of Gierut’s proposal is that if a child has acquired a marked sound, this child will logically have acquired all sounds considered to be typologically less marked.

As we will see, many of the relationships proposed by Gierut nicely fall within the general markedness considerations discussed above, for example that singleton syllable onsets are less marked, and are thus predicted to be acquired earlier, than complex onsets. However, Gierut also proposes less intuitive relationships, for example, that “spirantization implies place assimilation,” or that “stopping implies liquid gliding.” While such apparent relations have been documented in case studies of phonologically delayed learners of English (e.g. Dinnsen & O’Connor, 2002), it is difficult to pinpoint which formal or functional aspects of the child’s system could possibly yield these relationships. For example, how should spirantization, which consists of production of fricatives for stops, a manner substitution, enter in an implicational relationship with a process of assimilation affecting the place of articulation of (non-adjacent) consonants? To our knowledge, no theory of phonology or phonetics, from either formal or functional perspectives, can readily predict this relationship, even for individual languages.

Questions such as these initially motivated the research we summarize below. In a nutshell, we investigated the general applicability of 15 of the 22 implicational relationships proposed by Gierut (2007), through a cross-linguistic study of the development of consonants and consonant clusters across four different languages. Because of the developmental predictions made by these relationships, we opted to study individual learners longitudinally, which imposed practical limits on the range of our observations. In spite of this, clear trends emerged from the data, each of which follow the general logic of the complexity approach to phonological development. However, of the 15 implicational relationships we consider below, we could only validate the ones that involve directly comparable units or phonetic dimensions of speech, while the more apparently ad hoc relationships were either contradicted by the data or could not be verified based on the cross-linguistic corpus data we used in this research.

The 22 universal relationships proposed by Gierut are listed in (1). For the sake of clarity, we have grouped the relationships into three categories, anticipating on the outcomes of our study presented below. We begin, in (1a), with the relationships which have clear factual and related theoretical underpinnings and, thus, are likely to hold universally, all else being equal. Given markedness-through-mechanism, it is indeed possible that language- or learner-specific pressures, for example from usage frequency, counter some of these general effects, as Beckman et al. (2003) and Monnin et al. (2015) suggest from cross-linguistic evidence, for example concerning the velar-coronal implication in (1av). In (1b), we list relationships which are more debatable, as they involve theoretically unwarranted implications between units or structures. Finally, in (1c), we list the relationships that we could not verify based on the current data. We leave the investigation of these relationships for further research.

1. Implicational relationships (Gierut 2007)

   1. Emergent from universal mechanisms affecting phonological development

      1. Consonants imply vowels
      2. Affricates imply fricatives
      3. Fricatives imply stops
      4. Liquids imply nasals
      5. Velars imply coronals
      6. Stops in final position imply stops in initial position
      7. Clusters imply singletons
   2. Entail theoretically unwarranted implications

      1. Clusters imply affricates
      2. Liquid onset clusters imply a liquid in coda
      3. Word-initial /r/ implies post-vocalic /r/
      4. Fricatives in initial position imply fricatives in final position
      5. Voiced obstruents imply voiceless obstruents
      6. Fricative+liquid clusters imply stop+liquid clusters
      7. Clusters with a small sonority difference imply clusters with a greater difference
      8. A stridency and/or laterality distinction implies the phonetic occurrence of a liquid, which implies a fricative and/or affricate, which implies a voice distinction among cognate stops, which implies a nasal and glide
   3. Unverifiable from the current data or methods

      1. Stopping implies liquid gliding
      2. Manner assimilation implies liquid gliding
      3. Spirantization implies place assimilation
      4. Progressive place assimilation implies regressive place assimilation
      5. The absence of a voice contrast in final position implies the absence of a voice contrast in initial position
      6. Velar fronting word-finally implies velar fronting word-initially
      7. Errors of weak syllable deletion in syllables beginning with an obstruent imply like errors in syllables beginning with a sonorant

After we establish the developmental facts from our cross-linguistic study, we discuss some of the potential underpinnings, or lack thereof, of each relationship in (1a-b). As we will see, the most central generalisation emerging from our data is that only the relationships that involve formally similar units are the most clearly supported by our cross-linguistic investigation.

Cross-linguistic survey of implicational relationships

In order to study the unfolding of production patterns over time, and to track how phonological patterns may interact within individual phonological systems, it is necessary to observe these patterns within their developmental time spans (Rose & Inkelas, 2011). Working toward this goal, we used data from longitudinal studies. For each of the six case studies we considered for analysis, we derived the developmental trajectories of all relevant phones within singleton onsets, singletons codas, and complex onsets. We then verified whether the implicational relationships stated in (1a-b) could capture the developmental trajectories obtained for each child.

Corpus and language selection

We selected datasets from English, French, German, and Portuguese based on the typological similarities and differences that these languages offer, a summary of the phonological characteristics of each language is in AppendixA. For example, all of these four languages have lateral and rhotic consonants in syllable-initial, -medial, and -final positions. However, these liquids also display phonetic similarities and differences across the four languages which allow for a more detailed analysis of how phonetic detail may influence acquisition. For example, in English, the clear [l] of leaf is phonetically different from that of the dark (or velarized) [ɫ] of feel. European Portuguese patterns in a similar way, while the laterals in German and French do not; these two languages display clear [l] across all syllable positions. We introduce these datasets in the next section.

Datasets

All of the datasets we used for analysis were obtained through the PhonBank database project (https://phonbank.talkbank.org; Rose & MacWhinney, 2014).We summarize these datasets in Table II.

Table II:

Participants

Language	Name	Age Range	# of sessions	Sex	Type of Study
English	William	1;04.12 – 3;04.18	44	M	Naturalistic
French	Adrien	1;11.14 – 4;03.27	31	M	Naturalistic
German	Eleonora	1;00.07 – 1;10.25	30	F	Naturalistic
German	Wiglaf	1;03.21 – 2;01.21	24	M	Naturalistic
Portuguese	Inês	0;11.13 – 4;02.17	30	F	Naturalistic
English	Ben	3;09.27 – 4;03.05	17	M	Clinical

The typical English dataset comes from the Providence corpus (e.g. Demuth et al., 2006). We used spontaneous speech data from child William, recorded 44 times over a 2-year period, between 1;04.12 and 3;04.18. The French data are from child Adrien, a learner of Parisian French recorded across 31 sessions between 1;11.14 and 4;03.27 (Yamaguchi, 2012). The German data were obtained from two children documented as part of the Grimm (2007) study: Eleonora (30 sessions between 1;00.07 and 1;10.25) and Wiglaf (24 sessions between 1;03.21 and 2;01.21). Note that we considered the data from these two learners of German in order to better balance our empirical coverage, given that these datasets cover relatively smaller developmental periods than the other datasets from typically developing children we used for our study. The Portuguese data are from child Inês, who was recorded 30 times between the ages 0;11.13 and 4;02.17, as part of the larger Portuguese-CCF corpus (e.g. Correia, 2009; Costa, 2010). Finally, the clinical English data come from McAllister Byun’s (2009) case study of Ben, a child who has a speech sound disorder with features of Childhood Apraxia of Speech (CAS).

Data preparation

Each of these corpora was fully transcribed by their original authors and/or co-members of their research teams, all native speakers of the children’s developing languages. While we are aware of the limitations posed by phonetic transcription, especially in the absence of corroborating acoustic verifications, we believe that the level of granularity offered by these transcription data is sufficient to address the current research questions, especially given that the background research and related clinical considerations are also based on phonetically transcribed data. We analysed these corpus data using the computer program Phon (e.g. Rose & MacWhinney, 2014).

Among other functions, Phon automatically labels phonetic transcriptions for syllable positions within the word, crucial to identify consonants in different syllabic positions. Figure 1 shows the colour-coded syllabification of the child utterance ‘elephant big’.

Figure 1:

Syllabification and alignment in Phon

This figure also illustrates phone-by-phone alignments between the target (model) form attempted by the child and the actual form produced. These syllable-level annotations and phone alignments are at the centre of our analyses, as they enable the systematic compilation of segmental patterning across different positions within the syllable and word.

Data compilation

In our assessment of segmental development, we limited our analyses to singleton onsets and codas in word initial and final positions, respectively, in order to avoid interactions with patterns of cluster production. We also ignored results from truncated syllables. For example, if banana was produced as nana, [b] deletion in this case was not reported, as patterns of weak syllable deletion may very well occur independently of the child’s overall segmental productive abilities (Macken & Ferguson, 1983). In order to assess onset cluster development, we compiled the data in all word positions (e.g. word-initial, -medial, -final), to maximise the number of examples. We also made sure to verify whether syllable position within the word or relative to stress had a systematic influence on individual patterns, which was not the case for any of the results reported below.

We classified a phone or cluster type as acquired when the child achieved a majority of accurate productions within a given session transcript, and the proportion of correct productions remained stable or increased in consecutive sessions (e.g. Ingram, 1981, for an early discussion). In order to perform the relevant verifications about the validity of each implicational relationship in (1a-b), we built our observations, for each phone and phonological context, into developmental timelines, as in Table III, which summarizes the development of Wiglaf’s singleton onsets, final codas, and onset clusters across the documented time period.

Table III:

Summary of Wiglaf’s phonological development

1;03.21	1;05.03	1;05.26	1;06.12	1;07.11	1;08.02	1;08.13	1;09.02	1;09.09	1;09.19	1;10.28	1;11.03	1;11.13	2;00.17	2;01.07	Not acquired
Onset
[p]	[n]	[j] [m] [h]	[b]	[t]		[v]				[l]	[f]	[k] [g] [ʀ]	[d]	[ts]	[z] [ʃ] [tʃ] $\left[\widehat{\text{pf}}\right]$
Coda
	[p] [f]		[s]		[t]		[k]	[x]	[l]	[ç]		[ŋ]			[ʃ] [ʀ] $\left[\widehat{\text{pf}}\right]$
Onset cluster
										Fricative-rhotic	Stop-lateral	Fricative-lateral			Stop-glide Stop-rhotic Nasal-glide

Building on these summaries, we then established whether the orders of acquisition suggested by the implicational relationships in (1a-b) were supported, based on each child’s developmental timeline. We summarize our results in the next section; we invite the reader interested in the detail of our classifications to consult Watts (2018).

Results of cross-linguistic survey

Table IV provides the summary of our results for the 15 relationships in (1a-b). As we can see in this table, only the relationships in (1a) above could be verified (marked as “✓”) through at least some of the children’s data, without being contradicted (as “X” indicates) by any developmental patterns. The table displays inconclusive results for a subset of our data, due to either the impossibility to assess the relationships for some datasets (marked as “?”), or to the irrelevance of some of the relationships for given languages (marked as “N/A”).

Table IV:

Summary of results (for the 15 relationships in (1a-b))

		Implicational relationships	Ben (English)	William (English)	Adrien (French)	Eleonora (German)	Wiglaf (German)	Inês (Portuguese)
a.	i.	Consonants imply vowels	✓	✓	✓	✓	✓	✓
	ii.	Affricates imply fricatives	✓	✓	N/A	✓	✓	N/A
	iii.	Fricatives imply stops	✓	?	✓	X	✓	✓
	iv.	Liquids imply nasals	✓	✓	✓	✓	✓	✓
	v.	Velars imply coronals	✓	?	✓	✓	✓	✓
	vi.	Stops in final position imply stops in initial position	?	✓	✓	✓	✓	N/A
	vii.	Clusters imply singletons	✓	✓	✓	✓	✓	✓
b.	i.	Clusters imply affricates	?	✓	N/A	?	X	N/A
	ii.	Liquid onset clusters imply a liquid in coda	?	X	✓	?	✓	X
	iii.	Word-initial /r/ implies post-vocalic /r/	?	X	✓	X	X	X
	iv.	Fricatives in initial position imply fricatives in final position	✓	X	X	X	✓	✓
	v.	Voiced obstruents imply voiceless obstruents	X	?	✓	✓	✓	X
	vi.	Fricative+liquid clusters imply stop+liquid clusters	?	✓	?	?	X	✓
	vii.	Clusters with a small sonority difference imply clusters with a greater difference	?	✓	X	?	X	X
	viii.	A stridency and/or laterality distinction implies the phonetic occurrence of a liquid, which implies a fricative and/or affricate, which implies a voice distinction among cognate stops, which implies a nasal and glide	N/A	N/A	N/A	N/A	N/A	N/A

^*Legend: ✓ = validated; X = contradicted; N/A = not applicable; ? = inconclusive

In contrast to this, the relationships in (1b) could not be verified as universally valid, as these relationships were contradicted by at least one of the case studies. Finally, we ruled out the relationship in (1bviii) on grounds that it, as stated, is not applicable in practice.

Developmental evidence in support of implicational relationships

In this section, we summarize the developmental evidence in support of the classifications reported in Table IVa, for each of the relationships listed in (1a).

“Consonants imply vowels”

To the best of our knowledge, there has never been an attested case of a child that produces consonant sounds but does not have any vowel sounds in their inventory. There are however documented cases of children unable to produce consonants (e.g. Rialland et al., 2011). This can be predicted on given that vowel production does not require the generally more complex constrictions involved in consonant production.

“Affricates imply fricatives”

This implicational relationship is validated, in our study, by data from all the learners of languages which display affricates in their inventories. English-learning William acquired [tʃ] at 1;08.02, after he had acquired the fricative [s], at 1;04.12. German-learning Wiglaf acquired [ts] at 2;01.07 and the fricative [h] much earlier, at 1;05.26. While German-learner Eleonora had not yet acquired affricates by her latest recorded session, she acquired fricatives at an earlier age ([h] by 1;00.07). Atypical English-learner Ben follows a similar pattern as he acquired fricative [ʃ] at 3;11.18 but had not acquired affricates by his last recorded session.

“Fricatives imply stops”

Adrien, Wiglaf, Inês, and Ben’s acquisition data support this implicational relationship. Adrien acquired [v] at age 2;02.20, after he acquired stop [t], at 2;00.16. Wiglaf acquired [h] at 1;05.26, after [p], at 1;03.21. Inês had stop [d] by 1;00.25 but only mastered her first fricative ([f]) at 2;07.16. Ben acquired [ʃ] at 3;11.18, after he had mastered the stops [b] and [d], by 3;09.06. Finally, William (English) and Eleonora (German) had acquired both categories of fricatives and stops by their earliest recorded sessions; while their data do not offer ground for validation, they also do not contradict the relationship.

“Liquids imply nasals”

This implicational relationship is supported by the data from all six children. William had already acquired the nasal [m] by his earliest recorded session, 1;04.12, and acquired his first liquid, [l], slightly later, at 1;04.25. Similarly, Adrien had acquired the nasal [m] by his first recorded session, at 1;11.14, and acquired his first liquid, also [l], only at 2;04.16. Eleonora acquired [n] at 1;01.11 and her first liquid, [l], later, at 1;06.29. Wiglaf acquired [n] at 1;05.03 and the liquid [l] later, at 1;10.28. Inês had also acquired the nasal [m] by the beginning of her data collection at 0;11.14, and acquired [l] at 2;01.10. Ben had not acquired any liquid sounds by the end of his corpus data; however, he had acquired nasals [n] and [m] by his first recorded session, 3;09.06.

“Velars imply coronals”

French-learner Adrien acquired velar [k] at 2;05.23 and coronal [t] earlier, at 2;00.16. Eleonora acquired [k] at 1;06.05, after she acquired [t], at 1;01.11. Wiglaf acquired [k] at 1;11.13 and [t] at 1;07.11. Inês acquired [k] at 1;03.06, and had already [d] by 1;00.25. Ben did not acquire any velar sounds in onset position by the end of his corpus, but he had acquired coronal [d] in onset position by his first session, at 3;09.06. Ben acquired the velar sound [k] in coda at 3;10.25, however both his onset and coda acquisition of velars support this relationship as he acquired a coronal first in both positions. Finally, because William had acquired both the velar [k] and the coronal [d] by his earliest recorded session, at 1;04.12, his dataset neither validates nor contradicts this implicational relationship.

“Stops in final position imply stops in initial position”

This relationship is not applicable to Portuguese, a language where the only coda consonants allowed are continuant. This relationship is however supported by data from William, Adrien, Eleonora, and Wiglaf. William acquired the stop [k] in coda at 1;06.05, after he had acquired different stops in initial position by his first recorded session, at 1;04.12. Adrien acquired [t] in coda at 2;02.20, after [t] in onset, at 2;00.16. Eleonora acquired [p] in coda at 1;04.02, and had already acquired [b] in onset, at 1;01.11. Wiglaf acquired [p] in onset at 1;03.21 and [p] in coda at 1;05.03. Finally, we cannot evaluate this relationship using Ben’s data because he had acquired stops in both initial and final position by his first documented session.

“Clusters imply singletons”

All children were able to produce singleton onsets, but not complex ones, by their first recording sessions. They thus developed onset clusters later in development, in line with this implicational relationship, also in line with virtually all studies on the development of syllable structure, across different languages (e.g. Fikkert, 1994; Gierut & O’Connor, 2002; Freitas, 1997; Rose, 2000).

This completes our summary of the evidence in support for the implicational relationships in (1a). In the next section, we present further developmental evidence, which this time poses challenges to the relationships listed in (1b).

Developmental evidence challenging some of the implicational relationships

In this section, we highlight evidence which undermines the relationships listed in (1b), as summarized in our classification in Table IVb.

“Clusters imply affricates”

This relationship, irrelevant to French and Portuguese learners, could not be verified against the datasets for Eleonora and Ben, neither of whom acquired clusters or affricates during their documented periods. Of the two children remaining, William’s data appear to support this relationship, while Wiglaf’s contradict it. The earliest cluster type that William acquired is the stop-glide cluster, at age 1;09.25, which he acquired after the affricate [tʃ], at 1;08.02. In contrast to this, Wiglaf acquired fricative-rhotic clusters at 1;10.28, but only acquired his first affricate, [ts], a few months later, at 2;01.07. Given the cross-linguistic differences suggested by these observations, more research is needed, in order to uncover potential language-specific conditioning, as predicted under the markedness-through-mechanism approach. This question is particularly relevant given that the original evidence in support for this relationship, similar to William’s, also comes from a learner of English.

“Liquid onset clusters imply a liquid in coda”

This relationship, which relies on the term ‘liquid’ as a broadly defined lateral or rhotic consonant, is supported by at least some developmental patterns displayed by Adrien and Wiglaf. This relationship is however contradicted by William’s and Inês’s, while the data from Eleonora and Ben do not allow for an evaluation of this relationship. Adrien acquired both stop-lateral and stop-rhotic clusters at 4;01.13, long after he had acquired [l] in coda, at 2;02.20. Wiglaf acquired fricative-rhotic clusters at 1;10.28, stop-lateral clusters at 1;11.03, fricative-lateral clusters at 1;11.13, all a few weeks after he had acquired the liquid [l] in coda, at 1;09.19. In contrast to these two children, William acquired [l] in coda at 2;00.24, later than stop-lateral clusters, at 1;10.12, and fricative-lateral and stop-rhotic clusters, both at 2;00.12. Likewise, Inês acquired fricative-lateral and stop-lateral clusters at 2;08.23, stop-rhotic clusters at 3;10.01, and fricative-rhotic clusters at 3;11.12, in line with her development of liquids in singleton onsets. However, she had not acquired the liquid [ɫ] in coda by the end of her documented sessions. This asymmetry, too, calls for further cross-linguistic verifications, as the data are clearly supported in French and German, two languages which display clear [l] in syllable codas, while the relationship is challenged by learners of two languages which display dark [ɫ] in codas. Interesting in this regard is the observation that the evidence originally brought in support of this relationship also does not come from English, but primarily from Dutch syllable structure development (Fikkert, 1994; Baertsch, 2002).

“Word-initial /r/ implies post-vocalic /r/”

This relationship is not clearly borne out by any of the children’s data. It may minimally be supported by Adrien’s data, who acquired rhotics in onset and coda at the same age, at 4;01.13. William acquired word-initial [ɹ] at 1;07.08, but he did not acquire coda [ɹ] until 2;02.09. Both of the German children acquired word-initial [ʀ] relatively early (Eleonora at 1;10.02; Wiglaf at 1;11.13), but neither had acquired word-final [ʀ] by their last documented sessions. Similarly, for Portuguese, which displays different rhotics in word-initial onsets ([ʀ]) versus syllable codas ([ɾ]), Inês acquired word-initial [ʀ] by 3;00.15, but did not acquire word-final [ɾ] by her last documented session. Finally, Ben’s data do not permit an assessment of this implication, as he had not acquired [ɹ] in either position by the end of his recordings. The clear difference between most of these observations and the prediction made by this implicational relationship may stem from a methodological difference, given that the original observations brought in support for it come from a cross-sectional survey of developmental patterns in English (Smit 1993), as opposed to specific data from individual learners.

“Fricatives in initial position imply fricatives in final position”

This relationship is supported by data from Wiglaf, Inês, and Ben, but contradicted by data from William, Adrien, and Eleonora. Wiglaf acquired [h] in onset at 1;05.26, and [f] in coda slightly earlier, at 1;05.03. Inês acquired fricative [f] in initial position at 2;07.16 and fricative [ʃ] in coda much earlier at 2;00.11. Ben acquired fricative [ʃ] in onset position at 3;11.18 and had already acquired fricative [s] in coda by the earliest observation, at 3;09.06. In contrast to this, William acquired fricative [s] in onset position by 1;04.12, but acquired his first fricative in coda, [ʃ], almost two months later, at 1;06.05. Similarly, Adrien acquired his first fricative in onset position, [v], at age 2;02.20, while in coda, he acquired his first fricative, [f], only at age 3;00.16. Eleonora had acquired fricative [h] in onset position by 1;00.07, and acquired fricative [ç] in coda two months later, at 1;02.14. No trends emerge in these data, in line with the fact that, to our knowledge, no theories of syllabification provide grounds for this implicational relationship.

“Voiced obstruents imply voiceless obstruents”

To examine whether this relationship can be validated by the typical English data, earlier documentation of William’s development would have been necessary, as he had acquired both voiced and voiceless stops (namely, [b, d, k]) by 1;04.12, at the beginning of the recording period. Data from three of the remaining children, Adrien, Eleonora, and Wiglaf, support this implicational relationship, while data from the other two children, Inês and Ben, undermine it. Adrien acquired the voiceless stop [t] at 2;00.16, and its voiced counterpart [d] approximately one month later, at 2;01.13. Eleonora had acquired the voiceless fricative [h] by her earliest session 1;00.07, and acquired the voiced stop [b] later, at 1;01.11, while Wiglaf acquired the voiceless stop [p] at 1;03.21, and the voiced stop [b] at 1;06.12. In contrast to this, Inês and Ben did not acquire voiceless obstruents before voiced ones. Inês acquired the voiced stop [d] at 1;00.25, while her first voiceless stop ([p]) was acquired slightly later, at 1;01.30. In a similar way, Ben had acquired the voiced stops [b] and [d] by his earliest recorded session (3;09.06), but only acquired a voiceless obstruent ([ʃ]) a few months later, at 3;11.18. While it is difficult to discern any pattern in these data, we also suggest that a more systematic investigation of this relationship should involve acoustic analysis, as well as a consideration of the phonetic contrasts for voicing relevant to each language.

“Fricative+liquid clusters imply stop+liquid clusters”

This relationship is supported by the data from William and Inês, but contradicted by Wiglaf’s data, again with the broad definition of liquid weighing in the balance. The datasets from the other three children did not offer the evidence required to assess it. William acquired fricative-lateral clusters at 2;00.12, after he had acquired stop-lateral clusters, at 1;10.12. Inês acquired both fricative-lateral and stop-lateral clusters at 2;08.23. In contrast to this, Wiglaf’s acquisition of clusters follows a different pattern, as he acquired fricative-rhotic clusters at 1;10.28 but stop-lateral clusters only slightly later, at 1;11.03. Note as well that more detailed descriptions of the children’s data which include a distinction between lateral and rhotic liquids, are clearly not supported by the data (see Watts, 2018 for a detailed discussion).

“Clusters with a small sonority difference imply clusters with a greater difference”

To investigate this relationship, we evaluated several different types of clusters but excluded all sibilant+C clusters (sC for English, German, and French, and ʃC German and Portuguese), as these clusters, which can display flat or falling sonority profiles (e.g. [st, sp, sk]), tend to display their own developmental paths in comparison to other onset clusters (Barlow, 1997; Freitas, 1997; Goad & Rose, 2004). The data from Eleonora and Ben did not allow for an assessment of this relationship, as neither acquired consonant clusters within the periods documented by their corpora. William’s data loosely support this relationship, which is however contradicted by the data from Adrien, Inês, and Wiglaf. William acquired stop-glide clusters at 1;09.25, stop-lateral clusters at 1;10.12, stop-rhotic clusters at 2;00.12, fricative-lateral clusters at 2;00.12, fricative-glide clusters at 2;04.03, fricative-rhotic clusters at 2;09.05, and nasal-glide clusters at 2;11.14. The cluster type with the smallest sonority difference (nasal-glide) was the last acquired, and the cluster with the largest sonority difference was acquired first. The clusters acquired in between these however only show weak adherence to the relationship, also depending on how one classifies the relative sonority of fricatives and different liquids.

Adrien acquired fricative-rhotic and stop-rhotic clusters at 4;01.13, an observation which provides minimal support for the relationship. Inês acquired stop-glide clusters at age 2;01.10, and both fricative-lateral clusters and stop-lateral clusters at 2;08.23. Fricative-glide clusters may have been acquired at age 3;02.03, however she made only two attempts at this cluster in the whole dataset. Inês acquired stop-rhotic clusters at 3;10.01 and fricative-rhotic clusters at 3;11.12. While Inês acquired the cluster with the greatest sonority first, the specifics of this relationship apply to each cluster type, whether they involve a lateral or a rhotic consonant. More generally, Inês’s late acquisition of the stop-rhotic cluster relates to her late acquisition of the rhotic, which highlights that this relationship might hinge, at least in part, on language-specific properties of individual phones.

Wiglaf acquired fricative-rhotic clusters at 1;10.28, stop-lateral clusters at 1;11.03, and fricative-lateral clusters at 1;11.13. Wiglaf did not acquire the other three rising-sonority clusters, as summarized in Table III. Wiglaf’s acquisition of clusters thus contradicts this implicational relationship, as stop-glide clusters were not acquired by the last documented session, while their acquisition should have been implied by the other clusters he acquired.

Relationship between stridency, liquids, stops, and glides

This last relationship is, in its original formulation (in 1bviii), rather challenging to apply to any dataset, as it requires comparison of the age of acquisition for up to eight different phonological categories. This relationship thus involves too many dependencies to be applicable in practice; we are thus ruling it out in its current formulation and instead suggest that it be broken into its subparts, many of which would then more likely be compatible with similar relationships in (1a).

Discussion

In summary, the implicational relationships that could be validated for all children, as suggested by our classification in (1), appear to be warranted by either phonological theory or aspects of developmental phonetics. For example, the statement that “consonants imply vowels” in (1ai) is in line with common knowledge about vowels versus consonants in phonetics and phonology, from the earliest emergence of vowels in infant vocalizations (Davis & MacNeilage, 1995; Macken & Ferguson, 1983; Vihman et al., 1985) to the typological observation that while all languages display syllables that include vowels, very few of the world’s languages allow for phonetic syllables comprised of consonants only (cf. Berber; Dell & Elmedlaoui, 1985). Similar logic applies to the relationships in (1aii-vi), where both the main tendencies observed in acquisition data and across adult languages point to the relative markedness of different places and manners of articulation. For example, it is well documented that obstruent and nasal stops typically are, alongside glides and vowels, the earliest manners of articulation to emerge in babbling (de Boysson-Bardies et al., 1989; Locke, 1983; Stoel-Gammon, 1985; Winitz & Irwin, 1958). Stops are indeed less complex than fricatives, because they require only a ballistic closure gesture whereas fricatives require precise control of a narrow constriction (Edwards & Beckman, 2008; see, also, Gayraud et al., 2018). Finally, the last relationship that applies to all children, whereby “clusters imply singletons,” in (1avii) makes perfect sense at the level of syllable structure complexity, whereby a singleton syllable position is inherently less complex than one harbouring a sequence of phones, as illustrated for syllable onsets in Figure 2.

Figure 2:

Simple vs. branching onsets

Turning now to the relationships that are contradicted by at least some of the results from our cross-linguistic survey, we attribute those to two different reasons. Starting with the relationships introduced in (1b), we could not establish their validity as they involve units which cannot be formally related. For example, Gierut states in (1bi) that the presence of clusters implies the presence of affricates, but not vice versa. However, this logic ignores the fact that affricates also involve a certain degree of segmental complexity, given the stop-fricative aperture sequence that defines this class of phones. These two levels of complexity are illustrated in Figure 3.

Figure 3:

Affricate vs. branching onset

While the different structures in Figure 3 may be compared on independent grounds, for example in terms of their relative frequency in usage, or at the level of speech phonetics, the nature of these comparisons transcend the current discussion.

The same holds true of the remainder of the relationships in (1b). The three relationships in (1bii-iii) attempt to draw formal relationships between syllable onsets and syllable codas, which are considered to be formally independent units within syllable structure (e.g. Baertsch, 2002; Fikkert, 1994; Rose, 2000, and references therein), for which the only markedness relation is that syllable onsets are typologically less marked than codas (Blevins, 1995). Likewise, while the implicational relationship in (1biv) captures asymmetries relevant to positional effects conditioning the development of fricatives (e.g. Marshall & Chiat, 2003), it has no foundations in phonological theory, given the same formal distinction between syllable onsets and codas.

In contrast to the above relationships in (1bi-iv), the one in (1bv) could potentially be revised by adding references to positional determination, given the widespread, positionally-determined patterns of voicing and devoicing in different phonological contexts we observe both across languages (e.g. Wetzels & Mascaró, 2001) and in phonological development (e.g. Smit, 1993).

The two statements in (1bvi-vii) suggest implications that cannot be generalized across all languages, where sonority relations appear to vary both formally and in terms of usage frequency (e.g. Blevins, 1995). As we noted above, the types of consonants, and especially that of the liquids, involved in each cluster, can also govern patterns of cluster development, also in relation to the consonant that precedes them within the cluster. Such effects can only be predicted on a language-specific basis.

Finally, building on the logic of the current discussion, the implicational relationship in (1bviii) involves a number of unwarranted relations between individual sounds and/or sound classes, without specifying any positional detail. This relationship is, in practical terms, virtually impossible to test. As we suggested above, breaking it into its multiple subparts would enable a better assessment of its relevant, from both theoretical and practical perspectives.

Conclusion

Returning to the role of markedness in driving aspects of phonological and phonetic development, our results also emphasize that this notion is both valid and useful, so long as it is applied in ways which are constrained by aspects of phonological or phonetic complexity. Our results also suggest that while markedness can be invoked on universal grounds to address the most basic aspects of phonological systems (e.g. the distinction between singleton and complex syllable constituents such as syllable onsets and codas), markedness relations should ideally be established based on language-specific markedness, also in light of structural aspects of phonological systems, in conjunction with how these relations are expressed phonetically within the relevant languages. These results are fully in line with Hume’s (2011) view of markedness, which highlights the relevance of language-specific factors for all analyses of phonological development and developmental speech disorders. In turn, a better theoretical understanding of implicational relationships has the potential to make the complexity approach more predictive of developmental patterns of phonology and applicable within clinical settings, toward both the selection of initial therapy targets and the determination of efficient treatment options.

Appendix A:

Phonological Characteristics of English, French, German, and Portuguese

Characteristics	Language: English Language Family: Indo-European	Language: French Language Family: Indo-European	Language: German Language Family: Indo-European	Language: European Portuguese Language Family: Romance
Syllable shapes	Syllables can be as small as a single vowel or diphthong (e.g. I) and as large as three consonants before and after the vowel (C(0–3)VC(0–3)) (e.g. sprints) (Smit, 2007), with words like sixths [sɪksθs] a more extreme, rarely occurring case. In terms of internal syllable structure, English allows both vowels and sonorant consonants (such as [ɹ] or [n]) to occur in the nucleus of the syllable.	French displays, at the phonetic level, a syllable structure of C(0–3)VC(0–3) (Rose & Wauquier-Gravelines, 2007). In syllable nucleus, only vowels are allowed. French allows for branching onsets, which must have rising sonority, as per the Sonority Sequencing Principle.	German syllables can be as small as a single vowel and as large as three consonants in prevocalic and postvocalic position in a monosyllabic word (i.e. C(0–3)VC(0–3)) (e.g. Strasse /ʃtʁ/; Wiese, 1996).	Portuguese has a syllable structure of C(0–2)VC(0–2) and allows for up to two consonants in prevocalic and postvocalic position. The rhyme always contains a vowel or diphthong within the nucleus; Portuguese does not have syllabic consonants (Mateus & d’Andrade, 2000).
Tones	None	None	None	None
Syllable stress	English has lexical stress, which implies that every content word must have one stressed syllable while most multi-syllabic words show alternating stress patterns. Stress is often assigned to the first syllable of words, however a multitude of factors, such as word’s grammatical category or the number of affixes present in a word, can affect stress placement (Smit, 2007).	Stress consistently falls on the final syllable of phrases (e.g. Tranel, 1981; Kaye & Lowenstamm, 1984; Charette 1991). With the exception of schwa, vowels that occur in the last syllable of a phrase (or isolated word) consistently receive stress.	Penultimate stress placement is considered the regular stress pattern in German (Kohler, 1977; Fox, 2007). A number of factors affect stress placement, including morphological affixation. However, one of the last three syllables of the word always receives stress (Wiese 1996).	Stress assignment in Portuguese is affected by word category and morphological inflection. Stress always falls on one of the last three syllables (i.e. the final, penultimate, or antepenultimate syllable). In general (about 80% of the native vocabulary), stress falls on the final syllable of bare stems and on the penultimate syllable if there is a class marker (Mateus & d’Andrade, 2000).
Vowels and diphthongs	[i, ɪ, e, ɛ, æ, ʌ, ɑ, o, ʊ, u, ə, aɪ, aʊ, ɔɪ]	[i, y, u, e, ø, o, ɛ, $\tilde{\epsilon }$, œ, $\widetilde{œ}$, ə, ɔ, $\widetilde{ɔ}$, a, $\tilde{\text{a}}$, j, ɥ, w]	[iː, ɪ, eː, ɛ, ɛː, ʊ, uː, oː, ɔ, yː, ʏ, øː, œ, ə, ɐ, a, aː, j, w] *[ɐ] is a common allophone of post-vocalic [ʀ] (Wiese 1996); a diphthong ending in [ɐ] may be formed by syllables ending in [ʀ]	[i, $\tilde{\text{I}}$, u, $\tilde{\text{u}}$, ɯ, e, $\tilde{\text{e}}$, o, $\tilde{\text{o}}$, ɐ, $\widetilde{ɐ}$, ɛ, a, ɔ, j, w]
Consonants	[p, b, t, d, k, g, w, r, l, j, m, n, ŋ, f, v, θ, ð, s, z, ʃ, ʒ, h, tʃ, dʒ, ʔ]	[p, b, t, d, k, ɡ, f, v, s, z, ʃ, ʒ, m, n, ɲ, l, ʁ, w, ɥ, j]	[p, b, t, d, k, ɡ, (ʔ), f, v, s, z, ʃ, (ʒ), (ç), (x), (h), $\left(\widehat{\text{pf}}\right)$, (ts), (tʃ), (dʒ), ʁ, m, n, (ŋ), l] *German contains 15 uncontroversial consonant phonemes (there is debate over the allophonic vs. phonemic status of the sounds listed between parenthesis).	[p, b, t, d, k, g, f, v, s, z, ʃ, ʒ, m, n, ɲ, l, ʎ, ɾ, ʀ
Clusters	English allows clusters in word-initial, -medial, and -final position. Word-initially, /tl/, /dl/, /pw/, and /bw/ are not permitted. Additionally, voiced fricatives cannot occur as the first member of a branching onset.	Before a vowel, clusters are restricted to /s/ followed by an obstruent and a liquid (e.g. splendide $\left[\text{spl}\tilde{\text{a}}\text{did}\right]$) or they can consist of an obstruent followed by a liquid-glide combination (e.g. pluie [plɥi]), where the glide [ɥ] is part of the nucleus (Kaye & Lowenstamm, 1984). After the nucleus, French does not allow nasal codas, as already mentioned, but allows obstruent and liquid codas. Adjacent obstruents agree in terms of voicing values (i.e. adjacent obstruents are voiced or voiceless) (e.g. opter [ɔpte], but not *[ɔbte]). Word-initial clusters: [pl, pr, bl, br, fl, fr, vl, vr, tr, dr, kl, kr, ɡl, ɡr, sp, st, sk, spl, spr, str, skl, skr, pn, ps, pf, pt, kn, km, kv, ks, kt, tl, tm, ts, tʃ, sl, sm, sn, sv, sf, ʃl, ʃr, ʃn, ʃv, ʃpr, psk, ɡn, ɡz, dz, dʒ, zl, zv, zb, zɡr, mn, ft] Sonorant-obstruent clusters: [ʀl, ʀm, ʀn, ʀf, ʀs, ʀʃ, ʀç, ʀp, ʀt, ʀk, lm, ln, lf, ls, lʃ, lç, lp, lt, lk, mf, ms, mʃ, mp, mt, nf, ns, nʃ, nç, nt, ŋs, ŋʃ, ŋt, ŋk] Obstruent-obstruent clusters: [sf, sp, st, sk, fs, ft, χs, χt, ʃs, ʃt, ts, tʃ, ks, kt, pf, ps, pʃ, pt]	Syllable initial: [pl, pʀ, pn, ps, tʀ, tv, kl, kʀ, kn, km, ks, kv, bl, bʀ, dʀ, ɡl, ɡʀ, ɡn, ɡm, fl, fʀ, vl, vʀ, tsv, pfl, pfʀ, ʃl, ʃʀ, ʃn, ʃm, ʃv] Syllable final: [ʀl, ʀm, ʀn, ʀf, ʀs, ʀʃ, ʀç, ʀp, ʀt, ʀk, m, ln, lf, ls, lʃ, lç, lp, lt, lk, mf, ms, mʃ, mp, mt, nf, ns, nʃ, nç, nt, ŋs, ŋʃ, ŋt, ŋk, sf, sp, st, sk, fs, ft, χs, χt, ʃs, ʃt, ts, tʃ, ks, kt, pf, ps, pʃ, pt]	Syllable initial: [pɾ, bɾ, tɾ, dɾ, kɾ, ɡɾ, fɾ, pl, bl, fl, kl, ɡl, pt, bd, kt, ps, pn, tm, ɡn, mn] Syllable final: [pt, bt, bd, dk, kt, bs, bv, bʒ, tz, dv, ks, pn, bn, tm, tn, dm, dn, ɡm, ɡn, mn]
Examples of phonological constraints	The flap [ɾ] and glottal stop [ʔ] are frequently occurring allophones in English, with flap especially prominent in North-American dialects. The flap [ɾ] is an allophone of /t, d/ and occurs in between vowels in onsets of unstressed syllables (e.g. atom [æɾəm]); the glottal stop [ʔ] occurs frequently as an allophone of post-vocalic /t/ when it occurs before an alveolar (syllabic) nasal (e.g. button [bʌʔən]) (Hammond, 1999). Syllable- and word-final /t/ is also commonly expressed as [ʔ] in words such as department [dɪpɑʔmənt], foot [fʊʔ], and start [stɑɹʔ]. English also has two /l/ sounds. /l/ is velarized as [ɫ] when it occurs after a vowel or before a consonant at the end of a word (e.g. pool [puɫ] and help [hɛɫp]).	A word-initial onset in French can contain any consonant except /ɲ/. Medial codas can contain any single consonant with the exception of nasals; nasals occur in word-final position (e.g. canne [ka.n]), where they are syllabified as onsets of empty-headed syllables, similar to all consonants in this position (Piggott, 1999; Rose, 2000).