Children ’ s Consonant Acquisition in 27 Languages : A Cross-Linguistic Review

Purpose: The aim of this study was to provide a crosslinguistic review of acquisition of consonant phonemes to inform speech-language pathologists’ expectations of children’s developmental capacity by (a) identifying characteristics of studies of consonant acquisition, (b) describing general principles of consonant acquisition, and (c) providing case studies for English, Japanese, Korean, and Spanish. Method: A cross-linguistic review was undertaken of 60 articles describing 64 studies of consonant acquisition by 26,007 children from 31 countries in 27 languages: Afrikaans, Arabic, Cantonese, Danish, Dutch, English, French, German, Greek, Haitian Creole, Hebrew, Hungarian, Icelandic, Italian, Jamaican Creole, Japanese, Korean, Malay, Maltese, Mandarin (Putonghua), Portuguese, Setswana (Tswana), Slovenian, Spanish, Swahili, Turkish, and Xhosa. Results: Most studies were cross-sectional and examined single word production. Combining data from 27 languages,


C
hildren's acquisition of speech involves mastery of the perception and production of consonants, vowels, consonant clusters, tones, prosodic features, and phonological rules of the language(s) they speak, with the outcome of intelligible speech.Mastery of consonants is one of "the most widely used metrics of typical phonological acquisition and of phonological disorder" (Edwards & Beckman, 2008a, p. 937).Since the 1930s (e.g., Poole, 1934;Wellman, Case, Mengert, & Bradbury, 1931), many researchers have documented children's age of acquisition of consonants.In 1972, Sander published an article entitled, "When Are Speech Sounds Learned?", where he graphically summarized two studies of English consonant acquisition (Templin, 1957;Wellman et al., 1931) to describe customary versus mastery production and account for variability between studies.In his famous figure, the shorter bars indicated less variability in the age of acquisition (e.g., /p, m/) and the longer bars indicated greater variability (e.g., /t, s/).His documentation of customary and mastery production of consonants has been cited repeatedly and has been widely used by speech-language pathologists (SLPs) as a summary of English-speaking children's speech acquisition, despite the fact that there have been additional studies of children's acquisition of English published since this time (e.g., Dodd, Holm, Hua, & Crosbie, 2003;Smit, Hand, Freilinger, Bernthal, & Bird, 1990).
In the following decades, two additional summaries of consonant acquisition have been adopted by SLPs: percentage of consonants correct (PCC) and early-middle-late consonants.PCC was originally described by Shriberg and Kwiatkowski (1982) and was calculated by dividing the number of consonants produced correctly by the total number of consonants in a connected speech sample.PCC has been linked to descriptors of severity of involvement for children with speech sound disorders (SSD; e.g., mild, moderate, severe), and the construct has undergone minor modifications over the years (Shriberg, Austin, Lewis, McSweeny, & Wilson, 1997;Shriberg, Kwiatkowski, Best, Hengst, & Terselic-Weber, 1986).Since development, PCC and its variants (percentage of vowels correct [PVC] and percentage of phonemes correct [PPC]) have been widely used by SLPs to document speech acquisition of typically developing children (e.g., Dodd et al., 2003;Fabiano-Smith & Goldstein, 2010a) and those with SSD (e.g., McLeod, Harrison, McAllister, & McCormack, 2013) to support diagnosis of SSD.The third summary of consonant acquisition was also developed by Shriberg (1993), who grouped the 24 English consonants into early-8, middle-8, and late-8 consonants on the basis of data from 64 children with SSD aged 3-6 years.The early-middle-late construct has been used to describe the speech of typically developing children and children with SSD in English and Spanish (Fabiano-Smith & Goldstein, 2010b;Flipsen, Hammer, & Yost, 2005;Shriberg, Gruber, & Kwiatkowski, 1994) and has been used as a focus of intervention (Bleile, 2017).SLPs frequently use these three constructs (age of acquisition, PCC, and earlymiddle-late) to describe children's speech acquisition as they work with children in assessment, diagnosis, determination of intervention targets, and decision making regarding successful outcomes of intervention (Bernthal, Bankson, & Flipsen, 2017;McLeod & Baker, 2017).
English is not the first language of a large portion of the world's population (Simons & Fennig, 2018), and most English-dominant countries contain speakers of many different languages (e.g., 20.8% of the U.S. population speaks a language other than English at home, with the second most spoken language being Spanish; Ryan, 2013).Many studies have documented English-speaking SLPs' lack of confidence when providing services to children who speak languages that the SLPs do not speak (Caesar & Kohler, 2007;Guiberson & Atkins, 2012;Kritikos, 2003;Roseberry-McKibbin, Brice, & O'Hanlon, 2005;Williams & McLeod, 2012).In most of these studies, SLPs indicated that they do not have access to relevant research and resources to work effectively in cross-linguistic or multilingual contexts.However, in the case of children's speech acquisition, researchers across the world have been documenting the age of acquisition of consonants for languages other than English for many years.In the past decade, English-language readers have had greater access to normative studies of speech acquisition in a range of languages via books, book chapters, and websites (e.g., Hua & Dodd, 2006;McLeod, 2007McLeod, , 2010McLeod, , 2016a)).This increase in access to normative studies in languages other than English has answered recent calls to raise the visibility of scientific publications in languages other than English (Meneghini & Packer, 2007) and to answer criticism that much of the research about human behavior comes from western, educated, industrialized, rich, and democratic societies that are not representative of diverse cultures (Henrich, Heine, & Norenzayan, 2010).However, with over 7,000 documented languages in the world (Simons & Fennig, 2018), there will never be available speech acquisition data for every language and dialect.What SLPs need is a comprehensive cross-linguistic review to collate data about consonant acquisition (including age of acquisition, PCC, and early-middle-late constructs) to inform SLPs' expectations of children's acquisition of consonants across the world.

Theoretical Consideration of Consonant Acquisition Across Languages
A number of influential theorists and researchers have drawn on cross-linguistic studies to describe children's speech acquisition (Edwards & Beckman, 2008b;Jakobson, 1941Jakobson, /1968;;Locke, 1983;Stokes & Surendran, 2005;Vihman, 1996).Those who considered linguistic universals have focused on the acquisition of features of sounds by looking for generalizations across a large number of children.For example, Jakobson (1941Jakobson ( /1968) proposed a structuralist theory of phonological acquisition to describe the order of acquisition of feature classes (e.g., plosives are acquired before fricatives).Jakobson famously stated that "the relative chronological order of phonological acquisitions remains everywhere and at all times the same…the speed of this succession is, in contrast, exceedingly variable and indi-vidual…" (p.46).He argued that the universal properties of sounds contributed to the order and age of acquisition and features were considered to be unmarked (simple, earlier to be acquired) and marked (more complex, later acquired; Jakobson, 1941Jakobson, /1968).Locke (1983) similarly argued that children master phonetic features in a similar sequence in presenting his analysis of studies of English, German, Japanese, Russian, Italian, Arabic, Slovenian, Swedish, Norwegian, and Czech.In contrast, other linguists have challenged the concept of linguistic universals by demonstrating a wide variability between children and insisting on the importance of individual children's acquisition.For example, Ferguson and Farwell (1975) presented a cognitive model of speech acquisition to account for individual variability and argued against writing rules and generalizations.Vihman (1996) also argued for a cognitive model to describe individuals' segmental and prosodic acquisition and proposed a templatic phonology (Vihman & Croft, 2007).
Some researchers have considered children's speech acquisition by combining both general principles and individual capacity.Elbert (1984) reconciled these two seemingly divergent theoretical perspectives by applying them to SLPs' clinical practice.She suggested that SLPs should consider general overarching principles from speech acquisition norms concurrently with viewing individual children as creative learners who are "engaged in highly energetic pursuit of knowledge" (p.115).The biological account of speech acquisition proposed by Kent (1992) similarly combined both individual capacity and general principles.Kent outlined three broad principles of biological development (variability, reversals and revisions, and excess potential for development) and then described four sets of English consonants differentiated by articulatory complexity by drawing on the work of Sander (1972) and Set 4 [s, z, ʃ, ʒ, ʧ, ʤ, v, θ, ð].Each set was described regarding the motoric adjustments required to articulate each sound; for example, Set 4 requires adjustment of tongue place and configuration to produce fricatives.More recently, the emergence approach to speech acquisition has been described by Davis and Bedore (2013) combining both general principles and individual capacity by outlining how children's intrinsic capacities (perception, production, and cognitive) and interaction capacities (joint attention, turn-taking, and intention reading) intersect with the extrinsic context (ambient phonology and sociocultural influences).Combining both general principles and individual capacity is particularly important for children who learn to speak more than one language, especially because there are limited normative data regarding multilingual speech acquisition, and researchers are confounded by the need to account for differences in the age of exposure, level of proficiency, and use of each language for individual children.
To date, description of general principles of consonant acquisition has been undertaken in three ways: consonants within one language (e.g., Kent, 1992;Sander, 1972), features across languages (e.g., plosives develop before fricatives; Jakobson, 1941Jakobson, /1968;;Locke, 1983), or consonants across languages (e.g., Edwards, Beckman, & Munson, 2015;McLeod & Singh, 2009).With recent availability of consonant acquisition data across a wide range of languages, it is possible to expand our knowledge of general principles of consonant acquisition across languages.It is important to consider the consonant phoneme (contrastive speech sound within a language) as the unit of analysis rather than the phone (speech sound) or allophone (variations of phonemes) when undertaking a cross-linguistic review of consonant acquisition to provide generalizable information regarding similarities and differences between languages.A large-scale review is required to provide an overview of patterns of consonant phoneme acquisition across languages and general cross-linguistic patterns that could be applicable to languages not already studied.

Aims
The motivation for the current research was to inform SLPs' expectations of cross-linguistic consonant acquisition.Specifically, the aims of this review were to (a) identify and describe studies of consonant acquisition across languages; (b) provide general principles regarding the age of acquisition of consonant phonemes, PCC, early-middle-late consonants, and manner and place characteristics across languages; and (c) consider the application of these general principles to four languages as case studies: English, Korean, Japanese, and Spanish.

Method
A systematic literature search was undertaken, and a scoping review framework (Colquhoun et al., 2014) was used to examine a broad range of literature and synthesize cross-linguistic knowledge about children's acquisition of consonant phonemes.

Source 1
Thirteen databases were searched to identify the full range of published literature describing children's speech sound acquisition (Cambridge Journals Online; Cochrane; EBSCO Host; Google Scholar; The Scholarly Journal Archive [JSTOR]; Linguistics, Language, and Behavior Abstracts; Medline; Oxford Journals; PsycInfo; PubMed; Sage Journals; Springer Link; and Wiley Online Library).Four search terms were used across all databases-"children" AND "consonant" AND "acquisition" OR "development"-and 1,684 citations were located.Duplicates of citations were identified and removed, leaving 623 unique citations.

Source 2
The first author collated a database of studies of children's speech acquisition over 10+ years from journal articles and by attending conference presentations, visiting speech-language pathology clinics around the world, and contacting colleagues who work in different countries and speak languages other than English.In addition, authors of chapters in the International Guide to Speech Acquisition (McLeod, 2007) were asked to document speech acquisition studies that were developed for the languages they were writing about, and these were included in the review.Forty relevant unique articles were located that were not found during the database search (Source 1).

Source 3
Members of the International Expert Panel on Multilingual Children's Speech (McLeod, Verdon, & IEPMCS, 2017) were provided with the list of articles from Sources 1 and 2 and were invited to send additional articles to the authors.Five relevant additional articles were found.
A list of all articles is found on the Multilingual Children's Speech website (McLeod, 2016b).

Inclusion and Exclusion Criteria
The second author examined titles and abstracts of the 668 articles to determine whether they met the following criteria: (a) describe singleton consonant acquisition (n = 474 excluded); (b) present research data (n = 5 excluded); (c) describe typical speech and language development (n = 33 excluded); (d) be a journal article, book chapter, or dissertation (n = 13 excluded); and (e) describe 10 or more participants (n = 19 excluded).Exclusion of articles was hierarchical.After these inclusion criteria were applied, 124 articles remained.
Next, the entire articles were examined and excluded if (a) the full consonant repertoire of the language was not described (n = 16 excluded), (b) ages and/or consonants were not specified or not presented in a usable form (n = 25 excluded), (c) the only criterion for consonant acquisition was < 75% (n = 11 excluded), (d) data from monolingual and multilingual participants were not presented separately (n = 10 excluded), or (e) data reported within the article were inconsistent between the text, figures, and tables (n = 2 excluded).After the inclusion and exclusion criteria were applied to the articles, 60 articles remained describing 64 studies (some articles described more than one study).For articles where the inclusion/exclusion criteria were not clearly met, both authors examined and discussed the article until consensus about eligibility was reached.

Procedure
The 60 articles describing 64 unique studies were reviewed, and data were extracted describing article/study characteristics (year and language of publication), participant characteristics (number, age, sex, language/s, dialect, and country), research methods (speech sample type, study design, reliability, sensitivity and specificity, and acquisition criteria), and results (age of consonant acquisition, PCC/PVC/PPC) building on Smit (1986).
The most appropriate International Phonetic Alphabet (IPA; International Phonetic Association, 2015) symbol was selected when non-IPA symbols (e.g., orthographic symbols) were used by reading the entire article and considering additional literature that documented consonants of the language (e.g., International Phonetic Association, 1999;McLeod, 2007).For example, the Linares (1981) article about Spanish described ages of acquisition for "r" and "rr."These were entered in the current study as "r" = /ɾ/ (flap) and "rr" = /r/ (trill).In rare cases, if the non-IPA symbol was unable to be resolved, the consonant was not included.
Data were entered for all consonant phonemes described in each article relevant to each language and dialect.For example, Korean has three ways to distinguish plosives (lenis, fortis, and aspirated) on the basis of voice onset time and vowel onset fundamental frequency.Consequently, there were three Korean contrastive velar plosive phonemes included in the analysis: /k, k*, k h / (M.Kim & Pae, 2007).The four Cantonese contrastive velar plosive phonemes, distinguished by aspiration and labialization, were included in the analysis: /k, k h , k w , k wh / (Zee, 1999).In contrast, English has one voiceless velar phoneme /k/ ([k] and [k h ] are allophones and are not contrastive), so /k/ was the only voiceless velar plosive phoneme included in the review for English.Therefore, in the current review, data were entered separately for the phonemes /k, k*, k h , k w , k wh / to represent phonemes across languages and were not combined under /k/.Appendix A provides a list of all consonant phonemes assessed and acquired in each language.
Age-of-acquisition data were extracted from each of the reviewed studies.Typically, participants' age-of-acquisition data were reported in the studies as the minimum age (in years/months) when 75% of the participants had acquired a phoneme and/or when 90% of the participants had acquired a phoneme.If age of acquisition differed across word position or sex of the participants, then the youngest age was recorded.PCC and PVC were recorded in 6-month age intervals.If articles reported these data in 12-month intervals, then data were entered twice at each appropriate 6-month point.If an article was published in a language other than English, the authors used translated data about the article from McLeod (2007), asked the authors of the article (or SLPs) who were fluent in the language to translate specific sections, and/or used Google Translate for short segments (cf.McLeod & Verdon, 2014).

Data Analysis
All data were entered into SPSS Version 23.0 (International Business Machines, 2015) and analyzed in terms of frequency, central tendency (mean and median), and variability (standard deviation and range).The unit of data for analysis was the age of acquisition (in months) for each consonant phoneme, and age of acquisition was categorized as 75%-85% and 90%-100%.Criteria for inclusion within individual studies were not modified but were included in the analysis either in the 75%-85% and/or 90%-100% groups as appropriate.When children in the youngest age group of a study acquired a consonant (75%-85% and/or 90%-100% criteria), the data were included in the analysis, and this was noted.When children in the oldest age group of a study did not acquire a consonant (75%-85% and/or 90%-100% criteria), these data were not included in the analysis but were counted for reporting.For example, in the study of 999 children speaking Afrikaans by Lotter (1974), /s/ and /r/ were not acquired by 113 months old (90% criterion) and so were not included in the data analysis but were reported in Appendix A. Consonant phonemes that were not acquired are reported in the results and Appendix A.
All consonant phonemes were considered in analyses (as described previously), and data are presented in the text and tables.Classification of consonant phonemes into place and manner was based on categories from the IPA (International Phonetic Association, 2015).Because of the large number of consonants described in this article, the graphs contain a subset of consonants, whereas the tables and appendices include all consonants.Consonants presented in graphs are those represented by core, unmodified IPA symbols: consonants (pulmonic), consonants (nonpulmonic), and other symbols.Four case studies are presented for languages that were described in four or more studies: English, Korean, Japanese, and Spanish.

Interrater Reliability
The first author completed reliability checks on data extracted for each of the study characteristics (e.g., country, dialect) and data points for each consonant (e.g., age of acquisition) for seven studies (10.9%).A point-bypoint analysis of reliability was conducted, and interrater reliability was 96.7% across all variables and 1,145 data points.Discrepancies were discussed between authors until agreement was reached, and records were amended accordingly.

Demographic Data
The ages of the children in the 64 studies ranged from 5 months (0;5) to 155 months (12;11; see Appendix B).The minimum age examined within a study ranged from 5 to 60 months (M = 28.69months, median = 28.50 months, SD = 11.52 months).The maximum age studied ranged from 24 to 155 months (M = 73.50months, median = 72.00months, SD = 27.93 months).For the 40 studies that provided this information, there were data for 5,244 males and 5,267 females.The number of males and females was unable to be determined for 24 studies.

Data Collection Techniques
There were 45 studies (70.3%) that collected singleword speech samples, 11 (17.2%) that collected connected speech samples, and seven (10.9%) that collected singleword and connected speech samples, and the remaining study (1.6%) collected single words and isolated speech sounds (see Appendix B).

Word Position
Fifty-five (85.9%) studies described acquisition of consonants in the word-initial position, two (3.1%) did not study consonants in this position, and seven (10.9%) did not specify whether they did or not.Forty (62.5%) studies described acquisition of consonants in the within-word position, 11 (17.2%) did not, and 13 (20.3%)did not specify.Forty-two (65.6%) studies described acquisition of consonants in the word-final position, 10 (15.6%) did not, and 12 (18.8%)did not specify.On many occasions, the language being tested mediated the word positions that were tested.For example, consonants rarely occur in the initial or final position of Xhosa, so Mowrer and Burger (1991) only tested consonants in the within-word position.

Additional Data Collected
All 64 (100%) studies described consonant acquisition (because this was the main criterion for inclusion in this review).In addition, 29 (43.9%) also described consonant cluster acquisition, 22 (33.3%)described vowel acquisition, and three (4.5%)described tone acquisition;  however, these additional data were not analyzed in the current article.Nineteen studies (28.8%) provided information about PCC, eight (12.1%)provided information about PVC, and four (6.1%) provided information about PPC.

Description of the Data Analysis Techniques
Age-of-Acquisition Criteria The 64 studies reported the criteria used to document the age of acquisition of consonant phonemes.Fourteen studies reported two criteria (e.g., 75% and 90%), and the remainder reported one criterion.There were 32 studies (41.0%) that used a criterion of 75%, three (3.8%) that used an 80% criterion, one (1.3%) that used an 83% criterion, two (2.6%) that used an 85% criterion, 36 (46.2%) that used a 90% criterion, one that used a 95% criterion (1.3%), and three (3.8%) that used a 100% criterion.Supplemental Material S1 presents summary information about the age of acquisition of consonant phonemes at the 75%-85% and 90%-100% criteria levels across studies.Twenty studies (30.3%) required the consonant to be correct in one word position, five studies (7.6%) required the consonant to be correct in two word positions, and seven (10.6%) required the consonant to be correct in three word positions; however, this information was unable to be determined for 31 (47.0%)studies.It is important to note that accuracy based on word position needs to differ between languages; for example, in Cantonese, only two word (syllable) positions exist and most consonant phonemes are produced in the word-initial position.

Reliability
Twenty-nine (45.3%) studies provided interrater reliability, one (1.6%)provided intrarater reliability, and seven (10.9%) provided interrater and intrarater reliability.For the remaining 27 studies (42.2%), reliability measures were unable to be determined because some articles were not written in English or were published many years before reliability measures were standard practice (e.g., Poole, 1934).

Sensitivity and Specificity
Sensitivity is the proportion of children with SSD identified correctly.Specificity is the proportion of typically developing children not identified as having SSD.No articles described sensitivity or specificity.
Figure 1 demonstrates the mean age of acquisition of consonant phonemes across all languages ordered according to the 75%-85% and 90%-100% criteria.As can be seen, children acquire most of their consonant phonemes by 4 years old, with fewer consonants being acquired between 4 and 7 years old (demonstrated by the steeper gradient).Although the general pattern is that consonant phonemes are acquired by 90%-100% of the children at an older age than 75%-85% of the children, there are some consonants that appear to be acquired earlier by 90%-100% of the children (e.g., /ɟ, ɥ/).This may be an artifact of the different studies, methods, and languages examined.Only 14 of the 64 studies considered acquisition using both the 75%-85% and 90%-100% criteria.
Many consonant phonemes were acquired at a similar age across languages as demonstrated by the standard deviations being less than 12 months in Supplemental Material S1.For example, the voiceless labiodental fricative /f/ was acquired, on average, at 31.76 months old (75%-85% criteria), with a standard deviation of 5.92 months and a range of 18-43 months old, across 33 studies of 18 languages.Similarly, the voiceless palatal fricative /ç/ was acquired, on average, at 48.20 months old (90%-100% criteria), with a standard deviation of 0.45 months and a range of 48-49 months old, across six studies of three languages (German, Greek, and Japanese).There were also some consonant phonemes that were acquired at similar ages within languages.For example, acquisition of the affricate /ʨ*/ was described in four studies of Korean, and in each of these studies, it was documented as being acquired at 36 months (75%-85% criteria).However, there were some exceptions where the analysis demonstrated wide differences in the age of acquisition between studies.For example, the voiced labiodental fricative /v/ was acquired, on average, at 40.48 months (75%-85% criteria), with a standard deviation of 17.86 months and a range of 20-102 months old, across 25 studies of 14 languages.Similarly, there were wide differences in the age of acquisition for the voiced palatal stop /ɟ/, acquired on average at 51.50 months old (75%-85% criteria) with a standard deviation of 20.60 months and a range of 24-72 months, across four studies of three languages (Greek, Turkish, and Xhosa).
When interpreting these data, it is important to be aware that the youngest and oldest ages of the participants in each study may have contributed to basal and ceiling effects of the data.First, a basal effect may be seen in some studies because the youngest age of the participants in each study may influence the age of acquisition that is reported.Overall, 37.7% of consonant phonemes were documented as acquired by the youngest age of the participants.When considering data from studies that use the 75%-85% criteria, 42.8% (395/922 valid data points) were documented as being acquired at the youngest age of the participants (M = 25.87 months, SD = 9.01 months), and for studies that use the 90%-100% criteria, 32.3% (283/876 data points) were documented as being acquired at the youngest age of the participants (M = 30.00months, SD = 12.48 months).Therefore, the basal effect of the age of acquisition data reported within studies should be taken into consideration when interpreting children's competence, and it is possible that many consonant phonemes were acquired at an earlier age.Second, a ceiling effect may be seen in some studies.Overall, 11.6% of consonants were documented as not acquired by the oldest age of the participants.When considering data from studies that use the 75%-85% criteria, 8.4% (78/922 data points) were documented as not being acquired at the oldest age of the participants (M = 67.26months, SD = 21.17months), and for studies that use the 90%-100% criteria, 14.8% (130/876 data points) were documented as not being acquired at the oldest age of the participants (M = 76.10 months, SD = 30.87months).Appendix A lists phonemes not acquired in each language by the participants in the oldest age groups.

Percentage of Accuracy Across Languages
PCC was examined in 15 studies of 12 languages that presented the full repertoire of consonant phonemes for that language: Arabic, Danish, English, French, German, Hungarian, Malay, Portuguese, Setswana, Swahili, Turkish, and Xhosa (see Appendix B).Each of these studies used single-word samples to determine PCC.Eight of these studies examined PCC within six monthly age ranges.All 15 studies examined PCC at 48 months old (4;0), whereas fewer studies examined children aged between 12 months (1;0) and 102 months (8;6).Data from the studies demonstrate that children's PCC steadily increased as children grew older (see Table 2).By 2;0 years, children achieved an average PCC of 63.50, and by 5;0 years, children achieved an average PCC of 93.80.PVC was examined in seven studies of five languages that presented the full repertoire of vowels for that language: Danish, English, Setswana, Swahili, and Xhosa.Again, these data demonstrate that children's PVC increased.By 2;0 years, children achieved an average PVC of 88.20, and by 5;0 years, children achieved an average PVC of 98.02 (see Table 2).PPC was examined in four studies of two languages that presented the full repertoire of consonants and vowels for that language: English and German.By 2;0 years, children achieved an average PPC of 75.51, and by 5;0 years, children achieved an average PPC of 96.92 (see Table 2).

Manner and Place of Articulation
Table 3 and Supplemental Material S2 compare consonants produced using seven different manners of articulation (based on the IPA classification).On average, nasals, plosives, nonpulmonic consonants, approximants, and laterals were acquired earlier than trills, flaps, affricates, and fricatives.

Japanese Consonant Acquisition
Five articles describing four studies reported the age of acquisition of Japanese consonant phonemes in Japan (see Appendix B).The articles were published between 1967 and 1982 (M = 1971.40, median = 1970.50,SD = 6.66).The sum of all participants was 2,299, ranging from 10 to 1,689 (M = 486.40,median = 300.00,SD = 767.37).The ages of the children ranged from 12 months (1;0) to 83 months (6;9).All five studies (100.0%)elicited single-word data.All five studies (100.0%)described acquisition of consonants in the word-initial position.Four studies used a criterion of 90%, and one used a criterion of 100%.The use of interjudge and intrajudge reliability was unable to be determined.

Korean Consonant Acquisition
Four articles describing four studies reported the age of acquisition of Korean consonant phonemes in South Korea (see Appendix B).The articles were published between 1986 and 2005 (M = 1995.5,median = 1995.5,SD = 7.77).The sum of all participants was 845, ranging from 150 to 320 (M = 211.25,median = 187.50,SD = 79.20).The ages of the children ranged from 5 months (0;5) to 76 months (6;3).All four studies (100.0%)elicited single-word data and used a criterion of 75%.The word position and use of interjudge and intrajudge reliability were unable to be determined.
Figure 2d profiles the age of acquisition across the four studies of Spanish-speaking children.Using the 90% criterion (across four studies), no consonants were acquired at a mean age of between 2;0 and 2;11 years (24-35 months; although this age group was studied by De la Fuenta, 1985); eight consonants were acquired at a mean age of between 3;0 and 3;11 years (36-47 months), namely, /p, t, m, k, j, ɲ, l, ʧ/; 12 consonants were acquired at a mean age of between 4;0 and 4;11 years (48-59 months), namely, /ŋ, ʒ, ʤ, ɡ, n, b, d, f, x, ð, w, ɾ/; two consonants were acquired at a mean age of between 5;0 and 5;11 years (60-71 months), namely, /r, s/; and one consonant was acquired at a mean age of between 6;0 and 6;11 year (72-83 months), namely, /β/ (see Figure 2d).Using the 90% criterion across four studies of typical speech acquisition, the following Spanish consonants could be classified accordingly: Across the four studies of Spanish consonants, nasals, plosives, approximants, and laterals typically were acquired earlier than flaps and some fricatives.Spanish consonants produced with the pharynx and posterior tongue typically were acquired earlier than consonants articulated with the anterior tongue; however, there was an interaction between place and manner where anterior plosives and nasals were acquired earlier than anterior fricatives and trills.

Discussion
This article presents the world's largest analysis of consonant acquisition data to date: 60 articles describing 64 studies of consonant acquisition in 27 languages by 26,007 children from 31 countries.Most of the 64 studies of consonant acquisition reported cross-sectional data (81.3%) and elicited single-word speech samples (70.3%).Most described acquisition of consonant phonemes in the word-initial position (85.9%), with fewer describing the word-final position (65.6%) and the within-word position (62.5%).Although all described consonant acquisition, some also described consonant cluster acquisition (43.9%), vowel acquisition (33.3%), and tone acquisition (4.5%).Other studies that exclusively reported data for the acquisition of consonant clusters, vowels, and/or tones were not included in the current review (e.g., McLeod, van Doorn, & Reed, 2001).Some studies included PCC (28.8%),PVC (12.1%), and/or PPC (6.1%).Just over half included interjudge and/or intrajudge reliability measures (58.8%).
This article provides an overview of patterns of acquisition in 27 languages, across four case studies (English, Japanese, Korean, and Spanish), as well as general crosslinguistic patterns that could be applicable to working with a child with a language background that is not represented in the review.General principles of development were generated by considering consonant phonemes as the unit of analysis (rather than language).On average, almost all of the world's consonants were acquired by children's fifth birthdays (see Table 1).Most consonants and vowels were produced correctly by 5;0 years: 93.80 PCC (across 15 studies of 12 languages), 98.02 PVC (across seven studies of five languages), and 96.92 PPC (across four studies of two languages; see Table 2).On average, plosives, nasals, and nonpulmonic consonants (e.g., clicks) were acquired earlier than trills, flaps, fricatives, and affricates.Most labial (bilabial and labiodental), pharyngeal (pharyngeal, epiglottal, and glottal), and posterior lingual (palatal, velar, and uvular) consonants were acquired earlier than those using an anterior lingual placement (dental, alveolar, postalveolar, and retroflex); however, there was an interaction between place and manner.Nasal consonants were among the earliest to develop, and plosives (stops) were acquired earlier than fricatives.Approximant/lateral (liquid) consonants did not always precede fricatives.When individual languages were considered as case studies, the general principles of development were upheld; however, specific consonants not acquired by 5;0 differed slightly.Consonants that were not acquired by 5;0 years were /ɹ, ʒ, ð, θ/ (90%-100% criteria) across 15 studies of English consonants, /z/ (90%-100% criteria) across five studies of Japanese consonants, /s, l/ (75% criterion) across four studies of Korean consonants, and /r, s, β/ (90% criterion) across four studies of Spanish consonants.
In this study, the summarized acquisition of the 24 English consonants did not neatly fall into the early-8, middle-8, and late-8 consonants outlined by Shriberg (1993).Instead, Figures 2a and 3 depict a steady increase in acquisition for most consonants (early-middle), with four consonants, namely, /ɹ, ʒ, ð, θ/, being acquired last.Three of these consonants were included in the late consonants described by Shriberg, namely, /ɹ, ð, θ/, and the fourth consonant, /ʒ/, was excluded from Shriberg's analysis.The other consonants included in Shriberg's late-developing consonants were /ʃ, s, z/, but the current study indicates that these were acquired, on average, at a younger age.It is important to note that Shriberg's work was based on children with SSD, and the current study examined children with typically developing speech.The summarized acquisition of the Spanish consonants in Figure 2c did not neatly fall into the early, middle, and late consonants outlined by Fabiano-Smith and Goldstein (2010b) for eight monolingual Spanishspeaking children.The latest consonants to be acquired according to Fabiano-Smith and Goldstein (2010b) were /l, ð, r, ɾ/, whereas the latest consonants to be acquired according to the compilation of studies in the current article were /r, s, β/.These findings highlight the importance of combining general principles from the current study with individual data that are relevant to children in specific contexts (and dialects).The results from the current study should be used as general guidance, and SLPs are encouraged to source specific studies regarding children who speak the dialect and language within their communities (McLeod, 2016b).

Theoretical Implications
Children's consonant acquisition is a key feature of children's overall development, enabling them to perceive and produce intelligible speech and interact with members of society.The results of the current study support elements of the emergence approach to speech acquisition (Davis & Bedore, 2013) that describes how children's intrinsic capacities (e.g., children's production skills) intersect with the extrinsic context (e.g., the ambient phonology).The results of the current study can be interpreted to demonstrate that children master some categories of phonemes (manner/place) using a similar pattern of acquisition across languages, providing some support for theorists such as Locke (1983) and Jakobson (1941Jakobson ( /1968).However, the wide range and large standard deviations in the acquisition of some individual phonemes also provide support for individual variability and the cognitive model of speech acquisition by theorists such as Ferguson and Farwell (1975) and Vihman (1996).The current study upholds some (but not all) of the principles of markedness (cf.Jakobson, 1941Jakobson, / 1968) and the biological account of speech acquisition proposed by Kent (1992).The influence of functional load, phonetic frequency, and phonotactic probability was unable to be examined because of the lack of language-specific data for many of the languages studied (cf.Edwards & Beckman, 2008b;Ingram, 2012;Stokes & Surendran, 2005).Although the current study supported elements of the emergence approach to speech acquisition (Davis & Bedore, 2013), to comprehensively test this functionalist model of speech acquisition, future research should incorporate (a) children's production, perception, and cognition skills; (b) children's interaction capacities; (c) the extrinsic context including adults' ambient phonology and sociocultural influences; (d) the influence of phonetic complexity, functional load, phonetic frequency, and phonotactic probability; and (e) comparison of data from monolingual and multilingual children, keeping in mind Elbert's (1984) assertion that children are creative learners who are "engaged in highly energetic pursuit of knowledge" (p.115).To do this, researchers could consider undertaking cross-sectional and longitudinal studies of both languages of multilingual speakers, gathering a range of data (single words, nonwords, and connected speech) from children who were typically developing and with SSD, such as those documented in the comprehensive study undertaken by Albrecht (2017) to consider German-Turkish-speaking children's speech acquisition.

Clinical Implications
Children's consonant acquisition and accuracy are the main indicators used by SLPs to measure children's speech maturity and intelligibility.The current article draws together a large body of literature on consonant acquisition from across the world and adds to the information available to support SLPs' work with children who speak a range of languages.The article provides guidance to support SLPs' expectations of cross-linguistic consonant acquisition, including for languages where there are no data currently available.The current article provides a preliminary resource regarding consonant acquisition for nearly 7,000 languages that do not have speech acquisition data.In the current article, the traditional library database search only identified 25% (15/60) of the articles.To identify articles about consonant acquisition, SLPs need to access additional resources.Recently, international research collaborations have increased cross-linguistic access and knowledge about children's speech acquisition, assessment, and intervention.For example, the International Expert Panel on Multilingual Children's Speech (46 researchers who had worked in 43 countries using 27 languages) produced a tutorial to support SLPs to assess children's speech in a language that they do not speak (McLeod et al., 2017).The tutorial outlined a range of additional resources for SLPs including a review of 30 speech assessments in 19 languages (McLeod & Verdon, 2014), the Multilingual Children's Speech website (McLeod, 2016a) that provides resources in over 60 languages, and the crosslinguisticprojectmaterials website (Bernhardt & Stemberger, 2016) that provides assessments and analyses to support SLPs' cross-linguistic service provision.When SLPs are reading articles about consonant acquisition, particularly those that were published before the availability of typesetting of IPA symbols, additional resources including the IPA (International Phonetic Association, 1999, 2015) are required to decode some orthographic symbols within articles.
In the recent tutorial published to support SLPs to assess children's speech in language(s) not spoken by the SLPs (McLeod et al., 2017), elements of comprehensive and respectful assessments are described.One recommendation was to seek, where appropriate, typical speech acquisition data for the language(s) assessed.However, when typical speech acquisition data are unavailable or the data context/dialect is not appropriate, data from the current cross-linguistic review can be used to provide preliminary data to be compared with data from a family-member contrastive analysis (McLeod et al., 2017).Age of acquisition, PCC, and early-middle-late data from the current study can be used to inform SLPs' expectations of children's acquisition of consonants across the world.General principles to be gleaned from the current study are that most of the world's consonant phonemes are acquired (on average) by the time children are 5 years old, and, by this time, over 90% of consonants within words are produced correctly.Generally, the latest groups of consonants to be acquired are consonants that use the anterior tongue, particularly trills, flaps, affricates, and fricatives.This knowledge can be used to inform SLPs' expectations of children's developmental capacity and decision making regarding the need for intervention.A scenario using this knowledge could be that an SLP was asked to assess a 5-year-old child who spoke Hmong, a language that, to date, does not have a published study of consonant acquisition data available in English.Working with an interpreter, a speech sample was obtained that included a number of examples of each consonant phoneme.The child's parent also produced the same words, and the child's productions were compared using a family-member contrastive analysis, generating a list of phonemes that were not produced in an adultlike way, and by calculating the PCC.In the current scenario, the parents were concerned about their child's speech and had indicated that he was sometimes intelligible on the Intelligibility in Context Scale: Hmong (McLeod, Harrison, & McCormack, 2012).The child produced approximately 50% of consonants correct (whereas, in the current study, 5-year-old children had an average PCC of 93).The child had difficulty producing most types of consonants, including plosives and nasals (identified in the current study as being acquired early).Consequently, the child was referred for intervention to work with the parent and interpreter on accuracy of Hmong consonant production to increase his intelligibility in Hmong.

Limitations
Although this study presents an inclusive and diverse cross-linguistic view of children's consonant acquisition, there are a number of limitations that influence the findings.First, the mean age of acquisition as the primary data source can be influenced by the number of studies, the number of languages, the maximum and minimum ages studied, the elicited words, the criteria reported, the individual variability resulting in large standard deviations of scores (especially when there are small sample sizes), and the era of the studies (some were conducted > 50 years ago, so different theoretical and methodological conventions influenced data collection and analyses).The maximum and minimum ages studied influenced the age of acquisition; as mentioned in the results, data from 8.4% of the studies (75%-85% criteria) and 14.8% of the studies (90-100% criteria) were not included in the analysis because the consonants were not acquired by participants who were of the oldest ages in the study.As a result, Appendix A and Supplemental Material S1 provide additional data (standard deviation, range, number of studies, and languages studied) to assist with the interpretation of data.
Most data included in the current article report monolingual consonant acquisition, with a few reporting acquisition of consonants in children's first language or other languages of multilingual children (see Appendix B).Therefore, the summary data contained within this article cannot be extrapolated to all multilingual children's consonant acquisition.Within the world's literature, there are fewer studies of multilingual consonant acquisition than monolingual consonant acquisition, so disambiguating acquisition of consonant phonemes within multilingual children's languages and examining transfer of features from one language to another is challenging (see Hambly, Wren, McLeod, & Roulstone, 2013).However, there are some examples of carefully controlled studies of children's acquisition in more than one language that can provide a model for future studies (e.g., Albrecht, 2017;Fabiano-Smith & Goldstein, 2010a, 2010b).
Finally, most articles were published in English or had summary data available in English, presenting a limitation of the study because there may have been other studies that were not included because they were unable to be located with search strategies by English speakers.Although the current article increases the visibility of some scientific publications in languages other than English (Meneghini & Packer, 2007), it is likely that more articles could have been included in the review if the authors could search using non-Latin script (e.g., Chinese electronic databases were used by Li & To, 2017).In addition, although every attempt was made to collate data from the studies available in languages other than English, some information (e.g., reliability) could not be translated in all cases.

Future Research
During the collation and analysis of data in the current study, the authors developed guidelines to inform future researchers' reporting of data and to facilitate comparisons across studies of children's consonant acquisition (see Appendix C).These guidelines address reporting of demographic data, determining the age range of participants, selection of stimuli, reporting of consonant acquisition data, analysis, and documentation.The psychometric guidelines provided by McCauley and Swisher (1984) for creating norm-referenced assessments could supplement the guidelines in Appendix C.

Conclusion
This cross-linguistic review of 64 studies of 27 languages describes children's acquisition of consonant phonemes.By 5;0 years, most consonants were acquired and children were producing over 90% of consonants correctly.SLPs can draw upon these general principles as milestones of adultlike speech production but also should acknowledge children's individual speech acquisition journeys as a creative process.

Figure 1 .
Figure 1.Mean age of acquisition of pulmonic consonant phonemes across languages (a) from 29 studies using the 75%-85% criteria and (b) from 37 studies using the 90%-100% criteria.Reprinted with permission from McLeod and Crowe (2018).

Figure 3 .
Figure 3. Mean age of acquisition of consonants for English-speaking children (a) from nine studies using the 75%-85% criteria and (b) from eight studies using the 90%-100% criteria.Reprinted with permission from McLeod and Crowe (2018).

Table 1 .
Mean age of acquisition of pulmonic and nonpulmonic consonant phonemes across 27 languages using the 75%-85% and 90%-100% criteria organized according to age in years.

Table 2 .
Average percentage of consonants correct (PCC), percentage of vowels correct (PVC), and percentage of phonemes correct (PPC) across studies and languages.