Introducing WCM-SE: The word complexity measure phonetically justified and adapted to Swedish

ABSTRACT

This paper presents the Word Complexity Measure for Swedish (WCM-SE), an adaptation of the original WCM developed for English by Stoel-Gammon. These measures are used to calculate the phonological complexity of words or vocalizations, based on a number of phonological complexity parameters. Each production receives a complexity score based on how many of the parameters are present in the production.

Using phonological complexity scores to measure expressive phonology is suitable for assessing very young children, children with early phonology and children with phonological deficits. It is useful for both relational and independent analyses and enables comparisons between children and across development.

The original WCM uses eight phonological complexity parameters in three domains: word patterns, syllable structures and sound classes. The parameters selected are phonological characteristics that are acquired late in development among English-speaking children.

In the WCM-SE, complexity parameters in the domain sound classes were modified or added according to Swedish or universal patterns of phonology development. The parameters’ complexity is accounted for in terms of language-general phonetic characteristics.

KEYWORDS:

• Phonological complexity
• phonological development
• speech assessment
• Word Complexity Measure
• Swedish

Introduction

Assessing phonology of child speech production is an issue relevant both for language development research and for clinical diagnostics. A productive phonology measure should be able to assess children’s early vocalizations, as well as severely impaired speech of older children. This is the goal of the Word Complexity Measure (WCM; Stoel-Gammon, 2010). This paper documents an adaptation of the WCM to Swedish (WCM-SE) and justifies the selection of original measures on developmental and phonetic bases.

The WCM is used to calculate a complexity score for child productions, both individual words and longer utterances, depending on the type of analysis. It can be used for both independent and relational analyses and is suitable for both research and clinical assessment. Eight complexity parameters, organized in three domains, determine the total complexity score of a production (Table 1). For example, each fricative or affricate within a word is given one complexity point, as is the occurrence of a word-final consonant. The complexity parameters were chosen based on the developmental trajectory of productive phonology in English-learning children (Stoel-Gammon, 2010). The complexity measures have also been used in the design of formalized tests of productive phonology (Marklund, Lacerda, Persson, & Lohmander, 2018; Stoel-Gammon & Williams, 2013) and to compare different phonological tests to each other (Macrae, 2017).

Table 1. The complexity parameters in the original WCM (Stoel-Gammon, 2010) and in the WCM-SE, sorted by domain. Complexity parameters #6–#8 were modified for the Swedish adaptation, and #9 and #10 were added.

	Complexity parameters
Domains	#	WCM	WCM-SE
Word patterns	1	More than two syllables
	2	Stress on non-initial syllable
Syllable structures	3	Word-final consonant
	4	Consonant cluster
Sound classes	5	Velar consonant
	6	Liquid/syllabic liquid/rhotic vowel	Liquid
	7	Fricative/affricate	Fricative
	8	Voiced fricative/affricate	Voiced fricative
	9	-	Trill
	10	-	Long, front, rounded vowel

The WCM-SE, documented in the present paper, is the first tool for assessment of very early expressive phonology development in Swedish-learning children. It follows the same principle as the original measure, giving points for different complexity parameters adding up to a total score per word or utterance. The complexity parameters are modified to reflect Swedish phonemic classes and are motivated based on the phonological development of Swedish-learning children or patterns found across languages. The parameters are also considered in the light of phonetic-articulatory constraints, laying the groundwork for a language-general perspective on phonological complexity in the assessment of expressive phonology.

The Word Complexity Measure adapted to Swedish (WCM-SE)

In the adaptation of the WCM to Swedish, five complexity parameters from the original WCM were used without modifications (#1–#5 in Table 1), three were altered in the adaptation (#6–#8 in Table 1) and two new parameters were added (#9–#10 in Table 1). Parameters in the WCM-SE were grouped according to the same three domains as in the original WCM: word patterns, syllable structures and sound classes.

The complexity parameters regarding word patterns and syllable structures in the WCM-SE are kept as in the original WCM. The modifications in the adaptation all concern parameters in the domain sound classes. While velars, liquids, fricatives and voiced fricatives are the same in both versions, syllabic liquids, affricates and rhotic vowels were excluded because they are not part of the Standard Swedish phoneme inventory. Since no target productions exist in Swedish for these sound classes, they will not be given points in the WCM-SE. Trills and the long, front, rounded vowels [ʉ:], [y:], [ø:] were added as complexity parameters, based on data on typical acquisition of Swedish consonants and vowels.

Developmental rationale for the complexity parameters

Because research about early acquisition of Swedish phonology is scarce, the complexity parameters in the WCM-SE are to some extent also based on universal patterns of early phonological learning. When no data on acquisition of Swedish phonology are available, the parameter selection for the WCM-SE is based instead on developmental patterns found across languages, provided that the pattern is also ‘common knowledge’ based on clinical experience among Swedish speech–language pathologists. For example, only a few of the complexity parameters in the domains word patterns and syllable structures have been directly studied in Swedish-learning children, but they are all found across several languages.

More than two syllables

In general, multisyllabic words typically appear later in development than monosyllabic and disyllabic words. Children’s first words are typically monosyllabic or disyllabic (Vihman, 2014). Further, children’s early productions (at 16–17 months of age) for the most part target words that in their adult form are monosyllabic or disyllabic; this pattern is reported also for Swedish-learning children (de Boysson-Bardies & Vihman, 1991). Although the syllabic structure of early productions has not been directly investigated in Swedish, the above findings in combination with the fact that multisyllabic words are less frequent than monosyllabic and disyllabic words in genuine¹ Swedish words (Sigurd, 1965), motivates including the complexity parameter of more than two syllables in the WCM-SE without modification. In the WCM-SE, one point is given to any word with more than two syllables.

Stress on non-initial syllables

Deletion of weak syllables in multisyllabic words is a pattern seen across languages among children learning to speak (Johnson & Reimers, 2010). In English and Dutch children, deletion of weak syllables is more common when a non-initial syllable is stressed than when the initial syllable is stressed (Gerken, 1994; Wijnen, Krikhaar, & Den Os, 1994). In Swedish words, trochaic stress pattern is the most common (e.g. de Boysson-Bardies & Vihman, 1991; Lundeborg-Hammarström, 2017) and words with late stress are relatively uncommon and are generally of foreign origin (Sigurd, 1965). Based on the cross-linguistic pattern of deletion of weakly stressed syllables, the common stress patterns of Swedish, and the fact that infants already have phonotactic knowledge of stress patterns in the native language at 9 months (Jusczyk, Cutler, & Redanz, 1993), stress on non-initial syllables is included in WCM-SE without modification. WCM-SE gives one point to words with non-initial syllable stress.

Word-final consonant

In general, word-final deletion of consonants is more frequent than word-medial or word-initial deletion in child productions (Johnson & Reimers, 2010). Word-final deletion among English-learning children has usually disappeared by age 3 (Grunwell, 1981). For Swedish children, frequent omissions of word-final consonants have been reported for 3-year-olds, but not for 5-year-olds (Lohmander, Borell, Henningsson, Havstam, Lundeborg, & Persson, 2005). We, therefore, chose to include this complexity parameter without modification. Words with a final consonant are given one point.

Consonant cluster

In many languages, words with consonant clusters appear later in development compared to those without consonant clusters (e.g. McLeod, van Doorn, & Reed, 2001; Stemberger & Bernhardt, 2018). Words with clusters in different positions are highly frequent in Swedish (Linell & Jennische, 1980); however, the number of legal consonant combinations is high (Lundeborg-Hammarström, 2017) effectively leading to any particular combination not being very frequent. Swedish 3-year-olds can typically produce some types of clusters, but still have difficulty with others (Blumenthal & Jacobsson, 2013); 45% of Swedish-learning children make reductions in consonant clusters (Lohmander et al., 2005). At 5 years, no regular reduction pattern was found in consonant clusters (Lohmander et al., 2005). As Swedish children also produce consonant clusters later in development and thereby follow the general development of consonant cluster production, this complexity parameter is included in the WCM-SE without modification. Words receive one point per consonant cluster, regardless of how many consonants there are in the clusters.

Velar consonant

Front consonants are generally acquired shortly before back consonants, and velar fronting is a phenomenon found across languages in children’s productions (Johnson & Reimers, 2010). Stoel-Gammon proposes a developmental path for velars, according to which velars in all positions are substituted by coronals or dentals in the first stage of phonological development (1996). The Swedish consonant system consists of 18 consonants, and the velars are [k], [g], [ŋ] and in certain dialects [ɧ]. The adult target forms of Swedish children’s first words contain velars less frequently than they contain dentals (de Boysson-Bardies & Vihman, 1991). At 3 years of age, 100% of Swedish-learning children (n = 102) used bilabial and dental stops, whereas only 90% used velar stops (Lohmander, Lundeborg, & Persson, 2017). At the same age, [ɡ] and [ŋ] were among the small group of consonants that were not yet produced (Blumenthal & Lundeborg-Hammarström, 2014). Data from Swedish-learning children on velar acquisition are therefore in line with the proposed developmental trajectory, and this complexity parameter is included without modification in the WCM-SE. Each occurrence of a velar in a word is given one point.

Liquid

There are similarities in the acquisition of liquids across languages. In the first 25 words in Swedish, French, Japanese and American children’s speech, stops, fricatives and nasals were more frequent than liquids (de Boysson-Bardies & Vihman, 1991). The Swedish liquids are the lateral approximant [l], the retroflex lateral approximant [ɭ] and the alveolar approximant [ɹ]. The results on the development of liquids in Swedish-learning children are contradictory. According to Linell and Jennische, [l] can appear ‘early’ in Swedish children’s language development, unless substituted by [j] (1980), whereas Nettelbladt, on the other hand, reports that [l] is not established until about 4 to 6 years of age (2007). The grade of difficulty also varies between different liquids in Swedish, with the alveolar approximant [ɹ] being the most difficult (Linell & Jennische, 1980). Experienced speech-language pathologists often report late acquired liquids. Based on similar development of liquids across languages, some support from the literature for the late acquisition of liquids, and undocumented clinical experience, liquids are included as a complexity parameter in the WCM-SE. However, the syllabic liquids and rhotic vowels that are part of the parameter in the original WCM are excluded, as they are not included in the Swedish phoneme inventory. All liquids in a word are given one point per occurrence.

Fricative and voiced fricative

Fricatives have been studied in Swedish-learning children to a greater extent than many other speech sounds. Less than 50% of Swedish-learning 3-year-olds produce [s] and [ɕ], and 90% produce [f]. This is in contrast to dental stops and nasals, which all children typically produce at age 3 (Lohmander et al., 2017). Four out of the nine consonants that are not yet acquired at age 3 are the fricatives [v], [s], [ɕ] and [ɧ] (Blumenthal & Lundeborg-Hammarström, 2014). Two of the most difficult speech sounds to produce for typically developed 3–4-year-old Swedish-learning children are [ɧ] and [ɕ] (Henriksson & Lawrence, 2014), and the fricatives [v], [s], [ɕ], [ɧ] are still not established in 4–5-year-olds (Blumenthal & Jacobsson, 2013). Based on the developmental trajectory of fricatives in Swedish-learning children, they are included as a complexity parameter in the WCM-SE. We chose to follow the original WCM in treating fricatives and voiced fricatives as two separate parameters since there is a phonetic basis for this type of additive scoring in this particular case (see the following section on the phonetic rationale). For example, the production [vʊvə] (‘vovve’ doggy) is given four points in total: one per occurrence of a fricative in general, and one extra per occurrence of a voiced fricative. However, affricates and voiced affricates, although included in the parameters in the original WCM, are not included in the WCM-SE, as they are not part of the Swedish phoneme inventory.

Trill

Trills are acquired late in several languages (e.g. Stemberger & Bernhardt, 2018). Less than half of the Swedish-learning 3-year-olds use the trill [r] (Lohmander et al., 2017). It is also among the nine consonants not acquired among Swedish 3-year-olds (Blumenthal & Lundeborg-Hammarström, 2014). Trills are often some of the last phonemes acquired in Swedish (Lundeborg-Hammarström, 2017), and it is not unusual for Swedish-learning children with typical language development not to have established [r] at age 5 (Lundeborg-Hammarström, 2017; Nettelbladt, 2007). Since trills are among the speech sounds that are acquired last, they are added as a complexity parameter in the WCM-SE, and words containing a trill are awarded three complexity points per trill. For a discussion on why three points are awarded rather than one, see the Discussion section.

Long, front, rounded vowel

Swedish vowels exist in pairs of long and short versions, and the length covaries with the quality of the vowels, with short variants being more centred and lax. Although most vowels are present in Swedish-learning children at age 3 (Blumenthal & Jakobsson, 2013), the Swedish front, rounded vowels are known to be acquired late (Bjar, 2003; Linell & Jennische, 1980). The long, front, rounded vowels are therefore added as a complexity parameter in the WCM-SE. Each long, front, rounded vowel in a word is given one point per occurrence.

Phonetic rationale for the complexity parameters

The selection of phonological complexity parameters in both the original WCM and the WCM-SE is based on developmental trajectories, but in addition, a phonetic rationale for the WCM-SE is given. For the most part, the complexity parameters’ acoustic target forms require specific aerodynamic conditions that demand certain levels of articulatory control and coordination to be performed. For some parameters, we also describe perceptual and typological aspects and how they contribute to the complexity level.

More than two syllables

The production of multisyllabic words receives complexity points. Compared to canonical babbling, multisyllabic words are more variable in segments and stress patterns and therefore considered more complex. Variegated babbling, emerging at the end of the child’s first year, is phonetically more complex than canonical babbling because of greater variation in phoneme sequences. However, variegated babbling productions or even proto-words are phonetically not as stable as early words. Words, on the other hand, are relatively stable forms both regarding form and meaning. To produce a variegated string of sounds requires a high level of articulator coordination in the jaw, tongue, lips and velum.

Stress on non-initial syllables

A stressed syllable is typically louder, longer and has a higher fundamental frequency than an unstressed syllable. Placing stress on any syllable other than the first requires a fine and detailed control of sub-glottal pressure and vocal fold tension. In mature speech production, speakers control the sub-glottal pressure to keep it approximately constant in spite of all the aerodynamic changes involved in producing an utterance. It is more challenging for children to keep constant sub-glottal pressure and to control vocal fold tension, as long as their motoric abilities are immature. This means that for young children, the result is often both higher sub-glottal pressure and higher vocal pitch at the beginning of an utterance.

Word-final consonants

Word-final consonants are perceptually less prominent than word-initial or word-medial consonants, as the preceding vowel can mask them. For example, Dutch-learning infants are sensitive to final consonant omissions at 18 months of age, but not at 14 months of age (Levelt, 2012). Furthermore, at 24 months, English-learning children are less sensitive to final consonant mispronunciations than mispronunciations in other positions (Wang & Seidl, 2016). Word-final consonants are not as salient which results in them being less likely to be perceived, and therefore less likely to be included in early child productions.

Consonant cluster

Consonant clusters require fast, accurate and synchronized movements of the articulators to alter position quickly while maintaining closure or stationary noise generation (stops and fricatives, respectively). These rapid articulatory changes need to be accomplished without opening the vocal tract to a vowel configuration between the consonantal segments, which increases the level of difficulty and requires mature articulatory skill.

Velar consonant

Velars are visually less salient than more frontal consonants. Following the same line of reasoning as for word-final consonants, they are less likely to be produced simply because they are perceptually less salient. Especially during early speech development, velars tend to be drowned in a typical dento-alveolar dominance (Locke, 1983), presumably due to teething, potentially adding to an advantage for more fronted speech sounds.

Liquid

The liquids relevant for Swedish are the lateral approximant [l], the retroflex lateral approximant [ɭ] and the alveolar approximant [ɹ]. Liquids require fine-grained muscle control of the tongue apex or body to achieve partial constrictions of the vocal tract, and the ability to separate tongue movements from opening and closing of the mouth through moving the jaw.

Fricative and voiced fricative

In order to produce the stable noise source of a fricative, a critical balance is needed between the pressure drop over the constriction and its particular configuration appearance, in particular, its cross-sectional area and shape. This requires a high level of control over both the respiratory system and the articulators. Failure to maintain this critical balance will result in a stop, a semivowel or a flap. Production of a voiced fricative naturally requires the articulatory feats described above to achieve the stable noise source. On top of that, it also involves maintaining adequate trans-glottal pressure drop over the glottis to sustain vocal fold vibration throughout fricative production.

Trill

In a number of Swedish dialects, /r/ is realized as a trill, either the alveolar [r] or the uvular [ʀ]. Each movement cycle or pulse in a trill is caused by a stable aerodynamic oscillation. This requires a critical balance between the correct position of the tongue, relatively lax muscular tension in the apex or tongue body, and airflow strength. However, /r/ in Swedish can also be realized as voiced fricative or approximant, so in relational analysis using WCM-SE, the intended target production needs to be specified.

Long, front, rounded vowel

Typologically, rounding is the default for back vowels, but not for front vowels (Liljencrants & Lindblom, 1972; Maddieson, 1984), which suggests a general interaction of rounding and the front-back dimension. It is, therefore, reasonable to consider rounding of front vowels as more phonetically complex.

How to calculate phonological complexity score with WCM-SE

The complexity parameters of the WCM-SE and their points are summarized in Table 2. Examples of calculations for single word productions are presented in Table 3 (transcriptions of child productions from Renner, 2017). For various ways of applying the WCM-SE to larger speech samples in order to get a comprehensive picture of a child’s phonological development or proficiency, see Stoel-Gammon’s paper on the original WCM (2010).

Table 2. Summary of the complexity parameters in WCM-SE and the points they give for a production.

Complexity parameter		Points
1	More than two syllables	1 per word
2	Stress on non-initial syllable	1 per word
3	Word-final consonant	1 per word
4	Consonant cluster	1 per occurrence
5	Velar consonant	1 per occurrence
6	Liquid	1 per occurrence
7	Fricative	1 per occurrence
8	Voiced fricative	1 per occurrence
9	Trill	3 per occurrence
10	Long, front, rounded vowel	1 per occurrence

Table 3. Examples of how to calculate the WCM-SE score for individual words produced by children between 1 and 2 years of age. The complexity parameters are numbered as in Tables 1 and 2. Transcriptions of child productions from Renner (2017, pp. 69–70, 125, 131).

		Complexity parameter
Target word	Child production	1	2	3	4	5	6	7	8	9	10	Σ
[krʊkʊˈdiːl] ‘crocodile’ krokodil	[dɔdɔˈdiː]	1	1									2
[ˈlampa] ‘lamp’ lampa	[ˈpa]											0
[ˈlampa] ‘lamp’ lampa	[ˈblampa]				2		1					3
[ˈmøsːa] ‘cap’ mössa	[ˈmøːɕːa]								1		1	2
[ˈmøsːa] ‘cap’ mössa	[ˈmœɕan]			1				1				2
[ˈstrʉmpa] ‘sock’ strumpa	[ɡəˈɡa]		1			2						3

Validation

To validate the WCM-SE, data from typically developing children should be used. Comparisons could be made to other measures, such as the Percentage Consonants Correct (Shriberg, Austin, Lewis, McSweeny, & Wilson, 1997; Shriberg & Kwiatkowski, 1982), the Percentage Phonemes Correct (Dollaghan & Campbell, 1998) or the Phonological Mean Length of Utterance (Ingram, 2002).

Discussion

The original WCM is a valuable tool to assess productive phonology on segmental and supra-segmental levels, combining the potential of relational and independent analyses. An adaptation to other languages, such as the WCM-SE, is therefore inherently motivated. Adapting the measure to Swedish, in particular, is also motivated by the fact that there are currently no materials available to assess Swedish expressive phonology that is not word based.

An adaptation of a measure developed for English phonology to Swedish needs to take into account some critical differences between the two languages’ phonology regarding sound class parameters. Syllabic liquids, rhotic vowels and affricates (both voiced and unvoiced) are excluded from the WCM-SE, as they are not part of the Swedish phoneme inventory. Instead, trills ([r] and [R]) and long, front, rounded vowels ([ʉ], [y] and [ø]) are included as parameters.

In regard to the phoneme /r/, Swedish poses some challenges to the WCM-SE in terms of relational analysis. As /r/ can be realized in several ways, according to regional dialects, for example, this can result in different scores for a word depending on whether it is produced with a trill ([r]/[ʀ]), a voiced fricative ([ʐ]), a liquid ([ɹ]) or a tap/flap ([ɾ]). In order to compare complexity scores to target word forms in a relational analysis, we, therefore, suggest considering the regional dialect version of the target form (Table 4). Notably, trills are the only segments that are awarded more than one point per occurrence. The reason for this is that the other common realizations of /r/ are already awarded points on a graded scale, according to the scores in both WCM and WCM-SE. Taps/flaps are awarded zero points, liquids are awarded one point (for being liquids) and voiced fricatives result in two points (one point for being a fricative and one point for being a voiced fricative). Since the gradation is already inherent in the measure, it seems prudent to award points of a trill along the same scale. From an articulatory standpoint, a trill must be considered more complex than a liquid or a voiced fricative, thus motivating a higher score than any of those. In Swedish-learning children, trills are commonly substituted with an approximant or a voiced fricative supporting the notion that trills are the most complex /r/ realization, and motivating the score of three points.

Table 4. Illustration of different scores for target word containing /r/, depending on its realization.

Target word	Realization	Complexity parameter										Σ
		1	2	3	4	5	6	7	8	9	10
/kruːk/‘hook’ krok	[ˈkruːk]				1	2				3		6
	[ˈkʀuːk]				1	2				3		6
	[ˈkʐuːk]				1	2		1	1			5
	[ˈkɹuːk]				1	2	1					4
	[ˈkɾuːk]				1	2						3

Since the WCM-SE has added complexity parameters concerning vowels, there are some issues to discuss. In general, vowels are considered less difficult to produce compared to consonants, and vowel deviations or substitutions typically have less effect on intelligibility compared to consonant deviations and substitutions. However, among Swedish 3-year-olds (n = 30), 50% were found to have deviances in vowel production (Henriksson & Lawrence, 2014). Swedish is a language with a large vowel inventory, and the vowels uncommon in other languages but typical for Swedish, [ʉ:], [y:] and [ø:], are acquired late in development (Linell & Jennische, 1980). The size of the vowel inventory of the language may have an impact on the perceived accuracy of early vowel productions. In languages with few vowels, greater variation in vowel productions may be tolerated whereas in languages with large inventories more ‘fine calibration’ is required for a production to be classified as the target production. Additionally, as vowel production typically also shows a large variability in mature speech, vowels can be considered inherently more difficult to quantify, and therefore also to assess, than consonants.

The WCM is a useful tool that can easily be adapted to measure phonology in different ways (Stoel-Gammon, 2010). The WCM-SE extends the original WCM by providing phonetic foundations for parameter selection. However, establishing these bases for the WCM-SE has made it apparent that there is more to consider when assessing the phonological complexity of speech production. Although the current version of the WCM-SE is a good tool for the assessment of phonological complexity, there are obvious issues concerning the nature of the complexity scales used in the WCM-SE. For example, the articulatory demands of a specific segment may vary with context (e.g. stressed vs. unstressed syllables, word-final or word-initial position, adjacent speech sounds, etc.). It could also be argued that consonant clusters containing more segments should be scored higher than clusters containing fewer segments or that multisyllabic words should be given one point per extra syllable instead of just one point for the word being multisyllabic. The measure is additive in its nature, assuming equality of complexity scores across quite disparate areas, despite obvious variations in complexity between characteristics. This is an issue to be addressed in future updates of the measure. To be able to make comparisons across languages while at the same time taking language-specific phonology into account, a hierarchical model of phonological complexity would be useful, taking these nuances into account (Lacerda & Marklund, in preparation). Anchoring such a model firmly in phonetic reasoning is likely to make it applicable to different languages, and very useful for cross-linguistic comparison, since universal phonetics are the theoretical benchmark for discussions regarding the origins of phonological development, whereas developmental data show us the outcome. The relationship between phonetics and developmental phonology needs to be further evaluated in general.

Conclusion

The WCM-SE is an adaptation to Swedish of the original English WCM, based on complexity parameters that characterize the acquisition of expressive phonology, selected on the basis of acquisition age and motivated by articulatory, aerodynamic and typological aspects. The present adaptation has made apparent both a need for more research on the phonological development of Swedish-learning children and the possibility of a more general approach to measuring phonological complexity based on language-independent phonetic principles. The WCM-SE meets the need for a simple but valuable tool to assess early phonological development of Swedish-learning children.

Declaration of interest

The authors report no conflict of interest.

Notes

¹ Multisyllabic roots are generally of foreign origin; however, multisyllabic Swedish compound words are common.