ABSTRACT. Some children with speech sound disorders (SSD) have difficulty with literacyrelated

Transcription

1 ABSTRACT Some children with speech sound disorders (SSD) have difficulty with literacyrelated skills. In particular, they often have trouble with phonological processing, which is a robust predictor of early literacy. This study investigates the phonological processing abilities of preschoolers with SSD and uses a regression model to evaluate the degree to which these abilities can be concurrently predicted by types of speech sound errors. Forty-three English-speaking preschoolers (ages four to five) with SSD of unknown origin participated in an assessment of phonological processing skills and speech sound production. Productions elicited on a 125-item picture naming task were phonetically transcribed, and errors were coded in two ways: (1) according to Percent Consonants Correct (PCC), which weights all consonant errors equally, and (2) according to a three-category system: typical sound changes, atypical sound changes, and distortions. Phonological awareness (PA) was assessed via rhyme matching, onset (initial consonant) matching, onset segmentation and matching, and blending. Phonological memory was assessed using a syllable repetition task. Children also rapidly named pictures of monosyllabic and disyllabic words. Results showed that performance on a PA composite score could be predicted, in part, by vocabulary and age (about 33%). Atypical sound changes were found to account for additional variance in PA (another 6%), but distortions and typical errors did not account for significant variance in PA. Thus, use of more atypical sound changes was associated with poorer performance on PA tasks. When the same consonant errors were classified using PCC, speech sound errors were not found to predict significant variance in PA. Atypical sound changes also significantly predicted variance in phonological

2 memory (about 31%) and rapid naming (about 10%) tasks beyond what had already been predicted by vocabulary and age. The results support the notion that poorer performance on phonological processing tasks is associated with lower receptive vocabularies and production of more atypical speech sound changes. Results are interpreted in the context of the accuracy of phonological representations. Thus, atypical sound changes are seen as reflecting poorly specified internal representations of the sound features of words.

3 Phonological Processing and Speech Production in Preschoolers with Speech Sound Disorders By Jonathan Preston B.S. Elmira College M.S. Syracuse University DISSERTATION Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Speech-Language Pathology Department of Communication Sciences and Disorders Syracuse University August, 2008 Approved: Professor Mary Louise Edwards Date:

4 Copyright 2008 Jonathan Preston All rights reserved

5 ACKNOWLEDGEMENTS Thanks to the families that participated in this research, to the clinicians who referred children, and to my colleagues and friends in the field who offered encouragement and intellectual support. Thanks in particular to my advisor, Dr. Mary Louise Edwards, for her support. I am appreciative of comments and feedback from my committee members, Dr. Raymond Colton, Dr. Linda Milosky, Dr. Benita Blachman, and Dr. Annette Jenner-Matthews. I also would like to thank Renail Richards for assisting with reliability, and Dr. Lawrence Shriberg for providing the Power Point stimuli for the syllable repetition task. In addition, Dr. Beth Prieve s flexibility was important in making this project happen. This research was supported in part by the 2007 American Speech-Language- Hearing Foundation grant in Early Child Language awarded to the author.

6 TABLE OF CONTENTS CHAPTERS: I : INTRODUCTION... 1 II : METHODS III : RESULTS IV : DISCUSSION REFERENCES FIGURES Figure 1: Theoretical framework for the study... 5 Figure 2: Flow chart of procedures with number of participants Figure 3: Examples of PA stimuli Figure 4: Scatterplots of speech sound production error types and phonological awareness Principal Component Figure 5: Observed PA Principal Component scores and PA scores predicted by the regression (age, vocabulary, atypical sound changes) for the 43 children with SSD80 v

7 TABLES Table 1: Summary of speech sound error types and their suspected reflection of underlying phonological representations Table 2: Inclusionary criteria for the study Table 3: Descriptive statistics for the 43 preschoolers who participated in Part II and were used in the final analysis Table 4: Summary of speech sound (in)accuracy for 43 preschoolers with SSD Table 5: Pearson s correlation coefficients (r) of speech sound error types Table 6: Summary of the performance of 43 children on the phonological processing tasks Table 7: Pearson correlation coefficients (r) for the phonological awareness tasks for 43 children with speech sound disorders Table 8: Principal Component Analysis summary derived from the four Phonological Awareness tasks Table 9: Hierarchical regression used to predict PA Principal Component Table 10: Regression using PCC as the speech production variable to predict PA Table 11: Regression explaining variance in Phonological Memory (Syllable Repetition Task) Table 12: Regression explaining variance in Rapid Naming (average Z scores of two Rapid Naming tasks) vi

8 APPENDICES Appendix A: Transcription Rules and Coding Sound Changes Appendix B: Errors with Interacting Sound Changes: Which is Preferred? Appendix C: Words Used on the Picture Naming Task Appendix D: Phonological Awareness Tasks Appendix E: Syllable Repetition Task (from Shriberg et al, 2006) Appendix F: Rapid Naming Task Appendix G: Complete Correlation Matrix Appendix H: Measurement Issues Appendix I: Regression Diagnostics Appendix J: Caveats and Limitations: The Role of Children s Experiences Appendix K: Speech Perception vii

9 1 I : INTRODUCTION Literacy problems are a significant international concern, with as much as 15-20% of the world s population having some sort of reading difficulty (International Dyslexia Association, 2000). Early identification of such problems is essential so that early intervention can take place. Fortunately, it is now possible to identify skills in preschool that are good predictors of later literacy. This study will focus on preschoolers with speech sound disorders (SSD), who are known to be at risk for preliteracy and literacy problems (particularly phonological processing). Exactly how SSDs are related to preliteracy deficits is unclear. Therefore, to aid in the identification of early preliteracy problems, this study will explore the relationship between the specific types of speech sound errors produced by preschoolers with SSD and phonological processing skills, known to predict early literacy. Phonological processing, which is the ability to process speech sound information, is related to both speech production and literacy development (e.g., Stackhouse & Wells, 1997). Because phonological processing skills do not necessarily rely on alphabet knowledge, it is possible to assess these skills in preschool children (prior to formal literacy instruction). Phonological processing has been discussed as including three domains: phonological awareness (PA), phonological memory (PM), and phonological retrieval (as assessed by rapid naming, RN) (e.g., Wagner & Torgesen, 1987). Children with SSD have been reported to have weaknesses in each of these domains (e.g., Leitao et al., 1997). The degree to which variability in speech sound production is related to variability in each of these three components of phonological processing has not been thoroughly explored. This study addresses that issue.

10 2 The term speech sound disorder (SSD) will be used to refer to children who have clinically significant difficulties producing or using the speech sounds of their native language for their age and dialect groups (cf. NIDCD, 2006). Other reports have referred to these children as having articulation or phonological disorders and/or delays (e.g., Dodd, 1995; e.g., Gibbon, 1999). The current investigation will limit the definition to include children whose primary deficits are in speech communication, and who have no known oral structural problems (e.g., cleft palate) or developmental disorders (e.g., cerebral palsy). Approximately 8-9% of young children are diagnosed with a SSD (NIDCD, 2006); thus, the problem affects millions of children. There is emerging evidence that children who begin kindergarten with a SSD and poor phonological awareness are at particular risk for later literacy problems (e.g., Nathan et al., 2004); thus, early identification of these problems is crucial. The specific relationship between speech sound production patterns and phonological processing in children with SSD, however, remains unclear. Previous investigations have used measures of speech sound production (e.g., Percent Consonants Correct) that may not be sensitive to the nature of the errors a child makes. Therefore, the current study will examine the relationship between types of speech sound errors, quantified in a more precise manner than in previous investigations, and each domain of phonological processing in preschoolers with SSD. The primary focus will be the relationship between speech sound production and phonological awareness, with exploratory analyses examining the concurrent relationship between speech production and the other two domains of phonological processing, phonological memory and phonological retrieval/rapid naming.

11 3 The goals of this research are (1) to confirm previous assertions (which largely lack empirical support) regarding the strength of the relationship between various types of speech sound errors and measures of phonological processing in children with SSD; (2) to improve our understanding of how specific types of speech sound changes account for unique variance in phonological processing. The clinical contributions of the study include identifying speech production characteristics that may be indicative of risk for early literacy problems.

12 4 Review of the Literature The concept of phonological representations will be reviewed first, as phonological representations have been discussed as an underlying contributor to performance on phonological processing tasks as well as speech sound production. The literature concerning the relationship between phonological awareness (PA) and literacy development will be reviewed briefly to highlight the importance of being able to identify potential indicators of phonological processing difficulty. Also, the known connection between SSD and PA will be outlined, and limitations in our current understanding of this relationship will be addressed. The quantification of speech sound errors will be discussed, along with the justification for a more specific measurement system that could advance our understanding of the relationship between speech sound errors and phonological processing. Finally, literature related to two other domains of phonological processing, phonological memory and phonological retrieval/rapid naming, will be reviewed; the relationship between SSD and these two domains will be investigated by exploratory analyses. Figure 1 (similar to a model by Rvachew & Grawburg, 2006) was adapted for the current study to explicate the relationship between speech sound accuracy and phonological processing, and to show the presumed relationship of each to phonological representations. The literature review will use this figure as a guide in discussing the relationships among the concepts of interest.

13 5 Figure 1: Theoretical framework for the study PHONOLOGICAL REPRESENTATIONS Age Vocabulary SPEECH SOUND ACCURACY Atypical Sound Changes Typical Sound Changes Distortions PHONOLOGICAL PROCESSING Phonological Awareness Phonological Memory Phonological Retrieval Notes: The link between speech sound accuracy and phonological processing (heavy dotted line) remains unclear and will be examined here. The link between phonological processing and literacy is well established (not shown here). The variables in jagged boxes (age and vocabulary) are control variables that have been discussed as being associated with the accuracy of phonological representations. The shading of Speech Sound Accuracy indicates that there may be varying degrees of (in)accuracy of speech sound production. No significant relationship is generally reported between receptive vocabulary and speech production in preschoolers (Bishop & Adams, 1990; Rvachew & Grawburg, 2006); hence Figure 1 does not include a link between vocabulary and speech sound accuracy. Concurrent relationships will be explored in this study, not causality.

14 6 Phonological Representations Phonological representations are stored (internal) representations in the mental lexicon that contain the phonological (speech-sound related) features of words (Edwards, 1995; Pascoe et al., 2006; Rvachew, 2006; Stackhouse & Wells, 1997). These representations may include the constituent phonemes and phoneme combinations of words, and possibly the associated phonetic specifications of the segments, such as acoustic or motoric features (e.g., Shuster, 1998). Because these representations are internal, they cannot be directly measured. Therefore, researchers rely on measurable behaviors to make inferences about phonological representations. While some theorists hold that there are input representations and output representations (see Edwards, 1995 for a review), empirical data provide support for a strong relationship between the two (Foy & Mann, 2001; Shuster, 1998; Sutherland & Gillon, 2005). As most current models rely on a single underlying representation (Baddeley, 2003; Rvachew & Grawburg, 2006; Stackhouse & Wells, 1997), this is the view assumed in the current study. As in other studies (Elbro et al., 1998; Rvachew & Grawburg, 2006; Rvachew et al., 2003; Sénéchal et al., 2004; Strange & Broen, 1981), the current investigation will use speech sound production as one way of inferring the accuracy of phonological representations (see below). It is generally assumed that, as children get older, phonological representations develop and improve (i.e., become more adult-like) (Nathan et al., 2004; Sutherland & Gillon, 2005). Therefore, age must be taken into account when considering a child s phonological representations. However, not all children develop more accurate phonological representations at the same rate or with the same precision. Thus, some

15 7 children may have more accurate or stronger phonological representations than others (Rvachew & Grawburg, 2006; Rvachew et al., 2003; Snowling, 2000; Stackhouse, 1997; Swan & Goswami, 1997a). Weaknesses in the accuracy (or strength ) of phonological representations have been discussed as a basis for both impaired speech and poor phonological processing (and, by extension, poor literacy skills) (Elbro et al., 1998; Larivee & Catts, 1999; Rvachew, 2007; Sutherland & Gillon, 2005; Swan & Goswami, 1997a). For example, Senechal, Ouellette and Young (2004) suggest that "the quality of phonemic [phonological] representations may be reflected in children's expressive phonology or articulation" (p. 243). Similarly, Swan and Goswami (1997a) make the claim that weak phonological representations contribute to the poor phonological awareness skills of children with literacy problems. If a relationship is found between speech sound accuracy and phonological processing, this would provide support for the notion that phonological representations are an underlying factor in both speech sound production and phonological processing skills. The current study continues a line of research investigating the phonological deficit hypothesis, in which phonological processing is causally related to literacy skills (Snowling, 2000; Wagner & Torgesen, 1987). Accurate or precise phonological representations are considered to be important for the development of phonological processing skills (Fowler, 1991; Snowling, 2000). In fact, weaknesses in phonological representations have been discussed as the causal factor in poor performance on phonological processing tasks by children with preliteracy and literacy problems (Swan & Goswami, 1997a, 1997b). It has been presumed that children with inaccurate

16 8 phonological representations will have difficulty with tasks that require them to utilize those representations, such as comparing initial phonemes in words, or comparing rhymes. Phonological Representations, Phonological Awareness, and Literacy Phonological awareness (PA) refers to awareness of spoken units of speech, such as syllables and rhyming words (see explanations below). It includes phonemic awareness, which is the awareness of individual sounds (Report of the National Reading Panel, 2000). Converging evidence suggests that PA skills are related to spelling, reading decoding, reading comprehension, and reading fluency, both concurrently and longitudinally (Bradley & Bryant, 1983; Catts et al., 2001; National-Reading-Panel, 2000; Phillips & Torgesen, 2006; Snow et al., 1998; Wagner & Torgesen, 1987). Additionally, PA is a primary area of focus in this study because there is evidence that explicit instruction in PA can have positive benefits for literacy development in children both with and without SSD (Ball & Blachman, 1991; Bradley & Bryant, 1983; Gillon, 2000, 2005; Tangel & Blachman, 1992). Syllables are units of speech that must include a nucleus (typically a vowel), with optional consonants preceding the nucleus (the onset ) and/or following the nucleus (the coda ). Developmentally, awareness of syllables precedes awareness of rhyme (vowel plus coda), which precedes awareness of phonemes (individual consonants or vowels) (Liberman et al., 1974; Stackhouse, 1997). Hence, children become aware of smaller and smaller units of speech. While phonological awareness in preschoolers may help to predict later literacy, it is awareness of speech at the phoneme level (phonemic

17 9 awareness) that is most critical in learning to read and spell (Bradley & Bryant, 1983). This is because, in an alphabetic system such as English and many other written languages, letters represent phonemes, not rhymes or syllables. Preschoolers, who are the focus of this study, are often at the stage of learning to (a) identify and produce rhymes (a vowel plus coda, e.g, the at in hat), (b) identify and produce initial phonemes (e.g., the h in hat), and (c) blend units spoken separately to form words (e.g., blend h and at to form hat) (Bird et al., 1995; Catts, 1991; Gillon, 2000; Rvachew et al., 2003; Stackhouse, 1997; Stackhouse & Wells, 1997). Many studies have examined variables that relate to phonological awareness. These include receptive vocabulary (McDowell et al., 2007; Rvachew & Grawburg, 2006), expressive vocabulary (Elbro et al., 1998), letter naming (Elbro et al., 1998), socioeconomic status (McDowell et al., 2007; Nittrouer & Burton, 2005), and speech perception (Rvachew & Grawburg, 2006), etc. The present study will control for age and receptive vocabulary, the variables that have been most commonly discussed as relating to the development of phonological representations. The Role of Age and Vocabulary in PA Development As depicted in Figure 1, phonological processing skills (including PA) are related to age and vocabulary. As children get older and their vocabulary skills increase, they have a larger internal dataset from which to make inferences about phonological features of words. Therefore, phonological representations are thought to become more accurate (or precise) as vocabulary skills develop and as children get older (Metsala, 1999; Walley et al., 2003). It is also believed that as children become more attuned to

18 10 smaller phonological features of words, performance on PA tasks improves (Fowler, 1991; Liberman et al., 1974; Snowling, 2000). Age and PA. Longitudinal studies have reported growth in PA skills and literacy as children age (e.g., Caravolas et al., 2001; Nathan et al., 2004). In a cross-sectional study, Chafouleas et al. (1997) reported that age can account for as much as 60% of the variance in PA from kindergarten to second grade, providing evidence for rapid developmental growth in PA skills. This growth in PA and (pre-)literacy at a young age is often discussed as a function of more mature phonological representations (Fowler, 1991; Nathan et al., 2004; Swan & Goswami, 1997a). Thus, age is one important factor to consider when assessing PA. Vocabulary and PA. There is also a strong relationship between PA/literacy development and language skills in young children. For example, language impairment negatively impacts literacy development (Aram et al., 1984; Bishop & Adams, 1990; Bishop & Clarkson, 2003; Catts, 1993, 1997; Catts et al., 1994; Kamhi & Catts, 1986; Kamhi et al., 1988; Nathan et al., 2004), and for children with and without speech and language impairments, vocabulary has proven to be the most robust language measure when predicting PA. That is, vocabulary is reported to account for approximately 25-30% of the variance in PA in preschool and young school-age children (Bishop & Adams, 1990; Elbro et al., 1998; Rvachew, 2006; Rvachew & Grawburg, 2006; Rvachew et al., 2004). In the present study, the primary interest is in receptive vocabulary, in part because speech sound impairments influence the ability to reliably interpret a child s spoken vocabulary. There is empirical evidence that vocabulary and PA skills are positively

19 11 correlated (Rvachew, 2006; Rvachew & Grawburg, 2006; Swanson et al., 2003). For example, Metsala (1999) found that larger receptive vocabularies in children, as measured by the Peabody Picture Vocabulary Test-Revised (Dunn & Dunn, 1981), were correlated with better performance on phonological processing tasks (including blending, initial phoneme isolation, and rhyming). Vocabulary and PA were related even when the influence of age was controlled. She attributes this phenomenon to underlying representations that are more adult-like in their features and their organization. That is, children who know more words are thought to have more accurately defined phonological representations, because they must keep words separate from similarsounding words. Phonological representations are also related to speech sound production (Edwards et al., 2004; Hodson & Edwards, 1997; Shuster, 1998; Stackhouse & Wells, 1997). Therefore, there is reason to believe that phonological processing skills also relate to SSD. One important question is whether speech production can predict variance in PA, and whether it can predict variance in PA above and beyond the known contribution of receptive vocabulary and age. Suspected Causes of Poor Phonological Representations There is much speculation as to why phonological representations would be weak in some children, including many children with SSD and literacy difficulties. One possibility is that genetic factors, or a combination of genetic and environmental factors, play a role in speech and literacy difficulties (Lewis et al., 2002; Lewis et al., 2006; Raitano et al., 2004; Shriberg et al., 2005). Also, speech perception skills (including

20 12 temporal order judgment, phoneme discrimination, phoneme boundary identification, and amplitude envelope rise time) have been found to relate to phonological processing and literacy in several studies of children with different levels of reading skill (Lieberman et al., 1985; Mody et al., 1997; Richardson et al., 2004; Savage et al., 2005; Sénéchal et al., 2004; Watson & Miller, 1993) and children with SSD (Bridgeman & Snowling, 1988; Jamieson & Rvachew, 1992; Ohde & Sharf, 1988; Rvachew, 1994; Rvachew & Grawburg, 2006; Rvachew et al., 2004; Rvachew et al., 2003; Rvachew et al., 1999; Sharf et al., 1988). (Appendix K provides further discussion of this topic.) The current study is continuing a research line that presumes that phonological representations may be impaired, but does not attempt to explain why they are weak in some children with poor PA and/or speech sound production problems. Regardless of the mechanism responsible for weak phonological representations, it remains clear that performance on phonological processing tasks varies widely in young children, including those with SSD. The current study seeks to determine if a new measure of speech sound accuracy can provide additional explanation for variance in phonological processing, because both speech production and phonological processing are presumed to rely on phonological representations. This new measure could help to provide a clinical indicator of PA skills in young children with SSD. Phonological Awareness in Children with Speech Sound Disorders Children with SSD, as a group, have been found to have poor PA. Therefore, they are generally considered at risk for later literacy problems. For example, Lewis and Freebairn (1992) compared preschoolers, school-age children, adolescents, and adults

21 13 with histories of SSD to age-matched peers without such histories on a variety of PA tasks. Significant differences were found between the groups at all age levels, suggesting that a history of SSD constitutes risk for PA/literacy problems. However, because this was a retrospective study, specific speech sound production characteristics were not considered when evaluating PA/literacy outcomes. Raitano et al. (2004) found that five to six year olds with SSD performed below age-matched controls on a PA factor score which included rhyme, elision (segment deletion), blending, and sound matching. Bird et al. (1995) also found that five to seven year olds with SSD performed below controls on measures of rhyme, initial consonant matching, initial consonant segmentation and matching, and nonword reading and spelling, regardless of whether or not they had concomitant language impairments. In studies examining the effects of PA intervention for children with SSD, Gillon (2000, 2005) found large group differences between children with SSD and controls on measures of PA prior to intervention. Specifically, she found that children with SSD who were receiving intervention that did not include a PA component had a slower rate of literacy skill acquisition compared to typically developing control children without SSD. However, children with SSD who received PA intervention improved PA skills at a rate similar to typically developing control children. Leitao et al. (1997) reported that six year olds with SSD performed below typically developing children on PA measures such as elision (deletion of sounds), blending, segmentation, and invented spelling. They found that some (but not all) children with SSD perform below the range of typically developing children. Although no statistical analyses were performed to address the issue, they suggested that the

22 14 children with SSD performed differently based on the types of speech sound errors that they exhibited (see below for further discussion). This is one of the few attempts that has been made to (qualitatively) relate the variability in PA to types of speech sound errors. One exception to the above findings of low PA performance by children with SSD was provided by Catts (1993), who reported that a group of 15 kindergarten children who had articulation impairments performed as well as typically developing children on several early reading measures when assessed in second grade. However, these children were identified based only on the number of errors on a widely used articulation test, the Goldman Fristoe Test of Articulation (Goldman & Fristoe, 1986). No further speech analysis was reported. This greatly limits the ability to interpret the speech sound characteristics of the sample. This also highlights the possibility that some children with SSD may perform within normal limits on phonological processing and literacy measures. Again, the within-group variance among children with SSD has yet to be thoroughly explained. In summary, there is evidence that children with SSD, as a group, often have below-average PA, putting them at risk for later literacy problems. However, this is not the case for every child with SSD, and the variability in PA skills in this population is largely unexplained. Of interest in the present study is whether this variability can be partly explained by the relative occurrence of the different types of speech sound errors the child exhibits. Measuring Speech Sound Errors There is no universally accepted way of quantifying the accuracy of speech sound

23 15 production. Several different methods of measuring speech sound errors have been used. Research evaluating the relationship between phonological processing and speech sound disorders, quantified by the total number of errors or standard scores on a standardized articulation test, have yielded mixed results (Catts, 1993; Larivee & Catts, 1999; Rvachew & Grawburg, 2006). Drawbacks to the use of standardized tests include (a) the speech sample is often small (under 60 words), (b) the sample often includes just one occurrence of each sound in each word position, and (c) all types of errors are equally weighted (e.g., speech sound distortions may be counted the same as phoneme substitutions or omissions or unusual sound changes). One way to measure general speech sound accuracy is to (statistically) combine multiple measures of speech sound production to approximate or estimate speech sound production as a global construct. For example, Nathan et al. (2004) explored preschool speech sound production and its relationship with early literacy skills using path analysis. Speech sound production was measured as percent consonants correct (PCC) derived from naming 20 pictures and also repeating several real words and nonsense words. This composite of speech production was not a significant predictor of PA and literacy skills over the next two years. However, limitations of this study include the small speech sample, the use of PCC to measure speech sound accuracy (see below), the use of a repetition task (i.e., a phonological memory task) to evaluate speech sound accuracy, and the fact that different types of speech sound errors were not considered. Rvachew and Grawburg (2006) used structural equation modeling to examine whether PA could be predicted from speech production, estimated by PCC in connected speech and scores on the Goldman-Fristoe Test of Articulation-2 (Goldman & Fristoe,

24 ). They found that a model without a link between PA and speech production (estimated by PCC and GFTA-2 scores) was preferred to a model that used speech sound production to predict PA. Thus, the speech sound production-pa relationship was not confirmed using a global estimate of speech production. However, this study failed to evaluate the types of speech sound errors, which is argued to be an important difference in speech production between children. McDowell et al. (2007) used the GFTA-2 along with a measure of nonsense word repetition to estimate speech sound accuracy in 700 children between the ages of two and five. PA was measured by rhyming tasks, blending tasks, and elision (sound deletion) tasks. The combined GFTA-2 and nonword repetition measure was found to account for significant variance (5%) in PA beyond receptive vocabulary. However, limitations of this study include the use of a small speech sample, and the use of a phonological memory task (nonword repetition) to assess speech sound accuracy. It is also unclear how many of these children had a speech sound disorder. Importantly, this study also did not evaluate the types of speech sound errors made by the children. Frequency of Speech Sound Errors: Percent Consonants Correct (PCC) PCC is a widely used method of assessing severity of speech sound disorders (Shriberg et al., 1997a; Shriberg & Kwiatkowski, 1982). In this calculation, the number of correct consonants in a sample is divided by the number of attempted consonants. All consonant errors are therefore equally weighted. Although PCC in conversational speech is said to be related to severity of speech production problems (Shriberg & Kwiatkowski, 1982), it may not be the best measure for evaluating the relationship between speech

25 17 sound accuracy and phonological processing. That is, while it captures the frequency of consonant errors, it does not distinguish between types of errors (distortions, substitutions, omissions). In some instances, PCC based on a picture naming task has been found to predict PA and early literacy. For instance, Bishop and Adams (1990) reported that speech production measured by PCC at age five-and-a-half predicted later reading accuracy and spelling (known to be related to PA), although the contribution of speech sound errors to follow-up prediction of reading was relatively modest (PCC at four-and-a-half years explained 5.4% of the variance in reading accuracy at eight years of age, beyond vocabulary and IQ). Bird et al. (1995) found PCC in a picture naming task to contribute to predicting later literacy difficulty among five to seven year olds with SSD. Larivee and Catts (1999) also found PCC in multisyllabic words at the end of kindergarten to predict reading in first grade. The variance in reading ability explained by PCC overlapped with the variance in reading that was explained by PA; the authors hypothesized that this is evidence that PCC in multisyllabic words taps similar skills to PA, specifically the quality of phonological representations. In contrast, Gillon (2005) found no significant correlation between PCC in conversation and several measures of PA (rhyme oddity, phoneme matching, letter recognition, alliteration, syllable segmentation, letter-sound knowledge, phoneme isolation) during five assessment periods between three and six years of age. Additionally, Rvachew and Grawburg (2006) found that PCC in conversation was not related to PA in a large study of 95 preschoolers with SSD. In conclusion, the results of studies that have used PCC to predict PA and/or

26 18 literacy are mixed. One limitation is that PCC weights all speech sound errors the same, regardless of the type of error 1. Thus, PCC does not capture differences between speech sound patterns in children. Therefore, a new procedure for measuring speech sound errors will be used. It is hypothesized that this procedure will be more sensitive to PA problems than the standard PCC measure, in part because it takes into consideration the presumed relationship between the type of error and phonological representations. It is hypothesized that errors representing relatively weak phonological representations will make a significant contribution to the variance in phonological processing, while errors representing minor deviations from a target (and presumably more accurate phonological representations) will not make a significant contribution to the variance in phonological processing. Types of Speech Sound Errors Difficulty in learning to produce speech sounds correctly can be manifested in a variety of types of speech sound errors. However, not all errors are necessarily equivalent, as would appear to be the case when using PCC or raw number of errors on a standardized test. It is possible to consider speech sound errors and error patterns differently, as has been done by researchers and clinicians since the 1970s (Edwards & Shriberg, 1983; Ingram, 1976; Khan, 1982). Thus, the current study will categorize speech sound errors according to typical and atypical sound changes, often referred to as phonological processes. In this type of analysis, errors are analyzed in terms of place, 1 One alternative to PCC would be to use a revised measure (PCC-R) (Shriberg et al., 1997a), which considers phoneme omissions and substitutions as errors, but ignores distortions. However, this would not capture differences in sound changes involving one feature (e.g., [t] for /k/) and sound changes involving two or more features (e.g., [d] for /k/), nor would it differentiate between typical sound changes (e.g., [t] for /k/) and atypical sound changes (e.g., [s] for /k/). This distinction is further discussed later.

27 19 manner, voicing, and syllable structure. Such sound changes have been used in the literature for many years to describe speech sound errors produced in typically developing children and those with SSD, but previous investigations have often used these sound changes to describe the types of errors individual children (or small groups) make. There have been relatively few attempts to use such errors patterns to quantitatively describe children s speech sound accuracy. In this study, each speech sound error exhibited by each child will be classified according to the types of individual (component) changes involved: distortions, typical sound changes, and atypical sound changes. It is hypothesized that the types of sound changes represent different degrees of similarity between a target representation for a phoneme and the child s actual production. Distortion Errors. Errors that are typically referred to as distortions involve productions that are in the correct phoneme category, but are produced without phonetic precision or accuracy. Distortions, which reflect a slight alteration in the production of a sound (such as a slight problem with tongue shape or placement), are prevalent in the speech of young children with typically developing speech as well as those with SSD (Shriberg & Kwiatkowski, 1994; Smit et al., 1990). For example, the voiceless alveolar fricative /s/ in Sue could be produced with the tongue blade or tip too close to the teeth, resulting in a dentalized production of /s/, transcribed as [sʝu]. Such a production would still be recognized as belonging to the /s/ phoneme category. It has been suggested that distortions (e.g., dentalized or lateralized /s/, labialized /r/) may represent a breakdown in motoric processes (Dodd, 1995; Dworkin, 1980; Fletcher et al., 1961; Hall, 1989;

28 20 Shriberg et al., 2005). It is hypothesized that, because such motor differences are not likely to be related to phonological representations, distortion errors will not be closely related to phonological processing. In fact, Shriberg (1997) states, Unlike phoneme deletions and phoneme substitutions, phoneme distortions have not been associated with deficits in the phonological skills underlying reading, writing, and other verbal skills (p. 107). One investigation that empirically evaluated the relationship between distortions and phonological awareness in children with SSD was by Rvachew et al. (2007). These authors found no significant group differences between four to five year olds with normally developing PA and those with delayed PA in the number of distortions produced on the Goldman-Fristoe Test of Articulation-2 (Goldman & Fristoe, 2000). Preston and Edwards (2007) also reported that speech sound distortions, when counted as errors, reduce the correlation between speech sound errors and phonological awareness in adolescents. Thus, it is hypothesized that distortions are not indicative of weak phonological representations and therefore will not be related to phonological processing. Phonemic Sound Changes (Typical and Atypical). Phonemic sound changes, in which the target phoneme is not produced, may be considered less accurate productions than distortions. Phonemic sound changes include substitutions, in which a different phoneme is produced. For example, cat /kæt/ could be produced as [tæt] (k t). Patterns of omissions may also be observed; for example [kæ] for /kæt/ (t Ø). Some of these sound changes may be atypical; that is, they are found rarely, if at all, in normal development (Dodd, 1995, 2005; Dodd & Iacano, 1989; Dodd et al., 1989; Edwards & Shriberg, 1983; Ingram, 1976; Leitao & Fletcher, 2004). Therefore, atypical

29 21 sound changes are thought of as being less accurate than typical sound changes and are hypothesized to be related to weak phonological representations. Typical Sound Changes. Typical sound changes represent systematic substitutions or omissions that affect a class of sounds (e.g., velars or fricatives) or a sound sequence (e.g., /s/ plus stop clusters) (Edwards & Shriberg, 1983). For example, children with typically developing speech as well as children with SSD may produce the name Sue (/su/) as [tu], replacing the fricative /s/ with the presumably easier stop [t], with which it shares several features (a pattern often called stopping of fricatives ). It is also possible for a child to produce errors that involve more than one feature change at a time. Such changes may be considered interacting or overlapping (Edwards & Shriberg, 1983). For example Sue could be produced as [du] by stopping the fricative and adding voicing. This more complex two-feature change would not be captured using Percent Consonants Correct, because in both [tu] and [du], the one consonant that is assessed (/s/) is produced incorrectly, thus both productions are counted the same. Children with SSD may continue to use these typical phonemic sound changes beyond the ages at which they should have been outgrown (Edwards & Shriberg, 1983). The continued use of these sound changes may reflect a delay in learning linguistic rules for speech production, which could also be reflected in other phonological abilities such as phonological awareness. That is, frequent use of these typical sound errors may reflect a delay in phonological development for both speech production and phonological processing. Atypical Sound Changes: Some speech sound errors exhibited by children with

30 22 SSD represent sound changes that are found rarely, if at all, in typical phonological development. For example, children with SSD may delete the initial consonant in a word, producing Sue as [u] (Dodd & Iacano, 1989), or they may replace the /s/ with a sound produced further back in the mouth, as in [gu] for Sue. Such errors have been characterized as unusual, deviant, atypical, nondevelopmental, or different from those of normally developing English-speaking children (Dodd, 2005; Dodd & Iacano, 1989; Dodd et al., 1989; Edwards & Shriberg, 1983; Ingram, 1976; Klein & Spector, 1985; Leonard, 1985; Lowe, 1994). However, there is no complete list of typical and atypical changes, and there are some sound changes that are less clear-cut. Other changes are uncommon, but still phonetically plausible. In this study, an effort was made to define atypical errors based on existing literature. Definitions of typical and atypical sound changes as adapted for this study are found in Appendix A, along with examples. One of the goals of this study was to investigate the hypothesis that atypical sound changes may represent a greater degree of phonological impairment than other sound changes. According to Dodd and Iacano (1989), A child who follows the normal course of development, albeit slowly, is less linguistically impaired than a child who produces (atypical) errors (p. 334). They also suggest that, The use of (atypical) processes reflects a linguistic deficit, i.e., an impaired ability to abstract the rules governing phonology (p. 335). If this is the case, then atypical sound changes should be more strongly related to poor PA than typical sound changes. Indeed, there is some evidence that atypical phonemic sound changes may be associated with poorer PA outcomes. For example, Dodd et al. (1989) grouped children based on the nature of their speech error patterns. The authors reported that preschoolers

31 23 who consistently used atypical sound changes had an impaired ability to detect whether a word was phonologically legal (e.g., /zmebi/ is not phonologically legal, because it violates rules of English that prohibit initial consonant sequences such as /zm/). While this study generally lends support to the notion that atypical errors may reflect a poorer understanding of phonological rules, examination of the data indicates that the children who were grouped as having atypical speech errors had more errors overall than children who were in the group that used primarily typical errors. Additionally, vocabulary and age were not considered when the groups were compared. Hence, it is unclear whether atypical sound changes, more typical sound errors, more distortions, or other factors (e.g., vocabulary or age differences) were indicative of low performance on the phonological processing task. This is a common problem that is encountered when subgroups of children are compared. Leitao et al. (1997) compared typically developing, speech impaired, language impaired, and speech and language impaired six year olds on several measures of phonological processing. The authors noticed a range in the data, with a possible trend for a bimodal distribution on phonological awareness tasks among six year olds with speech impairment (i.e., possibly two separate subpopulations). They noted that children who frequently used atypical sound changes performed more poorly on PA tasks than those who frequently used typical sound changes. In a follow-up study, Leitao and Fletcher (2004) examined two cohorts of children with SSD at age six, and followed them prospectively until ages They discovered that children in the group that used more atypical sound changes when they were young (i.e., had 10% or more of their sound changes classified as atypical) performed significantly more poorly on phonological

32 24 awareness and literacy measures at follow-up than children who had few atypical phonemic sound changes (less than 10%). However, there were only seven children in each group, making it difficult to generalize findings. Among the studies most relevant to the current project is the work done by Rvachew, Chiang, and Evans (2007). They made an attempt to elucidate the relationship between PA and speech sound errors by analyzing children s consonant errors on the GFTA-2. The participants were 58 children with SSD ages four to five, divided into two groups: those with and without PA problems. The groups were compared on the types of speech sound errors they produced. Errors were classified as distortions, typical syllable structure errors (e.g., final consonant deletion), typical segmental errors (e.g., /s/ [t]), atypical syllable structure errors (e.g., initial consonant deletion), and atypical segmental errors (e.g., t [k]). When in preschool, the only significant group difference was that the children with PA problems produced more typical syllable structure changes. When in kindergarten, the only significant difference was that children with PA problems produced more atypical segmental errors. A limitation in the existing research has been the attempt to categorize children into discrete groups when, in fact, the variable(s) on which they were classified are continuous. For example, to evaluate the relationship between PA and speech sound errors, Rvachew et al. (2007) used a grouping variable to divide children according to their score above vs. below a cut point (one standard deviation below the mean of a group of control children) on a PA task. Dodd (1995) recommended using qualitative judgments for grouping children based on the presence/absence of atypical errors (as well

33 25 as the consistency of those errors). Related to this, Leitao and Fletcher (2004) grouped children based on percentage of atypical phonemic sound changes. This may have resulted in assigning children who had more speech sound errors overall into the atypical group (i.e., the children who had at least 10% of their sound changes defined as atypical may have also used more typical sound changes, so the relative contribution of atypical sound changes remains in question). Hence, it would be necessary to control for the use of all other sound changes when examining the effects of atypical sound changes. An analysis that predicts phonological processing from the relative occurrence of different types of speech sound errors has the advantage of being able to examine the separate influences of these errors, and it does not rely on grouping definitions to predict variance. Given the above descriptions of speech sound errors, Table 1 summarizes how speech sound error types are thought to relate to underlying phonological representations. Measurement System for Quantifying Sound Changes in the Present Study In the current study, a summary of each child s speech will include a score within each of the following categories, determined through narrow phonetic transcription. Appendix A provides definitions and examples of the types of sound changes and examples to show how the sound changes are quantified.

34 26 Table 1: Summary of speech sound error types and their suspected reflection of underlying phonological representations Error Type Distortions Typical Sound Changes Atypical Sound Changes Proposed Reflection of Phonological Representations Relatively accurate, because phonemically correct. Closest to the adult form. Moderately accurate; phonetically motivated and found in the speech of many typically developing children Poorly represented; uncommon and relatively far from the adult form; not phonetically plausible Proposed Statistical Relationship with Phonological Processing Weakest Moderate Strongest 1) Distortions Per Consonant: The number of consonants distorted divided by the total number of consonants attempted. Sound changes that are dialectally acceptable (e.g., partial devoicing of voiced final consonants) are not considered errors. 2) Typical Sound Changes Per Consonant: The number of typical sound changes divided by the total number of consonants attempted (an adaptation of the Process Density Index described by Edwards, 1992, and the Relative Influence on Unintelligibility by Dodd & Iacano, 1989). 3) Atypical Sound Changes Per Consonant: The number of atypical sound changes divided by the total number or consonants attempted (based on the Relative Influence on Unintelligibility by Dodd & Iacano, 1989). Whenever possible, these atypical sound changes are identified based on previous research; they are outlined in Appendix A.

35 27 The main advantages of this system are as follows. Both the types of sound changes and their frequency can be specifically defined using this three-category system. It should be noted that the current classification system also captures sound changes that co-occur on the same phoneme (i.e., are interacting or overlapping ) (Edwards & Shriberg, 1983). That is, if the word cap /kæp/ is produced as [dæp], two sound changes affect the initial phoneme (Velar Fronting [k t] and Initial Voicing [t d]). Both of the constituent (component) changes of this error are counted in the present analysis, whereas only one error would be counted using PCC. In addition, it is important to note that a particular sound error may require coding in more than one category. That is, a child s production of a phoneme may be comprised of more than one type of sound change. For example, if zipper /zǻpǫ/ is said as [sʝǻpǫ], both an atypical error (devoicing of the /z/ to [s] in word-initial position) and a distortion (dentalization) occur. Speech Samples Spontaneous (i.e., non-imitated) speech production samples are considered to provide good evidence of what a child is independently capable of producing. Speech samples taken from conversational speech, although useful for evaluating severity in a clinical setting, would be inadequate for purposes of this study. This is because such samples may fail to elicit a variety of syllable structures and phonemes, may be confounded by morphosyntactic and pragmatic elements, and inherently provide different samples from different children (Campbell & Shriberg, 1982; Paul & Shriberg, 1982). Thus, a picture naming task that controls the speech sounds and word structures sampled

36 28 would be the most representative and equivalent across children. Additionally, a naming task minimizes the complications associated with glossing a child s conversational speech (i.e., determining what the child intended to say), which may be difficult if the child is hard to understand. Because sound changes may affect both syllable/word structure (e.g., Final Consonant Deletion, Consonant Cluster Reduction) and individual phoneme production (e.g., Velar Fronting, Stopping), extensive samples containing a variety of syllable structures and phonemes in different word positions are needed. Larivee and Catts (1999) reported that the production of multisyllabic words is more sensitive to the prediction of reading than is the production of single-syllable words, so a complete sample would also include several multisyllabic words. These are not extensively sampled in many standardized articulation tests. In addition, all consonants should be sampled more than once across multiple word positions, to be certain that there are ample opportunities for observing any of the child s error patterns. Therefore, this study utilizes a 125 item picture naming task adapted from earlier research (Wolk et al., 1993) to meet the requirements outlined above. Exploratory Analyses Phonological awareness has been discussed as one component of phonological processing. Two other areas that have received attention in the literature (and that are also related to reading ability), as described earlier, are phonological memory and phonological retrieval/rapid naming. Both of these skills are also thought to rely, in part, on the accuracy of a child s phonological representations. Although children with SSD

37 29 have been found to perform more poorly than typically developing peers on phonological memory and rapid naming tasks, there is a significant lack of research addressing the relationship between types of speech sound errors and performance on these tasks in preschoolers. Therefore, exploratory analyses will address this issue. The resulting data could aid in the interpretation of the main findings and could provide insights into directions for future research. First Exploratory Analysis: Phonological Memory Phonological memory (PM) is the ability to retain phonological information in short-term memory. It has been argued that this ability is essential for children who are learning to read and spell (Brady, 1991; Metsala, 1999; Wagner & Torgesen, 1987). For example, children who are attempting to sound out (decode) a printed word with which they are unfamiliar often rehearse the sounds associated with the letters, either overtly or covertly. Once they reach the end of the word, they must recall all of those sounds. In fact, phonological memory has been found to be related to literacy skills, including reading and spelling accuracy, and may be weak in poor readers (Elbro et al., 1998; Griffiths & Snowling, 2002; Kamhi et al., 1988; Wagner & Torgesen, 1987). While phonological memory skills have been discussed as being related to phonological representations (Metsala, 1999), phonological memory tasks are not intended to draw upon stored phonological representations of words. Instead, they rely on temporary retention of phonological information. Similar to other domains of phonological processing, phonological memory has been found to be related to a child s age and receptive vocabulary skills (e.g., Edwards et al., 2004; Metsala, 1999; Munson et

38 30 al., 2005). Two common ways of assessing phonological memory involve repetition of numbers and repetition of nonsense words (nonwords). Number repetition assesses an individual s ability to immediately recall sequences of random numbers (e.g., 7, 4, 9; Elbro et al., 1998). This may not be appropriate for preschool children, as some children may have significantly more familiarity with numerical concepts than others, and because a semantic component is involved in the task. Therefore, number repetition is not used in this study. Nonword repetition skills have been found to separate good and poor readers, and to relate to literacy skills such as decoding of nonwords and spelling (e.g., Griffiths & Snowling, 2002; Kamhi et al., 1988; Lewis et al., 2004). Performance on nonword repetition tasks has been shown to be related to age (Metsala, 1999; Roy & Chiat, 2004), as well as receptive vocabulary ability (Edwards et al., 2004; Metsala, 1999; Munson et al., 2005). Because nonword repetition is more appropriate for preschoolers than number repetition, the present study will utilizes a nonword repetition task. Phonological Memory in Children with SSD. It has been argued that a child s ability to hold speech sound information in memory should be related to speech production development (Brady, 1991; Locke & Scott, 1979) and, in fact, weaknesses in phonological memory have been reported for children with SSD compared to their typically-developing peers (e.g., Munson et al., 2005; Preston & Edwards, 2007). Several limitations exist with nonword repetition tasks for children with SSD. Nonword repetition requires the ability to recall phonological input, establish a temporary

39 31 representation, plan the motor movements of the articulators necessary for the sound sequence, and execute those motor movements. That is, the ability to accurately repeat nonwords not only requires the ability to recall phonemes, but also the ability to perform complex motor movements, which may be influenced by adjacent phonemes (i.e., coarticulation effects). Hence, it is unclear which of these processes are disrupted in children with SSD. Therefore, it might be beneficial to use a purer task to assess phonological memory (i.e., one that simplifies motor demands and coarticulatory effects). In such a task, children would be required to repeat simple syllables that are likely to be within their repertoire of production abilities (e.g., /ma/, /da/, /ba/). Thus, a task is used in this study that assesses phonological memory in children with SSD without some of the complications associated with previous nonword repetition tasks (Shriberg et al., 2006). This will help to determine whether the ability to remember speech sounds is problematic for children with SSD. It is possible that difficulty recalling phonological information could be related to the ability to form (store) phonological representations such that children who have trouble retaining phonological information in short-term memory (as evidenced by poor nonword repetition abilities) might be expected to have trouble forming accurate long-term phonological representations. It is therefore hypothesized that performance on the nonword repetition task will be related to the accuracy of phonological representations, as indicated by the use of atypical speech sound changes. That is, atypical sound changes will be more strongly related to nonword repetition than will lower-level errors (i.e., distortions). To date, only one investigation with adolescents has investigated the potential relationship between speech sound accuracy and phonological memory. Preston and

40 32 Edwards (2007) found a significant relationship (r = 0.65) between percent of consonant errors and nonword repetition. However, this relationship was found in adolescents, not preschoolers, and the speech error analysis and phonological memory task differed from the current study. The present study will examine the contributions of different types of speech sound errors to variance in phonological memory in preschoolers with SSD. Second Exploratory Analysis: Rapid Naming Phonological retrieval is often assessed using rapid naming tasks. Rapid naming (RN) tasks require children to name a series of pictures/objects/ letters/ numbers as rapidly as possible. These tasks are frequently used to assess the ability to retrieve phonological information quickly. RN has been found to predict literacy skills, both concurrently and longitudinally (Allor, 2002; Catts et al., 2001; Kirby et al., 2003; Schatschneider et al., 2004). RN tasks have also been reported to separate good and poor readers (Denckla & Rudel, 1976), and to separate children at higher risk of reading problems from those at lower risk (Cardoso-Martins & Pennington, 2004). Similar to other domains of phonological processing, age is also a significant predictor of performance on RN tasks (Troia et al., 1996). However, RN and vocabulary tend not to be highly correlated, as reported in a recent meta-analysis of school-age children (r = 0.26) (Swanson et al., 2003). It has been argued that slow performance on naming tasks is due to poor phonological skills (Denckla & Rudel, 1976; Kirby et al., 2003; Raitano et al., 2004; Stringer et al., 2004; Swan & Goswami, 1997b; Troia et al., 1996). That is, when children are slow to name pictures, objects, numbers, or letters, the deficit may be because of poor

41 33 access to the phonological features of the word. Hence, the ability to quickly access phonological representations could impact the speed of naming. However, it is still unclear whether naming speed is a function of poorly stored representations and/or poor retrieval of phonological information. Debate exists as to whether RN should be considered a phonologically-based task (Wolf & Bowers, 1999), but its predictive value in literacy development is well documented. Therefore, predicting performance on RN in preschool children could provide some insight into processes that underlie literacy development. However, it should be noted that RN tasks have not been frequently used with preschoolers, hence, the exploratory nature of the RN component of the study. Rapid Naming in Children with SSD. A small body of research suggests that children with SSD may perform more slowly on RN tasks than their typically developing peers (Leitao et al., 1997; Preston & Edwards, 2006). However, the research is limited with this population, and the underlying reason for this difference in naming speed is unclear. Recent research indicates that rapid naming of phonologically complex words may be more challenging for adolescents with SSD than for their normally speaking peers; however, no group difference was observed on naming of monosyllabic stimuli (Preston & Edwards, 2006). Leitao et al. (1997) found that six year olds with SSD (as well as those with language impairments) performed below typically developing peers on several rapid naming tasks: letters, numbers, objects, and colors. Catts (1993) also reported that, for children with speech and language impairment, rapid naming of animals in kindergarten was moderately correlated with word reading in second grade.

42 34 Unfortunately, none of these studies examined specific speech production errors relative to rapid naming, and none used this task with preschoolers. One contradiction to the above evidence has been research reported by Raitano et al. (2004), who found no difference between five to six year olds with SSD and control participants on a rapid naming factor score which included naming of colors and objects. However, if syllable length plays a role, as suggested by Preston and Edwards (2006), then words of more than one syllable should be included in the stimuli. This limitation will be addressed in the present study by using both a monosyllabic and a disyllabic RN task. Because Rapid Naming is thought to rely on rapid access to phonological representations, it is hypothesized that the speech error measurement system based on the presumed accuracy of phonological representations will significantly predict variance in RN. Primary Goals of the Study and Hypotheses The importance of understanding how phonological processing skills vary in children with SSD has been described. Because types of sound changes are presumed to reflect the accuracy of phonological representations, it is possible that types of speech sound changes can explain variance in phonological processing. This hypothesis will be tested, and a new speech error classification system will be compared to the commonly used Percent Consonants Correct (PCC). Thus, while previous studies have examined how the frequency of speech errors relates to phonological awareness using PCC, this study is unique because it evaluates

43 35 both the frequency and the types of sound changes that are involved in children s speech errors. The following hypotheses are investigated in the current study: Hypothesis 1: Phonological awareness (PA) will be related to (correlated with) speech sound error types in preschoolers with SSD, according to the proposed accuracy of phonological representations. Hypothesis 2: Types of speech sound errors thought to reflect weak phonological representations will predict variance in PA above and beyond receptive vocabulary and age in preschoolers with SSD. Hypothesis 3: An analysis that characterizes sound changes according to the relative accuracy of phonological representations will provide a better explanation of the variance in PA than an analysis that considers all consonant errors to be equal (PCC). (Exploratory) Hypothesis 4: A speech production analysis that considers three types of sound changes will predict variance in phonological memory beyond the contribution of age and receptive vocabulary. (Exploratory) Hypothesis 5: A speech production analysis that considers three types of sound changes will predict variance in rapid naming beyond the contribution of age and receptive vocabulary. Summary To reiterate, phonological processing has been defined to include phonological awareness, phonological memory, and rapid naming. Phonological processing skills are

44 36 important in predicting literacy development. The goal of the present investigation is to determine whether speech sound accuracy can predict concurrent performance on phonological processing tasks in children with SSD. Two procedures for analyzing speech will be compared: (1) Percent Consonants Correct (PCC), and (2) an analysis that represents both the frequency and type of speech sound errors.

45 37 II : METHODS This study was approved by the Institutional Review Board at Syracuse University. General results (e.g., test scores) were made available to the parents of children who participated at the end of each session. Children were given books for their participation, and parents were financially compensated for their time. Participants Children sought for the study were preschoolers, ages four to five years, with speech sound disorders (SSD) of unknown cause (i.e., functional or idiopathic). No attempt was made to include or exclude children based on the type of SSD (articulation or phonological disorder, suspected childhood apraxia of speech, deviant or delayed speech sound production, etc.) because of the lack of agreed-upon criteria for such diagnoses. Children who were eligible for the study met the following criteria (described below in more detail): 1. Diagnosed by a speech-language pathologist with a SSD (articulation/ phonological disorder, suspected childhood apraxia of speech) 2. Primary language and dialect was General American English 3. Had no known developmental, neurological, or oral structural difficulties (such as mental retardation, cerebral palsy, pervasive developmental disorder/autism, cleft palate, permanent hearing loss, etc.) that might cause the SSD. A history of ear infections was acceptable. 4. Four or five years old and had not yet begun kindergarten. 5. Did not have a moderate or severe receptive language delay. Mild receptive

46 38 language delay was acceptable. Children were not excluded from the study for expressive language concerns. Recruiting The primary method of recruiting relied on referrals from speech-language clinicians in Upstate New York. These professionals were contacted via addresses that were publicly available on the internet (American Speech-Language-Hearing Association, New York State Speech-Language-Hearing Association, local agency web sites), presentations to local agencies, personal contacts, advertisements in professional newsletters, and direct mailings to agencies and preschools. A description about the study went out to these professionals, indicating the need for children who met the criteria listed above. Flyers were made available to these clinicians to pass along to parents of children who might qualify. In addition to clinical referrals, announcements were made available to the public in newspapers, the Syracuse University SUNews list serve, the Gebbie Speech-Language-Hearing Clinic, the Gebbie Clinic web site, and posters in local preschools. Parents then directly contacted the researcher if they were interested in obtaining further information. Parent Phone Interview Once the parent contacted the researcher about the study, a phone interview was conducted to confirm that the child was of the appropriate age and that the child had difficulty with speech sound production. Most of the children (all but two) were in

47 39 speech therapy 2. Once the study procedures were explained to the parent, they were asked if they wished to participate. All parents of children who met the criteria indicated that they wished to participate, so a screening session was scheduled (n = 53). Two of these parents contacted the researcher and scheduled the Part I Screening session, but cancelled the session and did not reschedule. Additionally, two parents contacted the researcher about participating, but the children were not of the appropriate age for the study, so they were not included. All parents reported that their child had no known permanent hearing loss or developmental disabilities that might cause a SSD (such as cleft palate, autism, cerebral palsy). Parents also confirmed that none of the children were exposed to a parent/guardian who spoke a language other than English at home, and all parents reported that the adults in the home were speakers of General American English. This was also informally confirmed in the home visit. Part I: Screening A screening was first conducted to determine eligibility for the study. Informed consent was obtained from one parent prior to the screening, and children provided oral assent for participation. Screenings took place either at the child s home (n = 49) or at a quiet room used for child research at Syracuse University (n = 2), based on parents preferences. Parents were allowed to observe if they wished. 2 One child (P47) was not in speech-language therapy but the parent expressed concerns about the child s articulation. The child was seen for the study but was later excluded because of high articulation score. A second child (P40) was not in therapy, although parents indicated that he did qualify for services. He achieved low speech sound production scores on the GFTA-2 and was included in the study.

48 40 A case history form was completed by parents during this session. Parents provided additional detail about the child s developmental history (medical, social, educational, speech/language) and family background. Socioeconomic status (SES), collected for descriptive purposes, was measured by the number of years of parental education, similar to other studies. This variable has been found to relate both to the prevalence of SSD and to PA skills (cf. Campbell et al., 2003; cf. Catts et al., 2001; Nittrouer & Burton, 2005). The entire screening protocol for Part I was pilot tested with two typically developing preschoolers, and portions of the protocol were pilot tested with two other children. This was done to obtain a time estimate of the length of the sessions, and to familiarize the examiner with the administration procedures for the tests. It took between minutes to administer all the screening tasks for Part I. Task Order for Part I Four tasks were randomly ordered: Goldman-Fristoe Test of Articulation-2 (GFTA-2) (Goldman & Fristoe, 2000); Concepts and Following Directions subtest and the Sentence Structure subtest of the Clinical Evaluation of Language Fundamentals: Preschool-2 (CELF:P-2) (Wiig et al., 2004); Peabody Picture Vocabulary Test-4 (PPVT- 4) (Dunn & Dunn, 2007); Pattern Construction subtest of the Differential Ability Scales (DAS) (Elliott, 1990). An oral mechanism screening, devised for this study, was either the fourth or fifth task. This was late in the session so that rapport had been established, in case some children might find it embarrassing to make movements with their mouth.

49 41 In addition, fatigue should have very little effect on the pass/fail outcome of the oral mechanism screening. Participants were offered a short break after two or three tasks. Speech Sound Production The Sounds-in-Words subtest of the GFTA-2 (Goldman & Fristoe, 2000) was chosen to screen speech because it is a commonly used test for the clinical diagnosis of speech sound disorders in preschoolers. This requires children to name pictures on 34 plates eliciting 53 target words, with online judgment of the accuracy of production of 61 consonants in initial, medial, and final position and consonant clusters. To qualify for the study, children had to achieve a standard score below 90 on this test. This task was audio recorded following procedures described below, but the task was scored online (live) in order to determine eligibility at the time of the screening. Audio recordings were consulted only if the child s score was 80 or above because the misidentification of a few errors would impact eligibility. According to the manual, the median test-retest reliability for phonemes in the initial, medial, and final positions of words is 98% agreement. Median inter-rater agreement for the presence of errors is 93% in the initial position, and 90% in the medial and final positions. The alpha reliabilities for the age groups in this study range from Approximately 10 communities from Upstate New York are represented in the standardization sample. An informal oral peripheral screening was also used to confirm that there were no gross structural or functional problems contributing to the SSD. This involved having the child imitate the examiner s mouth movements: close lips, purse lips, smile, elevate tongue, protrude tongue, and lateralize tongue. Oral structures were also observed for

50 42 abnormalities (teeth, hard and soft palate, lips, face). All participants demonstrated adequate structural/functional integrity of the oral peripheral mechanism. Language Because the children would be required to participate in phonological processing tasks, participants were required to demonstrate adequate receptive language skills. This was operationally defined as achieving scores not lower than one and one-third SD below the mean on at least two of three receptive language tasks: the Peabody Picture Vocabulary Test-IV (PPVT-4), the Concepts and Following Directions subtest of the Clinical Evaluation of Language Fundamentals: Preschool-2 (CELF:P-2), or the Sentence Structure subtest of the CELF:P-2. This was believed to be a reasonable means of not excluding children who might have subtle receptive language difficulties, but who would still be likely to follow directions and understand vocabulary well enough to participate in research tasks. Expressive language was not formally evaluated, because none of the experimental tasks required more than single word responses, and because the theoretical justification for the study did not rely on a child s expressive language skills. Two subtests of the CELF: P-2 (Wiig et al., 2004) were used to screen receptive language skills. The Concepts and Following Directions subtest requires children to follow verbal directions by pointing to pictures of animals, usually in a specified order. For example, Point to the big dog, then point to the little monkey. Items increase in length and complexity. There are 22 items, and testing is discontinued after five consecutive errors. The manual indicates that test-retest correlation is 0.83 for 4 year

51 43 olds and 0.88 for 5 year olds. Coefficient alpha for children in the age range seen here were , and split-half reliability is reported to be The CELF: P-2 Sentence Structure subtest requires children to point to a colored picture (from a field of 4) that accurately depicts a scene corresponding to the examiner s description. For example, Point to The girl who is standing in the front of the line is wearing a backpack, and the distracter pictures typically show slight variations, such as a girl in the back of the line with a backpack, the second girl in line wearing a backpack, etc. There are 22 items, and testing is discontinued after five consecutive errors. Testretest correlation is reported to be 0.85 for 4 year olds and 0.79 for 5 year olds. Coefficient alphas for children in the age range seen here were , and split-half reliability is reported to be The PPVT-4 (Dunn & Dunn, 2007) measures single word receptive vocabulary by requiring children to point to a colored picture (from a field of four) that corresponds with the single word spoken by the examiner. Items increase in complexity, and the testing continues until a ceiling is reached. Earlier versions of this instrument have been used in several studies to estimate receptive vocabulary skills in children with SSD (Rvachew, 2006; Rvachew & Grawburg, 2006). The newest version of this instrument was updated, in part, to improve reliability in preschoolers. Test-retest reliability for the age groups in this study range from r of Split-half reliability for the children ages 4;0-5;6 range from , and coefficient alphas are About 10 facilities from upstate New York are represented in the standardization sample.

52 44 Nonverbal Cognition The Pattern Construction subtest of the Differential Ability Scales (DAS, Elliott, 1990) was used as a brief screening of nonverbal intelligence. Children are shown pictures of patterns of yellow and black squares. They then try to manipulate and arrange the blocks to replicate the patterns shown in the picture. Both speed and accuracy of the pattern construction are considered in scoring. Of the nonverbal subtests in the DAS, the Pattern Construction subtest was chosen because it is a relatively efficient means of estimating nonverbal cognition (i.e., it can be scored live, and it has the highest correlation of all nonverbal subtests of the DAS with the Nonverbal Ability Composite). Children were included if they achieved a T score above 37. Test-retest correlations for the ages in this study are r = , and internal reliability is Table 2 shows a summary of the tasks from Part I, along with the criteria for inclusion in the study. Participants Included in Part II Fifty-one children participated in Part I (screening), and the 44 who met the criteria described above were invited to participate in Part II. Because one parent scheduled and then canceled the Part II session, a total of 43 children participated in the experimental tasks. Time between Part I and Part II ranged from 0-27 days, with an average of 10 days between sessions. Table 3 summarizes the performance on the Part I tasks for the 43 children who participated in Part II. The seven who did not qualify are excluded, as is the one who chose not to participate (see also Figure 2).

53 45 Table 2: Inclusionary criteria for the study Speech Diagnosed with a speech sound disorder Standard Score of <90 on GFTA-2 Exposed to General American English as the primary dialect, as reported by parents and observed in the screening Speech disorder not a result of permanent hearing loss or developmental disability, as reported by parents No obvious oral structural or functional problems Receptive Language (met at least two PPVT-4 Standard Score >80 CELF:P-2 Sentence Structure Scaled Score >6 CELF:P-2 Concepts & Following Directions Scaled Score >6 of three criteria) Nonverbal Cognition Differential Ability Scales: Pattern Construction subtest T score >37 Other No known developmental disabilities, as reported by the parent The 43 participants in Part II included 34 males and 9 females, a 3.78:1 gender ratio. This ratio is not statistically different than the 2.75:1 male: female ratio reported for children with SSD by Shriberg (1994) (χ 2 [1] =0.724, p = 0.395). All participants were Caucasian except for one female who was adopted from Asia. The average reported maternal education level was 16 years of formal schooling, or the equivalent of four years of college. The average paternal education level was 15 years of formal schooling, or about three years of college. It is evident from Table 3 that some of the participants had

54 46 relatively high vocabulary skills compared to the standardization sample of the PPVT-4, as well as relatively high nonverbal cognition, as measured by the DAS Pattern Construction subtest. Both the PPVT-4 and DAS Pattern Construction subtest were significantly above the expected mean based on one-sample t-tests (p s <0.01). Possible explanations for this include referral bias, as this project relied on SLPs to distribute information about the study, and self-selection bias, with families from higher socioeconomic homes perhaps being more likely to participate. Table 3: Descriptive statistics for the 43 preschoolers who participated in Part II and were included in the final analysis Mean SD Range Age at Part II In months GFTA-2 Sounds in Standard Score (mean 100, SD 15) Words Subtest Percentile DAS Pattern T score (mean 50, SD 10) Construction Percentile CELF:P-2 Sentence Structure Scaled Score (mean 10, SD 3) Concepts & Following Directions Scaled Score PPVT-4 Years of Parental Standard Score (mean 100, SD 15) Percentile Mother Education Father

55 47 Figure 2: Flow chart of procedures with number of participants Qualified through Phone Interview & Scheduled Part I Screening (n=53) Did Not Participate in Part I Screening (session cancelled, n=2) Participated in Part I Screening (n=51) GFTA-2 CELF:P-2 Sentence Structure CELF:P-2 Concepts & Following Directions PPVT-IV DAS Pattern Construction subtest Oral Peripheral Exam Did Not Participate in Part II (n=8) GFTA-2 > 90 (n=4) Did not complete one or more tasks (noncompliant; n=2) Language and nonverbal cognitive scores too low (n=1) Qualified but cancelled Part II (n=1) Participated in Part II Experimental Tasks (n=43) Hearing Screening Four Phonological Awareness Tasks Picture Naming Task Syllable Repetition Task Rapid Naming Tasks

56 48 Part II: Experimental Tasks Part II was conducted at Syracuse University for nine of 43 children; the remainder were seen at their homes. Part II took between minutes, and was split into two sessions if the child showed significant signs of fatigue or was distracted. Children were offered frequent breaks throughout Part II. Task order was pseudorandomized, with tasks being administered in the following way: 1. Hearing Screening 2. Introduce PA pictures: naming of/familiarization with 96 target words 3. Randomly chosen PA task 4. Randomly chosen PA task 5. Picture Naming task (for speech sample) 6. Randomly chosen PA task 7. Randomly chosen PA task 8. Rapid Naming or Syllable Repetition Task 9. Rapid Naming or Syllable Repetition Task This order was chosen because it was essential that children be familiarized with the phonological awareness (PA) task pictures before being exposed to them in the experimental tasks. All four PA tasks required use of the laptop and were similar in format (e.g., nonverbal response to stimuli); so these four tasks were split into groups of two, with the picture naming task between. Because the final two tasks were exploratory, they were completed at the end in case there was insufficient time to complete them. Only one child failed to complete one of the exploratory tasks.

57 49 Hearing Hearing was screened using a portable MAICO MA 27 audiometer. Behavioral responses were required (i.e., raising the hand when the tone is presented). Following a training/familiarization at about 65 db, pure tones were presented at 20 dbspl at 1000, 2000, and 4000 Hz (ASHA Audiologic Assessment Panel 1996, 1997). If tested at home, failure to respond at 20 db was followed by presentation of the same frequency at 25 db, and a response at this level was accepted as a pass due to presumed ambient noise levels in the home. Forty-one participants passed the screening. One participant (P38) was not screened because the audiometer was not available. One participant (P46) passed in the left ear but did not pass the screening in the right ear (right ear threshold of 30 db at 1000 Hz, passed at 25 db at 2000 Hz, threshold of 35 db at 4000 Hz.). He was kept in the study because there was no history of permanent hearing loss, and he did not appear to be an outlier in the dataset. Because this participant had a cold at the time of testing, failure to respond may have been due to otitis media. Care was taken so that all recorded stimuli were presented to this participant at a loudness level that he indicated was adequate. All analyses were repeated without this participant in the dataset, and the conclusions were unchanged. Speech Assessment Recording Procedure All tasks requiring verbal responses (Picture Naming Task, Syllable Repetition Task, Rapid Naming Tasks) were audio recorded. Two digital recorders were used, so as to have a backup recording if one device failed: (a) Zoom H4 Handy Recorder with two

58 50 studio quality X/Y pattern condenser microphones set to record as digital WAV files at 24-bit quantization and 48 khz sampling rate; (b) Olympus WS-331M digital voice recorder with built-in stereo microphone, recorded on extra-high-quality stereo mode with no low-cut filter. This device saved as Windows Media Audio (WMA) sound files with a 44.1 khz sampling rate. For later review of audio files, the WMA files were converted to WAV files so that they could be reviewed in the Praat (reference?) acoustic analysis software program. The clearest of the two recordings (usually the Olympus device) was used for transcription/analysis. For one participant (P35), the digital audio equipment was not brought to Part II; therefore, a cassette recording was made and this was later digitized. Speech Sample A 125 word picture naming task (PNT) adapted from Wolk, Edwards and Conture (1993) was used to assess all consonants in nearly every position in which they occur in words (initial, medial, and final). All vowels of General American English were included at least twice, as well as numerous consonant clusters/blends and multisyllabic words (see Appendix C). The entire sample consisted of 480 consonants, although this total was adjusted when necessary (e.g., if the child did not produce a particular word). Scripted prompts were used to elicit the target word if the child mislabeled a picture. For example, for the target splash, some children said, Jumping into the pool, so the examiner said, He jumped into the pool and it made a big. If a child failed to respond with the target word after several attempts at eliciting it, a delayed imitative response was allowed. That is, a model was provided by the examiner, followed by a

59 51 comment, then the child was again prompted to produced the word (e.g., He made a big splash. See? There s water going everywhere. He made a big. ) For half of the children, the PNT was administered in order from item 1 to item 125. For the other half of the children, the PNT was administered in reverse (i.e., from item 125 to item 1). The picture naming task was piloted with two typically developing children, ages four and five, and one seven year-old with a SSD. To the extent possible, pictures that were mislabeled by these children were replaced with newer or more explicit pictures to elicit the target words. Transcription Children s responses on the picture naming task were narrowly phonetically transcribed by the author. Praat software was used to play the digital files in free-field in a quiet room. Time between the initial assessment and the first phonetic transcriptions varied from one day to approximately 4 months, depending on the participant. To ensure accuracy of the transcriptions, audio files were reviewed by the author a minimum of three times for each participant. Transcriptions were entered directly into the Logical International Phonetic Programs software (LIPP, Oller & Delgado, 2001). For detailed phonetic variations, the author used the diacritics in this software program, and supplemented with the use of a nonspecific diacritic for clinical distortions (e.g., derhoticized /r/, lateralized /s/). Hence, any phoneme that had this distortion diacritic was counted as incorrect using PCC, and classified as a distortion using the threecategory system devised for this study. The transcriber who completed reliability

60 52 listened to the sound files using AKG K 240 headphones, and wrote out her detailed transcriptions rather than using LIPP (reliability details are provided later). If the child spoke a word more than once, the clearest recording of the two renditions was used; if both were clear, the first was chosen. When there was overlay with another speaker or there was background noise covering a portion of the word, the child was given credit for producing those overlaid sounds correctly. If a child added morphological endings, those were not analyzed (e.g., if a child said toys instead of toy, the plural was not scored). Further detail regarding transcription rules and procedures is included in Appendix A. Types of Speech Errors Using these transcriptions, two consonant analysis schemes were compared to see if either was better able to predict variance in phonological processing: 1) Percent Consonants Correct (PCC) was calculated from the picture naming task, with all consonant errors being weighted the same (i.e., substitutions, omissions, and distortions). Each consonant was therefore judged to be correct or incorrect. 2) Three types of speech sound changes: Distortions per consonant, Typical Sound Changes per consonant, and Atypical Sound Changes per consonant were calculated from the narrow transcription of the child s productions on the picture naming task. Note that speech errors for both analyses were computed by hand, rather than by computer, to allow for dialectal variations (e.g., partial devoicing, affrication of /tw, dw, tr, dr/ clusters, glottal stop replacement for final /t/, etc.) and for interacting sound

61 53 changes. This is because the LIPP program is limited in its ability to accurately code some of these sound changes (White, 1997). Initial coding of speech sound errors was completed by the author at the same time of the transcription. However, because it was necessary to refine some of the sound change definitions (Appendix A) as the study progressed, each participant s phonetic transcriptions were reviewed a minimum of three times to ensure accuracy and consistency of error coding. Typical and atypical sound changes were defined based on previous research. Changes in place of articulation, manner of articulation, voicing, and syllable structure that are commonly found in the speech sound development of children have been generally well described (Edwards & Shriberg, 1983; Ingram, 1976; Khan, 1982). In addition, there has been a moderate amount of discussion about what constitutes atypical or unusual sound changes. However, some sound changes have not been discussed adequately or the definitions are not fully agreed upon. For the present study, atypical sound changes were defined based on prior research, to the extent possible, but some definitions had to be refined to be sufficiently explicit (see Appendix A). A relatively conservative approach to defining sound changes as atypical was used. When there was lack of agreement in the literature, a general rule of phonetic plausibility was adopted. Thus, if a consonant sound change occurred that was potentially due to phonetic context, word position, or the influence of other consonants in the word, it was not considered atypical. Appendix A and Appendix B provide further detail about the coding of sound errors. To give a common example, velarization (or backing) of alveolar stops (e.g., d g) has often been considered atypical (e.g., Dodd & Iacano, 1989) because typically developing children generally replace back sounds with front sounds. Given the

62 54 definitions developed for this study, this sound change would be considered atypical only if it could not be accounted for by a typical sound change, such as velar assimilation. Thus, /d/ [g] in the word dinosaur would be considered atypical because there are no other velars in the word to trigger this change. However, if /d/ [g] occurred in the word pudding, it would be accounted for by the typical error of velar assimilation. (That is, /d/ assimilates to the velar feature of the /ŋ/.) Phonological Awareness While some PA tasks require spoken responses, this may confound results when assessing PA in children whose speech is often hard to understand (Sutherland & Gillon, 2005). Therefore, PA tasks that were selected for this study met the following criteria: (1) no spoken response was required; (2) the task has been shown to be related to later literacy development; and (3) the task was age-appropriate. PA assessment tasks and protocols were therefore based on prior research (see below). PA Stimuli Preparation and Presentation Ninety-six words (that were different from the picture naming task) were selected for use in the PA tasks. All 96 words were monosyllabic, and most were made up of CVC syllables (e.g., dog), with a few being CV (e.g., shoe) or CCVC (e.g., spoon). Words were chosen based on their phonological features (consonant and vowel components) and picturability/interpretability by four year olds. Most words were nouns, but there were two verbs (run, tap) and one adjective (red). To limit the number of items with which the children had to be familiar, each

63 55 word was used either two or three times, but no word was used more than twice in a given task, and never twice as the target response. For example, coat appeared once as a distracter item in the Onset Matching task, once as a correct target in the Rhyme Matching task, and once as a correct target in the Blending task. The stimuli for the four PA tasks are listed in Appendix D. Audio stimuli and instructions for the PA tasks were recorded by an adult male (the author) using a Sure WH22 head mounted microphone fed into a Rolls MX 54s Pro Mixer Plus in a double-walled soundproof booth. The signal was recorded at 44 khz sampling rate on a Dell Inspiron 8600 laptop in Praat v Stimuli were stored as WAV files. They were presented to the children using the same computer, and were imported into Microsoft Power Point. Audio stimuli were paired with visual stimuli, which were clip art pictures taken from a variety of sources (e.g., Microsoft Word, Google Images, and other internet sources). An external speaker was used to amplify the audio signal in environments where the internal speakers of the laptop were judged to be insufficient. PA tasks were pilot tested with one typically developing four year old, two typically developing five year olds, and a seven year old with a SSD. As with the picture naming task, if some of the children had difficulty identifying the pictures, different pictures were selected. Approximately five of the 96 pictures were replaced with newer clipart in order to better represent the target words. Familiarization Before any of the PA tasks were administered, children were familiarized with

64 56 the 96 target words to be used in the experimental PA tasks. Children were shown the pictures on the laptop. Instructions were, I am going to show you some pictures on the computer. Tell me the names of the pictures that you see. The examiner controlled the rate of presentation of the pictures (i.e., they were not time-controlled by the software). If a child was unfamiliar with the picture or provided the wrong label, a spoken model was provided, the child was asked to imitate the word, and then another model was provided. For example, when shown a picture of hen, if a child said rooster, the correct label was provided (e.g., That s a picture of a hen. Can I hear you try that word? Good. That s hen.). General Procedure for PA Tasks Children sat on the floor or at a table in front of the laptop. For each task, three or four pictures appeared together on the computer screen. These were arranged in a random configuration on the screen, so that the correct response picture was not consistently in the same position. Because three PA tasks used a field of four choices from which the child could select a response, the screen was divided into four quadrants. For the Blending task, three picture choices were arranged in a row. Figure 3 (shown after all tasks are described) provides examples of the visual layout for each of the PA tasks. Because all of the PA tasks required nonverbal responses to audio/visual stimuli, children were given a 12-inch magic wand with a soft end that was used as a pointer. They used this to lightly touch the computer screen to indicate their response. Some children chose to provide a verbal response, but they were encouraged to point as well

65 57 because verbal responses could be unintelligible. The recorded audio stimuli were played only once, unless the child failed to respond (e.g., if distracted) or requested repetition. The examiner pointed to the pictures on the screen as they were named. If a child changed his/her response, the final response was scored. There were five training items for all of the PA tasks, with feedback and instruction provided if the child responded incorrectly. All responses were noted online by the examiner. The first three PA tasks described below were adapted from Bird et al. (1995). These include rhyme matching, onset matching, and onset segmentation and matching. All three tasks have been used with preschoolers with SSD to predict early literacy skills (Rvachew, 2006; Rvachew & Grawburg, 2006). The tasks were adapted to be presented with recorded audio stimuli and clip art pictures on a laptop in PowerPoint (instead of using puppets, as in the original research). Additionally, target and distracter items were modified to control phonological similarity of distracter items to the targets, as described below. The stimuli used for all PA tasks are in Appendix D. Rhyme Matching. The rhyme matching task included 16 experimental items, with four blocks of four rhymes (i.e, four items that rhyme with the names Dan, Doug, Pete, Ned). For each trial, four pictures appeared on the computer screen at once, the correct picture and three distracters. Each block was introduced by presentation of a photo of a person paired with audio recording. For example, This is Dan. Dan likes things that rhyme with his name. Help Dan find things that rhyme with his name. The name was repeated during each item: Which one rhymes with Dan? spoon, cap, mouse, pan. Which one rhymes with Dan? (child points). For each item in the Rhyme Matching, one of the distracters had the same vowel as the target (here, /æ/ in cap), one

66 58 had the same final consonant (here, /n/ in spoon), and one had no phonemes in common with the target (here, mouse). A picture of the person whose name was to be rhymed always appeared in the upper-left hand portion of the screen (here, a picture of Dan). Five training items were provided with corrective feedback as necessary. Audio stimuli for each trial were recorded, and the examiner controlled when each item was presented. Onset Segmentation and Matching. A similar paradigm was used for the Onset Segmentation and Matching task. When presented with a field of four pictures, children were instructed to find a word that begins like a particular name. For example, Which one begins like Tom? Pin, juice, tie, door. Which one begins like Tom? (child points). Five training items were provided with corrective feedback. Prior to the training items, the children were shown a slide with examples of correct responses, such as Time and turtle begin like Tom. Now let s find some more. One of the distracter items always began with a phoneme that children frequently produce as a substitute for the target phoneme. For example, all of the matching items for Tom included a correct target beginning with /t/, but also a foil beginning with /d/ (e.g., door). There were five experimental items that begin with /t/ (to match Tom), and five that begin with /s/ (to match Sam). Onset Matching. The Onset Matching task required children to find a word from a field of four that began with a given sound. Unlike the Onset Segmentation and Matching task where children had to determine the initial sound of a word before matching it, in the Onset Matching task, children were given the phoneme they had to listen for. For example, Which one begins with /p/? Deer, kite, bug, pin. Which one begins with /p/? Five training items were provided (three with /r/, two with /m/) to

67 59 familiarize the child with the task. The experimental items included five where the child had to choose a word beginning with /p/, and five beginning with /tȓ/ (ch). As with the Onset Segmentation and Matching, one foil or distracter item in the Onset Matching began with a phoneme that children often produce as a substitute for the target. Thus, all /p/ matching items had a foil beginning with /b/, and all of the /tȓ/ (ch) items had a foil beginning with /ȓ/ (sh). The remaining two distracters began with phonemes that are less similar to the target (e.g., /d/ in deer differs from /p/ in both place and manner of articulation). Blending. To assess onset-rhyme and C-V-C phoneme blending (or synthesis), a task was adapted from previous research (Larivee & Catts, 1999). Children were presented with a set of three pictures on the computer screen (e.g., fan, fish, dish), and listened to a recorded presentation of the target word spoken in segments (e.g., /f--ǻ--ȓ/). There was approximately 1.0 second between phonemes. The child pointed to the picture to indicate a response. To introduce the task, children were shown a picture of a monster and told, This monster says things in a funny way. He says words in pieces. See if you can guess what he is saying. For each item, a carrier phrase spoken by a female ( Point to the one that you hear ) preceded the segments, spoken by the monster (a male). Twelve experimental items for the Blending task were presented in a game-like format in PowerPoint. The first six items required onset-rhyme blending (i.e., initial consonant [onset], then vowel-consonant pair [rhyme]). The last six required blending of individual phonemes (consonant, then vowel, then consonant). All targets were CVC words. Three training items with corrective feedback were presented before the six

68 60 Figure 3: Examples of PA stimuli Rhyme: Which one rhymes with Dan? Cat, fan, run, bike. Which one rhymes with Dan? Onset Segmentation & Matching: Which one begins like Tom? Pin, juice, tie, door? Which one begins like Tom?

69 61 Onset Matching: Which one begins with /p/? deer, kite, bug, pin. Which one begins with /p/? Blending: Female: Point to the one that you hear: Male: /m --- ațs/

70 62 onset-rhyme blending items (e.g., /f--ǻȓ/). Two additional training items with corrective feedback were used before the six C-V-C blending items (e.g., /f--ǻ--ȓ/). Both distracter words had phonological similarity to the target: one foil began with the same phoneme as the target (e.g., fan begins like fish), and one foil had the same vowel and/or final consonant as the target (e.g., dish has the same rhyme as fish). Exploratory Analyses Phonological Memory Phonological memory is an additional phonological processing domain that is discussed as being related to phonological representations, speech sound disorders, and literacy. In this study, phonological memory was assessed by the Syllable Repetition Task, which was developed for children with poor intelligibility (Shriberg et al., 2006). Shriberg et al. (2006) report data from 99 children confirming that the four consonants used in this task are in the phonetic inventories of children with SSD, that scores on the syllable repetition task correlate moderately with other nonword repetition tasks, and that the scores on this task met distributional requirements for parametric statistical analysis. Stimuli (provided by the first author of the original work) were two to four syllables in length, and were spoken by an adult female. For each item, the stimuli were produced with no pause between syllables. Four early developing consonants (Shriberg et al., 1994; Smit et al., 1990) were presented in combinations of CV syllables using the /a/ vowel (/ba, ma, da, na/). The stimuli therefore limited the articulatory demands of the task. The children were directed to imitate the examiner s productions of the individual syllables prior to beginning the syllable repetition task to be certain that they could

71 63 produce the sounds. All participants were able to imitate the single syllables. The recorded audio stimuli for this task were spoken by an adult female. They were presented in free field via Power Point on a laptop, using external speakers to amplify the signal if necessary. When the stimuli provided for this task were presented, the letters (e.g., bada ) appeared on the screen in conjunction with the auditory stimuli; therefore, the children were instructed to turn away and/or close their eyes so they could not see the laptop screen. As in Shriberg et al. (2006), items were repeated if the child failed to respond or requested a repetition. The instructions were as follows: You are going to hear the computer speak some funny words. Just say exactly what you hear. If the computer says /ba/, you say. What if the computer says /da/? How about /na/? What about /ma/? Good. Now listen to the lady on the computer say these silly words, and say exactly what she says. Similar to Shriberg et al. (2006), if the child failed to respond within several seconds of the presentation of a stimulus, or requested a repetition (e.g., what? ), the stimulus was played again. Appendix E lists the stimulus items for this task. Audio recordings of each child s productions were reviewed using Praat software and were phonetically transcribed by the author. Scoring procedures followed those outlined by Shriberg et al. (2006), and additional detail about scoring procedures was provided by the first author of the original work. Each consonant was scored as correct or incorrect, ignoring distortion errors. If the wrong number of syllables was produced, the child s productions were aligned with the target response to provide the highest score for that item. For example, if the child produced [bada] for the target [mabada], the child was given credit for two of three consonants in the word ([ba] and [da]). There were 50

72 64 total consonants in the stimuli, and percent consonants correct-revised (PCC-R, Shriberg et al., 1997a) was calculated from this task for each child. Phonological Retrieval/Rapid Naming A third domain of phonological processing is phonological retrieval. As in other studies, Rapid Naming (RN) tasks were used to assess retrieval of phonological forms. Because there is reason to believe that speed of recall may be influenced by the number of syllables (Preston & Edwards, 2006), two RN tasks were used: one with monosyllables and one with disyllables. For both tasks, stimuli were presented as color pictures on 8 ½ x 13 legal-sized paper. The pictures were arranged in five rows of six pictures (30 pictures). The two tasks (monosyllables and disyllables) were presented consecutively, but the order of the two RN tasks was randomly chosen for each participant. Appendix F has the RN stimuli. Children were first familiarized with the Rapid Naming paradigm and were briefly trained. The children named each of the four pictures (fish, cat, ball, book) and were given corrective feedback if they mislabeled them. Then, the children were shown the training page of the four color pictures repeated five times (20 pictures for the training). They were told, We are going to have a race to see how fast you can talk. The instructions were, You are going to name all of the pictures on this page as fast as you can. Start at the top and go through each row until you come to the end. Watch me do it first. During this familiarization trial, the examiner first modeled by naming the pictures quickly from left to right, then asked the child to do so. The examiner followed along by pointing to keep the child on the correct picture and to continue in a left-to-right

73 65 fashion. The instructions were repeated for the two experimental RN trials. The children were first familiarized with the individual pictures, and corrective feedback was given if a picture was mislabeled. The Monosyllabic RN task included colored pictures of the items dog, chair, hat, boat, and fire. Each picture appeared six times, for a total of 30 pictures (adapted from Torgesen & Wagner, 1997). The order of the five pictures differed in each of the six iterations. The Disyllabic RN task involved rapid naming of colored pictures of five twosyllable items: money, apple, finger, pencil, and table. Items were taken from the 3-4 year items from the PPVT-III and Expressive Vocabulary Test, as well as Carroll, Davies, and Richman (1971). All were two-syllable words with a trochaic (strong-weak) stress pattern. Each picture appeared six times, for a total of 30 pictures. The order of the five pictures differed from one iteration to the next. Digital sound files were used to score both of the Rapid Naming tasks. Acoustic waveforms were marked by the author using Praat, timing from the beginning of energy onset of the first word to the end of energy offset of the final word. If the child became distracted or went off task during the rapid naming (e.g., made a comment, asked a question, laughed), the duration of the off-task behavior was removed from the total naming time by subtracting the time from the beginning of the off-task behavior to the beginning of the naming of the next picture. This had to be done for five participants on the monosyllable RN task and nine participants on the disyllable RN task. For statistical analysis, the average z-score of the two RN tasks was used to summarize the construct of phonological retrieval.

74 66 Reliability of Measures Speech Production Reliability Analyzing the data derived from the picture naming task involved a two-step process. Each speech sample was narrowly phonetically transcribed; then the transcriptions were reviewed and coded for errors according to the scheme developed for this study (Appendix A). Therefore, reliability was obtained for both steps. A transcriber with more than 30 years of experience with phonetic transcription of children s speech completed reliability for these speech production measures. The first reliability measure evaluated the reliability of the error coding scheme. The reliability judge reviewed the author s narrow transcription of a randomly-selected sample of at least 20 words from each participant. She used the error coding system to classify each speech sound error (Appendix A). Word-by-word agreement was computed, scoring agree if the initial rater and the reliability judge completely agreed on the number of distortions, typical sound changes, and atypical sound changes in the word. Disagreements were reviewed and were used to further refine definitions of error patterns in Appendix A. Given the phonetic transcription of a child s speech, the two judges completely agreed on speech error coding of all of the sound changes in 834 of 903 words (92.4% of words; range % agreement on words from individual participants). Following adjustments to the coding system, 41 of those words that the two judges disagreed upon were independently coded a second time. Agreement was reached on 83% (34/41) of these words on which the judges had disagreed.

75 67 For the second reliability measure, the reliability judge independently transcribed 25 consecutive words of the 125 word speech sample from the picture naming task (20%) for 30 of the participants. This represents 14% of all words that were transcribed. The starting point for the 25 consecutive words was randomly chosen for each participant. The reliability judge then coded her transcriptions based on the definitions in Appendix A. This was a worst case scenario measure because differences in phonetic transcription could inherently result in different coding of speech sound changes. These 25 word samples ranged from consonants, depending on the specific words transcribed. For each 25 word sample, the number of distortion errors per consonant, typical sound changes per consonant, and atypical sound changes per consonant was computed. For these 30 participants, the mean (absolute) difference between the reliability judge s estimate and the original estimate for the 25 word sample was 2.7 atypical sound changes per consonant (SD 3.0, range 0-9.3), 3.6 typical sound changes per consonant (SD 4.7, range ), and 2.9 distortions per consonant (SD 2.8, range 0-8.0). The concordance correlation coefficient 3 was 0.73 for atypical errors, 0.94 for typical errors, and 0.73 for distortions. Syllable Repetition Task Reliability For 15 participants, productions elicited on the syllable repetition task (SRT) were independently transcribed by a trained research assistant, an undergraduate senior majoring in Communication Sciences and Disorders who had taken a course in applied phonetics, in which she learned phonetic transcription. Reliability was computed for 15 3 The concordance correlation coefficient (Lin, 1989) is similar to a Pearson s r but it provides an estimate of the departure of two ratings from exact agreement (i.e., 45 o line, or when both axes are an identical scale). Hence, it is a more conservative estimate of agreement than a Pearson s r.

76 68 participants by making correct/incorrect judgments on each consonant produced and computing a percent consonants correct (PCC) for the 50 consonants. SRT scores obtained by the reliability judge were within +/- 6% of the original estimate for all participants (mean difference 0.27%). There was no statistically significant difference between the original score and the score obtained by the reliability judge (t = 0.33, p = ), and the two scores were very highly correlated (r = 0.978, p< 0.001; concordance correlation coefficient 0.978). Rapid Naming Reliability For 14 participants, the durations for each of the two Rapid Naming (RN) tasks were independently re-timed by a trained research assistant using waveforms in Praat, as described above. The duration estimates between the two judges were very highly correlated for both the RN monosyllable task (r = 0.996; p < 0.001) and the RN disyllable task (r = 1.00; p < 0.001). The mean difference between the two judges in timing the RN monosyllable task was.04 sec (range of absolute differences sec). The mean difference between the two judges in timing the RN disyllable task was.01 sec (range of absolute differences sec). Paired t-tests revealed no statistically significant differences in the durations measured by the original measurement and the reliability judge for the RN monosyllable task (t = 0.11; p = 0.916) or the RN disyllable task (t = 0.14; p = 0.889). To ensure accuracy of the data, it was determined that a discrepancy in duration estimate of greater than +/- 0.5 sec would prompt a re-timing of the RN task. This was done for two participants on the RN monosyllable tasks and one participant on the RN disyllable tasks. In all three cases, the source of disagreement involved

77 69 measuring the duration of off-task behavior. The re-timing always agreed with one of the duration measures (the original or that of the reliability judge), so the retiming was used in the final data analysis. Data Analysis Statistics were computed using SPSS version 15.0 (SPSS, 2006). A correlational design was used to examine the concurrent relationship between measures of speech sound accuracy and phonological processing in children with SSD. Hierarchical multiple regression was used to evaluate the proportion of variance in phonological awareness that could be explained by speech sound errors. For all regressions, an alpha level of 0.05 was used as a guide for statistical significance testing. The study was designed to be able to predict variance in PA by detecting a change in R 2 (or R 2 ) of about 0.10 with power of approximately See Appendix H for a discussion of observed power.

78 70 III : RESULTS Summary of Speech Sound Production A primary goal of the present study was to evaluate appropriate methods of quantifying speech sound errors in children with SSD and to determine how those errors relate to phonological awareness (PA) skills. Speech sound accuracy scores were based on phonetic transcriptions of each child s consonant productions from the 125-item picture naming task. Percent Consonants Correct (PCC) was calculated for each child, with any phonemic change (substitution or omission) or clinical distortion being considered an error. Hence, each consonant was judged as correct or incorrect. The PCC scores for children in this study are shown in Table 4. Although normative data are not available for PCC in picture naming samples, they have been reported in connected speech samples. The mean in the present study is significantly lower than data reported elsewhere from connected speech samples in normally developing children and are near values reported for conversational samples from children with SSD (Campbell et al., 2007; Shriberg et al., 1997a; Shriberg & Kwiatkowski, 1982). The mean PCC is 4% lower than that reported by Bird and Bishop (1995) in a picture naming task with children with SSD who were, on average, 16 months older than the participants in this study. Wolk (1990) reported PCC on a similar picture naming task for 14 phonologically disordered children ages 4;2-5;11 (half of whom also stuttered); the mean PCC of the present study is 5.8% below the mean reported in that study. Therefore, the PCC scores appear to reflect a range of speech sound (in)accuracy and are consistent with values expected for children with speech sound disorders.

79 71 Table 4: Summary of speech sound (in)accuracy for 43 preschoolers with SSD Mean SD Range Percent Consonants Correct (PCC) Distortions per Consonant Typical Sound Changes per Cons Atypical Sound Changes per Cons From those same picture naming speech samples, all sound changes were also analyzed based on the three-category system described earlier: distortions, typical sound changes, and atypical sound changes. Many sound errors required more than one sound change (i.e., interactions) to explain the child s production (e.g., catch /kætȓ/ [dætȓ] requires both Velar Fronting and Initial Voicing to explain a single error /k/ [d]). Descriptive data are shown in Table 4. As expected, children produce significantly more typical sound changes per consonant than atypical sound changes per consonant. Distortions were produced relatively infrequently, as reported in other studies (Gruber, 1999). However, all children were found to produce at least some atypical sound changes. Higher values on the Typical Sound Changes per Consonant, Atypical Sound Changes per Consonant, and Distortions per Consonant indicate more errors and, therefore, less accurate speech production, while higher PCC values are indicative of greater speech sound accuracy (correct consonants). Therefore, one would expect PCC to be (negatively) related to these error types. 4 4 As described earlier, PCC is not simply a linear combination of the three error types, as PCC does not take into account the components/features of sound changes.

80 72 Table 5 reports correlations among the three categories of speech sound errors. Similar correlation matrices are not available from other studies, as this is the first study to numerically quantify all consonant errors according to this three-category system; however, these values are not unexpected. Typical and atypical sound changes were positively correlated (r = 0.344, p < 0.05), suggesting that children who have more atypical sound changes also have more typical sound changes. Distortions were negatively correlated with typical sound changes (r = , p <0.01). This is in accord with literature on speech development that has suggested that children may progress from making phonemic errors (substitutions and omissions) to distortion errors as their speech sound accuracy improves (Gruber, 1999). Hence, more distortions are associated with fewer typical sound changes. Table 5: Pearson s correlation coefficients (r) of speech sound error types Typical Changes Per Consonant Distortions Per Consonant Atypical Changes Per Consonant PCC (**) ** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed). Typical Sound Changes Per Consonant 0.302(*) (**) Distortions Per Consonant (**) 0.344(*) Summary of Phonological Awareness Table 6 summarizes the group performance of the preschoolers with SSD on the phonological processing tasks, which includes the four PA tasks: Rhyme Matching,

81 73 Onset Matching, Onset Segmentation and Matching, and Blending. As expected, there was a broad range in performance on the PA tasks among these 43 children with SSD. Therefore, there is interest in explaining this variability of PA skills. Table 6: Summary of the performance of 43 children on the phonological processing tasks Task Mean SD Range Rhyme Matching (out of 16 items) Onset Matching (out of 10 items) Onset Segmentation & Matching (out of 10 items) Blending (out of 12 items) Syllable Repetition (PCC-R) Monosyllable Rapid Naming (in seconds )* Disyllable Rapid Naming (in seconds) *One participant did not complete the Monosyllable Rapid Naming There was no evidence of floor or ceiling effects, indicating the appropriateness of these tasks for detecting differences in PA skills. This provides support for the use of these tasks with this age group, and indicates that they may be sensitive to differences in PA skills. The means and standard deviations are generally in agreement (within +/- 1 items correct) with those reported in other studies that used similar tasks with 4 to 6 year olds with SSD (Bird et al., 1995; Rvachew & Grawburg, 2006). All variables were normally distributed based on Kolmogorov-Smirnov tests for normality (all p s > 0.15) and visual inspection of histograms.

82 74 As shown in Table 7 and as anticipated from other studies, significant positive correlations were found among the phonological awareness variables. That is, children who performed relatively well on a given PA task were likely to perform relatively well on the other tasks. A more complete correlation matrix is available in Appendix G. Table 7: Pearson correlation coefficients (r) for the phonological awareness tasks for 43 children with speech sound disorders Onset Matching Onset Segmentation & Matching Blending Rhyme.621(**).508(**).356(*) Onset Matching.637(**).401(**) Onset Segmentation & Matching.490(**) ** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed) A Phonological Awareness (PA) composite score was calculated by using a Principal Component Analysis to summarize the four PA tasks. This is a multivariate technique used to derive a linear combination of several variables while retaining the maximum possible variance. Each child therefore ends up with a single composite score for PA (with a mean of 0 and SD of 1). For these data, the principal component derived from the four PA tasks retained 63% of the variance of the tasks. The factor loading/pearson s correlation coefficient of each PA task with the overall phonological awareness principal component is summarized in Table 8, along with the communality (or the proportion of variance of a variable that is retained in the principal component).

83 75 It can be seen in Table 7 and Table 8 that the Blending task has the lowest correlations with the other three PA tasks, and also has the lowest correlation with the overall PA composite. Blending, therefore, may require somewhat different skills or have demands that differ from the other tasks (e.g., memory and synthesis, see Discussion). It is also possible that performance on this task was more variable because it could be more highly influenced by guessing; that is, each item had only three picture choices, compared to the other PA tasks (Rhyme, Onset Matching, Onset Segmentation & Matching) which had four. Hence, 33% correct on the Blending task was equivalent to random guessing, whereas 25% correct on the other tasks was equivalent to random guessing (see Appendix H: Measurement Issues for further discussion). Table 8: Principal Component Analysis summary derived from the four Phonological Awareness tasks Correlation with Principal Communality * Task Component Rhyme Onset Matching Onset Segmentation & Matching Blending Total 0.63 * The communality is the proportion of variance of that task that is retained in the PA Principal Component.

84 76 Hypothesis 1 Hypothesis 1 was that a relationship would be found between PA and speech sound error types. It was believed that PA would be most strongly predicted by the speech error category that represented the weakest phonological representations (Atypical Sound Changes per Consonant). Correlational analyses and visual inspection of scatterplots between the PA composite and speech production variables (Figure 4) showed that there was very little relationship between PA and distortions (bottom plot of Figure 4), and very little relationship between PA and typical sound changes (middle plot of Figure 4); both of these correlations were not statistically significant (p s > 0.05). However, a significant relationship was found between PA and Atypical Sound Changes per Consonant (r = , p = 0.009; top plot of Figure 4). That is, atypical sound changes predicted about 13% of the variance in PA. As anticipated, the negative correlation indicates that children with more atypical sound changes performed more poorly on the PA tasks. This supports the hypothesis that atypical speech errors are related to poor PA, presumably because both reflect weak phonological representations. Hypothesis 2 As reported earlier, PA skills are often related to vocabulary and age, and this study examined the extent to which PA variance can be predicted by speech sound errors when vocabulary and age are taken into account. To address this question, hierarchical multiple regression was used (Table 9). See Appendix I for regression diagnostics 5. 5 Briefly, there were no significant violations of the assumptions of normal distribution of the variables or the residuals; the interaction terms did not account for significant variance in the model; there were no cases with standardized residuals more than 2.0 SD from the mean; tolerance statistics were high, indicating that multicollinearity is not a significant concern.

85 77 Figure 4: Scatterplots of speech sound production error types and phonological awareness composite (principal component) Typical Changes Per Consonant Atypical Changes Per Consonant

86 78 Adj Table 9: Hierarchical regression used to predict PA Principal Component Model b Var. Method (SE) β Sig. F df p R 2 R 2 1 PPVT4 Enter.044 (.010) , Age Enter.058 (.023) PPVT4.040 (.010) Age.060 (.022) ATYP Stepwise (2.799) F df p R 2 Adj , R 2 Not in the Equation Sig. Distortions.557 Typical Errors.247 Total R 2 = Total Adjusted R 2 =0.392 Notes: b = Unstandardized coefficient (an estimate of the change in the PA Principal Component score for each 1-unit change in that variable; Keith, 2006); SE = standard error of the regression coefficient; β = standardized coefficient (coefficient when all variables are expressed in standardized [z-score] form; SPSS, 2006); R 2 = the variance explained in PA by the variables in the model (Keith, 2006); Adj. R 2 = Adjusted R 2 (an attempt to correct R 2 to more closely reflect the fit of the model to the population; SPSS, 2006); PPVT4 = Standard score of the Peabody Picture Vocabulary Test-4; ATYP = Atypical sound changes per consonant As in other studies, receptive vocabulary was correlated with PA, and in this study vocabulary accounted for about 27% of the variance in the PA composite (r = 0.517, p < 0.001). The bivariate relationship between age and the PA composite was not

87 79 statistically significant (r = 0.264, p > 0.05), possibly due to the restricted age range of the participants in this study (21 months). However, when included in a model with other variables, age is a significant predictor of the variance in PA (see below). In the first step of the regression, age (in months) and receptive vocabulary (PPVT-4 standard score) were used to predict PA. These two variables accounted for about 33.3% of the variance in PA (F [2, 40] = 11.5, p < 0.001, R 2 = 0.365, Adjusted R 2 = 0.333), and both are statistically significant predictors of PA (p <0.05). In the second step, the three speech production variables were tested in the model using stepwise entry (adding variables to the equation based on those with the smallest probability of F, if the probability is small enough) 6. Atypical Sound Changes per Consonant was the only significant speech production variable selected into the equation, and vocabulary and age remained significant predictors of PA as well. The new model accounted for additional variance in PA ( R 2 = 0.070, adjusted R 2 = 0.059, p = 0.033). Therefore, Hypothesis 2 was confirmed: atypical sound changes predicted approximately 5.9% of the variance in PA beyond what was already accounted for by vocabulary and age. Figure 5 displays a scatterplot of the observed PA values (those achieved by the participants) and the PA values predicted by the regression (age, receptive vocabulary, and atypical errors). 6 Stepwise entry was chosen for the second step of the regression (as opposed to forcing the three speech variables into the equation together) because it was presumed that some of the variables would not be related to PA. Therefore, only those speech production variables that contribute to the prediction of PA variance would be chosen. That is, the goal is to determine if certain variables are more robust predictors of PA, not to determine if all three speech variables together are robust predictors.

88 80 Figure 5: Observed PA Principal Component scores and PA scores predicted by the regression (age, vocabulary, atypical sound changes) for the 43 children with SSD Hypothesis 3 Because it was found that atypical sound changes predict significant variance in PA beyond variance explained by vocabulary and age, it is of interest to determine whether a similar result would be found using PCC (which does not distinguish between types of incorrect productions) as the speech sound accuracy variable (Hypothesis 3). The bivariate correlation between PCC and the PA composite was not statistically significant (r = 0.222, p = 0.153). However, a similar hierarchical multiple regression was performed to predict PA, with PCC forced to enter in the second step, after vocabulary and age. The results (Table 10) indicate that PCC does not explain any variance in PA beyond receptive vocabulary and age in these 43 children with SSD ( R 2 = 0.000, p = 0.923). Therefore, the speech analysis based on presumed reflection of phonological representations appears to provide a better explanation for the relationship between PA and speech sound production than does the analysis using PCC.

89 81 Table 10: Regression using PCC as the speech production variable to predict PA Model b Adj Var. Method (SE) β Sig. F df p R 2 R 2 1 PPVT4 Enter.044 (.010) , Age Enter.058 (.023) PPVT4.044 (.011) Age.058 (.024) PCC Enter.001 (.012) F df p R 2 Adj R , Notes: b = Unstandardized coefficient (an estimate of the change in the PA Principal Component for each 1-unit change in that variable; Keith, 2006); SE = standard error of the regression coefficient; β = standardized coefficient (coefficient when all variables are expressed in standardized [z-score] form; SPSS, 2006); R 2 = the variance explained in PA by the variables in the model (Keith, 2006); Adj. R 2 = Adjusted R 2 (an attempt to correct R 2 to more closely reflect the fit of the model to the population; SPSS, 2006); PPVT4 = Standard score of the Peabody Picture Vocabulary Test-4. Exploratory Hypotheses This study also investigated how phonological memory and phonological retrieval/rapid naming skills are related to speech sound errors. The above regressions were therefore repeated to explain variance in phonological memory (with scores on the Syllable Repetition Task as the dependent variable; Hypothesis 4) and to explain variance in rapid naming (with the average Z score on the two rapid naming tasks as the dependent variable; Hypothesis 5).

90 82 (Exploratory) Hypothesis 4 Syllable Repetition Task was strongly correlated with Atypical Sound changes per Consonant (r = , p <0.001), weakly correlated with Typical Sound Changes per Consonants (r = , p = 0.026), and not significantly correlated with Distortions per Consonant (r = 0.031, p = 0.842). The regression analysis then evaluated whether variance in phonological memory (as assessed by the Syllable Repetition Task) could be explained by types of speech sound errors (Hypothesis 4). Results of the regression are shown in Table 11. The initial model, which includes receptive vocabulary and age, does not predict a statistically significant amount of variance in phonological memory (F [2, 40] = 2.25, p = 0.118; R 2 = 0.101; adjusted R 2 = 0.056). This is somewhat unexpected, as it is in contrast to other studies that have shown nonword repetition to correlate with vocabulary skills and age (e.g., Edwards et al., 2004; Metsala, 1999). When the three speech variables are added in the next step, the overall regression model becomes significant (F [1, 39] = 22.6, p <0.001,R 2 =0.409, Adjusted R 2 = 0.364). Atypical Sound Changes per Consonant becomes the only significant predictor of variance in phonological memory (p < 0.001). Atypical changes explain about 30.8% of the unique variance in the PA composite ( R 2 = 0.308, adjusted R 2 = 0.308) and age and vocabulary remain nonsignificant predictors. Typical Sound Changes per Consonant and Distortions per Consonant are not selected for entry by the stepwise method (p > 0.05). Therefore, as was the case with PA, atypical speech errors explain a significant amount of the variance (30.8%) in phonological memory. This confirms Hypothesis 4.

91 83 Table 11: Regression explaining variance in Phonological Memory (Syllable Repetition Task) Model Var. Method b (SE) β Sig. F df p R 2 Adj R 2 1 PPVT4 Enter , (.190) Age Enter (.431) PPVT4.232 (.160) Age (.354) ATYP Stepwise (55.3) F df p R 2 Adj , R 2 Not in the Equation Sig. Distortions.416 Typical Errors.507 Total R 2 =0.409 Total Adjusted R 2 =0.364 Notes: b = Unstandardized coefficient (an estimate of the change in the syllable repetition task for each 1-unit change in that variable; Keith, 2006); SE = standard error of the regression coefficient; β = standardized coefficient (coefficient when all variables are expressed in standardized [z-score] form; SPSS, 2006); R 2 = the variance explained in the syllable repetition task by the variables in the model (Keith, 2006); Adj. R 2 = Adjusted R 2 (an attempt to correct R 2 to more closely reflect the fit of the model to the population; SPSS, 2006); PPVT4 = Standard score of the Peabody Picture Vocabulary Test-4; ATYP = Atypical sound changes per consonant

92 84 (Exploratory) Hypothesis 5 The two RN tasks, which were moderately correlated (r = 0.55, p < 0.001), were combined using average z-scores. A final hierarchical regression was run to predict variance in RN, and the results are shown in Table 12. Age and receptive vocabulary did not predict significant variance in the RN composite (F [2, 39] = 2.15, p = 0.130, Adjusted R 2 =0.053). Next, the three speech production variables (Distortions per Consonant, Typical Sound Changes per Consonant, Atypical Sound Changes per Consonant) were tested in the regression model, using stepwise entry, to determine if types of speech sound production errors could explain performance in rapid naming. Note that the bivariate correlations between RN and all three speech production variables were nonsignificant, but they were entered into the equation to address the theoretical question and to determine if any unique variance in RN could be explained. The overall regression model that includes atypical sound changes, age, and receptive vocabulary is significant (F [1, 38] = 7.71, p = 0.026; R 2 = 0.213; adjusted R 2 = 0.151). Atypical Sound Changes Per Consonant was the only speech production variable that explained significant variance in the RN composite (p = 0.024), and the result was an increase in adjusted R 2 of 9.9%. When Atypical Sound Changes per Consonant is added, receptive vocabulary becomes a significant predictor of RN as well, and the model accounts for about 15.2% of the variance in Rapid Naming. Therefore, atypical sound changes predict significant variance in RN beyond age and vocabulary, confirming Hypothesis 5.

93 85 Table 12: Regression explaining variance in Rapid Naming (average z scores of two Rapid Naming tasks) 7 Model Var. Method B (SE) β Sig. F df p R 2 Adj R 2 1 PPVT4 Enter , (.011) Age Enter (.025) PPVT (.011) Age (.024) ATYP Stepwise (2.924) F df p R 2 Adj R , Not in the Equation Sig. Distortions.783 Typical Errors.719 Total R 2 =0.213 Total Adjusted R 2 =0.152 Notes: b = Unstandardized coefficient (an estimate of the change in the rapid naming average z- score for each 1-unit change in that variable; Keith, 2006); SE = standard error of the regression coefficient; β = standardized coefficient (coefficient when all variables are expressed in standardized [z-score] form; SPSS, 2006); R 2 = the variance explained in rapid naming by the variables in the model (Keith, 2006); Adj. R 2 = Adjusted R 2 (an attempt to correct R 2 to more closely reflect the fit of the model to the population; SPSS, 2006); PPVT4 = Standard score of the Peabody Picture Vocabulary Test-4; ATYP = Atypical sound changes per consonant 7 Because one participant (P20) did not complete the Monosyllable Rapid Naming task, her data are not included in the regression reported. However, the regression was run again with her included (using the Z score from the disyllabic Rapid Naming task that she did complete) and the conclusions were unchanged.

94 86 Summary The results presented above generally confirm the five hypotheses. Atypical sound changes are significantly correlated with PA, but other speech sound error types (distortions and typical sound changes) are not significantly related to PA. Atypical sound changes account for a significant amount of variance in PA (about 5.9%) above and beyond the variance explained by receptive vocabulary and age. The model that categorizes speech errors into three types (based on the presumed accuracy of phonological representations) was found to be better in explaining variance in PA, compared to the model which considers all consonant errors equally (PCC). Finally, atypical sound changes help to explain significant variance in phonological memory and phonological retrieval/rapid naming skills in children with speech sound disorders. In all of the models, more atypical errors are associated with poorer performance on phonological processing tasks.

95 87 IV : DISCUSSION In this study, the relationship between speech sound errors and three domains of phonological processing (phonological awareness, phonological retrieval/rapid naming, and phonological memory) was assessed. Phonological processing skills, which are often reported to be weak in poor readers, are also weak in some children with SSD. The fact that there is often a wide range of performance on phonological processing tasks by children with SSD was confirmed in the present study. The variability in PA found in this study could be accounted for, in part, by vocabulary skills and age (about 33%). Both of these factors have been discussed in the past as contributing to the development of PA and also to the accuracy of phonological representations. However, as with prior studies, there remained much unexplained variance in the performance of the children in this study on phonological awareness tasks. One additional consideration, therefore, was that speech sound production, which is also thought to rely, in part, on phonological representations, could predict performance on phonological processing tasks. That is, certain types of speech sound errors may be indicative of poorly specified or inaccurate phonological representations, and therefore these errors may be related to a child s performance on PA tasks. The results of this study confirmed that prediction: a measure thought to reflect poorly specified phonological representations in speech sound production, the number of atypical sound changes per consonant, was found to account for significant variance in PA. The variance accounted for was above and beyond any variance explained by vocabulary and age. However, no additional variance was explained in PA when Percent Consonants

96 88 Correct (PCC) was used to measure speech sound accuracy, suggesting that PCC may not be sensitive to variation in PA skills. It was also found that variance in two other phonological processing domains, phonological memory and phonological retrieval, could be explained in part by atypical sound changes. Therefore, models intended to account for phonological processing performance in children with SSD can be informed by these findings. Further discussion is provided below regarding types of consonant errors, each domain of phonological processing (phonological awareness, phonological memory, and phonological retrieval), and the presumed theoretical link to phonological representations. Consonant Error Types Percent Consonants Correct (PCC), which weights all speech sound errors equally, was not found to be significantly correlated with the PA composite score, and it did not account for any variance in PA beyond age and receptive vocabulary. This finding is generally consistent with prior research and supports Hypothesis 3. It suggests that PCC may not be a sensitive indicator of the relationship between speech sound errors and PA. However, as discussed in Appendix H, larger samples would be required to have adequate power to reject the use of PCC in predicting variance in PA. One of the unique features of the current study is that it attempts to provide a more complete explanation of the component feature changes involved in children s speech sound errors than has been done in the past. Whereas PCC simply considers all speech sound errors as equal, the current study calls upon phonetically-motivated explanations of how those errors could be derived. That is, the three-category system

97 89 was designed to more fully account for all of the features of a child s errors. As expected, distortion errors, which involve phonetic (within-phoneme) changes, were unrelated to performance on any of the phonological processing tasks. This is in line with previous findings and supports the notion that these phonetic variations are not indicative of poorly specified phonological representations (Preston & Edwards, 2007; Rvachew et al., 2007; Shriberg et al., 2005). Typical sound changes were also found not to be correlated with phonological processing skills in this study. While it was predicted that these errors might have a moderate correlation with phonological processing skills, the correlations were low and not statistically significant. Thus, it appears that the occurrence of typical sound changes provides little information regarding a child s phonological processing skills. In contrast, atypical sound changes were found to account for significant variance in all three domains of phonological processing. The primary analysis was intended to predict variance in phonological awareness (PA); atypical sound changes predicted about 13% of the variance in PA, and about 5.9% of the unique variance in PA when controlling for age and receptive vocabulary. While this is not necessarily a robust explanation of the variance in PA skills, it may be indicative of a shared phonological deficit, speculated here to be weak underlying phonological representations. That is, children with SSD who use unusual sounds changes to produce words may also have trouble attending to the sound features of words in tasks such as rhyming, initial consonant matching, and blending. Children who use more of these atypical sound changes also tend to be less accurate in syllable repetition and slower on rapid naming tasks. Thus, the fact that this measure accounts for significant variance in three separate

98 90 domains of phonological processing provides support for this type of analysis of a child s speech sound errors. To avoid overestimation of what was considered atypical, the classification scheme developed for this study was relatively conservative. The decision was made a priori to score errors in a manner that would give children the most possible credit (i.e., counting the smallest number of errors possible, and considering typical errors rather than atypical errors when alternate accounts were possible, as described in Appendix B). However, even with this conservative estimate, all participants were found to have at least a few occurrences of atypical errors in their speech samples. It is acknowledged that defining atypical errors differently could result in different findings, and other ways of analyzing speech sound errors could yield different results. For example, the investigation by Rvachew et al. (2007) reported no significant relationship between atypical sound changes and phonological processing in preschoolers with SSD. One explanation might be differences in statistical techniques (e.g., the regression used in the current investigation, vs. the t-tests used by Rvachew et al. to compare two groups of children with normal and delayed PA). An alternate explanation is that the speech error coding scheme developed for this study was more fine-grained and took phonetic plausibility and the effects of nearby sounds into consideration when trying to logically account for errors. For example, Rvachew et al. considered /d/ [g] to be atypical regardless of phonetic context, whereas the current investigation counted that change as typical if it could be accounted for by the typical sound change of velar assimilation. For example, if pudding is produced as [pțgǻŋ], /d/ [g] is accounted for by velar assimilation, with the /d/ taking on the back feature of the /ŋ/. Thus, the notions of

99 91 phonetic plausibility and component sound changes that were applied in this study were less apparent in the work by Rvachew et al. (2007). Phonological Awareness As expected, variance in phonological awareness (PA) could be predicted, in part, by vocabulary and age in preschoolers with SSD. Also as hypothesized, additional variance in PA was explained by atypical sound changes, a speech production variable thought to be associated with weak phonological representations. The ability to develop accurate or refined phonological representations for PA tasks (to the point that they may be used for comparing and contrasting initial consonants and rhymes of words) had a modest (but significant) negative relationship with the production of atypical sound changes. The primary impact of low PA is likely to be on early decoding and spelling. That is, if children do not have clearly defined representations for the essential sound features of words, they may have difficulty using phonological information for sounding out (decoding) words and spelling. It could be speculated that imprecise phonological representations might additionally hinder the ability to associate an orthographic symbol with a phoneme. However, this hypothesis was not tested, as orthographic knowledge and sound-symbol associations were not assessed. Appendix J provides further discussion of this issue. There is mounting evidence that children who enter kindergarten with a SSD and weak PA skills are at particular risk for early literacy problems (Bird et al., 1995; Nathan et al., 2004). Thus, the results of this study could have diagnostic significance. Early identification of PA problems is essential for early intervention to take place. Clinically,

100 92 PA assessments are not routine for all children with SSD. This study provides support for the notion that children with numerous atypical sound changes will be at an elevated risk for PA problems and should therefore be assessed in that domain. Phonological Retrieval/Rapid Naming The finding that atypical sound changes account for 4.6% of the unique variance in rapid naming (beyond the contribution of age and vocabulary) provides tentative support for the notion that problems with quickly retrieving phonological representations are related to the production of more atypical sound changes. The implications are that children who produce more atypical speech sound errors may be at added risk for literacy difficulties, as rapid naming tasks have been found to relate to reading fluency, spelling, and decoding (Allor, 2002; Kamhi et al., 1988; Kirby et al., 2003; Wolf & Bowers, 1999; Wolf et al., 2002). It should be noted that rapid naming tasks are not frequently used with preschoolers, and the predictive validity of the specific rapid naming tasks used in this study has not been investigated. As reported earlier, some of the preschoolers found it difficult to attend (uninterrupted) to a series of 30 pictures. Hence, other processes beyond phonological retrieval are clearly involved in rapid naming (attention, visual recognition, inhibition of recently retrieved words, etc., Wolf & Bowers, 1999). These other processes may account for the relatively weak association between speech sound errors and rapid naming in the present study. As expected, it took most participants longer to rapidly name 30 disyllabic words than 30 monosyllabic words. The data from this study could now be compared to the

101 93 naming abilities of preschoolers without SSD to determine if naming of disyllabic words takes disproportionately longer for children with SSD (as noted for adolescents in Preston & Edwards, 2006). Phonological Memory One of the interesting relationships found in the exploratory analyses in this study was that atypical sound changes showed a relatively strong correlation (r = ) with the measure of phonological memory used here and contributed a relatively large proportion of variance to phonological memory beyond age and vocabulary (30.8%). Because performance on repetition tasks has been strongly tied to language and literacy performance (Brady, 1991; Dollaghan & Campbell, 1998; Metsala, 1999; Munson et al., 2005), this finding has significant implications for identification of children at risk for literacy problems. As explained below (Speculations on Phonological Representations), it is possible that poor phonological memory is causally connected with both atypical sound changes and poorly specified phonological representations. The predictive value of this syllable repetition task in children with SSD should be investigated to determine if growth in speech sound accuracy and/or PA development over time could be predicted by performance on this task. In this study, the ability to repeat syllables was unrelated to age and receptive vocabulary. This in contrast to previous studies that evaluated nonword repetition in preschoolers (Edwards et al., 2004; Metsala, 1999; Roy & Chiat, 2004). It is unclear why this might be, although one possible explanation could be the restricted age (4;0-5;9) and PPVT-4 standard scores (>84) in this study. An additional possibility is that previous

102 94 reports of nonword repetition have utilized stimuli that require more complex articulatory demands. Those stimuli may have been more sensitive to age than the stimuli in this study, which utilized only four early developing consonants paired with the same vowel. Clinical Implications As expected, there were several children with SSD who performed quite well on the phonological processing tasks. Thus, as reported in prior research, not all children with SSD are necessarily at risk for literacy problems. Based on the results of this study, the use of atypical sound changes can be considered an indicator of weak phonological processing skills and, in particular, poor phonological memory. It could be argued, then, that intervention or treatment that focuses on speech sound production and phonological processing should be implemented for children who exhibit atypical sound changes (cf. Gillon, 2005), perhaps targeting atypical errors. There are few studies investigating treatment of children with atypical sound changes, but those that exist suggest that these errors can be improved with standard phonological treatment techniques, such as minimal pair intervention and facilitating contexts (Dodd & Iacano, 1989; Leonard & Brown, 1984; Stringfellow & McLeod, 1994). Given the relatively strong relationship between atypical sound changes and phonological memory, we might speculate about whether treatment directed at the improvement of phonological memory would have any impact on speech sound production. That is, children who can better recall from working memory the phonological features they just heard might be better able to store accurate phonological representations. However, there is a lack of research addressing the question of whether

103 95 phonological memory can be improved by intervention in individuals with poor nonword repetition skills. Although phoneme-based interventions have shown some success in improving phonological memory in aphasics (Kendall et al., 2008), the training of working phonological memory in children with SSD seems to be an area in need of further investigation. Caveats and Limitations Interpretation of Results Several caveats related to the findings of the current study should be noted. For example, given the sample size (n = 43), the confidence interval around the model R 2 in the prediction of PA is relatively large (R 2 = 0.435, 95% CI = ). Replication of these results in other samples will help to clarify the effect size and the strength of the relationship between PA and types of speech sound errors. Additionally, in the analysis of behavioral data, there remains debate as to how best to interpret the size of R 2 change (Keith, 2006). The amount of variance in PA that is explained by adding Atypical Sound Changes per Consonant to the equation is relatively modest ( R 2 adjusted = 0.059). Thus, in comparison to the other variables (particularly receptive vocabulary), this does not appear to be a large effect. However, because a significant amount of additional variance can be accounted for by adding Atypical Sound Changes and because there is theoretical reason to include this variable (i.e., it is thought to be indicative of weak phonological representations), this suggests that the model is useful in explaining variance in PA for children with SSD.

104 96 Nevertheless, to keep the findings of the current study in perspective, it is important to consider the relative strength of the relationships. The bivariate correlations, as well as the size of the standardized coefficients in the regression, suggest that PA is more strongly related to receptive vocabulary than to speech sound errors. This may be because vocabulary is thought to be causally related to the development of accurate phonological representations. That is, increases in vocabulary size might result in refinement of phonological representations. In contrast, some speech sound errors might be considered a result of inaccurate phonological representations. This remains a matter of theoretical speculation, as there is no direct way to evaluate the (in)accuracy of a child s phonological representations. Additionally, the results should be interpreted within the scope of the participant characteristics (e.g., primarily middle class, monolingual English-speaking children with idiopathic SSD). Thus, the results may not be applicable to all children with phonological processing difficulties. For example, many children have problems with PA but do not have speech sound production difficulties. Therefore, it is unlikely that any additional variance in the phonological processing skills could be explained by atypical errors in children without SSD, as they, by definition, rarely (if ever) exhibit atypical sound changes. Caveats on Speech Sound Errors As discussed earlier, debate exists concerning how to categorize speech sound errors, and particularly which errors should be considered atypical. While attempts were made to consult the literature regarding such sound changes, relevant literature was

105 97 sometimes absent or contradictory. Thus, other researchers might reach slightly different conclusions about which sound changes should be considered atypical. However, the error coding system was intended to be comprehensive and replicable and was based on extant literature and the notion of phonetic plausibility. Therefore, is believed to be a valid system for categorizing consonant errors. The method of quantifying speech sound accuracy in this study, while theoretically motivated, is not the only method for analyzing speech sound production skills. Other transcription-based methods exist for examining speech production output, although they have not consistently revealed a relationship between speech production and phonological processing. For example, phonological processing skills have been found to be unrelated to phonological features (e.g., sonorant, labial, nasal, etc.; Rvachew et al., 2007) and some standardized tests of speech sound accuracy (e.g., Larivee & Catts, 1999). Therefore, it is possible that transcription-based methods might not be highly sensitive to subtleties in speech sound production that relate to phonological processing and/or phonological representations. Future studies could implement instrumental analysis of phonetic output, including segmental and suprasegmental analysis. For example, possible speech-related predictors of phonological processing skills might include subtle acoustic features such as voice onset time (Tyler et al., 1990) and vowel formants (cf. Elbro et al., 1998), or prosodic characteristics such as lexical stress (Shriberg et al., 2003a; Shriberg et al., 2003b) and speaking rate (Smith et al., 2006). Vowels. This study did not analyze vowel production errors, in part because vowel accuracy is generally thought to develop earlier than consonant accuracy (Lowe, 1994, but see Pollock, 1991) and because vowel errors are less often discussed as a

106 98 characteristic of speech sound disorders. One study with Danish children suggested that the quality of vowel productions may be related to later literacy achievement in kindergarteners (Elbro et al., 1998). However, because the current study attempted to remedy some of the limitations associated with the Percent Consonants Correct measure, vowel errors were not analyzed here. Vowels were transcribed and vowel errors did occur. Thus, data are available for exploratory analyses in the future. A measure that is capable of describing and adequately weighting both consonant and vowel errors might be the most comprehensive measure of phonological output accuracy (such a measure is currently being developed by the author). Subgroups. Although the current study examined types of errors and their frequency, it did not quantify the consistency with which errors occur. There has been some discussion that the consistency of errors across multiple attempts at the production of a word may be indicative of childhood apraxia of speech (e.g., producing "elephant" four different ways on four different attempts, Dodd, 1995). Children with a diagnosis of childhood apraxia of speech (CAS) have been found to have difficulty with phonological processing (Dodd, 1995; Lewis et al., 2004). However, this study did not take different subgroups of children with SSD of unknown origin into consideration. Therefore, children with suspected CAS were included but were not looked at separately. 8 Several classification systems exist for SSD, based on suspected etiology (Shriberg et al., 1997b), speech error patterns (Bradford & Dodd, 1996; Crary, 1984; Dodd, 2005; Gibbon, 1999), concomitant speech disorders (Wolk et al., 1993), or 8 Approximately half of the parents or referring clinicians of participants in this study indicated that CAS was diagnosed or suspected, which is well above prevalence data for CAS. However, this is consistent with the notion that CAS definitions are broad and that the disorder is often clinically over-diagnosed (American-Speech-Language-Hearing-Association, 2007).

107 99 concomitant language disorders (Bird et al., 1995; Bishop & Adams, 1990; Leitao et al., 1997), etc., but there is poor consensus on how to differentially diagnose particular subtypes of SSD. Because none of these classification systems have robust empirical support, they were not utilized here. Although many of the children in this study probably fell into one or more of the subgroups described in the literature, looking at subgroups was not the focus of this study. Moreover, it is unclear how the results found here would be influenced by particular subgroups, as there is no clear description of the use of different types of sound errors by subgroups of children with SSD. Speculations on Phonological Representations As previously discussed, the results reported above may be accounted for in part by the accuracy of phonological representations. Phonological representations are thought to develop with vocabulary and age and to rely on a child s ability to extract and/or infer linguistically meaningful sound patterns in the speech signal. As vocabulary skills increase, children develop a broader variety of words from which to draw inferences about the essential phonological features of words (Fowler, 1991; Metsala, 1999). These inferences are thought to help children recognize underlying contrasts (e.g., voiced-voiceless, nasal-nonnasal, etc) and to recognize sound patterns and appropriate sound combinations in the adult language. Phonological representations might then become more specified and closer to the adult target as the child has more experience with a word and with similar-sounding words (Fowler, 1991). When children s ability to extract salient phonological features of words (and use them to form phonological

108 100 representations) is weak, their ability to recognize some of the salient phonological features (such as rhymes or initial consonants) might be weak as well. Some children with SSD, especially those with lower vocabularies and those who produce relatively more atypical sound changes, were found in this study to have greater difficulty on the PA tasks which required them to focus on the linguistically meaningful sound patterns of the speech signal (i.e., identify rhyme, initial consonants, etc.). For these children, salient features in speech sound production may be poorly represented, resulting in unusual productions of words (e.g., deletions of initial consonants, strong syllables, or unmarked members of consonant clusters). The current investigation provides support for the notion that phonological processing and types of speech sound errors are linked in preschoolers with SSD (Figure 1). The assumed link, though not directly tested here, is poorly specified ( weak ) phonological representations (cf. Swan & Goswami, 1997a). It remains unclear exactly why these representations are weak. Some possible explanations include: (a) Speech perception problem: poor detection or encoding of the phonetic features of the rapidly-changing speech signal (e.g., poor speech perception) may result in insufficient, incomplete, or inaccurately perceived information to store in phonological representations. Appendix K provides further detail related to this issue. (b) Storage problem: Some children with SSD may perceive speech signals accurately but have difficulty making the appropriate inferences about the phonetic or phonological components of words to store them correctly. (c) Phonological rehearsal problem. Articulatory rehearsal (e.g., the phonological loop ) which involves the ability to subvocally repeat/rehearse verbal

109 101 input to keep it in temporary storage for a longer time, has been discussed in models of working memory (Baddeley, 2003; Baddeley et al., 1998). There is evidence that this rehearsal may be impaired in some children with SSD (Bishop et al., 1990; Locke & Scott, 1979). Given the strength of the relationship between the phonological memory task and atypical sound changes found here, one hypothesis is that a limitation in the ability to accurately rehearse phonological information to form temporary phonological representations results in a lack of information available to form accurate long-term representations. That is, there may be a (covert) rehearsal mechanism that is impaired in some children with SSD, and this may impact the ability to accurately retain speechrelated information in working memory as well as form a long-term representation. Hence, the phonological memory/rehearsal deficit would be viewed as a causal factor in weak phonological representations. This would have implications for the ability to learn and store new phonological forms (cf. Sutherland & Gillon, 2005). Although the premise discussed thus far has been that phonological representations influence speech sound production, it is possible that a bidirectional relationship exists between phonological representations and speech sound production. This would mean that producing speech sounds correctly in words could reinforce adultlike representations, whereas producing speech sounds incorrectly could inhibit the development of adult-like phonological representation (Bishop et al., 1990; Nicolson et al., 2001). Such a view could be considered in accord with the Motor Theory of Speech Perception (Galantucci, Foweler, & Turvey, 2006).

110 102 Ideally, longitudinal follow-up of the children in this study would be helpful in determining if the tasks and speech production measures used here are able to predict long-term growth of phonological processing and literacy skills. For example, as kindergarteners, invented spelling tasks (which have often been discussed as relating to PA development in somewhat older children) could provide insight into these children s early knowledge of phonemes and phoneme-grapheme correspondence (e.g., Ball & Blachman, 1991). Additionally, it would be of interest to determine if periodic assessments would reveal concurrent reduction of atypical sound changes and improvement in phonological processing skills, both presumably due to the refinement of phonological representations. SIGNIFICANCE AND CONCLUSIONS This study evaluated the relationship between phonological processing and speech sound errors in children with speech sound disorders (SSD), while addressing some of the limitations of previous studies. The relative influence of different types of speech sound errors had not been well-explored in a systematic fashion. This study appears to be the first to address within-group variability in phonological processing through a measurement system that separates all consonant errors based on both types and frequency. A three-category scheme for coding speech sound errors was developed that accounted for the component features and changes in the children s sound errors, and it was found that atypical sound changes were better predictors of phonological processing than distortions and typical sound changes. Some of the limitations of previous studies, which formed discrete groups based on one measure of speech

111 103 production to predict variance in phonological processing, were addressed by using a regression technique. Additionally, the study explored phonological processing and speech sound production in preschoolers with SSD, a population that has not previously been studied in this way. This study has both clinical and theoretical importance, as it has helped to advance our understanding of how certain types of speech sound errors relate to specific phonological processing domains known to be related to early literacy. Atypical sound changes were found to predict unique variance in three phonological processing domains, whereas distortions and typical sound changes were not. Poorly specified phonological representations have been discussed as the link between phonological processing difficulties and some speech sound errors. This research suggests that more frequent use of atypical sound changes is related to greater risk of preliteracy problems (to the extent that they are tapped by these tasks) in children with SSD. This research provides important evidence in light of the critical age hypothesis for literacy development, which suggests that children who enter kindergarten with speech sound production problems and PA problems are at significant risk for literacy problems (Bird & Bishop, 1992; Nathan et al., 2004). Thus, it would be prudent for clinicians to consider the specific types of speech sound errors that reflect relatively greater risk for phonological processing (and, by extension, literacy) when evaluating and treating preschool children. Children with SSD who exhibit frequent atypical sound changes would be appropriate candidates for further evaluation of phonological processing. Therefore, it is hoped that this research may help to further our

112 104 understanding of which children are at particular risk for preliteracy and literacy problems so that early intervention can be implemented.

113 105 REFERENCES Allor, J. H. (2002). The relationships of phonemic awareness and rapid naming to reading development. Learning Disability Quarterly, 25, American-Speech-Language-Hearing-Association. (2007). Childhood Apraxia of Speech [Technical Report]. available from Aram, D. M., Ekelman, B. L., & Nation, J. E. (1984). Preschoolers with language disorders: 10 years later. Journal of Speech & Hearing Research, 27(2), ASHA Audiologic Assessment Panel (1997). Guidelines for audiological screening. Rockville, MD: Author. Baddeley, A. (2003). Working memory and language: An overview. Journal of Communication Disorders, 36, Baddeley, A., Gathercole, S., & Papagno, C. (1998). The phonological loop as a language learning device. Psychological Review, 105(1), Baguley, T. (2004). Understanding statistical power in the context of applied research. Applied Ergonomics, 35, Ball, E. W., & Blachman, B. A. (1991). Does phoneme awareness training in kindergarten make a difference in early word recognition and developmental spelling. Reading Research Quarterly, 26(1), Bird, J., & Bishop, D. (1992). Perception and awareness of phonemes in phonologically impaired children. European Journal of Disorders of Communication, 27,

114 106 Bird, J., Bishop, D. V. M., & Freeman, N. H. (1995). Phonological awareness and literacy development in children with expressive phonological impairments. Journal of Speech and Hearing Research, 38, Bishop, D. V., Brown, B. B., & Robson, J. (1990). The relationship between phoneme discrimination, speech production, and language comprehension in cerebralpalsied individuals. Journal of Speech and Hearing Research, 33(2), Bishop, D. V. M., & Adams, C. (1990). A prospective study of the relationship between specific language impairment, phonological disorders and reading retardation. Journal of Child Psychology and Psychiatry, 31(7), Bishop, D. V. M., & Clarkson, B. W. (2003). Written language as a window into residual language deficits: A study of children with persistent and residual speech and language impairments. Cortex, 39(2), Bradford, A., & Dodd, B. (1996). Do all speech-disordered children have motor deficits? Clinical Linguistics and Phonetics, 10(2), Bradley, L., & Bryant, P. E. (1983). Categorizing sounds and learning to read - a causal connection. Nature, 301, Brady, S. A. (1991). The role of memory in reading disability. In S. Brady & D. Shankweiler (Eds.), Phonological processes in literacy: A tribute to Isabelle Y. Liberman (pp ). Hillsdale, NJ: Lawrence Erlbaum Ass. Bridgeman, E., & Snowling, M. (1988). The perception of phoneme sequence: A comparison of dyspraxic and normal children. British Journal of Disorders of Communication, 23,

115 107 Burgess, S. R., & Lonigan, C. J. (1998). Bidirectional relations of phonological sensitivity and prereading abilities: Evidence from a preschool sample. Journal of Experimental Child Psychology, 70, Campbell, T. F., Dollaghan, C., Janosky, J. E., & Adelson, P. D. (2007). A performance curve for assessing change in Percentage of Consonants Correct Revised (PCC- R). Journal of Speech, Language, and Hearing Research, 50(4), Campbell, T. F., Dollaghan, C. A., Rockette, H. E., Paradise, J. L., Feldman, H. M., Shriberg, L. D., et al. (2003). Risk factors for speech delay of unknown origin in 3-year-old children. Child Development, 74(2), Campbell, T. F., & Shriberg, L. D. (1982). Associations among pragmatic functions, linguistic stress, and natural phonological processes in speech-delayed children. Journal of Speech and Hearing Research, 25, Caravolas, M., Hulme, C., & Snowling, M. J. (2001). The foundations of spelling ability: Evidence from a 3-year longitudinal study. Journal of Memory and Language, 45, Cardoso-Martins, C., & Pennington, B. F. (2004). The relationship between phoneme awareness and rapid serial naming skills and literacy acquisition: The role of developmental period and reading ability. Scientific Studies of Reading, 8(1), Carroll, J. B., Davies, P., & Richman, B. (1971). Word frequency book. Boston: Houghton Mifflin Co. Catts, H. W. (1991). Facilitating phonological awareness: Role of speech-language pathologists. Language, Speech and Hearing Services in Schools, 22,

116 108 Catts, H. W. (1993). The relationship between speech-language impairments and reading disabilities. Journal of Speech and Hearing Research, 36, Catts, H. W. (1997). The early identification of language-based reading disabilites. Language, Speech and Hearing Services in Schools, 28, Catts, H. W., Fey, M. E., Zhang, X., & Tomblin, J. B. (2001). Estimating the risk of future reading difficulties in kindergarten children: A research-based model and its clinical implementation. Language, Speech and Hearing Services in Schools, 32(1), Catts, H. W., Hu, C.-F., Larrivee, L., & Swank, L. (1994). Early identification of reading disabilities in children with speech-language impairments. In R. V. Watkins & M. L. Rice (Eds.), Specific language impairments in children (pp ). Baltimore, MD: Paul H. Brookes Publishing. Crary, M. A. (1984). Phonological characteristics of developmental verbal dyspraxia. Seminars in Speech and Language, 5(2), Denckla, M. B., & Rudel, R. G. (1976). Rapid 'automatized' naming (R.A.N.): Dyslexia differentiated from other learning disabilities. Neuropsychologia, 14, Dodd, B. (1995). The differential diagnosis and treatment of children with speech disorders. San Diego: Singular Publishing Group. Dodd, B. (Ed.). (2005). Differential diagnosis and treatment of children with speech disorder (2 ed.). Philadelphia: Whurr Publishers. Dodd, B., & Iacano, T. (1989). Phonological disorders in children: Changes in phonological process use during treatment. British Journal of Disorders of Communication, 24,

117 109 Dodd, B., Leahy, J., & Hambly, G. (1989). Phonological disorders in children: Underlying cognitive deficits. British Journal of Developmental Psychology, 7, Dollaghan, C., & Campbell, T. F. (1998). Nonword repetition and child language impairment. Journal of Speech, Language, and Hearing Research, 41, Dunn, L. M., & Dunn, D. M. (2007). Peabody Picture Vocabulary Test (4 ed.). Minneapolis, MN: Pearson. Dunn, L. M., & Dunn, L. M. (1981). Peabody Picture Vocabulary Test-Revised. Circle Pines, MN: American Guidance Service. Dworkin, J. P. (1980). Characteristics of frontal lispers clustered according to severity. Journal of Speech and Hearing Disorders, 35, Edwards, J., Beckman, M. E., & Munson, B. (2004). The interaction between vocabulary size and phonotactic probability effects on children s production accuracy and fluency in nonword repetition. Journal of Speech, Language, and Hearing Research, 47, Edwards, M. L. (1992). In support of phonological processes. Language, Speech and Hearing Services in Schools, 23, Edwards, M. L. (1995). Developmental Phonology. In H. Winitz (Ed.), Human communication and its disorders: A review (Vol. IV). Timonium, MD: York Press. Edwards, M. L., & Shriberg, L. D. (1983). Phonology: Applications in communicative disorders. San Diego, CA: College-Hill Press.

118 110 Elbro, C., Borstrom, I., & Peterson, D. K. (1998). Predicting dyslexia from kindergarten: The importance of distinctness of phonological representations of lexical items. Reading Research Quarterly, 33(1), Elliott, C. D. (1990). Differential Ability Scales. San Antonio, TX: Psychological Corporation/Harcourt Brace. Fletcher, S. G., Casteel, R. L., & Bradley, D. P. (1961). Tongue-thrust swallow, speech articulation, and age. Journal of Speech & Hearing Disorders 26, Fowler, A. E. (1991). How early phonological development might set the stage for phoneme awareness. In S. A. Brady & D. P. Shankweiler (Eds.), Phonological processes in literacy: A tribute to Isabelle Y. Liberman (pp ). Hillsdale, NJ: Lawrence Erlbaum Associates, Publishers. Foy, J. G., & Mann, V. (2001). Does strength of phonological representations predict phonological awareness in preschool children? Applied Psycholinguistics, 22(3), Galantucci, B., Fowler, C.A. & Turvery, M.T. (2006). The motor theory of speech perception reviewed. Psychonomic Bulletin & Review, 13 (3), Gibbon, F. E. (1999). Undifferentiated lingual gestures in children with articulation/phonological disorders. Journal of Speech, Language, and Hearing Research, 42, Gillon, G. T. (2000). The efficacy of phonological awareness intervention for children with spoken language impairment. Language, Speech, & Hearing Services in Schools, 31(2),

119 111 Gillon, G. T. (2005). Facilitating phoneme awareness development in 3- and 4-year-old children with speech impairment. Language, Speech, and Hearing Services in Schools, 36(4), Goldman, R., & Fristoe, M. (1986). Goldman-Fristoe Test of Articulation. Circle Pines, MN: American Guidance Service. Goldman, R., & Fristoe, M. (2000). Goldman Fristoe Test of Articulation - Second Ed. Circle Pines, MN: AGS. Griffiths, Y. M., & Snowling, M. J. (2002). Predictors of exception word and nonword reading in dyslexic children: The severity hypothesis. Journal of Educational Psychology, 94(1), Gruber, F. A. (1999). Probability estimates and paths to consonant normalization in children with speech delay. Journal of Speech, Language, and Hearing Research, 42, Hall, P. K. (1989). The occurrence of developmental apraxia of speech in a mild articulation disorder: A case study. Journal of Communication Disorders, 22, Hall, S. L., & Moats, L. C. (1999). Straight talk about reading: How parents can make a difference during the early years. Chicago, Ill: Contemporary Books. Hauner, K. K. Y., Shriberg, L. D., Kwiatkowski, J., & Allen, C. T. (2005). A subtype of speech delay associated with developmental psychosocial involvement. Journal of Speech, Language, and Hearing Research, 48, Hesketh, A., Adams, C., Nightingale, C., & Hall, R. (2000). Phonological awareness therapy and articulatory training approaches for children with phonological

120 112 disorders: A comparative outcome study. International Journal of Language and Communication Disorders, 35(3), Hodson, B., & Edwards, M. L. (1997). Perspectives in applied phonology. Gaithersburg, MD: Aspen Publishers, Inc. Ingram, D. (1976). Phonological disability in children. New York: Elsevier. International-Dyslexia-Association. (2000). Dyslexia Basics: Fact Sheet #62. Jamieson, D. G., & Rvachew, S. (1992). Remediating speech production errors with sound identification training. Journal of Speech Language Pathology and Audiology, 16(3), Justice, L. M., & Pullen, P. C. (2003). Promising interventions for promoting emergent literacy skills: Three evidence-based approaches. Topics in Early Childhood Special Education, 23(3), Kamhi, A. G., & Catts, H. W. (1986). Toward an understanding of developmental language and reading disorders. Journal of Speech & Hearing Disorders, 51(4), Kamhi, A. G., Catts, H. W., Mauer, D., Apel, K., & Gentry, B. F. (1988). Phonological and spatial processing abilities in language- and reading- impaired children. Journal of Speech and Hearing Disorders, 53, Kazdin, A. E. (2003). Research design in clinical psychology. Boston: Allyn and Bacon. Keith, T. Z. (2006). Multiple regression and beyond. Boston: Pearson Education. Kendall, D. L., Rosenbek, J. C., Heilman, K. M., Conway, T., Klenberg, K., Gonzalez Rothi, L. J., et al. (2008). Phoneme-based rehabilitation of anomia in aphasia. Brain and Language, 105(1), 1-17.

121 113 Khan, L. M. L. (1982). A review of 16 major phonological processes. Language, Speech and Hearing Services in Schools, 13, Kirby, J. R., Parrilla, R. K., & Pfeiffer, S. L. (2003). Naming speed and phonological awareness as predictors of reading development. Journal of Educational Psychology, 95(3), Klein, H. B., & Spector, C. C. (1985). Effect of syllable stress and serial position on error variability in polysyllabic productions of speech-delayed children. Journal of Speech and Hearing Disorders, 50(4), Larivee, L. S., & Catts, H. W. (1999). Early reading achievement in children with expressive phonological disorders. American Journal of Speech-Language Pathology, 8, Law, J., Garrett, Z., & Nye, C. (2004). The efficacy of treatment for children with developmental speech and language delay/disorder: A meta-analysis. Journal of Speech, Language, & Hearing Research, 47(4), Leitao, S., & Fletcher, J. (2004). Literacy outcomes for students with speech impairment: Long-term follow-up. International Journal of Language & Communication Disorders, 39(2), Leitao, S., Hogben, J., & Fletcher, J. (1997). Phonological processing skills in speech and language impaired children. European Journal of Disorders of Communication, 32, Leonard, L. B. (1985). Unusual and subtle phonological behavior in the speech of phonologically disordered children. Journal of Speech and Hearing Disorders, 50(1), 4-13.

122 114 Leonard, L. B., & Brown, B. L. (1984). Nature and boundaries of phonologic categories: A case study of an unusual phonologic pattern in a language-impaired child. Journal of Speech and Hearing Disorders, 49(4), Leonard, L. B., & McGregor, K. K. (1991). Unusual phonological patterns and their underlying representations: A case study. Journal of Child Language, 18, Lewis, B. A., & Freebairn, L. (1992). Residual effects of preschool phonology disorders in grade school, adolescence, and adulthood. Journal of Speech and Hearing Research, 35, Lewis, B. A., Freebairn, L. A., Hansen, A. J., Iyengar, S. K., & Taylor, H. G. (2004). School-age follow-up of children with childhood apraxia of speech. Language, Speech, and Hearing Services in Schools, 35, Lewis, B. A., Freebairn, L. A., & Taylor, H. (2002). Correlates of spelling abilities in children with early speech sound disorders. Reading and Writing, 15(3-4), Lewis, B. A., Shriberg, L. D., Freebairn, L. A., Hansen, A. J., Stein, C. M., Taylor, H. G., et al. (2006). The genetic bases of speech sound disorders: Evidence from spoken and written language. Journal of Speech, Language, & Hearing Research, 49(6), Liberman, I. Y., Shankweiler, D., Fischer, F. W., & Carter, B. (1974). Explicit syllable and phoneme segmentation in the young child. Journal of Experimental Child Psychology, 18,

123 115 Lieberman, P., Meskill, R. H., Chatillon, M., & Schupack, H. (1985). Phonetic Speech Perception Deficits in Dyslexia. Journal of Speech and Hearing Research, 28(4), Lin, L. I. (1989). A concordance correlation coefficient to evaluate reproducibility. Biometrics, 45(1), Locke, J. L., & Scott, K. K. (1979). Phonetically mediated recall in the phonetically disordered child. Journal of Communication Disorders, 12, Lowe, R. J. (1994). Phonology assessment and intervention applications in speech pathology. Baltimore, MD: Williams & Wilkins. McDowell, K. D., Lonigan, C. J., & Goldstein, H. (2007). Relations Among Socioeconomic Status, Age, and Predictors of Phonological Awareness. Journal of Speech, Language, and Hearing Research, 50(4), McGrath, L. M., Pennington, B. F., Willcutt, E. G., Boada, R., Shriberg, L. D., & Smith, S. D. (2007). Gene x Environment interactions in speech sound disorder predict language and preliteracy outcomes. Developmental Psychopathology, 19(4), Metsala, J. L. (1999). Young children's phonological awareness and nonword repetition as a function of vocabulary development. Journal of Educational Psychology, 91(1), 3. Mody, M., Studdert-Kennedy, M., & Brady, S. (1997). Speech perception deficits in poor readers: Auditory processing or phonological coding. Journal of Experimental Child Psychology, 64,

124 116 Munson, B., Edwards, J., & Beckman, M. E. (2005). Relationships between nonword repetition accuracy and other measures of linguistic development in children with phonological disorders. Journal of Speech, Language, & Hearing Research, 48(1), Nathan, L., Stackhouse, J., Goulandris, N., & Snowling, M. J. (2004). The development of early literacy skills among children with speech difficulties: A test of the "critical age hypothesis." Journal of Speech, Language, and Hearing Research, 47, National-Reading-Panel. (2000). Teaching children to read: An evidence-based assessment of the scientific research literature on reading and its implications for reading instruction. Washington, DC: National Institutes of Health. Nicolson, R. I., Fawcett, A. J., & Dean, P. (2001). Developmental dyslexia: the cerebellar deficit hypothesis. Trends in Neurosciences, 24(9), NIDCD. (2006, Jan 13). Statistics on voice, speech, and language. Retrieved October 12, 2006, 2006, from Nittrouer, S., & Burton, L. T. (2005). The role of early language experience in the development of speech perception and phonological processing abilities: Evidence from 5-year-olds with histories of otitis media with effusion and low socioeconomic status. Journal of Communication Disorders, 38, Ohde, R. N., & Sharf, D. J. (1988). Perceptual categorization and consistency of synthesized /r-w/ continua by adults, normal children and /r/-misarticulating children. Journal of Speech and Hearing Research, 31,

125 117 Oller, D. K., & Delgado, R. E. (2001). Logical International Phonetic Programs (LIPP). Miami, FL: Intelligent Hearing Systems. Oller, D. K., & Ramsdell, H. L. (2006). A weighted reliability measure for phonetic transcription. Journal of Speech, Language, & Hearing Research, 49(6), Pascoe, M., Stackhouse, J., & Wells, B. (2006). Persisting speech difficulties in children. Chichester, England: Wiley. Paul, R., & Shriberg, L. D. (1982). Associations between phonology and syntax in speech delayed children. Journal of Speech and Hearing Research, 25, Phillips, B. M., & Torgesen, J. K. (2006). Phonemic awareness and reading: Beyond the growth of initial reading accuracy. In D. K. Dickinson & S. B. Neuman (Eds.), Handbook of early literacy research (pp ). New York, NY: Guilford Press. Preston, J. L., & Edwards, M. L. (2006). Rapid naming abilities of adolescents with residual speech sound errors, American Speech-Language-Hearing Association Annual Convention. Miami, FL. Preston, J. L., & Edwards, M. L. (2007). Phonological processing skills of adolescents with residual speech sound errors. Language, Speech and Hearing Services in Schools, 38, Raitano, N. A., Pennington, B. F., Tunick, R. A., Boada, R., & Shriberg, L. D. (2004). Pre-literacy skills of subgroups of children with speech sound disorders. Journal of Child Psychology and Psychiatry, 45(4),

126 118 Richardson, U., Thomson, J. M., Scott, S. K., & Goswami, U. (2004). Auditory processing skills and phonological representation in dyslexic children. Dyslexia: An International Journal of Research and Practice Selected Papers from the 6th Conference of the British Dyslexia Association, 10(3), Roy, P., & Chiat, S. (2004). A prosodically controlled word and nonword repetition task for 2- to 4-year-olds: Evidence from typically developing children. Journal of Speech, Language, and Hearing Research, 47, Rvachew, S. (1994). Speech perception training can facilitate sound production learning. Journal of Speech and Hearing Research, 37, Rvachew, S. (2006). Longitudinal predictors of implicit phonological awareness skills. American Journal of Speech-Language Pathology, 15(2), Rvachew, S. (2007). Phonological processing and reading in children with speech sound disorders. American Journal of Speech-Language Pathology, 16(3), Rvachew, S., Chiang, P.-Y., & Evans, N. (2007). Characteristics of speech errors produced by children with and without delayed phonological awareness skills. Language, Speech and Hearing Services in Schools, 38(1), Rvachew, S., & Grawburg, M. (2006). Correlates of phonological awareness in preschoolers with speech sound disorders. Journal of Speech, Language, & Hearing Research, 49(1), Rvachew, S., Nowak, M., & Cloutier, G. (2004). Effect of phonemic perception training on the speech production and phonological awareness skills of children with expressive phonological delay. American Journal of Speech-Language Pathology, 13(3),

127 119 Rvachew, S., Ohberg, A., Grawburg, M., & Heyding, J. (2003). Phonological awareness and phonemic perception in 4-year-old children with delayed expressive phonology skills. American Journal of Speech-Language Pathology, 12(4), Rvachew, S., Rafaat, S., & Martin, M. (1999). Stimulability, speech perception skills, and the treatment of phonological disorders. American Journal of Speech-Language Pathology, 8, Savage, R. S., Frederickson, N., Goodwin, R., Patni, U., Smith, N., & Tuersley, L. (2005). Relationships among rapid digit naming, phonological processing, motor automaticity, and speech perception in poor, average and good readers and spellers. Journal of Learning Disabilities, 38(1), Schatschneider, C., Fletcher, J. M., Francis, D. J., Carlson, C. D., & Foorman, B. R. (2004). Kindergarten prediction of reading skills: A longitudinal comparative analysis. Journal of Educational Psychology, 96(2), Sénéchal, M., Ouellette, G., & Young, L. (2004). Testing the concurrent and predictive relations among articulation accuracy, speech perception, and phoneme awareness. Journal of Experimental Child Psychology, 89(3), Sharf, D. J., Ohde, R. N., & Lehman, M. E. (1988). Relationship between the discrimination of /w-r/ and /t-d/ continua and the identification of distorted /r/. Journal of Speech and Hearing Research, 31(2), Shriberg, L. D. (1994). Five subtypes of developmental phonological disorders. Clinics in Communication Disorders, 4(1),

128 120 Shriberg, L. D. (1997). Developmental phonological disorders. In B. W. Hodson & M. L. Edwards (Eds.), Perspectives in Applied Phonology (pp ). Gaithersburg, MD: Aspen Publishers, Inc. Shriberg, L. D., Austin, D., Lewis, B. A., McSweeny, J. L., & Wilson, D. L. (1997a). The percentage of consonants correct (PCC) metric: Extensions and reliability data. Journal of Speech, Language, and Hearing Research, 40, Shriberg, L. D., Campbell, T. F., Karlsson, H. B., Brown, R. L., McSweeny, J. L., & Nadler, C. J. (2003a). A diagnostic marker for childhood apraxia of speech: The lexical stress ratio. Clinical Linguistics & Phonetics, 17(7), Shriberg, L. D., Green, J. R., Campbell, T. F., McSweeny, J. L., & Scheer, A. R. (2003b). A diagnostic marker for childhood apraxia of speech: the coefficient of variation ratio. Clinical Linguistics and Phonetics, 17(7), Shriberg, L. D., & Kwiatkowski, J. (1982). Phonological disorders III: A procedure for assessing severity of involvement. Journal of Speech and Hearing Disorders, 47, Shriberg, L. D., & Kwiatkowski, J. (1994). Developmental phonological disorders I: A clinical profile. Journal of Speech, Language, and Hearing Research. Shriberg, L. D., Kwiatkowski, J., & Gruber, F. A. (1994). Developmental phonological disorders II: Short-term speech-sound normalization. Journal of Speech and Hearing Research, 37. Shriberg, L. D., Kwiatkowski, J., & Hoffmann, K. (1984). A procedure for phonetic transcriptions by consensus. Journal of Speech & Hearing Research, 27(3),

129 121 Shriberg, L. D., Lewis, B. A., McSweeny, J. L., & Wilson, D. L. (1997b). The speech disorders classification system (SDCS): Extensions and lifespan reference data. Journal of Speech, Language, and Hearing Research, 40, Shriberg, L. D., Lewis, B. A., Tomblin, J. B., McSweeny, J. L., Karlsson, H. B., & Scheer, A. R. (2005). Toward diagnostic and phenotype markers for genetically transmitted speech delay. Journal of Speech, Language, & Hearing Research, 48(4), Shriberg, L. D., Lohmeier, H. L., McSweeny, J. L., & Kwiatkowski, J. (2006). A nonword repetition task for speakers with misarticulations: The Syllable Repetition Task. Poster presented at the Annual Convention of the American Speech Language Hearing Association, Miami, FL. Shuster, L. I. (1998). The perception of correctly and incorrectly produced /r/. Journal of Speech, Language, and Hearing Research, 41, Smit, A. B., Hand, L., Freilinger, J. J., Bernthal, J. E., & Bird, A. (1990). The Iowa articulation norms project and its Nebraska replication. Journal of Speech and Hearing Disorders, 55(4), Smith, A. B., Roberts, J., Lambrecht Smith, S., Locke, J. L., & Bennett, J. (2006). Reduced speaking rate as an early predictor of reading disability. American Journal of Speech-Language Pathology, 15(3), Snow, C. E., Burns, M. S., & Griffin, P. (1998). Preventing reading difficulties in young children. Washington, DC: National Academy Press. Snowling, M. J. (2000). Dyslexia (Vol. 2nd). Oxford: Blackwell Publishers. SPSS. (2006). SPSS for Windows, Rel

130 122 Stackhouse, J. (1997). Phonological awareness: Connecting speech and literacy problems. In B. W. Hodson & M. L. Edwards (Eds.), Perspectives in applied phonology (pp ). Gaithersburg, MD: Aspen Publishers, Inc. Stackhouse, J., & Wells, B. (1997). Children's speech and literacy difficulties. London: Whurr Publishers Ltd. Strange, W., & Broen, P. A. (1981). The relationship between perception and production of /w/, /r/, and /l/ by three-year-old children. Journal of Experimental Child Psychology, 31, Stringer, R. W., Toplak, M. E., & Stanovich, K. E. (2004). Differential relationships between RAN performance, behavior ratings, and executive function measures: Searching for a double dissociation. Reading and Writing: An Interdisciplinary Journal, 17, Stringfellow, K., & McLeod, S. (1994). Using a facilitating phonetic context to reduce an unusual form of gliding. Language, Speech and Hearing Services in Schools, 25, Sutherland, D., & Gillon, G. T. (2005). Assessment of phonological representations in children with speech impairment. Language, Speech and Hearing Services in Schools, 36, Swan, D., & Goswami, U. (1997a). Phonological awareness deficits in developmental dyslexia and the phonological representations hypothesis. Journal of Experimental Child Psychology, 66, Swan, D., & Goswami, U. (1997b). Picture naming deficits in developmental dyslexia: The phonological representations hypothesis. Brain and Language, 56,

131 123 Swanson, H. L., Trainin, G., Necoechea, D. M., & Hammill, D. D. (2003). Rapid naming, phonological awareness, and reading: A meta-analysis of the correlation evidence. Review of Educational Research, 73(4), Tangel, D. M., & Blachman, B. A. (1992). Effect of phoneme awareness instruction on kindergarten children's invented spelling. Journal of Reading Behavior, 24(2), Torgesen, J. K., & Wagner, R. K. (1997). Comprehensive Test of Phonological Processes in Reading. Austin, TX: PRO-ED. Troia, G. A., Roth, F. P., & Yeni-Komshian, G. H. (1996). Word frequency and age effects in normally developing children's phonological processing. Journal of Speech and Hearing Research, 39, Tyler, A. A., Edwards, M. L., & Saxman, J. H. (1990). Acoustic validation of phonological knowledge and its relationship to treatment. Journal of Speech and Hearing Disorders, 55(2), Usdan, V. L. (1978). Utilization of the "straw technique" for correction of the lateral lisp. Language, Speech and Hearing Services in Schools, 9(1), 5-7. Wagner, R. K., & Torgesen, J. K. (1987). The nature of phonological processing and its causal role in the acquisition of reading skills. Psychological Bulletin, 101(2), Walley, A. C., Metsala, J. L., & Garlock, V. M. (2003). Spoken vocabulary growth: Its role in the development of phoneme awareness and early reading ability. Reading and Writing: An Interdisciplinary Journal, 16(1-2), 5-20.

132 124 Watson, B. U., & Miller, T. K. (1993). Auditory perception, phonological processing, and reading ability/disability. Journal of Speech and Hearing Research, 36(4), White, H. R. (1997). Comparison of computer-assisted and by-hand phonological analyses for young children with disordered phonology. Unpublished Masters Thesis, Syracuse University, Syracuse, NY. Wiig, E. H., Secord, W. A., & Semel, E. (2004). Clinical Evaluation of Language Fundamentals Preschool (2 ed.). San Antonio, TX: Harcourt/Psych Corp. Wolf, M., & Bowers, P. G. (1999). The double-deficit hypothesis for developmental dyslexias. Journal of Educational Psychology, 91(3), Wolf, M., O'Rourke, A. G., Gidiney, C., Lovett, M., Cirino, P., & Morris, R. (2002). The second deficit: An investigation of the independence of phonological namingspeed deficits in developmental dyslexia. Reading and Writing: An Interdisciplinary Journal, 15, Wolk, L. (1990). An investigation of stuttering and phonological difficulties in young children. Unpublished Doctoral dissertation, Syracuse University, Syracuse. Wolk, L., Edwards, M. L., & Conture, E. G. (1993). Coexistence of stuttering and disordered phonology in young children. Journal of Speech and Hearing Research, 36,

133 125 Appendix A: Transcription Rules and Coding Sound Changes 1) Do not count errors on morphological endings that are added (e.g., toys, shrimps). 2) Transcribe and code any errors in the second part of a two-word phrase (e.g., plant it; this one). 3) If part of an utterance is overlaid by another speaker, do your best to transcribe accurately. However, if uncertain, give the child as much credit as possible (i.e., count overlaid sound as correct). 4) Do not penalize the child for dialectally acceptable forms a) Count t Ȏ as correct between stressed and unstressed vowels (e.g., hippopotamus; spaghetti). However, t d is a typical voicing error. b) Allow the following alternations in the target form (i.e., don t consider them as errors): i) Beige: /beȴ/ or /beȣ/ (#4) ii) Newspaper: /nuspepǫ/ or /nuzpepǫ/ (#13) iii) Garage: /gǩrǡȣ/ or /gǩrǡȴ/ (#70) 5) For Leaf allow: /lif/ or /liv/ (#72) (/liv/ is allowed due to back formation from the plural leaves) 6) For Quack allow /kwæk/ or /kwækwæk/ (#100) Adjust denominator (total number of consonants) accordingly

134 126 7) Do not penalize for affrication/palatalization of /tr, dr, tw, dw/ clusters, or if these could be intermediate steps in a sound change (due to dialect) a) Ex: /tri/ [tȓri] is no error b) Ex: /twǻnz/ [tȓwǻnz] is no error 8) Surface errors should be broken down into separate sound changes, each of which is coded as a typical sound change, an atypical sound change, and/or a distortion. That is, an error may involve interacting changes (more than one typical, atypical, and/or distortion errors). Code surface errors based on the smallest number of sound changes to arrive at the child s production. If multiple paths of interacting changes are possible to account for a child s production, choose the one with the fewest atypical changes (see Appendix B). 9) Use the list of error pattern descriptions below to account for a child s production of a word. To the extent possible, try to account for all component feature changes (e.g., manner, place, voicing). 10) Typical and atypical changes can be interacting, as can multiple atypical changes. a) Ex: ladder [mædǩ] would be gliding of /l/ /w/ (typical) + Nasalization of /w/ [m] (atypical) b) Ex: yawn [vǥn] would be Glide Interchange /j/ /w/ (atypical) and Frication of a Glide /w/ [v] (atypical)

135 127 11) Glottal stops: a) Not considered an error at the end of a word/compound word if replacing /t/, because that is an acceptable dialectal variant i) Ex: squirtgun [skwǭȥ gȝn] is not an error ii) Ex: basket [bǽskǻȥ] is not an error iii) Ex: elephant [ǫlǝfǻnȥ] is no error, but [ǫlǝfǻȥ]] is Nasal Cluster Reduction, but no penalization for glottal stop replacing /t/ b) Atypical sound changes: the intrusion of a glottal stop, or a glottal stop substitution in word-initial or intervocalic position or in place of a consonant (other than a final /t/) in a final cluster or final position i) Ex: teeth [Ȥiθ] is an atypical sound change ii) Ex: ladder [lǽȥǫ] is an atypical sound change (intervocalic substitution) iii) Ex: chicken [tȓǻȥkǫn] is an atypical sound change (intrusion) iv) Ex: ketchup [kǽtȓǝȥ] is an atypical sound change v) Ex: tractor [trǽȥtǫ] is an atypical sound change 12) Partial voicing/devoicing errors should be transcribed, but they are not considered as distortions; they will be considered acceptable phonetic variants. 13) Assimilation errors are considered typical, whether they involve partial assimilation (one or several features) or complete assimilation (all features) (see section below on assimilations).

136 128 TYPICAL PHONEMIC SOUND CHANGES: (based on Edwards & Shriberg, 1983; Lowe, 1994) Typical Syllable Structure Changes Examples Sound Change Definition Target Word Production Final Consonant Deletion (FCD) S Cluster Reduction (SCR) - initial S Cluster Reduction (SCR) - final Liquid Cluster Reduction (LCR) Final singleton consonants are deleted in words or compound words /s/ is deleted in a consonant cluster in syllable-initial position Either /s/ or another phoneme is deleted in word-final position Deletion of /r/ or /l/ in any liquid cluster (syllable initial or final) spoon /spun/ ice cream /aǻs krim/ spoon /spun/ snake /snek/ dentist /dǫntǻst/ mailbox /melbǥks/ three /ſri/ present /prǫzǻnt/ [spu] [aǻ krim] [pun] [nek] [dǫntǻs] or [dǫntǻt] [melbǥk] or [melbǥs] [ſi] [pǫzǻnt] skateboard /sketbord/ [sketbod] black /blæk/ [bæk] sled Glide Cluster Reduction (GCR) Deletion of /w/ or /j/ in a cluster /slǫd/ twins /twǻnz/ vacuum /vǽkjum/ [sǫd] [tǻnz] [vǽkum]

137 129 Nasal Cluster Reduction (NCR) Deletion of either element of a nasal cluster elephant /ǫlǩfǻnt/ [ǫlǩfǻt] or [ǫlǩfǻn] Consonant Sequence Reduction (CSR) (Hodson, 1997) Deletion of a consonant in a sequence that crosses syllable or word boundaries helicopter /hǫlǻkǥptǫ/ tractor /trǽktǫ/ [hǫlǻkǥpǫ] or [hǫlǻkǥtǫ] [trǽtǫ] or [trǽkǫ] Weak Syllable Deletion Deletion of unstressed syllable banana /bǩnǽnǩ/ [nǽnǩ] hippopotamus /hǻpǩpǡtǩmǻs/ [hǻpǩpǡmǻs] Epenthesis (EP) Segment Coalescence (Seg- COA) Insertion of a vowel, often a schwa, in consonant clusters. Does not include insertion of other phonemes in other positions (see atypical intrusive consonants, vowels, syllables) Features from two adjacent phonemes combine to form a new segment that retains features of both phonemes. black /blæk/ plate /plet/ black /blæk/ [bǩlæk] [pțlet] [væk] ([v] has labial feature of /b/ and continuant feature of /l/) spider/spaǻdǫ/ [faǻdǫ] In our definition, segment coalescence can involve place and manner features, but not voicing. zebra /zibrǩ/ BUT: plate /plet/ [zivǩ] [vet] is Seg- COA + Voicing

138 130 Syllable Coalescence (Syl- COA) Reduplication (REDUP) Segments from two adjacent syllables combine, with a weak vowel (and sometimes a following sonorant) being deleted Entire stressed syllable is repeated, or a simplified version of it garage /gǩrǡȣ/ banana /bǩnǽnǩ/ pudding /pțdǻŋ/ [gǡȣ] [bǽnǩ] [pțpț] Typical Place of Articulation Changes Examples Sound Change Definition Target Word Production Depalatalization/ Palatal Fronting (DEPAL) Palatal obstruent is replaced by an alveolar chocolate /tȓǥklǻt/ [tsǥklǻt] Velar Fronting Velar phoneme replaced by alveolar cage /keȴ/ cage /keȴ/ [kedz] [teȴ] Labialization (LAB) Alveolarization (ALV) Alveolar or interdental becomes labial. NOTE: labial replacing palatal or velar is atypical (see atypical place changes) Interdental or labial consonant is replaced by alveolar consonant green /grin/ three /θri/ toy /tǥǻ/ scissors /sǻzǫz/ thumb /θȝm/ beige /beȣ/ [drin] [fri] [pǥǻ] [fǻzǫz] [sȝm] [deȣ]

139 131 Typical Manner of Articulation Changes Examples Sound Change Definition Target Word Production Gliding of Liquids (GL) liquids /r, l/ become glides [w] or [j] rabbit /ræbǻt/ [wæbǻt] Gliding of Fricatives (GF) Homorganic glide replaces a fricative in intervocalic position only (if fricatives are glided in word-initial position, this is atypical) leaf /lif/ television /tǫlǩvǻȣǻn/ [wif] or [jif] [tǫlǩwǻȣǻn] or [tǫlǩvǻjǻn] Stopping (ST) Fricatives or affricates become homorganic stops (i.e., same place of articulation) zebra /zibrǩ/ leaf/lif/ [dibrǩ] [lip] Stopping of palatal fricatives and affricates is just one change (ST, not ST + DEPAL) cage /keȴ/ catch /kætȓ/ [ked] [kæt] Vocalization (VOC) Stopping of interdentals to alveolar is one change. If resulting stop is interdental, do not count as distortion Postvocalic /l, r/ and syllabic liquids are replaced by a vowel three /θri/ thimble /θǻmblʜ/ thimble /θǻmblʜ/ spider/spaǻdǫ/ [ti] or [tʝi] [tǻmblʜ] or [tʝǻmblʜ] [θǻmbo] [spaǻdǩ] Deaffrication (DEAFF) Affricates are replaced by homorganic fricative cage /keȴ/ chocolate /tȓǥklǻt/ [keȣ] [ȓǥklǻt]

140 132 Affrication of fricatives (AFF) Fricatives become homorganic affricates washing /wǥȓǻŋ/ [wǥtȓǻŋ] television [tǫlǩvǻȴǻn] /tǫlǩvǻȣǻn/ zebra /zibrǩ/ [dzibrǩ] leaf /lif/ [lipf] Typical Voicing Changes Examples Sound Change Definition Target Word Production Initial Voicing (IV) Voiceless obstruents become voiced before a sonorant. Note: partial voicing is not considered an error cage /keȴ/ truck /trȝk/ [geȴ] [drȝk] Final Devoicing (FD) Voiced obstruents become voiceless at the end of a word or syllable. Note: partial devoicing is not considered an error cage /keȴ/ garage /gǩrǡȣ/ [ketȓ ] [gǩrǡȓ]

141 133 Other Typical Changes: Examples Sound Change Definition Target Word Production Metathesis (MET) Two consonants in a word exchange positions animals /ænǻmǩlz/ [æmǻnǩlz] Note: if the change results in a phonotactic violation, consider it atypical BUT: spring /sprǻŋ/ [pswǻŋ] is atypical because [psw] clusters are not allowed in English

142 134 Assimilations (ASSIM) are Typical Changes One phoneme takes on one or more features of another phoneme in the word: velar, alveolar, labial, palatal, nasal, liquid, fricative, continuant, etc. It may involve place and/or manner, but not voicing, as defined for this study Even if assimilations involve more than one feature, they count as just one typical change Velar ASSIM: Palatal ASSIM: Nasal ASSIM: Liquid ASSIM: Frication ASSIM: ASSIM to place and manner: Complete ASSIM: guitar /gǩtar/ shovel /ȓȝvǩl/ banana /bǩnǽnǩ/ ladder /lædǫ/ beige /beȣ/ leaf /lif/ flag /flæg/ [gǩkar] [ȓȝȣǩl] [mǩnǽnǩ] [lælǫ] [veȣ] [vif] [glæg] Complete assimilation (in which all features assimilate) is preferred if it can reduce the total number of steps GL + Complete ASSIM is Preferred over GL + ST + Lab ASSIM to explain initial [p] shrimp /ȓrǻmp/ [pwǻmp]

143 135 ATYPICAL SOUND CHANGES: (Indicated by an asterisk *) Atypical Syllable Structure Changes Examples Sound Change Definition Target Word Production Atypical /s/ Cluster Reduction (*ASCR) In word-initial /s/ clusters, the /s/ remains and the stop or nasal is deleted This is not applied to /sl/ and /sw/ clusters snowman /snomæn/ school /skul/ BUT: sled /slǫd/ [somæn] [sul] [sǫd] is LCR swim /swǻm/ [sǻm] is GCR Atypical Liquid Cluster Reduction (*ALCR; Lowe, 1994) In liquid clusters, the liquid is retained This can also be applied to syllablefinal liquid clusters tree /tri/ plant /plænt/ twelve /twǫlv/ skateboard [ri] [lænt] [twǫl] /sketbord/ [sketbor] Atypical Glide Cluster Reduction (*AGCR) In stop + glide cluster, the glide is retained twelve /twǫlv/ twin /twǻn/ [wǫlv] [wǻn] Initial Consonant Deletion (*ICD; Dodd & Iacano, 1989) Word-initial singleton consonants are deleted This can be evident when both elements of an intial cluster are deleted toy /toǻ/ leaf /lif/ plant /plænt/ [oǻ] [if] [ænt] is LCR and ICD

144 136 Medial (intervocalic) Consonant Deletion (*MCD; Dodd & Iacano, 1989) Addition of consonants, vowels or syllables (*ADD) Intervocalic consonants are deleted Individual consonants, vowels, or whole syllables are added Note: not used between members of consonant clusters (see Epenthesis) ladder /lædǫ/ scissors /sǻzǫz/ shovel /ȓȝvǩl/ spring /sprǻŋ/ beige /beȣ/ [læǫ] [sǻǫz] [ȓȝvǩvǩl] [sprǻŋk] [beȣa] Migration (*MIG) (Leonard & McGregor, 1991) A consonant is moved to another part of the word soap /sop/ [ops] Strong Syllable Deletion (*SSD) Syllable/ vowel with primary or secondary stress is deleted. shovel /ȓȝvǩl/ basket /bǽskǻt/ [vo] [kǻt] hippopotamus /hǻpǩpǡtǩmǻs/ [pǡtǩmǻs] is WSD + *SSD

145 137 Atypical Place of Articulation Changes Examples Sound Change Definition Target Word Production Glottal Replacement (*GR; Dodd & Iacano, 1989) Glottal stop [Ȥ] replaces a consonant (except syllable final /t/, in which case no error is counted) leaf /lif/ chocolate /tȓǥklǻt/ [Ȥif] or [liȥ] [ȤǤklǺt] BUT: [tȓǥklǻȥ] is NOT an error This is also used if a glottal stop replaces /h/ hippo /hǻpo/ [ȤǺpo] is an atypical error Backing (*BACK) A labial, dental, alveolar, or palatal is backed to a velar. Used when velar assimilation is not possible. toy /toǻ/ banana /bǩnænǩ/ [koǻ] [gǩnænǩ] Palatalization (*PAL) A non-palatal fricative or affricate (usually, but not restricted to alveolar) becomes a palatal phoneme. Used when palatal assimilation is not possible. scissors /sǻzǫz/ zebra /zibrǩ/ teeth /tiθ/ [ȓǻzǫz] or [sǻȣǫz] [Ȣibrǩ] [tiȓ] BUT: [tȓiθ] is teeth /tiθ/ *Affrication + *PAL

146 138 Atypical Labialization (*ALAB) Velar or palatal phoneme becomes labial. Used when labial assimilation is not possible. Note that alveolar or interdentals becoming labial is typical guitar /gǩtar/ catch /kætȓ/ [bǩtar] [pætȓ] or [kæp] (ST+ *LAB) Glide Interchange (*GLINT) Interchange between /j/ and /w/. Used when complete assimilation is not possible. yawn /jǥn/ washing /wǥȓǻŋ/ [wǥn] [jǥȓǻŋ] BUT: [vækwum vacuum cleaner kwinǫ] is GL + /vækjum complete klinǫ/ Assim. yoyo /jojo/ [wowo] is GLINT + complete Assim Liquid Interchange (*LIQINT) Interchange between /r/ and /l/ rabbit /ræbǻt/ leaf /lif/ [læbǻt] [rif] green /grin/ [glin]

147 139 Atypical Manner of Articulation Changes Sound Change Definition Target Word Examples Production Denasalization (*DENAS) Nasal phoneme become homorganic voiced stop (Dodd & Iacano, 1989; Shriberg, 1993) nose /noz/ [doz] Nasalization (*NAS) Non-nasal phoneme becomes homorganic nasal. Occurs only when nasal assimilation is not possible (cf. Shriberg, 1993) leaf /lif/ beige /beȣ/ [nif] [meȣ] Fricatives Replace Stops (*FRS) Fricative replaces homorganic stop. Only occurs when assimilation of fricatives is not possible (cf. Lowe, 1994) toy /toǻ/ crib /krǻb/ [soǻ] [krǻv] Liquids Replacing Glides (*LIQ) Glides become liquids (Stringfellow & McLeod, 1994) you /ju/ twin /twǻn/ [lu] [trǻn] Tetism (*TET) (Edwards & Shriberg, 1983) /f/ [t] Used when assimilation to alveolar stop is not possible feather /fǫðǫ/ leaf /lif/ [tǫðǫ] [lit] Atypical Gliding of Intervocalic Consonants (*AGL) Intervocalic consonants (other than fricatives) are replaced by glides ladder /lǽdǫ/ rabbit /rǽbǻt/ [lǽjǫ] [rǽwǻt]

148 140 Atypical Stopping of Liquids or Glides (*AST) Glide or liquid becomes a homorganic stop. Note that place of articulation changes should be counted as separate sound changes. leaf /lif/ washing /wǥȓǻŋ/ BUT: yawn /jǥn/ /dif/ [bǥȓǻŋ] [bǥn] is *AST + LAB leaf /lif/ [gif] is *AST and *BACK Atypical Voicing Changes Examples Sound Change Definition Target Word Production Initial/Prevocalic Devoicing (*IDEV) Prevocalic obstruents becomes devoiced (Dodd & Iacano, 1989) dog /dǥg/ [tǥg] Final Voicing (*FV) Postvocalic/final obstruents become voiced hat /hæt/ [hæd]

149 141 Interacting Atypical Sound Changes Atypical changes may interact with atypical changes, typical changes or distortions. Examples Sound Change Definition Target Word Production Atypical + Atypical Atypical + Distortion Atypical + Typical If an atypical change is repeated more than once in a word, it is coded as one atypical change plus assimilation (typical) *Nasalization + *Atypical Labialization *Liquid Interchange + Distorted /r/ *Palatalization + Initial Voicing *LIQ + ASSIM *BACKING + VELAR ASSIM (twice) guitar /gǻtar/ [mǻtar] yellow /jǫlo/ [jǫr w o] scissors /sǻzǫz/ /ȢǺzǪz/ yoyo /jojo/ [lolo] dentist /dǫntǻst/ [gǫŋkǻst] Assimilations should be considered prior to considering atypical changes For /l/ [d], prefer ASSIM (to alveolar stop) over *AST For /b/ [v], prefer ASSIM (to labial fricative) over *FRS telephone /tǫlǩfon/ thimble /θǻmbǩl/ [tǫdǩfon] [fǻmvǩl]

150 142 Examples of Consonant Cluster Changes: Coalescence, Assimilation, other Changes Do not penalize for affrication/palatalization of /tr, dr, tw, dw/ clusters, or if these could be intermediate steps in a sound change. a) Ex: /θri/ [tȓi] can be coded as Stopping [tri], with no penalization for affrication [tȓri], followed by Liquid Cluster Reduction to [tȓi]. b) Ex: /twǻnz/ [tȓwǻnz] no error is coded /twǻnz/ [ȴǺnz] is Initial Voicing + Glide Cluster Reduction Stop + Liquid or Stop + Glide resulting in a homorganic Fricative + Liquid or Fricative + Glide is Continuant Assimilation Continuant ASSIM: Coalescence: (features from adjacent segments twin /twǻn/ plate /plet/ princess /prǻnsǫs/ flag /flæg/ plate /plet/ [swǻn] [flet] [frǻnsǫs] or [fwǻnsǫs] [sæg] [fet] combine) ASSIM to Place of flag /flæg/ [slæg] Artic: Gliding of Liquid + tree /tri/ [fi] Coalesc. drum/drȝm/ [bȝm] INTERACTIONS: Initial Devoicing + LCR + *Affric. of stop bridge /brǻȴ/ [pfǻȴ] Gliding + ASSIM crib /krǻb/ [fwǻb] (continuant and labial green /grin/ [vwin] features)

151 143 Gliding + ASSIM (continuant and labial features) + Initial Devoicing bridge /brǻȴ/ drive /draǻv/ [fwǻȴ] [fwaǻv] Gliding + Depalatalization tree /tri, tȓri/ drive /ȴraǺv/ [tswi] [dzwaǻv] Liquid Cluster Red + Deaff tree /tri, tȓri/ drive /ȴraǺv/ [ȓi] [ȢaǺv] /s/ Cluster Red + Deaffrication + Liquid Cluster Red string /stȓrǻŋ/ strawberry /strǥbǫri/ [ȓǻŋ] [ȓǥbǫri]

152 144 DISTORTION ERRORS: Only clinically significant distortions (not appropriate for the context) are considered as distortion errors. They can be marked for any consonant (not just sibilants and liquids). NOTES ON DISTORTIONS: o Partial voicing and partial devoicing are not considered to be distortion errors o Only one distortion is coded on a particular phoneme Sibilant Distortions Examples Sound Change Target Word Production Lateralization Dentalization/ Interdentalization Although this is sometimes considered atypical, it will be considered a distortion in this study because of the suspected motoric (rather than linguistic) involvement (Usdan, 1978) Includes substitution of interdental phonemes for sibilants soap /sop/ zebra /zibrǩ/ soap /sop/ zebra /zibrǩ/ [sʢop] or [Ǽop] [zʢibrǩ] [sʝop] or [θop] [zʝebra] [ðibrǩ] Other Includes salivary (wet), whistled, flat tongue position, and other/ nonspecific sibilant distortions

153 145 Rhotic Distortions Examples Sound Change Target Word Production Derhoticization of /r, Ǫ, ǭ/ (Shriberg, 1994) rabbit /ræbǻt/ cracker /kræǫ/ [rʢæbǻt] [krʢæǫʢ] Labialization of /r/ (Shriberg, 1993) Other rabbit /ræbǻt/ [r w æbǻt] Other specific or nonspecific rhotic distortions are possible Other Distortions Note: This list is not exhaustive, but is illustrative of the types of distortions observed Sound Change Examples Partly Nasalized rabbit guitar Partly Denasalized mailbox banana Rhoticization of /w/ washing twins Dentalization of nonsibilant alveolars screwdriver toy dinosaur Distortions Interacting with Other Sound Changes Target Word Examples Production Depalatalization + Sibilant Distortion shovel /ȓȝvǩl/ [θȝvǩl] or [sʝȝvǩl] Alveolarization + Sibilant Distortion leaf /lif/ [lisʝ]

154 146 Appendix B: Errors with Interacting Sound Changes: Which is Preferred? Sometimes more than one path could account for a particular sound change. Examples are shown below to show how one path was selected Most of these different paths relate to scoring consonant clusters. Note that this does not claim that the child is going through these steps, but just that these are phonetically plausible paths connecting the child s production to the corresponding adult form (called derivations). Abbreviations are found in Appendix A. An asterisk (*) indicates an atypical sound change. a. No difference in total scoring. If there is no difference in the resulting score in any of the categories, either path is deemed acceptable. Ex: P43 word #35 drive /draǻv/ [baǻf]. Path #1 Path #2 /draǻv/ GL /draǻv/ LCR [dwaǻv ] COA [daǻv] Labial ASSIM [baǻv] FD [baǻv] FD [baǻf] [baǻf] RESULT: 3 Typical RESULT: 3 Typical

155 147 b. One path results in more atypical sound changes than another. The selected path is always the one with fewer atypical sound changes (so as not to penalize the child). Ex: P35 #116 string /strǻŋ/ [ȓwǻŋ]. Recall /stȓrǻŋ/ is an allowable target Preferred Path Non-Preferred Path /stȓrǻŋ/ /s/ CR /stȓrǻŋ/ *ASCR [tȓrǻŋ] GL [srǻŋ] GL [tȓwǻŋ] DEAFF [swǻŋ] *PAL [ȓwǻŋ] [ȓwǻŋ] RESULT: 3 Typical RESULT: 2 Atypical, 1 Typical Ex: P43 word #36 clown / klațn/ [bațn]. Preferred Path Non-Preferred Path /klațn/ GL /klațn/ LCR [kwațn] COA [kațn] IV [bațn] [gațn] *LAB (atypical) [bațn] RESULT: 2 Typical RESULT: 2 Typical, 1 Atypical

156 148 Ex: P41 #20 zebra /zibrǩ/ [zivǩ] Preferred Path Non-Preferred Path /zibrǩ/ COA /zibrǩ/ ASSIM (Cont.) [zivǩ] [zivrǩ] LCR [zivǩ] RESULT: 1 Typical RESULT: 1 Typical, 1 Atypical c. If one path results in more typical sound changes than another (but atypical changes remain the same), chose the path with the smallest number of typical errors. Ex: P43 word #55 queen /kwin/ [bind] Preferred Path Non-Preferred Path /kwin/ COA /kwin/ ASSIM (labial) [bin] *ADD [pwin] IV [bind] [bwin] GCR [bin] *ADD [bind] RESULT: 1 Typical, 1 Atypical RESULT: 3 Typical, 1 Atypical

157 149 Ex: P 41 word #42 shrimp /ȓrǻmp/ [pwǻmp] Preferred Path Non-Preferred Path ȓrǻmp GL ȓrǻmp GL [ȓwǻmp] ASSIM to /p/ [ȓwǻmp] ASSIM (Lab) [pwǻmp] [fwǻmp] ST [pwǻmp] RESULT: 2 Typical, 0 Atypical RESULT: 3 Typical, 0 Atypical

158 150 Appendix C: Words Used on the Picture Naming Task (adapted from Wolk, Edwards & Conture, 1993) 1. parachute 43. spaghetti 2. baby carriage 44. sticker 3. bathtub 45. smooth 4. beige 46. snake 5. teeth 47. sleep 6. dinosaur 48. swing 7. toy 49. splash 8. ketchup 50. spread 9. cookie 51. strawberry 10. catch 52. screwdriver 11. guitar 53. squirrel 12. measuring cup 54. twelve 13. newspaper 55. queen 14. giraffe 56. three 15. fire truck 57. skateboard 16. valentine 58. ladybug 17. thimble 59. basket 18. this 60. chicken 19. scissors 61. pajamas 20. zebra 62. ice cream 21. xylophone 63. banana 22. shovel 64. telephone 23. hippopotamus 65. television 24. ladder 66. toothbrush 25. refrigerator 67. dishwasher 26. washing machine 68. cage 27. yoyo 69. cowboy 28. animals 70. garage 29. plant 71. mailbox 30. princess 72. leaf 31. black 73. nose 32. brother 74. chocolate 33. bridge 75. jump rope 34. tractor 76. jelly 35. drive 77. feather 36. clown 78. vacuum cleaner 37. cracker 79. thank you 38. glasses 80. thirsty 39. grasshopper 81. there 40. flag 82. sandwich 41. french-fries 83. zipper 42. shrimp 84. shampoo 85. helicopter 86. library 87. rabbit 88. window 89. yawn 90. elephant 91. plate 92. present 93. blanket 94. breathe 95. tree house 96. twins 97. pudding 98. dragon 99. crib 100. quack 101. glove 102. green 103. flower 104. frog 105. throw 106. shrunk 107. spider 108. stamp 109. school bus 110. smoke 111. snowman 112. slide 113. swimming pool 114. splinter 115. spring 116. string 117. scratch 118. squirtgun 119. clock 120. yellow 121. drum 122. dentist 123. washcloth 124. hanger 125. teacher

159 151 Appendix D: Phonological Awareness Tasks Blending Adapted from Larivee & Catts (1999) Onsetrhyme (C-VC) 3 phonemes (C-V-C) Item # BT1 BT2 BT3 B01 B02 B03 B04 B05 B06 BT4 BT5 Stimulus Pictured Choices Score t æg ȓ it w Ǻg f Ǻ ȓ top tag bag (training) shoes meat sheet (training) wig weed pig (training) dish fan fish 0 1 tȓ iz cheese knees chain 0 1 ȓ Ǻp chip ship shell 0 1 ȅ Ȝm thumb sun thief 0 1 m ațs mouse house mouth 0 1 f eǻs feet vase face 0 1 s i d spoon seed knees (training) r oț p soup rope rose (training) B07 v æ n van fan vase 0 1 B08 s Ȝ n sub one sun 0 1 B09 ȴ ǫ t jet net juice 0 1 B10 n aǻ t light night knife 0 1 B11 k aǻ t kite bike cup 0 1 B12 k oț-t cap coat boat 0 1 Total # correct: /12

160 152 Rhyme Matching Adapted from Bird, Bishop & Freeman (1995) Training Items Sue T1 bee Pat T3 witch fish shoe cow face cat T2 two T4 chair hat mud rose coat neck T5 bat ham jet leg Experimental Items Dan 1 mouse cap 0 1 Pete 9 feet cheese 0 1 spoon pan hat ham 2 cat fan doll bean 0 1 run bike sheet nut 3 bone tap keys bat 0 1 can night soap seat 4 van back knees meat 0 1 pin house light knife Doug 5 nut rug 0 1 Ned 13 top seed 0 1 wig soup neck bed 6 mug pig red mug 0 1 sub chin leg chair 7 pot bag pen thief 0 1 jug cup weed head 8 tag run food sled 0 1 door bug hen chain Total correct /16

161 153 Onset Matching Adapted from Bird, Bishop & Freeman (1995) Training Items Pictured Choices /r/ T1 red dog T2 house rug one T3 run chin wig /m/ T4 mouth tape sheep rose T5 doll rope weed mouse Experimental Items /p/ 01 deer kite 0 1 /tȓ/ 06 bike chair 0 1 bug pin ship head 02 sock nut coat pan 0 1 pan boat chain sheet 03 pig hen keys fish 0 1 light bone chip shell 04 can pen shoes rope 0 1 red back net cheese 05 tie pot chin cat 0 1 bean seat witch sheep Total correct: /10

162 154 Ben OST1 cow bone Onset Segmentation & Matching Adapted from Bird et al. (1995) OST2 mud saw boat OST3 bed kite seed door OST4 van net bug tea OST5 sled sock fan back Tom Sam OS01 pin juice 0 1 OS06 two bat 0 1 tie door rug sun OS02 jug ham 0 1 OS07 tape cow 0 1 deer top bee saw OS03 toes dish 0 1 OS08 toes pen 0 1 food hen sock meat OS04 tap ship 0 1 OS09 bag knife 0 1 dog leg soup thumb OS05 doll tea 0 1 OS10 bed tag 0 1 mug hat soap jet Total correct: /10

163 155 Appendix E: Syllable Repetition Task (from Shriberg et al., 2006) 2 Syllable (16 consonants) /bada/ /dama/ /bama/ /mada/ /naba/ /daba/ /nada/ /maba/ 3 Syllable (18 consonants) /bamana/ /dabama/ /madaba/ /nabada/ 4 Syllable (16 consonants) /bamadana/ /danabama/ /manabada/ /nadamaba/ /banada/ /manaba/ Total: 50 Consonants

164 156 Appendix F: Rapid Naming Task Monosyllable: Disyllable: