Mind, Brain, and Education

Prosody, Syllable Stress, and the Segmentation of Speech

Sensitivity to speech rhythm is present from birth (Mehler et al., 1988). By 9 months, infants are aware of prosodic cues, such as stress patterns, and use information such as pausing and durational differences to identify major phrasal units in speech (Jusczyk et al., 1992). Infants around this age can also use relative lexical stress to identify words that have a stressed syllable followed by an unstressed syllable (the dominant strong–weak pattern in English; such as in the word “kingdom”). However, they cannot recognize words with a nondominant weak‐strong pattern such as in the word “guitar” (Jusczyk, Houston, & Newsome, 1999). This selectivity likely aids speech segmentation, the process of identifying the boundaries between words, syllables, or phonemes. Cutler and Norris (1988), for example, suggest English speakers mark word onset by the presence of strong syllables.

Complementary evidence, highlighting the importance of prosodic cues in speech segmentation, is available from studies of metrical stress, the alternating of strong and weak syllables. Here, children's ability to process words with the natural syllable stress patterns reversed (demonstrating sensitivity to metrical stress), is related to individual differences in phonological awareness (Wood, 2006). Further support for the relationship between sensitivity to metrical stress and segmentation skill is available from event‐related potential (ERP) studies in Dutch (Böcker, Bastiaansen, Vroomen, Brunia, & Gelder, 1999) and behavioral studies in Spanish (Sebastian & Costa, 1997).

The common practice of speaking to infants with exaggerated stress and more “extreme”, or stretched vowels, provides a naturalistic demonstration of the importance of prosodic cues in the development of phonological awareness. Not only do infants prefer it (Cooper & Aslin, 1994), evidence suggests that infant‐directed speech (IDS) may in fact support speech segmentation (Leong, Kalashnikova, Burnham, & Goswami, 2014; Thiessen, Hill, & Saffran, 2005). ERP studies support the linguistic benefits of IDS, with infants demonstrating increased neural activity in regions associated with phonetic encoding (e.g., Zhang et al., 2011).

A Temporal Sampling Framework of Developmental Dyslexia

One explanation of how the brain processes spoken language, suggests neural oscillations package speech information into time‐relevant units (Giraud & Poeppel, 2012). Slow theta, and faster gamma oscillatory networks, perform together to encode both phonetic segments (ca. 20–50 Hz) and syllables (ca. 3–10 Hz, Poeppel, 2003). A temporal sampling framework (Goswami et al., 2014) suggests poor temporal sampling at low frequencies (Theta [4–10 Hz] & Delta [1.5–4 Hz]), explains the weak syllable parsing and difficulties with perceiving syllable stress observed in dyslexia (Goswami, Gerson, & Astruc, 2010). According to this framework, individuals with dyslexia possess a deficit in their ability to process changes in amplitude. In particular, this relates to amplitude rise time, which is the rate of onset in the modulation of the amplitude envelope. Amplitude rise time cues the start of the syllable and provides information on syllable stress (Scott, 1998). It also contributes to the perception of speech rhythm more generally and the development of accurate phonological representations (Thomson, Goswami, & Baldeweg, 2008). Many children with poor reading have more difficulty than controls in discriminating amplitude rise time and demonstrate related poor phonological awareness. They may also show difficulties with perceiving duration and pitch, further determinants of perceived syllable stress (examples from behavioral and electroencephalogram [EEG] experimentation, respectively, can be found in Goswami, Fosker, Huss, Mead, & Szűcs, 2011; Hämäläinen, Salminen, & Leppänen, 2013; Kuppen, Huss, Fosker, Fegan, & Goswami, 2011; Kuppen, Huss, & Goswami, 2014; Kuppen & Goswami, 2016; Poelmans et al., 2011; Stefanics et al., 2011; Thomson & Goswami, 2010).

Rhythmic Spoken and Music‐Based Interventions for Literacy

Children's engagement with musical activity has repeatedly been reported as conveying literacy benefits (for a meta‐analysis, see Gordon, Fehd, & Mccandliss, 2015). While many of these studies concern the learning of a musical instrument, education‐based musical interventions often involve singing. Singing is a universal activity which appears spontaneously, in children around the age of 1 (Peretz, Gagnon, Hébert, & Macoir, 2004). Experience of musical rhythm, while less often taught explicitly, is equally ubiquitous. This is also the case for speech rhythm, considered a universal cornerstone in the acquisition of all languages (Langus, Mehler, & Nespor, 2016).

While the singing and speaking of rhythmic rhymes share many features, learning a rhyme does not require the production or matching of a given melody, as necessary for singing. Despite this, the neural networks overlap considerably (Özdemir, Norton, & Schlaug, 2006). While early reports suggested singing was a right lateralized activity (with language on the left), more recent neuroimaging studies report bihemispheric activation for both activities (Peretz et al., 2004; Schlaug, Marchina, & Norton, 2010). It is still unclear whether tune and lyrics are learned and represented separately in the brain (Christiner & Reiterer, 2013).

Sung or spoken rhyming text provides a strong rhythmic structure. A key feature in the current programs is the expectation created by these prosodic regularities, which can provide phonological priming and facilitate word recognition (Chen et al., 2016). In song, EEG studies demonstrate an enhanced linguistic comprehension when patterns of linguistic stress and musical meter align (Gordon, Magne, & Large, 2011).

Intervention studies implicate a causal relationship between training children's sensitivity to rhythm, pitch and melody, and enhanced literacy development (Bolduc, 2009; Gromko, 2005; Overy, 2003; meta‐analysis Standley, 2008). Bhide et al. (2013) delivered a rhythmic program including tapping, clapping, and discriminating rhythms to children with poor reading skills. They observed literacy benefits similar to a comparison group receiving an intervention training program in phonological awareness. In another program developed for children with dyslexia (Thomson, Leong, & Goswami, 2013), the impact of rhythm‐based training, using speech and nonspeech stimuli, was compared with a second program that focused on phonetic training. Both programs demonstrated medium effect sizes on literacy outcomes. Finally, Degé and Schwarzer (2011) gave German preschool children one of three interventions: a musical program (e.g., singing, drumming, rhythmic exercises, and dancing), a phonological skills intervention (e.g., rhyming exercises, phoneme recognition, and syllable exercises), or a sports program, which functioned as a control. After controlling for age, SES, and intelligence quotient (IQ), pupils in the musical and phonological skills programs showed a significant increase in areas of phonological awareness. The performance of the sports group remained unchanged. While this study suggested that experience with music can support the development of phonological awareness, the mechanism was unclear. Both song and rhythmic learning were potential candidates for driving change.

The Current Project

The current project aimed to offer teachers an early literacy classroom program that was effective, scientifically sound, and experimentally tested. The specific project aims were (1) to use rhythmic sung and spoken rhyming verse to increase phonological awareness and (2) to support Year 1 children attending schools with a high proportion of children eligible for free school meals (a proxy for low SES). The bespoke Sung and Spoken programs used exaggerated rhythmic features. Prosodic expectation was highlighted through the use of repetition (whether using stress, timing, or melody) and facilitated large and small speech unit segmentation. There were two desired outcomes:

1. To test the application of a dyslexia framework to a program that was designed for all Year 1 beginning readers and delivered as part of the standard classroom curriculum.
2. To assess the potential impact of the programs to raise phonological awareness in classrooms where pupils were from an area of socioeconomic disadvantage.

It was predicted that both the Sung and Spoken programs would have medium to large effects on the development of small and large unit phonological awareness compared with Controls.

Participants

All participating schools were mixed gender, nonselective schools in Bedfordshire (United Kingdom), as a stipulation of the funding body. All schools had a higher than average number of children receiving free school meals (as detailed in individual Ofsted inspections available online). Previous U.K. analyses indicated that children receiving free school meals are most often members of families claiming income support, and are likely to be one‐parent families. Free school meals can be taken as a proxy for a family falling into the bottom quartile of income distribution (Hobbs & Vignoles, 2007).

Representing two individual grant awards, the project ran across two separate intakes of Year 1 children, with one academic year separating the two cohorts (Cohort 1 M_age = 65.71 months, age range 5–6 years; Cohort 2 M_age = 66.03 months, age range 5–6 years). In Cohort 1, six classrooms from three schools were each randomly allocated to one of three conditions (three Control, one Spoken and two Sung). In Cohort 2, there were 5 schools and 13 classrooms. Condition allocation was achieved through a random number generator available online. In Cohort 2, the allocation was Sung (four classrooms), Spoken (three classrooms), and Controls (six classrooms). Control children were Year 1 children from participating schools receiving no additional literacy programs. Prior to the initiation of the project, informed parental consent was obtained for pupils to complete the pre‐ and post‐intervention testing. This was achieved through a signed consent letter from home. The total number of children where continued parental consent was given (i.e., parents gave consent over the entire project) in Cohort 1 was 98 and in Cohort 2 was 136. Each cohort consisted of entirely separate samples (i.e., Cohort 2 was not a follow up of Cohort 1 at a later date). Descriptive statistics for each condition are presented by Cohort in Tables 1 and 2.

Intervention Materials

In Cohort 1, the materials consisted of nine rhymes/songs (songs were the spoken rhymes set to music), with two new compositions added for Cohort 2. To develop the program, the first author collaborated with academics from the Faculty of Education at Cambridge University, in addition to one semi‐professional and two professional musicians.

Each rhyme/song was built upon common themes from the Year 1 curriculum (e.g. building materials, body parts), with age appropriate vocabulary taken from the 100 highest frequency words in the Child's Printed Word Database. Each written line was short, with the rhyming word placed at the end. The majority of the rhymes used were covered in Phase 5 of “Letters and Sounds” (the U.K. Department for Education and Skills' synthetic phonics program, 2007).

The songs were composed with many segmentation‐supporting features and set to either out‐of‐copyright or bespoke tunes. In many cases, pitch change co‐occurred with syllable change or beat and pitch change underlined divisions at the phoneme level. Pauses were inserted at linguistically meaningful places and vowels were emphasized or even elongated in some places.

Spoken rhymes and songs were delivered over compact disc (CD). A professional singer performed both the rhymes and songs. Percussion sticks and actions were used and were replicated consistently across conditions. PowerPoint slides were provided for each rhyme/song, and presented the words in each, with corresponding images. A CD recording and song/rhyme book was provided for each teacher. Worksheets for the Spoken and Sung conditions in Cohort 2 required the teacher to read out the rhyme and complete the phonological‐based activities (not provided for Controls) Free access to the project materials is available at www.tunetime.co.uk.

Experimental Materials

Procedure

With the exception of the development of two additional songs and the worksheets, Cohorts 1 and 2 adopted the same procedure in school. A peripatetic music teacher initiated the project with a visit to the classroom teachers to introduce the program of rhymes/songs. Teachers then had the summer period to familiarize themselves with the program.

At the start of the academic year, the research assistant administered the baseline or pre‐test phonological awareness assessments and the verbal ability test for the consented sample. The program was then introduced and run for 10 min at the start and end of each day. Each classroom consisted of approximately 30 children. The Spoken and Sung groups received a schedule for delivery that rotated the rhymes/songs so that the children gradually learned the complete repertoire. In both cohorts the program was delivered for 12 weeks (one school term). The phonological tests (see Experimental Materials) were delivered as post‐tests at the end of the term. The research assistant also conducted the fidelity of intervention interview at the end of the program and distributed the parental questionnaires.

Statistical Procedure

An exploratory analysis indicated that the data departed significantly from a normal distribution (according to the kurtosis measure, Shapiro–Wilk test and histogram inspection). A nonparametric procedure was therefore undertaken. The current statistical procedure followed the methods of Thomson et al. (2013), who analyzed a similarly characterized intervention dataset. This particular statistical approach was chosen as it offered transparency for the reader, while providing a statistical technique appropriate for the current data.

A series of Kruskal‐Wallis one‐way analysis of variance tests indicated that there were no statistical differences in baseline measures for Cohort 2. However, in Cohort 1, the Sung group started at a significantly higher level in their Rhyme Detection performance (X² = 11.26, p = .00), than the Spoken group (Z = −3.18, p = .001), and on Rhyme Production (X² = 7.49, p = .02) the Sung group was significantly higher than the Controls (Z = −2.61, p < .01).

To specifically examine the change from pre‐ to post‐test, the procedure of Thomson et al. (2013), adapted from Hatcher (2000), was employed. First, a series of Wilcoxon‐signed rank tests was carried out for the pre‐ and post‐tests for Cohorts 1 and 2. Because the project cohorts were slightly different from one another (see Procedure section for details), analyses were undertaken individually.

Phonological Awareness Measures

As can be seen in Tables 3 and 4, for the Spoken and Sung groups, the majority of change from pre‐ to post‐test was statistically significant. The exceptions were in Cohort 1 where the Spoken group did not make statistically significant change on the Rhyme Detection test and in Cohort 2, where the Sung group did not make statistically significant change on the Rhyme Production task. In Cohort 2, statistically significant change was also present in the Controls on all the three measures. Effect sizes were calculated to assess whether gains made by the Spoken and Sung groups were larger than the gains made by the Controls. To do this, gain scores of the Controls were subtracted from the gains of the Spoken and Sung group for each measure. In almost all cases, the experimental groups made larger gains compared with the Controls. The exception to this was in Cohort 2, where membership of the Control group was related to a small positive effect on Rhyme Production and Phoneme Deletion, compared to those learning the Sung program. As a final step, the supplementary gain was divided by the standard deviation of the gain for the Control group to achieve the effect size (Hatcher, 2000).

In Cohort 1, there was a large effect of the Spoken program on the Rhyme Production measure, a medium effect on Rhyme Detection and a small effect for Phoneme Deletion. In the Sung group, effects were small or absent across measures; however, it is worth noting that on Rhyme Production this group started from a statistically higher point than the Controls. In Cohort 2, there were again effects of the Spoken program on Rhyme Production and Rhyme Detection, although they were smaller than in Cohort 1. The Sung program in Cohort 2 showed at best a negligible effect on the Rhyme Production and Phoneme Deletion tasks, it was membership of the Control group, which conveyed a positive small effect.

Phonics Screening Check

In Cohort 2, the phonics screening check data were gathered as an objective measure of literacy progress. The mean outcomes by group were highly similar (Spoken mean = 35 [SD 7.46], Sung mean = 35 [SD 4.38], and Control mean = 34 [SD 6.81]), perhaps indicating that the measure was not adequately sensitive.

Parental Questionnaire and Fidelity of Intervention Interviews

The provision of music across primary schools in the United Kingdom is highly variable and depends upon staff musical expertise. In the project schools, the children had little exposure to music within school. Additionally, less than five of the project children took music lessons, were part of a choir or had any exposure to music outside of the classroom. In regards to the fidelity of implementation of interviews, some teachers found it easier to administer the program at the end of the day in a 20‐min slot rather than two individual 10‐min sessions at the start and end of the day. A statistical analysis indicated no differential outcomes of note for these classrooms.