Consonant inventories in the spontaneous speech of young children: A bootstrapping procedure

Abstract

Consonant inventories are commonly drawn to assess the phonological acquisition of toddlers. However, the spontaneous speech data that are analysed often vary substantially in size and composition. Consequently, comparisons between children and across studies are fundamentally hampered. This study aims to examine the effect of sample size on the resulting consonant inventories. A spontaneous speech corpus of 30 Dutch-speaking 2-year-olds was used. The results indicate that in order to construct and compare inventories reliably, they should be drawn from speech samples that are equally large. A new consonant inventory procedure is introduced. The implementation of this procedure demonstrates considerably less variation in inventory size across children and word positions than reported previously. This finding has important implications for clinical studies that constructed and compared inventories of typically and atypically developing children without normalizing the sample size.

Vocalization Rate and Consonant Production in Toddlers at High and Low Risk for Autism

“…Using 60 min of speech collected monthly from 30 Dutch toddlers between the ages of 22-24 months, Van Severen et al (2012) confirmed that consonant inventory size was positively correlated with the number of words included in the sample. Their results showed that inventory size increased rapidly with sample size for small sample sizes, but reached a plateau with larger sample sizes.…”

“…First, consonants begin to emerge in babble and speech before the age when diagnoses of ASD are stable, and studies of phonological development in infants at risk for ASD have not always examined those who receive diagnoses of ASD separately from those who do not. Second, consonant inventory size has been shown to be positively related to size of speech sample in general (Van Severen, Van Den Berg, Molemans, & Gillis, 2012), so it is unclear whether smaller consonant inventories might be an artifact of lower vocalization rate. Last, not all studies of toddlers with or at risk of ASD controlled for language level, yet (as noted above; also see Pennington & Bishop, 2009) delays in consonant production are related to delays in expressive language.…”

“…When vocalization rates are the same across children, this method results in comparable inventory sizes. However, when individual vocalization rates differ, this method disadvantages toddlers who vocalize less often (Van Severen et al, 2012).…”

“…As a result, long assessment times (e.g., longer than 60 min) may be necessary to collect enough consonant tokens to analyze but such lengthy samples are difficult to obtain. Van Severen et al (2012) acknowledged this disadvantage to their method. An alternative to collecting extremely long speech samples and comparing the absolute consonant inventory size across children whose vocalization rate differs, however, is to investigate whether the relationship between consonant inventory size and vocalization rate differs between groups.…”

“…Last, because larger speech samples provided more chances to hear different consonants, the mathematical modeling procedure from Van Severen et al (2012) was used to determine whether the relationship between speechlike vocalization rate and number of different consonants differed across groups. First, curve-fitting (Garson, 2012) was used to find the best-fit relationship between speechlike vocalization rate and number of different consonants for each group.…”