Word Segmentation in Infants At High Risk for Autism

Background: Word segmentation –the ability to identify word boundaries in continuous speech– is a fundamental step during language development. Prior research showed that 7-month-olds can identify word boundaries in continuous speech by computing transitional probabilities between syllables (Saffran et al., 1996) and that by 8 months of age, they become increasingly sensitive to other speech cues (e.g., stress placed on the word-initial syllable) available in the input (Johnson & Jusczyk, 2001). Importantly, early word segmentation has been linked to word learning in 17-month-olds (Graf Estes et al., 2007), as well as to higher vocabulary scores at age two and better language skills in the preschool years (Newman et al., 2006).

Objectives: Given that delayed language development is a hallmark feature of autism, examining word segmentation in infants at high risk for autism may provide a means to detect early abnormalities in the language acquisition process.

Methods: Twenty-eight high-risk and 20 low-risk 9-month-old infants (exposed primarily to English input) participated in this study. During the familiarization phase, infants were exposed to a continuous speech stream created by repeatedly concatenating 4 trisyllabic words. To assess infants’ sensitivity to speech cues, a ‘stressed’ version was created for the first syllable of two words by increasing its duration/amplitude/pitch. Individual syllables and the speech stream were created following the exact same procedures used by Johnson and Jusczyk (2001). Following the familiarization phase, infants’ looking times were recorded in response to the presentation of 12 test trials, 3 for each of four test items (blocked and presented in random order): two trisyllabic combinations which occurred 45 times in the speech stream (i.e., ‘words’ by transitional probabilities) and two trisyllabic combinations which occurred only 15 times in the speech stream but with stress placed on the first syllable (i.e., ‘words’ by speech cues).

Results: Low-risk infants showed longer looking times for ‘words’ by the transitional probabilities than for ‘words’ by speech cues (i.e., the predicted pattern at this age).  In contrast, high-risk infants did not show a significant difference in looking times for the two types of words. Furthermore, the proportion of infants who showed longer looking times for ‘words’ by transitional probabilities vs. speech cues differed significantly in the two groups (80% vs. 45%, respectively).  Notably, in the high-risk group, the difference in looking time between ‘words’ by transitional probabilities and ‘words’ by speech cues was positively correlated with word comprehension at 18 and 24 months (MacArthur-Bates Communicative Development Inventory), as well as with receptive and expressive language, and verbal mental age at 36 months (Mullen Scales of Early Learning).

Conclusions: Consistent with prior findings, low-risk infants showed evidence of segmenting the speech stream by relying on the speech cues available in the input.  As a group, high-risk infants did not show this pattern; however, the extent to which these infants showed looking time differences in the expected direction predicted better language skills at 18, 24, and 36 months. These findings suggest that word segmentation may aid in the early identification of autism.