Using Gaze Contingent Technology to Characterize a Toddler's Preference for Motherese Speech: Towards Eye-Tracking Based Biomarkers of ASD within the First Years of Life
When given a choice, typically developing infants prefer to listen to “motherese” – a speech style characterized by slower tempo and exaggerated speech contours and positive affect. This style is thought to play an important role in enhancing language acquisition, stimulating joint attention, and improving affective reactivity and emotion regulation. One previous study by Kuhl and colleagues (2005) noted that ~70% of 4-year old children with ASD preferred to listen to a computer generated voice rather than to motherese.
The purpose of the present study was to determine if toddlers much younger than the participants in the Kuhl study would also show a reduced preference for motherese speech using a gaze-contingent eye tracking test battery, and, if so, to determine the validation statistics associated with each eye tracking test.
Using our general population early detection program, the Get SET Early Model, as the main recruitment mechanism, toddlers with ASD and other disorders (mean age 25 months; range 12-43) participated in 1-3 eye tracking tasks using gaze contingent technology wherein a toddler’s fixation location controlled what he/she sees and hears. Toddlers were presented with ~60 second videos that depicted a woman engaging in animated motherese speech on one side of the video, in contrast to a non-social sound (e.g., highway traffic sounds) and associated image on the other. In Test 1, a toddler’s preference for motherese speech was compared to his/her preference for flat affect speech (N=211; 94 ASD vs 117 non-ASD); in Test 2 versus highway noises (N=234; 109 ASD, 125 non-ASD); and in Test 3 (N=53; 23 ASD, 30 non-ASD), versus music. See Figure 1. Percent fixation within motherese and non-motherese AOIs was the main dependent variable. One-way anovas were conducted and significant findings followed up with t-tests. Validation statistics for specificity and PPV were calculated using traditional formulas, as were R0C curves.
Surprisingly, when a preference for motherese speech was isolated by comparing fixation towards motherese and flat affect speech while keeping the mother images constant (Test 1), toddlers with ASD performed similar to TD toddlers: they preferred motherese speech (F=1.03, p=.39; ASD mean motherese 63% vs 67% TD). However, when fixation towards motherese was compared to fixation towards non-social sounds and associated images in Tests 2 and 3, toddlers with ASD dropped their preference for motherese in favor of listening to non-social sounds and watching non-social images. Overall F tests were significant for both tests (F=10.5 and 7.6 respectively, p<.05). When cutoff thresholds were used to generate validation statistics (see dashed line in Figure), specificity and PPV were strong at 98% and 88% for Test 2, and 97% and 90% for Test 3, respectively.
Eye tracking technology can be a powerful early diagnostic tool. Our results suggest that abnormalities in attention patterns in toddlers with ASD are strongly driven by what they see rather than what they hear. The power of combining multiple eye tracking tests and examination of age effects will also be discussed.