Real-Time dynamics of Child-Directed Speech: Using Pupillometry to Evaluate Children’s Processing of Natural Pitch Contours

• In Event: 3-193 - The Moment-to-Moment Dynamics of Infant-Directed Speech: What Infants Hear and How They Learn

Integrative Statement

Young children prefer child-directed speech (CDS) to adult-directed speech (ADS) (Cooper & Aslin,1990), and structural and prosodic features of CDS facilitate learning (Thiessen, Hill & Saffran, 2005; Graf Estes & Hurley, 2013). However, little is known about the moment-to-moment features of CDS that drive engagement. Parents use variation in pitch in the course of a single word, and introduce new words on pitch peaks (Fernald & Mazzie, 1991; Aslin, 1993), but how do word-level prosody dynamics shape children’s processing? We extracted 4 common word-level pitch contours from natural CDS and used pupillometry to quantify children’s engagement with pitch contours in CDS. Synchrony in pupil size can be used to assess engagement with speech; when adults engage with particular moments in a stimulus, their pupil dilations synchronize (Kang & Wheatley, 2015, 2017). If pupil size synchrony measures engagement in toddlers, then we expect greater synchrony for CDS than ADS.

In Experiment 1, using CHILDES corpora, we automatically extracted pitch contours from natural CDS to one infant (6-12-m.o.) and two children (24-30-m.o.) (Soderstrom et al., 2008; Weist & Zevenbergen, 2008). Hierarchical clustering (Montero & Vilar, 2014) of noun pitch contours yielded 4 clusters (Fig.1): rises, falls, hill, and valleys.

In Experiment 2, we used an eye-tracking paradigm to examine pupil dilation synchrony for CDS vs. ADS and to compare synchrony for the 4 word-level contours. 24-30-month-olds (n=23) listened to the same children’s story (The Little Mouse Who Lost Her Squeak) twice, once in CDS and once in ADS (counterbalanced). Intermixed with stories were 20 individual sentences with target nouns that followed the above-described contours, plus a flat baseline contour. For each trial, we calculated the pairwise dynamic time-warping distance (Tormene et al., 2008) between the pupil size time-series of the participants. Valleys and flats elicited the least synchrony, hills elicited the most synchrony, and rises and falls fell in-between (Fig.2B). We validated these results in the CDS story by clustering the pitch contours of words into contour types and quantified the pairwise synchrony during each word in the story (Fig.2C). A likelihood ratio test showed that contour type improved model fit both in sentence trials (p<0.001) and natural stories (p<0.001) compared to a null model. The interaction between contour type and source type did not improve model fit (p~0.3), suggesting the relative synchrony of the 4 contours followed the same pattern in natural speech and controlled sentences. Synchrony was likely driven, in part, by the naturalness of the contours (see naturalness ratings for valleys and hills in Figure 2D).

This two-part investigation yields a new, subsecond framework for understanding how young children engage with a signal known to support language learning. We identified the most common pitch contours in CDS and revealed a physiological response sensitive to real-time dynamics. In particular, we observed high synchrony for hills, likely reflecting parents’ pervasive use of this contour when referring to key words in CDS (Fernald & Mazzie, 1991; Aslin, 1993). In ongoing research, we are examining young children’s word learning from high-synchrony contours.

Authors

• Mira Nencheva, Princeton University
  Presenting Author

• Elise Piazza, Princeton University
  Non-Presenting Author

• Casey Lew-Williams, Princeton University
  Non-Presenting Author