THE DEVELOPMENT OF COMMUNICATION ACROSS TIMESCALES

Abstract

How do young children learn to organize the statistics of communicative input across milliseconds and months? Developmental science has made progress in understanding how infants learn patterns in language and how infant-directed speech is engineered to ease short-timescale processing, but less is known about how they link perceptual experiences across multiple levels of processing within an interaction (from syllables to stories) and across development. In this article, we propose that three domains of research - statistical summary, neural processing hierarchies, and neural coupling - will be fruitful in uncovering the dynamic exchange of information between children and adults, both in the moment and in aggregate. In particular, we discuss how the study of brain-to-brain and brain-to-behavior coupling between children and adults will further our understanding of how children's neural representations become aligned with the increasingly complex statistics of communication across timescales.

Figure 1.

Schematic diagrams of three levels of communicative processing, described in terms of (a) statistical summary and (b) neural hierarchies. (a) Statistical summary representations of the word cat, split into three levels of processing. At the single-word level (bottom panel), cat is represented in terms of its acoustic features (e.g., pitch contour) and component phonemes. At the sentence level (middle panel), the word is integrated into its nearby context, including the surrounding words and their pitch contour (which indicates a question here). At the narrative level (top panel), the word is processed in terms of a full story arc about a lost cat, which contains several events whose gist is summarized in each of the four circles, surrounded by the local details that comprise them. (b) Example depiction of neural entrainment to each of the three levels in three age groups: adult, child, and infant. Early sensory regions, such as primary auditory cortex or A1 (bottom), which process fast dynamics of language (syllables, words), might be fairly well synchronized across age groups. However, higher-order, default mode network regions (top), which integrate over longer timescales, might be synchronized across multiple adults but not across members of the three age groups.

Figure 2.

Cartoon example of a communicative interaction between a mother and child during real-life play. Red and blue curves (bottom-right) depict possible neural time series from one early sensory brain region (primary auditory cortex, or A1) and a network of higher-order regions (default mode network, or DMN). As they progress through the interaction, the dyad flows through several states of neural coupling in each region. When they are not interacting (including no shared sensory input), their brains are uncoupled in both regions. When they are hearing the same speech, or viewing the same object, shared input drives sensory coupling. Frequently, the adult anticipates predictable content before the child does, since the adult has access to richer semantic associations and narrative schemas than the child. This happens both at relatively short timescales (e.g., a dark cloud will be followed by rain) and longer ones (e.g., a canonical ending to the “lost cat” schema is that the cat is stuck in a tree). Sometimes, the child reroutes the conversation in a surprising way (e.g., a kangaroo will help find the lost cat), and the adult adapts to this detour. These are all examples of leader-follower dynamics, facilitated by behaviors that guide the other person toward a joint state of understanding. Whenever the adult and child converge on that joint state (e.g., they represent rain in a related way), there is mirrored coupling between them. By the end of the interaction, each person has dynamically adapted to the other to create this story, so the interaction as a whole represents synergistic coupling.