This thesis examines the temporal architecture of adult-infant interactions (i.e. the
manner in which an interaction unfolds over time). In particular, we explore the existence and
nature of a common narrative temporal framework underpinning adult-infant interaction,
consisting of phases of arousal and intensity split into four distinct states: introduction,
development, climax and resolution. This framework is considered a fundamental structure of
human cognition, and to be central in human communication and learning. We began by
exploring the current state adult-infant interaction research, and identified that recent work
largely fails to consider an underlying narrative element. We then sought to address this
‘narrative gap’ through theoretical, empirical and methodological contributions. We first
applied narrative theory to the neonatal imitation paradigm, viewing the imitative exchange
between experimenter and infant as being inherently dialogical in nature. On this basis we
proposed that underlying successful displays of imitation by neonates was a narrative
framework. We then explored the development of narrative through infancy by conducting a
longitudinal examination of mother-infant interactions when infants were aged 4 months, 7
months and 10 months. We hypothesised that the duration of infant positive affect would be a
function of the narrative phase reached in an interaction, with older infants engaging in
longer interactions that reached more advanced narrative phases. Our results supported these
predictions (except for interaction durations that decreased with infant age). Finally, we
outlined a methodological pipeline to automate the identification of narrative phases in adultinfant engagements. We first described the training and evaluation of a deep learning based
markerless motion tracking model specifically tailored for the tracking of adult and infant
movement during dyadic engagements. We then proposed a machine learning analysis
pipeline for the clustering of this movement data according to narrative phases, thus
removing the potential for human bias in the identification of narrative.