Learning to remember: the early ontogeny of episodic memory

Abstract

Over the past 60 years the neural correlates of human episodic memory have been the focus of intense neuroscientific scrutiny. By contrast, neuroscience has paid substantially less attention to understanding the emergence of this neurocognitive system. In this review we consider how the study of memory development has evolved. In doing so, we concentrate primarily on the first postnatal year because it is within this time window that the most dramatic shifts in scientific opinion have occurred. Moreover, this time frame includes the critical age (∼9 months) at which human infants purportedly first begin to demonstrate rudimentary hippocampal-dependent memory. We review the evidence for and against this assertion, note the lack of direct neurocognitive data speaking to this issue, and question how demonstrations of exuberant relational learning and memory in infants as young as 3-months old can be accommodated within extant models. Finally, we discuss whether current impasses in the infant memory literature could be leveraged by making greater use of neuroimaging techniques, such as magnetic resonance imaging (MRI), which have been deployed so successfully in adults.

Keywords: Episodic memory; Hippocampus; Infantile amnesia; Memory development; Navigation; fMRI.

Fig. 1

The human hippocampus. The top panel shows the hippocampi circled in red on sagittal (left), coronal (middle) and axial (right) views from a structural MRI brain scan. The hippocampus is composed of a number of subfields, CA1, CA2, CA3, which are adjoined by neighbouring areas – the dentate gyrus (DG), the subiculum (SUB), presubiculum, parasubiculum, and entorhinal cortex – to form the extended hippocampal formation. Three-dimensional images of two example hippocampi are shown in the bottom panel with some of the subregions indicated. From Mullally and Maguire (2013). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

Fig. 2

A taxonomy of long-term memory together with the brain structures purported to be involved in supporting each system. As illustrated, declarative memory can be further decomposed into memory for facts (semantic memory) and memory for events (episodic memory). From Squire and Zola-Morgan (1996), © National Academy of Sciences.

Fig. 3

The operant conditioning paradigms. (A) The mobile conjugate reinforcement paradigm (Rovee-Collier et al., 1980; suitable for use in 2–7 month old infants). The left panel illustrates phase 1: the baseline condition. Here the ankle ribbon is not connected to the mobile so that when the infant kicks they do not move the mobile. The middle panel illustrates phase 2, the acquisition phase, where the ankle ribbon and the mobile are connected so that when the infant kicks, the mobile conjugately moves. The right panel illustrates phase 3, the retention phase. Here, as in phase 1, the ankle ribbon and the mobile are not connected. However, if the infant recognised the mobile, they should kick to move the mobile. Memory of the mobile is therefore indexed by an increased rate of kicking in phase 3 relative to phase 1. (B) The operant train task (Hartshorn and Rovee-Collier, 1997; suitable for use in 6–24 month old infants). As with the operant mobile task, phase 1 (left panel) provides a baseline measure. Here the lever is deactivated and therefore when the infant presses the lever the train does not move. In phase 2 (middle panel), each lever press made by the infant moves the toy train for 1 or 2 s (depending on the infant's age). In phase 3 (right panel–the retention phase) the lever is again deactivated and memory for the train is indexed by an increased rate of lever pressing relative to the baseline pressing rate in phase 1.

Fig. 4

Standardised reference functions for the maximum duration retention of infants on the operant mobile, operant train and deferred imitation puppet tasks. Maximum retention duration (x-axis) appears to increase linearly as a function of increasing age (y-axis). Note that the difference in the slope of the two functions is attributed to the different training parameters used in these paradigms. This graph has been redrawn exactly from Rovee-Collier and Cuevas (2009), and is reprinted with permission from the American Psychological Association.

Fig. 5

Sensory preconditioning and the deferred imitation puppet task (Barr et al., 2003). The left panel illustrates the sensory preconditioning whereby an infant receives paired pre-exposure to puppet A and puppet B. In phase 2 (middle panel) the target actions (remove the mitten from the puppet's hand, shake the mitten, replace the mitten) were demonstrated for the infant on puppet A. The deferred imitation test then occurs in phase 3 (right panel). Here the memory of the pairing between puppet A and puppet B is demonstrated if the infant models the target actions on puppet B. This phase can only be performed by infants aged 6 months and above, as younger infants are motorically incapable of performing the target actions themselves. Lower panels reproduced from Rovee-Collier and Giles (2010) with the permission of Elsevier.

Fig. 6

(A) The hypothesised associative representation containing each element of the event sequence depicted in Fig. 5 – the deferred imitation puppet task. (B) Here, this associative representation has been subsumed into a larger relational network which also contains the association between puppet A and puppet B (encoded by the infants during the sensory pre-conditioning phase), which subsequently enables the infant to transfer context-specific associations (associations bound to puppet A) across contexts (to puppet B).