Neocortical catastrophic interference in healthy and amnesic adults: A paradoxical matter of time

Abstract

The human cortex can accommodate overlapping semantic information, such as synonyms, homonyms, or overlapping concepts. However, neuronal models of cortical networks predict Catastrophic Interference in conditions of overlapping information, obliterating old associations and sometimes preventing formation of new ones. It has been proposed that Catastrophic Interference in declarative memory is never observed in biological systems because of hippocampal pattern separation of competing associations. Here, we tested neocortical Catastrophic Interference during acquisition of overlapping associations through Fast Mapping; an incidental, exclusion based learning mechanism, that can support hippocampal-independent learning. Young adults acquired picture-label associations, either through explicit encoding or through Fast Mapping and were tested after 24 h. Overlapping/competing associations were presented either minutes (Early), or 22 h (Delayed) after learning. Catastrophic Interference was evident only following Fast Mapping, and only in the Delayed competition. In a follow-up experiment, Medial Temporal Lobe (MTL) amnesic patients demonstrated retroactive Catastrophic Interference after the Early competition, despite normal memory for noninterfered Fast Mapping associations. Thus, following Fast Mapping, a biological system demonstrated susceptibility to Catastrophic Interference, as predicted by the neuronal-model. Early retroactive Interference, however, can be prevented by MTL integrity.

Rapid consolidation of new knowledge in adulthood via fast mapping

“…Even in subjects with normal hippocampal functioning, a prediction that arises from computational models of memory is that hippocampally-bypassed information should not benefit from hippocampally-mediated resistance to interference. Consistent with this, it has been shown that information learned through fast mapping can be particularly susceptible to interference[7]. …”

“…One additional study deviated from the first’s fast mapping paradigm in several ways (such as using a less elaborative task (clicking the new item) and introducing the study as a word learning investigation), which may explain the atypical pattern of recognition memory performance found, making the results difficult to interpret[6]. A later replication of the original study was successful[7]. In cases of milder hippocampal atrophy, through natural aging, fast mapping has not supported learning more than explicit encoding[8], although the high level of remaining hippocampal volume (88% of the younger participants’ volume) limits the implications for hippocampal dependence.…”

The Consolidation and Transformation of Memory

“…A potential example is ''Fast Mapping,'' the process that permits toddler's rapid acquisition of words following a single exposure (Carey and Bartlet, 1978). A version of the Fast Mapping task was reported to involve in adults rapid neocortical acquisition potentially independent of the hippocampus (Sharon et al, 2011; see also Merhav et al, 2014Merhav et al, , 2015; but see Smith et al, 2014, for reservations concerning explicitness and consolidation into long-term memory independent of hippocampus).…”

What Learning Systems do Intelligent Agents Need? Complementary Learning Systems Theory Updated

“…It is also interesting to consider the finding of 'splitter' cells in a task where animals must alternate between turning left and right on successive trials in a T maze [167][168][169][170][171][172][173][174][175][176][177][178][179]: here, some CA1 and CA3 place cells for locations on the central stem of the T maze are modulated by the trajectory of the rat (e.g., whether it will subsequently turn left or right) whereas others are trajectory-independent. This phenomenon, known as partial remapping [48,[170][171][172], is consistent with the idea that pattern separation is a matter of degree in our theory [27,37].…”

“…In this paradigm, human participants see pairs of pictures of objects -one familiar and one unfamiliar -and are asked a question such as 'is the numbat's tail pointing up', inferring that the unfamiliar name 'numbat' must refer to the unfamiliar object [177]. Some studies find that patients with extensive hippocampus damage show retention of the new object-name association at a delayed test [178,179], suggesting very rapid neocortical learning even without a hippocampus. However, the finding has proven difficult to replicate [180][181][182]; future studies should continue to investigate this issue.…”