Generalization of Semantic Learning in Amnesia

Johns Hopkins University, Baltimore MD

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): Individuals with amnesia caused by medial temporal lobe (MTL) damage have marked difficulties learning new semantic (factual) and episodic information. Furthermore, what semantic information they can learn typically seems to be hyperspecific and cannot be flexibly applied. This limitation has been theoretically important. It has been interpreted as indicating that one of the functions of the MTL is to generalize incoming information, by coarse coding and by distributing connections among related information. However, learning without the MTL may not be as hyperspecific as data from amnesia may suggest. The errorless training methods that have been used to teach factual information to individuals with amnesia may encourage narrow, inflexible learning. Moreover, it has been shown empirically that learning can be made more general and more flexible by introducing variance into the materials to be learned. Neural network theorizing has suggested an explanation for this, in terms of the creation and broadening of basins of attraction in memory. Theoretical considerations also suggest that the timing of such training may also be very important. Accordingly, we will study individuals with amnesia caused by MTL damage (confirmed by 3D MRI reconstructions) and matched controls on two types of material: factual information (three word, subject-verb-object sentences such as Harry loves Sally); and face-name associations. Training in the Variance Condition will introduce variations into the verb (loves -> adores) or into the face (different views) and be compared to training without such variations in the No-Variance Condition. Furthermore, the effect of the timing in training when variations are introduced (from the beginning or after the canonical item has been introduced) will be tested. We expect that introducing variation into training will help the material that is learned become more generalizable and more flexible. In addition to standard measures of recall and recognition accuracy, we will also assess learning using reaction time and looking time techniques. Preliminary data from one amnesic individual studied so far supports our predictions and provides evidence for the feasibility and potential theoretical and practical yields of the proposed research. Therefore, the proposed studies will apply evidence and innovative notions from several different areas of investigation to the scientifically important problem of amnesia due to MTL damage. The results from this study would be expected to help initiate new lines of investigation into the interrelated challenges as to what fundamental mechanisms are involved in memory and what the best approaches are for learning new semantic information.