A Simplified Hippocampal Model That Learns and Uses Three Kinds of Context

Abstract

Context plays a critical role in cognition. Previously Hirsh (’74), Kesner & Hardy (83), and Gray (’82) proposed that the hippocampus could learn context. Presented here is a simple hippocampal model that learns and uses three types of context: a context coming from the past, a context coming from the present, and a context concerned with the future. In all three of these situations, context is encoded by neurons that fire in a way analogous to hippocampal place cells. When these firing patterns do not appear, the network seems incapable of solving context dependent problems.

In psychology, the importance of context arises before the turn of the century (Boring, ’50), in particular Titchner advocated a critical role for context in perception. Context is important to the networks that learn language in the schemes of Pollack (’90), Elman (’90), Jordan (’86), and Mozer (’92). Although not part of the usual terminology, context is at the heart of frames and schemes used by other cognitive psychologists. More to the point here, context learning seems to be part of hippocampal function (Hirsh, ’74, Kesner & Hardy, ’83, Gray, ’82). Context learning is compatible with the Cohen and Eichenbaum theory of flexible memory (Cohen, ’84; Eichenbaum et al., ’92). Hirsh places the use of context at the center of proper encoding and recall of long-term memory. Context specifies the location of long-term memory storage. In this view, context is equivalent to episodic memory. Moreover, episodic memory associates disparate objects and events from single experiences; unfortunately, it is a lack of episodic memory that so hampers patients like H. M. and R. B. And, finally, even though not an explicit part of O’Keefe and Nadel’s (’78) cognitive mapping theory (but see Nadel and Willner, ’80), a coding that is analogous to hippocampal place cells (we call them context cells) — a coding that can be used to get from point A to point B — are context-based codes when viewed within the function of our model of the hippocampus.

To appreciate the role of context in memory, picture this one situation. You go to the hippocampal conference at Grand Cayman and for the first time you meet John Smith, a scientist from Seattle. A year or two later you visit the NIH and you see a vaguely familiar face; it’s John Smith but you cannot remember his name. (As always striving for politeness as well as wishing to avoid embarrassment, you struggle to come up with a name to match the face.) If you can only remember the place, the circumstances, the episode where you met him, then you will have a chance of remembering the name. You well up a vague association of the conference room where you met and at the same time comes the hotel, the beach, and thenchwr(133) “John, what a surprise seeing you here! How are you?”

In other words, the storage of unique events is intimately associated with the surrounding circumstances (context). Of course, the idea of context-dependent memory is a couple of thousand years old as exemplified by the Roman’s method of loci for memorizing long speeches. One sequence (the speech) is learned by associating it with another sequence of patterns (the sequence of statues you pass as you walk through a well-known museum) by using each successive statue and its locus as the context for successive words and phrases in the speech.

Less grandiose forms of context are useful in many other types of cognitive processes. Thus, many cortical regions would need to produce context codes. But, context-based codes do seem particularly important for hippocampal functions including setting up appropriately retrievable stores of memories.