In this section we outline two theoretical frameworks – one rooted in experimental psychology and the other in neurobiology – that have both been highly influential in guiding ideas about how episodic memory functions. Both frameworks propose that there is an intimate relationship between the processes engaged when an event is experienced and those that are engaged when it is later remembered. Although they come out of different experimental traditions, and are articulated at different levels of explanation, the two frameworks are highly complementary. Integrating the frame- works leads to an account of episodic encoding and retrieval that can be framed at the explanatory level of cognitive neuroscience. The account generates predictions, outlined below in the sections describing studies of encoding and retrieval, that are testable in healthy humans with functional neuroimaging methods such as event‐
related fMRI.
Experimental psychologists have long investigated the relationship between the encoding and retrieval of episodic information, emphasizing the interdependency of these seemingly distinct mnemonic functions (e.g., Fisher and Craik, 1977; Morris, Bransford, and Franks, 1977; Tulving and Thomson, 1973). One outcome of this line of research has been the principle of transfer‐appropriate processing (Morris, Bransford, and Franks 1977). Transfer‐appropriate processing (TAP) is predicated on the twin assumptions that memories are represented in terms of the cognitive opera- tions engaged by an event as it is initially processed, and that successful memory retrieval occurs when those earlier operations are recapitulated (Kolers, 1973; for review see Roediger, Gallo, and Geraci, 2002). According to the first of these assump- tions, ostensibly the same event will give rise to different memory representations depending on which aspects of the event are emphasized or attended at study.
According to the second assumption, the effectiveness of a retrieval cue will depend
on the similarity between the processing engaged by the cue and the processing that occurred during encoding: the greater the similarity (the “study–test overlap”), the greater the likelihood of successful retrieval (Roediger and Guynn, 1996; Roediger, Weldon, and Challis, 1989; see Nairne, 2002, for caveats). Thus the question of what constitutes the most effective way of encoding information into memory can be fully addressed only if the retrieval conditions are specified. Similarly, the question of what constitutes an effective retrieval cue can be answered only if the processes engaged during encoding have been defined.
The idea at the core of the TAP principle, namely that retrieval of an episodic memory involves the reinstatement or recapitulation of processes active at the time of encoding, is also found in neurobiologically based models of memory retrieval (e.g., Alvarez and Squire, 1994; Norman and O’Reilly, 2003; Rolls, 2000; Shastri, 2002).
According to these models, recollection of a recent event occurs if the pattern of cortical activity elicited by the event when it was initially experienced is reinstated by activation of a hippocampally stored representation of that pattern. Through this mechanism, anatomically distinct cortical regions that were concurrently active during the online processing of an event will also be co‐activated during its retrieval, preserving associations between the features of the event that, together, make it distinct from other similar occurrences. In the model of Norman and O’Reilly (2003), for example, memory encoding is a consequence of the formation in the CA3 region of the hippocampus of a sparsely encoded representation of the pattern of cortical activity elicited by an event. Retrieval occurs when this representation is reactivated, which in turn leads to the reinstatement of the pattern of cortical activity encoded in the representation. Crucially, reactivation of the hippocampal representation does not depend on perfect overlap between the originally encoded activity and the activity engendered during a retrieval attempt. Because CA3 is highly effective at “pattern completion” (Marr, 1971; Wallenstein, Hasselmo, and Eichenbaum, 1998), activity that only partially overlaps the encoded information can be sufficient to cause reacti- vation of the entire representation, and hence reinstatement of the original cortical pattern. Because of this pattern‐completion mechanism, memories can be retrieved in response to retrieval cues that elicit activity that only partially resembles the activity elicited by the original event.
The principles of TAP and cortical reinstatement share several key concepts. These include the idea that memory retrieval involves the recapitulation of processes and representations that were active during encoding, and that the likelihood of successful retrieval is a function of the extent to which the processing engaged by a retrieval cue overlaps with that engaged at encoding.
Figure 5.1 illustrates one way in which these ideas can be schematized in terms of large‐scale patterns of brain activity. The figure attempts to capture the twin ideas that the retrieval of a prior episode involves reinstatement of the pattern of neural activity engaged during the original experience, and that retrieval cues need only elicit a fraction of the original activity in order to trigger the reinstatement of the entire pattern. The figure also highlights the key role played by the hippocampus in both successful encoding and retrieval. Before discussing empirical findings relevant to this theoretical framework, some caveats and qualifications are in order:
1 Encoding–retrieval overlap is represented in Figure 5.1 in terms of co‐activation of cortical regions on a relatively coarse spatial scale, concordant with what can
typically be detected with fMRI. Cortical activity can, however, be differentiated on a much more fine‐grained spatial scale (ultimately, of course, to the level of individual or small populations of neurons), and with respect to its temporal as well as its spatial properties. For example, online processing of two different faces is differentiated not so much by activity in distinct cortical regions, but by differ- ences in the patterning of activity within a common region. Thus, the task of the hippocampus, or any other structure whose function is to capture and later rein- state patterns of activity associated with the processing of an event, is considerably more complex than simply registering which cortical areas were co‐activated by the event.
2 Figure 5.1 is far from being a complete specification of the component processes underlying successful retrieval. Indeed, if this were all that there was, it is unclear how we would be able to distinguish between the perception of an event and our later memory of it (although see Johnson, Hashtroudi, and Lindsay, 1993).
Furthermore, if episodic retrieval were elicited every time there was overlap bet- ween current and past processing, we would be in a state of almost continuous retrieval. This would be highly maladaptive: it is not helpful when struggling to park your car to be distracted by a vivid recollection of the last time you parked in the same lot. As was noted by Tulving (1983), these and related considerations suggest that episodic retrieval is subject to some kind of control mechanism.
Tulving proposed that stimulus events are processed as retrieval cues only when an individual adopts a specific cognitive state, which he termed retrieval mode (see Chapter 1). According to this proposal, depending on whether or not retrieval mode is engaged, the same stimulus event will be processed either as an episodic
Encoding (b)
(a)
Retrieval cue activation Reinstatement
(c) (d)
Delay
figure 5.1 Schematic depiction of the proposed relationship between encoding‐ and retrieval‐related processing in episodic memory. (a) During encoding, presentation of a stim- ulus event activates a diverse set of cortical regions. The resulting pattern of cortical activity is captured by the hippocampus. (b) Following encoding, the pattern of cortical activity elicited by the event is stored as a memory representation in the hippocampus. (c) Later presentation of a part of the event (the retrieval cue) leads to partial reinstatement of the original pattern of cortical activity, which feeds forward to the hippocampus. (d) Overlap between the activity elicited by the retrieval cue and the stored pattern of activity causes the hippocampal represen- tation to be reactivated, which in turn leads to full reinstatement of the original pattern of cor- tical activity.
retrieval cue or with respect to its online significance. It is currently unclear how this and related ideas about retrieval processing (Rugg and Wilding, 2000) should be incorporated into the framework outlined in Figure 5.1.
3 The framework outlined in Figure 5.1 implies that retrieval consists of little more than the “replaying” of the processing engaged by the original experience. If this were so, then recollection would be “all or nothing”; either everything that was registered in the brain as an event unfolded would be retrieved, or nothing would be. Moreover, memories would be largely veridical. Clearly, neither of these sce- narios is accurate. Memories are a very imperfect mirror of experience: they are invariably partial, and often highly distorted, records of the original event (e.g., Bartlett, 1932; Loftus and Palmer, 1974; Schacter, 2002; see also Chapter 8).
Among the many factors contributing to this imperfect relation between an event and our later memory of it, two stand out. First, the different features of an event are not equally likely to be successfully encoded. Other things being equal, those aspects of the event that are attended to the most fully are most likely to be later remembered (see below, and Moscovitch, 1992). Second, there is a wealth of evi- dence that episodic retrieval is a constructive process, in which retrieved information is combined with other knowledge about the event and the result interpreted in light of current expectations and biases (e.g., Bransford and Franks, 1971; Brewer, 1987; Brewer and Treyens, 1981; Schacter, Norman, and Koutstaal, 1998). Thus, retrieved episodic information may only partially determine the content of the resulting memory representation. Together, these two factors will act to reduce the amount of overlap between encoding‐ and retrieval‐related neural activity to well below the 100% illustrated in Figure 5.1. Specifically, as we discuss below, cortical activity elicited during an event by attended information is more likely to be incorporated into the resulting hippocampal representation than is the activity elicited by unattended information. And the constructive nature of memory retrieval means that retrieval‐related neural activity will reflect not only the reinstatement of activity elicited at encoding, but also any non‐episodic information that was incorporated into the memory representation.