Knowledge supports memory retrieval through familiarity, not recollection

Highlights

• Participants retrieved old and new known and unknown factual statements.

• Retrieval was quicker for known than unknown statements.

• Recollection-related regions were recruited during retrieval of unknown statements.

• Familiarity-related regions were recruited during retrieval of known statements.

• Suggests that known but not unknown statements can be retrieved based on familiarity.

Abstract

Semantic memory, or general knowledge of the world, guides learning and supports the formation and retrieval of new episodic memories. Behavioral evidence suggests that this knowledge effect is supported by recollection—a more controlled form of memory retrieval generally accompanied by contextual details—to a greater degree than familiarity—a more automatic form of memory retrieval generally absent of contextual details. In the current study, we used functional magnetic resonance imaging (fMRI) to investigate the role that regions associated with recollection and familiarity play in retrieving recent instances of known (e.g., The Summer Olympic Games are held four years apart) and unknown (e.g., A flaky deposit found in port bottles is beeswing) statements. Our results revealed a surprising pattern: Episodic retrieval of known statements recruited regions associated with familiarity, but not recollection. Instead, retrieval of unknown statements recruited regions associated with recollection. These data, in combination with quicker reaction times for the retrieval of known than unknown statements, suggest that known statements can be successfully retrieved on the basis of familiarity, whereas unknown statements were retrieved on the basis of recollection. Our results provide insight into how knowledge influences episodic retrieval and demonstrate the role of neuroimaging in providing insights into cognitive processes in the absence of explicit behavioral responses.

Introduction

One fundamental idea in memory research is the distinction between memory for personally-experienced events, or episodic memory, and general knowledge of the world, or semantic memory (Tulving, 1972, Tulving, 1984). While the field traditionally emphasizes how episodic memories and knowledge differ (in phenomenology, development, and vulnerability to change), more recent work highlights the complex relationship between the two. For example, knowledge about sports (e.g., Rawson and Van Overschelde, 2008), people (e.g., Van Overschelde and Healy, 2001), and aviation (e.g., Meade et al., 2009) support new learning (episodic memories) in those domains. Conversely, episodic memory supports the retrieval of knowledge, such as when drawing upon a personal memory of “my auntie's fruit bowl” to generate exemplars of fruits (Vallee-Tourangeau et al., 1998, p. 562).

Knowledge supports both the elaboration and organization of incoming information (Ericsson and Kintsch, 1995), with consequent benefits for later memory. A diehard baseball fan— with intimate knowledge of the rules, players’ tendencies, and strategies—experiences a baseball game quite differently than a cricket game. Our question involves understanding why the baseball game will later be remembered better than the cricket game. From a dual-process view of episodic memory retrieval (Yonelinas, 2002), this benefit could result from an increase in recollection, familiarity, or both (but see, Wixted, 2007). Recollection involves relatively more conscious effort to think back to a particular time and place (i.e., contextual details), such as vividly remembering a game winning walk-off home run. In contrast, familiarity is relatively more automatic and does not invoke reliving; instead one simply knows the information (e.g., the fact that a home run won the game, without a sense of reliving that moment in time).

Separating the contributions of recollection and familiarity to episodic retrieval is challenging, especially since both are likely involved when discriminating old from new events (Yonelinas, 2002). For example, a face may be correctly identified as “old” because one remembers seeing the person in a particular place and time, or because the person feels familiar. One strategy is to measure properties that suggest recollection, such as memory for context, as opposed to simply collecting old/new judgments. Notably, it is easier to remember the background context associated with a famous face than an unfamiliar one (Reder et al., 2013), suggesting a role for recollection in the benefits of knowledge.

A second strategy involves measuring the subjective phenomenology associated with retrieval by asking people to label retrieved events as “remembered” or “known.” While mapping processes onto remember-know judgments is imperfect (Donaldson, 1996, Wixted and Stretch, 2004), “remember” responses do capture at least part of the recollective process. For example, “remember” responses drive the memory benefits of meaning-based over perceptual- based encoding (Gardiner, 1988) and the advantage of pictures over words (Rajaram, 1996).

Critically, knowledge appears to benefit episodic memory through “remembering” rather than “knowing.” For example, people with more Star Trek knowledge (defined by the ability to discriminate Star Trek lifeforms from lures) were more likely to say that they “remembered” reading information from a Star Trek text than were novices, while “remember” responses were similar for the two groups when they were tested on a (control) psychology text (Long and Prat, 2002). Similar effects occur with vocabulary terms; after studying a list of technical terms, students were better able to discriminate old and new terms from their academic major, and that benefit reflected remembering rather than knowing (Brandt et al., 2005). While these studies suggest that knowledge involves recollection-based retrieval, this finding may be unique to expert domains, as experts process stimuli in their domain of expertise more deeply (Kawamura et al., 2007) and encode and retrieve information based on interitem associations within the domain of expertise (Kalakoski and Saariluoma, 2001). Therefore, expertise effects may not extend to general knowledge in a non-expert domain.

We took a different approach to disentangling recollection and familiarity by examining activation of brain areas associated with these two retrieval processes. Recollection-based retrieval has been associated with posterior midline, ventral parietal, anterior prefrontal, and hippocampal regions (Kim, 2013, Rugg and Vilberg, 2013, Skinner and Fernandes, 2007, Spaniol et al., 2009), particularly posterior hippocampus (Poppenk et al., 2013, Ranganath and Ritchey, 2012), whereas familiarity-based retrieval has most commonly been associated with perirhinal cortex (Henson et al., 2003, Skinner and Fernandes, 2007), but also lateral prefrontal, including inferior frontal gyrus (IFG), and temporal regions (for a review, see Skinner and Fernandes, 2007). The present study (1) investigated the brain areas involved when knowledge supports episodic retrieval, using a recognition memory test, with a direct manipulation of knowledge (as opposed to comparing two different tasks), and (2) examined the mechanism underlying the benefits of knowledge, by linking them to brain areas associated with recollection and familiarity. We used three converging approaches to answer these questions.

To better understand how general knowledge influences memory for events, we manipulated whether or not stimuli were known to participants. We used facts drawn from different domains (history, geography, science, etc.) that extensive piloting demonstrated were known (e.g., The composer who worked in deafness was Beethoven) or unknown (e.g., The stick used in the game of shinty is a caman). To examine episodic retrieval during a recognition memory test, we identified, for both known and unknown statements, regions showing greater activity for hits than correct rejections (repetition enhancement—RE)—as RE reflects the formation and retrieval of new episodic representations (Henson, 2003; Henson et al., 2002). We also identified, for both known and unknown statements, regions showing less activity for hits than correct rejections (repetition suppression—RS)—as RS reflects the fluent processing of pre-existing semantic representations (Henson, 2003, Henson et al., 2002).

Second, we examined whether regions showing RE or RS effects related to individual differences in recognition memory performance for unknown versus known statements. Without a clear relationship with behavior, the putative correlates of episodic retrieval may be conflated with other processes such as priming (Dew and Cabeza, 2011, Paller et al., 2007). Therefore, significant correlations between neural repetition effects and behavior would support the idea that different regions support memory retrieval for information with or without pre- existing semantic representations.

Finally, using multi-voxel pattern analysis (MVPA), we examined whether unknown and known stimuli involved different memory representations. Previous studies demonstrated that activity patterns can distinguish old and new recognition memory trials (Rissman et al., 2010). Using a searchlight procedure, we investigated whether representational differences could also classify hits and correct rejections during the recognition test for unknown and known statements. Such a finding would support the idea that episodic retrieval of unknown and known statements differ not only in terms of memory processes, but also in their memory representations.

In sum, we investigated how knowledge supports episodic retrieval. After studying known and unknown statements, participants made old/new recognition decisions about these statements intermixed with new known and unknown statements. We examined (1) the effects of this knowledge manipulation on regions showing RE or RS effects, (2) the correlation between RE or RS effects in these regions and individual differences in recognition memory performance for unknown and known statements, and (3) the difference in memory representations for unknown and known statements as detected by MVPA.