Remembering and Knowing: Using another’s subjective report to make inferences about memory strength and subjective experience

Highlights

• Participants can use others’ subjective reports to classify memory in the RK paradigm.

• Know and Familiar are distinguishable classifications of recognition memory success.

• Confidence and subjective experience are differentially attributed to others’ reports.

• Suggests confidence and subjective experience are two different types of evaluations.

Abstract

The Remember–Know paradigm is commonly used to examine experiential states during recognition. In this paradigm, whether a Know response is defined as a high-confidence state of certainty or a low-confidence state based on familiarity varies across researchers, and differences in definitions and instructions have been shown to influence participants’ responding. Using a novel approach, in three internet-based questionnaires participants were placed in the role of ‘memory expert’ and classified others’ justifications of recognition decisions. Results demonstrated that participants reliably differentiated between others’ memory experiences – both in terms of confidence and other inherent differences in the justifications. Furthermore, under certain conditions, manipulations of confidence were found to shift how items were assigned to subjective experience categories (Remember, Know, Familiar, and Guess). Findings are discussed in relation to the relationship between subjective experience and confidence, and the separation of Know and Familiar response categories within the Remember–Know paradigm.

Introduction

When the Remember–Know (RK) paradigm was first introduced by Tulving (1985), Remember and Know responses were posited as reflecting autonoetic and noetic consciousness based on retrieval from episodic and semantic memory respectively. In recognition memory tests, old items were categorised as Remembered when the participant retrieved from memory something they had thought or experienced at the time of encoding and categorised as Known when the participant was aware that the item had been on the study list but could not recall anything experienced for the item at that time (Gardiner and Richardson-Klavehn, 2000, Tulving, 1985). Since Tulving’s first examination of Remember and Know subjective experiences two main issues have dominated the RK literature: The relationship between subjective experience and confidence, and the relationship between the subjective states of Remembering and Knowing and the underlying processes of recollection and familiarity. The experiments presented here examined both these issues using a novel methodology.

The relationship between subjective experience and confidence has been acknowledged from the conception of the RK paradigm when Tulving (1985) demonstrated that Remembered items were given higher confidence ratings than were Known items. This has been replicated in numerous experiments (e.g., Brewer and Sampaio, 2006, Gardiner and Java, 1990, Rajaram, 1993, Rajaram et al., 2002, Rotello and Zeng, 2008, Tulving, 1985; and Yonelinas, 2001) with the suggestion that information from subjective awareness is used to judge how confident one is (Gardiner, 2001). The relationship between subjective experience and confidence has also been the subject of debate between advocates of dual-process and single-process models of recognition. Dual-process models (e.g., Mandler, 1980, Tulving, 1985, Yonelinas, 1994, Yonelinas, 2002) assume that two distinct processes underlie recognition and that successful recognition is determined by contributions from both processes. Conversely, single-process accounts propose that recognition relies on only one continuous dimension of familiarity, confidence, or memory strength, and successful recognition is determined by the strength of this single dimension (e.g., Donaldson, 1996, Dunn, 2004). For comprehensive reviews of the literature that summarise the opposing viewpoints see Yonelinas, 2002, Diana et al., 2006, Dunn, 2004, Dunn, 2008, Parks and Yonelinas, 2007, and Wixted and Stretch (2004). However, despite continued theoretical debate, how the layperson understands, conceptualises, and acts upon subjective experiences of memory and confidence in memory are still not well understood.

Experiments that have compared RK and confidence judgments have shown different patterns of responding elicited by these two judgment types. Early studies operationalised confidence as a two-category scale of Sure–Unsure and compared proportion of items assigned against proportion assigned to Remember and Know. Using this comparison, these two judgment types were shown to not elicit the same patterns of responses using word/non-word (Gardiner and Java, 1990, Rajaram et al., 2002) or masked priming manipulations (Rajaram, 1993; and with the addition of a Guess category Tunney & Fernie, 2007). More recently, differences between RK and confidence judgments have been demonstrated by comparison of verbal reports at recognition to ‘think-aloud’ verbalisations made during study (McCabe, Geraci, Boman, Sensenig, & Rhodes, 2011). Here justifications for Remembered items were more likely to contain recollection of details verbalised during encoding than were justifications for high-confidence items. Taken together, these findings demonstrate that while judgments of confidence and subjective experience may be interrelated, RK judgments are not made solely on the basis of confidence and the two are not “experimentally interchangeable” (Rajaram et al., 2002, p. 234).

RK and confidence judgments have also been compared using a larger scale to measure confidence and analysis has focused on the resulting receiver-operating-characteristic (ROC) curves. Using this methodology, higher confidence has been consistently found to be associated with Remember responses compared to Know responses (Rotello et al., 2004, Rotello et al., 2005, Slotnick, 2010, Wixted and Stretch, 2004, Yonelinas, 2001, Yonelinas et al., 1996). Different modelling approaches have aimed to determine whether the data are better explained by single- or dual-process models (e.g., Donaldson, 1996, Dunn, 2004, Dunn, 2008, Gardiner et al., 2002, Macmillan et al., 2005, Wixted and Mickes, 2010). In a review, Gardiner (2008) concluded that most of these available models provide a reasonably good fit to the data but that as the technical complexity of the models has increased it has become more and more difficult to see how to discriminate between them empirically (see also O’Connor, Guhl, Cox, & Dobbins, 2011). Rather than attempting to provide evidence supporting one or other of the models, the experiments presented here examine how people make and understand judgments of subjective experience and confidence. In particular, we were interested in whether the content and nature of people’s subjective reports was enough to ‘recover’ information about confidence and experiential state from those who actually performed the memory test. Our goal was to examine how people interpret evidence from experiential reports in terms of both confidence and categories of subjective experience.

A second issue of debate within the RK paradigm is the relationship between the states of Remembering and Knowing and the processes of recollection and familiarity. In their review, Gardiner and Richardson-Klavehn (2000) identified interpretation of Know responses as “the most vexatious problem in the remember/know paradigm” (p. 238). The root of this problem lies in, firstly, whether Know responses are defined to participants in terms of familiarity or certainty; and secondly, whether Know responses are interpreted as reflecting an underlying process of familiarity or a state of knowing. For example, some researchers choose to ask participants to make Remember–Familiar judgments instead of Remember–Know. Donaldson, MacKenzie, and Underhill (1996) assert “…familiar rather than know was used to indicate nonrecollection, because the word know carries a connotation of certainty that is inconsistent with a confidence rating that indicates lack of certainty. Participants find it hard to say that they are unsure that an item was there but that they know it was” (p. 487, italics in original). Other researchers choose to encompass both familiarity and knowing within one response category, for example, Kelley and Jacoby (1998) define Knowing as “…the inability to recollect any details of the study presentation in combination with a feeling of familiarity or certainty that the word was studied” (p. 134, italics added). The issue of how Know responses are interpreted in terms of a familiarity process or a subjective state of knowing is not helped by many research reports not including the exact wording used to define response categories to participants.

Some researchers have separated Know and Familiar as response options. In their study of student learning, Conway, Gardiner, Perfect, Anderson, and Cohen (1997) found that participants could differentiate K and F responses; a K response indicating that they ‘just knew’ the answer. Students took multiple-choice question exams following four psychology lecture courses and students assigned answers to one of four categories: Remember (R), Know (K), Familiar (F), or Guess (G). At initial testing, higher performing students designated more answers as Remember than did poorer performing students; however, at re-test, these students assigned more answers to Know than to Remember. Conway et al. discuss this ‘R-to-K shift’ as reflecting a change in knowledge representation from episodic to semantic memory brought about by loss of episodic details from memory and the emergence of conceptual organisation. Evidence of the R-to-K shift has been demonstrated in similar studies of student learning by Barber, Rajaram, and Marsh (2008) and Herbert and Burt, 2001, Herbert and Burt, 2003, Herbert and Burt, 2004 and a study exploring learning of rare word definitions by Dewhurst, Conway, and Brandt (2009). Dewhurst et al. also found that Remember and Know responses did not differ in accuracy or confidence, though both were significantly more accurate and associated with higher levels of confidence than Familiar and Guess responses.

It could be argued that the separation of Know and Familiar might only apply to learning of rich materials over a longer time period (e.g., Conway et al., 1997, Herbert and Burt, 2001, Herbert and Burt, 2003, Herbert and Burt, 2004). In typical recognition experiments with lists of unrelated words, where memory is measured shortly after study, it would be difficult for participants to integrate or semanticise the studied information into any body of knowledge and therefore the experiential state of ‘Knowing’ might not appear to be applicable. However, Dewhurst et al. (2009) asked participants to learn rare word definitions specifically because these were unrelated facts that would be less easy to integrate into a wider schema. While these materials are arguably more meaningful than word lists, they are still less meaningful than the academic material used previously. Dewhurst et al.’s (2009) participants assigned 10% of items to Know at first test – a situation analogous to a single-time-point recognition experiment. While this proportion is lower than the 22% assigned to Know in Conway et al. (1997), it demonstrates that even for less meaningful learning, participants considered the Know response to reflect their experiential state for some items only 5 min after study with no opportunity for integration into a wider body of knowledge.³ These 10% of unrelated facts were ‘just known’ without any recollection or feelings of familiarity regarding the study episode only a short time after said study episode. This suggests that the separation of Know and Familiar might well be applicable to the subjective experiences in typical episodic recognition paradigms where the materials have lower intrinsic meaning (e.g., lists of unrelated words). Indeed, unpublished experimental work by the current authors has demonstrated that the four categories can be used reliably in episodic tasks and that K and F responses differentiate on key characteristics such as accuracy (Williams et al., 2013a, Williams et al., 2013b, Williams et al., 2009).

Participants thus appear to appreciate the differences between the experiential states of Remember, Know, Familiar, and Guess. Also, for a range of materials, the subjective experiences associated with learning demonstrate consistent patterns over time. In the current experiments, we examined whether these concepts map onto the evidence provided in justifications of recognition that differ in subjective experience and confidence level. Here our focus was on how individuals assess subjective information relating to memory experiences outside the context of a memory task.

A related literature considers how definitions of subjective experience categories can influence participants’ RK judgments. Rotello et al. (2005) and McCabe and Geraci (2009) compared traditional Remember instructions (after Rajaram 1993) with more conservative Remember instructions that constrained recollection to the study episode. More conservative instructions led to participants using the Remember response less frequently. Examining Know definitions, Geraci, McCabe, and Guillory (2009) emphasised Know responses as either highly confident or as less confident. When confidence was emphasised, Remembering and Knowing were differently influenced by words and non-words whereas confidence judgments were not. When confidence was not emphasised, patterns were similar for Remember–Know and Sure–Unsure responses. Geraci et al. (2009) concluded that the wording of instructions can have important theoretical implications for our understanding of Remembering, Knowing, and confidence. In the current experiments, source-specific instructions were included in definitions of Remember, Know, Familiar, and Guess responses, and Remember, Know, and Familiar definitions were each accompanied by a real-world example of the subjective experience, see Table 1. Geraci et al. (2009) also advocated the use of manipulation checks post-recognition to ensure that participants are utilising the response categories as instructed. One common manipulation check is to ask participants to justify their recognition responses, i.e., state why they said they recognised that item. Gardiner, Richardson-Klavehn, and Ramponi (1997) used this form of manipulation check, analysed the content of the obtained statements, and published a list of 270 justifications of Remember, Know, and Guess recognition decisions in Gardiner, Ramponi, and Richardson-Klavehn (1998). These justifications were used as stimuli in the current experiments.⁴

In Gardiner et al. (1997), after participants had assigned each item to Remember, Know, or Guess, the experimenter chose two responses from each category at random and asked the participant to explain their response. Importantly, the emphasis was on what led to the recognition decision; participants were not asked to justify why they had assigned a word to a particular category of subjective experience. Two expert raters then classified the justifications by their salient characteristics (Gardiner et al., 1998). Remember justifications were found to involve: intra-list associations, extra-list associations, item-specific images, and the item’s physical features; which Gardiner et al. related to participants’ use of encoding strategies. Remember justifications also included self-reference; this was suggested as reflecting items automatically triggering awareness of a personal memory. For Know and Guess justifications, transcripts were shorter and used rather limited vocabulary. Know justifications lacked recollection of specific contextual details and were instead characterised by feelings of familiarity, just knowing, thinking a word occurred, or reporting of the absence of recollective details. Guess justifications demonstrated inferences, speculations, and other judgmental strategies employed by participants.

To assess the relationship between subjective experience and confidence expressed in the justifications, Gardiner et al. (1998) coded which Know and Guess justifications indicated certainty or uncertainty by counting phrases such as ‘sure’/‘not sure’, ‘confident’/‘not confident’, or ‘I know’/‘I think’. For Know responses, 25% of justifications were rated as indicating certainty, and 11–20% as indicating uncertainty. For Guess justifications, 72–77% were rated as uncertain and none were rated as certain. Through their exploration of how participants justified recognition decisions, Gardiner et al. demonstrated that justifications for different subjective experiences reflected access to different memory processes at study and retrieval. Furthermore, the justifications also revealed differences in confidence associated with different subjective experiences.

The current experiments explore how people understand reports of memory experiences in terms of confidence and categories of subjective experience. The novel approach here is that instead of asking participants about their own ‘first-person’ experiences during a memory task, we placed them in the role of ‘memory expert’ and asked them to classify other people’s reports of recognition decisions. This ‘third-person’ approach is adopted to examine whether categories of subjective experience can be thought of as natural conceptualisations of memory experiences that people readily use and apply. Here three internet-based questionnaires are reported, in which the justification statements of Gardiner et al., 1997, Gardiner et al., 1998 were used as stimuli. In Experiment 1, participants were asked to assign a confidence rating to each justification statement. These confidence ratings were then used to select justifications for use as stimuli in Experiments 2 and 3 where participants were asked to categorise justifications as Remember, Know, Familiar, or Guess when confidence was manipulated. Our goal was to examine whether, by changing the confidence associated with an item, it was possible to shift interpretations of the category of subjective experience.

In experimental tasks, Remember responses are typically associated with higher confidence than Know responses (Brewer and Sampaio, 2006, Gardiner and Java, 1990, Rajaram, 1993, Rajaram et al., 2002, Rotello and Zeng, 2008, Tulving, 1985; and Yonelinas, 2001). Experiment 1 was designed to examine whether this relationship was replicated when it is not the participants’ own contents of memory under study – whether ‘third-person’ understanding of subjective experiences links experiential state to confidence in the same manner that ‘first-person’ tasks have found. If people are able to interpret others’ justifications in terms of the underlying memory processes, and these are linked to confidence, then each category of justification statement should be assigned different confidence ratings.

Experiments 2 and 3 manipulated the confidence associated with a report of subjective experience in a method somewhat akin to the experimental paradigms of Geraci et al., 2009, McCabe and Geraci, 2009, and Rotello et al. (2005). In Experiment 2, confidence was manipulated through use of justification statements that had obtained High, Medium, and Low confidence ratings in Experiment 1. In Experiment 3, confidence was manipulated by each justification statement being accompanied by a confidence value. In both experiments, instead of giving a confidence rating for the item, participants assigned justifications to a subjective experience category. If participants’ interpretations of the memory processes underlying the justifications are influenced by the confidence associated with the justification, then this should be reflected in how the items are assigned to the categories of subjective experience.