Introduction
Memory is among the cognitive functions that change the most with aging. With the current aging of the population, the study of memory during normal and pathological aging has become a major focus in the neuroscience field (Park and Festini, 2017). Neuropsychological assessment of learning and memory in aging populations frequently relies on the use of learning tasks based on supraspan word-lists (i.e., list lengths that exceed memory span), such as those from the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD; Morris et al., 1989), the California Verbal Learning Test (CVLT; Delis et al., 1987) or the Rey Auditory Verbal Learning Test (RAVLT; Rey, 1970). These tests include measures of short- and long-term recall and recognition, vulnerability to proactive and retroactive interference and learning ability. Although the CVLT yields more information, particularly regarding the encoding strategies used by the participants (Stricker et al., 2002), the RAVLT has long offered more adequate normative data and alternative test forms (Spreen et al., 1998). As reviewed by Saury and Emanuelson (2017), the extended version of the RAVLT can distinguish among the following four domains of learning and memory: (a) short-term recall (STR), (b) learning, (c) interference, and (d) retrieval [immediate recall after interference, delayed recall (DR), and delayed recognition]. Thus, the RAVLT provides numerous memory scores and derivative indices (Spreen et al., 1998). At the neuropsychological level, RAVLT performance (STR, learning, and DR) can discriminate among normal aging, mild cognitive impairment (MCI) and Alzheimer’s disease (AD; Marra et al., 2000; Estévez-González et al., 2003; Goryawala et al., 2015; Bauer et al., 2018) and is a good independent predictor of dementia (Eckerström et al., 2013; Li et al., 2017). Moreover, RAVLT DR performance can discriminate AD from other dementias, such as the behavioral variant of frontotemporal dementia (Ricci et al., 2012) or dementia with Lewy bodies (Bussè et al., 2017). Therefore, the clinical use of the RAVLT to detect memory impairment is supported by the French social security system (ALQP006 CCAM). Taken together, these elements can explain why the RAVLT is currently highly popular as shown by its inclusion in cognitive batteries used for aging studies [e.g., the Alzheimer’s Disease Neuroimaging Initiative (ADNI, Park et al., 2012) and the Canadian Longitudinal Study on Aging (CLSA, Tuokko et al., 2017)].
The present study aimed to further characterize the role of verbal material, i.e., its meaning, in memory assessment using supraspan verbal list-learning tests such as the RAVLT. To achieve our goal, we adopted a two-pronged approach that involved a healthy adult population of 447 participants under the age of 60 to avoid the pronounced effects of aging on memory.
First, to study the role of the lexico-semantic component of the verbal material, we compared memory scores obtained from the RAVLT to those of an alternative version of this test using phonologically similar but meaningless items, i.e., pseudowords. To the best of our knowledge, this issue has not been explored in this manner since certain preliminary works were carried out by Ebbinghaus on himself and by Tulving on a group of six participants (Tulving, 1985). Pseudowords are orthographically legal and pronounceable letter strings without meaning that should mainly activate orthographic and phonological sources of information but only partially (or not at all) lexical information (Mazoyer et al., 1993; Ziegler et al., 1997). Therefore, we were able to evaluate the lexico-semantic benefit by comparing pseudoword from word memory performance. As a first assumption, we expected that word meaning would enable better word than pseudoword memory performance. Moreover, demographic factors have been shown to have a major impact on RAVLT scores in the following manner: performance declines during healthy aging (McMinn et al., 1988; Geffen et al., 1990; Selnes et al., 1991; Mortensen and Gade, 1993; Van Der Elst et al., 2005; Malloy-Diniz and Parreira, 2007; Messinis et al., 2007; Teruya et al., 2009), and women perform better than men (Bleecker et al., 1988; Geffen et al., 1990; Mortensen and Gade, 1993; Aartsen et al., 2004; Maitland et al., 2004; Malloy-Diniz and Parreira, 2007; Messinis et al., 2007). We therefore examined the variability of the RAVLT and the pseudoword memory scores as a function of age and sex.
Second, we collected the participants’ self-reports of the strategies they used during the task to assess whether the self-reported strategies could lead to a better understanding of the explicit mechanisms by which the learning and memory occur. The use of encoding strategies has been linked to greater recall in word-list learning tests (Unsworth et al., 2019). Compared to the CVLT, in which the words are experimentally grouped into four categories to encourage the use of a semantic clustering strategy (Sunderaraman et al., 2013), the RAVLT words do not have a clear semantic relationship. Some authors have argued that serial order might become a preferred strategy in such cases (Vakil and Blachstein, 1994; Meijs et al., 2013). Here, we investigated the nature of the participants’ declarative memory processes without any preconception by collecting their self-reported strategies. Therefore, our second hypothesis was that RAVLT performance could be influenced by the self-reported strategies that the participants used. We also investigated whether these strategies could mediate the well-known effect of age and sex on RAVLT performance.
Discussion
As might be expected, compared to learning pseudowords, learning words (known meaningful items) resulted in improved performance at each memory stage, from STR to DR, via LOT (Table 2). Since in supraspan lists, more items are a priori more difficult to memorize, the longest length of the list cannot explain the better memory scores for words. This word advantage could arise from a detrimental effect of pseudowords’ novelty, which would have an especially strong impact on STR. However, as the word advantage decreased in the following trials, as shown by the proximity between word and pseudoword LOT, the pseudowords became more familiar with repetition. The word advantage observed in the latest stage of memorization thus indicates that words benefit from an effect other than mere familiarity, one that is probably related to the meaning they carry.
Our observation of the age effects on word memory performance (Table 3 and Figure 1) reproduced numerous previous findings showing that age (16–86 years) was associated with a decrease in both the STR and DR of words but did not have a visible effect on LOT (Geffen et al., 1990; Mitrushina et al., 1991; Poreh, 2005; Van Der Elst et al., 2005). In addition, our results showed that age systematically led to a decrease in pseudoword memory performance for each memory stage. The dissimilar effects of age on word and pseudoword performance were particularly important for LOT suggesting that the word-related gain lasts longer, specifically during the learning process.
We also examined the sex effect since the literature has regularly reported for many years that women perform better than men on STR and DR (Bolla-Wilson and Bleecker, 1986; Geffen et al., 1990; Aartsen et al., 2004; Van Der Elst et al., 2005; Gale et al., 2007; Messinis et al., 2007; Badcock et al., 2011). From the perspective of this sex contrast (Table 4 and Figure 2), our results reproduced these previous findings: women performed better than men on the STR and DR of words, but there was no visible sex difference in word LOT performance (Teruya et al., 2009). Regarding the difference in performance for words and pseudowords, the gain provided by the words emerged only among the women on the STR (Figure 2). This result demonstrates that the word gain can appear during early memory stages, which supplements recent studies highlighting the nature and importance of short-term semantic memory (Campoy et al., 2015; Aizpurua and Koutstaal, 2018).
To better understand the explicit mnemonics involved in word and pseudoword memory, we determined the strategies the participants used to retain the lists of words and pseudowords. We found that words and pseudowords elicited the same types of strategies (Table 1). Two main strategies were reported, namely, a strategy based on associations of meanings (lexical/semantic binding) of the items and a strategy based on the temporal contiguity (position on the list) of the items. These two strategies, called Meaning and Position in the present study, correspond to memory strategies previously described in free recall studies as subjective and serial clustering (Meijs et al., 2013) or semantic proximity effects and temporal contiguity (Sederberg et al., 2010). These two strategies were contrasted with a third strategy, namely, the Listening strategy, which has also been previously described in free recall tasks as a rehearsal mnemonic technique (Herrmann, 1987). These two strategies were also used in combination, an approach that we called the Dual strategy. Thus, from the complexity perspective, the four strategies could be ordered as follows: the Listening strategy was the simplest, a minimal strategy induced by the item presentation; the Position strategy was equivalent to the Meaning strategy; and the Dual strategy was the most complex strategy, combining two strategies.
Regarding the distribution of the four strategies according to type of item (Table 5), word memory relied on the use of more complex strategies than pseudoword memory, which led to a reduced use of the Listening strategy and an increased use of the Dual strategy. The Meaning strategy alone or in combination was the dominant strategy used for word memory (64% of the respondents) and, quite surprisingly, pseudoword memory (57% of the respondents). The main difference between the words and pseudowords concerned the nature of the Meaning strategy, which consisted of thematic and taxonomic associations for the words and lexical associations for the pseudowords. This result suggests that auditory supraspan list-learning tests encourage memory based on associations with familiar and meaningful knowledge in a hierarchical manner (depending on the cues the item provides), from phonemes to lexicon for pseudowords and from lexicon to semantic binding for words. This interpretation is consistent with a previous finding that participants performed equally well on the RAVLT and CVLT when the semantic categorization of words was experimentally forced in the CVLT (Crossen and Wiens, 1994). We cannot excluded the possibility that the fixed order of list presentation (words followed by pseudowords ∼3 hours later) may have favored the use of a Meaning strategy for pseudowords and constitutes a potential limitation to this study. Nevertheless, the subtypes of the Meaning strategy were not strictly the same for the words and pseudowords (Table 1). Thus, it appears that exposure to known phonological entities, regardless of their meaning, encouraged the use of lexico-semantic associations.
Subsequently, we found that the effectiveness of the self-reported strategies varied between word and pseudoword memory tasks (Table 6 and Figure 3). The Listening strategy led to the lowest word and pseudoword memory performance. However, the Meaning strategy, used alone or in combination with the Position strategy, improved word memory, while the Dual strategy was the most efficient for the pseudoword memory, suggesting that for pseudowords, the Meaning and Position strategies potentiated each other. According to Worthen and Hunt, list-learning mnemonic techniques can be divided into organization mnemonic processes, which emphasize intralist associations (i.e., all information to be remembered is linked together), and elaboration mnemonic processes, which rely on extralist cues (e.g., meaningfulness, mental imagery) (Worthen and Hunt, 2017). Furthermore, these authors argued that the combined use of the organization and elaboration strategies is the most effective mnemonic. Here, the Position strategy was analogous to an organization mnemonic, while the Meaning strategy was analogous to an elaboration mnemonic. The combination of organization and elaboration mnemonics appeared to be the most effective strategy for pseudoword memory (Figure 3 right). However, this did not apply to word memory, as only the elaboration mnemonic enabled the best performance (Figure 3 left). Moreover, concerning word memory, the Position strategy led to performances as low as those resulting from the use of Listening strategy, which is congruent with the previous findings showing that rote rehearsal is the least effective method for free recall tasks (Herrmann, 1987). Taken together, these two observations strongly suggest that the Meaning strategy was an extremely efficient mnemonic that had the potential to maximize verbal list-learning memory performance.
Finally, we were interested in determining whether the age and sex effects on memory performance could have been mediated by the previous self-reported strategies. We found that aging significantly decreased the likelihood of using any of the advanced strategies, i.e., all strategies other than Listening, for word and pseudoword memory (Table 7), which, in turn, impaired memory performance, as assessed by the CMAs (Table 8). This result strongly suggests that an increase in age produces a nonspecific decrease in the use of advanced strategies. We also observed that being a woman increased the likelihood of using the Meaning and Dual strategies compared to the Position strategy, for word memory (Table 7) and that this effect could to some extent mediate women’s superior word STR performance (Table 8). Thus, the gain in STR performance provided by the words could be specifically related to the fact that women preferentially use the Meaning strategy (alone or in combination with the Position strategy). This result is congruent with previous observations obtained using the CVLT (Kramer et al., 1988, 1997) and supports Andreano and Cahill’s hypothesis that women’s verbal memory advantage depends on encoding at the semantic level (Andreano and Cahill, 2009).
From a neuropsychological perspective, the distinction between normal and pathological aging and the early detection and identification of neurodegenerative diseases are core features of investigations. Since we found that RAVLT scores can capture semantic strategies, we can assume that the RAVLT enables the detection of the early semantic impairment, particularly word-finding difficulties, that characterizes AD (Herlitz and Viitanen, 1991; Croisile et al., 1996; de Lira et al., 2011; Verma and Howard, 2012). Extending this idea, the semantic sensitivity of the RAVLT might explain its previously observed ability to detect the disease in its early stages and, notably, to discriminate AD from MCI (Marra et al., 2000; Estévez-González et al., 2003; Goryawala et al., 2015; Bauer et al., 2018). Moreover, we observed that the word memory advantage of women was mediated by the use of the Meaning strategy (Table 8). This sex-specific mediation can lead to a sex-specific cognitive advantage that compensates for the neurological evolution of AD and, as previously hypothesized by Sundermann et al. (2017), interferes with the early detection of the disease in women. Therefore, when the RAVLT is used for diagnostic purposes, we join the authors in advocating the establishment of sex-specific cutoff scores (Nebel et al., 2018) or sex-adapted performance gaps between healthy individuals and those with dementia (Chapman et al., 2011; Gale et al., 2016).
Since the introduction of the RAVLT in the 1950s, new theories about memory have emerged. One of the most widely recognized is Tulving’s theory, which distinguishes nondeclarative (implicit) memory from declarative (explicit) memory. Moreover, Tulving describes two types of declarative memory, as follows: (1) episodic memory, corresponding to personal memories (or contextualized events) and referring to recollection, a key concept related to autonoetic consciousness (a subjective sense of self over time) and (2) semantic memory, corresponding to general knowledge and referring to familiarity, which is associated with noetic consciousness (Tulving, 2002). At this stage, it could be useful to relate Tulving’s model of declarative memory to our observations and to discuss the ability of the RAVLT to assess Tulving’s memory systems in light of our results.
Tulving previously specified that “there is no necessary correlation between behavior and conscious experience and in that sense the traditional research (i.e., such sterile situations as list-learning experiments) was not concerned with episodic memory” (Tulving, 2002). Nevertheless, in the early and mid-1990s, free recall tasks (used as declarative memory tests) and RAVLT scores (DR in particular) began to be associated with the assessment of episodic memory (Roediger, 1990; Litvan et al., 1991; Moffot et al., 1994; Eustache et al., 1995; Kirkby et al., 1995), and this association became increasingly frequent over time, occurring in 13–20% (PubMed) or 33–43% (ScienceDirect) of the RAVLT literature over the last 6 years (keyword cross-search from January 2020). Currently, it is not unusual to find a direct association between the RAVLT and the assessment of verbal episodic memory (Moradi et al., 2017; Barulli et al., 2019; Putcha et al., 2019; Sudo et al., 2019) or even to find the RAVLT being used to validate other verbal episodic memory tests (Morrison et al., 2018). Since episodic memory implies explicit retrieval (Tulving, 1995), the exploration of self-reported strategies is a direct way to investigate whether episodic memory is assessed by the RAVLT. Indeed, a strategy linked to episodic memory could make some autonoetic references to the context of the test (e.g., a participant associating “handle” with the “handle of the door in the experiment room”) or an item’s association with a personal contextualized recollection (e.g., a participant associating “fireplace” with the “fireplace from his childhood”). This type of autonoetic strategy did not emerge as a regularly used strategy in our results, leading to the conclusion that the RAVLT seems unable to assess episodic memory to a reasonable degree, which is similar to the conclusion reached by previous authors (Tulving, 1985, 2002; Van der Linden, 2004; Desgranges and Eustache, 2011). Consistent with Gavett and Horwitz’s assertion that list-learning test interpretation lacks the construct validity to allowed unbiased estimates of episodic memory ability (Gavett and Horwitz, 2012), our results supplement those reported by Casaletto et al. (2017), who focused on total immediate recall performance during learning trials and concluded that interpreting learning scores as equivalent to episodic memory may be erroneous. To avoid any confusion regarding the purpose of the RAVLT, and considering its widespread use, we encourage the use of simple terminology, i.e., “(supraspan) verbal memory,” to briefly characterize the RAVLT assessment. In addition, it seems appropriate to extend this caution to all supraspan verbal list-learning tests.
Within the framework of Tulving’s memory model, the present results lead to consider the RAVLT scores as being affected by semantic memory. Indeed, the present results show that general knowledge of the meaning of words (the thematic and taxonomic associations used in the Meaning strategy) maximize performance when used as a mnemonic (Figure 3). Moreover, the Meaning strategy was widely used among the participants (Table 5) and was found to mediate women’s advantage regarding the STR of words (Table 8). Therefore, RAVLT performance appears to be highly dependent on this general knowledge and, thus, on semantic memory. Nonetheless, the self-reported strategies explained only 28% (5.6% imputable to the fixed effects) of the variance in RAVLT performance and mediated 2–8% of the age and sex effects on memory performance. These effect sizes suggest that some memory processes escaped the participants’ self-assessment and that the RAVLT scores surely assessed some implicit mechanisms. This hypothesis is compliant with Tulving’s definition of semantic memory, which is declarative knowledge characterized by implicit retrieval (Tulving, 1995). In addition, a quick review of the literature reveals the possible involvement of early implicit encoding processes, such as working memory and executive functions. Although the age and sex effects on memory performance that we observed were adjusted for WMC, other dimensions of working memory and executive functions may be involved. A recent study demonstrated that semantic encoding is automatic in verbal short-term memory (Campoy et al., 2015). Furthermore, the recency effect, which plays a significant role in RAVLT scores, is largely driven by verbal attention (Griffin et al., 2017). The recency effect has been conceptualized in immediate free recall as implicit learning coupled with a particular mode of retrieval that may, but need not, be conscious and explicit (Baddeley and Hitch, 1993). Finally, the present observation that aging nonspecifically decreases the use of advanced self-reported strategies (Table 7) also argues for the involvement of executive functions. This interpretation is consistent with previous observations that normal aging is associated with more difficulty accessing lexico-semantic operations and representations due to a slowdown in executive functions rather than concept loss per se (Baciu et al., 2016). A recent study also showed selective inability to recall RAVLT midlist items in patients with a selective mild executive function deficit (Consonni et al., 2017). Moreover, some authors noted the probable combined effect of attentional and short-term memory processes on the immediate recall score (Gavett and Horwitz, 2012) and a high degree of overlap between verbal memory and executive functioning (Duff et al., 2005). While further studies are needed to clarify the nature of these associations, our results contribute to previous studies suggesting the existence of an association between RAVLT scores and working memory or executive functions.
In conclusion, our results showed that word meaning provides a significant gain in recall performance on the RAVLT. In particular, word meaning enables access to meaning-based memory strategies that optimize verbal memory performance. Moreover, the beneficial role of the Position memory strategy for pseudoword recall was potentiated by a lexical-based memory strategy. These two observations strongly suggest that pseudoword and word-list learning memory performance depends on strategies that are based on general semantic knowledge. Within the framework of Tulving’s declarative memory model, our results indicate that the verbal list-learning test free recall scores are affected by semantic memory. Since the self-reported strategies were not autonoetic, we conclude that the RAVLT is not suitable for episodic memory assessment as it is increasingly referred to in the literature.