EEG-based Univariate and Multivariate Analyses Reveal that Multiple Processes Contribute to the Production Effect in Recognition

Reading words aloud enhances memory relative to reading them silently (see MacLeod & Bodner, 2017, for a brief review). A handful of early studies reported this effect (e.g., Hopkins & Edwards, 1972), but it was only after MacLeod et al. (2010) delineated this phenomenon and dubbed it the production effect (PE) that this effective encoding strategy received systematic attention from researchers. Since MacLeod et al. (2010), over 50 studies of the PE have been reported (e.g., Bodner et al., 2020; Forrin et al., 2012; López Assef et al., 2021; Lin & MacLeod, 2012; Ozubko et al., 2012a; Saint-Aubin et al., 2021; Zormpa et al., 2019), and meta-analyses have confirmed that it improves recognition memory (e.g., Bodner et al., 2014; Fawcett, 2013).

Current research on the PE has shifted from establishing the effect to understanding the mechanism by which it works. Much of this research has focused on whether the PE is the result of increased distinctiveness in memory or increased memory strength (e.g., Bodner et al., 2016; Fawcett & Ozubko, 2016; MacLeod & Bodner, 2017; Ozubko & MacLeod, 2010). The original and most common account of the PE is that reading aloud enhances distinctiveness. MacLeod et al. (2010) emphasized that reading aloud involves unique phonological and articulatory processing of items relative to silent reading. Retrieval of these distinctive elements of encoding can aid recognition later on, possibly also leading participants to recall having read them aloud. By a strength account, in contrast, reading aloud simply enhances the activation of the memory traces of studied items, rendering them more familiar at test (Bodner & Taikh, 2012).

Several studies have attempted to disentangle the distinctiveness and strength accounts (e.g., Bodner & Taikh, 2012; Bodner et al., 2020; Fawcett & Ozubko, 2016; Kelly et al., 2022; MacLeod et al., 2010; Ozubko et al., 2014; Roberts et al., 2022), with mixed success. More recently, Fawcett and Ozubko (2016) provided evidence that both distinctiveness and strength can contribute to the PE. Following up work by Ozubko et al. (2012), Fawcett and Ozubko collected remember/know judgments from participants during a recognition test. Based on a dual-process model of recognition, remember judgments were used to index recollection and know judgments were used to index familiarity (after Tulving, 1985). Familiarity is often characterized as an undifferentiated feeling that a stimulus was recently experienced, in line with increased memory familiarity. For example, we recognize that a person looks very familiar, even if we cannot recall any details about them. On the other hand, recollection is typically construed as the ability to vividly and consciously re-experience a past event and the context surrounding it, such as recollecting a person’s name and where and when we last encountered them when we recognize them. The common view is that distinctiveness promotes recollection, whereas memory strength promotes familiarity. Fawcett and Ozubko (2016) found that reading aloud increased both recollection and familiarity, suggesting that reading aloud increases both distinctiveness and strength, thus providing support for a dual-process account of the PE (see also MacLeod & Bodner, 2017; Yonelinas, 2000).

In addition, some researchers have posited that production might increase attention to items during encoding (MacDonald & MacLeod, 1998; Varao Sousa et al., 2013). Consistent with an attentional component, participants who read aloud report less distraction, and continuous noise has little effect on the PE (Mama et al., 2018). The sole functional magnetic resonance imaging (fMRI) study of the PE found that hippocampal and parahippocampal gyri activation was greater at encoding when reading aloud than when reading silently, consistent with reading aloud requiring greater attention allocation (Bailey et al., 2021).

Neural technologies such as fMRI and EEG can be used to clarify the mechanisms underlying the PE. To date, only Hassall et al. (2016) have used event-related potential (ERP) to explore the PE. ERP data were recorded during the study phase after participants were cued whether to read aloud or silently (i.e., the intention phase) rather than during execution of the task, to prevent contamination of EEG data by head movement. Hassall et al. found a larger P300 (more precisely, the P3b) for aloud than silent trials, which they interpreted as a feature of either more distinctive encoding or greater attentional engagement. However, this study did not directly analyze the ERP amplitude during the task execution phase or during the recognition test, nor were responses separated based on recollection versus familiarity.

Recently, Bailey et al. (2021) collected fMRI data during both study and test phases of a PE study. Importantly, aloud and silent conditions were tested alongside an aloud control in which participants said the word "check" aloud rather than the actual word on the screen, to control for phonological and articulatory processing. The aloud condition had stronger activation in areas related to motor, somatosensory, and auditory processing than the silent or control conditions (which did not differ), consistent with the distinctiveness account. Evidence for production increasing attention was also reported, as described earlier. In the recognition test phase, the aloud and control conditions activated areas associated with articulation and auditory processing. In addition, in the aloud-silent contrast during the test phase, the fusiform gyrus (containing the visual item form region) was more activated in the aloud condition, which may reflect more vivid recollection of aloud words. In conclusion, Bailey et al. found evidence for distinctiveness and attention underlying the PE.

Our study is the first to use ERP to attempt to separate and identify the contributions of recollection and familiarity to the PE. Numerous studies of recognition memory suggest that the late positive complex (LPC) component varies with recollection and that the FN400 component varies with familiarity; thus, the LPC old/new effect (difference) and FN400 old/new effect (difference) have become standard measures of recollection and familiarity, respectively (e.g., Bridger et al., 2012; Curran, 2000; Madore et al., 2020; Rugg & Curran, 2007; Wang et al., 2021). The LPC old/new effect reflects more positive amplitude for old items relative to new items in the left parietal area at 500-800 ms after stimulus onset (Curran & Friedman, 2004; Forester et al., 2019; Madore et al., 2020), indicative of enhanced recollection. The FN400 old/new effect is a frontally distributed negative ERP component from 300-500 ms after stimulus onset. This component exhibits more positive amplitude for correctly recognizing old items relative to new items, indicative of enhanced familiarity (Mecklinger & Jäger, 2009; Wang et al., 2021).

Following Bailey et al. (2021), our study included three encoding conditions: aloud, silent, and an aloud control task in which participants produced the equivalent of the word "check" in Chinese. Each study phase trial was divided into an intention subphase and an execution subphase (see Figure 1). Participants prepared their response during the intention subphase, then performed it during the execution subphase (cued by color change). We expected to obtain evidence in the EEG data for production enhancing attention in the intention subphase. Rather than using a random mixture of the three tasks, a blocked design was used to prevent participants from performing the wrong task or from needing to inhibit reading aloud (see Figure 1). After studying the items, participants made remember/know/new recognition judgments during the test phase (see Figure 1). We then analyzed the EEG data from the intention subphase of the study phase and from the recognition judgments in the test phase.

Behaviorally, based on Fawcett and Ozubko (2016), we expected to obtain more remember judgments (indexing greater distinctiveness/recollection) and higher familiarity estimates (indexing greater memory strength) for items that were studied in the aloud condition than in the silent and control conditions. In the ERP measures, if reading aloud also enhances attention, we expected larger P3b amplitudes in the study phase for aloud trials than for silent or control trials, representing top-down attentional processing (Goris et al., 2018; Strobel et al., 2015). During the test phase, in the aloud condition we expected to observe a larger LPC old/new effect, indicating greater distinctiveness, and a larger FN400 old/new effect, indicating greater memory strength, relative to the silent or control conditions.

A novel aspect of our study was that we used Multivariate Pattern Analysis (MVPA) to confirm the influences of production on ERP components. MVPA is a multivariate analytical technique and is typically used when referring to the practice of characterizing (decoding) the difference between experimental conditions based on their patterns of brain responses (Fahrenfort et al., 2018). MVPA involves training a classifier (a pattern classification algorithm) to distinguish different patterns of brain activity associated with different experimental variables of interest. MVPA is more sensitive than conventional ERP analysis because it uses whole-brain activity to depict neural activity patterns over time (de Vries et al., 2019; Li et al., 2022; Sharifian et al., 2021), and because the neural stability of decoding performance can be analyzed (King & Dehaene, 2014). MVPA was used to help determine whether the old/new effects associated with the PE arise early during the recognition judgment task when the FN400 is observed, or later when the LPC is observed. If the PE is more dependent on familiarity, we should observe early decoding between produced and new items, whereas if it is more dependent on recollection, decoding should occur later (i.e., more than 500 ms after stimulus onset; Larzabal et al., 2020; Turner et al., 2017).

留言 (0)

沒有登入
gif