Distinct FN400/N400 memory effects for perceptually fluent and disfluent words

Highlights

• More old responses (hits) were given to blurry words in more demanding Experiment 1.

• This was accompanied by larger FN400/N400 and LPC old/new effects for blurry words.

• No effects of fluency on recognition were found in less demanding Experiment 2.

• FN400 for blurry words and parietal N400 for clear words were found in Experiment 2.

• Early perceptual fluency ERP effects were found in both experiments.

Abstract

Recognition memory studies have shown that increased perceptual fluency results in more “old” responses and, presumably, increases familiarity. However, the exact neural mechanisms of these effects remain unresolved. We conducted two ERP experiments in which participants encoded words and performed a recognition test where fluency was manipulated by changing clarity of test words (half of them were clear or blurry). In the more demanding Experiment 1, we found a reversed effect of fluency on recognition (more hits for blurry words), which was accompanied by larger N400 and LPC old/new effects for blurry words. For high confidence responses, the topography of N400 shifted towards frontal electrodes (the FN400 for blurry words). In the less demanding Experiment 2, no behavioral differences between clear and blurry words were observed. However, there was a discrepancy in the ERP results, with the frontal FN400 for blurry words and the parietal N400 for clear words, suggesting that distinct neural pathways can support familiarity-based recognition for clear and blurry items. In both experiments, early perceptual fluency ERP effects were also observed. The results indicate that both semantic processing and familiarity can be enhanced by perceptual fluency and contribute to recognition judgments, depending on the interpretations of fluency.

Introduction

Fluency is typically regarded as the subjective experience of ease in information processing (Alter & Oppenheimer, 2009). This experience may arise from a wide range of mechanisms, but all of them produce similar consequences, such that more fluent stimuli gain more favorable assessments than less fluent stimuli. For example, easily retrieved stimuli seemed to belong to more populated categories (Tversky & Kahneman, 1973) and were liked more (Bornstein and D'Agostino, 1992) than stimuli that were difficult to retrieve, stimuli with higher contrast were judged as more pretty than stimuli with lower contrast (Reber, Winkielman, & Schwarz, 1998), and statements written in fluent fonts seemed to be truer (Reber & Schwarz, 1999) and induced more confidence (Alter, Oppenheimer, Epley, & Eyre, 2007) than statements written in disfluent fonts. It was even demonstrated that financial stocks with more easily pronounced names outperformed financial stocks with less easily pronounced names in the short term across real stock markets (Alter & Oppenheimer, 2006).

Apart from this wide range of judgments, fluency can also influence memory decisions. Jacoby and Dallas (1981) explained that items that are reencountered are processed more fluently than items that are encountered for the first time, so we learn by experience that fluent processing is an indicator of prior occurrence. This often results in increased feelings of familiarity for fluently processed items (Jacoby & Whitehouse, 1989). In recognition studies, increased familiarity for more fluent test stimuli produces more “old” responses for those stimuli, even when they were not studied previously, inducing more hits and false alarms. Such effects were observed in studies that used a variety of methods to increase different types of fluency. For example, repetition priming (e.g., Kurilla & Westerman, 2008 (Experiments 1 and 2); Westerman, 2001, Westerman, 2008 (Experiments 1 and 3); Westerman, Lloyd, & Miller, 2002) and visual clarity (e.g., Whittlesea, Jacoby, & Girard, 1990) manipulations were used to enhance perceptual fluency. Other studies utilized conceptual priming (e.g., Rajaram and Geraci, 2000, Taylor and Henson, 2012) and predictive sentence stems (e.g., Kurilla & Westerman, 2008 (Experiment 3); Westerman, 2008 (Experiments 2 and 4); Whittlesea and Williams, 2001a, Whittlesea and Williams, 2001b) to enhance conceptual fluency. Regardless of the method used, these studies jointly showed that fluency manipulations affected recognition in a similar way, yielding more hits and false alarms for more fluent stimuli.

Despite large body of evidence that fluency can enhance familiarity and increase positive recognition judgments, there are studies in which fluency effects on recognition were not found or were reversed. For example, Unkelbach, 2006, Olds and Westerman, 2012, Experiments 1 and 2) demonstrated that participants that were trained to associate high fluency with novelty responded opposite to the typical fluency effect, such that more fluent items were more likely to be endorsed as “new”. In addition, the control group of participants in the study of Unkelbach (2006) did not show any effect of fluency on recognition. These results indicate that the interpretation of fluency in terms of familiarity is context dependent and susceptible to change given the appropriate training. However, reversed effects of fluency on recognition were also found without any specific training. In a series of five repetition priming experiments, Huber, Clark, Curran, and Winkielman (2008) showed that short duration primes produced a recognition preference for primed words, replicating the standard relationship between fluency and recognition, but primes with long durations produced the opposite effect, such that accuracy was worse for primed items. Apparently, the prolonged exposure to primes eliminated the positive priming by the perceptual habituation that discounted the familiarity response. Altogether, these results can be reconciled with the discrepancy-attribution theory (Whittlesea and Williams, 2001a, Whittlesea and Williams, 2001b), a more general account of fluency, which posits that fluency leads to familiarity (and greater recognition) only when it is discrepant from expectations and attributed to the test items. When fluency is expected and attributed to a perceptual experience, the feelings of familiarity might be eliminated.

The fact that fluency does not always support familiarity judgments implies that fluency- and familiarity-based recognition are governed by distinct neural pathways. Neuropsychological data indicate that severely amnesic patients with lesions in medial temporal lobes (MTL) exhibit fluency effects, but at-chance performance on recognition memory tasks (Conroy, Hopkins, & Squire, 2005), whereas patients with occipital lobe lesions demonstrate a reversed pattern, with impaired fluency but normal recognition performance that was presumably based on familiarity (Gabrieli et al., 1995, Keane et al., 1995). These data conform with the well-documented role of the MTL structures in declarative memory, with recollection typically attributed to the hippocampus and familiarity to the perirhinal cortex (Cowell, Barense, & Sadil, 2019). Activity reductions in perirhinal cortex have also been linked to repetition-induced fluency and conceptual fluency, as revealed by the priming effects (Voss et al., 2009, Wang et al., 2014). On the other hand, perceptual fluency effects can be associated with more posterior cortical regions, like the occipito-temporal regions in the ventral visual stream that supports visual perception (Cowell et al., 2010, Danckert et al., 2007). Thus, it seems that neural substrates of fluency and familiarity-based recognition partially overlap, particularly when fluency results from repetition or conceptual processing (the perirhinal cortex), but are partially distinct, particularly when pure perceptual fluency is at hand (the occipital regions).

Nevertheless, identifying brain structures engaged in processing fluency and familiarity-based recognition does not explain the exact mechanism of how fluency influences recognition and more direct evidence of this influence is needed. Such evidence may come from studies in which event-related potentials (ERPs) are recorded. ERPs are electrophysiological measures of brain activity that provide a continuous measure of processing with excellent temporal resolution. This virtue of ERPs makes them especially suited to answer important questions regarding fast cognitions, like fluency. ERP fluency effects were observed as early as between 150 and 250 ms, such that ERP amplitudes in this time window were more positive for more fluent (repeated) items (Woollams, Taylor, Karayanidis, & Henson, 2008). Similar findings were reported by Voss and Paller (2010), who observed fluency ERP for geometric shapes between 100 and 300 ms at parietal electrodes, albeit the polarity of this effect was reversed. Recognition memory studies found distinct ERP components, the FN400 and the LPC (late positive component), which are regarded as electrophysiological correlates of familiarity and recollection, respectively (Rugg et al., 1998a, Rugg and Curran, 2007). The FN400 occurs from around 300 ms to 500 ms after stimulus onset at mid-frontal electrodes, while the LPC occurs from around 500 ms to 800 ms, typically at left parietal electrodes for words (but is more widespread for pictures). It could be predicted that, if increased fluency enhances familiarity, this should affect the amplitudes of the electrophysiological correlate of familiarity, namely the FN400. However, there are studies that suggest that although the ERP fluency effects were reported in the FN400 time window, they were still recorded at different scalp locations (central and parietal), resembling the N400, a well-known ERP component reflecting semantic processing, but not familiarity (Nessler et al., 2005, Kurilla and Gonsalves, 2012, Lucas et al., 2012).

The results of the above-mentioned ERP studies indicate that it is difficult to identify clear FN400 modulations resulting from increased fluency. However, this might be because these studies utilized repetition priming to enhance fluency, and repetition priming alters both perceptual and conceptual fluency. Item repetition increases perceptual fluency due to the complete overlap of perceptual features, but it also increases conceptual fluency due to the repeated concept. These simultaneous changes in perceptual and conceptual fluency make it difficult to determine whether the two forms have different effects. In addition, repetition priming makes it difficult to separate the FN400 from the N400 (Bridger et al., 2012, Stróżak et al., 2016, Voss and Federmeier, 2011). However, these issues can be resolved by using experimental procedures that are better suited to disentangle perceptual and conceptual fluency and capitalizing on ERPs temporal sensitivity.

A suitable method to capture the pure effects of perceptual fluency on recognition is to manipulate the perceptual identifiability of items that are presented in the recognition test. This method seems to be most appropriate because changing the visual clarity of test items, as long as these items are still identifiable, affects perceptual fluency without altering conceptual fluency. The procedure of manipulating the visual clarity of some of the items in the recognition test, together with the recording of ERPs, was first used by Leynes and Zish (2012). In this study, two experiments were conducted in which participants encoded words and in the recognition test half of the old words and half of the new words were blurred. Also, in one experiment the clarity of test probes varied randomly from trial to trial, whereas in the other experiment there were separate clear and blurry blocks of words during recognition. Random variations in clarity resulted in greater sensitivity, more liberal responses and greater hit rate for clear than for blurry items. Also in this condition, the FN400 old/new effect was observed, indicating familiarity-based recognition. On the other hand, blocking clarity made the perceptual fluency arising from the clarity manipulation irrelevant to recognition decisions, as confirmed by the absence of the FN400 in this condition. However, when clarity was blocked, the early old/new and clear/blurry effects were observed at parietal electrode sites from 280 to 400 ms, such that old words elicited more negative amplitudes than new words, and clear words elicited more negative amplitudes than blurry words. Thus, it seems that more fluent stimuli demand fewer cognitive resources and induce smaller ERP amplitudes than less fluent stimuli, irrespective of whether fluency is arising from repetition or perceptual properties.

The other method that can help to disentangle different types of fluency is to manipulate them independently within one procedure. Wang, Li, Gao, & Guo (2018) conducted an ERP experiment in which participants encoded words and then completed a recognition test in which old and new words were conceptually primed or not primed. Additionally, half of the test words was perceptually clear and the other half was blurred. Behavioral results indicated that the “know” hit rate (reflecting familiarity) was greater for clear than for blurry words, and the ERP results indicated that the difference between clear and blurry words was observed early, in the 100–200 ms time interval, such that ERPs for clear words were more positive than ERPs for blurry words. Also, basic memory effects were observed, such that the N400¹ and LPC were more positive for old words than for new words. These results demonstrated that perceptual and conceptual fluency differently affect behavioral responses and that these two types of fluency are associated with different ERP components.

The studies of Leynes & Zish, 2012, Wang et al., 2018 provided important findings regarding the relationships between perceptual fluency and repetition and between perceptual and conceptual fluency, respectively. However, in both of these studies, the perceptual fluency effects and their electrophysiological correlates were defined as the difference between clear and blurry retrieval cues, collapsed across old and new items in the recognition test. Also, basic memory effects were calculated as the difference between old and new test words, collapsed across clear and blurry items². This approach is appropriate given the logic behind these studies; however, if perceptual fluency effects on recognition memory are to be tested, it seems reasonable to analyze basic old/new memory effects and their corresponding ERP correlates separately for clear and blurry items³. Because perceptual fluency affects recognition judgments, such that the hit rates are greater for clear than for blurry items, these effects should be reflected in ERPs corresponding to the underlying memory processes. Following the logic of Leynes and Zish (2012) that processing items that are perceptually more fluent give rise to familiarity, we can assume that the FN400 old/new effects for clear items should be larger than the FN400 old/new effects for blurry items. Only the direct comparison of such ERP contrasts might provide evidence that the perceptual fluency effects on recognition are reflected at the neural level.

To this end, we conducted two ERP experiments in which participants encoded words and then performed a recognition test in which half of the old words and half of the new words were presented in perceptually clear or blurry format, replicating the procedure introduced by Leynes and Zish (2012) in their experiment with random variations in clarity. Our main aim was to calculate the FN400 old/new effects separately for clear and blurry test words in order to assess whether perceptual fluency affects recognition by altering the electrophysiological index of familiarity. We hypothesized that if enhanced perceptual fluency is interpreted as familiarity, recognition accuracy should be greater for clear than for blurry test words and this should be accompanied by larger FN400 old/new effect for clear items. We also aimed at identifying early ERP differences between processing clear and blurry test words. However, given the ambiguity of the previous findings, such that these early clear-blurry differences were found in different time intervals and were of different polarity (280–400 ms, blurry > clear in Leynes and Zish (2012); 100–200 ms, clear > blurry in Wang et al. (2018)), we refrained from formulating the exact hypotheses in that domain.

Moreover, we conducted two experiments that differed in the difficulty of the memory task, as this factor might be crucial when it comes to the contributions of perceptual fluency to recognition, but it has not been tested so far. In previous studies, the number of words that was presented at the encoding stage was set arbitrarily, which led to different memory requirements imposed on participants (e.g., Leynes and Zish (2012) used 150 words, whereas Wang et al. (2018) used seven study-test blocks with 60 words in each block). The influence of perceptual fluency on recognition relies on changing the expectations regarding fluency (Mecklinger & Bader, 2020), and the overall difficulty of the memory task might determine whether fluency would have any effect on these expectations. In addition, longer lists create more variability in the familiarity response and incline participants to adopt a more liberal familiarity criterion, which produces worse recognition performance (Cary and Reder, 2003, Shiffrin et al., 1995). Given that, it seems plausible that longer lists can also make participants to be more likely to change their expectations regarding fluency and perceive it as an indication of something familiar, whereas shorter lists might have no effect on these expectations. This should be reflected in different patterns of behavioral and ERP results for more difficult and less difficult memory tasks. To verify these assumptions, we designed one experiment with four long study-test blocks (more difficult memory requirements), and the other experiment with eight short study-test blocks (less difficult memory requirements). We hypothesized that the enhanced perceptual fluency should result in greater recognition accuracy and larger FN400 old/new effects for clear test words, and that this pattern of results should rather be observed in the first experiment with more difficult memory requirements than in the second experiment with less difficult memory requirements. Altogether, we presume that such experimental approach might be useful in delineating the perceptual fluency effects on recognition.