Recognition memory for emotionally negative and neutral words: an ERP study

Abstract

Scalp recorded event-related potentials were used to investigate the neural activity elicited by emotionally negative and emotionally neutral words during the performance of a recognition memory task. Behaviourally, the principal difference between the two word classes was that the false alarm rate for negative items was approximately double that for the neutral words. Correct recognition of neutral words was associated with three topographically distinct ERP memory ‘old/new’ effects: an early, bilateral, frontal effect which is hypothesised to reflect familiarity-driven recognition memory; a subsequent left parietally distributed effect thought to reflect recollection of the prior study episode; and a late onsetting, right-frontally distributed effect held to be a reflection of post-retrieval monitoring. The old/new effects elicited by negative words were qualitatively indistinguishable from those elicited by neutral items and, in the case of the early frontal effect, of equivalent magnitude also. However, the left parietal effect for negative words was smaller in magnitude and shorter in duration than that elicited by neutral words, whereas the right frontal effect was not evident in the ERPs to negative items. These differences between neutral and negative words in the magnitude of the left parietal and right frontal effects were largely attributable to the increased positivity of the ERPs elicited by new negative items relative to the new neutral items. Together, the behavioural and ERP findings add weight to the view that emotionally valenced words influence recognition memory primarily by virtue of their high levels of ‘semantic cohesion’, which leads to a tendency for ‘false recollection’ of unstudied items.

Introduction

Numerous studies have demonstrated that performance on episodic memory tasks is influenced by the emotional nature of the test items. Notably, recall of emotionally negative material (and, in some studies, positive material also) is enhanced relative to the recall of neutral items [5], [23], [26]. It has been suggested that differences in memory performance for emotional and non-emotional material arise through multiple mechanisms, two of the most important of which are ‘arousal’, and ‘semantic cohesiveness’ [22]. Items that are arousing (i.e. items that elicit significant autonomic responses as indexed, for example, by skin conductance) gain a mnemonic advantage over non-arousing items through the enhancement of their encoding and consolidation in memory. This advantage seems likely to be mediated at the neural level by the modulatory influence of the amygdala on hippocampal and cortical components of the network supporting episodic memory [10], [22]. By contrast, non-arousing emotional items (e.g. words with negative emotional valence) influence memory largely because, unlike emotionally neutral items, they tend to belong to categories that are semantically ‘cohesive’, that is, categories in which the constituent items share strong inter-item associations.

The foregoing distinction provides a useful framework in which to view findings of studies comparing memory for emotionally valenced and emotionally neutral words. With the exception of certain classes of item such as ‘taboo’ words, emotional and non-emotional words differ little in their arousing properties [19] raising the possibility that differences in memory for these two word classes are due mainly to differences in semantic cohesiveness. Consistent with this possibility, Phelps and LaBar (cited in LaBar and Phelps, [19]) found that when inter-item associations were controlled, the recall advantage normally found for emotionally valenced words over neutral items was eliminated. Also consistent are findings showing that the above-mentioned recall advantage does not extend to tests of recognition memory (e.g. see [5], [20]). In one of these studies [5], discrimination (d′) between studied (old) and unstudied (new) words was lower for emotional than for neutral items (hit and false alarm rates were not reported). In the second study [20], emotional words were associated with a higher hit rate but, because this effect was offset by an equivalent elevation of the false alarm rate, recognition accuracy did not differ for the two word types.

The disparate pattern of findings for recall and recognition fit well with the idea that emotional valence exerts much of its effect on memory for words through the mechanism of semantic cohesion. The fact that emotional words generally belong to a relatively ‘closed’ semantic category means that recall for these items benefits for the same reasons that underlie the recall advantage for words from any semantically categorised set relative to uncategorised words [2]. In the case of recognition, the findings of Leiphart et al. [20] suggest that emotionally valenced words may act like the ‘related lures’ that elicit high false alarm rates in studies of ‘false recollection’. In these studies (e.g. [24], [25], [35] see [15] for review) subjects typically study lists of semantically related items and subsequently attempt to discriminate between these items and two kinds of new word — semantic associates of studied words (related lures), and words semantically unrelated to items that had been shown at study (unrelated lures). Relative to unrelated lures, related lures generate high false alarm rates, with subjects reporting many of these responses to be based on a recollection of the lure as a member of the study list (‘Remember’ responses, e.g. [25]). Neuroimaging and electrophysiological evidence suggest that the illusory recollection of related lures relies on much of the same neural circuitry that underlies veridical recollection [7], [14], [34], [35]. Since sets of emotionally valenced words tend to be semantically and associatively related, it is easy to see how these items might also engage processes supporting false recollection.

According to the foregoing analysis, therefore, emotionally valenced words do not exert their effect on memory by virtue of their valence per se unless they are arousing. Rather, they do so because emotionally valenced words have, on average, stronger inter-item associations than do sets of unselected neutral words. Thus, memory for emotional words is mediated not by systems or processes to which these items have privileged access, but by the same cognitive operations that support memory for non-emotional material.

In the present experiment, we assess the foregoing proposal by investigating the neural correlates of recognition memory for emotionally valenced and neutral words using event-related brain potentials (ERPs). In general terms, the proposal that recognition memory for the two classes of word engages equivalent cognitive operations can be assessed by determining whether the memory-related ERP effects they elicit differ with respect to scalp distribution. If the effects for emotional and neutral words do differ in this respect, it would indicate that memory for the two kinds of material is neurally (and, therefore, most likely functionally) dissociable [29]. Such a finding would be inconsistent with the proposal that memory for emotional and neutral words relies on functionally equivalent memory mechanisms.

The results of previous research investigating the ERP correlates of recognition memory allow predictions arising from the ‘semantic cohesiveness’ hypothesis to be addressed more specifically. It has long been known that, relative to new words, ERPs elicited by correctly classified old words in tests of recognition memory are more positive-going — the so-called ERP ‘old/new’ effect (see [33] for review). Recent work (for review see [28]; see also [31]) has led to the identification of three old/new effects which appear to index functionally distinct aspects of recognition. For present purposes, the most important effect is one which onsets around 400–500 ms post-stimulus, and is maximal over the left parietal scalp (the ‘left parietal’ old/new effect). The effect appears to be a neural correlate of the episodic retrieval (recollection) of study items [39] and, in studies of false recollection [7], [14] is elicited by both genuinely old words and related lures. Another old/new effect involves an earlier (ca. 300–500 ms), bilateral shift with a frontal distribution. Rugg et al. [31] proposed that this effect was a neural correlate of familiarity-based recognition. This is a form of recognition memory held by some authors to be independent of recollection, and to underlie recognition judgements associated with ‘Know’ rather than ‘Remember’ responses [9], [38]. The third old/new effect onsets quite late (ca. 500–700 ms), persists for a second or more, and is distributed over the right frontal scalp [6], [39]. The ‘right frontal’ old/new effect is held to reflect processes that operate on the products of memory retrieval [28].

If, as suggested above, unstudied emotionally valenced words in tests of recognition memory act like associative lures in studies of false recollection, these items should be difficult to discriminate from studied emotional items because of their tendency to elicit ‘recollection’ of their study presentation. As a consequence, unstudied emotional items should be associated with an elevated false alarm rate relative to unstudied neutral words. Furthermore, even on trials on which unstudied emotional items are correctly classified as new, the tendency of these items to elicit illusory recollection might be expected to both impede the decision to respond ‘new’ and, critically, to be manifest in a pattern of neural activity — the left parietal ERP effect — signifying the engagement of neural systems supporting episodic retrieval. Thus, relative to the left parietal old/new effect elicited by neutral words, the effect elicited by emotionally valenced words should be smaller in magnitude. Furthermore, the difference in the magnitude of the left parietal effects elicited by the two word types should be carried mainly by the ERPs elicited by new items, those elicited by emotional words exhibiting the greater positivity.

We tested these predictions by recording ERPs while subjects discriminated between studied and unstudied emotionally negative and neutral words. For the reasons already noted, we expected negative words to give rise to an excess of false alarms, a smaller left parietal old/new effect, but no evidence for the recruitment of memory mechanisms additional to those engaged by neutral items. Further, to the extent that emotional valence influences recognition memory exclusively through its effects on recollection, the ERP old/new effect held by Rugg et al. [31] to index familiarity-driven recognition should be unaffected by this variable.