Emotional Content and Source Memory for Language: Impairment in an Incidental Encoding Task

Ferré, Pilar; Comesaña, Montserrat; Guasch, Marc

doi:10.3389/fpsyg.2019.00065

ORIGINAL RESEARCH article

Front. Psychol., 30 January 2019

Sec. Emotion Science

Volume 10 - 2019 | https://doi.org/10.3389/fpsyg.2019.00065

Emotional Content and Source Memory for Language: Impairment in an Incidental Encoding Task

$\r\nPilar Ferr*$ Pilar Ferré^1*

Montserrat Comesaña²

Marc Guasch¹

¹Department of Psychology and CRAMC, Universitat Rovira i Virgili, Tarragona, Spain
²Human Cognition Lab, Research Center on Psychology, School of Psychology, University of Minho, Braga, Portugal

Research into the effects of emotion on source memory (i.e., memory for certain contextual details of a stimulus, such as its location, color, or temporal context) has yielded inconsistent findings. Mather and her co-workers tried to account for such inconsistencies by pointing out the relevance of the characteristics of the feature examined. Specifically, they distinguished between intrinsic and extrinsic features (Mather, 2007) and between goal-relevant and goal-irrelevant information (Mather and Sutherland, 2011). In the current study, we investigated source memory for language, which is an intrinsic feature or words. Catalan-Spanish bilinguals were tested in three experiments involving a recognition task in which they were asked about the language of presentation (Catalan or Spanish) of emotional and neutral words. In Experiments 1 and 2, source memory for negative and neutral words was assessed. In Experiment 1 participants performed an intentional encoding task in which language was a goal-relevant feature. In Experiment 2, they did an incidental encoding task in which language was also goal-relevant. Experiment 3 replicated Experiment 2 but negative words were replaced by positive words. The results showed an impairment in source memory for the language of presentation of emotional words when the encoding task was incidental, but not when it was intentional. Such impairment was observed with both negative words and positive words. The results are discussed in relation to the proposals of Mather and co-workers and point to the relevance of modulating factors, such as the intentional/incidental nature of encoding.

Introduction

The last decades have witnessed a great interest in the study of the effects of emotion on episodic memory. The majority of the studies in the field have been concerned with item memory (i.e., memory for the content of the information), finding that emotional stimuli are better remembered than neutral stimuli (see LaBar and Cabeza, 2006; Mather and Sutherland, 2011; Phelps, 2012; Dolcos et al., 2017; Kensinger and Kark, 2018; for overviews). However, far less research has been devoted to the effects of emotion on source memory (see Chiu et al., 2013, for a review). Source memory refers to memory for certain contextual details of an event (Johnson et al., 1993). It is commonly assessed using recognition memory tasks in which participants are not only asked to recognize previously presented information as old, but are also required to retrieve encoding details about such information. Studies in this field have compared emotional stimuli (defined in terms of their valence and arousal levels) and neutral stimuli (commonly images or words) and have tested memory for different types of information, such as temporal or spatial information, cognitive operations performed during encoding, or visual-perceptual stimulus features, among others (Chiu et al., 2013). In all these studies, the amount of source information is much lower than the number of trials, as responses involve a limited number of alternative choices (e.g., participants have to decide between two colors, or two different spatial locations). A related line of research has focused on memory for relational/associative information, where there is a one-to-one relationship between trials and the relational information (e.g., participants have to remember the background scenes presented together with emotional/neutral items or the two items which were presented in pairs during encoding, see Chiu et al., 2013, for a detailed description). Although the terms “source memory” and “relational/associative memory” do not refer exactly to the same, the main findings of both lines of research have to be taken into consideration in order to have a complete picture of the field. Of note, the present study fails within the first group of studies and sought to test the effects of emotion on source memory for language.

Research into the effects of emotion on source memory and relational memory has yielded mixed findings: a number of studies have shown that emotional content enhances memory for location (D’Argembeau and Van der Linden, 2004; Mather and Nesmith, 2008; Nashiro and Mather, 2011; Schmidt et al., 2011; Rimmele et al., 2012), color (Doerksen and Shimamura, 2001; Kensinger and Corkin, 2003), temporal information (D’Argembeau and Van der Linden, 2005; Schmidt et al., 2011; Rimmele et al., 2012), the task performed with the items at encoding (Kensinger and Schacter, 2006), or the identity of objects associated to background scenes (Smith et al., 2005). This enhanced memory was associated, in the last study, with differential neural activity during the retrieval of information associated with emotional background scenes. A number of studies, however, have found a different pattern: an impairment of emotional stimuli in the memory for location (Mather et al., 2006; Mitchell et al., 2006; Mather and Knight, 2008; Maddock and Frein, 2009), color (Mackenzie et al., 2015) and the type of task performed during encoding (Cook et al., 2007; Mao et al., 2015). Such impairment was also observed at a neural level, where event-related potential (ERP) correlates of source memory were observed for neutral items, but not for emotional ones (Mao et al., 2015). Similarly, a series of studies have reported an impairment in memory for the peripheral information of scenes containing a central emotional item (e.g., Kensinger et al., 2007; Bisby and Burgess, 2014) as well as for items associated with emotional stimuli in paired designs (e.g., Mather and Knight, 2008; Nashiro and Mather, 2011; Pierce and Kensinger, 2011; Bisby and Burgess, 2014). Apart from these contrasting findings, other studies have failed to find any effect of emotion on source memory and relational memory at all (e.g., Sharot and Yonelinas, 2008; Mather et al., 2009; Meyer et al., 2015).

The reason of the inconsistent pattern of findings above reviewed is far from being established. However, they suggest that several factors can be modulating emotion effects. Mather and her colleagues have highlighted the role of certain aspects of the information. In her first proposal, the object-based framework, Mather (2007) argued that the crucial distinction was between intrinsic information (i.e., features that are an integral part of an item, such as its color) and extrinsic information (i.e., features that are not an integral part of an item, such as the context of an emotional scene). According to Mather, emotional (arousing) information would attract attention, enhancing item-context binding only when the context belongs to the same object as the emotional information (i.e., intrinsic information). This proposal has been supported both by studies reporting a memory enhancement with intrinsic information, such as color or location (Doerksen and Shimamura, 2001; D’Argembeau and Van der Linden, 2005; Mackay and Ahmetzanov, 2005; Mather and Nesmith, 2008; Nashiro and Mather, 2011) and by those reporting a memory impairment with peripheral details of scenes (e.g., Kensinger et al., 2007) or the binding between pairs of items (e.g., Nashiro and Mather, 2011; Pierce and Kensinger, 2011). However, the inconsistencies in the literature cannot be entirely resolved by Mather’s account. For instance, both null effects (Meyer et al., 2015) and detrimental effects (Mackenzie et al., 2015) of emotion on intrinsic features of the stimuli have recently been reported. Even more, the same type of information (e.g., location, an intrinsic feature) has shown an enhancement in some studies (D’Argembeau and Van der Linden, 2004; Mackay and Ahmetzanov, 2005) and an impairment in others (Mather et al., 2006; Maddock and Frein, 2009).

In light of the above, Mather and co-workers updated and extended their model by emphasizing the role of a different factor, namely stimulus priority. Given limited human attention capacity, it is necessary to prioritize one piece of information above others during processing. In the arousal-biased competition (ABC) theory, Mather and Sutherland (2011) argued that emotional arousal would have a role in such prioritization, by enhancing processing (and memory) of the highest priority information, while impairing processing (and memory) of the lowest priority information. Therefore, the ABC theory predicts an emotion enhancement in memory for high priority information, regardless of its intrinsic/extrinsic nature. In this way, it can account for previous evidence of emotion improvement in memory for what can be considered extrinsic sources of information. For instance, Guillet and Arndt (2009) reported enhanced memory for neutral words associated with emotional words. Of note, in this study participants were asked to learn word-word associations, thus making the associated neutral words become priority information. Similarly, instructing participants to intentionally connect the location of objects with background scenes led to an enhancement in memory for the identity of objects (Ventura-Bort et al., 2016) and for their location (Luck et al., 2014) when the background scenes were emotional. Interestingly, this effect was associated with an enhancement of one ERP indicator of recognition memory (the old/new effect, Ventura-Bort et al., 2016). It was also associated with increased activity in the amygdala during encoding (Luck et al., 2014), suggesting that the interaction between this structure and medial temporal lobe regions involved in memory binding might be responsible for the superior memory for goal-relevant information (Dolcos et al., 2017).

The majority of the studies reviewed in the above paragraphs (mainly those conducted before 2012) were not carried out to test the ABC theory. However, we can examine the extent to which the features examined were goal-relevant (i.e., relevant for the task to be performed) by looking at the encoding instructions. In doing so, the ABC does not seem to be entirely supported. In particular, although the enhancing emotion effects in memory for information that was relevant for a specific task have been reported (e.g., Doerksen and Shimamura, 2001), both null effects (Meyer et al., 2015) and detrimental effects (Maddock and Frein, 2009; Otani et al., 2012; Mackenzie et al., 2015) have also been obtained. On the contrary, enhancing effects have been reported with information which was not at all relevant during encoding (e.g., Nashiro and Mather, 2011; Schmidt et al., 2011; Rimmele et al., 2012).

Overall, it seems that the proposals of Mather and co-workers cannot account entirely for the discrepant findings in the field. A possible reason may be that other factors, such as the valence of the stimuli or the nature of the encoding task, can modulate the pattern of effects. Regarding valence, although some studies have reported an effect of emotion on source and relational memory for both negative and positive information (e.g., Mather and Nesmith, 2008; Schmidt et al., 2011; Mao et al., 2015), a number of studies have failed to find any effect with positive stimuli (e.g., D’Argembeau and Van der Linden, 2005; Cook et al., 2007; Maddock and Frein, 2009; Otani et al., 2012). These results suggest that the attraction of attention that can ultimately lead to an effect on source and relational memory seems to affect mainly negative information. With respect to the nature of the encoding task, some studies have relied on intentional encoding tasks, in which participants are encouraged to memorize both the item and the source (Otani et al., 2012; Mackenzie et al., 2015; Mao et al., 2015), or only the item (Cook et al., 2007). Other studies have relied on incidental tasks in which participants are not told about any subsequent memory test (Maddock and Frein, 2009; Nashiro and Mather, 2011; Schmidt et al., 2011; Bisby and Burgess, 2014). Arguably, the intention (vs. the lack of intention) to memorize both the content and the context of given information may affect the extent to which emotion affects memory of the context. In particular, if the influence of emotion on cognition involves the automatic activation of attention (D’Argembeau and Van der Linden, 2004), the effects of emotion on source memory might be more easily observed when participants do not intentionally memorize the source.

In the present study, we aimed to test the effects of emotion on source memory by using as source information a feature never before studied in the field: the language in which words were presented. Psycholinguistic research has demonstrated that there are distinct levels of word representation. Some of these have to do with the form of the word itself (i.e., orthography, phonology) and others with its meaning (e.g., Carreiras et al., 2014). When people know more than one language (i.e., they are bilinguals), each word in their lexicon must be identified in relation to the language to which it belongs. Therefore, language is part of word representation (Dijkstra and van Heuven, 2002). Since the meaning of words can include emotional connotations and their forms contain information about language, language constitutes a good means of testing the effect of emotion on source memory. Importantly, what makes language a feature worth to be investigated in relation to source memory is that it is highly integrated with item information, being clearly an intrinsic property of the words (see Bar-Elli, 2006).

Catalan-Spanish bilinguals were tested in this study in three memory recognition experiments in which Catalan and Spanish words were presented during the acquisition phase. Both valence and the type of encoding task (intentional vs. incidental) were manipulated. In Experiment 1, source memory for negative and neutral words intentionally encoded was assessed, language being a goal-relevant feature (i.e., participants were told to encode both the words and their language of presentation during the acquisition phase). In Experiment 2, the encoding task was incidental, although language was also a relevant feature (i.e., participants were asked to name the words in the correct language, without being informed that their memory of the words or the language would be tested afterward). Experiment 3 replicated Experiment 2, but negative words were replaced with positive ones in an attempt to evaluate the impact of valence on source memory. This manipulation was only done in the incidental task because the effects of emotion on memory were restricted to this type of codification, as will be described below. Considering that language was an intrinsic and goal-relevant feature in the three experiments, we might expect an improvement in memory for the language of presentation of emotional words in comparison to neutral words in all of them. However, modulations in the effect of emotion on source memory by the type of encoding task and the valence of the stimuli might be also expected if we bear in mind the literature here (Cook et al., 2007; Mather, 2007; Otani et al., 2012). In particular, we would expect a stronger effect of emotion on source memory in the incidental encoding task and with negative words.

Experiment 1

This experiment was carried out in accordance with the code of ethics of the World Medical Association (Declaration of Helsinki) and was approved by the Ethics Committee for Human Research (SECH 017/2015) of the Research Center on Psychology (CIPsi) at the University of Minho. Furthermore, participants gave written informed consent.

Methods

Participants

Thirty undergraduate students (23 females) from the University Rovira i Virgili (Tarragona, Spain) took part in the experiment either voluntarily or in exchange of academic credits. Their mean age was 22.27 years (SD = 4.36). All the participants had normal vision or corrected-to-normal vision and were bilinguals of Catalan and Spanish. Of note, the degree of bilingualism in Catalonia is very high. People living there acquire Catalan and Spanish in early childhood and are immersed throughout their lives in both languages. Hence, all the participants here were highly proficient early bilinguals of Catalan and Spanish, living in Catalonia at the time of the experiment, and using both Catalan and Spanish on a regular basis.

Participants were asked to fulfill a language history questionnaire to assess their degree of bilingualism. Data from the questionnaire confirmed that they were highly proficient in both languages. The average proficiency was 6.77 (SD = 0.39, Minimum = 5.75, Maximum = 7.00, Range = 1.25) for Catalan and 6.85 (SD = 0.30, Minimum = 5.75, Maximum = 7.00, Range = 1.25) for Spanish on a seven-point scale (being 1 a ‘very poor level,’ and 7 a ‘very good level’). The difference in self-rated proficiency between both languages was not significant; t(29) = 1.14, p = 0.265. Additionally, participants rated their frequency of language use on a seven-point scale (were 1 was ‘only Catalan’ and 7 was ‘only Spanish’), their mean being 4.02 (SD = 0.88, Minimum = 2.25, Maximum = 6.00, Range = 3.75). Finally, participants also rated their language preference using the same scale (i.e., 1 = ‘only Catalan’; 7 = ‘only Spanish’), the average rating being 4.18 (SD = 0.91, Minimum = 1.75, Maximum = 6.50, Range = 4.75). We carried out one-sampled t-tests to examine whether frequency and preference ratings were significantly different from the central point of the scale (i.e., ‘to the same extent in Catalan and Spanish’). These analyses failed to reveal a significant difference (both ps > 0.25), indicating that participants preferred and used both languages to the same extent.

Design and Materials

The experiment employed a 2x2 design. The factors involved were language (words presented in Catalan vs. Spanish) and emotion (negative vs. neutral words). The critical stimuli consisted of a set of 60 Spanish words and their Catalan translations (see Appendix). These were the words to be learned during the acquisition phase (i.e., old words). They were obtained from several Spanish normative databases, using the emoFinder online search engine (Fraga et al., 2018). Values were taken mainly from the Spanish adaptation of ANEW (Redondo et al., 2007), but some values were not available here and were obtained from Ferré et al. (2012), Guasch et al. (2016), and Hinojosa et al. (2016). Half the stimuli where negative words (e.g., mentira in Spanish and mentida in Catalan, meaning ‘lie’ in English) and the other half were neutral words (e.g., sendero in Spanish and sender in Catalan, meaning ‘path’ in English). A word was considered negative if it had a value of valence below 4 on a 9-point scale (1 = ‘very unpleasant,’ 9 = ‘very pleasant’), as well as a value of arousal above 5 on a 9-point scale (1 = ‘completely calm,’ 9 = ‘completely excited’). On the other hand, a word was considered neutral if it had a value of valence between 4 and 6, as well as a value of arousal below 5. Of note, we relied on affective ratings for Spanish words, since there are no normative affective ratings available for Catalan words. We assumed that the affective charge is the same for words in the two languages. The reasons for this are twofold: on the one hand, all the experimental words were cognates in the two languages (i.e., translation equivalents that are very similar in form, e.g., cicatriz-cicatriu, scar). On the other hand, the results of previous studies carried out with highly proficient early bilinguals of Catalan and Spanish (i.e., like those tested in this study) involving emotional and neutral words have repeatedly failed to find any interaction between the emotion effects and language (Ferré et al., 2010, 2013, 2018).

Negative and neutral words were matched on several psycholinguistic variables: length of the Catalan and the Spanish words, lexical frequency of the Catalan and the Spanish words, concreteness, imageability, subjective familiarity, and age of acquisition (AoA). The frequency of the Catalan words (i.e., word frequency per million) was obtained from the IEC (Rafel, 1998) through the online tool NIM (Guasch et al., 2013). The frequency of the Spanish words (also word frequency per million) was obtained from the subtitle corpus of EsPal (Duchon et al., 2013). EsPal was also the source of the subjective ratings of concreteness, imageability, and familiarity (with the exception of some unavailable data, which were obtained from Guasch et al., 2016). Age of acquisition was taken from Alonso et al. (2015). Of note, as above explained, all the critical stimuli were cognate words in Catalan and Spanish. The reason for this was to make the source memory task more difficult for participants. We reasoned that it would be too easy to remember the language in which words were presented if the form of the translation equivalents was different between the languages, the so-called non-cognate words (e.g., estiu-verano, ‘summer’). Notably, we excluded identical cognate translations in Catalan and Spanish (e.g., amor-amor, ‘love’) because this kind of stimuli would not have allowed participants to discriminate between the two languages. Hence, we matched the degree of orthographic similarity between translation equivalents across experimental conditions by computing the normalized Levenshtein distance (NLD) as described by Schepens et al. (2012). To that end, we used the online tool NIM (Guasch et al., 2013). Independent samples t-tests were used to check that the experimental conditions differed only in the manipulated variables. These tests showed significant differences between negative words and neutral words both in valence, t(58) = 19.80, p < 0.001, and arousal, t(58) = 14.34, p < 0.001. There were not significant differences in any of the controlled variables (all ps > 0.257). Table 1 shows the affective, semantic and lexical characteristics of the stimuli.

TABLE 1

Table 1. Affective, semantic and lexical characteristics of the stimuli (old and new words) used in Experiments 1–3 (standard deviations in parentheses).

We created two experimental lists. In each list, half the words appeared in Spanish and the other half appeared in Catalan. The language of presentation was counterbalanced across lists. Hence, any word that appeared in Spanish in one list, appeared in Catalan in the other list, and vice versa.

Finally, in addition to the 60 words to be learned, we selected another set of 60 words to be used in the test phase of the recognition task as new words. These items were selected according to the same criteria as the critical items, and followed the same distribution in experimental conditions. The statistical analyses showed that there were clear differences between negative and neutral new words in both valence, t(58) = 19.51, p < 0.001, and arousal, t(58) = 13.64, p < 0.001. There were no differences in any of the above mentioned controlled variables (all ps > 0.366). Furthermore, t-tests carried out between old and new negative words, and between old and new neutral words, confirmed that the old and new sets of words were correctly matched (all ps > 0.09).

Procedure

Participants performed a recognition memory task consisting of three phases. The first of these was the acquisition phase, followed by a distractor phase, and the test phase.

During the acquisition phase, participants had to memorize a list of words presented on the screen. Those words could appear either in Catalan or in Spanish. Participants knew that their memory of the words would be tested afterward. They were encouraged not only to memorize the words, but also the language in which they appeared. Participants were presented with a set of 60 words belonging to the critical conditions (i.e., 15 Spanish negative words, 15 Spanish neutral words, 15 Catalan negative words, and 15 Catalan neutral words). We created two experimental files. The language of presentation was counterbalanced across files. Hence, any word that appeared in Spanish in one file, appeared in Catalan in the other file, and vice versa. Apart from the 60 critical items, each file included 8 filler items: 4 at the beginning and 4 at the end. These words were included to control for serial position effects (primacy and recency effects, respectively) and had the same characteristics as the experimental words (i.e., there were 2 Spanish negative words, 2 Spanish neutral words, 2 Catalan negative words and 2 Catalan neutral words).

Words appeared continuously, remaining on the screen for 2 s with a gap of a further 2 s between each one. They were presented in the middle of the screen, lowercase, in black color against a white background. The font was Arial 40, and the screen resolution was set to 1680 × 1050. Words appeared completely at random, with a different order for each participant. The only exception was with the eight filler items, which appeared in the same order for all participants. Words were presented using SR Research Ltd’s Experiment Builder software.

Immediately after the acquisition phase, participants performed a distractor phase. They watched an animated movie in silence which lasted 5 min and 27 s. Immediately after the end of the movie, participants did the test phase.

In the test phase there were 120 trials, 60 involving old words (i.e., the critical words that had been presented during the acquisition phase) and 60 involving new words. Like the old words, half of the new words were negative and half were neutral words. Filler items belonging to the primacy and recency portions of the encoding list were not presented during the test phase. There was a single file, which was presented to all the participants. In each trial, a screen with three options was displayed. One option consisted of a Spanish word, and the other option was its Catalan translation. The Spanish word always appeared on the left side of the screen, and the Catalan word always appeared on the right side of the screen (see Figure 1). The third option was the word NUEVA (‘new’ in Spanish). Participants had to select from the three options with the mouse, to indicate whether the word had appeared in either Catalan or Spanish during the acquisition phase (i.e., it was an ‘old’ word) or whether it had not appeared (in this case, they had to select the option ‘new’). Words were presented in Arial 24, uppercase, to prevent any strategy based on the recall of the visual form of the words (note that words had been presented in lowercase during the acquisition phase). There was no time limit to respond. After participants’ response, there was an inter-trial interval of 500 ms. Stimuli were presented completely randomly with a different random order created for each participant. Words were presented and data were recorded using SR Research Ltd’s Experiment Builder software.

FIGURE 1

Figure 1. Screenshot of a trial in the test phase. Both words in the upper two boxes mean ‘tear,’ the Spanish word on the left and the Catalan word on the right. The box below contains the Spanish word for ‘new.’

After the main experimental task, participants were asked to fulfill an online language history questionnaire. The overall experimental session lasted approximately 30 min.

Results and Discussion

Across the three experiments, we first report item memory scores: hit rate, false alarm rate, A′ and $B_{D}^{″}$ . Hit rate is the proportion of studied items labeled old without regard to source accuracy (i.e., regardless of language). False alarm rate is the proportion of new items wrongly labeled old. A′ is an index of discriminability which varies from 0 to 1 with 0.5 indicating chance performance. $B_{D}^{″}$ is the corresponding measure of bias. Values greater than zero indicate conservative bias, and values less than zero indicate liberal bias (see Macmillan and Creelman, 2005). Subsequently, we report source memory performance. To obtain this, we calculated the unbiased hit rate (Wagner, 1993; Ventura-Bort et al., 2017). The Hit unbiased index (Hu) allows adjusting for response biases, because it not only takes into consideration the hit rates, but also the number of times a response has been given. It is defined as the conjoint probability of the correct identification of a stimulus and the correct use of a response (Wagner, 1993). For instance, the source memory score (Hu) for Spanish negative words was calculated as follows:

\frac{S p a N e g_L a n g_c o r r e c t}{S p a N e g_L a n g_c o r r e c t + S p a N e g_L a n g_i n c o r r e c t} * \frac{S p a N e g_L a n g_c o r r e c t}{S p a N e g_R e s p o n s e s}

SpaNeg_Lang_correct: Number of Spanish negative words correctly labeled as old whose language of presentation was correctly identified.

SpaNeg_Lang_incorrect: Number of Spanish negative words correctly labeled as old whose language of presentation was not correctly identified (i.e., participants identified the Catalan word as the presented one).

SpaNeg_Responses: Number of times that the identified language for a correctly labeled old word was Spanish (this includes both “SpaNeg_Lang_correct” responses and the number of times that Spanish was incorrectly identified as the language of presentation of a Catalan negative word).

In all cases, we carried out a 2 (Language) × 2 (Emotion) repeated measures ANOVA. Pairwise Bonferroni corrected comparisons were applied when the ANOVA yielded significant results. Item recognition scores and source memory scores are summarized in Table 2. Source memory scores collapsed by language in the three experiments are displayed in Figure 2.

TABLE 2

Table 2. Mean, standard deviation (in parentheses), minimum and maximum values (in square brackets) of the dependent variables across experimental conditions in Experiment 1.

FIGURE 2

Figure 2. Source memory scores (Hu) collapsed by language for emotional and neutral words in the three experiments.

The ANOVA revealed a main effect of emotion on hit rate, F(1,29) = 7.84, p = 0.009, $η_{p}^{2}$ = 0.21, MSE = 0.02, as well as on false alarm rate, F(1,29) = 11.55, p = 0.002, $η_{p}^{2}$ = 0.29, MSE = 0.01, where the proportion of hits and false alarms was higher for negative words (M = 0.77, SD = 0.17, and M = 0.39, SD = 0.20, for hits and false alarm rates, respectively) than for neutral words (M = 0.70, SD = 0.19 and M = 0.31, SD = 0.20). There was no effect of emotion in the A′ parameter, F(1,29) = 0.13, p = 0.725, $η_{p}^{2}$ = 0.004, MSE = 0.01. In contrast, the analyses showed an emotion effect in $B_{D}^{”}$ , F(1,29) = 15.87, p < 0.001, $η_{p}^{2}$ = 0.35, MSE = 0.13, revealing a more liberal criterion for negative words (M = -0.30, SD = 0.55) than for neutral words (M = -0.03, SD = 0.55). Neither language nor the interaction of emotion by language was significant in any of the analyses (all ps > 0.095).

The analysis of the source memory score (Hu) failed to show a significant effect of any of the factors or interactions involved (all ps > 0.314).

The results of this experiment demonstrated that emotion did not affect item memory. Although negative words showed a higher hit rate than neutral words, they also showed a higher false alarm rate. In fact, the false alarm rate was quite high in general in this experiment, suggesting that participants did not discriminate very well between old and new items. A possible reason for that might be that, due to our interest to match the old and new words as best as possible, there is a meaning similarity between the items of both sets. Apart from that, as above mentioned, false alarm rate was higher for negative words than for neutral words, leading discriminability (A′) to be the same in these two types of words. Furthermore, the criterion was more liberal for negative words than for neutral ones. This pattern of findings is in line with previous memory recognition studies (e.g., Doerksen and Shimamura, 2001; Pesta et al., 2001; Comblain et al., 2004; D’Argembeau and Van der Linden, 2004; Mackenzie et al., 2015), suggesting that participants have a tendency to label emotional items as old regardless of whether these had been presented during acquisition or not. However, the present results are at odds with past reports of an enhancing emotion effect in memory (e.g., D’Argembeau and Van der Linden, 2005; Cook et al., 2007; Schmidt et al., 2011; Rimmele et al., 2012; Bisby and Burgess, 2014; Meyer et al., 2015). It should be noted here that such an enhancing effect is much more common in free recall tasks (e.g., Doerksen and Shimamura, 2001; D’Argembeau and Van der Linden, 2004; Davidson et al., 2006; Maddock and Frein, 2009; Nashiro and Mather, 2011; Wang and Fu, 2011; Otani et al., 2012) than in recognition tasks, and in studies involving long retention intervals between the encoding and test phases (Kleinsmith and Kaplan, 1963; see Yonelinas and Ritchey, 2015; Dolcos et al., 2017, for overviews). Moreover, focusing on recognition tasks, the enhancing effect has been reported much more frequently with images (e.g., D’Argembeau and Van der Linden, 2005; Schmidt et al., 2011; Rimmele et al., 2012; Bisby and Burgess, 2014; Meyer et al., 2015) than with words (e.g., Doerksen and Shimamura, 2001; D’Argembeau and Van der Linden, 2004; Davidson et al., 2006). A possible reason for the above findings is that the higher similarity in meaning between emotional words in comparison to neutral words (Talmi and Moscovitch, 2004) benefits free recall but impairs recognition memory, making it more difficult to distinguish between old and new emotional words than between old and new neutral words (White et al., 2014). Such a difficulty might be reduced with images, because they are probably more distinctive than words. On the other hand, long retention intervals would allow an emotion advantage on item memory to emerge due to consolidation processes (LaBar and Cabeza, 2006). The use of an immediate recognition phase, as in the present study, involves testing memory before such consolidation processes have occurred.

A more relevant finding for the purpose of this study is that we failed to find any effect of emotion on source memory, in line with some past studies (Sharot and Phelps, 2004; Sharot and Yonelinas, 2008; Mather et al., 2009; however, see section “Introduction” for a review of the studies finding such an effect). This pattern of findings is not consistent with the proposals of Mather and her co-workers (Mather, 2007; Mather and Sutherland, 2011), considering that language is an intrinsic feature of words which is goal-relevant for the participants during encoding (i.e., they had to encode both the words and the language of presentation). Bearing these two factors in mind, an enhancement should have been clearly observed.

A possible reason for the lack of effects of emotion on source memory might be the nature of the encoding task. During encoding, participants were told that their memory would be tested afterward. It might be that the intention to learn both the words and the language in which they were presented overrides any effect that emotion could have had on source memory (i.e., by increasing performance with both neutral and emotional words). Hence, if we assume that the influence of emotion on cognition may involve automatic processes (e.g., an automatic activation of attention; D’Argembeau and Van der Linden, 2004), it is possible that the effects of emotion on source memory are more easily observed when participants do not intend to memorize the information. In fact, this was what D’Argembeau and Van der Linden (2004) found in a study on source memory for color in which incidental and intentional source encoding were compared. In order to further explore this possibility, we conducted Experiment 2, in which participants were not asked to memorize either the words or the language of presentation, although source information was still goal-relevant, as participants were required to name the words in the correct language.

Experiment 2

This experiment was carried out in accordance with the code of ethics of the World Medical Association (Declaration of Helsinki) and was approved by the Ethics Committee for Human Research (SECH 017/2015) of the Research Center on Psychology (CIPsi) at the University of Minho. Furthermore, participants gave written informed consent.