Seventeen right-handed students (9 females; aged 19–25 years; mean age = 22.4 years) who are all native Chinese speakers participated in the experiment. One dataset was excluded due to excessive muscle artifacts and electrode drift, leaving 16 for analyses. All the participants have normal or corrected visual acuity, and reported no current or past neurological or psychiatric disease. Each subject signed an informed consent form before experiment and received monetary compensation after experiment. This research was approved by the Human Research Ethics Committee at Capital Normal University, methods were carried out in accordance with its relevant guidelines.

The stimuli consisted of a set of 388 two-character Chinese full names, such as “王筠” (Wang Jun) and “刘蔷” (Liu Qiang) (Figure 1). It should be noted that family name (such as, “Wang” and “Liu”) were selected from a book named “The four hundred most common Chinese family name”, while first name (such as, “Jun” and “Qiang”) were obtained from various Internet sources. For all names, 8 names were used for practice and another 20 names were used for buffer. The remaining 360 names were equally divided into 2 lists: list 1 and list 2, both of which were rearranged as list 3 and list 4, respectively. The left-right study order of the names was also preserved in both old and rearranged pairs. For example, when “王筠” (Wang Jun) and “刘蔷” (Liu Qiang) were presented in list 1, “王蔷” (Wang Qiang) and “刘筠” (Liu Jun) were presented in list 3 (Figure 1). Likewise, when “李瑾” (Li Jin) and “张娟” (Zhang Juan) were presented in list 2, “李娟” (Li Juan) and “张瑾” (Zhang Jin) were presented in list 4. In test phase, for half of participants, list 1, list 3 and list 2 were used as old, rearranged and new stimuli respectively (eg., “王筠”, “王蔷”, “李瑾”), whereas for the other halves, list 2, list 4 and list 1 were used as old, rearranged and new stimuli respectively (eg., “李瑾”, “李瑾”, “王筠”). The test order of old and rearranged trials was controlled by ABBA balance method. Eleven students who did not take part in the formal experiment rated the familiarity of the names on a 5-point Likert scale ranging from 1 = lowest familiarity to 5 = highest familiarity. According to the statistical analysis of rating results, the four lists equaled in terms of familiarity (mean: list 1 = 2.57, list 2 = 2.51, list 3 = 2.63, list 4 = 2.48; F(3, 30) = 1.29, p > 0.05).

The experiment included a study phase and a test phase (Figure 1). In the study phase, participants firstly received a practice before formal experiment. They were informed that they needed remember or forget names according to instruction. Formal study consisted of 5 blocks, each of which contained 44 names bounded by filler names (two primacy buffers and two recency buffers). For the remaining 40 names, 20 names were instructed to remember, the other 20 names were instructed to forget. Each study trial initiated by a 500 ms presentation of a fixation cross, and then followed by a 500 ms presentation of a name. After names, a blank screen lasted for 2000 ms was inserted. Then, an instruction “记住” (denotes Remember) or “忘记” (denotes Forget) was displayed for 500 ms. Finally, another blank screen was shown for 1000 ms. The order of trials was pseudo-random with the constraint of no more than three consecutive trials for each type of instruction appearing in sequence. All stimuli were presented in white against a black background. After the study phase, participants had a rest of 2 min.

During the test phase, participants also firstly received practice in order to adapt to the task demanding. Formal test consisted of 5 blocks, each with 108 names (36 old, 36 rearranged, 36 new). Each test trial began with a fixation cross lasting for 500 ms, then a name (visual angle 5.30° × 1.43°) was presented for 2000 ms followed by a 1000 ms blank screen (Figure 1). Participants had to make a response before the next fixation cross presented. If both family name and first name in pairs were studied and encoded together at study, they would be judged as “old”; if both of them were studied, but not encoded together at study, they would be judged as “rearranged”; if both of them were not studied at all, they would be judged as “new.” Participants were reminded that they could not make a judgment for old and rearranged pairs according to instruction (“记住” or ”忘记”). The order of trials was also pseudo-random with the constraint of no more than three consecutive trials coming from the same type. Both accuracy and speed of response were emphasized. Response buttons for “old” and “new” were counterbalanced across participants.

Electroencephalographic (EEG) data were recorded in test phase of the experiment and measured from 62 Ag/AgCl electrodes embedded in an elastic cap with NeuroScan SynAmps system (NeuroScan Inc. Sterling, Virginia, USA). The electrode locations adhered to the extended international 10–20 system. All channels were referenced to left mastoid on-line and re-referenced off-line to averaged mastoids. EEG and EOG were amplified using a 0.01–100 Hz band pass and sampled at 500 Hz. Impedance was kept below 5 kΩ. Data were band pass filtered from 0.05 to 40 Hz off-line. EOG blink artifacts were corrected using a linear regression estimate. The averaged epoch was 1200 ms, including 200 ms prior to stimulus onset. Baseline corrections were performed using mean amplitudes of pre-stimulus onset. Trials exceeding ±90 μV were rejected.

We analyzed all data with SPSS 21.0. Repeated-measures ANOVA adjusted by Greenhouse-Geisser corrections when assumptions of sphericity in the repeated measures analyses were violated. The alpha level was 0.05. The analysis of behavioral data focused on accuracy and reaction times (RTs). The ERP amplitudes were averaged at two latency intervals in the test phase (300–500 ms and 500–800 ms) and over sets of midline electrode clusters along the anterior-posterior axis (frontal: F3/Fz/F4; central: C3/Cz/C4; parietal: P3/Pz/P4), which were selected on the basis of previous studies (Van Hooff et al., 2009; Tibon et al., 2014). The ERP amplitude for a certain electrode site was the average of three selected electrodes, for instance, the ERP amplitude for frontal area was the average of F3, Fz, and F4. Likewise, the ERP amplitude for a certain hemisphere was the average of three selected electrodes, for instance, the ERP amplitude for left hemisphere was the average of F3, C3, and P3.

This experiment constituted a 2 (Instruction: remember vs. forget) × 3 (Stimulus Type: old vs. rearranged vs. new) × 3 (Response Type: old vs. rearranged vs. new) incomplete within-subjects design. This design was necessarily incomplete because the instruction manipulation was not varied for new stimulus. Thus, we sorted the EEG data into 15 experimental conditions to examine the event-related potentials elicited by the names. For old names in TBR condition, if they were correctly remembered and judged by participants as “old”—TBR_R; if they were forgotten and were judged as “new”—TBR_F; finally, they could be judged as “rearranged”—TBR_r. For old names in TBF condition, if they were remembered and judged by participants as “old”—TBF_R, if they were forgotten and were judged as “new”—TBF_F; finally, they could be judged as “rearranged”—TBF_r. For rearranged names which consisted of items from names following by the same Remember or Forget instruction, they were denoted as TBRr and TBFr respectively. They could be judged as “old” (TBRr_R, TBFr_R), “new” (TBRr_F, TBFr_F) or “rearranged” (TBRr_r, TBFr_r). For new names, they could be judged as “old” (N_R), “new” (correctly rejected; N_CR) or “rearranged” (N_r). All abbreviations for each condition were shown in Table 1.

According to our purpose, we selectively focused on 6 conditions: TBR_R, TBF_R, TBF_r, TBRr_r, TBFr_r, and N_CR condition. The EEG of the selected 6 conditions were separately overlapped and averaged. The mean numbers (standard deviation) of artifact-free trials for them were as follows: TBR_R 61 (10), TBF_R 26 (9), TBF_r 33 (10), TBRr_r 39 (7), TBFr_r 29 (6), and N_CR 130 (25). The mean trial numbers (standard deviation) of the rest 9 conditions were as fol1ows: TBR_F 7 (5), TBR_r 19 (6), TBF_F 29 (12), TBRr_R 13 (5), TBRr_F 7 (5), TBFr_R 10 (5), TBFr_F 19 (6), N_R 9 (7), N_r 35 (20). There were two main reasons why we abandoned these conditions: firstly, the correlation of most conditions with our experiment purpose was low; secondly, trial numbers of these conditions were not sufficient to support further analysis (less than 16).

The analysis of ERP results was conducted including the following three steps. First, in order to investigate the dissociation between item retrieval and associative retrieval, we examined overall differences among TBR_R, TBRr_r, and N_CR condition. Both the stimuli and responses in these conditions differed from each other, but participants made correct responses in each condition. In Lyu et al. (2015) and Tibon et al. (2014), associative recognition was indexed by activation for old vs. rearranged pairs (i.e., a less negative deflection in response to old pairs relative to rearranged pairs). Item recognition was indexed by activation for old/rearranged pairs vs. new pairs, (i.e., greater negativity for new pairs compared with old and rearranged pairs). In this study, associative retrieval was defined by the ERP difference between TBR_R and TBRr_r condition. Item retrieval was defined by the ERP difference between TBR_R/TBRr_r, and N_CR condition.

Second, in order to investigate the DF effect, we examined overall differences among TBR_R, TBF_R, and N_CR condition. In Nowicka et al. (2009) and Van Hooff et al. (2009), directed forgetting for single items was defined by a more negative deflection in response to remembered TBF item relative to remembered TBR item. Meanwhile, they did not find difference between remembered TBF item and new item. In this study, DF effect was defined by the ERP difference both between TBF_R and TBR_R condition and between TBF_R and N_CR condition. As both TBF_R and TBR_R condition contained item and associative information, the ERP difference only reflected some uncertain kind of information was directed-forgotten (item, associative information, or both).

Third, we examined the DF effect for associative information. In Nowicka et al. (2009) and Van Hooff et al. (2009), directed forgetting for single items was also defined by a more negative deflection in response to forgotten TBF item relative to remembered TBF item and new item. In this study, we conducted three contrasts concerning associative information by using the similar logic to the “reversed old/new effect” in item recognition: (1) TBF_r vs. TBF_R condition. Participants in TBF_r condition correctly hit item information but wrongly missed associative information in the pair while they both correctly hit item and associative information in TBF_R condition. Therefore, the ERP difference between TBF_r and TBF_R condition reflected pure associative retrieval. Here, DF effect for associative information can be defined by a more negative deflection in response to TBF_r relative to TBF_R condition. (2) TBF_r vs. TBFr_r condition. Participants in TBFr_r condition not only correctly hit item information but also correctly rejected associative information in the pair. Therefore, the ERP difference between TBF_r and TBFr_r condition also reflected pure associative retrieval. DF effect for associative information can also be indexed by a more negative deflection in response to TBF_r relative to TBFr_r condition. (3) TBF_r vs. TBRr_r condition. Participants in TBRr_r condition not only correctly hit item information but also correctly rejected associative information in the pair. Although participants made correct responses to item information in TBF_r and TBRr_r conditions, the context of responses was different: to-be-forgotten in TBF_r condition while to-be-remembered in TBRr_r condition. Thus, the ERP difference between TBF_r and TBRr_r condition partly reflected mental process with respect to item information. However, they also partly reflected directed forgetting for associative information (i.e., a less positivity for TBF_r compared with TBRr_r condition).

In addition, to exclude the possibility that the between-condition difference in the 500–800 ms time window may be correlated with the difference in the 300–500 ms time window, we planned to confirm the independency between the two time windows by dissociating the two ERP components (i.e., FN400 and LPC). According to previous studies (Curran and Hancock, 2007), we determined to conduct topographical analyses for old/new effects. If the interaction of time window × location was significant, we can refer that the neuronal sources concerning the two components were different.

The mean accuracy and RTs in each condition are presented in Table 2 and Figure 2. According to our goal, we conducted planned comparison for these selected conditions. Because the following five contrasts between different conditions in accuracy and RTs were independent on each other, we did not perform correction for the multiple comparisons.

For mean accuracy, paired t-test showed that mean accuracy in TBR_R condition was greater than TBRr_r condition [t₍₁₅₎ = 10.406, p < 0.001]. Mean accuracy in TBF_R condition was lower than TBR_R condition [t₍₁₅₎ = −10.613, p < 0.001]. Mean accuracy in TBF_r condition was greater than TBF_R condition [t₍₁₅₎ = 2.121, p = 0.05]. Mean accuracy in TBF_r condition was greater than TBFr_r condition [t₍₁₅₎ = 2.535, p = 0.023]. Mean accuracy in TBF_r condition was lower than TBRr_r condition [t₍₁₅₎ = −2.869, p = 0.012].

For RTs, paired t-test showed that RTs in TBR_R condition was shorter than TBRr_r condition [t₍₁₅₎ = −7.766, p < 0.001]. RTs in TBF_R condition was shorter than TBR_R condition [t₍₁₅₎ = −6.379, p < 0.001]. RTs in TBF_r condition was longer than TBF_R condition [t₍₁₅₎ = 4.138, p = 0.001]. RTs in TBF_r condition was not shorter than TBFr_r condition [t₍₁₅₎ = 0.737, p = 0.473]. RTs in TBF_r condition was longer than TBRr_r condition [t₍₁₅₎ = 2.857, p = 0.012].

Participants in TBR_R condition made response more accurately and quickly than in TBRr_r condition, which indicated identifying rearranged stimuli was a difficult task. Overall, performance in TBF_R condition was inferior to that in TBR_R condition, indicating that DF effect was present. Associative information may be directed-forgotten by comparing TBF_r condition with TBF_R, TBFr_r, and TBRr_r condition.

The behavioral results showed that participants in TBR_R condition were more accurate and faster than in TBRr_r condition, which was consistent with a number of researches on comparing old pairs with rearranged pairs (Donaldson and Rugg, 1998; Liang and Guo, 2012; Lyu et al., 2015). In accordance with the Encoding Specificity Principle (Tulving and Thomson, 1973), item retrieval is more likely to be successful if items are presented at test in the context in which they are studied (old pairing) rather than in a new context (rearranged pairing). In this study, items at test in TBR_R condition appeared in the same context as encoding, whereas items in TBRr_r condition occurred in new context.

For ERP data, it can be inferred that item retrieval and associative retrieval dissociated in different time courses, that is, item retrieval appeared at 300–500 ms interval and associative retrieval appeared at 500–800 ms interval. We found TBR_R condition elicited more positive potentials than N_CR condition both in early and late interval, indicating that some kind of information was retrieved. However, which type of information (item or association) was uncertain. The ERP difference between rearranged and new pairs is thought to reflect item information as both associations in the two type of pairs were not studied in the study phase (Hockley, 1992). Likewise, the ERP difference between old and rearranged pairs is thought to reflect associative information on the base of two hypotheses. One is item information in the two types of pairs has equal extent familiarity (Hockley, 1992), the other one is associative information probably is an all-or-none variable (Kelley and Wixed, 2001). In 300–500 ms interval, the ERP difference between TBR_R condition and TBRr_r condition did not reach significance while TBRr_r condition elicited more positive potentials than N_CR condition, which indicated that participants only retrieved item information in early recognition. In 500–800 ms interval, we obtained a reversal pattern: the ERP difference between TBRr_r condition and N_CR condition was insignificant while TBR_R condition elicited more positive potentials than TBRr_r condition, indicating only associative information was retrieved in late recognition. This pattern of results was consistent with that of Lyu et al. (2015) and Tibon et al. (2014). Tibon et al. (2014) found differences between old and new pairs, and between rearranged and new pairs, but not between old and rearranged pairs by pairwise comparisons in 350–550 ms interval. For the later time window (550–750 ms), this analysis revealed a significant difference between old and rearranged pairs.

The idea that associative retrieval was indexed by the ERP difference between old and rearranged pairs was supported by indirect evidence. Donaldson and Rugg (1998) showed that old and rearranged pairs were associated with different magnitude old/new effects in three time windows (600–900 ms, 900–1200 ms, 1200–1434 ms). They explained that old pairs are more likely to engender recollection of associative information while rearranged pairs may mainly be formed from trials made by default strategy lack of recollection. Similarly, Rugg et al. (1996) found that recognized/recalled items elicited larger old/new effect than those for recognized/unrecalled items from 500 to 800 ms. They attributed this divergence to different source of information: information of recognized/recalled items is in terms of the successful recollection of study episode, whereas information of recognized/unrecalled items bases largely on familiarity. Furthermore, Jäger et al. (2006) found that face elicited old/new effect at 400–700 ms interval in inter-item condition when it was recognized and then forced-choice judged correctly. It suggested that recollection enables retrieving association between arbitrarily different faces.

Item retrieval was demonstrated to occur earlier than associative retrieval. Meanwhile, we provided evidence for the viewpoint that item retrieval relied only on familiarity and associative retrieval relied only on recollection. A lot of researches have revealed that associative retrieval depended only on recollection. Yonelinas (1997) reported that associative recognition is only based on recollection through comparing the ROC curve of item recognition with those of associative recognition. Hockley and Consoli (1999), employing “R/K” paradigm, subsequently found more “remember” response than “know” response in associative recognition. It is generally agreed that item retrieval can be supported by both familiarity and recollection (Yonelinas, 1997; Donaldson and Rugg, 1998; Jäger et al., 2006), but this idea has been challenged by Jacoby (1991) who posited that item retrieval depends only on familiarity. Liang (2013) further posited that there may be more familiar sensation in item retrieval and that recollection may be involved in retrieving the inherent characteristic or contextual details of items. In our experiment, participants made correct judgments via the pronunciation (or spelling) of names without retrieving details like the font, color, position, meaning, and so on. Thus, item retrieval relies merely on familiarity.

The amplitudes of ERP for TBF_r condition were lower than amplitudes of TBF_R, TBFr_r, TBRr_r condition at 500–800 ms interval (the reversed effect), which was similar to the findings of Nowicka et al. (2009) and Van Hooff et al. (2009). Combined with behavioral results, we concluded that associative information was directed-forgotten.

According to selective rehearsal account, the rote rehearsal for associative information is stopped at the time of presentation of Forget instruction (Bailey and Chapman, 2012; Yang et al., 2012). In other words, Forget instruction results in inhibition of normal encoding processes for TBF associative information. However, some researchers argued that Forget instruction initiates attentional inhibition mechanism, which may withdraw attention from TBF associative information and prevent the return of attention to their representations (Zacks et al., 1996; Nowicka et al., 2009). Both passive stop of rote rehearsal and active inhibition of attention return cause the inferiority of the encoding for associative information of TBF compared to that of TBR. Nevertheless, in accordance with the retrieval inhibitory account (Bjork and Bjork, 2003), the Forget instruction only blocked the access of retrieval for TBF associative information, which is available via successfully encoding but not accessible. In summary, the reversed effect for forgotten TBF names may result from either successful inhibition of encoding processes or retrieval processes or even both.

We found that ERP of TBF_R condition was more negative than that of TBR_R condition at the two selected time window. It can be inferred that the whole name can be directed-forgotten. Unfortunately, we cannot determine which kind of information was directed-forgotten. As directed forgetting for associative information was indexed by three reversed effect at 500–800 ms interval in our experiment, the ERP difference between TBF_R and TBR_R condition at 300–500 ms interval can be taken as an indication of directed forgetting for item information. In addition, the ERP difference between TBF_r condition and TBRr_r condition in 300–500 ms interval reflected the success of item retrieval under Forget instruction. Meanwhile, there was no ERP difference between TBF_r condition and TBFr_r condition in the time window of 300–500 ms. These results in concert reflected that item information was retrieved in the early time window.

In ERP results, the above evidence strongly suggested that remembered item information in TBF context evoked more negative-going potentials relative to that in TBR context. This was in line with two previous studies, which reflected that remembered TBR items are more likely to give rise to a sense of conscious recollection than remembered TBF items (Nowicka et al., 2009; Van Hooff et al., 2009). In behavioral results, we also replicated the critical finding from a body of studies (Paz-Caballero and Menor, 1999; Ullsperger et al., 2000; Nowicka et al., 2009; Van Hooff et al., 2009; Yang et al., 2012) in which participants were more accurate and faster in response to TBR item than to TBF item (i.e., typical DF effect).

The amplitudes of ERP for TBF_R condition were higher than amplitudes of N_CR condition at 300–500 ms interval, that is, remembered item information in TBF context evoked more positive potentials than new item information. This is inconsistent with Nowicka et al. (2009) and Van Hooff et al. (2009). In their study, ERP for remembered TBF items did not differ from that for new items, which was interpreted as inhibition of items in TBF context.

The discrepancy between the studies of Nowicka et al. (2009) and Van Hooff et al. (2009) and the current study might result from the types of stimuli and the encoding tasks. In their study, stimuli were concrete nouns (e.g., “desk,” “news”) which can quickly generate mental imagery corresponding to the referred objects. In contrast, in our experiment, the family name (e.g., “王” “刘”) and first name (e.g., “筠” “蔷”) were both abstract nouns. The concreteness effect of words indicated that concrete words are processed more quickly and accurately, and then better remembered than abstract words (Zhang et al., 2006). Furthermore, there was only one item for each trial in their studies, whereas there were two items in our study, which might aggravate working memory load (Alvarez and Cavanagh, 2004). Finally, for experimental procedure, Van Hooff et al. (2009) inserted a lexical decision task in which all studied words were presented repeatedly between study and test, resulting in deeply process for studied words. In contrast, participants received test immediately in our experiment, which ultimately resulted in weaker memory strength of items. Altogether, memory performance for items in our experiment was worse than theirs, which indicated by different old/new effect.

The amplitudes of ERP for TBF_r condition were lower than amplitudes of TBFr_r condition at 500–800 ms interval, that is, forgotten associative information in TBF context evoked more negative potentials than new associative information. This seemed to indicate that the neural pattern of directed forgetting for associative information differed from that for item information. As already discussed, in our experiment, item retrieval only based on familiarity which is lack of details, but associative retrieval only engendered recollection which contains numerous details. Thus, associative retrieval retrieves more details than item retrieval. In addition, there is a possibility that the degree of inhibition for item retrieval is equal to that for associative retrieval at the very beginning. Because it was acknowledged by Lyu et al. (2015) and our results that item retrieval occurs before associative retrieval and item retrieval has priority for the use of resources. Thus, the resources are adequate for item retrieval so that the implementation of Forget instruction is blocked. However, according to limited resources theory, the remaining resources for associative retrieval are not sufficient for conquering suppression of Forget instruction, accompanied by competition for the same limited resources between the two types of retrieval (Kahneman, 2003). Finally, associative retrieval is more inhibited than item retrieval, which is demonstrated by different ERP effect.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.