Working memory (WM) allows us to hold and manipulate information that is relevant to our current task goals and plan objectives over a short period of time (Baddeley, 1986). However, the capacity of WM is highly limited (Luck and Vogel, 1997; Cowan, 2001) – only about three or four pieces of information can be retained at any given time. The underlying processes and mechanisms that support the constraints of capacity are largely investigated in the WM literature (Jonides et al., 2008; Edin et al., 2009; Luck and Vogel, 2013; Ma et al., 2014; D’Esposito and Postle, 2015; Stokes, 2015; Postle, 2006, 2015). Evidence from this body of research has revealed that attention is of importance in controlling the contents of WM. Whereas past selection history affects the allocation of attention on target selection (Awh et al., 2012; Belopolsky and Awh, 2016), it remains unexplored whether selection history can influence WM. The goal of this study aims to investigate whether the history of attentional processing can modulate WM capacity.

Accumulating evidence has revealed that WM capacity is modulated by attention (Awh and Jonides, 2001; Gazzaley, 2011; Gazzaley and Nobre, 2012). These studies show that top-down attention serves as the selective mechanism for encoding task-relevant information into WM (Schmidt et al., 2002; Murray et al., 2011) and introducing bias toward cued items during WM maintenance (Griffin and Nobre, 2003; Lepsien and Nobre, 2007; Kuo et al., 2011, 2012, 2014; Astle et al., 2012; Pertzov et al., 2013; Rerko and Oberauer, 2013). These findings suggest that attention mechanisms may determine the total amount of information that can be effectively stored in WM (Edin et al., 2009; Oberauer and Hein, 2012). For example, evidence from electrophysiological recordings in humans revealed that these top-down mechanisms are biased toward a subset of relevant items and prevented irrelevant distractors from entering WM (Vogel et al., 2005).

Recently, an inspiring attention framework suggests that the history of attentional control affects the selective biasing of information processing (Awh et al., 2012; Belopolsky and Awh, 2016). In this framework, recent history of attentional selection (i.e., selection history) elicits a lingering consequence of past selection episodes or goals (Awh et al., 2012, p. 437). The lingering effect of past selection can form a bias toward recently activated representations or templates, exerting a powerful influence on the selection priority for a target item to accomplish the current task. Behavioral and electrophysiological studies in humans have shown that selection history may cause an inter-trial priming effect for the processing of the predefined target features (Maljkovic and Nakayama, 1994; Hillstrom, 2000; Fecteau and Munoz, 2003; Wolfe et al., 2003; Geyer et al., 2006, 2011; Eimer et al., 2010; Kristjánsson and Campana, 2010; Brascamp et al., 2011; Töllner et al., 2012; Gokce et al., 2014; Feldmann-Wüstefeld and Schubö, 2016). For example, when participants were required to judge the shape of a pop-out color target amid multiple distractors, they showed faster search times when the target color repeated the one shown in the preceding trial relative to the non-repeated trials (Maljkovic and Nakayama, 1994). A recent study using a flanker task also showed that the context-driven selection history modulated distractor processing so that the processing mode in one block persisted in the subsequent block (Yeh et al., 2014). These findings indicated that an implicit setting of selection, inter-trial repetition or consistent processing mode across trials, can be embedded within a task context to influence the perceptual processing of a target item.

Given the influences of selection history on attentional control and the importance of attention in WM, selection history should influence WM performance. Yet, no prior study has investigated the impact of selection history on WM. This study aims to test the influences of context-driven selection history on WM capacity. Participants performed a delayed response WM task for colors (Experiments 1 and 3) and shapes (Experiments 2A and 2B) with both high (three items in Experiments 1 and 2; four items in Experiment 3) and low WM loads (one item). They were first presented with one item (low WM load) or three/four items (high WM load) within a memory array, and following a short retention interval, they responded to a test probe based on the retained information. Selection history was operationally defined as the number of items that had been attended across trials in a block, manipulated by the stimulus set-size (4-item and 8-item contexts in Experiments 1 and 2, and 5-item and 9-item contexts in Experiment 3) from which the memorized content was selected. Although the memory array always contained one or three/four items, the WM task was performed in two different contexts. In one context (4-item or 5-item), fewer memorized contents from the previous trials could interfere with the processing of task-relevant items in the current trial. In another context (8-item or 9-item), more representations from previous trials could interfere with stimulus processing in the current trial. The lingering effect refers to the influence of previous memorized content on the efficacy of attentional control on performing the WM task. Attentional control should be less effective when more representations have been activated unless only one item needs to be encoded in the WM task. Thereby, I hypothesized that WM capacity would be reduced in the context with more possible stimuli unless attention focuses on a single item to protect it from interference caused by the lingering effect of previous trials.

Previous research has shown that WM capacity may vary across stimulus categories (Alvarez and Cavanagh, 2004). The goal of Experiments 2A and 2B was to replicate the results of Experiment 1 based on a different stimulus category – shape.

The results from the previous experiments clearly showed that context-driven selection history can influence WM capacity, especially with high WM load. It is noted that high WM load was three items in these three experiments. The goal of Experiment 3 aimed to test whether the influences of selection history on WM capacity can also be observed when the memory set-size was increased to reach the limit in capacity – four items (Luck and Vogel, 1997; Cowan, 2001) for high WM load.

Goal-directed behaviors depend upon the allocation of attention toward a subset of relevant information from the external environment and within the internal representations (Awh and Jonides, 2001; Chun et al., 2011; Gazzaley, 2011; Gazzaley and Nobre, 2012). A growing body of evidence has revealed that the context-driven selection history of attentional deployment can generate a lingering bias in selection (Eimer et al., 2010; Kristjánsson and Campana, 2010; Geyer et al., 2011; Awh et al., 2012; Belopolsky and Awh, 2016; Jost and Mayr, 2016). The present study highlighted the influences of selection history on the efficacy of attentional control on WM representations and affect WM capacity in a delayed response task. The context-driven selection history was defined as the number of items that had been attended across trials within a block while performing the WM task. On each trial of different blocks, one (low WM load) or three/four items (high WM load) were randomly selected from a subset of fewer possible stimuli (4-item or 5-item context) or more possible stimuli (8-item or 9-item context) as the to-be-remembered item(s). Participants were instructed to match a test probe with the memorized content after a retention interval and were not informed about the manipulation of the task context. The main finding was that WM capacity was significantly reduced in the context with more possible stimuli relative to the context with fewer possible stimuli. More importantly, this reduction in capacity was observed when WM load was high and was not observed when attention was focused on only one item for WM maintenance. Finally, this detrimental effect in capacity was obtained when the memory set-size reached the capacity limit, e.g., four objects, in WM.

Prior research has examined the mechanisms underlying the impact of selection history on attentional allocation at the perceptual level. These investigations demonstrated that context-driven attention allocation elicited inter-trial priming in visual search tasks (Hillstrom, 2000; Fecteau and Munoz, 2003; Wolfe et al., 2003; Geyer et al., 2006; Becker et al., 2009; Kristjánsson and Campana, 2010; Belopolsky and Awh, 2016). Maljkovic and Nakayama (1994) proposed that the search history of attention-driving features influences search in the subsequent trials. Such inter-trial priming can facilitate the perceptual processing of the search targets and yield selection benefits in response times on repeated trials compared with non-repeated trials. A similar inter-trial priming effect was found in a recent study in which participants’ response to a target singleton was interfered with by an irrelevant color singleton (Theeuwes and Burg, 2011). However, this interference was diminished when the selected feature remained the same from one trial to the next trial. Inter-trial priming could accumulate, through repeating the same spatial layout of stimuli across blocks, to guide spatial attention to the target location and facilitate target search (Chun and Jiang, 1998; Olson and Chun, 2001). This contextual cueing benefit was observed even though explicit memory performance measured by recognition was at the chance level.

Recent electrophysiological studies in humans that exploited an event-related potential marker of attentional selection (e.g., N2pc) (Luck and Hillyard, 1994; Eimer, 1996) in a visual search task provided neural evidence in support of inter-trial priming (Töllner et al., 2012; Gokce et al., 2014). For example, the onset latency of N2pc was delayed when target and distractor colors were changed compared to when they were repeated (Eimer et al., 2010). The subcomponent of the N2pc (e.g., Nd, the negative part of the N2pc contralateral to distractors) was also influenced by the repetition of distractor feature (Feldmann-Wüstefeld and Schubö, 2016). The repetition of distractors can decrease attentional capture by task-irrelevant stimuli in the blocks where participants can expect a specific distractor color compared to the blocks where participants cannot predict the distractor color. Together, these results suggest that the repetition of a feature attribute across trials allows the mechanisms of attention more efficient processing.

The implicit and autonomous influences of selection history on target processing have also been observed in the research line of statistical regularities (Chun and Turk-Browne, 2007; Turk-Browne et al., 2010; Umemoto et al., 2010). In these studies, statistical regularity was usually manipulated in a stream of items with embedded sequences so that certain stimuli precede or follow with other stimuli repeatedly. Selection history in this experimental context refers to the contingency of stimulus arrangement across trials. The extraction of the regularities or patterns can bias the allocation of attention in guiding goal-directed behaviors and thus modulate perceptual and mnemonic operations. For example, regularities triggered implicit perceptual anticipations for perceiving face and scene stimuli (Turk-Browne et al., 2010). In this functional magnetic resonance imaging study, sequential contingencies (e.g., paired images) were embedded in a continuous stream of visual images while participants performed categorical responses. Although participants were not aware of the existence of the trial structure, significant neural activity was observed to the predictive stimuli in the learning-related brain area (e.g., hippocampus).

Regularities also facilitated the encoding of the to-be-remembered items in the location where the targets were presented more likely, even though the participants were not aware of the contingency in a WM task (Umemoto et al., 2010). A recent study also showed that memory of a stream of objects in a structured sequence (e.g., triplet element) based on temporal co-occurrence was better than memory of the objects in random sequences (Otsuka and Saiki, 2016). In this study, the regularities from the structured sequence induced implicit allocation of attention toward the triplets and improved memory performance for each triplet constituent. Moreover, they showed that the benefits of regularities were eliminated when an unexpected distractor was inserted in the structured sequence of the triplets. The benefits of statistical regularities on memory support the notion that the history of stimulus contingency across trials influences memory performance without awareness of the contingency.

The current experimental context differs from those adopted in the study of inter-trial priming and the study of statistical regularity. The memorized content was randomly selected on each trial so that inter-trial priming could not function effectively. No structured sequence was embedded across trials so that statistical regularity could not be formed. Selection history, in a broad sense, refers to past episodic traces of attended representations for accomplishing the WM task goal. A small number of representations were activated in the 4-item/5-item context whereas a large number of representations were activated in the 8-item/9-item context. The history was implicitly accrued in the context for performing the memory task. The effect of context on WM capacity is history-driven, that is not relevant to the current top-down behavioral goal and is not related to stimulus-driven aspects of attentional processing. For the delayed responses, the participants were always presented with a memory array that consisted of one or three/four target items. They were not informed about the manipulation of the task contexts and the task required no explicit knowledge of the past trials. There was no visual information other than the WM targets within the memory array. This experimental design is important for ensuring that the observed effects of selection history are unlikely to be the results of perceptual interferences from the visual cues or distractors. The novel finding of the impact of selection history on WM capacity is significant.

The results of the current study showed null effect when only a single item was focused on and maintained in WM. The attentional demand for encoding one item in WM is relatively low and a single item can be effectively processed and represented in WM. This finding supports the notion that only one item can be focused in attention for selecting goal-directed responses (Oberauer, 2002; Olivers et al., 2011). In this case, attentional mechanisms can effectively modulate this focused representation and resolve the competition from the previously memorized items. Thus, whether the context contained a small or large number of previously memorized items does not matter. Alternatively, this null effect of selection history for the low WM load may result from a ceiling effect as the task is relatively easy.

When WM load was high with three or four items, selection history showed a detrimental effect on WM capacity. In the context with fewer possible stimuli, experience accrued to build episodic traces of four or five different stimuli that had been attended and tagged as task-relevant target templates for WM decision across trials. When a display of three/four items was shown on a trial, only one once-attended target template did not occur. Participants may adopt a discarding strategy to perform high WM load trials by remembering which item was not shown in the memory array. For example, if red, green, blue and yellow were the possible stimuli in the 4-item context, and the to-be-remembered items in a given trial were red, green and blue, participants might just remember “it’s not yellow.” Although participants would need to alter their decision rules between trials for the match and no-match judgments based on WM load, they just have to remember one item in this context. Thus, WM capacity was higher compared with the context that contained more possible items. In the latter context, episodic traces contained eight or nine once-attended target templates across trials. Among these templates, five did not occur in a display of three/four items. The discarding strategy could not function effectively and hence, these five once-attended target templates could cause strong interference on WM operations. This alternative interpretation cannot be fully excluded, because of an increase in false alarm rates for high WM load in the context with more possible items. The discarding strategy, however, cannot explain the results of Experiment 2A because context did not significantly affect false alarm rates. Moreover, this discarding strategy should have led to high K measures for the high load condition as only one item (e.g., “it’s not yellow”) is tagged in WM. The results of Experiments 2A and 2B did not support this prediction because K was smaller than 2 when three items must be remembered in the 4-item context.

The differential effects of selection history on WM capacity in the current study are in line with the three-embedded components model of WM that distinguishes three states of memory representation in declarative WM (Oberauer, 2002; Lewis-Peacock et al., 2012; Oberauer and Hein, 2012; Larocque et al., 2014). In this model, WM emerges from the interaction between attention and long-term memory (LTM; i.e., activated part of LTM). The region of direct access stores the relevant information selected from the activated LTM representations that is required for immediate access (Woodman et al., 2007; Carlisle et al., 2011). The direct-access region consists of only a subset of elements (e.g., three or four items) and their relationships, which are temporarily bound to the current context. Finally, the state of the focus of attention serves as a selection device. It selects a single item or chunk from the set of information in the direct-access region for action selection. The present results suggest that the once-attended items activate representations in LTM and could cause interference only when multiple items must be retrieved from activated LTM into the region of direct access. The lingering effect does not affect the state in the focus of attention. This theoretical speculation requires future work to examine how the selection history and focus of attention influence WM using various techniques and paradigms. Future research should also test the generalizability of the current results to WM load higher than the capacity limitation of WM (e.g., six items in the memory array).

The findings from the current study suggest that the lingering effects of selection history for recently attended stimuli can cause strong interferences with currently relevant WM targets and reduce WM capacity, especially when the inter-item competition is strong between recently attended target templates and currently maintained representations. The lingering selection bias is considered an implicit context-driven effect. The present results highlight the influences of the history of selection and focused attention in WM capacity. These findings also bolster the notion that WM representations are highly flexible and susceptible to the different task contexts.

The author confirms being the sole contributor of this work and approved it for publication.

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