Strategy adoption in the n-back task

The n-back procedure is a complex task that may require (1) maintenance of the previous n items in memory, (2) updating of new items for active maintenance, (3) creation of bindings between each stimulus and its temporal context, and (4) resolving interference from non-relevant trials (Chatham et al., 2011; Cohen et al., 1997;

Oberauer, 2005). However, strategy adoption in the task can vary between participants or task demands (Botvinick, Braver, Barch, Carter, & Cohen, 2001), and the strategy used can mediate the working memory resources engaged (Juvina & Taatgen, 2007).

The Dual Mechanism of Control framework (Braver, 2012) describes a proactive control process that involves activating the target n-back item in advance, ready for comparison to the anticipated stimulus. That is, for each trial the ‘correct’ n-back item is dynamically prepared prior to presentation of the item to which a response must be made. Specifically, participants activate the n^th back item in memory prior to the trial to decide if the forthcoming item is a ‘hit’. In comparison, reactive control initiates attentive mechanisms upon presentation of the stimulus, after which competing responses are activated and the correct response must be selected. Specifically, following presentation of the item, participants attempt to retrieve the n^th back item in memory to establish if a correct match exists.

Several models have been produced that attempt to explain the precise processes involved in effective n-back task performance (e.g. Chatham et al., 2011; Gosmann &

Eliasmith, 2015; Juvina & Taatgen, 2007; Szmalec et al., 2011). A common component in such models is an active rehearsal strategy, where a rehearsal window of size n is maintained (Harbison et al., 2011; Juvina & Taatgen, 2007; Szmalec et al., 2011). These rehearsal strategies involve updating, where new items are added to the rehearsal window and the now-irrelevant items removed (see also Wilhelm et al., 2013).

However, rehearsal strategies may themselves differ in whether proactive or reactive control processes are employed (Ralph, 2014).

A proactive, static rehearsal process, is proposed in Chatham et al. (2011) where the n serial positions are held in memory separately to the n memory items. In this model, each item is allocated to a serial position, but only the task-relevant serial position is

under the focus of attention when a new item is presented. This new item is then compared to the item in that serial position, a response is made, and the new item then replaces the previous memory item in that same position. Importantly, the attentional focus then allocates the next serial position as task-relevant, and the same matching process is completed. Such a method is efficient because it does not require updating of the position of every item, and does not require any updating should the new item match the old (a target) (Ralph, 2014).

In comparison, Juvina and Taatgen (2007) describe a rolling rehearsal strategy where items are rehearsed in the phonological loop to increase their activation strengths. At each trial, new items are appended to the list whilst the first item is removed. That is, when an item falls out of the maintenance window this item becomes irrelevant, and this highly activated item must be removed and inhibited to prevent its reappearance in the rehearsal list. However, interference may still arise from this removed item due to the limited capacity of a suppression mechanism (Harbison et al., 2011; Jonides & Nee, 2006; Juvina & Taatgen, 2007; Szmalec et al., 2011). This strategy reflects a reactive control process, where all the items in the rehearsal window are retrieved in response to a new stimulus, and the strongest activation in this list is then selected and compared to the presented item (Ralph, 2014).

Though the above strategies may differ in their use of proactive or reactive control, they are similar in their requirement for active maintenance of the stimulus. Importantly, it should also be noted that such maintenance is not necessarily specific to rehearsal within the phonological loop, as rehearsal may instead be attributed to a multi-modal refreshing mechanism (Cohen et al., 1997; M. R. Johnson et al., 2015). Indeed, there is evidence that suppressing an articulatory rehearsal process (e.g. via repetition of the word ‘the’) will cause participants to recruit additional, domain-specific, maintenance

resources (Chein & Fiez, 2010; Gruber & Cramon, 2003). Consequently, these control mechanisms may accommodate n-back performance for non-verbal stimuli such as fractals, abstract shapes (Nystrom et al., 2000; Ragland et al., 2002), faces (Dade et al., 2001), and olfactory stimuli (Dade et al., 2001; Jönsson et al., 2011).

Further distinction between n-back strategies considers the use of control. That is, whilst rehearsal strategies reflect high control, other strategies modelled for the task propose low control processes that allow effective performance without rehearsal of the memoranda (e.g. Juvina & Taatgen, 2007; B McElree, 2001). For example, a slow, reactive search process has been proposed to aid recovery of order information in circumstances where the n-back item is not maintained in focal attention (McElree, 2001). In this model, the most recent item is retrieved, which cues the next in the sequence, and so on until a match to the probe item is found. Alternatively, Juvina and Taatgen (2007) propose a low control strategy that compares a temporal-based estimation of a target item’s encoding (a ‘time-tag’) to the approximate age of a target item (see also Nijboer, Borst, van Rijn, & Taatgen, 2016). To be clear, rather than actively maintaining the n items in memory, participants rely upon residual memory traces of the items to judge whether the current item matches the activation level expected of a target (i.e. a temporal familiarity judgement).

A variable task demand that may affect the employment of a particular n-back strategy is the prevalence of recent lures (Ralph, 2014). In his thesis, Ralph (2014) assessed whether the ratio of targets to lures influences the control strategies adopted in the n-back task, hypothesising that a lure-heavy sequence will increase the use of proactive control (i.e. engaging attention to the stored target item prior to presentation of the probe item). In direct contrast to the prediction, fewer target trials in a sequence resulted in a decrease in the use of a proactive control method. In explanation, he proposes the

effort involved in keeping targets active is not worth the reward of getting relatively few trials correct when the number of targets is low. Consequently, his findings provide evidence that participants are able to extract information about the nature of the task, and to make adjustments to their control strategy accordingly (Ralph, 2014).

It is also possible that participants may make a strategic decision to make n-back judgments based only on familiarity (Juvina & Taatgen, 2007). Above-chance performance during the n-back task is possible if a strategy is adopted to accept probes when a familiarity-strength criterion is exceeded, and this method is likely the primary process used to reject non-recent lure items (Harbison et al., 2011). However, the inclusion of recent lures mean familiarity cannot be used to accurately identify a target item, and a cognitive control process must be adopted to discriminate these lures from targets to prevent inflated false alarm rates (Juvina & Taatgen, 2007; Szmalec et al., 2011). It appears that participants will typically opt for control strategies (i.e. a strategy that attempts to explicitly compare the probe item to information linked to the n-back serial position) to maximise accuracy (see McElree, 2001), though sequences with a low number of recent lure items can increase reliance on familiarity (Harbison et al., 2011).

This strategy may also occur if there are failures in recollection that prevent a judgement of the probe’s position (Juvina & Taatgen, 2007). Recollection-based decisions must be made to ensure a target or recent-lure decision is correct, and it is this process that is proposed to engage cognitive control processes (Juvina & Taatgen, 2007;

Smith & Jonides, 1999).

How cognitive control strategies in the back task may be applied to an olfactory n-back task is unclear. Verbal recoding would allow the use of proactive or reactive high-control rehearsal strategy (Chatham et al., 2011; Juvina & Taatgen, 2007), and could explain a relationship between odour verbalisability and n-back performance (Jönsson et

al., 2011). However, the dissociation of verbal and olfactory short-term memory indicates that a purely verbal rehearsal process may not be appropriate (Andrade &

Donaldson, 2007). A high-control strategy may instead be applied to an olfactory representation through a refreshing process (e.g. M. R. Johnson et al., 2015). This refreshing process involves directing attention to one of several active representations (M. R. Johnson et al., 2015; Raye et al., 2007), and is proposed to drive non-verbal rehearsal (Baddeley, 2012). However, olfactory imagery has been proposed unique in its inability to give rise to a conscious olfactory representation (Stevenson & Attuquayefio, 2013; Zucco, 2003), meaning such rehearsal-based strategies may not be possible. If this is the case, the strategy might be mediated by whether an odour is identified or named. For hard-to-name odours, this may involve the adoption of a low-control strategy, which involves a reactive memory search, or a familiarity-based temporal estimation (Juvina & Taatgen, 2007; B McElree, 2001). Alternatively, failure to recollect these low verbalisability odours may result in the adoption of a familiarity that does not include any control process (i.e. acceptance of an item if a familiarity strength signal exceeds a fixed criterion, Juvina & Taatgen, 2007).