Attention to Information in Working Memory

The Three-Embedded-Components Model of Working Memory

This three-embedded-components model (Oberauer, 2002, 2009) is an extension of the model proposed by Cowan (1995). It distinguishes three components in working memory: the activated part of long-term memory (LTM), the region of direct access, and a single-item focus of attention (Fig. 1). The activated part of LTM keeps information available that could be relevant to the current task. For instance, during performance of an arithmetic-updating task, the activated part of LTM would hold digits and mathematical operators available. Some of the representations in activated LTM are held in the region of direct access, which roughly corresponds to the focus of attention in Cowan’s model; we therefore refer to it here as the broad focus. In contrast to the activated part of LTM, the broad focus of attention has a limited capacity. It is assumed to hold about four items or chunks of information available at a time and to bind them into new structures. For instance, when a person holds a running sum of 324 in mind and has to add 12 to it, the three digits of the running sum are held in the broad focus, and each digit is bound to its role (i.e., ones, tens, and hundreds). The narrower, single-item focus of attention serves to select one item or chunk as the target of the next cognitive operation (Oberauer, 2002). In the running-addition task, this item would be the single digit that is to be updated by the next arithmetic operation.

OPEN IN VIEWER

Fig. 1.

An illustration of the three-embedded-component model of working memory. Within the network of representations in long-term memory, a subset is activated above baseline (depicted by gray circles). Of these representations, a small number (a, b, and c) are held in the region of direct access (or the broad focus). These items are directly accessible through temporary bindings (indicated by broken lines) that link them to locations in a mental space (e.g., to ordinal positions in a list or to spatial positions on the screen) that serve as cues. The narrow focus of attention (depicted by the bold circles) accesses a single item, using one location in mental space as a cue. Adapted from Psychology of Learning and Motivation (Vol. 51, p. 50), by B. H. Ross (Ed.), 2009, San Diego, CA: Academic Press. Copyright 2009, Elsevier. Adapted with permission.

Evidence for the Broad Focus of Attention and Activated Long-Term Memory

In conditions that largely exclude contributions from LTM, people can hold approximately three to four independent items in working memory (Fukuda, Awh, & Vogel, 2010; Oberauer & Kliegl, 2006). This ability reflects the capacity of the broad focus. In many circumstances, it can be supplemented by the maintenance of currently unneeded information in activated LTM.

This was demonstrated by an experiment using the paradigm illustrated in Figure 2 (Oberauer, 2005b). On each trial, participants were required to memorize two short lists of words; the set size of the two lists (i.e., the number of words) was varied independently. Participants then made two successive recognition decisions; each decision was preceded by the presentation of a cue indicating which list was relevant for that decision. The probe followed the cue after either 100 or 2,000 ms. At short (100-ms) cue-probe intervals, the set size of both lists affected response times (RTs), whereas at long (2,000-ms) cue-probe intervals, only the size of the relevant list affected RTs, a pattern of results suggesting that within 1 to 2 seconds, the irrelevant list was removed from the capacity-limited broad focus. However, the irrelevant list was not entirely forgotten: Rejecting probes from the irrelevant list was harder for participants and took them more time than rejecting new word probes did. In a previous experiment using this paradigm, this intrusion cost lasted for 5 seconds after the list cue had been presented (Oberauer, 2001). Moreover, when the second cue indicated that the previously irrelevant list was now relevant, participants could accurately respond to the second cue, and their RTs for the second probe were influenced again by the set size of the previously irrelevant, now relevant list (see Fig. 3).

OPEN IN VIEWER

Fig. 2.

The two-cue, two-probe recognition paradigm used by Oberauer (2005b). After two lists, presented in different colors, are encoded in working memory by participants, one list is cued as relevant for the first probe. The first probe must be rejected, which is difficult because it was included in the irrelevant list, which is still activated in LTM. The second cue indicates that the previously irrelevant list is now relevant. The second probe is to be accepted.

OPEN IN VIEWER

Fig. 3.

Response times in the two-cue, two-probe recognition experiment of Oberauer (2005b). CPI = cue-probe interval. Data for the second probe come from trials on which the first cue and the second cue indicated that different studied lists were relevant. Thus, the set size labeled “relevant” for the second probe is the set size of the list that was irrelevant for the first probe, and vice versa.

These results imply that people can flexibly select a subset of information in working memory to which immediate access is required. This is the function of the broad focus of attention: It provides a candidate set of items for direct access—for instance, a subset of items to be compared with a probe. Increasing the number of items in the broad focus slows down access because the items interfere with each other. While the broad focus maintains this set of items, other information can be maintained outside the broad focus in activated LTM. People can flexibly outsource a temporarily irrelevant subset of their working memory contents into activated LTM and retrieve it back into the broad focus if it is needed later.

In a recent study, Lewis-Peacock, Drysdale, Oberauer, and Postle (2011) traced the neural substrate of this process. The researchers asked people to remember two items from two different content categories (phonological, semantic, or visual-spatial). This task was followed by two recognition tests, each preceded by a cue indicating which of the two items was relevant for the next recognition decision. Lewis-Peacock et al. used a pattern classifier to identify the content category reflected in people’s fMRI blood-oxygen-level-dependent (BOLD) signals at different time intervals during a trial. Following each cue, only the content cued as relevant was reflected in ongoing neuronal activity (as indicated by the BOLD pattern). Neuronal activity corresponding to the currently irrelevant content dropped to baseline, but it resurfaced when the second cue indicated that the previously irrelevant content was now relevant. Thus, it appears that only the contents of the broad focus are maintained by persistent neural activity during a delay period, whereas information outside the focus is maintained without ongoing neural activity. Hence, the term “activated LTM” might be a misnomer, because with regard to neural activity, this is the content of working memory that is least active.

Evidence for the Narrow Focus of Attention

Evidence for a single-item focus of attention in working memory comes from four experimental findings: recognition–retrieval dynamics, performance costs resulting from switching between items in working memory (i.e., object-switch costs), retrieval benefits from cueing individual items, and the limitation of visual search to a single search template.

Studies measuring speed-accuracy trade-off (SAT) functions for immediate recognition in a Sternberg (1966) recognition paradigm have provided evidence that the last item in a studied list has privileged status in working memory (e.g., McElree & Dosher, 1989). Measuring SAT functions enables an estimation of a person’s rate of processing (in this case, of retrieval and a recognition decision) independent of his or her chosen trade-off between speed and accuracy. The last list item was accessible at a higher rate than any other item from the list, which indicates that the last item was already in the focus of attention when the probe was presented, so no retrieval process was required. All other list items had statistically indistinguishable retrieval rates, supporting the assumption that only the last item was held in the narrow focus of attention.

In the object-switch paradigm, participants carry out a sequence of cognitive operations on individual items (or objects) encoded in working memory; each successive operation is applied either to the same object as before or to a different object stored in working memory. For instance, Garavan (1998) presented geometrical shapes (triangles and rectangles) one by one to participants and asked them to keep a running count of the two categories. On each trial, participants had to either update the previously updated counter (object-repetition trial) or the other counter (object-switch trial). RTs were approximately 300 ms slower after object switches than after object repetitions. Such switch costs (or repetition benefits) indicate that the object processed last is still held within the focus of attention when the next processing operation commences. If this object is the target of the next action, it does not need to be retrieved into the focus. Conversely, in case of an object switch, the focus requires time to access the new target object from the candidates held in the broad focus.

Further research with the object-switch paradigm has shown that the focus is not strictly limited to a single item. When two items are needed for a cognitive operation (e.g., addition of two digits), they can be accessed simultaneously by the focus by being chunked (Oberauer & Bialkova, 2009). Moreover, if those items are highly distinct so that they cannot be confused with each other, the focus might be able to maintain them simultaneously (Gilchrist & Cowan, 2011; Oberauer & Bialkova, 2011). Thus, the limitation of the narrow focus to a single item appears to be not structural but functional: It serves to unambiguously single out the information needed for an operation. For instance, holding two separate digits in the focus would leave ambiguity as to which digit filled which role in an operation. In contrast, when the task is to transform a digit using arithmetic and a spatial location using mental shifts, there is no ambiguity as to which operation to apply to which object, and both can be held in the focus simultaneously.

Makovski, Sussman, and Jiang (2008) and Griffin and Nobre (2003) used the change-detection task (Luck & Vogel, 1997) to investigate the narrow focus in visual working memory. Participants encoded into working memory a variable number of visual objects displayed simultaneously in a prime display. After a short interval, a probe display was presented that either matched the first display (match trials) or included a change in one of the objects. Between the offset of the prime display and the onset of the probe display, a cue was presented that marked one location from the prime display, indicating which item, if any, would be different in the probe display. Providing such a retro-cue led to faster RTs and better accuracy in change detection compared to providing no cue or compared to invalid cues. When two retro-cues were presented simultaneously, highlighting two different locations associated with different objects, no cueing benefit was observed (Makovski & Jiang, 2007). This result indicates that only one item can be focused on at a time.

Further evidence for a single-item focus has come from visual-search tasks. These tasks require participants to search for a target among a number of distractors in a visual display. To do this, people need to hold a template of the search target (i.e., a representation of what they are looking for) in working memory and compare it with the visual input until they find a stimulus that matches the search template (Duncan & Humphreys, 1989). People use only one item in working memory as a search template at any given time (for a review, see Olivers, Peters, Houtkamp, & Roelfsema, 2011). When participants are required to compare sequentially presented visual stimuli with two targets simultaneously, accuracy decreases, and the magnitude of this decrease is predicted by the assumption that only one search template can be used at any time (Houtkamp & Roelfsema, 2009).

One might ask why people don’t hold both templates in the narrow focus even though it would presumably improve performance, particularly if the two targets are highly distinct, as they were in one of Houtkamp and Roelfsema’s experiments. The answer may be that doing so is not advantageous after all. Holding both targets in the narrow focus simultaneously would mean comparing them both with each visual stimulus simultaneously. On match trials, this would lead to an ambiguous signal: a match to one template together with a mismatch to the other template. In contrast, comparing each template separately with the input would return an unambiguous “match” signal on match trials. Therefore, again, the limitation of the narrow focus might be functional, not structural.

Is There a Single Zoom-Lens Focus in Working Memory?

Is it really necessary to assume two different foci of attention in working memory? As an alternative, Cowan et al. (2005) proposed that there is a single focus that can either “zoom out” to include up to four chunks or “zoom in” on just a single chunk, depending on task demands. A single focus that can flexibly zoom in or zoom out can behave like the broad focus or the narrow focus, but not at the same time. Therefore, finding effects that support the notion of a broad focus and effects that support the notion of a narrow focus at the same time would make it difficult to account for the whole pattern by assuming only one flexible focus.

Such evidence has come from an experiment of Oberauer (2002). Participants encoded in working memory two lists of digits, each of which varied in set size. A cue then identified one list as “active” and the other as “passive.” About 2 seconds after the cue, arithmetic latencies depended only on the set size of the active list, a result showing that the passive list had been (temporarily) removed from the broad focus. In the same experiment, successive arithmetic operations on the digits of the active list were carried out faster when they required repeated access to the same digit than when they required a switch to another digit in the same list. This object-switch cost is one of the findings supporting the notion of a single-item focus.

Individual Differences

Individual differences in measures of working memory capacity reflect to a large extent differences in the ability to hold representations in the broad focus (Cowan et al., 2005; Fukuda, Vogel, Mayr, & Awh, 2010). The capacity of the broad focus can be quantified by the number of items held in it. Whether this number reflects an underlying constant (i.e., a fixed number of units in memory) across different materials and methods of measurement is still a matter of debate (Fougnie & Marois, 2011). In our view, the main function of the broad focus is to relate multiple representations to each other by binding them into a common cognitive coordinate system. Supporting this idea, individual differences in working memory capacity are related to variation in the ability to maintain temporary bindings between items and their contexts (Oberauer, 2005a; Öztekin & McElree, 2010).

Complementary to the ability to hold relevant information in the broad focus is the ability to keep irrelevant information out of it and to remove no-longer-relevant information from it. Variation in this ability to control the contents of the broad focus might be a second source of individual differences in working memory capacity (Vogel, McCollough, & Machizawa, 2005).

In contrast, variation in parameters of the narrow focus appear to be unrelated to individual differences in working memory capacity. The speed of switching between objects in working memory did not correlate with performance on complex working memory capacity span tasks (Unsworth & Engle, 2008). Likewise, individuals with high complex span and low complex span did not differ in their speed of access to the last item in the Sternberg task, which is assumed to be in the narrow focus (Öztekin & McElree, 2010).

Conclusions and Outlook

Humans can maintain only a limited amount of information in working memory. Selecting a subset of this information is a crucial aspect of cognitive processing, such as solving problems, doing mental arithmetic, and understanding language. There is converging evidence from several experimental paradigms for a limited-capacity component of working memory that can hold a small number of chunks available for direct access. There is also converging evidence that within that small set, a single chunk is often selected for processing and thereby achieves privileged status.

The three-embedded-components framework (Oberauer, 2002, 2009) is an attempt to explain both sets of findings. It distinguishes between the functions of the broad focus and the narrow focus. The broad focus serves to select from among the representations in activated LTM those that require imminent access. The broad focus also serves as a “blackboard” for assembling new structures out of the selected representations. The narrow focus serves to select at any moment the information that is needed for an upcoming cognitive operation. Whereas the broad focus has a limited capacity, most likely because of the interference between multiple content-context bindings (Oberauer & Kliegl, 2006; Oberauer & Lewandowsky, 2008), the narrow focus is limited by its function: When one specific item in working memory is needed—for instance, when one out of two counters needs to be increased—holding more than one item in the narrow focus would not serve cognition well.

There are numerous directions for further research on attention to information in working memory; here, we sketch just one of them. The embedded-components framework was initially applied only to declarative working memory—that is, working memory for objects of thought and action (i.e., symbols, physical objects, events, etc.). The framework was recently extended to procedural working memory—that is, working memory for intended mental or physical actions (Oberauer, 2009). The broad focus of procedural working memory is assumed to hold the task set relevant for the ongoing activity; this task set consists of bindings between stimulus categories and corresponding responses. The narrow focus serves to select a single response to be made on a given trial. The assumption of analogous structures and mechanisms in declarative and procedural working memory motivates the prediction of analogous patterns of effects in both areas. This prediction has already received some support—for instance, we found a pattern of time costs for switching between lists that mirrors the pattern of costs for switching between tasks (Souza, Oberauer, Gade, & Druey, 2012). These initial findings raise the possibility that the cognitive system uses similar solutions for the attentional selection of both declarative and procedural representations in memory.

Recommended Reading

Lepsien, J., & Nobre, A. (2006). Cognitive control of attention in the human brain: Insights from orienting attention to mental representations. Brain Research, 1105, 20–31. A review of behavioral and neuroscience research by the authors on directing the focus of attention in visual working memory.

Oberauer, K. (2009). (See References). A more elaborate outline of the three-embedded-components model of working memory, with its extension to procedural working memory.

Olivers, C. N. L., Peters, J., Houtkamp, R., & Roelfsema, P. R. (2011). (See References). A highly accessible review of the evidence from visual-search experiments for the distinction between a single-item focus and a more encompassing working memory.