Simon Farrell

An endogenous distributed model of ordering in serial recall

We introduce a distributed model of memory for serial order, called SOB, that produces ordered serial recall by relying on encoding and retrieval processes that are endogenous to the model. SOB explains the basic shape of the serial position curve, the pattern of errors during recall (including the balance between transpositions, omissions, intrusions, and erroneous repetitions), the effects of list length on the distribution of errors, the overall level of recall and response latency, and the effects of natural language frequency on recall performance. In addition, contrary to several recent suggestions, SOB demonstrates that distributed representations can support unambiguous recall, selective response suppression, and novelty-sensitive encoding.

What limits working memory capacity

We review the evidence for the 3 principal theoretical contenders that vie to explain why and how working memory (WM) capacity is limited. We examine the possibility that capacity limitations arise from temporal decay; we examine whether they might reflect a limitation in cognitive resources; and we ask whether capacity might be limited because of mutual interference of representations in WM. We evaluate each hypothesis against a common set of findings reflecting the capacity limit: The set-size effect and its modulation by domain-specificity and heterogeneity of the memory set; the effects of unfilled retention intervals and of distractor processing in the retention interval; and the pattern of correlates of WM tests. We conclude that-at least for verbal memoranda-a decay explanation is untenable. A resource-based view remains tenable but has difficulty accommodating several findings. The interference approach has its own set of difficulties but accounts best for the set of findings, and therefore, appears to present the most promising approach for future development. (PsycINFO Database Record

A connectionist model of complacency and adaptive recovery under automation.

Automation is intended to reduce the demands on operators in complex environments, thereby enhancing overall system performance. Although automation usually reduces workload, it is often accompanied by a decline in monitoring performance, an effect known as complacency. The circumstances under which complacency occurs and how it can be prevented, for example by intermittently returning control to the operator, are empirically well understood. To date, that empirical knowledge has not been accompanied by strong psychological theory. This article presents a computational model of human performance under automation based on connectionist principles. The model is shown to explain several benchmark findings, among them the basic complacency effect; the effect of the variability of automation reliability on complacency; the effect of task complexity; and the effect of intermittently returning control to the operator.

Interference between maintenance and processing in working memory: the effect of item-distractor similarity in complex span.

Four experiments examined the effect of phonological similarity between items and distractors on complex span performance. Item-distractor similarity benefited serial recall when distractors followed the items they were similar to, but not when distractors preceded the items they were similar to. These findings are predicted by C-SOB (contextual serial order in a box), a computational model of complex span. The model assumes that distractors are involuntarily encoded into memory, being associated to the preceding items list position. Distractors interfere with items by superposition of distributed representations that are associated to the same position. Superposition distorts item memory; this distortion is less severe when the distractor is similar to the item. Further support for the assumption that distractors are encoded at the position of the preceding item comes from the finding that intrusions of distractors at recall tended to come from the position of the target item. In addition, intruding distractors tend to replace items to which they are similar, showing that lack of distinctiveness also contributes to interference.

Multiple Roles for Time in Short-Term Memory: Evidence from Serial Recall of Order and Timing.

Three experiments are reported that examine the relationship between short-term memory for time and order information, and the more specific claim that order memory is driven by a timing signal. Participants were presented with digits spaced irregularly in time and postcued (Experiments 1 and 2) or precued (Experiment 3) to recall the order or timing of the digits. The primary results of interest were as follows: (a) Instructing participants to group lists had similar effects on serial and timing recall in inducing a pause in recall between suggested groups; (b) the timing of recall was predicted by the timing of the input lists in both serial recall and timing recall; and (c) when the recall task was precued, there was a tendency for temporally isolated items to be more accurately recalled than temporally crowded items. The results place constraints on models of serial recall that assume a timing signal generates positional representations and suggest an additional role for information about individual durations in short-term memory.

Dissociating conditional recency in immediate and delayed free recall: a challenge for unitary models of recency.

Temporal distinctiveness models of recency in free recall predict that increasing the delay between the end of sequence and attempting recall of items from that sequence will reduce recency. An empirical dissociation is reported here that violates this prediction when the delay is introduced by the act of recall itself. Analysis of data from a number of previously published free recall studies shows that when the assumed availability of final list items is taken into account, recency increases across the first few output positions in immediate recall despite the delay introduced by recalling items; no such change, with a trend to decreasing recency, is observed in delayed recall. Simulations are presented, showing that 2 models accounting for recency in free recall, the temporal context model (M. W. Howard & M. J. Kahana, 2002) and the SIMPLE model (G. D. A. Brown, I. Neath, & N. Chater, 2007), are unable to account for this novel pattern of data. Further simulations show that the results are consistent with a short-term buffer contributing to recency in immediate free recall and that ordered probing of items may also contribute to this effect; both of these are consistent with the formulation of a short-term buffer akin to models of serial recall.

The Dynamics of Access to Groups in Working Memory

The finding that participants leave a pause between groups when attempting serial recall of temporally grouped lists has been taken to indicate access to a hierarchical representation of the list in working memory. An alternative explanation is that the dynamics of serial recall solely reflect output (rather than memorial) processes, with the temporal pattern at input merely suggesting a basis for the pattern of output buffering. Three experiments are presented here that disentangle input structure from output buffering in serial recall. In Experiment 1, participants were asked to recall a subset of visually presented digits from a temporally grouped list in their original order, where either within-group position or group position was kept constant. In Experiment 2, participants performed more standard serial recall of spoken digits, and input and output position were dissociated by asking participants to initiate recall from a post-cued position in the list. In Experiment 3, participants were asked to serially recall temporally grouped lists of visually presented digits where the grouping structure was unpredictable, under either articulatory suppression or silent conditions. The 3 experiments point to a tight linkage between implied memorial structures (i.e., the pattern of grouping at encoding) and the output structure implied by retrieval times and call into question a purely motoric account of the dynamics of recall.

Working memory for cross-domain sequences

How is information from different content domains bound together into a representation of the whole sequence? Several theories predict that mixing information from different domains specifically impairs the ordering of information from different domains, whereas ordering within domains might be enhanced. In contrast, domain-general models—in which items from different domains are simply assumed to be less confusable—predict that mixing items from different domains enhances ordering, as the list items will on average be less confusable. The results of an experiment showed an overall advantage for mixed over pure lists in ordering information, supporting the domain-general viewpoint. Simulations with a representative domain-general model—the start–end model of Henson [(1998). Short-term memory for serial order: The start-end model. Cognitive Psychology, 36, 73–137] —showed that the model gave a satisfactory account of the data. Together, the data and simulations lend evidence to the idea that a domain-general mechanism is responsible for ordering stimuli from different domains, and that domain-specific effects are attributable to the relative similarity of item representations.

Is scanning in probed order recall articulatory

We consider how theories of serial recall might apply to other short-term memory tasks involving recall of order. In particular, we consider the possibility that when participants are cued to recall an item at an arbitrary position in a sequence, they covertly serially recall the list up to the cued position. One question is whether such “scanning” is articulatory in nature. Two experiments are presented in which the syllabic length of words preceding and following target positions were manipulated, to test the prediction of an articulatory-based mechanism that time to recall an item at a particular position will depend on the number of preceding long words. Although latency was dependent on target position, no word length effects on latency were observed. Additionally, the effects of word length on accuracy replicate recent demonstrations in serial recall that recall accuracy is dependent on the word length of all list items, not just that of target items, in line with distinctiveness assumptions. It is concluded that if scanning does occur, it is not carried out by covert or overt articulation.

Benchmarks for Models of Short Term and Working Memory

Any mature field of research in psychology-such as short-term/working memory-is characterized by a wealth of empirical findings. It is currently unrealistic to expect a theory to explain them all; theorists must satisfice with explaining a subset of findings. The aim of the present article is to make the choice of that subset less arbitrary and idiosyncratic than is current practice. We propose criteria for identifying benchmark findings that every theory in a field should be able to explain: Benchmarks should be reproducible, generalize across materials and methodological variations, and be theoretically informative. We propose a set of benchmarks for theories and computational models of short-term and working memory. The benchmarks are described in as theory-neutral a way as possible, so that they can serve as empirical common ground for competing theoretical approaches. Benchmarks are rated on three levels according to their priority for explanation. Selection and ratings of the benchmarks is based on consensus among the authors, who jointly represent a broad range of theoretical perspectives on working memory, and they are supported by a survey among other experts on working memory. The article is accompanied by a web page providing an open forum for discussion and for submitting proposals for new benchmarks; and a repository for reference data sets for each benchmark. (PsycINFO Database Record