Functional similarities and differences between the coding of positional information in verbal and spatial short-term order memory

ABSTRACT

Temporal grouping effects in verbal and spatial serial recall suggest that the representation of serial order in verbal and spatial short-term memory (STM) incorporates positional information. However, not all effects of grouping are created equal in the verbal and spatial domains. Although grouping a sequence of verbal items engenders an increase in between-group transpositions that maintain their within-group position, grouping a sequence of spatial items does not engender an increase in these so-called interposition errors. Here I present experimental and computational modeling evidence which suggests that positional information is represented in subtly different ways in verbal and spatial STM. Specifically, the findings indicate that in verbal STM, groups are coded for their position in a sequence and items are coded for their position in a group. By contrast, in spatial STM groups are coded for their position in a sequence, but items are coded for their position in a sequence, rather than in a group. Findings support the notion that positional information in verbal and spatial STM is represented by modality-specific mechanisms rather than a domain-general system.

KEYWORDS:

• Positional
• serial order
• short-term memory
• spatial
• verbal

Acknowledgments

This paper is based on part of the author's doctoral dissertation completed at the University of York, England. The author is now based at the University of Western Australia.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. I do not simulate the dynamics of recall since like most other models of serial recall, the SEM does not incorporate a mechanism for generating response time predictions. Although such a mechanism can easily be introduced by augmenting the recall process in the SEM with a set of competitive decision accumulators (e.g. Usher & McClelland, 2001), additional assumptions about the hierarchical nature of retrieval of positional information would also need to be incorporated to account for the long recall times at the beginning of the sequence and at group boundaries (Farrell, 2012). In the interest of parsimony, I chose not to model these phenomenon, since the latency data are only reported here to verify that grouping exerted its usual effects on the dynamics of recall.

2. It transpires that this is a consequence of the low parameter settings of the start and end group markers ($S_G=0.05$; $E_G=0.20$). Lower values of these parameters render the group markers near the start and end of the sequence highly distinctive, meaning that between-group transpositions are highly unlikely to occur over the first and last few serial positions (e.g. when cueing items at positions one, two, and three with the group marker for the first group, items at these positions will have much higher activations than items in the second and third groups). However, the group marker for the middle group will be less distinctive due to its greater distance from the start and end of the sequence, meaning that between-group transpositions are more probable at medial serial positions (e.g. when cueing items at positions four, five, and six with the group marker for the second group, items at these positions will have only slightly higher activations than items in the first and third groups). Because of this, when the item at position six is cued with its group and item markers it will be particularly vulnerable to confusion with the item at position seven from the third group (not only because of the poor resolution of the group marker for the second group but because of high positional overlap between the item markers for positions six and seven). Suppose that item seven is prematurely recalled at position six and then suppressed (suppose also that the first five items in the sequence have been recalled and suppressed and, like item seven, are therefore unlikely to be recalled again). Bearing in mind the high resolution of group markers near the beginning and end of the sequence noted previously, when the group and item markers are used to cue recall at position seven, item six will only be a weak recall competitor as it emanates from a different group to the one being cued, therefore the items from positions eight and nine will be the strongest competitors, with item eight being the item that will most likely be recalled given the suppression of item seven after it was prematurely recalled at position six, combined with the higher degree of overlap between the item markers for positions seven and eight than for positions seven and nine. When the group and item markers are used to cue recall at position eight, item six will still remain a weak recall competitor and the most likely item that will be recalled is item nine given the suppression of item eight after it was prematurely recalled at position seven. Finally, when the group and item markers are used to cue recall at position nine, the only item left unrecalled and unsuppressed will be item six, which when recalled will force an interposition error.

3. I also examined the prevalence of interpositions in grouped sequences by comparing the frequency of three-apart interpositions with the frequency of two-apart transpositions. This analysis revealed that the proportion of interpositions was greater for 94% of the parameter setting combinations simulated for the GPS + IPG model, compared to only 2% for the GPS + IPS model. This analysis confirms that interpositions are a representative feature of the GPS + IPG model, but not the GPS + IPS model, which only very rarely generates these errors.

4. During the review of the current paper one reviewer noted that there is some circularity to this account. The syllabic structure of normal language gives rise to constraints on how segments might be mis-ordered, which means that interposition errors are more likely. However, at the same time, some basic ordering mechanism is proposed to give rise to the structure of language. Accordingly, we are still left with the question of why language should become ordered in one way, and other types of information are ordered in another way. Put another way, the syllabic structure of language is presented as the cause of the interposition errors in verbal serial recall, thereby implicating a group-position-in-sequence + item-position-in-group ordering process, but that process is also argued to give rise to the syllabic structure of language. I acknowledge this criticism but leave it to the reader to evaluate the credits and debits of my explanation, which is the best I have been able to put forward to explain the error pattern discrepancy between the verbal and spatial domains.

Additional information

Funding

This work was supported by a research studentship from the Economic and Social Research Council of the United Kingdom.