Retention-error patterns in complex alphanumeric serial-recall tasks

Abstract

We propose a new method based on an algorithm usually dedicated to DNA sequence alignment in order to both reliably score short-term memory performance on immediate serial-recall tasks and analyse retention-error patterns. There can be considerable confusion on how performance on immediate serial list recall tasks is scored, especially when the to-be-remembered items are sampled with replacement. We discuss the utility of sequence-alignment algorithms to compare the stimuli to the participants' responses. The idea is that deletion, substitution, translocation, and insertion errors, which are typical in DNA, are also typical putative errors in short-term memory (respectively omission, confusion, permutation, and intrusion errors). We analyse four data sets in which alphanumeric lists included a few (or many) repetitions. After examining the method on two simple data sets, we show that sequence alignment offers 1) a compelling method for measuring capacity in terms of chunks when many regularities are introduced in the material (third data set) and 2) a reliable estimator of individual differences in short-term memory capacity. This study illustrates the difficulty of arriving at a good measure of short-term memory performance, and also attempts to characterise the primary factors underpinning remembering and forgetting.

Keywords:

• Memory
• Span
• Scoring
• Errors
• Short-term memory
• Working memory
• Chunks
• Chunking

View correction statement:

Erratum

This research was supported in part by the Agence Nationale de la Recherche Grant # ANR-09-JCJC-0131-01 and in part by the Centre National de la Recherche Scientifique (CNRS) and the Institut National des Sciences Humaines et Sociales (INSHS) research grant (PEPS) to Fabien Mathy in 2009. The authors wish to express thanks to Nicolas Heller for providing valuable assistance with analyzing data.

This research was supported in part by the Agence Nationale de la Recherche Grant # ANR-09-JCJC-0131-01 and in part by the Centre National de la Recherche Scientifique (CNRS) and the Institut National des Sciences Humaines et Sociales (INSHS) research grant (PEPS) to Fabien Mathy in 2009. The authors wish to express thanks to Nicolas Heller for providing valuable assistance with analyzing data.

Notes

¹ One of our participants did not have sufficient time to finish the experiment. Because there was no indication of the number of lists completed on the screen, the experimenter could not know that the participant was one trial short of finishing the experiment.

² We think that a global alignment can fit serial position functions in which good performance for early and late items are observed (local alignment is more complex and can over-prioritize mid-list items).

³ In general, the algorithms leave a blank for simple substitutions and produce a substitution (:) symbol for conservative substitutions of amino acids whose side chains have similar biochemical properties, but this distinction was not necessary to explore the fundamental capabilities of the algorithms. We simply considered every mismatch as a substitution.

⁴ These extrapolations are only given to indicate where performance is supposed to be asymptotic, given the fit, but there is a possibility that lengthier lists would in reality worsen performance and make the task resemble a running-memory span task.

⁵ The other peculiar feature is that even though the response hardly matched the stimulus at all, the number of items recalled was correct.

⁶ Ranschburg effects in our experiments were naturally favoured since items were sometimes repeated, but our goal was not to detail this effect.