Even an activated long-term memory system still needs a separate short-term store: A reply to Cowan (2019).

Dennis Norris

doi:10.1037/bul0000204

Abstract

In Norris (2017), I explained why the notion of activated LTM (long-term memory) combined with a focus of attention was unable to perform the computations required to support short-term memory (STM) and argued that those extra computations must require a separate STM system. Cowan (2019) made the alternative proposal that this full set of computations is better conceptualized as a unitary system of activated LTM. To this he added a pointer system, the ability to perform variable binding, and an unspecified model of STM that acts as a front end to LTM. This appears to be simply an exercise in relabeling. Furthermore, without a computational specification of how the components work, the model lacks the ability to simulate even the most basic STM phenomena. If the model were specified in more detail it seems almost inevitable that it would contain something instantly recognizable as an STM system.

Highlights

Norris (2017) highlighted the inadequacy of models that embody the claim that short-term memory (STM) can be supported entirely by aLTM
Even if the conclusion were that the hippocampus, and only the hippocampus, plays a role in binding, any further conclusion about the role of aLTM depends on the additional assumption that the hippocampus is exclusively involved in LTM and could not be construed as implementing any part of a separate STM process
When Cowan suggests that a model of STM might form the front end of LTM learning it is not clear whether he has a particular model of STM in mind

Summary

Storage of new configurations is needed in STM

Everyone recognizes that there must be new, rapid learning of information in STM tasks (e.g., Keppel & Underwood, 1962), and the newly learned information is typically still in an activated state, aLTM, at the time of test (Cowan, 1999). 2. Token representations cannot aLTM includes rapid learning of information, be represented in aLTM, and can include the same episodic only types information about tokens that one adds to LTM (Cowan, 1999; Nairne & Neath, 2001). 3. No extant model of STM Including new learning as part of aLTM changes performance based on aLTM the need because separate STM copy theories might be reclassified as the front end of longterm learning. A few models deal explicitly with aspects of aLTM and new learning

STM recall differs from

11. Neuroimaging as a correlation fallacy

A Model of STM as the Front end of LTM

Conclusion