Cowan - 2019 - Short-term Memory Based on Activated Long-term Mem

Abstract

Short-term memory (STM) functions as a limited-capacity information store, holding a small amount of data in an accessible state temporarily. This note discusses Norris (2017) theory that suggests STM operates with separate copies of information but argues for an alternative view. It posits that STM relies heavily on activated long-term memory (aLTM), which plays a vital role in how we retain and recall information.

Introduction

Norris's review of STM theories indicates a potential separation between STM and long-term memory, with a particular focus on serial recall tasks. The review highlights significant frameworks in memory research, including Cowan's embedded-processes framework, which offers a different perspective on how memory is structured and function.

Points of Contention with Norris' Argument

Common Assumption Overlooked:

Cowan advocates that aLTM is capable of incorporating new records formed rapidly through STM tasks, suggesting a more integrated relationship between the two types of memory rather than a stark division.

Pathological Cases:

Norris' assertions regarding STM impairment present limited empirical evidence. Many issues can be better accounted for by exploring the encoding and retrieval challenges linked to long-term memory (LTM) instead.

Pointer Structures and Learning:

Although Norris discusses pointers to aLTM being involved in long-term learning, it is postulated that these pointers are more complex and multifaceted than previously considered.

Memory Storage Models:

Cowan elaborates that STM can interact with aLTM within a broader learning context. This view integrates several functions of memory, signaling that STM should not be viewed in isolation.

Overview of the Embedded-Processes Theory

Definition of aLTM:

Activated long-term memory (aLTM) refers to memories that are currently activated and accessible, containing newly learned information relevant to tasks at hand. This concept diverges from the traditional view of having a distinct store dedicated solely to STM.

Focus of Attention (FoA):

The focus of attention can only accommodate a limited amount of information simultaneously. This focus allows for active processing, the binding of information, and the efficient retrieval of relevant memories.

Key Distinctions Between Memory Theories

  1. Separate-STM-Copy ApproachesTwo theories proposed by Norris suggest different mechanisms: one posits that environmental information is stored in a separate STM, while the other relies on pointers that direct attention to LTM.

  2. Unitary Memory ApproachThis approach indicates that a single mechanism governs both STM and LTM, emphasizing interference effects that occur when these two memory types interact.

  3. Embedded-Processes ApproachMemory is conceptualized as a singular construct of LTM, with a temporary subset activated for the items currently being attended to. This model includes characteristics of aLTM performance that enable rapid learning.

Distinctions and Challenges Raised by Norris

New Configurations in STM:

Norris raises questions regarding the capability of aLTM to manage complex spatial and temporal configurations. Cowan counters by indicating that aLTM's ability to absorb and represent new learning allows for handling diverse configurations effectively.

Token Versus Type Representation:

Norris contends that aLTM lacks the ability to represent multiple instances of the same item. However, Cowan argues that the allowance for new learning broadens this capability, permitting diverse representations.

Model Availability and Predictions:

Cowan asserts that various successful models demonstrate aLTM's functionality in STM-related scenarios, challenging Norris’ view of distinct boundaries.

Different Memory Properties:

Norris identifies differences in recall characteristics between STM and LTM; however, the argument neglects to acknowledge the common processes shared across various tasks.

Decay Rates:

Norris claims that separate memory processes exhibit differing decay rates, thereby affecting memory performance. Alternatively, Cowan emphasizes that decay rates fluctuate depending on the quality and context of learning.

Importance of Short Timeframes:

The significance of short durations in STM is emphasized, where Cowan introduces the concept that new learning incorporated during tasks drastically alters retention capabilities.

Neuropathologies:

Norris refers to selective impairment cases; Cowan emphasizes the necessity for interpretations of memory issues to consider the specific context of the tasks involved.

Variable Binding in STM:

Norris highlights the challenges posed by maintaining relational bindings among items. Cowan argues that rapid learning facilitates effective binding of information.

Task Impurities:

The potential evidence regarding concurrent influences on tasks can obfuscate the distinctions between STM and LTM; this overlap complicates the understanding of distinct memory functionalities.

Neuroimaging Correlations:

Norris views neuroimaging results as merely correlational. In contrast, Cowan incorporates concrete evidence illustrating consistent neural patterns suggesting shared mechanisms between STM and LTM.

Conclusion: An Integrated Framework

The embedded-processes view posits a straightforward model of memory without necessitating the independent separation of STM from LTM. It proposes that both short-term and long-term memory are interconnected systems, reliant on common neural resources for their effective function. This integrated framework advances the understanding of memory, emphasizing the collaborative nature of different memory types and the neurological basis behind them.

Acknowledgments

This paper was supported by NIH Grant 4R01-HD-21338.