(Online study) Characterising the learning function in recognition memory

Background

While it is well recognised that performance improves with increased study time in episodic memory tasks, it is unclear what is the nature of this learning process. Specifically, how does learning occurs and what is the functional form of this learning process.

Research Questions / Hypotheses

The aim of this study is to explore the functional form (i.e. exponential or power) and the mechanism (i.e. incremental, all-or-none, or hybrid) of learning in relation to study time in episodic recognition memory.

Participants

62 participants (in semester 1, Experiment 1) and 56 participants (in semester 2, Experiment 2) completed this study. Two participants from semester 1 were excluded from data analysis because of near-to-chance performance.

Methods

Two experiments were conducted whereby Experiment 2 was conducted to replicate results from Experiment 1. We employed a single-word recognition memory task and a within-subject design in both experiments with 11 presentation time conditions (ranging from 50 ms to 4000 ms). Experiment 2 only deviated from Experiment 1 in the incorporation of a 50 ms visual mask after the presentation of each study item. Different learning mechanisms and functions were implemented with the Linear Ballistic Accumulator to jointly model recognition accuracy and latency at individual-subject level. A 3 x 2 factorial model comparison was conducted, varying learning mechanism (incremental account, all-or-none account, or hybrid account) and function (exponential or power).

Results

Results from two experiments showed consistent support for the all-or-none mechanism and power function.

Implications

The results suggest that learning in recognition memory occurs as a power increase in the encoding probability of items. However, given the best model still misfit the data and the close predictions between exponential and power functions, converging evidence for the all-or-none model and more stringent model selection methods are required in future investigations in order to draw conclusive inference about all-or-none learning and better distinguish between different functions.