[Online] Effect of Strength on Response Latency in Free Recall

Background

In memory testing, it is commonly agreed that decision-making is involved in recognition tasks as people have to decide on whether a tested item is encountered previously in study phase. However, it is less known that decision-making process is also involved in free recall tasks where instruction is to recall as many items as possible in any order from a sequence of studied items. Osth and Farrell (2019) successfully used the the Linear Ballistic Accumulator (LBA) model to jointly model serial position function and complete RT distribution in free recall tasks, yet they made no attempt to model inhibition between items of different strengths, despite theoretical evidence showing slowdown in RT due to increased interference between items.    The LBA model assumes that each list item receives its own accumulator and evidence races to a retrieval threshold from a starting point that is drawn from a uniform distribution. Evidence accumulates deterministically but each drift rate is sampled from a distribution, such that accumulators with low mean drift rates can by chance sample a high drift rate and win the competition.

Research Questions / Hypotheses

We hypothesised that item inhibition is needed to be modelled within the Linear Ballistic Accumulator (LBA) model in order to model list strength effect (LSE) -- strongly learned items impair memory of weakly learned items when tested with recall tasks. This leaded us to further hypothesised a type of model in which multiple presentations of items creates a stronger accumulator and individual strengths of accumulators are normalised based on the summed strength of all accumulators -- stronger drift rate with normalisation model.

Participants

51 participants participated and 50 of them completed both sessions.

Methods

We presented participants with three types of 6-item study lists: 1) pure-weak lists with each item presented once; 2) pure-strong lists with each item randomly repeated three times; and 3) mixed lists with half of the items presented once and another half randomly presented three times. A 12-second delay containing distractor tasks was inserted before participants were prompted to freely recall. Participants were then given 16s to recall the lists they just studied in any order they like.

Results

It was found that items that were presented more than once (strong items) were recalled more frequently than items presented once (weak items). A weak positive LSE was also found yet a null LSE was also found in mixed1 condition where strong items were presented before weak items. Large primacy effect was shown in probability of first recall curves and a second primacy effect was also shown due to our block design (6-item lists presented in blocks of 3). The median RT for strong and weak items in each condition appeared to be fairly similar, with RT of pure-strong condition being slightly faster. Model fitting showed that stronger drift rate without normalisation model fitted better to our data.

Implications

Our data could be strongly affected by the primacy effect which was not successfully eliminated by our distractor tasks. Hence, a second experiment would be conducted for longer study lists (10-item lists) to further attenuate primacy effect. To this point, no conclusion could be made about whether model with or without normalisation would better capture free recall sequences of human memory.    Planned communication of results would be in the form of honours thesis and mini conference presentation.