by Michael Diamond, Courtney O’Brien, Hyungwook Yim and Simon Dennis
Experience sampling methodologies are set to revolutionise psychological science, as these techniques overcome some of the major limitations of laboratory experiments. This blog post explains how we utilised experience sampling techniques in a memory experiment investigating the Word Frequency Effect in recognition memory.
The Word Frequency Effect refers to the phenomenon that low-frequency (uncommon) words are better discriminated than high-frequency words (words that are commonly used in language), with low-frequency words having higher hit rates and lower false alarm rates (Glanzer et al., 1993; Glanzer & Adams, 1990). The word frequency effect has been of theoretical interest since (1) even though low-frequency words are less well represented in memory they are recognized better, and (2) that the pattern is the opposite in a recall task, where high-frequency words are better recalled than low-frequency words.
There are two views of the mechanism of the Word Frequency Effect in recognition memory: The Retrieving Effectively from Memory Model (Shiffrin & Steyvers, 1997) which is an Item Noise Account, and the Bind-Cue-Decide Model of Episodic Memory, which is a Context Noise Account. (Dennis & Humphreys, 2001). The two models agree that the Word Frequency Effect is due to interference in memory, however, they differ in their explanations for why this interference occurs. The item noise account argues that low frequency words are more distinguishable from other words than high frequency words and therefore are not as affected by item interference, while for the context noise account the contexts that the item is experienced in cause interference. As low frequency words are seen in fewer contexts they experience less interference. There have been studies that attempt to untangle the sources of interference by controlling for item distinctiveness and manipulating context variability in different ways (Chalmers et al., 1997; Chalmers & Humphreys, 2003; Malmberg et al., 2002; Reder et al., 2002). However, it is not trivial to completely test this idea in laboratory, because one cannot measure the experience that the participant has had with a word before entering the experiment.
One way to resolve the issue is considering recency. Distinguishing the study context from the context in which one most recently saw the word will be more difficult if the word has been seen recently. A couple of recent studies have provided evidence that recency matters by pre-exposing the to-be-recognized stimuli prior to the main recognition memory task (Dye et al., 2017). However, pre-exposing the materials in these studies involves an unnatural way of manipulating recency and possibly introduces confounds. Moreover, the time scale that these studies examine (the time between pre-exposure to testing) is relatively short (e.g., 20mins).
In a recent study, we overcame some of these challenges using experience sampling methods. Using the www.unforgettable.me service, we collected participant’s (N=65) daily email for two months. This experimental paradigm meant that we had access to the words that each individual experienced and this allowed us to create a personal corpus for each participant. We knew when each email was received, so recency of the words in the personal corpus was able to be measured, rather than being manipulated as in previous studies. We could then generate individualized word frequencies rather than relying on normative frequency measures. Since individualized frequency provide a more customized window – looking into one participant’s experience – it follows that individualized frequencies should provide a better account for understanding the Word Frequency Effect compared to the normative frequencies, which rely on the assumption that the normative frequency is a good approximation to the frequency experienced by each participant. Participants went through a study-test recognition memory test using their personalised corpus of the words that occur in their emails.
The context noise account of the word frequency effect predicts that more recently experienced words will cause more interference than words that were experienced earlier. Notably, we found that recency has a graded effect on recognition memory that extends for at least two months (see Figure 1a).
The results support the context noise account of the Word Frequency Effect, and challenges the item noise account.
It is also interesting to see that the individualized frequency captures the Word Frequency Effect well. On one hand, it is an obvious result since the individualized frequency, which is calculated from an individualized corpus, is personalized and will handle the individual variability better. However, it is also worth noting that the individual corpus is based on (1) only two-month worth of emails, and (2) only from received emails. Considering how many words people experience in their daily life from diverse sources (e.g., TV, messenger, books, etc.) and also compared to the size of the normative corpus that has been used (50,000 words vs. 51 million words), the information that the individual frequency is providing is striking. Therefore, it is highly probable that having a longer length of data collecting period with a more diverse set of data streams to generate each personalised corpus may provide more interesting information about one’s pre-experimental word experience.
Finally, this study provides an interesting way to use experience sampling methods in memory research. Memory studies have preferred controlled laboratory studies as they generally have less noise in the data and tighter manipulations for examining the effect of concern. However, our recent study showed that some phenomena which are greatly debated in memory literature, can not be resolved in the laboratory, and experience sampling methods provide a promising way of conducting these studies. We therefore propose a more ecologically valid way of examining the effect of recency in recognition memory, once again highlighting the power of using experience sampling methods to conduct psychological research ‘in the wild’.
The paper on this work can be found here.
Simon Dennis 