Research Briefing: The associative structure of memory for multi-element events

Inspired by Chris Chambers and Mark Stokes, this is a research briefing about a recent paper published by myself and Neil Burgess. My hope is that it explains why we did the research, what we found, and why we feel this is important. As with the research briefings from the names above, it is aimed at a scientifically-curious individual without any prior knowledge in the subject. In relation to this, I hope I pitch it correctly – do let me know one way or the other!

Horner, A.J., & Burgess, N. (2013) The associative structure of memory for multi-element events, Journal of Experimental Psychology: General, 142(4), 1370-1383. [abstract & PDF]

When we remember previous events in our lives we bring to mind a wealth of information. We remember the layout of the living-room, the person we were talking to, the music that was playing in the background and the smell that wafted through from the kitchen. When we first experienced this event, each element within this multi-sensory experience was processed in different regions of the brain. For example, the visual information will have been processed in visual cortex in the occipital and temporal lobes and the auditory information in auditory cortex in the temporal lobes. How is it that we are able to remember all these multiple elements despite them being processed in different regions of the brain? 

Theories of episodic memory (memory for events) suggest that all the multiple elements of an event are bound together in an ‘event’ memory (or, more precisely, and event ‘engram’ or event ‘memory trace’) [1]. By binding all this information into a single engram, all the elements from an event can be retrieved and re-experienced at a later time. The hippocampi, located in the temporal lobes, are thought to support this binding as they receive input from multiple regions of the cortex. In this sense, they act as ‘convergence zones’, binding information across our multiple senses [2].

Although most theories of episodic memory presume the existence of bound event engrams, little evidence has been presented in support of this idea [though see 3]. One prediction that stems from this proposal is that the retrieval of elements from the same event should be related. If you retrieve information about the location you were in, you should be more likely to retrieve information about who was there and what they were saying at the time. In other words, retrieval of an event should be ‘all-or-none’ in nature. We tested this prediction, providing support for the existence of bound event engrams.

The experiments consisted of a learning and a testing phase (Figure 1). During learning, participants were presented with a series of ‘events’. Each event consisted of three elements – a location, famous person and object. For example, an event might be: Kitchen, Barack Obama and Hammer. The three elements of each event were presented on a computer screen, as words. Participants were required to construct a mental image of the three elements interacting and imagine it as vividly as possible. For example, they might imagine Barack Obama smashing kitchen cupboards with a hammer. After learning several events we tested their memory for each pairwise association within each three element event. In the above example, this meant testing the association between Kitchen and Barack Obama; Kitchen and Hammer; and Barack Obama and Hammer. 

If all three elements of an event are bound in a single event engram, performance for each pair should be related. If you remember Kitchen and Barack Obama, you should also remember Kitchen and Hammer. Equally, if you don’t remember one pair, you shouldn’t remember the other pairs from the same event. We tested this ‘dependency’ between retrievals from the same event. In order to control for differences in accuracy and the level of guessing across participants, we developed models that predicted the level of dependency if performance of pairs for an event were unrelated (the Independent Model) or if performance of pairs for an event were completely related (the Dependent Model). We then compared the amount of dependency for each participant with each of these models to see if the retrieval of elements from the same event were related. Across three experiments we showed strong evidence for this predicted dependency. The level of dependency was consistently greater than the Independent Model (Figure 2), and in some situations did not differ from the Dependent Model (see paper).

Thus, we provided evidence that the multiple elements of an event are bound within a single ‘event’ engram. Event engrams are thought to be the fundamental building blocks of episodic memory. They allow us to retrieve information about a single event, without retrieving information from other similar events. Despite the common assumption that these event engrams exist, we provide some of the first behavioural evidence for their existence. Our future work is focussing on revealing the exact neural mechanism that allows for this event binding, as well as investigating which regions of the brain support this binding process.

 

1. Tulving, E. (1983). Elements of episodic memory. Oxford: Clarendon Press.
2. Damasio, A. R. (1989). The Brain Binds Entities and Events by Multiregional Activation from Convergence Zones. Neural Computation, 1(1), 123–132.
3. Jones, G. V. (1976). A fragmentation hypothesis of memory: Cued recall of pictures and of sequential position. Journal of Experimental Psychology: General, 105(3), 277–293. 

Changing attitudes to benefits?

This should be a very quick post. There were a couple of articles in the Guardian yesterday on the British Social Attitudes Survery [1][2]. Here are some quotes:

"The percentage of Britons believing that "benefits for unemployed people are too high and discourage work" has seen the biggest fall in thirty years"

"Since last year, and perhaps in stark contradiction to most people's expectations, attitudes towards benefit claimants have become strikingly more sympathetic. The difference in the 2012 findings is the largest single change on this scale since the survey began. Last year 62% felt that benefits were too high and discouraged work. This year that has dropped to 52%."

So let's look at the data.

On the x-axis we have year and the y-axis the percentage of people agree with either the statement "benefits for unemployed people are too high and discourage work" (in green) or "benefits for unemployed people are too low and cause hardship" (in blue).

The 'softening' in attitudes the quotes are referring to is the change from 2011 to 2012 (look at the dip in the green line at the far right of the graph). This is technically correct, but come on, there is signal and there is noise. Compared to the overall trend over the past 25 years it seems like the journalists are clutching at straws. Of course, this dip could be meaningful, but it is impossible to tell from 2 data points. If the trend continues next year then we might be able to start discussing it.

As an example, here are two curves fitted to the data set. For the geeks, the first is a polynomial of degree 2 (quadratic; left panel) and the second of degree 3 (cubic; right panel). The dots are the actual data, the lines the fitted curves.

The only point I want to make is that the lines (fitted curves) go either up or down at the end depending on the particular model. I did this analysis in 5 minutes in MATLAB (the polyfit function for those interested) so it is obviously crude. In short, we don't really know if the change from 2011-2012 will continue in the same direction or even reverse next year.

At this point I should add that both journalists provided caveats to their original statements by saying:

"Although this is the biggest softening in attitudes so far, it doesn't represent the highest level of support for benefits to the unemployed. In fact, looking at all three decades of data reveals a very different picture of how attitudes to welfare have changed."

"I would urge caution. As the Guardian's Data Blog shows this morning, the graph on this question is what statisticians call, in technical jargon, a spiky little bugger." 

My only point is that both pieces seemed to lead with noise and caveat with the signal. The latter should be the focus and the former worthy of note, not the other way round.

Journal Club #1

One of the reasons for starting a blog was that I wanted to occasionally write brief summaries of papers that have influenced me along the way. In particular, I wanted to write about papers that either might not have received a great deal of attention when published, or older papers that I think still have a great deal to offer.  I’m starting with a paper I hold dear, as it was the rock that I built my ramshackle PhD upon.

Logan, G. (1990) Repetition priming and automaticity: common underlying mechanisms?, Cognitive Psychology, 22, 1-35

Repetition priming is simply a change in behaviour to a stimulus you have previously experienced relative to one you have not. For example, if I showed you an object and asked you to classify it according to whether it was man-made you would respond more quickly and accurately on the second compared to the first presentation. So what? Firstly, repetition priming can be a big effect. You can be as much as 200msecs faster on the second presentation of an object, a potential 20% increase in speed [1]. Secondly, amnesics show intact priming, despite being impaired at consciously remembering specific events (such as what they had for breakfast that morning; referred to as ‘episodic’ memory)[2]. Thirdly, priming is long lasting – effects have been shown despite a gap of several years between the first and second presentation of a stimulus [3]. So priming seemingly taps into a mechanism(s) that learns quickly (following a single presentation of a stimulus), lasts a long time (several years) and seems to be distinct from ‘conscious’ or ‘explicit’ forms of memory.

What causes repetition priming? There is plenty of healthy debate about the underlying cause(s) of priming. Perhaps the most common idea though relates to the learning of perceptual information when a stimulus is first encountered [4]. For example, when we see an object for the first time we construct a perceptual representation of that object. On second presentation the existence of the previously learned perceptual representation allows us to recognise the object more quickly, speeding up any subsequent task performed on the object. Although many argue about the specifics of such theories [5-6], most would agree that one locus of repetition priming is perceptual in nature. If this is the case, it shouldn’t necessarily matter what particular task we perform on the object each time we see it; faster recognition = faster reaction times = repetition priming.

Logan proposed a completely different idea [7]. He suggested that when we see an object for the first time and respond to the object, we learn a direct mapping between the stimulus and response (an S-R association). When the object is seen for a second time, we simply retrieve the S-R association learned on the first presentation and quickly make the same response. Put simply, we don’t need to figure out what response to make anymore as we just remember what we did last time. This idea makes a clear prediction that ‘perceptual’ theories of repetition priming do not. It predicts that if you have to respond differently on the second presentation of an object relative to the first presentation, repetition priming should decrease. In other words, you should only see repetition priming when you are able to make the same response on the second presentation. Across a series of 4 elegant experiments Logan clearly showed this to be the case. When participants had to perform a different response between stimulus presentations, or compute a new response following a switch in task, repetition priming significantly decreased. He therefore provided strong evidence for his view of repetition priming. 

Again, why should we care? Well, the two theories of priming outlined above are fundamentally distinct learning mechanisms. One relates to perceptual learning, one to associative stimulus-response learning. A phenomenon as robust and ubiquitous as repetition priming deserves to be fully understood. This debate is particularly important for fMRI studies of “neural priming”, where people use an effect called repetition suppression or fMR-adaptation (a decrease in the fMRI BOLD response to a repeated relative to a novel stimulus). Many studies have presumed “neural priming” relates to perceptual learning within visual regions. If Logan is correct, however, the effect they are measuring may actually relate to the learning/retrieval of S-R associations and therefore may not be telling us anything about perceptual learning per se [8]. Without being aware of this experimental psychology research, it would be easy to confound perceptual learning with S-R learning by not including a response manipulation within your experimental design. This is, therefore, a classic example of how cognitive psychology and clever experimental design can help inform more ‘neuroscientific’ studies of the same underlying phenomena.

Finally, just a quick note to say Logan wasn’t completely correct. It seems, as is often the case, that repetition priming results from multiple learning mechanisms. One does relate to the learning of perceptual (and conceptual) representations and another relates to the learning of S-R associations [9]. As a result of Logan’s original work we now have much clearer understanding of the multiple mechanisms underlying repetition priming, as well as having much tighter experimental control over these different contributions.

References:

1. Horner, A. J., & Henson, R. N. (2009). Bindings between stimuli and multiple response codes dominate long-lag repetition priming in speeded classification tasks. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35(3), 757–779.
2. Warrington, E. K., & Weiskrantz, L. (1968). New Method of Testing Long-term Retention with Special Reference to Amnesic Patients. Nature, 217, 972–972.
3. Mitchell, D. B. (2006). Nonconscious priming after 17 years: invulnerable implicit memory? Psychological Science, 17(11), 925–929. 
4. Schacter, D. L. (1990). Perceptual representation systems and implicit memory. Annals of the New York Academy of Sciences, 608, 543–571. 
5. Tenpenny, P. L. (1995). Abstractionist versus episodic theories of repetition priming and word identification. Psychonomic Bulletin & Review, 2, 339–363.
6. Bowers, J. S. (2000). In defense of abstractionist theories of repetition priming and word identification. Psychonomic Bulletin & Review, 7(1), 83–99. 
7. Logan, G. D. (1990). Repetition priming and automaticity: common underlying mechanisms? Cognitive Psychology, 22, 1–35.
8. Dobbins, I. G., Schnyer, D. M., Verfaellie, M., & Schacter, D. L. (2004). Cortical activity reductions during repetition priming can result from rapid response learning. Nature, 428(6980), 316–319.
9. Horner, A. J., & Henson, R. N. (2008). Priming, response learning and repetition suppression. Neuropsychologia, 46(7), 1979–1991.