The Brain’s Seahorse, the Hippocampus, Helps us Remember Exactly what we Read

Previous research has shown that the hippocampus – a brain structure only around 3–5 centimetres long, named after its seahorse-like shape – can encode highly similar images in ways that allow them to remain distinguishable from one another. Researchers call this specific encoding mechanism pattern separation. In a recent study, researchers from the Brain Imaging Centre at the HUN-REN Research Centre for Natural Sciences and the Wigner Data Centre at the HUN-REN Wigner Research Centre for Physics have shown that hippocampal pattern separation also operates on the basis of meaning: it enables our memory to distinguish between texts with very similar semantic content.

How do episodic and semantic memory work?

Two broad types of memory system are generally distinguished. Our knowledge about the world – for example, knowing that Paris is the capital of France – is referred to as semantic memory. By contrast, remembering as a personal experience who we were with, when we visited Paris and what the weather was like is an example of episodic memory.

The role of the hippocampus in episodic memory has been recognised since the 1950s, following the first bilateral removal of the hippocampus performed to alleviate severe epileptic symptoms. Henry Molaison, who later became one of the most famous cases in the history of memory research – along with several other people who underwent similar surgery – was able to communicate and think normally, his attention was intact, and he had no major motor or perceptual difficulties. However, he was unable to form vivid memories of unique events that had happened only moments earlier. He could learn new skills but could not remember who had taught him those skills, when he had learnt them or how the learning had taken place.

Until relatively recently, by contrast, the processing of meaning in memory was described by cognitive scientists as a separate process, primarily associated with language and semantic memory.

“Our study is novel in identifying a neural operation shared by the episodic and semantic memory systems and implemented by a tiny region within the hippocampus,” emphasised Dr Attila Keresztes, Senior Research Fellow at HUN-REN TTK and Assistant Professor at the Institute of Psychology at Eötvös Loránd University (ELTE).

How the researchers examined the encoding of similar word combinations

The experiment involved 30 healthy young adults, who first read adjective–noun phrases such as “exotic zoo”. At this stage, the researchers simply asked them to decide whether they would include each phrase in an imaginary newspaper they were editing.

The participants were then given an unexpected memory task. They again saw adjective–noun phrases – sometimes exactly the same ones they had seen before and sometimes very similar alternatives, such as “strange zoo” instead of “exotic zoo” – and had to decide whether they had previously seen each expression in exactly that form.

Throughout the experiment, the participants’ brains were scanned using high-resolution functional magnetic resonance imaging (fMRI) to determine whether the hippocampus encoded highly similar expressions differently.

The researchers also distinguished between individual hippocampal subfields using high-resolution anatomical imaging, which reveals characteristic tissue features along the hippocampal formation. The activity identified was most prominent in the dentate gyrus, consistent with its proposed role in pattern separation.

Another important feature of the experiment was that the researchers selected the word combinations so that half were semantically more similar and half less similar to one another. “We were able to quantify this using a Hungarian distributional language model, in which the similarity between words is represented by a numerical value calculated from how often the model encounters the words together or in similar textual contexts,” said Alex Ilyés, Research Assistant at HUN-REN TTK and Assistant Lecturer at ELTE’s Institute of Psychology.

Stronger pattern separation was associated with more accurate memory

The results showed that participants whose hippocampi displayed stronger fMRI signatures consistent with pattern separation were better able to encode highly similar word combinations as distinct representations. As a result, they were subsequently better at distinguishing previously seen expressions from new but highly similar ones in the memory test.

The finding is even more striking given that, during encoding, the participants did not know that their memory would later be tested.

Although in everyday terms we tend to think of remembering as something that happens when we retrieve a past experience, the accuracy of a memory is already shaped by the way information is encoded. The work of Alex Ilyés and colleagues contributes to a better understanding of the pattern separation processes that operate during encoding and help preserve the specificity of memories.

Their findings suggest that the hippocampus contributes to the formation of a wide range of highly specific memories. It continues to separate overlapping representations even when we are simply reading and the brain appears to be engaged in “nothing more” than language processing. In other words, if we later wanted to recount the contents of this article word for word, our accuracy would depend in part on how reliably our hippocampus was working while we were reading it.