Event segmentation: how we understand and remember a story

What is an event while reading?

Think of an invented scene: a woman walks into an archive, looks for a letter and asks the caretaker for help. The actions belong to a situation you can follow. If the story moves to the next morning, in another city, with a new character, you have to update your bearings.

A narrative event does not necessarily coincide with a page or a chapter. What counts in recognising the transitions between events are the changes in the situation described. Studies on the comprehension of texts and films examine precisely how these changes relate to segmentation. Zacks, Speer & Reynolds (2009).

This organisation also works on several levels: in the example, “the visit to the archive” contains smaller actions, such as searching a shelf or talking to the caretaker.

What changes at the boundary between two events?

According to Event Segmentation Theory, we maintain representations that help us interpret what is happening and form expectations. When activity becomes less predictable, the increase in prediction error can contribute to updating the model. This prediction concerns what comes immediately next in the activity, not only the final outcome of the plot. Zacks et al. (2007).

In 2007 Speer, Zacks and Reynolds studied people reading short stories during an fMRI recording. At the points later identified as event boundaries, transient increases in activity emerged in several brain regions, associated with changes in the situation being narrated. The result supports the idea that segmenting events is a spontaneous component of reading. Speer, Zacks & Reynolds (2007).

Boundaries are not simply points where we lose information, though. A later study linked event boundaries to hippocampal activity and to the reactivation of representations during recall. Breaking experience up also helps to organise it in memory. Baldassano et al. (2017).

Why recap accuracy matters for the reader's memory

A summary is not necessarily neutral with respect to what we remember. If it contains incorrect information, it can influence later memory of the original experience.

In 2026 Mattea Sim, Yael Eiger and Tadayoshi Kohno studied this effect using an audiovisual event. Participants watched a video of a car accident and, 24–48 hours later, read an AI-generated summary that contained either accurate information or a deliberately misleading detail. Their memory of the event was then tested.

For one critical detail of the video, 83.6% of participants who had read the accurate summary remembered the information correctly, compared with 44.8% of those who had received the misleading summary.

The result suggests that an error in a summary can do something other than simply provide wrong information: it can interfere with what a person later believes they remember about the original experience.

An important limit

The study concerns memory for an audiovisual event, not the reading of novels. It therefore does not show that an incorrect summary of a book produces the same effect, nor does it measure the effectiveness of Fabulè.

It does, however, offer a relevant indication for designing tools that reconstruct past experiences: when a system returns a recap, accuracy, the provenance of information and the distinction between facts and inferences are important, because the summary can itself become a new source of information for the user.

For a narrative assistant this introduces two distinct requirements:

• not anticipating what the reader has not yet read, respecting the anti-spoiler boundary;
• not altering what they have already read, avoiding presenting as facts events, relationships or motives not supported by the available text.

For Fabulè, this evidence does not demonstrate a benefit of the product: rather, it reinforces a design question. A recap designed for Narrative Re-entry should try to recover the necessary context without adding unverified details and without turning a plausible interpretation into a fact of the story.

Fabulè is an AI assistant for readers, designed to help you recover characters, relationships and events you have already come across. The starting point is the position you have reached in the book and the question that is blocking your understanding.

Fabulè builds the AI's narrative context exclusively from the text already read up to the point you indicate. The later pages are not supplied to the model for that answer. It is the operating principle by which the product keeps its support within the path the reader has already covered.

The connection with the research guides the choice to focus on context. It is not experimental proof that the assistant improves memory or restores a specific mental state.

Fabulè uses the expression Narrative Re-entry to describe this recovery of context in practical terms. To explore the problem and the product's approach further, read Picking up a book you left unfinished.

To put it into practice while you read, follow the guide on how to remember the plot of a book in 5 steps.

Event segmentation and narrative transportation: how are they related?

Segmentation concerns the organisation of what happens in the story. Narrative transportation concerns the reader's involvement, with attention, emotions and imagination directed towards the narrated world.

They are complementary perspectives. Recovering a missing piece of information can help you carry on reading, but on its own it does not guarantee that immersion will return.

To explore this second perspective, read The Science of Narrative Transportation.